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 Aliases are not allowed to point to aliases with linkages that can be
685 overridden. Since they are only a second name, the possibility of the
686 intermediate alias being overridden cannot be represented in an object file.
688 .. _namedmetadatastructure:
693 Named metadata is a collection of metadata. :ref:`Metadata
694 nodes <metadata>` (but not metadata strings) are the only valid
695 operands for a named metadata.
699 ; Some unnamed metadata nodes, which are referenced by the named metadata.
700 !0 = metadata !{metadata !"zero"}
701 !1 = metadata !{metadata !"one"}
702 !2 = metadata !{metadata !"two"}
704 !name = !{!0, !1, !2}
711 The return type and each parameter of a function type may have a set of
712 *parameter attributes* associated with them. Parameter attributes are
713 used to communicate additional information about the result or
714 parameters of a function. Parameter attributes are considered to be part
715 of the function, not of the function type, so functions with different
716 parameter attributes can have the same function type.
718 Parameter attributes are simple keywords that follow the type specified.
719 If multiple parameter attributes are needed, they are space separated.
724 declare i32 @printf(i8* noalias nocapture, ...)
725 declare i32 @atoi(i8 zeroext)
726 declare signext i8 @returns_signed_char()
728 Note that any attributes for the function result (``nounwind``,
729 ``readonly``) come immediately after the argument list.
731 Currently, only the following parameter attributes are defined:
734 This indicates to the code generator that the parameter or return
735 value should be zero-extended to the extent required by the target's
736 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
737 the caller (for a parameter) or the callee (for a return value).
739 This indicates to the code generator that the parameter or return
740 value should be sign-extended to the extent required by the target's
741 ABI (which is usually 32-bits) by the caller (for a parameter) or
742 the callee (for a return value).
744 This indicates that this parameter or return value should be treated
745 in a special target-dependent fashion during while emitting code for
746 a function call or return (usually, by putting it in a register as
747 opposed to memory, though some targets use it to distinguish between
748 two different kinds of registers). Use of this attribute is
751 This indicates that the pointer parameter should really be passed by
752 value to the function. The attribute implies that a hidden copy of
753 the pointee is made between the caller and the callee, so the callee
754 is unable to modify the value in the caller. This attribute is only
755 valid on LLVM pointer arguments. It is generally used to pass
756 structs and arrays by value, but is also valid on pointers to
757 scalars. The copy is considered to belong to the caller not the
758 callee (for example, ``readonly`` functions should not write to
759 ``byval`` parameters). This is not a valid attribute for return
762 The byval attribute also supports specifying an alignment with the
763 align attribute. It indicates the alignment of the stack slot to
764 form and the known alignment of the pointer specified to the call
765 site. If the alignment is not specified, then the code generator
766 makes a target-specific assumption.
772 .. Warning:: This feature is unstable and not fully implemented.
774 The ``inalloca`` argument attribute allows the caller to take the
775 address of outgoing stack arguments. An ``inalloca`` argument must
776 be a pointer to stack memory produced by an ``alloca`` instruction.
777 The alloca, or argument allocation, must also be tagged with the
778 inalloca keyword. Only the past argument may have the ``inalloca``
779 attribute, and that argument is guaranteed to be passed in memory.
781 An argument allocation may be used by a call at most once because
782 the call may deallocate it. The ``inalloca`` attribute cannot be
783 used in conjunction with other attributes that affect argument
784 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
785 ``inalloca`` attribute also disables LLVM's implicit lowering of
786 large aggregate return values, which means that frontend authors
787 must lower them with ``sret`` pointers.
789 When the call site is reached, the argument allocation must have
790 been the most recent stack allocation that is still live, or the
791 results are undefined. It is possible to allocate additional stack
792 space after an argument allocation and before its call site, but it
793 must be cleared off with :ref:`llvm.stackrestore
796 See :doc:`InAlloca` for more information on how to use this
800 This indicates that the pointer parameter specifies the address of a
801 structure that is the return value of the function in the source
802 program. This pointer must be guaranteed by the caller to be valid:
803 loads and stores to the structure may be assumed by the callee
804 not to trap and to be properly aligned. This may only be applied to
805 the first parameter. This is not a valid attribute for return
808 This indicates that pointer values :ref:`based <pointeraliasing>` on
809 the argument or return value do not alias pointer values which are
810 not *based* on it, ignoring certain "irrelevant" dependencies. For a
811 call to the parent function, dependencies between memory references
812 from before or after the call and from those during the call are
813 "irrelevant" to the ``noalias`` keyword for the arguments and return
814 value used in that call. The caller shares the responsibility with
815 the callee for ensuring that these requirements are met. For further
816 details, please see the discussion of the NoAlias response in `alias
817 analysis <AliasAnalysis.html#MustMayNo>`_.
819 Note that this definition of ``noalias`` is intentionally similar
820 to the definition of ``restrict`` in C99 for function arguments,
821 though it is slightly weaker.
823 For function return values, C99's ``restrict`` is not meaningful,
824 while LLVM's ``noalias`` is.
826 This indicates that the callee does not make any copies of the
827 pointer that outlive the callee itself. This is not a valid
828 attribute for return values.
833 This indicates that the pointer parameter can be excised using the
834 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
835 attribute for return values and can only be applied to one parameter.
838 This indicates that the function always returns the argument as its return
839 value. This is an optimization hint to the code generator when generating
840 the caller, allowing tail call optimization and omission of register saves
841 and restores in some cases; it is not checked or enforced when generating
842 the callee. The parameter and the function return type must be valid
843 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
844 valid attribute for return values and can only be applied to one parameter.
848 Garbage Collector Names
849 -----------------------
851 Each function may specify a garbage collector name, which is simply a
856 define void @f() gc "name" { ... }
858 The compiler declares the supported values of *name*. Specifying a
859 collector which will cause the compiler to alter its output in order to
860 support the named garbage collection algorithm.
867 Prefix data is data associated with a function which the code generator
868 will emit immediately before the function body. The purpose of this feature
869 is to allow frontends to associate language-specific runtime metadata with
870 specific functions and make it available through the function pointer while
871 still allowing the function pointer to be called. To access the data for a
872 given function, a program may bitcast the function pointer to a pointer to
873 the constant's type. This implies that the IR symbol points to the start
876 To maintain the semantics of ordinary function calls, the prefix data must
877 have a particular format. Specifically, it must begin with a sequence of
878 bytes which decode to a sequence of machine instructions, valid for the
879 module's target, which transfer control to the point immediately succeeding
880 the prefix data, without performing any other visible action. This allows
881 the inliner and other passes to reason about the semantics of the function
882 definition without needing to reason about the prefix data. Obviously this
883 makes the format of the prefix data highly target dependent.
885 Prefix data is laid out as if it were an initializer for a global variable
886 of the prefix data's type. No padding is automatically placed between the
887 prefix data and the function body. If padding is required, it must be part
890 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
891 which encodes the ``nop`` instruction:
895 define void @f() prefix i8 144 { ... }
897 Generally prefix data can be formed by encoding a relative branch instruction
898 which skips the metadata, as in this example of valid prefix data for the
899 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
903 %0 = type <{ i8, i8, i8* }>
905 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
907 A function may have prefix data but no body. This has similar semantics
908 to the ``available_externally`` linkage in that the data may be used by the
909 optimizers but will not be emitted in the object file.
916 Attribute groups are groups of attributes that are referenced by objects within
917 the IR. They are important for keeping ``.ll`` files readable, because a lot of
918 functions will use the same set of attributes. In the degenerative case of a
919 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
920 group will capture the important command line flags used to build that file.
922 An attribute group is a module-level object. To use an attribute group, an
923 object references the attribute group's ID (e.g. ``#37``). An object may refer
924 to more than one attribute group. In that situation, the attributes from the
925 different groups are merged.
927 Here is an example of attribute groups for a function that should always be
928 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
932 ; Target-independent attributes:
933 attributes #0 = { alwaysinline alignstack=4 }
935 ; Target-dependent attributes:
936 attributes #1 = { "no-sse" }
938 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
939 define void @f() #0 #1 { ... }
946 Function attributes are set to communicate additional information about
947 a function. Function attributes are considered to be part of the
948 function, not of the function type, so functions with different function
949 attributes can have the same function type.
951 Function attributes are simple keywords that follow the type specified.
952 If multiple attributes are needed, they are space separated. For
957 define void @f() noinline { ... }
958 define void @f() alwaysinline { ... }
959 define void @f() alwaysinline optsize { ... }
960 define void @f() optsize { ... }
963 This attribute indicates that, when emitting the prologue and
964 epilogue, the backend should forcibly align the stack pointer.
965 Specify the desired alignment, which must be a power of two, in
968 This attribute indicates that the inliner should attempt to inline
969 this function into callers whenever possible, ignoring any active
970 inlining size threshold for this caller.
972 This indicates that the callee function at a call site should be
973 recognized as a built-in function, even though the function's declaration
974 uses the ``nobuiltin`` attribute. This is only valid at call sites for
975 direct calls to functions which are declared with the ``nobuiltin``
978 This attribute indicates that this function is rarely called. When
979 computing edge weights, basic blocks post-dominated by a cold
980 function call are also considered to be cold; and, thus, given low
983 This attribute indicates that the source code contained a hint that
984 inlining this function is desirable (such as the "inline" keyword in
985 C/C++). It is just a hint; it imposes no requirements on the
988 This attribute suggests that optimization passes and code generator
989 passes make choices that keep the code size of this function as small
990 as possible and perform optimizations that may sacrifice runtime
991 performance in order to minimize the size of the generated code.
993 This attribute disables prologue / epilogue emission for the
994 function. This can have very system-specific consequences.
996 This indicates that the callee function at a call site is not recognized as
997 a built-in function. LLVM will retain the original call and not replace it
998 with equivalent code based on the semantics of the built-in function, unless
999 the call site uses the ``builtin`` attribute. This is valid at call sites
1000 and on function declarations and definitions.
1002 This attribute indicates that calls to the function cannot be
1003 duplicated. A call to a ``noduplicate`` function may be moved
1004 within its parent function, but may not be duplicated within
1005 its parent function.
1007 A function containing a ``noduplicate`` call may still
1008 be an inlining candidate, provided that the call is not
1009 duplicated by inlining. That implies that the function has
1010 internal linkage and only has one call site, so the original
1011 call is dead after inlining.
1013 This attributes disables implicit floating point instructions.
1015 This attribute indicates that the inliner should never inline this
1016 function in any situation. This attribute may not be used together
1017 with the ``alwaysinline`` attribute.
1019 This attribute suppresses lazy symbol binding for the function. This
1020 may make calls to the function faster, at the cost of extra program
1021 startup time if the function is not called during program startup.
1023 This attribute indicates that the code generator should not use a
1024 red zone, even if the target-specific ABI normally permits it.
1026 This function attribute indicates that the function never returns
1027 normally. This produces undefined behavior at runtime if the
1028 function ever does dynamically return.
1030 This function attribute indicates that the function never returns
1031 with an unwind or exceptional control flow. If the function does
1032 unwind, its runtime behavior is undefined.
1034 This function attribute indicates that the function is not optimized
1035 by any optimization or code generator passes with the
1036 exception of interprocedural optimization passes.
