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 The ``inalloca`` argument attribute allows the caller to take the
769 address of outgoing stack arguments. An ``inalloca`` argument must
770 be a pointer to stack memory produced by an ``alloca`` instruction.
771 The alloca, or argument allocation, must also be tagged with the
772 inalloca keyword. Only the past argument may have the ``inalloca``
773 attribute, and that argument is guaranteed to be passed in memory.
775 An argument allocation may be used by a call at most once because
776 the call may deallocate it. The ``inalloca`` attribute cannot be
777 used in conjunction with other attributes that affect argument
778 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
779 ``inalloca`` attribute also disables LLVM's implicit lowering of
780 large aggregate return values, which means that frontend authors
781 must lower them with ``sret`` pointers.
783 When the call site is reached, the argument allocation must have
784 been the most recent stack allocation that is still live, or the
785 results are undefined. It is possible to allocate additional stack
786 space after an argument allocation and before its call site, but it
787 must be cleared off with :ref:`llvm.stackrestore
790 See :doc:`InAlloca` for more information on how to use this
794 This indicates that the pointer parameter specifies the address of a
795 structure that is the return value of the function in the source
796 program. This pointer must be guaranteed by the caller to be valid:
797 loads and stores to the structure may be assumed by the callee
798 not to trap and to be properly aligned. This may only be applied to
799 the first parameter. This is not a valid attribute for return
802 This indicates that pointer values :ref:`based <pointeraliasing>` on
803 the argument or return value do not alias pointer values which are
804 not *based* on it, ignoring certain "irrelevant" dependencies. For a
805 call to the parent function, dependencies between memory references
806 from before or after the call and from those during the call are
807 "irrelevant" to the ``noalias`` keyword for the arguments and return
808 value used in that call. The caller shares the responsibility with
809 the callee for ensuring that these requirements are met. For further
810 details, please see the discussion of the NoAlias response in `alias
811 analysis <AliasAnalysis.html#MustMayNo>`_.
813 Note that this definition of ``noalias`` is intentionally similar
814 to the definition of ``restrict`` in C99 for function arguments,
815 though it is slightly weaker.
817 For function return values, C99's ``restrict`` is not meaningful,
818 while LLVM's ``noalias`` is.
820 This indicates that the callee does not make any copies of the
821 pointer that outlive the callee itself. This is not a valid
822 attribute for return values.
827 This indicates that the pointer parameter can be excised using the
828 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
829 attribute for return values and can only be applied to one parameter.
832 This indicates that the function always returns the argument as its return
833 value. This is an optimization hint to the code generator when generating
834 the caller, allowing tail call optimization and omission of register saves
835 and restores in some cases; it is not checked or enforced when generating
836 the callee. The parameter and the function return type must be valid
837 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
838 valid attribute for return values and can only be applied to one parameter.
842 Garbage Collector Names
843 -----------------------
845 Each function may specify a garbage collector name, which is simply a
850 define void @f() gc "name" { ... }
852 The compiler declares the supported values of *name*. Specifying a
853 collector which will cause the compiler to alter its output in order to
854 support the named garbage collection algorithm.
861 Prefix data is data associated with a function which the code generator
862 will emit immediately before the function body. The purpose of this feature
863 is to allow frontends to associate language-specific runtime metadata with
864 specific functions and make it available through the function pointer while
865 still allowing the function pointer to be called. To access the data for a
866 given function, a program may bitcast the function pointer to a pointer to
867 the constant's type. This implies that the IR symbol points to the start
870 To maintain the semantics of ordinary function calls, the prefix data must
871 have a particular format. Specifically, it must begin with a sequence of
872 bytes which decode to a sequence of machine instructions, valid for the
873 module's target, which transfer control to the point immediately succeeding
874 the prefix data, without performing any other visible action. This allows
875 the inliner and other passes to reason about the semantics of the function
876 definition without needing to reason about the prefix data. Obviously this
877 makes the format of the prefix data highly target dependent.
879 Prefix data is laid out as if it were an initializer for a global variable
880 of the prefix data's type. No padding is automatically placed between the
881 prefix data and the function body. If padding is required, it must be part
884 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
885 which encodes the ``nop`` instruction:
889 define void @f() prefix i8 144 { ... }
891 Generally prefix data can be formed by encoding a relative branch instruction
892 which skips the metadata, as in this example of valid prefix data for the
893 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
897 %0 = type <{ i8, i8, i8* }>
899 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
901 A function may have prefix data but no body. This has similar semantics
902 to the ``available_externally`` linkage in that the data may be used by the
903 optimizers but will not be emitted in the object file.
910 Attribute groups are groups of attributes that are referenced by objects within
911 the IR. They are important for keeping ``.ll`` files readable, because a lot of
912 functions will use the same set of attributes. In the degenerative case of a
913 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
914 group will capture the important command line flags used to build that file.
916 An attribute group is a module-level object. To use an attribute group, an
917 object references the attribute group's ID (e.g. ``#37``). An object may refer
918 to more than one attribute group. In that situation, the attributes from the
919 different groups are merged.
921 Here is an example of attribute groups for a function that should always be
922 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
926 ; Target-independent attributes:
927 attributes #0 = { alwaysinline alignstack=4 }
929 ; Target-dependent attributes:
930 attributes #1 = { "no-sse" }
932 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
933 define void @f() #0 #1 { ... }
940 Function attributes are set to communicate additional information about
941 a function. Function attributes are considered to be part of the
942 function, not of the function type, so functions with different function
943 attributes can have the same function type.
945 Function attributes are simple keywords that follow the type specified.
946 If multiple attributes are needed, they are space separated. For
951 define void @f() noinline { ... }
952 define void @f() alwaysinline { ... }
953 define void @f() alwaysinline optsize { ... }
954 define void @f() optsize { ... }
957 This attribute indicates that, when emitting the prologue and
958 epilogue, the backend should forcibly align the stack pointer.
959 Specify the desired alignment, which must be a power of two, in
962 This attribute indicates that the inliner should attempt to inline
963 this function into callers whenever possible, ignoring any active
964 inlining size threshold for this caller.
966 This indicates that the callee function at a call site should be
967 recognized as a built-in function, even though the function's declaration
968 uses the ``nobuiltin`` attribute. This is only valid at call sites for
969 direct calls to functions which are declared with the ``nobuiltin``
972 This attribute indicates that this function is rarely called. When
973 computing edge weights, basic blocks post-dominated by a cold
974 function call are also considered to be cold; and, thus, given low
977 This attribute indicates that the source code contained a hint that
978 inlining this function is desirable (such as the "inline" keyword in
979 C/C++). It is just a hint; it imposes no requirements on the
982 This attribute suggests that optimization passes and code generator
983 passes make choices that keep the code size of this function as small
984 as possible and perform optimizations that may sacrifice runtime
985 performance in order to minimize the size of the generated code.
987 This attribute disables prologue / epilogue emission for the
988 function. This can have very system-specific consequences.
990 This indicates that the callee function at a call site is not recognized as
991 a built-in function. LLVM will retain the original call and not replace it
992 with equivalent code based on the semantics of the built-in function, unless
993 the call site uses the ``builtin`` attribute. This is valid at call sites
994 and on function declarations and definitions.
996 This attribute indicates that calls to the function cannot be
997 duplicated. A call to a ``noduplicate`` function may be moved
998 within its parent function, but may not be duplicated within
1001 A function containing a ``noduplicate`` call may still
1002 be an inlining candidate, provided that the call is not
1003 duplicated by inlining. That implies that the function has
1004 internal linkage and only has one call site, so the original
1005 call is dead after inlining.
1007 This attributes disables implicit floating point instructions.
1009 This attribute indicates that the inliner should never inline this
1010 function in any situation. This attribute may not be used together
1011 with the ``alwaysinline`` attribute.
1013 This attribute suppresses lazy symbol binding for the function. This
1014 may make calls to the function faster, at the cost of extra program
1015 startup time if the function is not called during program startup.
1017 This attribute indicates that the code generator should not use a
1018 red zone, even if the target-specific ABI normally permits it.
1020 This function attribute indicates that the function never returns
1021 normally. This produces undefined behavior at runtime if the
1022 function ever does dynamically return.
1024 This function attribute indicates that the function never returns
1025 with an unwind or exceptional control flow. If the function does
1026 unwind, its runtime behavior is undefined.
1028 This function attribute indicates that the function is not optimized
1029 by any optimization or code generator passes with the
1030 exception of interprocedural optimization passes.
1031 This attribute cannot be used together with the ``alwaysinline``
1032 attribute; this attribute is also incompatible
1033 with the ``minsize`` attribute and the ``optsize`` attribute.
1035 This attribute requires the ``noinline`` attribute to be specified on
1036 the function as well, so the function is never inlined into any caller.
1037 Only functions with the ``alwaysinline`` attribute are valid
1038 candidates for inlining into the body of this function.
1040 This attribute suggests that optimization passes and code generator
1041 passes make choices that keep the code size of this function low,
1042 and otherwise do optimizations specifically to reduce code size as
1043 long as they do not significantly impact runtime performance.
1045 On a function, this attribute indicates that the function computes its
1046 result (or decides to unwind an exception) based strictly on its arguments,
1047 without dereferencing any pointer arguments or otherwise accessing
1048 any mutable state (e.g. memory, control registers, etc) visible to
1049 caller functions. It does not write through any pointer arguments
1050 (including ``byval`` arguments) and never changes any state visible
1051 to callers. This means that it cannot unwind exceptions by calling
1052 the ``C++`` exception throwing methods.
1054 On an argument, this attribute indicates that the function does not
1055 dereference that pointer argument, even though it may read or write the
1056 memory that the pointer points to if accessed through other pointers.
1058 On a function, this attribute indicates that the function does not write
1059 through any pointer arguments (including ``byval`` arguments) or otherwise
1060 modify any state (e.g. memory, control registers, etc) visible to
1061 caller functions. It may dereference pointer arguments and read
1062 state that may be set in the caller. A readonly function always
1063 returns the same value (or unwinds an exception identically) when
1064 called with the same set of arguments and global state. It cannot
1065 unwind an exception by calling the ``C++`` exception throwing
1068 On an argument, this attribute indicates that the function does not write
1069 through this pointer argument, even though it may write to the memory that
1070 the pointer points to.
1072 This attribute indicates that this function can return twice. The C
1073 ``setjmp`` is an example of such a function. The compiler disables
1074 some optimizations (like tail calls) in the caller of these
1076 ``sanitize_address``
1077 This attribute indicates that AddressSanitizer checks
1078 (dynamic address safety analysis) are enabled for this function.
1080 This attribute indicates that MemorySanitizer checks (dynamic detection
1081 of accesses to uninitialized memory) are enabled for this function.
1083 This attribute indicates that ThreadSanitizer checks
1084 (dynamic thread safety analysis) are enabled for this function.
1086 This attribute indicates that the function should emit a stack
1087 smashing protector. It is in the form of a "canary" --- a random value
1088 placed on the stack before the local variables that's checked upon
1089 return from the function to see if it has been overwritten. A
1090 heuristic is used to determine if a function needs stack protectors
1091 or not. The heuristic used will enable protectors for functions with:
1093 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1094 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1095 - Calls to alloca() with variable sizes or constant sizes greater than
1096 ``ssp-buffer-size``.
1098 Variables that are identified as requiring a protector will be arranged
1099 on the stack such that they are adjacent to the stack protector guard.
1101 If a function that has an ``ssp`` attribute is inlined into a
1102 function that doesn't have an ``ssp`` attribute, then the resulting
1103 function will have an ``ssp`` attribute.
1105 This attribute indicates that the function should *always* emit a
1106 stack smashing protector. This overrides the ``ssp`` function
1109 Variables that are identified as requiring a protector will be arranged
1110 on the stack such that they are adjacent to the stack protector guard.
1111 The specific layout rules are:
1113 #. Large arrays and structures containing large arrays
1114 (``>= ssp-buffer-size``) are closest to the stack protector.
1115 #. Small arrays and structures containing small arrays
1116 (``< ssp-buffer-size``) are 2nd closest to the protector.
1117 #. Variables that have had their address taken are 3rd closest to the
1120 If a function that has an ``sspreq`` attribute is inlined into a
1121 function that doesn't have an ``sspreq`` attribute or which has an
1122 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1123 an ``sspreq`` attribute.
1125 This attribute indicates that the function should emit a stack smashing
1126 protector. This attribute causes a strong heuristic to be used when
1127 determining if a function needs stack protectors. The strong heuristic
1128 will enable protectors for functions with:
1130 - Arrays of any size and type
1131 - Aggregates containing an array of any size and type.
1132 - Calls to alloca().
1133 - Local variables that have had their address taken.
1135 Variables that are identified as requiring a protector will be arranged
1136 on the stack such that they are adjacent to the stack protector guard.
1137 The specific layout rules are:
1139 #. Large arrays and structures containing large arrays
1140 (``>= ssp-buffer-size``) are closest to the stack protector.
1141 #. Small arrays and structures containing small arrays
1142 (``< ssp-buffer-size``) are 2nd closest to the protector.
1143 #. Variables that have had their address taken are 3rd closest to the
1146 This overrides the ``ssp`` function attribute.
1148 If a function that has an ``sspstrong`` attribute is inlined into a
1149 function that doesn't have an ``sspstrong`` attribute, then the
1150 resulting function will have an ``sspstrong`` attribute.
1152 This attribute indicates that the ABI being targeted requires that
1153 an unwind table entry be produce for this function even if we can
1154 show that no exceptions passes by it. This is normally the case for
1155 the ELF x86-64 abi, but it can be disabled for some compilation
1160 Module-Level Inline Assembly
1161 ----------------------------
1163 Modules may contain "module-level inline asm" blocks, which corresponds
1164 to the GCC "file scope inline asm" blocks. These blocks are internally
1165 concatenated by LLVM and treated as a single unit, but may be separated
1166 in the ``.ll`` file if desired. The syntax is very simple:
1168 .. code-block:: llvm
1170 module asm "inline asm code goes here"
1171 module asm "more can go here"
1173 The strings can contain any character by escaping non-printable
1174 characters. The escape sequence used is simply "\\xx" where "xx" is the
1175 two digit hex code for the number.
1177 The inline asm code is simply printed to the machine code .s file when
1178 assembly code is generated.
1180 .. _langref_datalayout:
1185 A module may specify a target specific data layout string that specifies
1186 how data is to be laid out in memory. The syntax for the data layout is
1189 .. code-block:: llvm
1191 target datalayout = "layout specification"
1193 The *layout specification* consists of a list of specifications
1194 separated by the minus sign character ('-'). Each specification starts
1195 with a letter and may include other information after the letter to
1196 define some aspect of the data layout. The specifications accepted are
1200 Specifies that the target lays out data in big-endian form. That is,
1201 the bits with the most significance have the lowest address
1204 Specifies that the target lays out data in little-endian form. That
1205 is, the bits with the least significance have the lowest address
1208 Specifies the natural alignment of the stack in bits. Alignment
1209 promotion of stack variables is limited to the natural stack
1210 alignment to avoid dynamic stack realignment. The stack alignment
1211 must be a multiple of 8-bits. If omitted, the natural stack
1212 alignment defaults to "unspecified", which does not prevent any
1213 alignment promotions.
1214 ``p[n]:<size>:<abi>:<pref>``
1215 This specifies the *size* of a pointer and its ``<abi>`` and
1216 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1217 bits. The address space, ``n`` is optional, and if not specified,
1218 denotes the default address space 0. The value of ``n`` must be
1219 in the range [1,2^23).
1220 ``i<size>:<abi>:<pref>``
1221 This specifies the alignment for an integer type of a given bit
1222 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1223 ``v<size>:<abi>:<pref>``
1224 This specifies the alignment for a vector type of a given bit
1226 ``f<size>:<abi>:<pref>``
1227 This specifies the alignment for a floating point type of a given bit
1228 ``<size>``. Only values of ``<size>`` that are supported by the target
1229 will work. 32 (float) and 64 (double) are supported on all targets; 80
1230 or 128 (different flavors of long double) are also supported on some
1233 This specifies the alignment for an object of aggregate type.
1235 If present, specifies that llvm names are mangled in the output. The
1238 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1239 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1240 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1241 symbols get a ``_`` prefix.
1242 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1243 functions also get a suffix based on the frame size.
1244 ``n<size1>:<size2>:<size3>...``
1245 This specifies a set of native integer widths for the target CPU in
1246 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1247 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1248 this set are considered to support most general arithmetic operations
1251 On every specification that takes a ``<abi>:<pref>``, specifying the
1252 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1253 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1255 When constructing the data layout for a given target, LLVM starts with a
1256 default set of specifications which are then (possibly) overridden by
1257 the specifications in the ``datalayout`` keyword. The default
1258 specifications are given in this list:
1260 - ``E`` - big endian
1261 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1262 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1263 same as the default address space.
1264 - ``S0`` - natural stack alignment is unspecified
1265 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1266 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1267 - ``i16:16:16`` - i16 is 16-bit aligned
1268 - ``i32:32:32`` - i32 is 32-bit aligned
1269 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1270 alignment of 64-bits
1271 - ``f16:16:16`` - half is 16-bit aligned
1272 - ``f32:32:32`` - float is 32-bit aligned
1273 - ``f64:64:64`` - double is 64-bit aligned
1274 - ``f128:128:128`` - quad is 128-bit aligned
1275 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1276 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1277 - ``a:0:64`` - aggregates are 64-bit aligned
1279 When LLVM is determining the alignment for a given type, it uses the
1282 #. If the type sought is an exact match for one of the specifications,
1283 that specification is used.
1284 #. If no match is found, and the type sought is an integer type, then
1285 the smallest integer type that is larger than the bitwidth of the
1286 sought type is used. If none of the specifications are larger than
1287 the bitwidth then the largest integer type is used. For example,
1288 given the default specifications above, the i7 type will use the
1289 alignment of i8 (next largest) while both i65 and i256 will use the
1290 alignment of i64 (largest specified).
1291 #. If no match is found, and the type sought is a vector type, then the
1292 largest vector type that is smaller than the sought vector type will
1293 be used as a fall back. This happens because <128 x double> can be
1294 implemented in terms of 64 <2 x double>, for example.
1296 The function of the data layout string may not be what you expect.
1297 Notably, this is not a specification from the frontend of what alignment
1298 the code generator should use.
1300 Instead, if specified, the target data layout is required to match what
1301 the ultimate *code generator* expects. This string is used by the
1302 mid-level optimizers to improve code, and this only works if it matches
1303 what the ultimate code generator uses. If you would like to generate IR
1304 that does not embed this target-specific detail into the IR, then you
1305 don't have to specify the string. This will disable some optimizations
1306 that require precise layout information, but this also prevents those
1307 optimizations from introducing target specificity into the IR.
1314 A module may specify a target triple string that describes the target
1315 host. The syntax for the target triple is simply:
1317 .. code-block:: llvm
1319 target triple = "x86_64-apple-macosx10.7.0"
1321 The *target triple* string consists of a series of identifiers delimited
1322 by the minus sign character ('-'). The canonical forms are:
1326 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1327 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1329 This information is passed along to the backend so that it generates
1330 code for the proper architecture. It's possible to override this on the
1331 command line with the ``-mtriple`` command line option.
1333 .. _pointeraliasing:
1335 Pointer Aliasing Rules
1336 ----------------------
1338 Any memory access must be done through a pointer value associated with
1339 an address range of the memory access, otherwise the behavior is
1340 undefined. Pointer values are associated with address ranges according
1341 to the following rules:
1343 - A pointer value is associated with the addresses associated with any
1344 value it is *based* on.
1345 - An address of a global variable is associated with the address range
1346 of the variable's storage.
1347 - The result value of an allocation instruction is associated with the
1348 address range of the allocated storage.
1349 - A null pointer in the default address-space is associated with no
1351 - An integer constant other than zero or a pointer value returned from
1352 a function not defined within LLVM may be associated with address
1353 ranges allocated through mechanisms other than those provided by
1354 LLVM. Such ranges shall not overlap with any ranges of addresses
1355 allocated by mechanisms provided by LLVM.
1357 A pointer value is *based* on another pointer value according to the
1360 - A pointer value formed from a ``getelementptr`` operation is *based*
1361 on the first operand of the ``getelementptr``.
1362 - The result value of a ``bitcast`` is *based* on the operand of the
1364 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1365 values that contribute (directly or indirectly) to the computation of
1366 the pointer's value.
1367 - The "*based* on" relationship is transitive.
1369 Note that this definition of *"based"* is intentionally similar to the
1370 definition of *"based"* in C99, though it is slightly weaker.
1372 LLVM IR does not associate types with memory. The result type of a
1373 ``load`` merely indicates the size and alignment of the memory from
1374 which to load, as well as the interpretation of the value. The first
1375 operand type of a ``store`` similarly only indicates the size and
1376 alignment of the store.
1378 Consequently, type-based alias analysis, aka TBAA, aka
1379 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1380 :ref:`Metadata <metadata>` may be used to encode additional information
1381 which specialized optimization passes may use to implement type-based
1386 Volatile Memory Accesses
1387 ------------------------
1389 Certain memory accesses, such as :ref:`load <i_load>`'s,
1390 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1391 marked ``volatile``. The optimizers must not change the number of
1392 volatile operations or change their order of execution relative to other
1393 volatile operations. The optimizers *may* change the order of volatile
1394 operations relative to non-volatile operations. This is not Java's
1395 "volatile" and has no cross-thread synchronization behavior.
1397 IR-level volatile loads and stores cannot safely be optimized into
1398 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1399 flagged volatile. Likewise, the backend should never split or merge
1400 target-legal volatile load/store instructions.
1402 .. admonition:: Rationale
1404 Platforms may rely on volatile loads and stores of natively supported
1405 data width to be executed as single instruction. For example, in C
1406 this holds for an l-value of volatile primitive type with native
1407 hardware support, but not necessarily for aggregate types. The
1408 frontend upholds these expectations, which are intentionally
1409 unspecified in the IR. The rules above ensure that IR transformation
1410 do not violate the frontend's contract with the language.
1414 Memory Model for Concurrent Operations
1415 --------------------------------------
1417 The LLVM IR does not define any way to start parallel threads of
1418 execution or to register signal handlers. Nonetheless, there are
1419 platform-specific ways to create them, and we define LLVM IR's behavior
1420 in their presence. This model is inspired by the C++0x memory model.
1422 For a more informal introduction to this model, see the :doc:`Atomics`.
1424 We define a *happens-before* partial order as the least partial order
1427 - Is a superset of single-thread program order, and
1428 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1429 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1430 techniques, like pthread locks, thread creation, thread joining,
1431 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1432 Constraints <ordering>`).
1434 Note that program order does not introduce *happens-before* edges
1435 between a thread and signals executing inside that thread.