1037 This attribute cannot be used together with the ``alwaysinline``
1038 attribute; this attribute is also incompatible
1039 with the ``minsize`` attribute and the ``optsize`` attribute.
1041 This attribute requires the ``noinline`` attribute to be specified on
1042 the function as well, so the function is never inlined into any caller.
1043 Only functions with the ``alwaysinline`` attribute are valid
1044 candidates for inlining into the body of this function.
1046 This attribute suggests that optimization passes and code generator
1047 passes make choices that keep the code size of this function low,
1048 and otherwise do optimizations specifically to reduce code size as
1049 long as they do not significantly impact runtime performance.
1051 On a function, this attribute indicates that the function computes its
1052 result (or decides to unwind an exception) based strictly on its arguments,
1053 without dereferencing any pointer arguments or otherwise accessing
1054 any mutable state (e.g. memory, control registers, etc) visible to
1055 caller functions. It does not write through any pointer arguments
1056 (including ``byval`` arguments) and never changes any state visible
1057 to callers. This means that it cannot unwind exceptions by calling
1058 the ``C++`` exception throwing methods.
1060 On an argument, this attribute indicates that the function does not
1061 dereference that pointer argument, even though it may read or write the
1062 memory that the pointer points to if accessed through other pointers.
1064 On a function, this attribute indicates that the function does not write
1065 through any pointer arguments (including ``byval`` arguments) or otherwise
1066 modify any state (e.g. memory, control registers, etc) visible to
1067 caller functions. It may dereference pointer arguments and read
1068 state that may be set in the caller. A readonly function always
1069 returns the same value (or unwinds an exception identically) when
1070 called with the same set of arguments and global state. It cannot
1071 unwind an exception by calling the ``C++`` exception throwing
1074 On an argument, this attribute indicates that the function does not write
1075 through this pointer argument, even though it may write to the memory that
1076 the pointer points to.
1078 This attribute indicates that this function can return twice. The C
1079 ``setjmp`` is an example of such a function. The compiler disables
1080 some optimizations (like tail calls) in the caller of these
1082 ``sanitize_address``
1083 This attribute indicates that AddressSanitizer checks
1084 (dynamic address safety analysis) are enabled for this function.
1086 This attribute indicates that MemorySanitizer checks (dynamic detection
1087 of accesses to uninitialized memory) are enabled for this function.
1089 This attribute indicates that ThreadSanitizer checks
1090 (dynamic thread safety analysis) are enabled for this function.
1092 This attribute indicates that the function should emit a stack
1093 smashing protector. It is in the form of a "canary" --- a random value
1094 placed on the stack before the local variables that's checked upon
1095 return from the function to see if it has been overwritten. A
1096 heuristic is used to determine if a function needs stack protectors
1097 or not. The heuristic used will enable protectors for functions with:
1099 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1100 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1101 - Calls to alloca() with variable sizes or constant sizes greater than
1102 ``ssp-buffer-size``.
1104 Variables that are identified as requiring a protector will be arranged
1105 on the stack such that they are adjacent to the stack protector guard.
1107 If a function that has an ``ssp`` attribute is inlined into a
1108 function that doesn't have an ``ssp`` attribute, then the resulting
1109 function will have an ``ssp`` attribute.
1111 This attribute indicates that the function should *always* emit a
1112 stack smashing protector. This overrides the ``ssp`` function
1115 Variables that are identified as requiring a protector will be arranged
1116 on the stack such that they are adjacent to the stack protector guard.
1117 The specific layout rules are:
1119 #. Large arrays and structures containing large arrays
1120 (``>= ssp-buffer-size``) are closest to the stack protector.
1121 #. Small arrays and structures containing small arrays
1122 (``< ssp-buffer-size``) are 2nd closest to the protector.
1123 #. Variables that have had their address taken are 3rd closest to the
1126 If a function that has an ``sspreq`` attribute is inlined into a
1127 function that doesn't have an ``sspreq`` attribute or which has an
1128 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1129 an ``sspreq`` attribute.
1131 This attribute indicates that the function should emit a stack smashing
1132 protector. This attribute causes a strong heuristic to be used when
1133 determining if a function needs stack protectors. The strong heuristic
1134 will enable protectors for functions with:
1136 - Arrays of any size and type
1137 - Aggregates containing an array of any size and type.
1138 - Calls to alloca().
1139 - Local variables that have had their address taken.
1141 Variables that are identified as requiring a protector will be arranged
1142 on the stack such that they are adjacent to the stack protector guard.
1143 The specific layout rules are:
1145 #. Large arrays and structures containing large arrays
1146 (``>= ssp-buffer-size``) are closest to the stack protector.
1147 #. Small arrays and structures containing small arrays
1148 (``< ssp-buffer-size``) are 2nd closest to the protector.
1149 #. Variables that have had their address taken are 3rd closest to the
1152 This overrides the ``ssp`` function attribute.
1154 If a function that has an ``sspstrong`` attribute is inlined into a
1155 function that doesn't have an ``sspstrong`` attribute, then the
1156 resulting function will have an ``sspstrong`` attribute.
1158 This attribute indicates that the ABI being targeted requires that
1159 an unwind table entry be produce for this function even if we can
1160 show that no exceptions passes by it. This is normally the case for
1161 the ELF x86-64 abi, but it can be disabled for some compilation
1166 Module-Level Inline Assembly
1167 ----------------------------
1169 Modules may contain "module-level inline asm" blocks, which corresponds
1170 to the GCC "file scope inline asm" blocks. These blocks are internally
1171 concatenated by LLVM and treated as a single unit, but may be separated
1172 in the ``.ll`` file if desired. The syntax is very simple:
1174 .. code-block:: llvm
1176 module asm "inline asm code goes here"
1177 module asm "more can go here"
1179 The strings can contain any character by escaping non-printable
1180 characters. The escape sequence used is simply "\\xx" where "xx" is the
1181 two digit hex code for the number.
1183 The inline asm code is simply printed to the machine code .s file when
1184 assembly code is generated.
1186 .. _langref_datalayout:
1191 A module may specify a target specific data layout string that specifies
1192 how data is to be laid out in memory. The syntax for the data layout is
1195 .. code-block:: llvm
1197 target datalayout = "layout specification"
1199 The *layout specification* consists of a list of specifications
1200 separated by the minus sign character ('-'). Each specification starts
1201 with a letter and may include other information after the letter to
1202 define some aspect of the data layout. The specifications accepted are
1206 Specifies that the target lays out data in big-endian form. That is,
1207 the bits with the most significance have the lowest address
1210 Specifies that the target lays out data in little-endian form. That
1211 is, the bits with the least significance have the lowest address
1214 Specifies the natural alignment of the stack in bits. Alignment
1215 promotion of stack variables is limited to the natural stack
1216 alignment to avoid dynamic stack realignment. The stack alignment
1217 must be a multiple of 8-bits. If omitted, the natural stack
1218 alignment defaults to "unspecified", which does not prevent any
1219 alignment promotions.
1220 ``p[n]:<size>:<abi>:<pref>``
1221 This specifies the *size* of a pointer and its ``<abi>`` and
1222 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1223 bits. The address space, ``n`` is optional, and if not specified,
1224 denotes the default address space 0. The value of ``n`` must be
1225 in the range [1,2^23).
1226 ``i<size>:<abi>:<pref>``
1227 This specifies the alignment for an integer type of a given bit
1228 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1229 ``v<size>:<abi>:<pref>``
1230 This specifies the alignment for a vector type of a given bit
1232 ``f<size>:<abi>:<pref>``
1233 This specifies the alignment for a floating point type of a given bit
1234 ``<size>``. Only values of ``<size>`` that are supported by the target
1235 will work. 32 (float) and 64 (double) are supported on all targets; 80
1236 or 128 (different flavors of long double) are also supported on some
1239 This specifies the alignment for an object of aggregate type.
1241 If present, specifies that llvm names are mangled in the output. The
1244 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1245 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1246 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1247 symbols get a ``_`` prefix.
1248 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1249 functions also get a suffix based on the frame size.
1250 ``n<size1>:<size2>:<size3>...``
1251 This specifies a set of native integer widths for the target CPU in
1252 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1253 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1254 this set are considered to support most general arithmetic operations
1257 On every specification that takes a ``<abi>:<pref>``, specifying the
1258 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1259 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1261 When constructing the data layout for a given target, LLVM starts with a
1262 default set of specifications which are then (possibly) overridden by
1263 the specifications in the ``datalayout`` keyword. The default
1264 specifications are given in this list:
1266 - ``E`` - big endian
1267 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1268 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1269 same as the default address space.
1270 - ``S0`` - natural stack alignment is unspecified
1271 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1272 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1273 - ``i16:16:16`` - i16 is 16-bit aligned
1274 - ``i32:32:32`` - i32 is 32-bit aligned
1275 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1276 alignment of 64-bits
1277 - ``f16:16:16`` - half is 16-bit aligned
1278 - ``f32:32:32`` - float is 32-bit aligned
1279 - ``f64:64:64`` - double is 64-bit aligned
1280 - ``f128:128:128`` - quad is 128-bit aligned
1281 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1282 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1283 - ``a:0:64`` - aggregates are 64-bit aligned
1285 When LLVM is determining the alignment for a given type, it uses the
1288 #. If the type sought is an exact match for one of the specifications,
1289 that specification is used.
1290 #. If no match is found, and the type sought is an integer type, then
1291 the smallest integer type that is larger than the bitwidth of the
1292 sought type is used. If none of the specifications are larger than
1293 the bitwidth then the largest integer type is used. For example,
1294 given the default specifications above, the i7 type will use the
1295 alignment of i8 (next largest) while both i65 and i256 will use the
1296 alignment of i64 (largest specified).
1297 #. If no match is found, and the type sought is a vector type, then the
1298 largest vector type that is smaller than the sought vector type will
1299 be used as a fall back. This happens because <128 x double> can be
1300 implemented in terms of 64 <2 x double>, for example.
1302 The function of the data layout string may not be what you expect.
1303 Notably, this is not a specification from the frontend of what alignment
1304 the code generator should use.
1306 Instead, if specified, the target data layout is required to match what
1307 the ultimate *code generator* expects. This string is used by the
1308 mid-level optimizers to improve code, and this only works if it matches
1309 what the ultimate code generator uses. If you would like to generate IR
1310 that does not embed this target-specific detail into the IR, then you
1311 don't have to specify the string. This will disable some optimizations
1312 that require precise layout information, but this also prevents those
1313 optimizations from introducing target specificity into the IR.
1320 A module may specify a target triple string that describes the target
1321 host. The syntax for the target triple is simply:
1323 .. code-block:: llvm
1325 target triple = "x86_64-apple-macosx10.7.0"
1327 The *target triple* string consists of a series of identifiers delimited
1328 by the minus sign character ('-'). The canonical forms are:
1332 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1333 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1335 This information is passed along to the backend so that it generates
1336 code for the proper architecture. It's possible to override this on the
1337 command line with the ``-mtriple`` command line option.
1339 .. _pointeraliasing:
1341 Pointer Aliasing Rules
1342 ----------------------
1344 Any memory access must be done through a pointer value associated with
1345 an address range of the memory access, otherwise the behavior is
1346 undefined. Pointer values are associated with address ranges according
1347 to the following rules:
1349 - A pointer value is associated with the addresses associated with any
1350 value it is *based* on.
1351 - An address of a global variable is associated with the address range
1352 of the variable's storage.
1353 - The result value of an allocation instruction is associated with the
1354 address range of the allocated storage.
1355 - A null pointer in the default address-space is associated with no
1357 - An integer constant other than zero or a pointer value returned from
1358 a function not defined within LLVM may be associated with address
1359 ranges allocated through mechanisms other than those provided by
1360 LLVM. Such ranges shall not overlap with any ranges of addresses
1361 allocated by mechanisms provided by LLVM.
1363 A pointer value is *based* on another pointer value according to the
1366 - A pointer value formed from a ``getelementptr`` operation is *based*
1367 on the first operand of the ``getelementptr``.
1368 - The result value of a ``bitcast`` is *based* on the operand of the
1370 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1371 values that contribute (directly or indirectly) to the computation of
1372 the pointer's value.
1373 - The "*based* on" relationship is transitive.
1375 Note that this definition of *"based"* is intentionally similar to the
1376 definition of *"based"* in C99, though it is slightly weaker.
1378 LLVM IR does not associate types with memory. The result type of a
1379 ``load`` merely indicates the size and alignment of the memory from
1380 which to load, as well as the interpretation of the value. The first
1381 operand type of a ``store`` similarly only indicates the size and
1382 alignment of the store.
1384 Consequently, type-based alias analysis, aka TBAA, aka
1385 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1386 :ref:`Metadata <metadata>` may be used to encode additional information
1387 which specialized optimization passes may use to implement type-based
1392 Volatile Memory Accesses
1393 ------------------------
1395 Certain memory accesses, such as :ref:`load <i_load>`'s,
1396 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1397 marked ``volatile``. The optimizers must not change the number of
1398 volatile operations or change their order of execution relative to other
1399 volatile operations. The optimizers *may* change the order of volatile
1400 operations relative to non-volatile operations. This is not Java's
1401 "volatile" and has no cross-thread synchronization behavior.
1403 IR-level volatile loads and stores cannot safely be optimized into
1404 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1405 flagged volatile. Likewise, the backend should never split or merge
1406 target-legal volatile load/store instructions.
1408 .. admonition:: Rationale
1410 Platforms may rely on volatile loads and stores of natively supported
1411 data width to be executed as single instruction. For example, in C
1412 this holds for an l-value of volatile primitive type with native
1413 hardware support, but not necessarily for aggregate types. The
1414 frontend upholds these expectations, which are intentionally
1415 unspecified in the IR. The rules above ensure that IR transformation
1416 do not violate the frontend's contract with the language.
1420 Memory Model for Concurrent Operations
1421 --------------------------------------
1423 The LLVM IR does not define any way to start parallel threads of
1424 execution or to register signal handlers. Nonetheless, there are
1425 platform-specific ways to create them, and we define LLVM IR's behavior
1426 in their presence. This model is inspired by the C++0x memory model.
1428 For a more informal introduction to this model, see the :doc:`Atomics`.
1430 We define a *happens-before* partial order as the least partial order
1433 - Is a superset of single-thread program order, and
1434 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1435 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1436 techniques, like pthread locks, thread creation, thread joining,
1437 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1438 Constraints <ordering>`).
1440 Note that program order does not introduce *happens-before* edges
1441 between a thread and signals executing inside that thread.
1443 Every (defined) read operation (load instructions, memcpy, atomic
1444 loads/read-modify-writes, etc.) R reads a series of bytes written by
1445 (defined) write operations (store instructions, atomic
1446 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1447 section, initialized globals are considered to have a write of the
1448 initializer which is atomic and happens before any other read or write
1449 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1450 may see any write to the same byte, except:
1452 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1453 write\ :sub:`2` happens before R\ :sub:`byte`, then
1454 R\ :sub:`byte` does not see write\ :sub:`1`.
1455 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1456 R\ :sub:`byte` does not see write\ :sub:`3`.
1458 Given that definition, R\ :sub:`byte` is defined as follows:
1460 - If R is volatile, the result is target-dependent. (Volatile is
1461 supposed to give guarantees which can support ``sig_atomic_t`` in
1462 C/C++, and may be used for accesses to addresses which do not behave
1463 like normal memory. It does not generally provide cross-thread
1465 - Otherwise, if there is no write to the same byte that happens before
1466 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1467 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1468 R\ :sub:`byte` returns the value written by that write.
1469 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1470 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1471 Memory Ordering Constraints <ordering>` section for additional
1472 constraints on how the choice is made.
1473 - Otherwise R\ :sub:`byte` returns ``undef``.
1475 R returns the value composed of the series of bytes it read. This
1476 implies that some bytes within the value may be ``undef`` **without**
1477 the entire value being ``undef``. Note that this only defines the
1478 semantics of the operation; it doesn't mean that targets will emit more
1479 than one instruction to read the series of bytes.
1481 Note that in cases where none of the atomic intrinsics are used, this
1482 model places only one restriction on IR transformations on top of what
1483 is required for single-threaded execution: introducing a store to a byte
1484 which might not otherwise be stored is not allowed in general.
1485 (Specifically, in the case where another thread might write to and read
1486 from an address, introducing a store can change a load that may see
1487 exactly one write into a load that may see multiple writes.)
1491 Atomic Memory Ordering Constraints
1492 ----------------------------------
1494 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1495 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1496 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1497 ordering parameters that determine which other atomic instructions on
1498 the same address they *synchronize with*. These semantics are borrowed
1499 from Java and C++0x, but are somewhat more colloquial. If these
1500 descriptions aren't precise enough, check those specs (see spec
1501 references in the :doc:`atomics guide <Atomics>`).
1502 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1503 differently since they don't take an address. See that instruction's
1504 documentation for details.
1506 For a simpler introduction to the ordering constraints, see the
1510 The set of values that can be read is governed by the happens-before
1511 partial order. A value cannot be read unless some operation wrote
1512 it. This is intended to provide a guarantee strong enough to model
1513 Java's non-volatile shared variables. This ordering cannot be
1514 specified for read-modify-write operations; it is not strong enough
1515 to make them atomic in any interesting way.
1517 In addition to the guarantees of ``unordered``, there is a single
1518 total order for modifications by ``monotonic`` operations on each
1519 address. All modification orders must be compatible with the
1520 happens-before order. There is no guarantee that the modification
1521 orders can be combined to a global total order for the whole program
1522 (and this often will not be possible). The read in an atomic
1523 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1524 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1525 order immediately before the value it writes. If one atomic read
1526 happens before another atomic read of the same address, the later
1527 read must see the same value or a later value in the address's
1528 modification order. This disallows reordering of ``monotonic`` (or
1529 stronger) operations on the same address. If an address is written
1530 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1531 read that address repeatedly, the other threads must eventually see
1532 the write. This corresponds to the C++0x/C1x
1533 ``memory_order_relaxed``.
1535 In addition to the guarantees of ``monotonic``, a
1536 *synchronizes-with* edge may be formed with a ``release`` operation.
1537 This is intended to model C++'s ``memory_order_acquire``.
1539 In addition to the guarantees of ``monotonic``, if this operation
1540 writes a value which is subsequently read by an ``acquire``
1541 operation, it *synchronizes-with* that operation. (This isn't a
1542 complete description; see the C++0x definition of a release
1543 sequence.) This corresponds to the C++0x/C1x
1544 ``memory_order_release``.
1545 ``acq_rel`` (acquire+release)
1546 Acts as both an ``acquire`` and ``release`` operation on its
1547 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1548 ``seq_cst`` (sequentially consistent)
1549 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1550 operation which only reads, ``release`` for an operation which only
1551 writes), there is a global total order on all
1552 sequentially-consistent operations on all addresses, which is
1553 consistent with the *happens-before* partial order and with the
1554 modification orders of all the affected addresses. Each
1555 sequentially-consistent read sees the last preceding write to the
1556 same address in this global order. This corresponds to the C++0x/C1x
1557 ``memory_order_seq_cst`` and Java volatile.
1561 If an atomic operation is marked ``singlethread``, it only *synchronizes
1562 with* or participates in modification and seq\_cst total orderings with
1563 other operations running in the same thread (for example, in signal
1571 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1572 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1573 :ref:`frem <i_frem>`) have the following flags that can set to enable
1574 otherwise unsafe floating point operations
1577 No NaNs - Allow optimizations to assume the arguments and result are not
1578 NaN. Such optimizations are required to retain defined behavior over
1579 NaNs, but the value of the result is undefined.
1582 No Infs - Allow optimizations to assume the arguments and result are not
1583 +/-Inf. Such optimizations are required to retain defined behavior over
1584 +/-Inf, but the value of the result is undefined.
1587 No Signed Zeros - Allow optimizations to treat the sign of a zero
1588 argument or result as insignificant.
1591 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1592 argument rather than perform division.
1595 Fast - Allow algebraically equivalent transformations that may
1596 dramatically change results in floating point (e.g. reassociate). This
1597 flag implies all the others.
1604 The LLVM type system is one of the most important features of the
1605 intermediate representation. Being typed enables a number of
1606 optimizations to be performed on the intermediate representation
1607 directly, without having to do extra analyses on the side before the
1608 transformation. A strong type system makes it easier to read the
1609 generated code and enables novel analyses and transformations that are
1610 not feasible to perform on normal three address code representations.
1620 The void type does not represent any value and has no size.
1638 The function type can be thought of as a function signature. It consists of a
1639 return type and a list of formal parameter types. The return type of a function
1640 type is a void type or first class type --- except for :ref:`label <t_label>`
1641 and :ref:`metadata <t_metadata>` types.
1647 <returntype> (<parameter list>)
1649 ...where '``<parameter list>``' is a comma-separated list of type
1650 specifiers. Optionally, the parameter list may include a type ``...``, which
1651 indicates that the function takes a variable number of arguments. Variable
1652 argument functions can access their arguments with the :ref:`variable argument
1653 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1654 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1658 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1659 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1660 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1661 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1662 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1663 | ``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. |
1664 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1665 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1666 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1673 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1674 Values of these types are the only ones which can be produced by
1682 These are the types that are valid in registers from CodeGen's perspective.
1691 The integer type is a very simple type that simply specifies an
1692 arbitrary bit width for the integer type desired. Any bit width from 1
1693 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1701 The number of bits the integer will occupy is specified by the ``N``
1707 +----------------+------------------------------------------------+
1708 | ``i1`` | a single-bit integer. |
1709 +----------------+------------------------------------------------+
1710 | ``i32`` | a 32-bit integer. |
1711 +----------------+------------------------------------------------+
1712 | ``i1942652`` | a really big integer of over 1 million bits. |
1713 +----------------+------------------------------------------------+
1717 Floating Point Types
1718 """"""""""""""""""""
1727 - 16-bit floating point value
1730 - 32-bit floating point value
1733 - 64-bit floating point value
1736 - 128-bit floating point value (112-bit mantissa)
1739 - 80-bit floating point value (X87)
1742 - 128-bit floating point value (two 64-bits)
1749 The x86_mmx type represents a value held in an MMX register on an x86
1750 machine. The operations allowed on it are quite limited: parameters and
1751 return values, load and store, and bitcast. User-specified MMX
1752 instructions are represented as intrinsic or asm calls with arguments
1753 and/or results of this type. There are no arrays, vectors or constants
1770 The pointer type is used to specify memory locations. Pointers are
1771 commonly used to reference objects in memory.