1437 Every (defined) read operation (load instructions, memcpy, atomic
1438 loads/read-modify-writes, etc.) R reads a series of bytes written by
1439 (defined) write operations (store instructions, atomic
1440 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1441 section, initialized globals are considered to have a write of the
1442 initializer which is atomic and happens before any other read or write
1443 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1444 may see any write to the same byte, except:
1446 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1447 write\ :sub:`2` happens before R\ :sub:`byte`, then
1448 R\ :sub:`byte` does not see write\ :sub:`1`.
1449 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1450 R\ :sub:`byte` does not see write\ :sub:`3`.
1452 Given that definition, R\ :sub:`byte` is defined as follows:
1454 - If R is volatile, the result is target-dependent. (Volatile is
1455 supposed to give guarantees which can support ``sig_atomic_t`` in
1456 C/C++, and may be used for accesses to addresses which do not behave
1457 like normal memory. It does not generally provide cross-thread
1459 - Otherwise, if there is no write to the same byte that happens before
1460 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1461 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1462 R\ :sub:`byte` returns the value written by that write.
1463 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1464 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1465 Memory Ordering Constraints <ordering>` section for additional
1466 constraints on how the choice is made.
1467 - Otherwise R\ :sub:`byte` returns ``undef``.
1469 R returns the value composed of the series of bytes it read. This
1470 implies that some bytes within the value may be ``undef`` **without**
1471 the entire value being ``undef``. Note that this only defines the
1472 semantics of the operation; it doesn't mean that targets will emit more
1473 than one instruction to read the series of bytes.
1475 Note that in cases where none of the atomic intrinsics are used, this
1476 model places only one restriction on IR transformations on top of what
1477 is required for single-threaded execution: introducing a store to a byte
1478 which might not otherwise be stored is not allowed in general.
1479 (Specifically, in the case where another thread might write to and read
1480 from an address, introducing a store can change a load that may see
1481 exactly one write into a load that may see multiple writes.)
1485 Atomic Memory Ordering Constraints
1486 ----------------------------------
1488 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1489 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1490 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1491 ordering parameters that determine which other atomic instructions on
1492 the same address they *synchronize with*. These semantics are borrowed
1493 from Java and C++0x, but are somewhat more colloquial. If these
1494 descriptions aren't precise enough, check those specs (see spec
1495 references in the :doc:`atomics guide <Atomics>`).
1496 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1497 differently since they don't take an address. See that instruction's
1498 documentation for details.
1500 For a simpler introduction to the ordering constraints, see the
1504 The set of values that can be read is governed by the happens-before
1505 partial order. A value cannot be read unless some operation wrote
1506 it. This is intended to provide a guarantee strong enough to model
1507 Java's non-volatile shared variables. This ordering cannot be
1508 specified for read-modify-write operations; it is not strong enough
1509 to make them atomic in any interesting way.
1511 In addition to the guarantees of ``unordered``, there is a single
1512 total order for modifications by ``monotonic`` operations on each
1513 address. All modification orders must be compatible with the
1514 happens-before order. There is no guarantee that the modification
1515 orders can be combined to a global total order for the whole program
1516 (and this often will not be possible). The read in an atomic
1517 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1518 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1519 order immediately before the value it writes. If one atomic read
1520 happens before another atomic read of the same address, the later
1521 read must see the same value or a later value in the address's
1522 modification order. This disallows reordering of ``monotonic`` (or
1523 stronger) operations on the same address. If an address is written
1524 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1525 read that address repeatedly, the other threads must eventually see
1526 the write. This corresponds to the C++0x/C1x
1527 ``memory_order_relaxed``.
1529 In addition to the guarantees of ``monotonic``, a
1530 *synchronizes-with* edge may be formed with a ``release`` operation.
1531 This is intended to model C++'s ``memory_order_acquire``.
1533 In addition to the guarantees of ``monotonic``, if this operation
1534 writes a value which is subsequently read by an ``acquire``
1535 operation, it *synchronizes-with* that operation. (This isn't a
1536 complete description; see the C++0x definition of a release
1537 sequence.) This corresponds to the C++0x/C1x
1538 ``memory_order_release``.
1539 ``acq_rel`` (acquire+release)
1540 Acts as both an ``acquire`` and ``release`` operation on its
1541 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1542 ``seq_cst`` (sequentially consistent)
1543 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1544 operation which only reads, ``release`` for an operation which only
1545 writes), there is a global total order on all
1546 sequentially-consistent operations on all addresses, which is
1547 consistent with the *happens-before* partial order and with the
1548 modification orders of all the affected addresses. Each
1549 sequentially-consistent read sees the last preceding write to the
1550 same address in this global order. This corresponds to the C++0x/C1x
1551 ``memory_order_seq_cst`` and Java volatile.
1555 If an atomic operation is marked ``singlethread``, it only *synchronizes
1556 with* or participates in modification and seq\_cst total orderings with
1557 other operations running in the same thread (for example, in signal
1565 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1566 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1567 :ref:`frem <i_frem>`) have the following flags that can set to enable
1568 otherwise unsafe floating point operations
1571 No NaNs - Allow optimizations to assume the arguments and result are not
1572 NaN. Such optimizations are required to retain defined behavior over
1573 NaNs, but the value of the result is undefined.
1576 No Infs - Allow optimizations to assume the arguments and result are not
1577 +/-Inf. Such optimizations are required to retain defined behavior over
1578 +/-Inf, but the value of the result is undefined.
1581 No Signed Zeros - Allow optimizations to treat the sign of a zero
1582 argument or result as insignificant.
1585 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1586 argument rather than perform division.
1589 Fast - Allow algebraically equivalent transformations that may
1590 dramatically change results in floating point (e.g. reassociate). This
1591 flag implies all the others.
1598 The LLVM type system is one of the most important features of the
1599 intermediate representation. Being typed enables a number of
1600 optimizations to be performed on the intermediate representation
1601 directly, without having to do extra analyses on the side before the
1602 transformation. A strong type system makes it easier to read the
1603 generated code and enables novel analyses and transformations that are
1604 not feasible to perform on normal three address code representations.
1614 The void type does not represent any value and has no size.
1632 The function type can be thought of as a function signature. It consists of a
1633 return type and a list of formal parameter types. The return type of a function
1634 type is a void type or first class type --- except for :ref:`label <t_label>`
1635 and :ref:`metadata <t_metadata>` types.
1641 <returntype> (<parameter list>)
1643 ...where '``<parameter list>``' is a comma-separated list of type
1644 specifiers. Optionally, the parameter list may include a type ``...``, which
1645 indicates that the function takes a variable number of arguments. Variable
1646 argument functions can access their arguments with the :ref:`variable argument
1647 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1648 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1652 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1653 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1654 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1655 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1656 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1657 | ``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. |
1658 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1659 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1660 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1667 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1668 Values of these types are the only ones which can be produced by
1676 These are the types that are valid in registers from CodeGen's perspective.
1685 The integer type is a very simple type that simply specifies an
1686 arbitrary bit width for the integer type desired. Any bit width from 1
1687 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1695 The number of bits the integer will occupy is specified by the ``N``
1701 +----------------+------------------------------------------------+
1702 | ``i1`` | a single-bit integer. |
1703 +----------------+------------------------------------------------+
1704 | ``i32`` | a 32-bit integer. |
1705 +----------------+------------------------------------------------+
1706 | ``i1942652`` | a really big integer of over 1 million bits. |
1707 +----------------+------------------------------------------------+
1711 Floating Point Types
1712 """"""""""""""""""""
1721 - 16-bit floating point value
1724 - 32-bit floating point value
1727 - 64-bit floating point value
1730 - 128-bit floating point value (112-bit mantissa)
1733 - 80-bit floating point value (X87)
1736 - 128-bit floating point value (two 64-bits)
1743 The x86_mmx type represents a value held in an MMX register on an x86
1744 machine. The operations allowed on it are quite limited: parameters and
1745 return values, load and store, and bitcast. User-specified MMX
1746 instructions are represented as intrinsic or asm calls with arguments
1747 and/or results of this type. There are no arrays, vectors or constants
1764 The pointer type is used to specify memory locations. Pointers are
1765 commonly used to reference objects in memory.
1767 Pointer types may have an optional address space attribute defining the
1768 numbered address space where the pointed-to object resides. The default
1769 address space is number zero. The semantics of non-zero address spaces
1770 are target-specific.
1772 Note that LLVM does not permit pointers to void (``void*``) nor does it
1773 permit pointers to labels (``label*``). Use ``i8*`` instead.
1783 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1784 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1785 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1786 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1787 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1788 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1789 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1798 A vector type is a simple derived type that represents a vector of
1799 elements. Vector types are used when multiple primitive data are
1800 operated in parallel using a single instruction (SIMD). A vector type
1801 requires a size (number of elements) and an underlying primitive data
1802 type. Vector types are considered :ref:`first class <t_firstclass>`.
1808 < <# elements> x <elementtype> >
1810 The number of elements is a constant integer value larger than 0;
1811 elementtype may be any integer or floating point type, or a pointer to
1812 these types. Vectors of size zero are not allowed.
1816 +-------------------+--------------------------------------------------+
1817 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1818 +-------------------+--------------------------------------------------+
1819 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1820 +-------------------+--------------------------------------------------+
1821 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1822 +-------------------+--------------------------------------------------+
1823 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1824 +-------------------+--------------------------------------------------+
1833 The label type represents code labels.
1848 The metadata type represents embedded metadata. No derived types may be
1849 created from metadata except for :ref:`function <t_function>` arguments.
1862 Aggregate Types are a subset of derived types that can contain multiple
1863 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1864 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1874 The array type is a very simple derived type that arranges elements
1875 sequentially in memory. The array type requires a size (number of
1876 elements) and an underlying data type.
1882 [<# elements> x <elementtype>]
1884 The number of elements is a constant integer value; ``elementtype`` may
1885 be any type with a size.
1889 +------------------+--------------------------------------+
1890 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1891 +------------------+--------------------------------------+
1892 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1893 +------------------+--------------------------------------+
1894 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1895 +------------------+--------------------------------------+
1897 Here are some examples of multidimensional arrays:
1899 +-----------------------------+----------------------------------------------------------+
1900 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1901 +-----------------------------+----------------------------------------------------------+
1902 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1903 +-----------------------------+----------------------------------------------------------+
1904 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1905 +-----------------------------+----------------------------------------------------------+
1907 There is no restriction on indexing beyond the end of the array implied
1908 by a static type (though there are restrictions on indexing beyond the
1909 bounds of an allocated object in some cases). This means that
1910 single-dimension 'variable sized array' addressing can be implemented in
1911 LLVM with a zero length array type. An implementation of 'pascal style
1912 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1922 The structure type is used to represent a collection of data members
1923 together in memory. The elements of a structure may be any type that has
1926 Structures in memory are accessed using '``load``' and '``store``' by
1927 getting a pointer to a field with the '``getelementptr``' instruction.
1928 Structures in registers are accessed using the '``extractvalue``' and
1929 '``insertvalue``' instructions.
1931 Structures may optionally be "packed" structures, which indicate that
1932 the alignment of the struct is one byte, and that there is no padding
1933 between the elements. In non-packed structs, padding between field types
1934 is inserted as defined by the DataLayout string in the module, which is
1935 required to match what the underlying code generator expects.
1937 Structures can either be "literal" or "identified". A literal structure
1938 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1939 identified types are always defined at the top level with a name.
1940 Literal types are uniqued by their contents and can never be recursive
1941 or opaque since there is no way to write one. Identified types can be
1942 recursive, can be opaqued, and are never uniqued.
1948 %T1 = type { <type list> } ; Identified normal struct type
1949 %T2 = type <{ <type list> }> ; Identified packed struct type
1953 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1954 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1955 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1956 | ``{ 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``. |
1957 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1958 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1959 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1963 Opaque Structure Types
1964 """"""""""""""""""""""
1968 Opaque structure types are used to represent named structure types that
1969 do not have a body specified. This corresponds (for example) to the C
1970 notion of a forward declared structure.
1981 +--------------+-------------------+
1982 | ``opaque`` | An opaque type. |
1983 +--------------+-------------------+
1988 LLVM has several different basic types of constants. This section
1989 describes them all and their syntax.
1994 **Boolean constants**
1995 The two strings '``true``' and '``false``' are both valid constants
1997 **Integer constants**
1998 Standard integers (such as '4') are constants of the
1999 :ref:`integer <t_integer>` type. Negative numbers may be used with
2001 **Floating point constants**
2002 Floating point constants use standard decimal notation (e.g.
2003 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2004 hexadecimal notation (see below). The assembler requires the exact
2005 decimal value of a floating-point constant. For example, the
2006 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2007 decimal in binary. Floating point constants must have a :ref:`floating
2008 point <t_floating>` type.
2009 **Null pointer constants**
2010 The identifier '``null``' is recognized as a null pointer constant
2011 and must be of :ref:`pointer type <t_pointer>`.
2013 The one non-intuitive notation for constants is the hexadecimal form of
2014 floating point constants. For example, the form
2015 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2016 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2017 constants are required (and the only time that they are generated by the
2018 disassembler) is when a floating point constant must be emitted but it
2019 cannot be represented as a decimal floating point number in a reasonable
2020 number of digits. For example, NaN's, infinities, and other special
2021 values are represented in their IEEE hexadecimal format so that assembly
2022 and disassembly do not cause any bits to change in the constants.
2024 When using the hexadecimal form, constants of types half, float, and
2025 double are represented using the 16-digit form shown above (which
2026 matches the IEEE754 representation for double); half and float values
2027 must, however, be exactly representable as IEEE 754 half and single
2028 precision, respectively. Hexadecimal format is always used for long
2029 double, and there are three forms of long double. The 80-bit format used
2030 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2031 128-bit format used by PowerPC (two adjacent doubles) is represented by
2032 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2033 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2034 will only work if they match the long double format on your target.
2035 The IEEE 16-bit format (half precision) is represented by ``0xH``
2036 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2037 (sign bit at the left).
2039 There are no constants of type x86_mmx.
2041 .. _complexconstants:
2046 Complex constants are a (potentially recursive) combination of simple
2047 constants and smaller complex constants.
2049 **Structure constants**
2050 Structure constants are represented with notation similar to
2051 structure type definitions (a comma separated list of elements,
2052 surrounded by braces (``{}``)). For example:
2053 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2054 "``@G = external global i32``". Structure constants must have
2055 :ref:`structure type <t_struct>`, and the number and types of elements
2056 must match those specified by the type.
2058 Array constants are represented with notation similar to array type
2059 definitions (a comma separated list of elements, surrounded by
2060 square brackets (``[]``)). For example:
2061 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2062 :ref:`array type <t_array>`, and the number and types of elements must
2063 match those specified by the type.
2064 **Vector constants**
2065 Vector constants are represented with notation similar to vector
2066 type definitions (a comma separated list of elements, surrounded by
2067 less-than/greater-than's (``<>``)). For example:
2068 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2069 must have :ref:`vector type <t_vector>`, and the number and types of
2070 elements must match those specified by the type.
2071 **Zero initialization**
2072 The string '``zeroinitializer``' can be used to zero initialize a
2073 value to zero of *any* type, including scalar and
2074 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2075 having to print large zero initializers (e.g. for large arrays) and
2076 is always exactly equivalent to using explicit zero initializers.
2078 A metadata node is a structure-like constant with :ref:`metadata
2079 type <t_metadata>`. For example:
2080 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2081 constants that are meant to be interpreted as part of the
2082 instruction stream, metadata is a place to attach additional
2083 information such as debug info.
2085 Global Variable and Function Addresses
2086 --------------------------------------
2088 The addresses of :ref:`global variables <globalvars>` and
2089 :ref:`functions <functionstructure>` are always implicitly valid
2090 (link-time) constants. These constants are explicitly referenced when
2091 the :ref:`identifier for the global <identifiers>` is used and always have
2092 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2095 .. code-block:: llvm
2099 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2106 The string '``undef``' can be used anywhere a constant is expected, and
2107 indicates that the user of the value may receive an unspecified
2108 bit-pattern. Undefined values may be of any type (other than '``label``'
2109 or '``void``') and be used anywhere a constant is permitted.
2111 Undefined values are useful because they indicate to the compiler that
2112 the program is well defined no matter what value is used. This gives the
2113 compiler more freedom to optimize. Here are some examples of
2114 (potentially surprising) transformations that are valid (in pseudo IR):
2116 .. code-block:: llvm
2126 This is safe because all of the output bits are affected by the undef
2127 bits. Any output bit can have a zero or one depending on the input bits.
2129 .. code-block:: llvm
2140 These logical operations have bits that are not always affected by the
2141 input. For example, if ``%X`` has a zero bit, then the output of the
2142 '``and``' operation will always be a zero for that bit, no matter what
2143 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2144 optimize or assume that the result of the '``and``' is '``undef``'.
2145 However, it is safe to assume that all bits of the '``undef``' could be
2146 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2147 all the bits of the '``undef``' operand to the '``or``' could be set,
2148 allowing the '``or``' to be folded to -1.
2150 .. code-block:: llvm
2152 %A = select undef, %X, %Y
2153 %B = select undef, 42, %Y
2154 %C = select %X, %Y, undef
2164 This set of examples shows that undefined '``select``' (and conditional
2165 branch) conditions can go *either way*, but they have to come from one
2166 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2167 both known to have a clear low bit, then ``%A`` would have to have a
2168 cleared low bit. However, in the ``%C`` example, the optimizer is
2169 allowed to assume that the '``undef``' operand could be the same as
2170 ``%Y``, allowing the whole '``select``' to be eliminated.
2172 .. code-block:: llvm
2174 %A = xor undef, undef
2191 This example points out that two '``undef``' operands are not
2192 necessarily the same. This can be surprising to people (and also matches
2193 C semantics) where they assume that "``X^X``" is always zero, even if
2194 ``X`` is undefined. This isn't true for a number of reasons, but the
2195 short answer is that an '``undef``' "variable" can arbitrarily change
2196 its value over its "live range". This is true because the variable
2197 doesn't actually *have a live range*. Instead, the value is logically
2198 read from arbitrary registers that happen to be around when needed, so
2199 the value is not necessarily consistent over time. In fact, ``%A`` and
2200 ``%C`` need to have the same semantics or the core LLVM "replace all
2201 uses with" concept would not hold.
2203 .. code-block:: llvm
2211 These examples show the crucial difference between an *undefined value*
2212 and *undefined behavior*. An undefined value (like '``undef``') is
2213 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2214 operation can be constant folded to '``undef``', because the '``undef``'
2215 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2216 However, in the second example, we can make a more aggressive
2217 assumption: because the ``undef`` is allowed to be an arbitrary value,
2218 we are allowed to assume that it could be zero. Since a divide by zero
2219 has *undefined behavior*, we are allowed to assume that the operation
2220 does not execute at all. This allows us to delete the divide and all
2221 code after it. Because the undefined operation "can't happen", the
2222 optimizer can assume that it occurs in dead code.
2224 .. code-block:: llvm
2226 a: store undef -> %X
2227 b: store %X -> undef
2232 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2233 value can be assumed to not have any effect; we can assume that the
2234 value is overwritten with bits that happen to match what was already
2235 there. However, a store *to* an undefined location could clobber
2236 arbitrary memory, therefore, it has undefined behavior.
2243 Poison values are similar to :ref:`undef values <undefvalues>`, however
2244 they also represent the fact that an instruction or constant expression
2245 which cannot evoke side effects has nevertheless detected a condition
2246 which results in undefined behavior.
2248 There is currently no way of representing a poison value in the IR; they
2249 only exist when produced by operations such as :ref:`add <i_add>` with
2252 Poison value behavior is defined in terms of value *dependence*:
2254 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2255 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2256 their dynamic predecessor basic block.
2257 - Function arguments depend on the corresponding actual argument values
2258 in the dynamic callers of their functions.
2259 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2260 instructions that dynamically transfer control back to them.
2261 - :ref:`Invoke <i_invoke>` instructions depend on the
2262 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2263 call instructions that dynamically transfer control back to them.
2264 - Non-volatile loads and stores depend on the most recent stores to all
2265 of the referenced memory addresses, following the order in the IR
2266 (including loads and stores implied by intrinsics such as
2267 :ref:`@llvm.memcpy <int_memcpy>`.)
2268 - An instruction with externally visible side effects depends on the
2269 most recent preceding instruction with externally visible side
2270 effects, following the order in the IR. (This includes :ref:`volatile
2271 operations <volatile>`.)
2272 - An instruction *control-depends* on a :ref:`terminator
2273 instruction <terminators>` if the terminator instruction has
2274 multiple successors and the instruction is always executed when
2275 control transfers to one of the successors, and may not be executed
2276 when control is transferred to another.
2277 - Additionally, an instruction also *control-depends* on a terminator
2278 instruction if the set of instructions it otherwise depends on would
2279 be different if the terminator had transferred control to a different
2281 - Dependence is transitive.
2283 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2284 with the additional affect that any instruction which has a *dependence*
2285 on a poison value has undefined behavior.
2287 Here are some examples:
2289 .. code-block:: llvm
2292 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2293 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2294 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2295 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2297 store i32 %poison, i32* @g ; Poison value stored to memory.
2298 %poison2 = load i32* @g ; Poison value loaded back from memory.
2300 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2302 %narrowaddr = bitcast i32* @g to i16*
2303 %wideaddr = bitcast i32* @g to i64*
2304 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2305 %poison4 = load i64* %wideaddr ; Returns a poison value.
2307 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2308 br i1 %cmp, label %true, label %end ; Branch to either destination.
2311 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2312 ; it has undefined behavior.
2316 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2317 ; Both edges into this PHI are
2318 ; control-dependent on %cmp, so this
2319 ; always results in a poison value.
2321 store volatile i32 0, i32* @g ; This would depend on the store in %true
2322 ; if %cmp is true, or the store in %entry
2323 ; otherwise, so this is undefined behavior.
2325 br i1 %cmp, label %second_true, label %second_end
2326 ; The same branch again, but this time the
2327 ; true block doesn't have side effects.
2334 store volatile i32 0, i32* @g ; This time, the instruction always depends
2335 ; on the store in %end. Also, it is
2336 ; control-equivalent to %end, so this is
2337 ; well-defined (ignoring earlier undefined
2338 ; behavior in this example).
2342 Addresses of Basic Blocks
2343 -------------------------
2345 ``blockaddress(@function, %block)``
2347 The '``blockaddress``' constant computes the address of the specified
2348 basic block in the specified function, and always has an ``i8*`` type.