1773 Pointer types may have an optional address space attribute defining the
1774 numbered address space where the pointed-to object resides. The default
1775 address space is number zero. The semantics of non-zero address spaces
1776 are target-specific.
1778 Note that LLVM does not permit pointers to void (``void*``) nor does it
1779 permit pointers to labels (``label*``). Use ``i8*`` instead.
1789 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1790 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1791 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1792 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1793 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1794 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1795 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1804 A vector type is a simple derived type that represents a vector of
1805 elements. Vector types are used when multiple primitive data are
1806 operated in parallel using a single instruction (SIMD). A vector type
1807 requires a size (number of elements) and an underlying primitive data
1808 type. Vector types are considered :ref:`first class <t_firstclass>`.
1814 < <# elements> x <elementtype> >
1816 The number of elements is a constant integer value larger than 0;
1817 elementtype may be any integer or floating point type, or a pointer to
1818 these types. Vectors of size zero are not allowed.
1822 +-------------------+--------------------------------------------------+
1823 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1824 +-------------------+--------------------------------------------------+
1825 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1826 +-------------------+--------------------------------------------------+
1827 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1828 +-------------------+--------------------------------------------------+
1829 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1830 +-------------------+--------------------------------------------------+
1839 The label type represents code labels.
1854 The metadata type represents embedded metadata. No derived types may be
1855 created from metadata except for :ref:`function <t_function>` arguments.
1868 Aggregate Types are a subset of derived types that can contain multiple
1869 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1870 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1880 The array type is a very simple derived type that arranges elements
1881 sequentially in memory. The array type requires a size (number of
1882 elements) and an underlying data type.
1888 [<# elements> x <elementtype>]
1890 The number of elements is a constant integer value; ``elementtype`` may
1891 be any type with a size.
1895 +------------------+--------------------------------------+
1896 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1897 +------------------+--------------------------------------+
1898 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1899 +------------------+--------------------------------------+
1900 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1901 +------------------+--------------------------------------+
1903 Here are some examples of multidimensional arrays:
1905 +-----------------------------+----------------------------------------------------------+
1906 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1907 +-----------------------------+----------------------------------------------------------+
1908 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1909 +-----------------------------+----------------------------------------------------------+
1910 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1911 +-----------------------------+----------------------------------------------------------+
1913 There is no restriction on indexing beyond the end of the array implied
1914 by a static type (though there are restrictions on indexing beyond the
1915 bounds of an allocated object in some cases). This means that
1916 single-dimension 'variable sized array' addressing can be implemented in
1917 LLVM with a zero length array type. An implementation of 'pascal style
1918 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1928 The structure type is used to represent a collection of data members
1929 together in memory. The elements of a structure may be any type that has
1932 Structures in memory are accessed using '``load``' and '``store``' by
1933 getting a pointer to a field with the '``getelementptr``' instruction.
1934 Structures in registers are accessed using the '``extractvalue``' and
1935 '``insertvalue``' instructions.
1937 Structures may optionally be "packed" structures, which indicate that
1938 the alignment of the struct is one byte, and that there is no padding
1939 between the elements. In non-packed structs, padding between field types
1940 is inserted as defined by the DataLayout string in the module, which is
1941 required to match what the underlying code generator expects.
1943 Structures can either be "literal" or "identified". A literal structure
1944 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1945 identified types are always defined at the top level with a name.
1946 Literal types are uniqued by their contents and can never be recursive
1947 or opaque since there is no way to write one. Identified types can be
1948 recursive, can be opaqued, and are never uniqued.
1954 %T1 = type { <type list> } ; Identified normal struct type
1955 %T2 = type <{ <type list> }> ; Identified packed struct type
1959 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1960 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1961 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1962 | ``{ 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``. |
1963 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1964 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1965 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1969 Opaque Structure Types
1970 """"""""""""""""""""""
1974 Opaque structure types are used to represent named structure types that
1975 do not have a body specified. This corresponds (for example) to the C
1976 notion of a forward declared structure.
1987 +--------------+-------------------+
1988 | ``opaque`` | An opaque type. |
1989 +--------------+-------------------+
1994 LLVM has several different basic types of constants. This section
1995 describes them all and their syntax.
2000 **Boolean constants**
2001 The two strings '``true``' and '``false``' are both valid constants
2003 **Integer constants**
2004 Standard integers (such as '4') are constants of the
2005 :ref:`integer <t_integer>` type. Negative numbers may be used with
2007 **Floating point constants**
2008 Floating point constants use standard decimal notation (e.g.
2009 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2010 hexadecimal notation (see below). The assembler requires the exact
2011 decimal value of a floating-point constant. For example, the
2012 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2013 decimal in binary. Floating point constants must have a :ref:`floating
2014 point <t_floating>` type.
2015 **Null pointer constants**
2016 The identifier '``null``' is recognized as a null pointer constant
2017 and must be of :ref:`pointer type <t_pointer>`.
2019 The one non-intuitive notation for constants is the hexadecimal form of
2020 floating point constants. For example, the form
2021 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2022 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2023 constants are required (and the only time that they are generated by the
2024 disassembler) is when a floating point constant must be emitted but it
2025 cannot be represented as a decimal floating point number in a reasonable
2026 number of digits. For example, NaN's, infinities, and other special
2027 values are represented in their IEEE hexadecimal format so that assembly
2028 and disassembly do not cause any bits to change in the constants.
2030 When using the hexadecimal form, constants of types half, float, and
2031 double are represented using the 16-digit form shown above (which
2032 matches the IEEE754 representation for double); half and float values
2033 must, however, be exactly representable as IEEE 754 half and single
2034 precision, respectively. Hexadecimal format is always used for long
2035 double, and there are three forms of long double. The 80-bit format used
2036 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2037 128-bit format used by PowerPC (two adjacent doubles) is represented by
2038 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2039 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2040 will only work if they match the long double format on your target.
2041 The IEEE 16-bit format (half precision) is represented by ``0xH``
2042 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2043 (sign bit at the left).
2045 There are no constants of type x86_mmx.
2047 .. _complexconstants:
2052 Complex constants are a (potentially recursive) combination of simple
2053 constants and smaller complex constants.
2055 **Structure constants**
2056 Structure constants are represented with notation similar to
2057 structure type definitions (a comma separated list of elements,
2058 surrounded by braces (``{}``)). For example:
2059 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2060 "``@G = external global i32``". Structure constants must have
2061 :ref:`structure type <t_struct>`, and the number and types of elements
2062 must match those specified by the type.
2064 Array constants are represented with notation similar to array type
2065 definitions (a comma separated list of elements, surrounded by
2066 square brackets (``[]``)). For example:
2067 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2068 :ref:`array type <t_array>`, and the number and types of elements must
2069 match those specified by the type.
2070 **Vector constants**
2071 Vector constants are represented with notation similar to vector
2072 type definitions (a comma separated list of elements, surrounded by
2073 less-than/greater-than's (``<>``)). For example:
2074 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2075 must have :ref:`vector type <t_vector>`, and the number and types of
2076 elements must match those specified by the type.
2077 **Zero initialization**
2078 The string '``zeroinitializer``' can be used to zero initialize a
2079 value to zero of *any* type, including scalar and
2080 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2081 having to print large zero initializers (e.g. for large arrays) and
2082 is always exactly equivalent to using explicit zero initializers.
2084 A metadata node is a structure-like constant with :ref:`metadata
2085 type <t_metadata>`. For example:
2086 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2087 constants that are meant to be interpreted as part of the
2088 instruction stream, metadata is a place to attach additional
2089 information such as debug info.
2091 Global Variable and Function Addresses
2092 --------------------------------------
2094 The addresses of :ref:`global variables <globalvars>` and
2095 :ref:`functions <functionstructure>` are always implicitly valid
2096 (link-time) constants. These constants are explicitly referenced when
2097 the :ref:`identifier for the global <identifiers>` is used and always have
2098 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2101 .. code-block:: llvm
2105 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2112 The string '``undef``' can be used anywhere a constant is expected, and
2113 indicates that the user of the value may receive an unspecified
2114 bit-pattern. Undefined values may be of any type (other than '``label``'
2115 or '``void``') and be used anywhere a constant is permitted.
2117 Undefined values are useful because they indicate to the compiler that
2118 the program is well defined no matter what value is used. This gives the
2119 compiler more freedom to optimize. Here are some examples of
2120 (potentially surprising) transformations that are valid (in pseudo IR):
2122 .. code-block:: llvm
2132 This is safe because all of the output bits are affected by the undef
2133 bits. Any output bit can have a zero or one depending on the input bits.
2135 .. code-block:: llvm
2146 These logical operations have bits that are not always affected by the
2147 input. For example, if ``%X`` has a zero bit, then the output of the
2148 '``and``' operation will always be a zero for that bit, no matter what
2149 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2150 optimize or assume that the result of the '``and``' is '``undef``'.
2151 However, it is safe to assume that all bits of the '``undef``' could be
2152 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2153 all the bits of the '``undef``' operand to the '``or``' could be set,
2154 allowing the '``or``' to be folded to -1.
2156 .. code-block:: llvm
2158 %A = select undef, %X, %Y
2159 %B = select undef, 42, %Y
2160 %C = select %X, %Y, undef
2170 This set of examples shows that undefined '``select``' (and conditional
2171 branch) conditions can go *either way*, but they have to come from one
2172 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2173 both known to have a clear low bit, then ``%A`` would have to have a
2174 cleared low bit. However, in the ``%C`` example, the optimizer is
2175 allowed to assume that the '``undef``' operand could be the same as
2176 ``%Y``, allowing the whole '``select``' to be eliminated.
2178 .. code-block:: llvm
2180 %A = xor undef, undef
2197 This example points out that two '``undef``' operands are not
2198 necessarily the same. This can be surprising to people (and also matches
2199 C semantics) where they assume that "``X^X``" is always zero, even if
2200 ``X`` is undefined. This isn't true for a number of reasons, but the
2201 short answer is that an '``undef``' "variable" can arbitrarily change
2202 its value over its "live range". This is true because the variable
2203 doesn't actually *have a live range*. Instead, the value is logically
2204 read from arbitrary registers that happen to be around when needed, so
2205 the value is not necessarily consistent over time. In fact, ``%A`` and
2206 ``%C`` need to have the same semantics or the core LLVM "replace all
2207 uses with" concept would not hold.
2209 .. code-block:: llvm
2217 These examples show the crucial difference between an *undefined value*
2218 and *undefined behavior*. An undefined value (like '``undef``') is
2219 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2220 operation can be constant folded to '``undef``', because the '``undef``'
2221 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2222 However, in the second example, we can make a more aggressive
2223 assumption: because the ``undef`` is allowed to be an arbitrary value,
2224 we are allowed to assume that it could be zero. Since a divide by zero
2225 has *undefined behavior*, we are allowed to assume that the operation
2226 does not execute at all. This allows us to delete the divide and all
2227 code after it. Because the undefined operation "can't happen", the
2228 optimizer can assume that it occurs in dead code.
2230 .. code-block:: llvm
2232 a: store undef -> %X
2233 b: store %X -> undef
2238 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2239 value can be assumed to not have any effect; we can assume that the
2240 value is overwritten with bits that happen to match what was already
2241 there. However, a store *to* an undefined location could clobber
2242 arbitrary memory, therefore, it has undefined behavior.
2249 Poison values are similar to :ref:`undef values <undefvalues>`, however
2250 they also represent the fact that an instruction or constant expression
2251 which cannot evoke side effects has nevertheless detected a condition
2252 which results in undefined behavior.
2254 There is currently no way of representing a poison value in the IR; they
2255 only exist when produced by operations such as :ref:`add <i_add>` with
2258 Poison value behavior is defined in terms of value *dependence*:
2260 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2261 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2262 their dynamic predecessor basic block.
2263 - Function arguments depend on the corresponding actual argument values
2264 in the dynamic callers of their functions.
2265 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2266 instructions that dynamically transfer control back to them.
2267 - :ref:`Invoke <i_invoke>` instructions depend on the
2268 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2269 call instructions that dynamically transfer control back to them.
2270 - Non-volatile loads and stores depend on the most recent stores to all
2271 of the referenced memory addresses, following the order in the IR
2272 (including loads and stores implied by intrinsics such as
2273 :ref:`@llvm.memcpy <int_memcpy>`.)
2274 - An instruction with externally visible side effects depends on the
2275 most recent preceding instruction with externally visible side
2276 effects, following the order in the IR. (This includes :ref:`volatile
2277 operations <volatile>`.)
2278 - An instruction *control-depends* on a :ref:`terminator
2279 instruction <terminators>` if the terminator instruction has
2280 multiple successors and the instruction is always executed when
2281 control transfers to one of the successors, and may not be executed
2282 when control is transferred to another.
2283 - Additionally, an instruction also *control-depends* on a terminator
2284 instruction if the set of instructions it otherwise depends on would
2285 be different if the terminator had transferred control to a different
2287 - Dependence is transitive.
2289 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2290 with the additional affect that any instruction which has a *dependence*
2291 on a poison value has undefined behavior.
2293 Here are some examples:
2295 .. code-block:: llvm
2298 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2299 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2300 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2301 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2303 store i32 %poison, i32* @g ; Poison value stored to memory.
2304 %poison2 = load i32* @g ; Poison value loaded back from memory.
2306 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2308 %narrowaddr = bitcast i32* @g to i16*
2309 %wideaddr = bitcast i32* @g to i64*
2310 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2311 %poison4 = load i64* %wideaddr ; Returns a poison value.
2313 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2314 br i1 %cmp, label %true, label %end ; Branch to either destination.
2317 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2318 ; it has undefined behavior.
2322 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2323 ; Both edges into this PHI are
2324 ; control-dependent on %cmp, so this
2325 ; always results in a poison value.
2327 store volatile i32 0, i32* @g ; This would depend on the store in %true
2328 ; if %cmp is true, or the store in %entry
2329 ; otherwise, so this is undefined behavior.
2331 br i1 %cmp, label %second_true, label %second_end
2332 ; The same branch again, but this time the
2333 ; true block doesn't have side effects.
2340 store volatile i32 0, i32* @g ; This time, the instruction always depends
2341 ; on the store in %end. Also, it is
2342 ; control-equivalent to %end, so this is
2343 ; well-defined (ignoring earlier undefined
2344 ; behavior in this example).
2348 Addresses of Basic Blocks
2349 -------------------------
2351 ``blockaddress(@function, %block)``
2353 The '``blockaddress``' constant computes the address of the specified
2354 basic block in the specified function, and always has an ``i8*`` type.
2355 Taking the address of the entry block is illegal.
2357 This value only has defined behavior when used as an operand to the
2358 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2359 against null. Pointer equality tests between labels addresses results in
2360 undefined behavior --- though, again, comparison against null is ok, and
2361 no label is equal to the null pointer. This may be passed around as an
2362 opaque pointer sized value as long as the bits are not inspected. This
2363 allows ``ptrtoint`` and arithmetic to be performed on these values so
2364 long as the original value is reconstituted before the ``indirectbr``
2367 Finally, some targets may provide defined semantics when using the value
2368 as the operand to an inline assembly, but that is target specific.
2372 Constant Expressions
2373 --------------------
2375 Constant expressions are used to allow expressions involving other
2376 constants to be used as constants. Constant expressions may be of any
2377 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2378 that does not have side effects (e.g. load and call are not supported).
2379 The following is the syntax for constant expressions:
2381 ``trunc (CST to TYPE)``
2382 Truncate a constant to another type. The bit size of CST must be
2383 larger than the bit size of TYPE. Both types must be integers.
2384 ``zext (CST to TYPE)``
2385 Zero extend a constant to another type. The bit size of CST must be
2386 smaller than the bit size of TYPE. Both types must be integers.
2387 ``sext (CST to TYPE)``
2388 Sign extend a constant to another type. The bit size of CST must be
2389 smaller than the bit size of TYPE. Both types must be integers.
2390 ``fptrunc (CST to TYPE)``
2391 Truncate a floating point constant to another floating point type.
2392 The size of CST must be larger than the size of TYPE. Both types
2393 must be floating point.
2394 ``fpext (CST to TYPE)``
2395 Floating point extend a constant to another type. The size of CST
2396 must be smaller or equal to the size of TYPE. Both types must be
2398 ``fptoui (CST to TYPE)``
2399 Convert a floating point constant to the corresponding unsigned
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 ``fptosi (CST to TYPE)``
2405 Convert a floating point constant to the corresponding signed
2406 integer constant. TYPE must be a scalar or vector integer type. CST
2407 must be of scalar or vector floating point type. Both CST and TYPE
2408 must be scalars, or vectors of the same number of elements. If the
2409 value won't fit in the integer type, the results are undefined.
2410 ``uitofp (CST to TYPE)``
2411 Convert an unsigned 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 ``sitofp (CST to TYPE)``
2417 Convert a signed integer constant to the corresponding floating
2418 point constant. TYPE must be a scalar or vector floating point type.
2419 CST must be of scalar or vector integer type. Both CST and TYPE must
2420 be scalars, or vectors of the same number of elements. If the value
2421 won't fit in the floating point type, the results are undefined.
2422 ``ptrtoint (CST to TYPE)``
2423 Convert a pointer typed constant to the corresponding integer
2424 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2425 pointer type. The ``CST`` value is zero extended, truncated, or
2426 unchanged to make it fit in ``TYPE``.
2427 ``inttoptr (CST to TYPE)``
2428 Convert an integer constant to a pointer constant. TYPE must be a
2429 pointer type. CST must be of integer type. The CST value is zero
2430 extended, truncated, or unchanged to make it fit in a pointer size.
2431 This one is *really* dangerous!
2432 ``bitcast (CST to TYPE)``
2433 Convert a constant, CST, to another TYPE. The constraints of the
2434 operands are the same as those for the :ref:`bitcast
2435 instruction <i_bitcast>`.
2436 ``addrspacecast (CST to TYPE)``
2437 Convert a constant pointer or constant vector of pointer, CST, to another
2438 TYPE in a different address space. The constraints of the operands are the
2439 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2440 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2441 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2442 constants. As with the :ref:`getelementptr <i_getelementptr>`
2443 instruction, the index list may have zero or more indexes, which are
2444 required to make sense for the type of "CSTPTR".
2445 ``select (COND, VAL1, VAL2)``
2446 Perform the :ref:`select operation <i_select>` on constants.
2447 ``icmp COND (VAL1, VAL2)``
2448 Performs the :ref:`icmp operation <i_icmp>` on constants.
2449 ``fcmp COND (VAL1, VAL2)``
2450 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2451 ``extractelement (VAL, IDX)``
2452 Perform the :ref:`extractelement operation <i_extractelement>` on
2454 ``insertelement (VAL, ELT, IDX)``
2455 Perform the :ref:`insertelement operation <i_insertelement>` on
2457 ``shufflevector (VEC1, VEC2, IDXMASK)``
2458 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2460 ``extractvalue (VAL, IDX0, IDX1, ...)``
2461 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2462 constants. The index list is interpreted in a similar manner as
2463 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2464 least one index value must be specified.
2465 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2466 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2467 The index list is interpreted in a similar manner as indices in a
2468 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2469 value must be specified.
2470 ``OPCODE (LHS, RHS)``
2471 Perform the specified operation of the LHS and RHS constants. OPCODE
2472 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2473 binary <bitwiseops>` operations. The constraints on operands are
2474 the same as those for the corresponding instruction (e.g. no bitwise
2475 operations on floating point values are allowed).
2482 Inline Assembler Expressions
2483 ----------------------------
2485 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2486 Inline Assembly <moduleasm>`) through the use of a special value. This
2487 value represents the inline assembler as a string (containing the
2488 instructions to emit), a list of operand constraints (stored as a
2489 string), a flag that indicates whether or not the inline asm expression
2490 has side effects, and a flag indicating whether the function containing
2491 the asm needs to align its stack conservatively. An example inline
2492 assembler expression is:
2494 .. code-block:: llvm
2496 i32 (i32) asm "bswap $0", "=r,r"
2498 Inline assembler expressions may **only** be used as the callee operand
2499 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2500 Thus, typically we have:
2502 .. code-block:: llvm
2504 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2506 Inline asms with side effects not visible in the constraint list must be
2507 marked as having side effects. This is done through the use of the
2508 '``sideeffect``' keyword, like so:
2510 .. code-block:: llvm
2512 call void asm sideeffect "eieio", ""()
2514 In some cases inline asms will contain code that will not work unless
2515 the stack is aligned in some way, such as calls or SSE instructions on
2516 x86, yet will not contain code that does that alignment within the asm.
2517 The compiler should make conservative assumptions about what the asm
2518 might contain and should generate its usual stack alignment code in the
2519 prologue if the '``alignstack``' keyword is present:
2521 .. code-block:: llvm
2523 call void asm alignstack "eieio", ""()
2525 Inline asms also support using non-standard assembly dialects. The
2526 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2527 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2528 the only supported dialects. An example is:
2530 .. code-block:: llvm
2532 call void asm inteldialect "eieio", ""()
2534 If multiple keywords appear the '``sideeffect``' keyword must come
2535 first, the '``alignstack``' keyword second and the '``inteldialect``'
2541 The call instructions that wrap inline asm nodes may have a
2542 "``!srcloc``" MDNode attached to it that contains a list of constant
2543 integers. If present, the code generator will use the integer as the
2544 location cookie value when report errors through the ``LLVMContext``
2545 error reporting mechanisms. This allows a front-end to correlate backend
2546 errors that occur with inline asm back to the source code that produced
2549 .. code-block:: llvm
2551 call void asm sideeffect "something bad", ""(), !srcloc !42
2553 !42 = !{ i32 1234567 }
2555 It is up to the front-end to make sense of the magic numbers it places
2556 in the IR. If the MDNode contains multiple constants, the code generator
2557 will use the one that corresponds to the line of the asm that the error
2562 Metadata Nodes and Metadata Strings
2563 -----------------------------------
2565 LLVM IR allows metadata to be attached to instructions in the program
2566 that can convey extra information about the code to the optimizers and
2567 code generator. One example application of metadata is source-level
2568 debug information. There are two metadata primitives: strings and nodes.