2349 Taking the address of the entry block is illegal.
2351 This value only has defined behavior when used as an operand to the
2352 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2353 against null. Pointer equality tests between labels addresses results in
2354 undefined behavior --- though, again, comparison against null is ok, and
2355 no label is equal to the null pointer. This may be passed around as an
2356 opaque pointer sized value as long as the bits are not inspected. This
2357 allows ``ptrtoint`` and arithmetic to be performed on these values so
2358 long as the original value is reconstituted before the ``indirectbr``
2361 Finally, some targets may provide defined semantics when using the value
2362 as the operand to an inline assembly, but that is target specific.
2366 Constant Expressions
2367 --------------------
2369 Constant expressions are used to allow expressions involving other
2370 constants to be used as constants. Constant expressions may be of any
2371 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2372 that does not have side effects (e.g. load and call are not supported).
2373 The following is the syntax for constant expressions:
2375 ``trunc (CST to TYPE)``
2376 Truncate a constant to another type. The bit size of CST must be
2377 larger than the bit size of TYPE. Both types must be integers.
2378 ``zext (CST to TYPE)``
2379 Zero extend a constant to another type. The bit size of CST must be
2380 smaller than the bit size of TYPE. Both types must be integers.
2381 ``sext (CST to TYPE)``
2382 Sign extend a constant to another type. The bit size of CST must be
2383 smaller than the bit size of TYPE. Both types must be integers.
2384 ``fptrunc (CST to TYPE)``
2385 Truncate a floating point constant to another floating point type.
2386 The size of CST must be larger than the size of TYPE. Both types
2387 must be floating point.
2388 ``fpext (CST to TYPE)``
2389 Floating point extend a constant to another type. The size of CST
2390 must be smaller or equal to the size of TYPE. Both types must be
2392 ``fptoui (CST to TYPE)``
2393 Convert a floating point constant to the corresponding unsigned
2394 integer constant. TYPE must be a scalar or vector integer type. CST
2395 must be of scalar or vector floating point type. Both CST and TYPE
2396 must be scalars, or vectors of the same number of elements. If the
2397 value won't fit in the integer type, the results are undefined.
2398 ``fptosi (CST to TYPE)``
2399 Convert a floating point constant to the corresponding signed
2400 integer constant. TYPE must be a scalar or vector integer type. CST
2401 must be of scalar or vector floating point type. Both CST and TYPE
2402 must be scalars, or vectors of the same number of elements. If the
2403 value won't fit in the integer type, the results are undefined.
2404 ``uitofp (CST to TYPE)``
2405 Convert an unsigned integer constant to the corresponding floating
2406 point constant. TYPE must be a scalar or vector floating point type.
2407 CST must be of scalar or vector integer type. Both CST and TYPE must
2408 be scalars, or vectors of the same number of elements. If the value
2409 won't fit in the floating point type, the results are undefined.
2410 ``sitofp (CST to TYPE)``
2411 Convert a signed integer constant to the corresponding floating
2412 point constant. TYPE must be a scalar or vector floating point type.
2413 CST must be of scalar or vector integer type. Both CST and TYPE must
2414 be scalars, or vectors of the same number of elements. If the value
2415 won't fit in the floating point type, the results are undefined.
2416 ``ptrtoint (CST to TYPE)``
2417 Convert a pointer typed constant to the corresponding integer
2418 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2419 pointer type. The ``CST`` value is zero extended, truncated, or
2420 unchanged to make it fit in ``TYPE``.
2421 ``inttoptr (CST to TYPE)``
2422 Convert an integer constant to a pointer constant. TYPE must be a
2423 pointer type. CST must be of integer type. The CST value is zero
2424 extended, truncated, or unchanged to make it fit in a pointer size.
2425 This one is *really* dangerous!
2426 ``bitcast (CST to TYPE)``
2427 Convert a constant, CST, to another TYPE. The constraints of the
2428 operands are the same as those for the :ref:`bitcast
2429 instruction <i_bitcast>`.
2430 ``addrspacecast (CST to TYPE)``
2431 Convert a constant pointer or constant vector of pointer, CST, to another
2432 TYPE in a different address space. The constraints of the operands are the
2433 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2434 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2435 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2436 constants. As with the :ref:`getelementptr <i_getelementptr>`
2437 instruction, the index list may have zero or more indexes, which are
2438 required to make sense for the type of "CSTPTR".
2439 ``select (COND, VAL1, VAL2)``
2440 Perform the :ref:`select operation <i_select>` on constants.
2441 ``icmp COND (VAL1, VAL2)``
2442 Performs the :ref:`icmp operation <i_icmp>` on constants.
2443 ``fcmp COND (VAL1, VAL2)``
2444 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2445 ``extractelement (VAL, IDX)``
2446 Perform the :ref:`extractelement operation <i_extractelement>` on
2448 ``insertelement (VAL, ELT, IDX)``
2449 Perform the :ref:`insertelement operation <i_insertelement>` on
2451 ``shufflevector (VEC1, VEC2, IDXMASK)``
2452 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2454 ``extractvalue (VAL, IDX0, IDX1, ...)``
2455 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2456 constants. The index list is interpreted in a similar manner as
2457 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2458 least one index value must be specified.
2459 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2460 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2461 The index list is interpreted in a similar manner as indices in a
2462 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2463 value must be specified.
2464 ``OPCODE (LHS, RHS)``
2465 Perform the specified operation of the LHS and RHS constants. OPCODE
2466 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2467 binary <bitwiseops>` operations. The constraints on operands are
2468 the same as those for the corresponding instruction (e.g. no bitwise
2469 operations on floating point values are allowed).
2476 Inline Assembler Expressions
2477 ----------------------------
2479 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2480 Inline Assembly <moduleasm>`) through the use of a special value. This
2481 value represents the inline assembler as a string (containing the
2482 instructions to emit), a list of operand constraints (stored as a
2483 string), a flag that indicates whether or not the inline asm expression
2484 has side effects, and a flag indicating whether the function containing
2485 the asm needs to align its stack conservatively. An example inline
2486 assembler expression is:
2488 .. code-block:: llvm
2490 i32 (i32) asm "bswap $0", "=r,r"
2492 Inline assembler expressions may **only** be used as the callee operand
2493 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2494 Thus, typically we have:
2496 .. code-block:: llvm
2498 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2500 Inline asms with side effects not visible in the constraint list must be
2501 marked as having side effects. This is done through the use of the
2502 '``sideeffect``' keyword, like so:
2504 .. code-block:: llvm
2506 call void asm sideeffect "eieio", ""()
2508 In some cases inline asms will contain code that will not work unless
2509 the stack is aligned in some way, such as calls or SSE instructions on
2510 x86, yet will not contain code that does that alignment within the asm.
2511 The compiler should make conservative assumptions about what the asm
2512 might contain and should generate its usual stack alignment code in the
2513 prologue if the '``alignstack``' keyword is present:
2515 .. code-block:: llvm
2517 call void asm alignstack "eieio", ""()
2519 Inline asms also support using non-standard assembly dialects. The
2520 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2521 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2522 the only supported dialects. An example is:
2524 .. code-block:: llvm
2526 call void asm inteldialect "eieio", ""()
2528 If multiple keywords appear the '``sideeffect``' keyword must come
2529 first, the '``alignstack``' keyword second and the '``inteldialect``'
2535 The call instructions that wrap inline asm nodes may have a
2536 "``!srcloc``" MDNode attached to it that contains a list of constant
2537 integers. If present, the code generator will use the integer as the
2538 location cookie value when report errors through the ``LLVMContext``
2539 error reporting mechanisms. This allows a front-end to correlate backend
2540 errors that occur with inline asm back to the source code that produced
2543 .. code-block:: llvm
2545 call void asm sideeffect "something bad", ""(), !srcloc !42
2547 !42 = !{ i32 1234567 }
2549 It is up to the front-end to make sense of the magic numbers it places
2550 in the IR. If the MDNode contains multiple constants, the code generator
2551 will use the one that corresponds to the line of the asm that the error
2556 Metadata Nodes and Metadata Strings
2557 -----------------------------------
2559 LLVM IR allows metadata to be attached to instructions in the program
2560 that can convey extra information about the code to the optimizers and
2561 code generator. One example application of metadata is source-level
2562 debug information. There are two metadata primitives: strings and nodes.
2563 All metadata has the ``metadata`` type and is identified in syntax by a
2564 preceding exclamation point ('``!``').
2566 A metadata string is a string surrounded by double quotes. It can
2567 contain any character by escaping non-printable characters with
2568 "``\xx``" where "``xx``" is the two digit hex code. For example:
2571 Metadata nodes are represented with notation similar to structure
2572 constants (a comma separated list of elements, surrounded by braces and
2573 preceded by an exclamation point). Metadata nodes can have any values as
2574 their operand. For example:
2576 .. code-block:: llvm
2578 !{ metadata !"test\00", i32 10}
2580 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2581 metadata nodes, which can be looked up in the module symbol table. For
2584 .. code-block:: llvm
2586 !foo = metadata !{!4, !3}
2588 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2589 function is using two metadata arguments:
2591 .. code-block:: llvm
2593 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2595 Metadata can be attached with an instruction. Here metadata ``!21`` is
2596 attached to the ``add`` instruction using the ``!dbg`` identifier:
2598 .. code-block:: llvm
2600 %indvar.next = add i64 %indvar, 1, !dbg !21
2602 More information about specific metadata nodes recognized by the
2603 optimizers and code generator is found below.
2608 In LLVM IR, memory does not have types, so LLVM's own type system is not
2609 suitable for doing TBAA. Instead, metadata is added to the IR to
2610 describe a type system of a higher level language. This can be used to
2611 implement typical C/C++ TBAA, but it can also be used to implement
2612 custom alias analysis behavior for other languages.
2614 The current metadata format is very simple. TBAA metadata nodes have up
2615 to three fields, e.g.:
2617 .. code-block:: llvm
2619 !0 = metadata !{ metadata !"an example type tree" }
2620 !1 = metadata !{ metadata !"int", metadata !0 }
2621 !2 = metadata !{ metadata !"float", metadata !0 }
2622 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2624 The first field is an identity field. It can be any value, usually a
2625 metadata string, which uniquely identifies the type. The most important
2626 name in the tree is the name of the root node. Two trees with different
2627 root node names are entirely disjoint, even if they have leaves with
2630 The second field identifies the type's parent node in the tree, or is
2631 null or omitted for a root node. A type is considered to alias all of
2632 its descendants and all of its ancestors in the tree. Also, a type is
2633 considered to alias all types in other trees, so that bitcode produced
2634 from multiple front-ends is handled conservatively.
2636 If the third field is present, it's an integer which if equal to 1
2637 indicates that the type is "constant" (meaning
2638 ``pointsToConstantMemory`` should return true; see `other useful
2639 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2641 '``tbaa.struct``' Metadata
2642 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2644 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2645 aggregate assignment operations in C and similar languages, however it
2646 is defined to copy a contiguous region of memory, which is more than
2647 strictly necessary for aggregate types which contain holes due to
2648 padding. Also, it doesn't contain any TBAA information about the fields
2651 ``!tbaa.struct`` metadata can describe which memory subregions in a
2652 memcpy are padding and what the TBAA tags of the struct are.
2654 The current metadata format is very simple. ``!tbaa.struct`` metadata
2655 nodes are a list of operands which are in conceptual groups of three.
2656 For each group of three, the first operand gives the byte offset of a
2657 field in bytes, the second gives its size in bytes, and the third gives
2660 .. code-block:: llvm
2662 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2664 This describes a struct with two fields. The first is at offset 0 bytes
2665 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2666 and has size 4 bytes and has tbaa tag !2.
2668 Note that the fields need not be contiguous. In this example, there is a
2669 4 byte gap between the two fields. This gap represents padding which
2670 does not carry useful data and need not be preserved.
2672 '``fpmath``' Metadata
2673 ^^^^^^^^^^^^^^^^^^^^^
2675 ``fpmath`` metadata may be attached to any instruction of floating point
2676 type. It can be used to express the maximum acceptable error in the
2677 result of that instruction, in ULPs, thus potentially allowing the
2678 compiler to use a more efficient but less accurate method of computing
2679 it. ULP is defined as follows:
2681 If ``x`` is a real number that lies between two finite consecutive
2682 floating-point numbers ``a`` and ``b``, without being equal to one
2683 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2684 distance between the two non-equal finite floating-point numbers
2685 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2687 The metadata node shall consist of a single positive floating point
2688 number representing the maximum relative error, for example:
2690 .. code-block:: llvm
2692 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2694 '``range``' Metadata
2695 ^^^^^^^^^^^^^^^^^^^^
2697 ``range`` metadata may be attached only to loads of integer types. It
2698 expresses the possible ranges the loaded value is in. The ranges are
2699 represented with a flattened list of integers. The loaded value is known
2700 to be in the union of the ranges defined by each consecutive pair. Each
2701 pair has the following properties:
2703 - The type must match the type loaded by the instruction.
2704 - The pair ``a,b`` represents the range ``[a,b)``.
2705 - Both ``a`` and ``b`` are constants.
2706 - The range is allowed to wrap.
2707 - The range should not represent the full or empty set. That is,
2710 In addition, the pairs must be in signed order of the lower bound and
2711 they must be non-contiguous.
2715 .. code-block:: llvm
2717 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2718 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2719 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2720 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2722 !0 = metadata !{ i8 0, i8 2 }
2723 !1 = metadata !{ i8 255, i8 2 }
2724 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2725 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2730 It is sometimes useful to attach information to loop constructs. Currently,
2731 loop metadata is implemented as metadata attached to the branch instruction
2732 in the loop latch block. This type of metadata refer to a metadata node that is
2733 guaranteed to be separate for each loop. The loop identifier metadata is
2734 specified with the name ``llvm.loop``.
2736 The loop identifier metadata is implemented using a metadata that refers to
2737 itself to avoid merging it with any other identifier metadata, e.g.,
2738 during module linkage or function inlining. That is, each loop should refer
2739 to their own identification metadata even if they reside in separate functions.
2740 The following example contains loop identifier metadata for two separate loop
2743 .. code-block:: llvm
2745 !0 = metadata !{ metadata !0 }
2746 !1 = metadata !{ metadata !1 }
2748 The loop identifier metadata can be used to specify additional per-loop
2749 metadata. Any operands after the first operand can be treated as user-defined
2750 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2751 by the loop vectorizer to indicate how many times to unroll the loop:
2753 .. code-block:: llvm
2755 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2757 !0 = metadata !{ metadata !0, metadata !1 }
2758 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2763 Metadata types used to annotate memory accesses with information helpful
2764 for optimizations are prefixed with ``llvm.mem``.
2766 '``llvm.mem.parallel_loop_access``' Metadata
2767 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2769 For a loop to be parallel, in addition to using
2770 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2771 also all of the memory accessing instructions in the loop body need to be
2772 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2773 is at least one memory accessing instruction not marked with the metadata,
2774 the loop must be considered a sequential loop. This causes parallel loops to be
2775 converted to sequential loops due to optimization passes that are unaware of
2776 the parallel semantics and that insert new memory instructions to the loop
2779 Example of a loop that is considered parallel due to its correct use of
2780 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2781 metadata types that refer to the same loop identifier metadata.
2783 .. code-block:: llvm
2787 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
2789 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2791 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2795 !0 = metadata !{ metadata !0 }
2797 It is also possible to have nested parallel loops. In that case the
2798 memory accesses refer to a list of loop identifier metadata nodes instead of
2799 the loop identifier metadata node directly:
2801 .. code-block:: llvm
2805 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
2807 br label %inner.for.body
2811 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2813 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
2815 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2819 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
2821 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2823 outer.for.end: ; preds = %for.body
2825 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2826 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2827 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2829 '``llvm.vectorizer``'
2830 ^^^^^^^^^^^^^^^^^^^^^
2832 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2833 vectorization parameters such as vectorization factor and unroll factor.
2835 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2836 loop identification metadata.
2838 '``llvm.vectorizer.unroll``' Metadata
2839 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2841 This metadata instructs the loop vectorizer to unroll the specified
2842 loop exactly ``N`` times.
2844 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2845 operand is an integer specifying the unroll factor. For example:
2847 .. code-block:: llvm
2849 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2851 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2854 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2855 determined automatically.
2857 '``llvm.vectorizer.width``' Metadata
2858 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2860 This metadata sets the target width of the vectorizer to ``N``. Without
2861 this metadata, the vectorizer will choose a width automatically.
2862 Regardless of this metadata, the vectorizer will only vectorize loops if
2863 it believes it is valid to do so.
2865 The first operand is the string ``llvm.vectorizer.width`` and the second
2866 operand is an integer specifying the width. For example:
2868 .. code-block:: llvm
2870 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2872 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2875 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2878 Module Flags Metadata
2879 =====================
2881 Information about the module as a whole is difficult to convey to LLVM's
2882 subsystems. The LLVM IR isn't sufficient to transmit this information.
2883 The ``llvm.module.flags`` named metadata exists in order to facilitate
2884 this. These flags are in the form of key / value pairs --- much like a
2885 dictionary --- making it easy for any subsystem who cares about a flag to
2888 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2889 Each triplet has the following form:
2891 - The first element is a *behavior* flag, which specifies the behavior
2892 when two (or more) modules are merged together, and it encounters two
2893 (or more) metadata with the same ID. The supported behaviors are
2895 - The second element is a metadata string that is a unique ID for the
2896 metadata. Each module may only have one flag entry for each unique ID (not
2897 including entries with the **Require** behavior).
2898 - The third element is the value of the flag.
2900 When two (or more) modules are merged together, the resulting
2901 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2902 each unique metadata ID string, there will be exactly one entry in the merged
2903 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2904 be determined by the merge behavior flag, as described below. The only exception
2905 is that entries with the *Require* behavior are always preserved.
2907 The following behaviors are supported:
2918 Emits an error if two values disagree, otherwise the resulting value
2919 is that of the operands.
2923 Emits a warning if two values disagree. The result value will be the
2924 operand for the flag from the first module being linked.
2928 Adds a requirement that another module flag be present and have a
2929 specified value after linking is performed. The value must be a
2930 metadata pair, where the first element of the pair is the ID of the
2931 module flag to be restricted, and the second element of the pair is
2932 the value the module flag should be restricted to. This behavior can
2933 be used to restrict the allowable results (via triggering of an
2934 error) of linking IDs with the **Override** behavior.
2938 Uses the specified value, regardless of the behavior or value of the
2939 other module. If both modules specify **Override**, but the values
2940 differ, an error will be emitted.
2944 Appends the two values, which are required to be metadata nodes.
2948 Appends the two values, which are required to be metadata
2949 nodes. However, duplicate entries in the second list are dropped
2950 during the append operation.
2952 It is an error for a particular unique flag ID to have multiple behaviors,
2953 except in the case of **Require** (which adds restrictions on another metadata
2954 value) or **Override**.
2956 An example of module flags:
2958 .. code-block:: llvm
2960 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2961 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2962 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2963 !3 = metadata !{ i32 3, metadata !"qux",
2965 metadata !"foo", i32 1
2968 !llvm.module.flags = !{ !0, !1, !2, !3 }
2970 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2971 if two or more ``!"foo"`` flags are seen is to emit an error if their
2972 values are not equal.
2974 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2975 behavior if two or more ``!"bar"`` flags are seen is to use the value
2978 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2979 behavior if two or more ``!"qux"`` flags are seen is to emit a
2980 warning if their values are not equal.
2982 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2986 metadata !{ metadata !"foo", i32 1 }
2988 The behavior is to emit an error if the ``llvm.module.flags`` does not
2989 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2992 Objective-C Garbage Collection Module Flags Metadata
2993 ----------------------------------------------------
2995 On the Mach-O platform, Objective-C stores metadata about garbage
2996 collection in a special section called "image info". The metadata
2997 consists of a version number and a bitmask specifying what types of
2998 garbage collection are supported (if any) by the file. If two or more
2999 modules are linked together their garbage collection metadata needs to
3000 be merged rather than appended together.
3002 The Objective-C garbage collection module flags metadata consists of the
3003 following key-value pairs:
3012 * - ``Objective-C Version``
3013 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3015 * - ``Objective-C Image Info Version``
3016 - **[Required]** --- The version of the image info section. Currently
3019 * - ``Objective-C Image Info Section``
3020 - **[Required]** --- The section to place the metadata. Valid values are
3021 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3022 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3023 Objective-C ABI version 2.
3025 * - ``Objective-C Garbage Collection``
3026 - **[Required]** --- Specifies whether garbage collection is supported or
3027 not. Valid values are 0, for no garbage collection, and 2, for garbage
3028 collection supported.
3030 * - ``Objective-C GC Only``
3031 - **[Optional]** --- Specifies that only garbage collection is supported.
3032 If present, its value must be 6. This flag requires that the
3033 ``Objective-C Garbage Collection`` flag have the value 2.
3035 Some important flag interactions:
3037 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3038 merged with a module with ``Objective-C Garbage Collection`` set to
3039 2, then the resulting module has the
3040 ``Objective-C Garbage Collection`` flag set to 0.
3041 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3042 merged with a module with ``Objective-C GC Only`` set to 6.
3044 Automatic Linker Flags Module Flags Metadata
3045 --------------------------------------------
3047 Some targets support embedding flags to the linker inside individual object
3048 files. Typically this is used in conjunction with language extensions which
3049 allow source files to explicitly declare the libraries they depend on, and have
3050 these automatically be transmitted to the linker via object files.
3052 These flags are encoded in the IR using metadata in the module flags section,
3053 using the ``Linker Options`` key. The merge behavior for this flag is required
3054 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3055 node which should be a list of other metadata nodes, each of which should be a
3056 list of metadata strings defining linker options.
3058 For example, the following metadata section specifies two separate sets of
3059 linker options, presumably to link against ``libz`` and the ``Cocoa``
3062 !0 = metadata !{ i32 6, metadata !"Linker Options",
3064 metadata !{ metadata !"-lz" },
3065 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3066 !llvm.module.flags = !{ !0 }
3068 The metadata encoding as lists of lists of options, as opposed to a collapsed
3069 list of options, is chosen so that the IR encoding can use multiple option
3070 strings to specify e.g., a single library, while still having that specifier be
3071 preserved as an atomic element that can be recognized by a target specific
3072 assembly writer or object file emitter.
3074 Each individual option is required to be either a valid option for the target's
3075 linker, or an option that is reserved by the target specific assembly writer or
3076 object file emitter. No other aspect of these options is defined by the IR.
3078 .. _intrinsicglobalvariables:
3080 Intrinsic Global Variables
3081 ==========================
3083 LLVM has a number of "magic" global variables that contain data that
3084 affect code generation or other IR semantics. These are documented here.
3085 All globals of this sort should have a section specified as
3086 "``llvm.metadata``". This section and all globals that start with
3087 "``llvm.``" are reserved for use by LLVM.