2569 All metadata has the ``metadata`` type and is identified in syntax by a
2570 preceding exclamation point ('``!``').
2572 A metadata string is a string surrounded by double quotes. It can
2573 contain any character by escaping non-printable characters with
2574 "``\xx``" where "``xx``" is the two digit hex code. For example:
2577 Metadata nodes are represented with notation similar to structure
2578 constants (a comma separated list of elements, surrounded by braces and
2579 preceded by an exclamation point). Metadata nodes can have any values as
2580 their operand. For example:
2582 .. code-block:: llvm
2584 !{ metadata !"test\00", i32 10}
2586 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2587 metadata nodes, which can be looked up in the module symbol table. For
2590 .. code-block:: llvm
2592 !foo = metadata !{!4, !3}
2594 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2595 function is using two metadata arguments:
2597 .. code-block:: llvm
2599 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2601 Metadata can be attached with an instruction. Here metadata ``!21`` is
2602 attached to the ``add`` instruction using the ``!dbg`` identifier:
2604 .. code-block:: llvm
2606 %indvar.next = add i64 %indvar, 1, !dbg !21
2608 More information about specific metadata nodes recognized by the
2609 optimizers and code generator is found below.
2614 In LLVM IR, memory does not have types, so LLVM's own type system is not
2615 suitable for doing TBAA. Instead, metadata is added to the IR to
2616 describe a type system of a higher level language. This can be used to
2617 implement typical C/C++ TBAA, but it can also be used to implement
2618 custom alias analysis behavior for other languages.
2620 The current metadata format is very simple. TBAA metadata nodes have up
2621 to three fields, e.g.:
2623 .. code-block:: llvm
2625 !0 = metadata !{ metadata !"an example type tree" }
2626 !1 = metadata !{ metadata !"int", metadata !0 }
2627 !2 = metadata !{ metadata !"float", metadata !0 }
2628 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2630 The first field is an identity field. It can be any value, usually a
2631 metadata string, which uniquely identifies the type. The most important
2632 name in the tree is the name of the root node. Two trees with different
2633 root node names are entirely disjoint, even if they have leaves with
2636 The second field identifies the type's parent node in the tree, or is
2637 null or omitted for a root node. A type is considered to alias all of
2638 its descendants and all of its ancestors in the tree. Also, a type is
2639 considered to alias all types in other trees, so that bitcode produced
2640 from multiple front-ends is handled conservatively.
2642 If the third field is present, it's an integer which if equal to 1
2643 indicates that the type is "constant" (meaning
2644 ``pointsToConstantMemory`` should return true; see `other useful
2645 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2647 '``tbaa.struct``' Metadata
2648 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2650 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2651 aggregate assignment operations in C and similar languages, however it
2652 is defined to copy a contiguous region of memory, which is more than
2653 strictly necessary for aggregate types which contain holes due to
2654 padding. Also, it doesn't contain any TBAA information about the fields
2657 ``!tbaa.struct`` metadata can describe which memory subregions in a
2658 memcpy are padding and what the TBAA tags of the struct are.
2660 The current metadata format is very simple. ``!tbaa.struct`` metadata
2661 nodes are a list of operands which are in conceptual groups of three.
2662 For each group of three, the first operand gives the byte offset of a
2663 field in bytes, the second gives its size in bytes, and the third gives
2666 .. code-block:: llvm
2668 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2670 This describes a struct with two fields. The first is at offset 0 bytes
2671 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2672 and has size 4 bytes and has tbaa tag !2.
2674 Note that the fields need not be contiguous. In this example, there is a
2675 4 byte gap between the two fields. This gap represents padding which
2676 does not carry useful data and need not be preserved.
2678 '``fpmath``' Metadata
2679 ^^^^^^^^^^^^^^^^^^^^^
2681 ``fpmath`` metadata may be attached to any instruction of floating point
2682 type. It can be used to express the maximum acceptable error in the
2683 result of that instruction, in ULPs, thus potentially allowing the
2684 compiler to use a more efficient but less accurate method of computing
2685 it. ULP is defined as follows:
2687 If ``x`` is a real number that lies between two finite consecutive
2688 floating-point numbers ``a`` and ``b``, without being equal to one
2689 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2690 distance between the two non-equal finite floating-point numbers
2691 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2693 The metadata node shall consist of a single positive floating point
2694 number representing the maximum relative error, for example:
2696 .. code-block:: llvm
2698 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2700 '``range``' Metadata
2701 ^^^^^^^^^^^^^^^^^^^^
2703 ``range`` metadata may be attached only to loads of integer types. It
2704 expresses the possible ranges the loaded value is in. The ranges are
2705 represented with a flattened list of integers. The loaded value is known
2706 to be in the union of the ranges defined by each consecutive pair. Each
2707 pair has the following properties:
2709 - The type must match the type loaded by the instruction.
2710 - The pair ``a,b`` represents the range ``[a,b)``.
2711 - Both ``a`` and ``b`` are constants.
2712 - The range is allowed to wrap.
2713 - The range should not represent the full or empty set. That is,
2716 In addition, the pairs must be in signed order of the lower bound and
2717 they must be non-contiguous.
2721 .. code-block:: llvm
2723 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2724 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2725 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2726 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2728 !0 = metadata !{ i8 0, i8 2 }
2729 !1 = metadata !{ i8 255, i8 2 }
2730 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2731 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2736 It is sometimes useful to attach information to loop constructs. Currently,
2737 loop metadata is implemented as metadata attached to the branch instruction
2738 in the loop latch block. This type of metadata refer to a metadata node that is
2739 guaranteed to be separate for each loop. The loop identifier metadata is
2740 specified with the name ``llvm.loop``.
2742 The loop identifier metadata is implemented using a metadata that refers to
2743 itself to avoid merging it with any other identifier metadata, e.g.,
2744 during module linkage or function inlining. That is, each loop should refer
2745 to their own identification metadata even if they reside in separate functions.
2746 The following example contains loop identifier metadata for two separate loop
2749 .. code-block:: llvm
2751 !0 = metadata !{ metadata !0 }
2752 !1 = metadata !{ metadata !1 }
2754 The loop identifier metadata can be used to specify additional per-loop
2755 metadata. Any operands after the first operand can be treated as user-defined
2756 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2757 by the loop vectorizer to indicate how many times to unroll the loop:
2759 .. code-block:: llvm
2761 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2763 !0 = metadata !{ metadata !0, metadata !1 }
2764 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2769 Metadata types used to annotate memory accesses with information helpful
2770 for optimizations are prefixed with ``llvm.mem``.
2772 '``llvm.mem.parallel_loop_access``' Metadata
2773 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2775 For a loop to be parallel, in addition to using
2776 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2777 also all of the memory accessing instructions in the loop body need to be
2778 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2779 is at least one memory accessing instruction not marked with the metadata,
2780 the loop must be considered a sequential loop. This causes parallel loops to be
2781 converted to sequential loops due to optimization passes that are unaware of
2782 the parallel semantics and that insert new memory instructions to the loop
2785 Example of a loop that is considered parallel due to its correct use of
2786 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2787 metadata types that refer to the same loop identifier metadata.
2789 .. code-block:: llvm
2793 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
2795 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2797 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2801 !0 = metadata !{ metadata !0 }
2803 It is also possible to have nested parallel loops. In that case the
2804 memory accesses refer to a list of loop identifier metadata nodes instead of
2805 the loop identifier metadata node directly:
2807 .. code-block:: llvm
2811 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
2813 br label %inner.for.body
2817 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2819 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
2821 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2825 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
2827 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2829 outer.for.end: ; preds = %for.body
2831 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2832 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2833 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2835 '``llvm.vectorizer``'
2836 ^^^^^^^^^^^^^^^^^^^^^
2838 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2839 vectorization parameters such as vectorization factor and unroll factor.
2841 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2842 loop identification metadata.
2844 '``llvm.vectorizer.unroll``' Metadata
2845 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2847 This metadata instructs the loop vectorizer to unroll the specified
2848 loop exactly ``N`` times.
2850 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2851 operand is an integer specifying the unroll factor. For example:
2853 .. code-block:: llvm
2855 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2857 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2860 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2861 determined automatically.
2863 '``llvm.vectorizer.width``' Metadata
2864 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2866 This metadata sets the target width of the vectorizer to ``N``. Without
2867 this metadata, the vectorizer will choose a width automatically.
2868 Regardless of this metadata, the vectorizer will only vectorize loops if
2869 it believes it is valid to do so.
2871 The first operand is the string ``llvm.vectorizer.width`` and the second
2872 operand is an integer specifying the width. For example:
2874 .. code-block:: llvm
2876 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2878 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2881 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2884 Module Flags Metadata
2885 =====================
2887 Information about the module as a whole is difficult to convey to LLVM's
2888 subsystems. The LLVM IR isn't sufficient to transmit this information.
2889 The ``llvm.module.flags`` named metadata exists in order to facilitate
2890 this. These flags are in the form of key / value pairs --- much like a
2891 dictionary --- making it easy for any subsystem who cares about a flag to
2894 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2895 Each triplet has the following form:
2897 - The first element is a *behavior* flag, which specifies the behavior
2898 when two (or more) modules are merged together, and it encounters two
2899 (or more) metadata with the same ID. The supported behaviors are
2901 - The second element is a metadata string that is a unique ID for the
2902 metadata. Each module may only have one flag entry for each unique ID (not
2903 including entries with the **Require** behavior).
2904 - The third element is the value of the flag.
2906 When two (or more) modules are merged together, the resulting
2907 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2908 each unique metadata ID string, there will be exactly one entry in the merged
2909 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2910 be determined by the merge behavior flag, as described below. The only exception
2911 is that entries with the *Require* behavior are always preserved.
2913 The following behaviors are supported:
2924 Emits an error if two values disagree, otherwise the resulting value
2925 is that of the operands.
2929 Emits a warning if two values disagree. The result value will be the
2930 operand for the flag from the first module being linked.
2934 Adds a requirement that another module flag be present and have a
2935 specified value after linking is performed. The value must be a
2936 metadata pair, where the first element of the pair is the ID of the
2937 module flag to be restricted, and the second element of the pair is
2938 the value the module flag should be restricted to. This behavior can
2939 be used to restrict the allowable results (via triggering of an
2940 error) of linking IDs with the **Override** behavior.
2944 Uses the specified value, regardless of the behavior or value of the
2945 other module. If both modules specify **Override**, but the values
2946 differ, an error will be emitted.
2950 Appends the two values, which are required to be metadata nodes.
2954 Appends the two values, which are required to be metadata
2955 nodes. However, duplicate entries in the second list are dropped
2956 during the append operation.
2958 It is an error for a particular unique flag ID to have multiple behaviors,
2959 except in the case of **Require** (which adds restrictions on another metadata
2960 value) or **Override**.