3091 The '``llvm.used``' Global Variable
3092 -----------------------------------
3094 The ``@llvm.used`` global is an array which has
3095 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3096 pointers to named global variables, functions and aliases which may optionally
3097 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3100 .. code-block:: llvm
3105 @llvm.used = appending global [2 x i8*] [
3107 i8* bitcast (i32* @Y to i8*)
3108 ], section "llvm.metadata"
3110 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3111 and linker are required to treat the symbol as if there is a reference to the
3112 symbol that it cannot see (which is why they have to be named). For example, if
3113 a variable has internal linkage and no references other than that from the
3114 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3115 references from inline asms and other things the compiler cannot "see", and
3116 corresponds to "``attribute((used))``" in GNU C.
3118 On some targets, the code generator must emit a directive to the
3119 assembler or object file to prevent the assembler and linker from
3120 molesting the symbol.
3122 .. _gv_llvmcompilerused:
3124 The '``llvm.compiler.used``' Global Variable
3125 --------------------------------------------
3127 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3128 directive, except that it only prevents the compiler from touching the
3129 symbol. On targets that support it, this allows an intelligent linker to
3130 optimize references to the symbol without being impeded as it would be
3133 This is a rare construct that should only be used in rare circumstances,
3134 and should not be exposed to source languages.
3136 .. _gv_llvmglobalctors:
3138 The '``llvm.global_ctors``' Global Variable
3139 -------------------------------------------
3141 .. code-block:: llvm
3143 %0 = type { i32, void ()* }
3144 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3146 The ``@llvm.global_ctors`` array contains a list of constructor
3147 functions and associated priorities. The functions referenced by this
3148 array will be called in ascending order of priority (i.e. lowest first)
3149 when the module is loaded. The order of functions with the same priority
3152 .. _llvmglobaldtors:
3154 The '``llvm.global_dtors``' Global Variable
3155 -------------------------------------------
3157 .. code-block:: llvm
3159 %0 = type { i32, void ()* }
3160 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3162 The ``@llvm.global_dtors`` array contains a list of destructor functions
3163 and associated priorities. The functions referenced by this array will
3164 be called in descending order of priority (i.e. highest first) when the
3165 module is loaded. The order of functions with the same priority is not
3168 Instruction Reference
3169 =====================
3171 The LLVM instruction set consists of several different classifications
3172 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3173 instructions <binaryops>`, :ref:`bitwise binary
3174 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3175 :ref:`other instructions <otherops>`.
3179 Terminator Instructions
3180 -----------------------
3182 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3183 program ends with a "Terminator" instruction, which indicates which
3184 block should be executed after the current block is finished. These
3185 terminator instructions typically yield a '``void``' value: they produce
3186 control flow, not values (the one exception being the
3187 ':ref:`invoke <i_invoke>`' instruction).
3189 The terminator instructions are: ':ref:`ret <i_ret>`',
3190 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3191 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3192 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3196 '``ret``' Instruction
3197 ^^^^^^^^^^^^^^^^^^^^^
3204 ret <type> <value> ; Return a value from a non-void function
3205 ret void ; Return from void function
3210 The '``ret``' instruction is used to return control flow (and optionally
3211 a value) from a function back to the caller.
3213 There are two forms of the '``ret``' instruction: one that returns a
3214 value and then causes control flow, and one that just causes control
3220 The '``ret``' instruction optionally accepts a single argument, the
3221 return value. The type of the return value must be a ':ref:`first
3222 class <t_firstclass>`' type.
3224 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3225 return type and contains a '``ret``' instruction with no return value or
3226 a return value with a type that does not match its type, or if it has a
3227 void return type and contains a '``ret``' instruction with a return
3233 When the '``ret``' instruction is executed, control flow returns back to
3234 the calling function's context. If the caller is a
3235 ":ref:`call <i_call>`" instruction, execution continues at the
3236 instruction after the call. If the caller was an
3237 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3238 beginning of the "normal" destination block. If the instruction returns
3239 a value, that value shall set the call or invoke instruction's return
3245 .. code-block:: llvm
3247 ret i32 5 ; Return an integer value of 5
3248 ret void ; Return from a void function
3249 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3253 '``br``' Instruction
3254 ^^^^^^^^^^^^^^^^^^^^
3261 br i1 <cond>, label <iftrue>, label <iffalse>
3262 br label <dest> ; Unconditional branch
3267 The '``br``' instruction is used to cause control flow to transfer to a
3268 different basic block in the current function. There are two forms of
3269 this instruction, corresponding to a conditional branch and an
3270 unconditional branch.
3275 The conditional branch form of the '``br``' instruction takes a single
3276 '``i1``' value and two '``label``' values. The unconditional form of the
3277 '``br``' instruction takes a single '``label``' value as a target.
3282 Upon execution of a conditional '``br``' instruction, the '``i1``'
3283 argument is evaluated. If the value is ``true``, control flows to the
3284 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3285 to the '``iffalse``' ``label`` argument.
3290 .. code-block:: llvm
3293 %cond = icmp eq i32 %a, %b
3294 br i1 %cond, label %IfEqual, label %IfUnequal
3302 '``switch``' Instruction
3303 ^^^^^^^^^^^^^^^^^^^^^^^^
3310 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3315 The '``switch``' instruction is used to transfer control flow to one of
3316 several different places. It is a generalization of the '``br``'
3317 instruction, allowing a branch to occur to one of many possible
3323 The '``switch``' instruction uses three parameters: an integer
3324 comparison value '``value``', a default '``label``' destination, and an
3325 array of pairs of comparison value constants and '``label``'s. The table
3326 is not allowed to contain duplicate constant entries.
3331 The ``switch`` instruction specifies a table of values and destinations.
3332 When the '``switch``' instruction is executed, this table is searched
3333 for the given value. If the value is found, control flow is transferred
3334 to the corresponding destination; otherwise, control flow is transferred
3335 to the default destination.
3340 Depending on properties of the target machine and the particular
3341 ``switch`` instruction, this instruction may be code generated in
3342 different ways. For example, it could be generated as a series of
3343 chained conditional branches or with a lookup table.
3348 .. code-block:: llvm
3350 ; Emulate a conditional br instruction
3351 %Val = zext i1 %value to i32
3352 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3354 ; Emulate an unconditional br instruction
3355 switch i32 0, label %dest [ ]
3357 ; Implement a jump table:
3358 switch i32 %val, label %otherwise [ i32 0, label %onzero
3360 i32 2, label %ontwo ]
3364 '``indirectbr``' Instruction
3365 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3372 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3377 The '``indirectbr``' instruction implements an indirect branch to a
3378 label within the current function, whose address is specified by
3379 "``address``". Address must be derived from a
3380 :ref:`blockaddress <blockaddress>` constant.
3385 The '``address``' argument is the address of the label to jump to. The
3386 rest of the arguments indicate the full set of possible destinations
3387 that the address may point to. Blocks are allowed to occur multiple
3388 times in the destination list, though this isn't particularly useful.
3390 This destination list is required so that dataflow analysis has an
3391 accurate understanding of the CFG.
3396 Control transfers to the block specified in the address argument. All
3397 possible destination blocks must be listed in the label list, otherwise
3398 this instruction has undefined behavior. This implies that jumps to
3399 labels defined in other functions have undefined behavior as well.
3404 This is typically implemented with a jump through a register.
3409 .. code-block:: llvm
3411 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3415 '``invoke``' Instruction
3416 ^^^^^^^^^^^^^^^^^^^^^^^^
3423 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3424 to label <normal label> unwind label <exception label>
3429 The '``invoke``' instruction causes control to transfer to a specified
3430 function, with the possibility of control flow transfer to either the
3431 '``normal``' label or the '``exception``' label. If the callee function
3432 returns with the "``ret``" instruction, control flow will return to the
3433 "normal" label. If the callee (or any indirect callees) returns via the
3434 ":ref:`resume <i_resume>`" instruction or other exception handling
3435 mechanism, control is interrupted and continued at the dynamically
3436 nearest "exception" label.
3438 The '``exception``' label is a `landing
3439 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3440 '``exception``' label is required to have the
3441 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3442 information about the behavior of the program after unwinding happens,
3443 as its first non-PHI instruction. The restrictions on the
3444 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3445 instruction, so that the important information contained within the
3446 "``landingpad``" instruction can't be lost through normal code motion.
3451 This instruction requires several arguments:
3453 #. The optional "cconv" marker indicates which :ref:`calling
3454 convention <callingconv>` the call should use. If none is
3455 specified, the call defaults to using C calling conventions.
3456 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3457 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3459 #. '``ptr to function ty``': shall be the signature of the pointer to
3460 function value being invoked. In most cases, this is a direct
3461 function invocation, but indirect ``invoke``'s are just as possible,
3462 branching off an arbitrary pointer to function value.
3463 #. '``function ptr val``': An LLVM value containing a pointer to a
3464 function to be invoked.
3465 #. '``function args``': argument list whose types match the function
3466 signature argument types and parameter attributes. All arguments must
3467 be of :ref:`first class <t_firstclass>` type. If the function signature
3468 indicates the function accepts a variable number of arguments, the
3469 extra arguments can be specified.
3470 #. '``normal label``': the label reached when the called function
3471 executes a '``ret``' instruction.
3472 #. '``exception label``': the label reached when a callee returns via
3473 the :ref:`resume <i_resume>` instruction or other exception handling
3475 #. The optional :ref:`function attributes <fnattrs>` list. Only
3476 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3477 attributes are valid here.
3482 This instruction is designed to operate as a standard '``call``'
3483 instruction in most regards. The primary difference is that it
3484 establishes an association with a label, which is used by the runtime
3485 library to unwind the stack.
3487 This instruction is used in languages with destructors to ensure that
3488 proper cleanup is performed in the case of either a ``longjmp`` or a
3489 thrown exception. Additionally, this is important for implementation of
3490 '``catch``' clauses in high-level languages that support them.
3492 For the purposes of the SSA form, the definition of the value returned
3493 by the '``invoke``' instruction is deemed to occur on the edge from the
3494 current block to the "normal" label. If the callee unwinds then no
3495 return value is available.
3500 .. code-block:: llvm
3502 %retval = invoke i32 @Test(i32 15) to label %Continue
3503 unwind label %TestCleanup ; {i32}:retval set
3504 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3505 unwind label %TestCleanup ; {i32}:retval set
3509 '``resume``' Instruction
3510 ^^^^^^^^^^^^^^^^^^^^^^^^
3517 resume <type> <value>
3522 The '``resume``' instruction is a terminator instruction that has no
3528 The '``resume``' instruction requires one argument, which must have the
3529 same type as the result of any '``landingpad``' instruction in the same
3535 The '``resume``' instruction resumes propagation of an existing
3536 (in-flight) exception whose unwinding was interrupted with a
3537 :ref:`landingpad <i_landingpad>` instruction.
3542 .. code-block:: llvm
3544 resume { i8*, i32 } %exn
3548 '``unreachable``' Instruction
3549 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3561 The '``unreachable``' instruction has no defined semantics. This
3562 instruction is used to inform the optimizer that a particular portion of
3563 the code is not reachable. This can be used to indicate that the code
3564 after a no-return function cannot be reached, and other facts.
3569 The '``unreachable``' instruction has no defined semantics.
3576 Binary operators are used to do most of the computation in a program.
3577 They require two operands of the same type, execute an operation on
3578 them, and produce a single value. The operands might represent multiple
3579 data, as is the case with the :ref:`vector <t_vector>` data type. The
3580 result value has the same type as its operands.
3582 There are several different binary operators:
3586 '``add``' Instruction
3587 ^^^^^^^^^^^^^^^^^^^^^
3594 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3595 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3596 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3597 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3602 The '``add``' instruction returns the sum of its two operands.
3607 The two arguments to the '``add``' instruction must be
3608 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3609 arguments must have identical types.
3614 The value produced is the integer sum of the two operands.
3616 If the sum has unsigned overflow, the result returned is the
3617 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3620 Because LLVM integers use a two's complement representation, this
3621 instruction is appropriate for both signed and unsigned integers.
3623 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3624 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3625 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3626 unsigned and/or signed overflow, respectively, occurs.
3631 .. code-block:: llvm
3633 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3637 '``fadd``' Instruction
3638 ^^^^^^^^^^^^^^^^^^^^^^
3645 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3650 The '``fadd``' instruction returns the sum of its two operands.
3655 The two arguments to the '``fadd``' instruction must be :ref:`floating
3656 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3657 Both arguments must have identical types.
3662 The value produced is the floating point sum of the two operands. This
3663 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3664 which are optimization hints to enable otherwise unsafe floating point
3670 .. code-block:: llvm
3672 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3674 '``sub``' Instruction
3675 ^^^^^^^^^^^^^^^^^^^^^
3682 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3683 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3684 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3685 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3690 The '``sub``' instruction returns the difference of its two operands.
3692 Note that the '``sub``' instruction is used to represent the '``neg``'
3693 instruction present in most other intermediate representations.
3698 The two arguments to the '``sub``' instruction must be
3699 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3700 arguments must have identical types.
3705 The value produced is the integer difference of the two operands.
3707 If the difference has unsigned overflow, the result returned is the
3708 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3711 Because LLVM integers use a two's complement representation, this
3712 instruction is appropriate for both signed and unsigned integers.
3714 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3715 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3716 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3717 unsigned and/or signed overflow, respectively, occurs.
3722 .. code-block:: llvm
3724 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3725 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3729 '``fsub``' Instruction
3730 ^^^^^^^^^^^^^^^^^^^^^^
3737 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3742 The '``fsub``' instruction returns the difference of its two operands.
3744 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3745 instruction present in most other intermediate representations.
3750 The two arguments to the '``fsub``' instruction must be :ref:`floating
3751 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3752 Both arguments must have identical types.
3757 The value produced is the floating point difference of the two operands.
3758 This instruction can also take any number of :ref:`fast-math
3759 flags <fastmath>`, which are optimization hints to enable otherwise
3760 unsafe floating point optimizations:
3765 .. code-block:: llvm
3767 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3768 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3770 '``mul``' Instruction
3771 ^^^^^^^^^^^^^^^^^^^^^
3778 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3779 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3780 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3781 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3786 The '``mul``' instruction returns the product of its two operands.
3791 The two arguments to the '``mul``' instruction must be
3792 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3793 arguments must have identical types.
3798 The value produced is the integer product of the two operands.
3800 If the result of the multiplication has unsigned overflow, the result
3801 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3802 bit width of the result.
3804 Because LLVM integers use a two's complement representation, and the
3805 result is the same width as the operands, this instruction returns the
3806 correct result for both signed and unsigned integers. If a full product
3807 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3808 sign-extended or zero-extended as appropriate to the width of the full
3811 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3812 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3813 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3814 unsigned and/or signed overflow, respectively, occurs.
3819 .. code-block:: llvm
3821 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3825 '``fmul``' Instruction
3826 ^^^^^^^^^^^^^^^^^^^^^^
3833 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3838 The '``fmul``' instruction returns the product of its two operands.
3843 The two arguments to the '``fmul``' instruction must be :ref:`floating
3844 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3845 Both arguments must have identical types.
3850 The value produced is the floating point product of the two operands.
3851 This instruction can also take any number of :ref:`fast-math
3852 flags <fastmath>`, which are optimization hints to enable otherwise
3853 unsafe floating point optimizations:
3858 .. code-block:: llvm
3860 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3862 '``udiv``' Instruction
3863 ^^^^^^^^^^^^^^^^^^^^^^
3870 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3871 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3876 The '``udiv``' instruction returns the quotient of its two operands.
3881 The two arguments to the '``udiv``' instruction must be
3882 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3883 arguments must have identical types.
3888 The value produced is the unsigned integer quotient of the two operands.
3890 Note that unsigned integer division and signed integer division are
3891 distinct operations; for signed integer division, use '``sdiv``'.
3893 Division by zero leads to undefined behavior.
3895 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3896 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3897 such, "((a udiv exact b) mul b) == a").
3902 .. code-block:: llvm
3904 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3906 '``sdiv``' Instruction
3907 ^^^^^^^^^^^^^^^^^^^^^^
3914 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3915 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3920 The '``sdiv``' instruction returns the quotient of its two operands.
3925 The two arguments to the '``sdiv``' instruction must be
3926 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3927 arguments must have identical types.
3932 The value produced is the signed integer quotient of the two operands
3933 rounded towards zero.
3935 Note that signed integer division and unsigned integer division are
3936 distinct operations; for unsigned integer division, use '``udiv``'.
3938 Division by zero leads to undefined behavior. Overflow also leads to
3939 undefined behavior; this is a rare case, but can occur, for example, by
3940 doing a 32-bit division of -2147483648 by -1.
3942 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3943 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3948 .. code-block:: llvm
3950 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3954 '``fdiv``' Instruction
3955 ^^^^^^^^^^^^^^^^^^^^^^
3962 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3967 The '``fdiv``' instruction returns the quotient of its two operands.
3972 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3973 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3974 Both arguments must have identical types.
3979 The value produced is the floating point quotient of the two operands.
3980 This instruction can also take any number of :ref:`fast-math
3981 flags <fastmath>`, which are optimization hints to enable otherwise
3982 unsafe floating point optimizations:
3987 .. code-block:: llvm
3989 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3991 '``urem``' Instruction
3992 ^^^^^^^^^^^^^^^^^^^^^^
3999 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
4004 The '``urem``' instruction returns the remainder from the unsigned
4005 division of its two arguments.
4010 The two arguments to the '``urem``' instruction must be
4011 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4012 arguments must have identical types.
4017 This instruction returns the unsigned integer *remainder* of a division.
4018 This instruction always performs an unsigned division to get the
4021 Note that unsigned integer remainder and signed integer remainder are
4022 distinct operations; for signed integer remainder, use '``srem``'.
4024 Taking the remainder of a division by zero leads to undefined behavior.
4029 .. code-block:: llvm
4031 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
4033 '``srem``' Instruction
4034 ^^^^^^^^^^^^^^^^^^^^^^
4041 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
4046 The '``srem``' instruction returns the remainder from the signed
4047 division of its two operands. This instruction can also take
4048 :ref:`vector <t_vector>` versions of the values in which case the elements
4054 The two arguments to the '``srem``' instruction must be
4055 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4056 arguments must have identical types.
4061 This instruction returns the *remainder* of a division (where the result
4062 is either zero or has the same sign as the dividend, ``op1``), not the
4063 *modulo* operator (where the result is either zero or has the same sign
4064 as the divisor, ``op2``) of a value. For more information about the
4065 difference, see `The Math
4066 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4067 table of how this is implemented in various languages, please see
4069 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4071 Note that signed integer remainder and unsigned integer remainder are
4072 distinct operations; for unsigned integer remainder, use '``urem``'.
4074 Taking the remainder of a division by zero leads to undefined behavior.
4075 Overflow also leads to undefined behavior; this is a rare case, but can
4076 occur, for example, by taking the remainder of a 32-bit division of
4077 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4078 rule lets srem be implemented using instructions that return both the
4079 result of the division and the remainder.)
4084 .. code-block:: llvm
4086 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4090 '``frem``' Instruction
4091 ^^^^^^^^^^^^^^^^^^^^^^
4098 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4103 The '``frem``' instruction returns the remainder from the division of
4109 The two arguments to the '``frem``' instruction must be :ref:`floating
4110 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4111 Both arguments must have identical types.
4116 This instruction returns the *remainder* of a division. The remainder
4117 has the same sign as the dividend. This instruction can also take any
4118 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4119 to enable otherwise unsafe floating point optimizations:
4124 .. code-block:: llvm
4126 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4130 Bitwise Binary Operations
4131 -------------------------
4133 Bitwise binary operators are used to do various forms of bit-twiddling
4134 in a program. They are generally very efficient instructions and can
4135 commonly be strength reduced from other instructions. They require two
4136 operands of the same type, execute an operation on them, and produce a
4137 single value. The resulting value is the same type as its operands.
4139 '``shl``' Instruction
4140 ^^^^^^^^^^^^^^^^^^^^^
4147 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4148 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4149 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4150 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4155 The '``shl``' instruction returns the first operand shifted to the left
4156 a specified number of bits.
4161 Both arguments to the '``shl``' instruction must be the same
4162 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4163 '``op2``' is treated as an unsigned value.
4168 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4169 where ``n`` is the width of the result. If ``op2`` is (statically or
4170 dynamically) negative or equal to or larger than the number of bits in
4171 ``op1``, the result is undefined. If the arguments are vectors, each
4172 vector element of ``op1`` is shifted by the corresponding shift amount
4175 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4176 value <poisonvalues>` if it shifts out any non-zero bits. If the
4177 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4178 value <poisonvalues>` if it shifts out any bits that disagree with the
4179 resultant sign bit. As such, NUW/NSW have the same semantics as they
4180 would if the shift were expressed as a mul instruction with the same
4181 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4186 .. code-block:: llvm
4188 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4189 <result> = shl i32 4, 2 ; yields {i32}: 16
4190 <result> = shl i32 1, 10 ; yields {i32}: 1024
4191 <result> = shl i32 1, 32 ; undefined
4192 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4194 '``lshr``' Instruction
4195 ^^^^^^^^^^^^^^^^^^^^^^
4202 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4203 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4208 The '``lshr``' instruction (logical shift right) returns the first
4209 operand shifted to the right a specified number of bits with zero fill.
4214 Both arguments to the '``lshr``' instruction must be the same
4215 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4216 '``op2``' is treated as an unsigned value.
4221 This instruction always performs a logical shift right operation. The
4222 most significant bits of the result will be filled with zero bits after
4223 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4224 than the number of bits in ``op1``, the result is undefined. If the
4225 arguments are vectors, each vector element of ``op1`` is shifted by the
4226 corresponding shift amount in ``op2``.
4228 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4229 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4235 .. code-block:: llvm
4237 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4238 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4239 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4240 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4241 <result> = lshr i32 1, 32 ; undefined
4242 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4244 '``ashr``' Instruction
4245 ^^^^^^^^^^^^^^^^^^^^^^
4252 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4253 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4258 The '``ashr``' instruction (arithmetic shift right) returns the first
4259 operand shifted to the right a specified number of bits with sign
4265 Both arguments to the '``ashr``' instruction must be the same
4266 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4267 '``op2``' is treated as an unsigned value.
4272 This instruction always performs an arithmetic shift right operation,
4273 The most significant bits of the result will be filled with the sign bit
4274 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4275 than the number of bits in ``op1``, the result is undefined. If the
4276 arguments are vectors, each vector element of ``op1`` is shifted by the
4277 corresponding shift amount in ``op2``.
4279 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4280 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4286 .. code-block:: llvm
4288 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4289 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4290 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4291 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4292 <result> = ashr i32 1, 32 ; undefined
4293 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4295 '``and``' Instruction
4296 ^^^^^^^^^^^^^^^^^^^^^
4303 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4308 The '``and``' instruction returns the bitwise logical and of its two
4314 The two arguments to the '``and``' instruction must be
4315 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4316 arguments must have identical types.