2962 An example of module flags:
2964 .. code-block:: llvm
2966 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2967 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2968 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2969 !3 = metadata !{ i32 3, metadata !"qux",
2971 metadata !"foo", i32 1
2974 !llvm.module.flags = !{ !0, !1, !2, !3 }
2976 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2977 if two or more ``!"foo"`` flags are seen is to emit an error if their
2978 values are not equal.
2980 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2981 behavior if two or more ``!"bar"`` flags are seen is to use the value
2984 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2985 behavior if two or more ``!"qux"`` flags are seen is to emit a
2986 warning if their values are not equal.
2988 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2992 metadata !{ metadata !"foo", i32 1 }
2994 The behavior is to emit an error if the ``llvm.module.flags`` does not
2995 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2998 Objective-C Garbage Collection Module Flags Metadata
2999 ----------------------------------------------------
3001 On the Mach-O platform, Objective-C stores metadata about garbage
3002 collection in a special section called "image info". The metadata
3003 consists of a version number and a bitmask specifying what types of
3004 garbage collection are supported (if any) by the file. If two or more
3005 modules are linked together their garbage collection metadata needs to
3006 be merged rather than appended together.
3008 The Objective-C garbage collection module flags metadata consists of the
3009 following key-value pairs:
3018 * - ``Objective-C Version``
3019 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3021 * - ``Objective-C Image Info Version``
3022 - **[Required]** --- The version of the image info section. Currently
3025 * - ``Objective-C Image Info Section``
3026 - **[Required]** --- The section to place the metadata. Valid values are
3027 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3028 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3029 Objective-C ABI version 2.
3031 * - ``Objective-C Garbage Collection``
3032 - **[Required]** --- Specifies whether garbage collection is supported or
3033 not. Valid values are 0, for no garbage collection, and 2, for garbage
3034 collection supported.
3036 * - ``Objective-C GC Only``
3037 - **[Optional]** --- Specifies that only garbage collection is supported.
3038 If present, its value must be 6. This flag requires that the
3039 ``Objective-C Garbage Collection`` flag have the value 2.
3041 Some important flag interactions:
3043 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3044 merged with a module with ``Objective-C Garbage Collection`` set to
3045 2, then the resulting module has the
3046 ``Objective-C Garbage Collection`` flag set to 0.
3047 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3048 merged with a module with ``Objective-C GC Only`` set to 6.
3050 Automatic Linker Flags Module Flags Metadata
3051 --------------------------------------------
3053 Some targets support embedding flags to the linker inside individual object
3054 files. Typically this is used in conjunction with language extensions which
3055 allow source files to explicitly declare the libraries they depend on, and have
3056 these automatically be transmitted to the linker via object files.
3058 These flags are encoded in the IR using metadata in the module flags section,
3059 using the ``Linker Options`` key. The merge behavior for this flag is required
3060 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3061 node which should be a list of other metadata nodes, each of which should be a
3062 list of metadata strings defining linker options.
3064 For example, the following metadata section specifies two separate sets of
3065 linker options, presumably to link against ``libz`` and the ``Cocoa``
3068 !0 = metadata !{ i32 6, metadata !"Linker Options",
3070 metadata !{ metadata !"-lz" },
3071 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3072 !llvm.module.flags = !{ !0 }
3074 The metadata encoding as lists of lists of options, as opposed to a collapsed
3075 list of options, is chosen so that the IR encoding can use multiple option
3076 strings to specify e.g., a single library, while still having that specifier be
3077 preserved as an atomic element that can be recognized by a target specific
3078 assembly writer or object file emitter.
3080 Each individual option is required to be either a valid option for the target's
3081 linker, or an option that is reserved by the target specific assembly writer or
3082 object file emitter. No other aspect of these options is defined by the IR.
3084 .. _intrinsicglobalvariables:
3086 Intrinsic Global Variables
3087 ==========================
3089 LLVM has a number of "magic" global variables that contain data that
3090 affect code generation or other IR semantics. These are documented here.
3091 All globals of this sort should have a section specified as
3092 "``llvm.metadata``". This section and all globals that start with
3093 "``llvm.``" are reserved for use by LLVM.
3097 The '``llvm.used``' Global Variable
3098 -----------------------------------
3100 The ``@llvm.used`` global is an array which has
3101 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3102 pointers to named global variables, functions and aliases which may optionally
3103 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3106 .. code-block:: llvm
3111 @llvm.used = appending global [2 x i8*] [
3113 i8* bitcast (i32* @Y to i8*)
3114 ], section "llvm.metadata"
3116 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3117 and linker are required to treat the symbol as if there is a reference to the
3118 symbol that it cannot see (which is why they have to be named). For example, if
3119 a variable has internal linkage and no references other than that from the
3120 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3121 references from inline asms and other things the compiler cannot "see", and
3122 corresponds to "``attribute((used))``" in GNU C.
3124 On some targets, the code generator must emit a directive to the
3125 assembler or object file to prevent the assembler and linker from
3126 molesting the symbol.
3128 .. _gv_llvmcompilerused:
3130 The '``llvm.compiler.used``' Global Variable
3131 --------------------------------------------
3133 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3134 directive, except that it only prevents the compiler from touching the
3135 symbol. On targets that support it, this allows an intelligent linker to
3136 optimize references to the symbol without being impeded as it would be
3139 This is a rare construct that should only be used in rare circumstances,
3140 and should not be exposed to source languages.
3142 .. _gv_llvmglobalctors:
3144 The '``llvm.global_ctors``' Global Variable
3145 -------------------------------------------
3147 .. code-block:: llvm
3149 %0 = type { i32, void ()* }
3150 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3152 The ``@llvm.global_ctors`` array contains a list of constructor
3153 functions and associated priorities. The functions referenced by this
3154 array will be called in ascending order of priority (i.e. lowest first)
3155 when the module is loaded. The order of functions with the same priority
3158 .. _llvmglobaldtors:
3160 The '``llvm.global_dtors``' Global Variable
3161 -------------------------------------------
3163 .. code-block:: llvm
3165 %0 = type { i32, void ()* }
3166 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3168 The ``@llvm.global_dtors`` array contains a list of destructor functions
3169 and associated priorities. The functions referenced by this array will
3170 be called in descending order of priority (i.e. highest first) when the
3171 module is loaded. The order of functions with the same priority is not
3174 Instruction Reference
3175 =====================
3177 The LLVM instruction set consists of several different classifications
3178 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3179 instructions <binaryops>`, :ref:`bitwise binary
3180 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3181 :ref:`other instructions <otherops>`.
3185 Terminator Instructions
3186 -----------------------
3188 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3189 program ends with a "Terminator" instruction, which indicates which
3190 block should be executed after the current block is finished. These
3191 terminator instructions typically yield a '``void``' value: they produce
3192 control flow, not values (the one exception being the
3193 ':ref:`invoke <i_invoke>`' instruction).
3195 The terminator instructions are: ':ref:`ret <i_ret>`',
3196 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3197 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3198 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3202 '``ret``' Instruction
3203 ^^^^^^^^^^^^^^^^^^^^^
3210 ret <type> <value> ; Return a value from a non-void function
3211 ret void ; Return from void function
3216 The '``ret``' instruction is used to return control flow (and optionally
3217 a value) from a function back to the caller.
3219 There are two forms of the '``ret``' instruction: one that returns a
3220 value and then causes control flow, and one that just causes control
3226 The '``ret``' instruction optionally accepts a single argument, the
3227 return value. The type of the return value must be a ':ref:`first
3228 class <t_firstclass>`' type.
3230 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3231 return type and contains a '``ret``' instruction with no return value or
3232 a return value with a type that does not match its type, or if it has a
3233 void return type and contains a '``ret``' instruction with a return
3239 When the '``ret``' instruction is executed, control flow returns back to
3240 the calling function's context. If the caller is a
3241 ":ref:`call <i_call>`" instruction, execution continues at the
3242 instruction after the call. If the caller was an
3243 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3244 beginning of the "normal" destination block. If the instruction returns
3245 a value, that value shall set the call or invoke instruction's return
3251 .. code-block:: llvm
3253 ret i32 5 ; Return an integer value of 5
3254 ret void ; Return from a void function
3255 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3259 '``br``' Instruction
3260 ^^^^^^^^^^^^^^^^^^^^
3267 br i1 <cond>, label <iftrue>, label <iffalse>
3268 br label <dest> ; Unconditional branch
3273 The '``br``' instruction is used to cause control flow to transfer to a
3274 different basic block in the current function. There are two forms of
3275 this instruction, corresponding to a conditional branch and an
3276 unconditional branch.
3281 The conditional branch form of the '``br``' instruction takes a single
3282 '``i1``' value and two '``label``' values. The unconditional form of the
3283 '``br``' instruction takes a single '``label``' value as a target.
3288 Upon execution of a conditional '``br``' instruction, the '``i1``'
3289 argument is evaluated. If the value is ``true``, control flows to the
3290 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3291 to the '``iffalse``' ``label`` argument.
3296 .. code-block:: llvm
3299 %cond = icmp eq i32 %a, %b
3300 br i1 %cond, label %IfEqual, label %IfUnequal
3308 '``switch``' Instruction
3309 ^^^^^^^^^^^^^^^^^^^^^^^^
3316 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3321 The '``switch``' instruction is used to transfer control flow to one of
3322 several different places. It is a generalization of the '``br``'
3323 instruction, allowing a branch to occur to one of many possible
3329 The '``switch``' instruction uses three parameters: an integer
3330 comparison value '``value``', a default '``label``' destination, and an
3331 array of pairs of comparison value constants and '``label``'s. The table
3332 is not allowed to contain duplicate constant entries.
3337 The ``switch`` instruction specifies a table of values and destinations.
3338 When the '``switch``' instruction is executed, this table is searched
3339 for the given value. If the value is found, control flow is transferred
3340 to the corresponding destination; otherwise, control flow is transferred
3341 to the default destination.
3346 Depending on properties of the target machine and the particular
3347 ``switch`` instruction, this instruction may be code generated in
3348 different ways. For example, it could be generated as a series of
3349 chained conditional branches or with a lookup table.
3354 .. code-block:: llvm
3356 ; Emulate a conditional br instruction
3357 %Val = zext i1 %value to i32
3358 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3360 ; Emulate an unconditional br instruction
3361 switch i32 0, label %dest [ ]
3363 ; Implement a jump table:
3364 switch i32 %val, label %otherwise [ i32 0, label %onzero
3366 i32 2, label %ontwo ]
3370 '``indirectbr``' Instruction
3371 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3378 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3383 The '``indirectbr``' instruction implements an indirect branch to a
3384 label within the current function, whose address is specified by
3385 "``address``". Address must be derived from a
3386 :ref:`blockaddress <blockaddress>` constant.
3391 The '``address``' argument is the address of the label to jump to. The
3392 rest of the arguments indicate the full set of possible destinations
3393 that the address may point to. Blocks are allowed to occur multiple
3394 times in the destination list, though this isn't particularly useful.
3396 This destination list is required so that dataflow analysis has an
3397 accurate understanding of the CFG.