4321 The truth table used for the '``and``' instruction is:
4338 .. code-block:: llvm
4340 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4341 <result> = and i32 15, 40 ; yields {i32}:result = 8
4342 <result> = and i32 4, 8 ; yields {i32}:result = 0
4344 '``or``' Instruction
4345 ^^^^^^^^^^^^^^^^^^^^
4352 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4357 The '``or``' instruction returns the bitwise logical inclusive or of its
4363 The two arguments to the '``or``' instruction must be
4364 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4365 arguments must have identical types.
4370 The truth table used for the '``or``' instruction is:
4389 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4390 <result> = or i32 15, 40 ; yields {i32}:result = 47
4391 <result> = or i32 4, 8 ; yields {i32}:result = 12
4393 '``xor``' Instruction
4394 ^^^^^^^^^^^^^^^^^^^^^
4401 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4406 The '``xor``' instruction returns the bitwise logical exclusive or of
4407 its two operands. The ``xor`` is used to implement the "one's
4408 complement" operation, which is the "~" operator in C.
4413 The two arguments to the '``xor``' instruction must be
4414 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4415 arguments must have identical types.
4420 The truth table used for the '``xor``' instruction is:
4437 .. code-block:: llvm
4439 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4440 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4441 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4442 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4447 LLVM supports several instructions to represent vector operations in a
4448 target-independent manner. These instructions cover the element-access
4449 and vector-specific operations needed to process vectors effectively.
4450 While LLVM does directly support these vector operations, many
4451 sophisticated algorithms will want to use target-specific intrinsics to
4452 take full advantage of a specific target.
4454 .. _i_extractelement:
4456 '``extractelement``' Instruction
4457 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4464 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4469 The '``extractelement``' instruction extracts a single scalar element
4470 from a vector at a specified index.
4475 The first operand of an '``extractelement``' instruction is a value of
4476 :ref:`vector <t_vector>` type. The second operand is an index indicating
4477 the position from which to extract the element. The index may be a
4483 The result is a scalar of the same type as the element type of ``val``.
4484 Its value is the value at position ``idx`` of ``val``. If ``idx``
4485 exceeds the length of ``val``, the results are undefined.
4490 .. code-block:: llvm
4492 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4494 .. _i_insertelement:
4496 '``insertelement``' Instruction
4497 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4504 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4509 The '``insertelement``' instruction inserts a scalar element into a
4510 vector at a specified index.
4515 The first operand of an '``insertelement``' instruction is a value of
4516 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4517 type must equal the element type of the first operand. The third operand
4518 is an index indicating the position at which to insert the value. The
4519 index may be a variable.
4524 The result is a vector of the same type as ``val``. Its element values
4525 are those of ``val`` except at position ``idx``, where it gets the value
4526 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4532 .. code-block:: llvm
4534 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4536 .. _i_shufflevector:
4538 '``shufflevector``' Instruction
4539 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4546 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4551 The '``shufflevector``' instruction constructs a permutation of elements
4552 from two input vectors, returning a vector with the same element type as
4553 the input and length that is the same as the shuffle mask.
4558 The first two operands of a '``shufflevector``' instruction are vectors
4559 with the same type. The third argument is a shuffle mask whose element
4560 type is always 'i32'. The result of the instruction is a vector whose
4561 length is the same as the shuffle mask and whose element type is the
4562 same as the element type of the first two operands.
4564 The shuffle mask operand is required to be a constant vector with either
4565 constant integer or undef values.
4570 The elements of the two input vectors are numbered from left to right
4571 across both of the vectors. The shuffle mask operand specifies, for each
4572 element of the result vector, which element of the two input vectors the
4573 result element gets. The element selector may be undef (meaning "don't
4574 care") and the second operand may be undef if performing a shuffle from
4580 .. code-block:: llvm
4582 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4583 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4584 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4585 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4586 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4587 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4588 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4589 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4591 Aggregate Operations
4592 --------------------
4594 LLVM supports several instructions for working with
4595 :ref:`aggregate <t_aggregate>` values.
4599 '``extractvalue``' Instruction
4600 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4607 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4612 The '``extractvalue``' instruction extracts the value of a member field
4613 from an :ref:`aggregate <t_aggregate>` value.
4618 The first operand of an '``extractvalue``' instruction is a value of
4619 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4620 constant indices to specify which value to extract in a similar manner
4621 as indices in a '``getelementptr``' instruction.
4623 The major differences to ``getelementptr`` indexing are:
4625 - Since the value being indexed is not a pointer, the first index is
4626 omitted and assumed to be zero.
4627 - At least one index must be specified.
4628 - Not only struct indices but also array indices must be in bounds.
4633 The result is the value at the position in the aggregate specified by
4639 .. code-block:: llvm
4641 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4645 '``insertvalue``' Instruction
4646 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4653 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4658 The '``insertvalue``' instruction inserts a value into a member field in
4659 an :ref:`aggregate <t_aggregate>` value.
4664 The first operand of an '``insertvalue``' instruction is a value of
4665 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4666 a first-class value to insert. The following operands are constant
4667 indices indicating the position at which to insert the value in a
4668 similar manner as indices in a '``extractvalue``' instruction. The value
4669 to insert must have the same type as the value identified by the
4675 The result is an aggregate of the same type as ``val``. Its value is
4676 that of ``val`` except that the value at the position specified by the
4677 indices is that of ``elt``.
4682 .. code-block:: llvm
4684 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4685 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4686 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4690 Memory Access and Addressing Operations
4691 ---------------------------------------
4693 A key design point of an SSA-based representation is how it represents
4694 memory. In LLVM, no memory locations are in SSA form, which makes things
4695 very simple. This section describes how to read, write, and allocate
4700 '``alloca``' Instruction
4701 ^^^^^^^^^^^^^^^^^^^^^^^^
4708 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields {type*}:result
4713 The '``alloca``' instruction allocates memory on the stack frame of the
4714 currently executing function, to be automatically released when this
4715 function returns to its caller. The object is always allocated in the
4716 generic address space (address space zero).
4721 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4722 bytes of memory on the runtime stack, returning a pointer of the
4723 appropriate type to the program. If "NumElements" is specified, it is
4724 the number of elements allocated, otherwise "NumElements" is defaulted
4725 to be one. If a constant alignment is specified, the value result of the
4726 allocation is guaranteed to be aligned to at least that boundary. If not
4727 specified, or if zero, the target can choose to align the allocation on
4728 any convenient boundary compatible with the type.
4730 '``type``' may be any sized type.
4735 Memory is allocated; a pointer is returned. The operation is undefined
4736 if there is insufficient stack space for the allocation. '``alloca``'d
4737 memory is automatically released when the function returns. The
4738 '``alloca``' instruction is commonly used to represent automatic
4739 variables that must have an address available. When the function returns
4740 (either with the ``ret`` or ``resume`` instructions), the memory is
4741 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4742 The order in which memory is allocated (ie., which way the stack grows)
4748 .. code-block:: llvm
4750 %ptr = alloca i32 ; yields {i32*}:ptr
4751 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4752 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4753 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4757 '``load``' Instruction
4758 ^^^^^^^^^^^^^^^^^^^^^^
4765 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4766 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4767 !<index> = !{ i32 1 }
4772 The '``load``' instruction is used to read from memory.
4777 The argument to the ``load`` instruction specifies the memory address
4778 from which to load. The pointer must point to a :ref:`first
4779 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4780 then the optimizer is not allowed to modify the number or order of
4781 execution of this ``load`` with other :ref:`volatile
4782 operations <volatile>`.
4784 If the ``load`` is marked as ``atomic``, it takes an extra
4785 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4786 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4787 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4788 when they may see multiple atomic stores. The type of the pointee must
4789 be an integer type whose bit width is a power of two greater than or
4790 equal to eight and less than or equal to a target-specific size limit.
4791 ``align`` must be explicitly specified on atomic loads, and the load has
4792 undefined behavior if the alignment is not set to a value which is at
4793 least the size in bytes of the pointee. ``!nontemporal`` does not have
4794 any defined semantics for atomic loads.
4796 The optional constant ``align`` argument specifies the alignment of the
4797 operation (that is, the alignment of the memory address). A value of 0
4798 or an omitted ``align`` argument means that the operation has the ABI
4799 alignment for the target. It is the responsibility of the code emitter
4800 to ensure that the alignment information is correct. Overestimating the
4801 alignment results in undefined behavior. Underestimating the alignment
4802 may produce less efficient code. An alignment of 1 is always safe.
4804 The optional ``!nontemporal`` metadata must reference a single
4805 metadata name ``<index>`` corresponding to a metadata node with one
4806 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4807 metadata on the instruction tells the optimizer and code generator
4808 that this load is not expected to be reused in the cache. The code
4809 generator may select special instructions to save cache bandwidth, such
4810 as the ``MOVNT`` instruction on x86.
4812 The optional ``!invariant.load`` metadata must reference a single
4813 metadata name ``<index>`` corresponding to a metadata node with no
4814 entries. The existence of the ``!invariant.load`` metadata on the
4815 instruction tells the optimizer and code generator that this load
4816 address points to memory which does not change value during program
4817 execution. The optimizer may then move this load around, for example, by
4818 hoisting it out of loops using loop invariant code motion.
4823 The location of memory pointed to is loaded. If the value being loaded
4824 is of scalar type then the number of bytes read does not exceed the
4825 minimum number of bytes needed to hold all bits of the type. For
4826 example, loading an ``i24`` reads at most three bytes. When loading a
4827 value of a type like ``i20`` with a size that is not an integral number
4828 of bytes, the result is undefined if the value was not originally
4829 written using a store of the same type.
4834 .. code-block:: llvm
4836 %ptr = alloca i32 ; yields {i32*}:ptr
4837 store i32 3, i32* %ptr ; yields {void}
4838 %val = load i32* %ptr ; yields {i32}:val = i32 3
4842 '``store``' Instruction
4843 ^^^^^^^^^^^^^^^^^^^^^^^
4850 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4851 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4856 The '``store``' instruction is used to write to memory.
4861 There are two arguments to the ``store`` instruction: a value to store
4862 and an address at which to store it. The type of the ``<pointer>``
4863 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4864 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4865 then the optimizer is not allowed to modify the number or order of
4866 execution of this ``store`` with other :ref:`volatile
4867 operations <volatile>`.
4869 If the ``store`` is marked as ``atomic``, it takes an extra
4870 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4871 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4872 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4873 when they may see multiple atomic stores. The type of the pointee must
4874 be an integer type whose bit width is a power of two greater than or
4875 equal to eight and less than or equal to a target-specific size limit.
4876 ``align`` must be explicitly specified on atomic stores, and the store
4877 has undefined behavior if the alignment is not set to a value which is
4878 at least the size in bytes of the pointee. ``!nontemporal`` does not
4879 have any defined semantics for atomic stores.
4881 The optional constant ``align`` argument specifies the alignment of the
4882 operation (that is, the alignment of the memory address). A value of 0
4883 or an omitted ``align`` argument means that the operation has the ABI
4884 alignment for the target. It is the responsibility of the code emitter
4885 to ensure that the alignment information is correct. Overestimating the
4886 alignment results in undefined behavior. Underestimating the
4887 alignment may produce less efficient code. An alignment of 1 is always
4890 The optional ``!nontemporal`` metadata must reference a single metadata
4891 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4892 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4893 tells the optimizer and code generator that this load is not expected to
4894 be reused in the cache. The code generator may select special
4895 instructions to save cache bandwidth, such as the MOVNT instruction on
4901 The contents of memory are updated to contain ``<value>`` at the
4902 location specified by the ``<pointer>`` operand. If ``<value>`` is
4903 of scalar type then the number of bytes written does not exceed the
4904 minimum number of bytes needed to hold all bits of the type. For
4905 example, storing an ``i24`` writes at most three bytes. When writing a
4906 value of a type like ``i20`` with a size that is not an integral number
4907 of bytes, it is unspecified what happens to the extra bits that do not
4908 belong to the type, but they will typically be overwritten.
4913 .. code-block:: llvm
4915 %ptr = alloca i32 ; yields {i32*}:ptr
4916 store i32 3, i32* %ptr ; yields {void}
4917 %val = load i32* %ptr ; yields {i32}:val = i32 3
4921 '``fence``' Instruction
4922 ^^^^^^^^^^^^^^^^^^^^^^^
4929 fence [singlethread] <ordering> ; yields {void}
4934 The '``fence``' instruction is used to introduce happens-before edges
4940 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4941 defines what *synchronizes-with* edges they add. They can only be given
4942 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4947 A fence A which has (at least) ``release`` ordering semantics
4948 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4949 semantics if and only if there exist atomic operations X and Y, both
4950 operating on some atomic object M, such that A is sequenced before X, X
4951 modifies M (either directly or through some side effect of a sequence
4952 headed by X), Y is sequenced before B, and Y observes M. This provides a
4953 *happens-before* dependency between A and B. Rather than an explicit
4954 ``fence``, one (but not both) of the atomic operations X or Y might
4955 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4956 still *synchronize-with* the explicit ``fence`` and establish the
4957 *happens-before* edge.
4959 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4960 ``acquire`` and ``release`` semantics specified above, participates in
4961 the global program order of other ``seq_cst`` operations and/or fences.
4963 The optional ":ref:`singlethread <singlethread>`" argument specifies
4964 that the fence only synchronizes with other fences in the same thread.
4965 (This is useful for interacting with signal handlers.)
4970 .. code-block:: llvm
4972 fence acquire ; yields {void}
4973 fence singlethread seq_cst ; yields {void}
4977 '``cmpxchg``' Instruction
4978 ^^^^^^^^^^^^^^^^^^^^^^^^^
4985 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields {ty}
4990 The '``cmpxchg``' instruction is used to atomically modify memory. It
4991 loads a value in memory and compares it to a given value. If they are
4992 equal, it stores a new value into the memory.
4997 There are three arguments to the '``cmpxchg``' instruction: an address
4998 to operate on, a value to compare to the value currently be at that
4999 address, and a new value to place at that address if the compared values
5000 are equal. The type of '<cmp>' must be an integer type whose bit width
5001 is a power of two greater than or equal to eight and less than or equal
5002 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5003 type, and the type of '<pointer>' must be a pointer to that type. If the
5004 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5005 to modify the number or order of execution of this ``cmpxchg`` with
5006 other :ref:`volatile operations <volatile>`.
5008 The success and failure :ref:`ordering <ordering>` arguments specify how this
5009 ``cmpxchg`` synchronizes with other atomic operations. The both ordering
5010 parameters must be at least ``monotonic``, the ordering constraint on failure
5011 must be no stronger than that on success, and the failure ordering cannot be
5012 either ``release`` or ``acq_rel``.
5014 The optional "``singlethread``" argument declares that the ``cmpxchg``
5015 is only atomic with respect to code (usually signal handlers) running in
5016 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5017 respect to all other code in the system.
5019 The pointer passed into cmpxchg must have alignment greater than or
5020 equal to the size in memory of the operand.
5025 The contents of memory at the location specified by the '``<pointer>``'
5026 operand is read and compared to '``<cmp>``'; if the read value is the
5027 equal, '``<new>``' is written. The original value at the location is
5030 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5031 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5032 load with an ordering parameter determined the second ordering parameter.
5037 .. code-block:: llvm
5040 %orig = atomic load i32* %ptr unordered ; yields {i32}
5044 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5045 %squared = mul i32 %cmp, %cmp
5046 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields {i32}
5047 %success = icmp eq i32 %cmp, %old
5048 br i1 %success, label %done, label %loop
5055 '``atomicrmw``' Instruction
5056 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5063 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
5068 The '``atomicrmw``' instruction is used to atomically modify memory.
5073 There are three arguments to the '``atomicrmw``' instruction: an
5074 operation to apply, an address whose value to modify, an argument to the
5075 operation. The operation must be one of the following keywords:
5089 The type of '<value>' must be an integer type whose bit width is a power
5090 of two greater than or equal to eight and less than or equal to a
5091 target-specific size limit. The type of the '``<pointer>``' operand must
5092 be a pointer to that type. If the ``atomicrmw`` is marked as
5093 ``volatile``, then the optimizer is not allowed to modify the number or
5094 order of execution of this ``atomicrmw`` with other :ref:`volatile
5095 operations <volatile>`.
5100 The contents of memory at the location specified by the '``<pointer>``'
5101 operand are atomically read, modified, and written back. The original
5102 value at the location is returned. The modification is specified by the
5105 - xchg: ``*ptr = val``
5106 - add: ``*ptr = *ptr + val``
5107 - sub: ``*ptr = *ptr - val``
5108 - and: ``*ptr = *ptr & val``
5109 - nand: ``*ptr = ~(*ptr & val)``
5110 - or: ``*ptr = *ptr | val``
5111 - xor: ``*ptr = *ptr ^ val``
5112 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5113 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5114 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5116 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5122 .. code-block:: llvm
5124 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5126 .. _i_getelementptr:
5128 '``getelementptr``' Instruction
5129 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5136 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5137 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5138 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5143 The '``getelementptr``' instruction is used to get the address of a
5144 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5145 address calculation only and does not access memory.
5150 The first argument is always a pointer or a vector of pointers, and
5151 forms the basis of the calculation. The remaining arguments are indices
5152 that indicate which of the elements of the aggregate object are indexed.
5153 The interpretation of each index is dependent on the type being indexed
5154 into. The first index always indexes the pointer value given as the
5155 first argument, the second index indexes a value of the type pointed to
5156 (not necessarily the value directly pointed to, since the first index
5157 can be non-zero), etc. The first type indexed into must be a pointer
5158 value, subsequent types can be arrays, vectors, and structs. Note that
5159 subsequent types being indexed into can never be pointers, since that
5160 would require loading the pointer before continuing calculation.
5162 The type of each index argument depends on the type it is indexing into.
5163 When indexing into a (optionally packed) structure, only ``i32`` integer
5164 **constants** are allowed (when using a vector of indices they must all
5165 be the **same** ``i32`` integer constant). When indexing into an array,
5166 pointer or vector, integers of any width are allowed, and they are not
5167 required to be constant. These integers are treated as signed values
5170 For example, let's consider a C code fragment and how it gets compiled
5186 int *foo(struct ST *s) {
5187 return &s[1].Z.B[5][13];
5190 The LLVM code generated by Clang is:
5192 .. code-block:: llvm
5194 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5195 %struct.ST = type { i32, double, %struct.RT }
5197 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5199 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5206 In the example above, the first index is indexing into the
5207 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5208 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5209 indexes into the third element of the structure, yielding a
5210 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5211 structure. The third index indexes into the second element of the
5212 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5213 dimensions of the array are subscripted into, yielding an '``i32``'
5214 type. The '``getelementptr``' instruction returns a pointer to this
5215 element, thus computing a value of '``i32*``' type.
5217 Note that it is perfectly legal to index partially through a structure,
5218 returning a pointer to an inner element. Because of this, the LLVM code
5219 for the given testcase is equivalent to:
5221 .. code-block:: llvm
5223 define i32* @foo(%struct.ST* %s) {
5224 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5225 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5226 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5227 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5228 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5232 If the ``inbounds`` keyword is present, the result value of the
5233 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5234 pointer is not an *in bounds* address of an allocated object, or if any
5235 of the addresses that would be formed by successive addition of the
5236 offsets implied by the indices to the base address with infinitely
5237 precise signed arithmetic are not an *in bounds* address of that
5238 allocated object. The *in bounds* addresses for an allocated object are
5239 all the addresses that point into the object, plus the address one byte
5240 past the end. In cases where the base is a vector of pointers the
5241 ``inbounds`` keyword applies to each of the computations element-wise.
5243 If the ``inbounds`` keyword is not present, the offsets are added to the
5244 base address with silently-wrapping two's complement arithmetic. If the
5245 offsets have a different width from the pointer, they are sign-extended
5246 or truncated to the width of the pointer. The result value of the
5247 ``getelementptr`` may be outside the object pointed to by the base
5248 pointer. The result value may not necessarily be used to access memory
5249 though, even if it happens to point into allocated storage. See the
5250 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5253 The getelementptr instruction is often confusing. For some more insight
5254 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5259 .. code-block:: llvm
5261 ; yields [12 x i8]*:aptr
5262 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5264 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5266 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5268 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5270 In cases where the pointer argument is a vector of pointers, each index
5271 must be a vector with the same number of elements. For example:
5273 .. code-block:: llvm
5275 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5277 Conversion Operations
5278 ---------------------
5280 The instructions in this category are the conversion instructions
5281 (casting) which all take a single operand and a type. They perform
5282 various bit conversions on the operand.
5284 '``trunc .. to``' Instruction
5285 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5292 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5297 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5302 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5303 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5304 of the same number of integers. The bit size of the ``value`` must be
5305 larger than the bit size of the destination type, ``ty2``. Equal sized
5306 types are not allowed.
5311 The '``trunc``' instruction truncates the high order bits in ``value``
5312 and converts the remaining bits to ``ty2``. Since the source size must
5313 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5314 It will always truncate bits.
5319 .. code-block:: llvm
5321 %X = trunc i32 257 to i8 ; yields i8:1
5322 %Y = trunc i32 123 to i1 ; yields i1:true
5323 %Z = trunc i32 122 to i1 ; yields i1:false
5324 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5326 '``zext .. to``' Instruction
5327 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5334 <result> = zext <ty> <value> to <ty2> ; yields ty2
5339 The '``zext``' instruction zero extends its operand to type ``ty2``.
5344 The '``zext``' instruction takes a value to cast, and a type to cast it
5345 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5346 the same number of integers. The bit size of the ``value`` must be
5347 smaller than the bit size of the destination type, ``ty2``.
5352 The ``zext`` fills the high order bits of the ``value`` with zero bits
5353 until it reaches the size of the destination type, ``ty2``.
5355 When zero extending from i1, the result will always be either 0 or 1.
5360 .. code-block:: llvm
5362 %X = zext i32 257 to i64 ; yields i64:257
5363 %Y = zext i1 true to i32 ; yields i32:1
5364 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5366 '``sext .. to``' Instruction
5367 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5374 <result> = sext <ty> <value> to <ty2> ; yields ty2
5379 The '``sext``' sign extends ``value`` to the type ``ty2``.
5384 The '``sext``' instruction takes a value to cast, and a type to cast it
5385 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5386 the same number of integers. The bit size of the ``value`` must be
5387 smaller than the bit size of the destination type, ``ty2``.
5392 The '``sext``' instruction performs a sign extension by copying the sign
5393 bit (highest order bit) of the ``value`` until it reaches the bit size
5394 of the type ``ty2``.
5396 When sign extending from i1, the extension always results in -1 or 0.
5401 .. code-block:: llvm
5403 %X = sext i8 -1 to i16 ; yields i16 :65535
5404 %Y = sext i1 true to i32 ; yields i32:-1
5405 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5407 '``fptrunc .. to``' Instruction
5408 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5415 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5420 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5425 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5426 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5427 The size of ``value`` must be larger than the size of ``ty2``. This
5428 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5433 The '``fptrunc``' instruction truncates a ``value`` from a larger
5434 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5435 point <t_floating>` type. If the value cannot fit within the
5436 destination type, ``ty2``, then the results are undefined.
5441 .. code-block:: llvm
5443 %X = fptrunc double 123.0 to float ; yields float:123.0
5444 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5446 '``fpext .. to``' Instruction
5447 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5454 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5459 The '``fpext``' extends a floating point ``value`` to a larger floating
5465 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5466 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5467 to. The source type must be smaller than the destination type.
5472 The '``fpext``' instruction extends the ``value`` from a smaller
5473 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5474 point <t_floating>` type. The ``fpext`` cannot be used to make a
5475 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5476 *no-op cast* for a floating point cast.
5481 .. code-block:: llvm
5483 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5484 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5486 '``fptoui .. to``' Instruction
5487 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5494 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5499 The '``fptoui``' converts a floating point ``value`` to its unsigned
5500 integer equivalent of type ``ty2``.
5505 The '``fptoui``' instruction takes a value to cast, which must be a
5506 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5507 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5508 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5509 type with the same number of elements as ``ty``
5514 The '``fptoui``' instruction converts its :ref:`floating
5515 point <t_floating>` operand into the nearest (rounding towards zero)
5516 unsigned integer value. If the value cannot fit in ``ty2``, the results
5522 .. code-block:: llvm
5524 %X = fptoui double 123.0 to i32 ; yields i32:123
5525 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5526 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5528 '``fptosi .. to``' Instruction
5529 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5536 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5541 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5542 ``value`` to type ``ty2``.
5547 The '``fptosi``' instruction takes a value to cast, which must be a
5548 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5549 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5550 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5551 type with the same number of elements as ``ty``
5556 The '``fptosi``' instruction converts its :ref:`floating
5557 point <t_floating>` operand into the nearest (rounding towards zero)
5558 signed integer value. If the value cannot fit in ``ty2``, the results
5564 .. code-block:: llvm
5566 %X = fptosi double -123.0 to i32 ; yields i32:-123
5567 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5568 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5570 '``uitofp .. to``' Instruction
5571 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5578 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5583 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5584 and converts that value to the ``ty2`` type.
5589 The '``uitofp``' instruction takes a value to cast, which must be a
5590 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5591 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5592 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5593 type with the same number of elements as ``ty``
5598 The '``uitofp``' instruction interprets its operand as an unsigned
5599 integer quantity and converts it to the corresponding floating point
5600 value. If the value cannot fit in the floating point value, the results
5606 .. code-block:: llvm
5608 %X = uitofp i32 257 to float ; yields float:257.0
5609 %Y = uitofp i8 -1 to double ; yields double:255.0
5611 '``sitofp .. to``' Instruction
5612 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5619 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5624 The '``sitofp``' instruction regards ``value`` as a signed integer and
5625 converts that value to the ``ty2`` type.
5630 The '``sitofp``' instruction takes a value to cast, which must be a
5631 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5632 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5633 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5634 type with the same number of elements as ``ty``
5639 The '``sitofp``' instruction interprets its operand as a signed integer
5640 quantity and converts it to the corresponding floating point value. If
5641 the value cannot fit in the floating point value, the results are
5647 .. code-block:: llvm
5649 %X = sitofp i32 257 to float ; yields float:257.0
5650 %Y = sitofp i8 -1 to double ; yields double:-1.0
5654 '``ptrtoint .. to``' Instruction
5655 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5662 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5667 The '``ptrtoint``' instruction converts the pointer or a vector of
5668 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5673 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5674 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5675 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5676 a vector of integers type.
5681 The '``ptrtoint``' instruction converts ``value`` to integer type
5682 ``ty2`` by interpreting the pointer value as an integer and either
5683 truncating or zero extending that value to the size of the integer type.
5684 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5685 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5686 the same size, then nothing is done (*no-op cast*) other than a type
5692 .. code-block:: llvm
5694 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5695 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5696 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5700 '``inttoptr .. to``' Instruction
5701 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5708 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5713 The '``inttoptr``' instruction converts an integer ``value`` to a
5714 pointer type, ``ty2``.
5719 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5720 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5726 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5727 applying either a zero extension or a truncation depending on the size
5728 of the integer ``value``. If ``value`` is larger than the size of a
5729 pointer then a truncation is done. If ``value`` is smaller than the size
5730 of a pointer then a zero extension is done. If they are the same size,
5731 nothing is done (*no-op cast*).
5736 .. code-block:: llvm
5738 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5739 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5740 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5741 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5745 '``bitcast .. to``' Instruction
5746 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5753 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5758 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5764 The '``bitcast``' instruction takes a value to cast, which must be a
5765 non-aggregate first class value, and a type to cast it to, which must
5766 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5767 bit sizes of ``value`` and the destination type, ``ty2``, must be
5768 identical. If the source type is a pointer, the destination type must
5769 also be a pointer of the same size. This instruction supports bitwise
5770 conversion of vectors to integers and to vectors of other types (as
5771 long as they have the same size).
5776 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5777 is always a *no-op cast* because no bits change with this
5778 conversion. The conversion is done as if the ``value`` had been stored
5779 to memory and read back as type ``ty2``. Pointer (or vector of
5780 pointers) types may only be converted to other pointer (or vector of
5781 pointers) types with the same address space through this instruction.
5782 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5783 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5788 .. code-block:: llvm
5790 %X = bitcast i8 255 to i8 ; yields i8 :-1
5791 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5792 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5793 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5795 .. _i_addrspacecast:
5797 '``addrspacecast .. to``' Instruction
5798 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5805 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5810 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5811 address space ``n`` to type ``pty2`` in address space ``m``.
5816 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5817 to cast and a pointer type to cast it to, which must have a different
5823 The '``addrspacecast``' instruction converts the pointer value
5824 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5825 value modification, depending on the target and the address space
5826 pair. Pointer conversions within the same address space must be
5827 performed with the ``bitcast`` instruction. Note that if the address space
5828 conversion is legal then both result and operand refer to the same memory
5834 .. code-block:: llvm
5836 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5837 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5838 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5845 The instructions in this category are the "miscellaneous" instructions,
5846 which defy better classification.
5850 '``icmp``' Instruction
5851 ^^^^^^^^^^^^^^^^^^^^^^
5858 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5863 The '``icmp``' instruction returns a boolean value or a vector of
5864 boolean values based on comparison of its two integer, integer vector,
5865 pointer, or pointer vector operands.
5870 The '``icmp``' instruction takes three operands. The first operand is
5871 the condition code indicating the kind of comparison to perform. It is
5872 not a value, just a keyword. The possible condition code are:
5875 #. ``ne``: not equal
5876 #. ``ugt``: unsigned greater than
5877 #. ``uge``: unsigned greater or equal
5878 #. ``ult``: unsigned less than
5879 #. ``ule``: unsigned less or equal
5880 #. ``sgt``: signed greater than
5881 #. ``sge``: signed greater or equal
5882 #. ``slt``: signed less than
5883 #. ``sle``: signed less or equal
5885 The remaining two arguments must be :ref:`integer <t_integer>` or
5886 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5887 must also be identical types.
5892 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5893 code given as ``cond``. The comparison performed always yields either an
5894 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5896 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5897 otherwise. No sign interpretation is necessary or performed.
5898 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5899 otherwise. No sign interpretation is necessary or performed.
5900 #. ``ugt``: interprets the operands as unsigned values and yields
5901 ``true`` if ``op1`` is greater than ``op2``.
5902 #. ``uge``: interprets the operands as unsigned values and yields
5903 ``true`` if ``op1`` is greater than or equal to ``op2``.
5904 #. ``ult``: interprets the operands as unsigned values and yields
5905 ``true`` if ``op1`` is less than ``op2``.
5906 #. ``ule``: interprets the operands as unsigned values and yields
5907 ``true`` if ``op1`` is less than or equal to ``op2``.
5908 #. ``sgt``: interprets the operands as signed values and yields ``true``
5909 if ``op1`` is greater than ``op2``.
5910 #. ``sge``: interprets the operands as signed values and yields ``true``
5911 if ``op1`` is greater than or equal to ``op2``.
5912 #. ``slt``: interprets the operands as signed values and yields ``true``
5913 if ``op1`` is less than ``op2``.
5914 #. ``sle``: interprets the operands as signed values and yields ``true``
5915 if ``op1`` is less than or equal to ``op2``.
5917 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5918 are compared as if they were integers.
5920 If the operands are integer vectors, then they are compared element by
5921 element. The result is an ``i1`` vector with the same number of elements
5922 as the values being compared. Otherwise, the result is an ``i1``.
5927 .. code-block:: llvm
5929 <result> = icmp eq i32 4, 5 ; yields: result=false
5930 <result> = icmp ne float* %X, %X ; yields: result=false
5931 <result> = icmp ult i16 4, 5 ; yields: result=true
5932 <result> = icmp sgt i16 4, 5 ; yields: result=false
5933 <result> = icmp ule i16 -4, 5 ; yields: result=false
5934 <result> = icmp sge i16 4, 5 ; yields: result=false
5936 Note that the code generator does not yet support vector types with the
5937 ``icmp`` instruction.
5941 '``fcmp``' Instruction
5942 ^^^^^^^^^^^^^^^^^^^^^^
5949 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5954 The '``fcmp``' instruction returns a boolean value or vector of boolean
5955 values based on comparison of its operands.
5957 If the operands are floating point scalars, then the result type is a
5958 boolean (:ref:`i1 <t_integer>`).
5960 If the operands are floating point vectors, then the result type is a
5961 vector of boolean with the same number of elements as the operands being
5967 The '``fcmp``' instruction takes three operands. The first operand is
5968 the condition code indicating the kind of comparison to perform. It is
5969 not a value, just a keyword. The possible condition code are:
5971 #. ``false``: no comparison, always returns false
5972 #. ``oeq``: ordered and equal
5973 #. ``ogt``: ordered and greater than
5974 #. ``oge``: ordered and greater than or equal
5975 #. ``olt``: ordered and less than
5976 #. ``ole``: ordered and less than or equal
5977 #. ``one``: ordered and not equal
5978 #. ``ord``: ordered (no nans)
5979 #. ``ueq``: unordered or equal
5980 #. ``ugt``: unordered or greater than
5981 #. ``uge``: unordered or greater than or equal
5982 #. ``ult``: unordered or less than
5983 #. ``ule``: unordered or less than or equal
5984 #. ``une``: unordered or not equal
5985 #. ``uno``: unordered (either nans)
5986 #. ``true``: no comparison, always returns true
5988 *Ordered* means that neither operand is a QNAN while *unordered* means
5989 that either operand may be a QNAN.
5991 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5992 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5993 type. They must have identical types.
5998 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5999 condition code given as ``cond``. If the operands are vectors, then the
6000 vectors are compared element by element. Each comparison performed
6001 always yields an :ref:`i1 <t_integer>` result, as follows:
6003 #. ``false``: always yields ``false``, regardless of operands.
6004 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6005 is equal to ``op2``.
6006 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6007 is greater than ``op2``.
6008 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6009 is greater than or equal to ``op2``.
6010 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6011 is less than ``op2``.
6012 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6013 is less than or equal to ``op2``.
6014 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6015 is not equal to ``op2``.
6016 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6017 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6019 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6020 greater than ``op2``.
6021 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6022 greater than or equal to ``op2``.
6023 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6025 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6026 less than or equal to ``op2``.
6027 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6028 not equal to ``op2``.
6029 #. ``uno``: yields ``true`` if either operand is a QNAN.
6030 #. ``true``: always yields ``true``, regardless of operands.
6035 .. code-block:: llvm
6037 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6038 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6039 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6040 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6042 Note that the code generator does not yet support vector types with the
6043 ``fcmp`` instruction.
6047 '``phi``' Instruction
6048 ^^^^^^^^^^^^^^^^^^^^^
6055 <result> = phi <ty> [ <val0>, <label0>], ...
6060 The '``phi``' instruction is used to implement the φ node in the SSA
6061 graph representing the function.
6066 The type of the incoming values is specified with the first type field.
6067 After this, the '``phi``' instruction takes a list of pairs as
6068 arguments, with one pair for each predecessor basic block of the current
6069 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6070 the value arguments to the PHI node. Only labels may be used as the
6073 There must be no non-phi instructions between the start of a basic block
6074 and the PHI instructions: i.e. PHI instructions must be first in a basic
6077 For the purposes of the SSA form, the use of each incoming value is
6078 deemed to occur on the edge from the corresponding predecessor block to
6079 the current block (but after any definition of an '``invoke``'
6080 instruction's return value on the same edge).
6085 At runtime, the '``phi``' instruction logically takes on the value
6086 specified by the pair corresponding to the predecessor basic block that
6087 executed just prior to the current block.
6092 .. code-block:: llvm
6094 Loop: ; Infinite loop that counts from 0 on up...
6095 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6096 %nextindvar = add i32 %indvar, 1
6101 '``select``' Instruction
6102 ^^^^^^^^^^^^^^^^^^^^^^^^
6109 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6111 selty is either i1 or {<N x i1>}
6116 The '``select``' instruction is used to choose one value based on a
6117 condition, without IR-level branching.
6122 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6123 values indicating the condition, and two values of the same :ref:`first
6124 class <t_firstclass>` type. If the val1/val2 are vectors and the
6125 condition is a scalar, then entire vectors are selected, not individual
6131 If the condition is an i1 and it evaluates to 1, the instruction returns
6132 the first value argument; otherwise, it returns the second value
6135 If the condition is a vector of i1, then the value arguments must be
6136 vectors of the same size, and the selection is done element by element.
6141 .. code-block:: llvm
6143 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6147 '``call``' Instruction
6148 ^^^^^^^^^^^^^^^^^^^^^^
6155 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6160 The '``call``' instruction represents a simple function call.
6165 This instruction requires several arguments:
6167 #. The optional "tail" marker indicates that the callee function does
6168 not access any allocas or varargs in the caller. Note that calls may
6169 be marked "tail" even if they do not occur before a
6170 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6171 function call is eligible for tail call optimization, but `might not
6172 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6173 The code generator may optimize calls marked "tail" with either 1)
6174 automatic `sibling call
6175 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6176 callee have matching signatures, or 2) forced tail call optimization
6177 when the following extra requirements are met:
6179 - Caller and callee both have the calling convention ``fastcc``.
6180 - The call is in tail position (ret immediately follows call and ret
6181 uses value of call or is void).
6182 - Option ``-tailcallopt`` is enabled, or
6183 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6184 - `Platform specific constraints are
6185 met. <CodeGenerator.html#tailcallopt>`_
6187 #. The optional "cconv" marker indicates which :ref:`calling
6188 convention <callingconv>` the call should use. If none is
6189 specified, the call defaults to using C calling conventions. The
6190 calling convention of the call must match the calling convention of
6191 the target function, or else the behavior is undefined.
6192 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6193 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6195 #. '``ty``': the type of the call instruction itself which is also the
6196 type of the return value. Functions that return no value are marked
6198 #. '``fnty``': shall be the signature of the pointer to function value
6199 being invoked. The argument types must match the types implied by
6200 this signature. This type can be omitted if the function is not
6201 varargs and if the function type does not return a pointer to a
6203 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6204 be invoked. In most cases, this is a direct function invocation, but
6205 indirect ``call``'s are just as possible, calling an arbitrary pointer
6207 #. '``function args``': argument list whose types match the function
6208 signature argument types and parameter attributes. All arguments must
6209 be of :ref:`first class <t_firstclass>` type. If the function signature
6210 indicates the function accepts a variable number of arguments, the
6211 extra arguments can be specified.
6212 #. The optional :ref:`function attributes <fnattrs>` list. Only
6213 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6214 attributes are valid here.
6219 The '``call``' instruction is used to cause control flow to transfer to
6220 a specified function, with its incoming arguments bound to the specified
6221 values. Upon a '``ret``' instruction in the called function, control
6222 flow continues with the instruction after the function call, and the
6223 return value of the function is bound to the result argument.
6228 .. code-block:: llvm
6230 %retval = call i32 @test(i32 %argc)
6231 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6232 %X = tail call i32 @foo() ; yields i32
6233 %Y = tail call fastcc i32 @foo() ; yields i32
6234 call void %foo(i8 97 signext)
6236 %struct.A = type { i32, i8 }
6237 %r = call %struct.A @foo() ; yields { 32, i8 }
6238 %gr = extractvalue %struct.A %r, 0 ; yields i32
6239 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6240 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6241 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6243 llvm treats calls to some functions with names and arguments that match
6244 the standard C99 library as being the C99 library functions, and may
6245 perform optimizations or generate code for them under that assumption.
6246 This is something we'd like to change in the future to provide better
6247 support for freestanding environments and non-C-based languages.
6251 '``va_arg``' Instruction
6252 ^^^^^^^^^^^^^^^^^^^^^^^^
6259 <resultval> = va_arg <va_list*> <arglist>, <argty>
6264 The '``va_arg``' instruction is used to access arguments passed through
6265 the "variable argument" area of a function call. It is used to implement
6266 the ``va_arg`` macro in C.
6271 This instruction takes a ``va_list*`` value and the type of the
6272 argument. It returns a value of the specified argument type and
6273 increments the ``va_list`` to point to the next argument. The actual
6274 type of ``va_list`` is target specific.
6279 The '``va_arg``' instruction loads an argument of the specified type
6280 from the specified ``va_list`` and causes the ``va_list`` to point to
6281 the next argument. For more information, see the variable argument
6282 handling :ref:`Intrinsic Functions <int_varargs>`.
6284 It is legal for this instruction to be called in a function which does
6285 not take a variable number of arguments, for example, the ``vfprintf``
6288 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6289 function <intrinsics>` because it takes a type as an argument.
6294 See the :ref:`variable argument processing <int_varargs>` section.
6296 Note that the code generator does not yet fully support va\_arg on many
6297 targets. Also, it does not currently support va\_arg with aggregate
6298 types on any target.
6302 '``landingpad``' Instruction
6303 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6310 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6311 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6313 <clause> := catch <type> <value>
6314 <clause> := filter <array constant type> <array constant>
6319 The '``landingpad``' instruction is used by `LLVM's exception handling
6320 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6321 is a landing pad --- one where the exception lands, and corresponds to the
6322 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6323 defines values supplied by the personality function (``pers_fn``) upon
6324 re-entry to the function. The ``resultval`` has the type ``resultty``.
6329 This instruction takes a ``pers_fn`` value. This is the personality
6330 function associated with the unwinding mechanism. The optional
6331 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6333 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6334 contains the global variable representing the "type" that may be caught
6335 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6336 clause takes an array constant as its argument. Use
6337 "``[0 x i8**] undef``" for a filter which cannot throw. The
6338 '``landingpad``' instruction must contain *at least* one ``clause`` or
6339 the ``cleanup`` flag.
6344 The '``landingpad``' instruction defines the values which are set by the
6345 personality function (``pers_fn``) upon re-entry to the function, and
6346 therefore the "result type" of the ``landingpad`` instruction. As with
6347 calling conventions, how the personality function results are
6348 represented in LLVM IR is target specific.
6350 The clauses are applied in order from top to bottom. If two
6351 ``landingpad`` instructions are merged together through inlining, the
6352 clauses from the calling function are appended to the list of clauses.
6353 When the call stack is being unwound due to an exception being thrown,
6354 the exception is compared against each ``clause`` in turn. If it doesn't
6355 match any of the clauses, and the ``cleanup`` flag is not set, then
6356 unwinding continues further up the call stack.
6358 The ``landingpad`` instruction has several restrictions:
6360 - A landing pad block is a basic block which is the unwind destination
6361 of an '``invoke``' instruction.
6362 - A landing pad block must have a '``landingpad``' instruction as its
6363 first non-PHI instruction.
6364 - There can be only one '``landingpad``' instruction within the landing
6366 - A basic block that is not a landing pad block may not include a
6367 '``landingpad``' instruction.
6368 - All '``landingpad``' instructions in a function must have the same
6369 personality function.
6374 .. code-block:: llvm
6376 ;; A landing pad which can catch an integer.
6377 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6379 ;; A landing pad that is a cleanup.
6380 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6382 ;; A landing pad which can catch an integer and can only throw a double.
6383 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6385 filter [1 x i8**] [@_ZTId]
6392 LLVM supports the notion of an "intrinsic function". These functions
6393 have well known names and semantics and are required to follow certain
6394 restrictions. Overall, these intrinsics represent an extension mechanism
6395 for the LLVM language that does not require changing all of the
6396 transformations in LLVM when adding to the language (or the bitcode
6397 reader/writer, the parser, etc...).
6399 Intrinsic function names must all start with an "``llvm.``" prefix. This
6400 prefix is reserved in LLVM for intrinsic names; thus, function names may
6401 not begin with this prefix. Intrinsic functions must always be external
6402 functions: you cannot define the body of intrinsic functions. Intrinsic
6403 functions may only be used in call or invoke instructions: it is illegal
6404 to take the address of an intrinsic function. Additionally, because
6405 intrinsic functions are part of the LLVM language, it is required if any
6406 are added that they be documented here.
6408 Some intrinsic functions can be overloaded, i.e., the intrinsic
6409 represents a family of functions that perform the same operation but on
6410 different data types. Because LLVM can represent over 8 million
6411 different integer types, overloading is used commonly to allow an
6412 intrinsic function to operate on any integer type. One or more of the
6413 argument types or the result type can be overloaded to accept any
6414 integer type. Argument types may also be defined as exactly matching a
6415 previous argument's type or the result type. This allows an intrinsic
6416 function which accepts multiple arguments, but needs all of them to be
6417 of the same type, to only be overloaded with respect to a single
6418 argument or the result.
6420 Overloaded intrinsics will have the names of its overloaded argument
6421 types encoded into its function name, each preceded by a period. Only
6422 those types which are overloaded result in a name suffix. Arguments
6423 whose type is matched against another type do not. For example, the
6424 ``llvm.ctpop`` function can take an integer of any width and returns an
6425 integer of exactly the same integer width. This leads to a family of
6426 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6427 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6428 overloaded, and only one type suffix is required. Because the argument's
6429 type is matched against the return type, it does not require its own
6432 To learn how to add an intrinsic function, please see the `Extending
6433 LLVM Guide <ExtendingLLVM.html>`_.
6437 Variable Argument Handling Intrinsics
6438 -------------------------------------
6440 Variable argument support is defined in LLVM with the
6441 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6442 functions. These functions are related to the similarly named macros
6443 defined in the ``<stdarg.h>`` header file.
6445 All of these functions operate on arguments that use a target-specific
6446 value type "``va_list``". The LLVM assembly language reference manual
6447 does not define what this type is, so all transformations should be
6448 prepared to handle these functions regardless of the type used.
6450 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6451 variable argument handling intrinsic functions are used.
6453 .. code-block:: llvm
6455 define i32 @test(i32 %X, ...) {
6456 ; Initialize variable argument processing
6458 %ap2 = bitcast i8** %ap to i8*
6459 call void @llvm.va_start(i8* %ap2)
6461 ; Read a single integer argument
6462 %tmp = va_arg i8** %ap, i32
6464 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6466 %aq2 = bitcast i8** %aq to i8*
6467 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6468 call void @llvm.va_end(i8* %aq2)
6470 ; Stop processing of arguments.
6471 call void @llvm.va_end(i8* %ap2)
6475 declare void @llvm.va_start(i8*)
6476 declare void @llvm.va_copy(i8*, i8*)
6477 declare void @llvm.va_end(i8*)
6481 '``llvm.va_start``' Intrinsic
6482 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6489 declare void @llvm.va_start(i8* <arglist>)
6494 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6495 subsequent use by ``va_arg``.
6500 The argument is a pointer to a ``va_list`` element to initialize.
6505 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6506 available in C. In a target-dependent way, it initializes the
6507 ``va_list`` element to which the argument points, so that the next call
6508 to ``va_arg`` will produce the first variable argument passed to the
6509 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6510 to know the last argument of the function as the compiler can figure
6513 '``llvm.va_end``' Intrinsic
6514 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6521 declare void @llvm.va_end(i8* <arglist>)
6526 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6527 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6532 The argument is a pointer to a ``va_list`` to destroy.
6537 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6538 available in C. In a target-dependent way, it destroys the ``va_list``
6539 element to which the argument points. Calls to
6540 :ref:`llvm.va_start <int_va_start>` and
6541 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6546 '``llvm.va_copy``' Intrinsic
6547 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6554 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6559 The '``llvm.va_copy``' intrinsic copies the current argument position
6560 from the source argument list to the destination argument list.
6565 The first argument is a pointer to a ``va_list`` element to initialize.
6566 The second argument is a pointer to a ``va_list`` element to copy from.
6571 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6572 available in C. In a target-dependent way, it copies the source
6573 ``va_list`` element into the destination ``va_list`` element. This
6574 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6575 arbitrarily complex and require, for example, memory allocation.
6577 Accurate Garbage Collection Intrinsics
6578 --------------------------------------
6580 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6581 (GC) requires the implementation and generation of these intrinsics.
6582 These intrinsics allow identification of :ref:`GC roots on the
6583 stack <int_gcroot>`, as well as garbage collector implementations that
6584 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6585 Front-ends for type-safe garbage collected languages should generate
6586 these intrinsics to make use of the LLVM garbage collectors. For more
6587 details, see `Accurate Garbage Collection with
6588 LLVM <GarbageCollection.html>`_.
6590 The garbage collection intrinsics only operate on objects in the generic
6591 address space (address space zero).
6595 '``llvm.gcroot``' Intrinsic
6596 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6603 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6608 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6609 the code generator, and allows some metadata to be associated with it.
6614 The first argument specifies the address of a stack object that contains
6615 the root pointer. The second pointer (which must be either a constant or
6616 a global value address) contains the meta-data to be associated with the
6622 At runtime, a call to this intrinsic stores a null pointer into the
6623 "ptrloc" location. At compile-time, the code generator generates
6624 information to allow the runtime to find the pointer at GC safe points.
6625 The '``llvm.gcroot``' intrinsic may only be used in a function which
6626 :ref:`specifies a GC algorithm <gc>`.
6630 '``llvm.gcread``' Intrinsic
6631 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6638 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6643 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6644 locations, allowing garbage collector implementations that require read
6650 The second argument is the address to read from, which should be an
6651 address allocated from the garbage collector. The first object is a
6652 pointer to the start of the referenced object, if needed by the language
6653 runtime (otherwise null).
6658 The '``llvm.gcread``' intrinsic has the same semantics as a load
6659 instruction, but may be replaced with substantially more complex code by
6660 the garbage collector runtime, as needed. The '``llvm.gcread``'
6661 intrinsic may only be used in a function which :ref:`specifies a GC
6666 '``llvm.gcwrite``' Intrinsic
6667 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6674 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6679 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6680 locations, allowing garbage collector implementations that require write
6681 barriers (such as generational or reference counting collectors).
6686 The first argument is the reference to store, the second is the start of
6687 the object to store it to, and the third is the address of the field of
6688 Obj to store to. If the runtime does not require a pointer to the
6689 object, Obj may be null.
6694 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6695 instruction, but may be replaced with substantially more complex code by
6696 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6697 intrinsic may only be used in a function which :ref:`specifies a GC
6700 Code Generator Intrinsics
6701 -------------------------
6703 These intrinsics are provided by LLVM to expose special features that
6704 may only be implemented with code generator support.
6706 '``llvm.returnaddress``' Intrinsic
6707 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6714 declare i8 *@llvm.returnaddress(i32 <level>)
6719 The '``llvm.returnaddress``' intrinsic attempts to compute a
6720 target-specific value indicating the return address of the current
6721 function or one of its callers.
6726 The argument to this intrinsic indicates which function to return the
6727 address for. Zero indicates the calling function, one indicates its
6728 caller, etc. The argument is **required** to be a constant integer
6734 The '``llvm.returnaddress``' intrinsic either returns a pointer
6735 indicating the return address of the specified call frame, or zero if it
6736 cannot be identified. The value returned by this intrinsic is likely to
6737 be incorrect or 0 for arguments other than zero, so it should only be
6738 used for debugging purposes.
6740 Note that calling this intrinsic does not prevent function inlining or
6741 other aggressive transformations, so the value returned may not be that
6742 of the obvious source-language caller.
6744 '``llvm.frameaddress``' Intrinsic
6745 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6752 declare i8* @llvm.frameaddress(i32 <level>)
6757 The '``llvm.frameaddress``' intrinsic attempts to return the
6758 target-specific frame pointer value for the specified stack frame.
6763 The argument to this intrinsic indicates which function to return the
6764 frame pointer for. Zero indicates the calling function, one indicates
6765 its caller, etc. The argument is **required** to be a constant integer
6771 The '``llvm.frameaddress``' intrinsic either returns a pointer
6772 indicating the frame address of the specified call frame, or zero if it
6773 cannot be identified. The value returned by this intrinsic is likely to
6774 be incorrect or 0 for arguments other than zero, so it should only be
6775 used for debugging purposes.
6777 Note that calling this intrinsic does not prevent function inlining or
6778 other aggressive transformations, so the value returned may not be that
6779 of the obvious source-language caller.
6783 '``llvm.stacksave``' Intrinsic
6784 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6791 declare i8* @llvm.stacksave()
6796 The '``llvm.stacksave``' intrinsic is used to remember the current state
6797 of the function stack, for use with
6798 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6799 implementing language features like scoped automatic variable sized
6805 This intrinsic returns a opaque pointer value that can be passed to
6806 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6807 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6808 ``llvm.stacksave``, it effectively restores the state of the stack to
6809 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6810 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6811 were allocated after the ``llvm.stacksave`` was executed.
6813 .. _int_stackrestore:
6815 '``llvm.stackrestore``' Intrinsic
6816 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6823 declare void @llvm.stackrestore(i8* %ptr)
6828 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6829 the function stack to the state it was in when the corresponding
6830 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6831 useful for implementing language features like scoped automatic variable
6832 sized arrays in C99.
6837 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6839 '``llvm.prefetch``' Intrinsic
6840 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6847 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6852 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6853 insert a prefetch instruction if supported; otherwise, it is a noop.
6854 Prefetches have no effect on the behavior of the program but can change
6855 its performance characteristics.
6860 ``address`` is the address to be prefetched, ``rw`` is the specifier
6861 determining if the fetch should be for a read (0) or write (1), and
6862 ``locality`` is a temporal locality specifier ranging from (0) - no
6863 locality, to (3) - extremely local keep in cache. The ``cache type``
6864 specifies whether the prefetch is performed on the data (1) or
6865 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6866 arguments must be constant integers.
6871 This intrinsic does not modify the behavior of the program. In
6872 particular, prefetches cannot trap and do not produce a value. On
6873 targets that support this intrinsic, the prefetch can provide hints to
6874 the processor cache for better performance.
6876 '``llvm.pcmarker``' Intrinsic
6877 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6884 declare void @llvm.pcmarker(i32 <id>)
6889 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6890 Counter (PC) in a region of code to simulators and other tools. The
6891 method is target specific, but it is expected that the marker will use
6892 exported symbols to transmit the PC of the marker. The marker makes no
6893 guarantees that it will remain with any specific instruction after
6894 optimizations. It is possible that the presence of a marker will inhibit
6895 optimizations. The intended use is to be inserted after optimizations to
6896 allow correlations of simulation runs.
6901 ``id`` is a numerical id identifying the marker.
6906 This intrinsic does not modify the behavior of the program. Backends
6907 that do not support this intrinsic may ignore it.
6909 '``llvm.readcyclecounter``' Intrinsic
6910 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6917 declare i64 @llvm.readcyclecounter()
6922 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6923 counter register (or similar low latency, high accuracy clocks) on those
6924 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6925 should map to RPCC. As the backing counters overflow quickly (on the
6926 order of 9 seconds on alpha), this should only be used for small
6932 When directly supported, reading the cycle counter should not modify any
6933 memory. Implementations are allowed to either return a application
6934 specific value or a system wide value. On backends without support, this
6935 is lowered to a constant 0.
6937 Note that runtime support may be conditional on the privilege-level code is
6938 running at and the host platform.
6940 '``llvm.clear_cache``' Intrinsic
6941 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6948 declare void @llvm.clear_cache(i8*, i8*)
6953 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
6954 in the specified range to the execution unit of the processor. On
6955 targets with non-unified instruction and data cache, the implementation
6956 flushes the instruction cache.
6961 On platforms with coherent instruction and data caches (e.g. x86), this
6962 intrinsic is a nop. On platforms with non-coherent instruction and data
6963 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropiate
6964 instructions or a system call, if cache flushing requires special
6967 The default behavior is to emit a call to ``__clear_cache'' from the run
6970 This instrinsic does *not* empty the instruction pipeline. Modifications
6971 of the current function are outside the scope of the intrinsic.
6973 Standard C Library Intrinsics
6974 -----------------------------
6976 LLVM provides intrinsics for a few important standard C library
6977 functions. These intrinsics allow source-language front-ends to pass
6978 information about the alignment of the pointer arguments to the code
6979 generator, providing opportunity for more efficient code generation.
6983 '``llvm.memcpy``' Intrinsic
6984 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6989 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6990 integer bit width and for different address spaces. Not all targets
6991 support all bit widths however.
6995 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6996 i32 <len>, i32 <align>, i1 <isvolatile>)
6997 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6998 i64 <len>, i32 <align>, i1 <isvolatile>)
7003 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7004 source location to the destination location.
7006 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
7007 intrinsics do not return a value, takes extra alignment/isvolatile
7008 arguments and the pointers can be in specified address spaces.
7013 The first argument is a pointer to the destination, the second is a
7014 pointer to the source. The third argument is an integer argument
7015 specifying the number of bytes to copy, the fourth argument is the
7016 alignment of the source and destination locations, and the fifth is a
7017 boolean indicating a volatile access.
7019 If the call to this intrinsic has an alignment value that is not 0 or 1,
7020 then the caller guarantees that both the source and destination pointers
7021 are aligned to that boundary.
7023 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
7024 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7025 very cleanly specified and it is unwise to depend on it.
7030 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7031 source location to the destination location, which are not allowed to
7032 overlap. It copies "len" bytes of memory over. If the argument is known
7033 to be aligned to some boundary, this can be specified as the fourth
7034 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
7036 '``llvm.memmove``' Intrinsic
7037 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7042 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7043 bit width and for different address space. Not all targets support all
7048 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7049 i32 <len>, i32 <align>, i1 <isvolatile>)
7050 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7051 i64 <len>, i32 <align>, i1 <isvolatile>)
7056 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7057 source location to the destination location. It is similar to the
7058 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7061 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7062 intrinsics do not return a value, takes extra alignment/isvolatile
7063 arguments and the pointers can be in specified address spaces.
7068 The first argument is a pointer to the destination, the second is a
7069 pointer to the source. The third argument is an integer argument
7070 specifying the number of bytes to copy, the fourth argument is the
7071 alignment of the source and destination locations, and the fifth is a
7072 boolean indicating a volatile access.
7074 If the call to this intrinsic has an alignment value that is not 0 or 1,
7075 then the caller guarantees that the source and destination pointers are
7076 aligned to that boundary.
7078 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7079 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7080 not very cleanly specified and it is unwise to depend on it.
7085 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7086 source location to the destination location, which may overlap. It
7087 copies "len" bytes of memory over. If the argument is known to be
7088 aligned to some boundary, this can be specified as the fourth argument,
7089 otherwise it should be set to 0 or 1 (both meaning no alignment).
7091 '``llvm.memset.*``' Intrinsics
7092 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7097 This is an overloaded intrinsic. You can use llvm.memset on any integer
7098 bit width and for different address spaces. However, not all targets
7099 support all bit widths.
7103 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7104 i32 <len>, i32 <align>, i1 <isvolatile>)
7105 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7106 i64 <len>, i32 <align>, i1 <isvolatile>)
7111 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7112 particular byte value.
7114 Note that, unlike the standard libc function, the ``llvm.memset``
7115 intrinsic does not return a value and takes extra alignment/volatile
7116 arguments. Also, the destination can be in an arbitrary address space.
7121 The first argument is a pointer to the destination to fill, the second
7122 is the byte value with which to fill it, the third argument is an
7123 integer argument specifying the number of bytes to fill, and the fourth
7124 argument is the known alignment of the destination location.
7126 If the call to this intrinsic has an alignment value that is not 0 or 1,
7127 then the caller guarantees that the destination pointer is aligned to
7130 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7131 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7132 very cleanly specified and it is unwise to depend on it.
7137 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7138 at the destination location. If the argument is known to be aligned to
7139 some boundary, this can be specified as the fourth argument, otherwise
7140 it should be set to 0 or 1 (both meaning no alignment).
7142 '``llvm.sqrt.*``' Intrinsic
7143 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7148 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7149 floating point or vector of floating point type. Not all targets support
7154 declare float @llvm.sqrt.f32(float %Val)
7155 declare double @llvm.sqrt.f64(double %Val)
7156 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7157 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7158 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7163 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7164 returning the same value as the libm '``sqrt``' functions would. Unlike
7165 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7166 negative numbers other than -0.0 (which allows for better optimization,
7167 because there is no need to worry about errno being set).
7168 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7173 The argument and return value are floating point numbers of the same
7179 This function returns the sqrt of the specified operand if it is a
7180 nonnegative floating point number.
7182 '``llvm.powi.*``' Intrinsic
7183 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7188 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7189 floating point or vector of floating point type. Not all targets support
7194 declare float @llvm.powi.f32(float %Val, i32 %power)
7195 declare double @llvm.powi.f64(double %Val, i32 %power)
7196 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7197 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7198 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7203 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7204 specified (positive or negative) power. The order of evaluation of
7205 multiplications is not defined. When a vector of floating point type is
7206 used, the second argument remains a scalar integer value.
7211 The second argument is an integer power, and the first is a value to
7212 raise to that power.
7217 This function returns the first value raised to the second power with an
7218 unspecified sequence of rounding operations.
7220 '``llvm.sin.*``' Intrinsic
7221 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7226 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7227 floating point or vector of floating point type. Not all targets support
7232 declare float @llvm.sin.f32(float %Val)
7233 declare double @llvm.sin.f64(double %Val)
7234 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7235 declare fp128 @llvm.sin.f128(fp128 %Val)
7236 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7241 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7246 The argument and return value are floating point numbers of the same
7252 This function returns the sine of the specified operand, returning the
7253 same values as the libm ``sin`` functions would, and handles error
7254 conditions in the same way.
7256 '``llvm.cos.*``' Intrinsic
7257 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7262 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7263 floating point or vector of floating point type. Not all targets support
7268 declare float @llvm.cos.f32(float %Val)
7269 declare double @llvm.cos.f64(double %Val)
7270 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7271 declare fp128 @llvm.cos.f128(fp128 %Val)
7272 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7277 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7282 The argument and return value are floating point numbers of the same
7288 This function returns the cosine of the specified operand, returning the
7289 same values as the libm ``cos`` functions would, and handles error
7290 conditions in the same way.
7292 '``llvm.pow.*``' Intrinsic
7293 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7298 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7299 floating point or vector of floating point type. Not all targets support
7304 declare float @llvm.pow.f32(float %Val, float %Power)
7305 declare double @llvm.pow.f64(double %Val, double %Power)
7306 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7307 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7308 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7313 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7314 specified (positive or negative) power.
7319 The second argument is a floating point power, and the first is a value
7320 to raise to that power.
7325 This function returns the first value raised to the second power,
7326 returning the same values as the libm ``pow`` functions would, and
7327 handles error conditions in the same way.
7329 '``llvm.exp.*``' Intrinsic
7330 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7335 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7336 floating point or vector of floating point type. Not all targets support
7341 declare float @llvm.exp.f32(float %Val)
7342 declare double @llvm.exp.f64(double %Val)
7343 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7344 declare fp128 @llvm.exp.f128(fp128 %Val)
7345 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7350 The '``llvm.exp.*``' intrinsics perform the exp function.
7355 The argument and return value are floating point numbers of the same
7361 This function returns the same values as the libm ``exp`` functions
7362 would, and handles error conditions in the same way.
7364 '``llvm.exp2.*``' Intrinsic
7365 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7370 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7371 floating point or vector of floating point type. Not all targets support
7376 declare float @llvm.exp2.f32(float %Val)
7377 declare double @llvm.exp2.f64(double %Val)
7378 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7379 declare fp128 @llvm.exp2.f128(fp128 %Val)
7380 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7385 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7390 The argument and return value are floating point numbers of the same
7396 This function returns the same values as the libm ``exp2`` functions
7397 would, and handles error conditions in the same way.
7399 '``llvm.log.*``' Intrinsic
7400 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7405 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7406 floating point or vector of floating point type. Not all targets support
7411 declare float @llvm.log.f32(float %Val)
7412 declare double @llvm.log.f64(double %Val)
7413 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7414 declare fp128 @llvm.log.f128(fp128 %Val)
7415 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7420 The '``llvm.log.*``' intrinsics perform the log function.
7425 The argument and return value are floating point numbers of the same
7431 This function returns the same values as the libm ``log`` functions
7432 would, and handles error conditions in the same way.
7434 '``llvm.log10.*``' Intrinsic
7435 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7440 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7441 floating point or vector of floating point type. Not all targets support
7446 declare float @llvm.log10.f32(float %Val)
7447 declare double @llvm.log10.f64(double %Val)
7448 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7449 declare fp128 @llvm.log10.f128(fp128 %Val)
7450 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7455 The '``llvm.log10.*``' intrinsics perform the log10 function.
7460 The argument and return value are floating point numbers of the same
7466 This function returns the same values as the libm ``log10`` functions
7467 would, and handles error conditions in the same way.
7469 '``llvm.log2.*``' Intrinsic
7470 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7475 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7476 floating point or vector of floating point type. Not all targets support
7481 declare float @llvm.log2.f32(float %Val)
7482 declare double @llvm.log2.f64(double %Val)
7483 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7484 declare fp128 @llvm.log2.f128(fp128 %Val)
7485 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7490 The '``llvm.log2.*``' intrinsics perform the log2 function.
7495 The argument and return value are floating point numbers of the same
7501 This function returns the same values as the libm ``log2`` functions
7502 would, and handles error conditions in the same way.
7504 '``llvm.fma.*``' Intrinsic
7505 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7510 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7511 floating point or vector of floating point type. Not all targets support
7516 declare float @llvm.fma.f32(float %a, float %b, float %c)
7517 declare double @llvm.fma.f64(double %a, double %b, double %c)
7518 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7519 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7520 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7525 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7531 The argument and return value are floating point numbers of the same
7537 This function returns the same values as the libm ``fma`` functions
7538 would, and does not set errno.
7540 '``llvm.fabs.*``' Intrinsic
7541 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7546 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7547 floating point or vector of floating point type. Not all targets support
7552 declare float @llvm.fabs.f32(float %Val)
7553 declare double @llvm.fabs.f64(double %Val)
7554 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7555 declare fp128 @llvm.fabs.f128(fp128 %Val)
7556 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7561 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7567 The argument and return value are floating point numbers of the same
7573 This function returns the same values as the libm ``fabs`` functions
7574 would, and handles error conditions in the same way.
7576 '``llvm.copysign.*``' Intrinsic
7577 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7582 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7583 floating point or vector of floating point type. Not all targets support
7588 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7589 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7590 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7591 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7592 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7597 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7598 first operand and the sign of the second operand.
7603 The arguments and return value are floating point numbers of the same
7609 This function returns the same values as the libm ``copysign``
7610 functions would, and handles error conditions in the same way.
7612 '``llvm.floor.*``' Intrinsic
7613 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7618 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7619 floating point or vector of floating point type. Not all targets support
7624 declare float @llvm.floor.f32(float %Val)
7625 declare double @llvm.floor.f64(double %Val)
7626 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7627 declare fp128 @llvm.floor.f128(fp128 %Val)
7628 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7633 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7638 The argument and return value are floating point numbers of the same
7644 This function returns the same values as the libm ``floor`` functions
7645 would, and handles error conditions in the same way.
7647 '``llvm.ceil.*``' Intrinsic
7648 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7653 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7654 floating point or vector of floating point type. Not all targets support
7659 declare float @llvm.ceil.f32(float %Val)
7660 declare double @llvm.ceil.f64(double %Val)
7661 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7662 declare fp128 @llvm.ceil.f128(fp128 %Val)
7663 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7668 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7673 The argument and return value are floating point numbers of the same
7679 This function returns the same values as the libm ``ceil`` functions
7680 would, and handles error conditions in the same way.
7682 '``llvm.trunc.*``' Intrinsic
7683 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7688 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7689 floating point or vector of floating point type. Not all targets support
7694 declare float @llvm.trunc.f32(float %Val)
7695 declare double @llvm.trunc.f64(double %Val)
7696 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7697 declare fp128 @llvm.trunc.f128(fp128 %Val)
7698 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7703 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7704 nearest integer not larger in magnitude than the operand.
7709 The argument and return value are floating point numbers of the same
7715 This function returns the same values as the libm ``trunc`` functions
7716 would, and handles error conditions in the same way.
7718 '``llvm.rint.*``' Intrinsic
7719 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7724 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7725 floating point or vector of floating point type. Not all targets support
7730 declare float @llvm.rint.f32(float %Val)
7731 declare double @llvm.rint.f64(double %Val)
7732 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7733 declare fp128 @llvm.rint.f128(fp128 %Val)
7734 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7739 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7740 nearest integer. It may raise an inexact floating-point exception if the
7741 operand isn't an integer.
7746 The argument and return value are floating point numbers of the same
7752 This function returns the same values as the libm ``rint`` functions
7753 would, and handles error conditions in the same way.
7755 '``llvm.nearbyint.*``' Intrinsic
7756 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7761 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7762 floating point or vector of floating point type. Not all targets support
7767 declare float @llvm.nearbyint.f32(float %Val)
7768 declare double @llvm.nearbyint.f64(double %Val)
7769 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7770 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7771 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7776 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7782 The argument and return value are floating point numbers of the same
7788 This function returns the same values as the libm ``nearbyint``
7789 functions would, and handles error conditions in the same way.
7791 '``llvm.round.*``' Intrinsic
7792 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7797 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7798 floating point or vector of floating point type. Not all targets support
7803 declare float @llvm.round.f32(float %Val)
7804 declare double @llvm.round.f64(double %Val)
7805 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7806 declare fp128 @llvm.round.f128(fp128 %Val)
7807 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7812 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7818 The argument and return value are floating point numbers of the same
7824 This function returns the same values as the libm ``round``
7825 functions would, and handles error conditions in the same way.
7827 Bit Manipulation Intrinsics
7828 ---------------------------
7830 LLVM provides intrinsics for a few important bit manipulation
7831 operations. These allow efficient code generation for some algorithms.
7833 '``llvm.bswap.*``' Intrinsics
7834 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7839 This is an overloaded intrinsic function. You can use bswap on any
7840 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7844 declare i16 @llvm.bswap.i16(i16 <id>)
7845 declare i32 @llvm.bswap.i32(i32 <id>)
7846 declare i64 @llvm.bswap.i64(i64 <id>)
7851 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7852 values with an even number of bytes (positive multiple of 16 bits).
7853 These are useful for performing operations on data that is not in the
7854 target's native byte order.
7859 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7860 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7861 intrinsic returns an i32 value that has the four bytes of the input i32
7862 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7863 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7864 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7865 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7868 '``llvm.ctpop.*``' Intrinsic
7869 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7874 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7875 bit width, or on any vector with integer elements. Not all targets
7876 support all bit widths or vector types, however.
7880 declare i8 @llvm.ctpop.i8(i8 <src>)
7881 declare i16 @llvm.ctpop.i16(i16 <src>)
7882 declare i32 @llvm.ctpop.i32(i32 <src>)
7883 declare i64 @llvm.ctpop.i64(i64 <src>)
7884 declare i256 @llvm.ctpop.i256(i256 <src>)
7885 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7890 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7896 The only argument is the value to be counted. The argument may be of any
7897 integer type, or a vector with integer elements. The return type must
7898 match the argument type.
7903 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7904 each element of a vector.
7906 '``llvm.ctlz.*``' Intrinsic
7907 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7912 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7913 integer bit width, or any vector whose elements are integers. Not all
7914 targets support all bit widths or vector types, however.
7918 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7919 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7920 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7921 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7922 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7923 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7928 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7929 leading zeros in a variable.
7934 The first argument is the value to be counted. This argument may be of
7935 any integer type, or a vectory with integer element type. The return
7936 type must match the first argument type.
7938 The second argument must be a constant and is a flag to indicate whether
7939 the intrinsic should ensure that a zero as the first argument produces a
7940 defined result. Historically some architectures did not provide a
7941 defined result for zero values as efficiently, and many algorithms are
7942 now predicated on avoiding zero-value inputs.
7947 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7948 zeros in a variable, or within each element of the vector. If
7949 ``src == 0`` then the result is the size in bits of the type of ``src``
7950 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7951 ``llvm.ctlz(i32 2) = 30``.
7953 '``llvm.cttz.*``' Intrinsic
7954 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7959 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7960 integer bit width, or any vector of integer elements. Not all targets
7961 support all bit widths or vector types, however.
7965 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7966 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7967 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7968 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7969 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7970 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7975 The '``llvm.cttz``' family of intrinsic functions counts the number of
7981 The first argument is the value to be counted. This argument may be of
7982 any integer type, or a vectory with integer element type. The return
7983 type must match the first argument type.
7985 The second argument must be a constant and is a flag to indicate whether
7986 the intrinsic should ensure that a zero as the first argument produces a
7987 defined result. Historically some architectures did not provide a
7988 defined result for zero values as efficiently, and many algorithms are
7989 now predicated on avoiding zero-value inputs.
7994 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7995 zeros in a variable, or within each element of a vector. If ``src == 0``
7996 then the result is the size in bits of the type of ``src`` if
7997 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7998 ``llvm.cttz(2) = 1``.
8000 Arithmetic with Overflow Intrinsics
8001 -----------------------------------
8003 LLVM provides intrinsics for some arithmetic with overflow operations.
8005 '``llvm.sadd.with.overflow.*``' Intrinsics
8006 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8011 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
8012 on any integer bit width.
8016 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8017 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8018 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8023 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8024 a signed addition of the two arguments, and indicate whether an overflow
8025 occurred during the signed summation.
8030 The arguments (%a and %b) and the first element of the result structure
8031 may be of integer types of any bit width, but they must have the same
8032 bit width. The second element of the result structure must be of type
8033 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8039 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8040 a signed addition of the two variables. They return a structure --- the
8041 first element of which is the signed summation, and the second element
8042 of which is a bit specifying if the signed summation resulted in an
8048 .. code-block:: llvm
8050 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8051 %sum = extractvalue {i32, i1} %res, 0
8052 %obit = extractvalue {i32, i1} %res, 1
8053 br i1 %obit, label %overflow, label %normal
8055 '``llvm.uadd.with.overflow.*``' Intrinsics
8056 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8061 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8062 on any integer bit width.
8066 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8067 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8068 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8073 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8074 an unsigned addition of the two arguments, and indicate whether a carry
8075 occurred during the unsigned summation.
8080 The arguments (%a and %b) and the first element of the result structure
8081 may be of integer types of any bit width, but they must have the same
8082 bit width. The second element of the result structure must be of type
8083 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8089 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8090 an unsigned addition of the two arguments. They return a structure --- the
8091 first element of which is the sum, and the second element of which is a
8092 bit specifying if the unsigned summation resulted in a carry.
8097 .. code-block:: llvm
8099 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8100 %sum = extractvalue {i32, i1} %res, 0
8101 %obit = extractvalue {i32, i1} %res, 1
8102 br i1 %obit, label %carry, label %normal
8104 '``llvm.ssub.with.overflow.*``' Intrinsics
8105 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8110 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8111 on any integer bit width.
8115 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8116 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8117 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8122 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8123 a signed subtraction of the two arguments, and indicate whether an
8124 overflow occurred during the signed subtraction.
8129 The arguments (%a and %b) and the first element of the result structure
8130 may be of integer types of any bit width, but they must have the same
8131 bit width. The second element of the result structure must be of type
8132 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8138 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8139 a signed subtraction of the two arguments. They return a structure --- the
8140 first element of which is the subtraction, and the second element of
8141 which is a bit specifying if the signed subtraction resulted in an
8147 .. code-block:: llvm
8149 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8150 %sum = extractvalue {i32, i1} %res, 0
8151 %obit = extractvalue {i32, i1} %res, 1
8152 br i1 %obit, label %overflow, label %normal
8154 '``llvm.usub.with.overflow.*``' Intrinsics
8155 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8160 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8161 on any integer bit width.
8165 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8166 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8167 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8172 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8173 an unsigned subtraction of the two arguments, and indicate whether an
8174 overflow occurred during the unsigned subtraction.
8179 The arguments (%a and %b) and the first element of the result structure
8180 may be of integer types of any bit width, but they must have the same
8181 bit width. The second element of the result structure must be of type
8182 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8188 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8189 an unsigned subtraction of the two arguments. They return a structure ---
8190 the first element of which is the subtraction, and the second element of
8191 which is a bit specifying if the unsigned subtraction resulted in an
8197 .. code-block:: llvm
8199 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8200 %sum = extractvalue {i32, i1} %res, 0
8201 %obit = extractvalue {i32, i1} %res, 1
8202 br i1 %obit, label %overflow, label %normal
8204 '``llvm.smul.with.overflow.*``' Intrinsics
8205 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8210 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8211 on any integer bit width.
8215 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8216 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8217 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8222 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8223 a signed multiplication of the two arguments, and indicate whether an
8224 overflow occurred during the signed multiplication.
8229 The arguments (%a and %b) and the first element of the result structure
8230 may be of integer types of any bit width, but they must have the same
8231 bit width. The second element of the result structure must be of type
8232 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8238 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8239 a signed multiplication of the two arguments. They return a structure ---
8240 the first element of which is the multiplication, and the second element
8241 of which is a bit specifying if the signed multiplication resulted in an
8247 .. code-block:: llvm
8249 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8250 %sum = extractvalue {i32, i1} %res, 0
8251 %obit = extractvalue {i32, i1} %res, 1
8252 br i1 %obit, label %overflow, label %normal
8254 '``llvm.umul.with.overflow.*``' Intrinsics
8255 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8260 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8261 on any integer bit width.
8265 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8266 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8267 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8272 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8273 a unsigned multiplication of the two arguments, and indicate whether an
8274 overflow occurred during the unsigned multiplication.
8279 The arguments (%a and %b) and the first element of the result structure
8280 may be of integer types of any bit width, but they must have the same
8281 bit width. The second element of the result structure must be of type
8282 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8288 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8289 an unsigned multiplication of the two arguments. They return a structure ---
8290 the first element of which is the multiplication, and the second
8291 element of which is a bit specifying if the unsigned multiplication
8292 resulted in an overflow.
8297 .. code-block:: llvm
8299 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8300 %sum = extractvalue {i32, i1} %res, 0
8301 %obit = extractvalue {i32, i1} %res, 1
8302 br i1 %obit, label %overflow, label %normal
8304 Specialised Arithmetic Intrinsics
8305 ---------------------------------
8307 '``llvm.fmuladd.*``' Intrinsic
8308 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8315 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8316 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8321 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8322 expressions that can be fused if the code generator determines that (a) the
8323 target instruction set has support for a fused operation, and (b) that the
8324 fused operation is more efficient than the equivalent, separate pair of mul
8325 and add instructions.
8330 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8331 multiplicands, a and b, and an addend c.
8340 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8342 is equivalent to the expression a \* b + c, except that rounding will
8343 not be performed between the multiplication and addition steps if the
8344 code generator fuses the operations. Fusion is not guaranteed, even if
8345 the target platform supports it. If a fused multiply-add is required the
8346 corresponding llvm.fma.\* intrinsic function should be used
8347 instead. This never sets errno, just as '``llvm.fma.*``'.
8352 .. code-block:: llvm
8354 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8356 Half Precision Floating Point Intrinsics
8357 ----------------------------------------
8359 For most target platforms, half precision floating point is a
8360 storage-only format. This means that it is a dense encoding (in memory)
8361 but does not support computation in the format.
8363 This means that code must first load the half-precision floating point
8364 value as an i16, then convert it to float with
8365 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8366 then be performed on the float value (including extending to double
8367 etc). To store the value back to memory, it is first converted to float
8368 if needed, then converted to i16 with
8369 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8372 .. _int_convert_to_fp16:
8374 '``llvm.convert.to.fp16``' Intrinsic
8375 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8382 declare i16 @llvm.convert.to.fp16(f32 %a)
8387 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8388 from single precision floating point format to half precision floating
8394 The intrinsic function contains single argument - the value to be
8400 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8401 from single precision floating point format to half precision floating
8402 point format. The return value is an ``i16`` which contains the
8408 .. code-block:: llvm
8410 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8411 store i16 %res, i16* @x, align 2
8413 .. _int_convert_from_fp16:
8415 '``llvm.convert.from.fp16``' Intrinsic
8416 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8423 declare f32 @llvm.convert.from.fp16(i16 %a)
8428 The '``llvm.convert.from.fp16``' intrinsic function performs a
8429 conversion from half precision floating point format to single precision
8430 floating point format.
8435 The intrinsic function contains single argument - the value to be
8441 The '``llvm.convert.from.fp16``' intrinsic function performs a
8442 conversion from half single precision floating point format to single
8443 precision floating point format. The input half-float value is
8444 represented by an ``i16`` value.
8449 .. code-block:: llvm
8451 %a = load i16* @x, align 2
8452 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8457 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8458 prefix), are described in the `LLVM Source Level
8459 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8462 Exception Handling Intrinsics
8463 -----------------------------
8465 The LLVM exception handling intrinsics (which all start with
8466 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8467 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8471 Trampoline Intrinsics
8472 ---------------------
8474 These intrinsics make it possible to excise one parameter, marked with
8475 the :ref:`nest <nest>` attribute, from a function. The result is a
8476 callable function pointer lacking the nest parameter - the caller does
8477 not need to provide a value for it. Instead, the value to use is stored
8478 in advance in a "trampoline", a block of memory usually allocated on the
8479 stack, which also contains code to splice the nest value into the
8480 argument list. This is used to implement the GCC nested function address
8483 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8484 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8485 It can be created as follows:
8487 .. code-block:: llvm
8489 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8490 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8491 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8492 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8493 %fp = bitcast i8* %p to i32 (i32, i32)*
8495 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8496 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8500 '``llvm.init.trampoline``' Intrinsic
8501 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8508 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8513 This fills the memory pointed to by ``tramp`` with executable code,
8514 turning it into a trampoline.
8519 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8520 pointers. The ``tramp`` argument must point to a sufficiently large and
8521 sufficiently aligned block of memory; this memory is written to by the
8522 intrinsic. Note that the size and the alignment are target-specific -
8523 LLVM currently provides no portable way of determining them, so a
8524 front-end that generates this intrinsic needs to have some
8525 target-specific knowledge. The ``func`` argument must hold a function
8526 bitcast to an ``i8*``.
8531 The block of memory pointed to by ``tramp`` is filled with target
8532 dependent code, turning it into a function. Then ``tramp`` needs to be
8533 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8534 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8535 function's signature is the same as that of ``func`` with any arguments
8536 marked with the ``nest`` attribute removed. At most one such ``nest``
8537 argument is allowed, and it must be of pointer type. Calling the new
8538 function is equivalent to calling ``func`` with the same argument list,
8539 but with ``nval`` used for the missing ``nest`` argument. If, after
8540 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8541 modified, then the effect of any later call to the returned function
8542 pointer is undefined.
8546 '``llvm.adjust.trampoline``' Intrinsic
8547 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8554 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8559 This performs any required machine-specific adjustment to the address of
8560 a trampoline (passed as ``tramp``).
8565 ``tramp`` must point to a block of memory which already has trampoline
8566 code filled in by a previous call to
8567 :ref:`llvm.init.trampoline <int_it>`.
8572 On some architectures the address of the code to be executed needs to be
8573 different to the address where the trampoline is actually stored. This
8574 intrinsic returns the executable address corresponding to ``tramp``
8575 after performing the required machine specific adjustments. The pointer
8576 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8581 This class of intrinsics exists to information about the lifetime of
8582 memory objects and ranges where variables are immutable.
8586 '``llvm.lifetime.start``' Intrinsic
8587 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8594 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8599 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8605 The first argument is a constant integer representing the size of the
8606 object, or -1 if it is variable sized. The second argument is a pointer
8612 This intrinsic indicates that before this point in the code, the value
8613 of the memory pointed to by ``ptr`` is dead. This means that it is known
8614 to never be used and has an undefined value. A load from the pointer
8615 that precedes this intrinsic can be replaced with ``'undef'``.
8619 '``llvm.lifetime.end``' Intrinsic
8620 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8627 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8632 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8638 The first argument is a constant integer representing the size of the
8639 object, or -1 if it is variable sized. The second argument is a pointer
8645 This intrinsic indicates that after this point in the code, the value of
8646 the memory pointed to by ``ptr`` is dead. This means that it is known to
8647 never be used and has an undefined value. Any stores into the memory
8648 object following this intrinsic may be removed as dead.
8650 '``llvm.invariant.start``' Intrinsic
8651 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8658 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8663 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8664 a memory object will not change.
8669 The first argument is a constant integer representing the size of the
8670 object, or -1 if it is variable sized. The second argument is a pointer
8676 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8677 the return value, the referenced memory location is constant and
8680 '``llvm.invariant.end``' Intrinsic
8681 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8688 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8693 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8694 memory object are mutable.
8699 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8700 The second argument is a constant integer representing the size of the
8701 object, or -1 if it is variable sized and the third argument is a
8702 pointer to the object.
8707 This intrinsic indicates that the memory is mutable again.
8712 This class of intrinsics is designed to be generic and has no specific
8715 '``llvm.var.annotation``' Intrinsic
8716 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8723 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8728 The '``llvm.var.annotation``' intrinsic.
8733 The first argument is a pointer to a value, the second is a pointer to a
8734 global string, the third is a pointer to a global string which is the
8735 source file name, and the last argument is the line number.
8740 This intrinsic allows annotation of local variables with arbitrary
8741 strings. This can be useful for special purpose optimizations that want
8742 to look for these annotations. These have no other defined use; they are
8743 ignored by code generation and optimization.
8745 '``llvm.ptr.annotation.*``' Intrinsic
8746 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8751 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8752 pointer to an integer of any width. *NOTE* you must specify an address space for
8753 the pointer. The identifier for the default address space is the integer
8758 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8759 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8760 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8761 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8762 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8767 The '``llvm.ptr.annotation``' intrinsic.
8772 The first argument is a pointer to an integer value of arbitrary bitwidth
8773 (result of some expression), the second is a pointer to a global string, the
8774 third is a pointer to a global string which is the source file name, and the
8775 last argument is the line number. It returns the value of the first argument.
8780 This intrinsic allows annotation of a pointer to an integer with arbitrary
8781 strings. This can be useful for special purpose optimizations that want to look
8782 for these annotations. These have no other defined use; they are ignored by code
8783 generation and optimization.
8785 '``llvm.annotation.*``' Intrinsic
8786 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8791 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8792 any integer bit width.
8796 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8797 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8798 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8799 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8800 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8805 The '``llvm.annotation``' intrinsic.
8810 The first argument is an integer value (result of some expression), the
8811 second is a pointer to a global string, the third is a pointer to a
8812 global string which is the source file name, and the last argument is
8813 the line number. It returns the value of the first argument.
8818 This intrinsic allows annotations to be put on arbitrary expressions
8819 with arbitrary strings. This can be useful for special purpose
8820 optimizations that want to look for these annotations. These have no
8821 other defined use; they are ignored by code generation and optimization.
8823 '``llvm.trap``' Intrinsic
8824 ^^^^^^^^^^^^^^^^^^^^^^^^^
8831 declare void @llvm.trap() noreturn nounwind
8836 The '``llvm.trap``' intrinsic.
8846 This intrinsic is lowered to the target dependent trap instruction. If
8847 the target does not have a trap instruction, this intrinsic will be
8848 lowered to a call of the ``abort()`` function.
8850 '``llvm.debugtrap``' Intrinsic
8851 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8858 declare void @llvm.debugtrap() nounwind
8863 The '``llvm.debugtrap``' intrinsic.
8873 This intrinsic is lowered to code which is intended to cause an
8874 execution trap with the intention of requesting the attention of a
8877 '``llvm.stackprotector``' Intrinsic
8878 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8885 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8890 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8891 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8892 is placed on the stack before local variables.
8897 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8898 The first argument is the value loaded from the stack guard
8899 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8900 enough space to hold the value of the guard.
8905 This intrinsic causes the prologue/epilogue inserter to force the position of
8906 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8907 to ensure that if a local variable on the stack is overwritten, it will destroy
8908 the value of the guard. When the function exits, the guard on the stack is
8909 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8910 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8911 calling the ``__stack_chk_fail()`` function.
8913 '``llvm.stackprotectorcheck``' Intrinsic
8914 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8921 declare void @llvm.stackprotectorcheck(i8** <guard>)
8926 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8927 created stack protector and if they are not equal calls the
8928 ``__stack_chk_fail()`` function.
8933 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8934 the variable ``@__stack_chk_guard``.
8939 This intrinsic is provided to perform the stack protector check by comparing
8940 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8941 values do not match call the ``__stack_chk_fail()`` function.
8943 The reason to provide this as an IR level intrinsic instead of implementing it
8944 via other IR operations is that in order to perform this operation at the IR
8945 level without an intrinsic, one would need to create additional basic blocks to
8946 handle the success/failure cases. This makes it difficult to stop the stack
8947 protector check from disrupting sibling tail calls in Codegen. With this
8948 intrinsic, we are able to generate the stack protector basic blocks late in
8949 codegen after the tail call decision has occurred.
8951 '``llvm.objectsize``' Intrinsic
8952 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8959 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8960 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8965 The ``llvm.objectsize`` intrinsic is designed to provide information to
8966 the optimizers to determine at compile time whether a) an operation
8967 (like memcpy) will overflow a buffer that corresponds to an object, or
8968 b) that a runtime check for overflow isn't necessary. An object in this
8969 context means an allocation of a specific class, structure, array, or
8975 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8976 argument is a pointer to or into the ``object``. The second argument is
8977 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8978 or -1 (if false) when the object size is unknown. The second argument
8979 only accepts constants.
8984 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8985 the size of the object concerned. If the size cannot be determined at
8986 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8987 on the ``min`` argument).
8989 '``llvm.expect``' Intrinsic
8990 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8995 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
9000 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
9001 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9002 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9007 The ``llvm.expect`` intrinsic provides information about expected (the
9008 most probable) value of ``val``, which can be used by optimizers.
9013 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
9014 a value. The second argument is an expected value, this needs to be a
9015 constant value, variables are not allowed.
9020 This intrinsic is lowered to the ``val``.
9022 '``llvm.donothing``' Intrinsic
9023 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9030 declare void @llvm.donothing() nounwind readnone
9035 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
9036 only intrinsic that can be called with an invoke instruction.
9046 This intrinsic does nothing, and it's removed by optimizers and ignored
9049 Stack Map Intrinsics
9050 --------------------
9052 LLVM provides experimental intrinsics to support runtime patching
9053 mechanisms commonly desired in dynamic language JITs. These intrinsics
9054 are described in :doc:`StackMaps`.