3402 Control transfers to the block specified in the address argument. All
3403 possible destination blocks must be listed in the label list, otherwise
3404 this instruction has undefined behavior. This implies that jumps to
3405 labels defined in other functions have undefined behavior as well.
3410 This is typically implemented with a jump through a register.
3415 .. code-block:: llvm
3417 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3421 '``invoke``' Instruction
3422 ^^^^^^^^^^^^^^^^^^^^^^^^
3429 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3430 to label <normal label> unwind label <exception label>
3435 The '``invoke``' instruction causes control to transfer to a specified
3436 function, with the possibility of control flow transfer to either the
3437 '``normal``' label or the '``exception``' label. If the callee function
3438 returns with the "``ret``" instruction, control flow will return to the
3439 "normal" label. If the callee (or any indirect callees) returns via the
3440 ":ref:`resume <i_resume>`" instruction or other exception handling
3441 mechanism, control is interrupted and continued at the dynamically
3442 nearest "exception" label.
3444 The '``exception``' label is a `landing
3445 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3446 '``exception``' label is required to have the
3447 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3448 information about the behavior of the program after unwinding happens,
3449 as its first non-PHI instruction. The restrictions on the
3450 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3451 instruction, so that the important information contained within the
3452 "``landingpad``" instruction can't be lost through normal code motion.
3457 This instruction requires several arguments:
3459 #. The optional "cconv" marker indicates which :ref:`calling
3460 convention <callingconv>` the call should use. If none is
3461 specified, the call defaults to using C calling conventions.
3462 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3463 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3465 #. '``ptr to function ty``': shall be the signature of the pointer to
3466 function value being invoked. In most cases, this is a direct
3467 function invocation, but indirect ``invoke``'s are just as possible,
3468 branching off an arbitrary pointer to function value.
3469 #. '``function ptr val``': An LLVM value containing a pointer to a
3470 function to be invoked.
3471 #. '``function args``': argument list whose types match the function
3472 signature argument types and parameter attributes. All arguments must
3473 be of :ref:`first class <t_firstclass>` type. If the function signature
3474 indicates the function accepts a variable number of arguments, the
3475 extra arguments can be specified.
3476 #. '``normal label``': the label reached when the called function
3477 executes a '``ret``' instruction.
3478 #. '``exception label``': the label reached when a callee returns via
3479 the :ref:`resume <i_resume>` instruction or other exception handling
3481 #. The optional :ref:`function attributes <fnattrs>` list. Only
3482 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3483 attributes are valid here.
3488 This instruction is designed to operate as a standard '``call``'
3489 instruction in most regards. The primary difference is that it
3490 establishes an association with a label, which is used by the runtime
3491 library to unwind the stack.
3493 This instruction is used in languages with destructors to ensure that
3494 proper cleanup is performed in the case of either a ``longjmp`` or a
3495 thrown exception. Additionally, this is important for implementation of
3496 '``catch``' clauses in high-level languages that support them.
3498 For the purposes of the SSA form, the definition of the value returned
3499 by the '``invoke``' instruction is deemed to occur on the edge from the
3500 current block to the "normal" label. If the callee unwinds then no
3501 return value is available.
3506 .. code-block:: llvm
3508 %retval = invoke i32 @Test(i32 15) to label %Continue
3509 unwind label %TestCleanup ; {i32}:retval set
3510 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3511 unwind label %TestCleanup ; {i32}:retval set
3515 '``resume``' Instruction
3516 ^^^^^^^^^^^^^^^^^^^^^^^^
3523 resume <type> <value>
3528 The '``resume``' instruction is a terminator instruction that has no
3534 The '``resume``' instruction requires one argument, which must have the
3535 same type as the result of any '``landingpad``' instruction in the same
3541 The '``resume``' instruction resumes propagation of an existing
3542 (in-flight) exception whose unwinding was interrupted with a
3543 :ref:`landingpad <i_landingpad>` instruction.
3548 .. code-block:: llvm
3550 resume { i8*, i32 } %exn
3554 '``unreachable``' Instruction
3555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3567 The '``unreachable``' instruction has no defined semantics. This
3568 instruction is used to inform the optimizer that a particular portion of
3569 the code is not reachable. This can be used to indicate that the code
3570 after a no-return function cannot be reached, and other facts.
3575 The '``unreachable``' instruction has no defined semantics.
3582 Binary operators are used to do most of the computation in a program.
3583 They require two operands of the same type, execute an operation on
3584 them, and produce a single value. The operands might represent multiple
3585 data, as is the case with the :ref:`vector <t_vector>` data type. The
3586 result value has the same type as its operands.
3588 There are several different binary operators:
3592 '``add``' Instruction
3593 ^^^^^^^^^^^^^^^^^^^^^
3600 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3601 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3602 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3603 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3608 The '``add``' instruction returns the sum of its two operands.
3613 The two arguments to the '``add``' instruction must be
3614 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3615 arguments must have identical types.
3620 The value produced is the integer sum of the two operands.
3622 If the sum has unsigned overflow, the result returned is the
3623 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3626 Because LLVM integers use a two's complement representation, this
3627 instruction is appropriate for both signed and unsigned integers.
3629 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3630 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3631 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3632 unsigned and/or signed overflow, respectively, occurs.
3637 .. code-block:: llvm
3639 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3643 '``fadd``' Instruction
3644 ^^^^^^^^^^^^^^^^^^^^^^
3651 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3656 The '``fadd``' instruction returns the sum of its two operands.
3661 The two arguments to the '``fadd``' instruction must be :ref:`floating
3662 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3663 Both arguments must have identical types.
3668 The value produced is the floating point sum of the two operands. This
3669 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3670 which are optimization hints to enable otherwise unsafe floating point
3676 .. code-block:: llvm
3678 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3680 '``sub``' Instruction
3681 ^^^^^^^^^^^^^^^^^^^^^
3688 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3689 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3690 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3691 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3696 The '``sub``' instruction returns the difference of its two operands.
3698 Note that the '``sub``' instruction is used to represent the '``neg``'
3699 instruction present in most other intermediate representations.
3704 The two arguments to the '``sub``' instruction must be
3705 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3706 arguments must have identical types.
3711 The value produced is the integer difference of the two operands.
3713 If the difference has unsigned overflow, the result returned is the
3714 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3717 Because LLVM integers use a two's complement representation, this
3718 instruction is appropriate for both signed and unsigned integers.
3720 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3721 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3722 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3723 unsigned and/or signed overflow, respectively, occurs.
3728 .. code-block:: llvm
3730 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3731 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3735 '``fsub``' Instruction
3736 ^^^^^^^^^^^^^^^^^^^^^^
3743 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3748 The '``fsub``' instruction returns the difference of its two operands.
3750 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3751 instruction present in most other intermediate representations.
3756 The two arguments to the '``fsub``' instruction must be :ref:`floating
3757 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3758 Both arguments must have identical types.
3763 The value produced is the floating point difference of the two operands.
3764 This instruction can also take any number of :ref:`fast-math
3765 flags <fastmath>`, which are optimization hints to enable otherwise
3766 unsafe floating point optimizations:
3771 .. code-block:: llvm
3773 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3774 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3776 '``mul``' Instruction
3777 ^^^^^^^^^^^^^^^^^^^^^
3784 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3785 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3786 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3787 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3792 The '``mul``' instruction returns the product of its two operands.
3797 The two arguments to the '``mul``' instruction must be
3798 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3799 arguments must have identical types.
3804 The value produced is the integer product of the two operands.
3806 If the result of the multiplication has unsigned overflow, the result
3807 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3808 bit width of the result.
3810 Because LLVM integers use a two's complement representation, and the
3811 result is the same width as the operands, this instruction returns the
3812 correct result for both signed and unsigned integers. If a full product
3813 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3814 sign-extended or zero-extended as appropriate to the width of the full
3817 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3818 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3819 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3820 unsigned and/or signed overflow, respectively, occurs.
3825 .. code-block:: llvm
3827 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3831 '``fmul``' Instruction
3832 ^^^^^^^^^^^^^^^^^^^^^^
3839 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3844 The '``fmul``' instruction returns the product of its two operands.
3849 The two arguments to the '``fmul``' instruction must be :ref:`floating
3850 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3851 Both arguments must have identical types.
3856 The value produced is the floating point product of the two operands.
3857 This instruction can also take any number of :ref:`fast-math
3858 flags <fastmath>`, which are optimization hints to enable otherwise
3859 unsafe floating point optimizations:
3864 .. code-block:: llvm
3866 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3868 '``udiv``' Instruction
3869 ^^^^^^^^^^^^^^^^^^^^^^
3876 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3877 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3882 The '``udiv``' instruction returns the quotient of its two operands.
3887 The two arguments to the '``udiv``' instruction must be
3888 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3889 arguments must have identical types.
3894 The value produced is the unsigned integer quotient of the two operands.
3896 Note that unsigned integer division and signed integer division are
3897 distinct operations; for signed integer division, use '``sdiv``'.
3899 Division by zero leads to undefined behavior.
3901 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3902 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3903 such, "((a udiv exact b) mul b) == a").
3908 .. code-block:: llvm
3910 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3912 '``sdiv``' Instruction
3913 ^^^^^^^^^^^^^^^^^^^^^^
3920 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3921 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3926 The '``sdiv``' instruction returns the quotient of its two operands.
3931 The two arguments to the '``sdiv``' instruction must be
3932 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3933 arguments must have identical types.
3938 The value produced is the signed integer quotient of the two operands
3939 rounded towards zero.
3941 Note that signed integer division and unsigned integer division are
3942 distinct operations; for unsigned integer division, use '``udiv``'.
3944 Division by zero leads to undefined behavior. Overflow also leads to
3945 undefined behavior; this is a rare case, but can occur, for example, by
3946 doing a 32-bit division of -2147483648 by -1.
3948 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3949 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3954 .. code-block:: llvm
3956 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3960 '``fdiv``' Instruction
3961 ^^^^^^^^^^^^^^^^^^^^^^
3968 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3973 The '``fdiv``' instruction returns the quotient of its two operands.
3978 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3979 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3980 Both arguments must have identical types.
3985 The value produced is the floating point quotient of the two operands.
3986 This instruction can also take any number of :ref:`fast-math
3987 flags <fastmath>`, which are optimization hints to enable otherwise
3988 unsafe floating point optimizations:
3993 .. code-block:: llvm
3995 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3997 '``urem``' Instruction
3998 ^^^^^^^^^^^^^^^^^^^^^^
4005 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
4010 The '``urem``' instruction returns the remainder from the unsigned
4011 division of its two arguments.
4016 The two arguments to the '``urem``' instruction must be
4017 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4018 arguments must have identical types.
4023 This instruction returns the unsigned integer *remainder* of a division.
4024 This instruction always performs an unsigned division to get the
4027 Note that unsigned integer remainder and signed integer remainder are
4028 distinct operations; for signed integer remainder, use '``srem``'.
4030 Taking the remainder of a division by zero leads to undefined behavior.
4035 .. code-block:: llvm
4037 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
4039 '``srem``' Instruction
4040 ^^^^^^^^^^^^^^^^^^^^^^
4047 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
4052 The '``srem``' instruction returns the remainder from the signed
4053 division of its two operands. This instruction can also take
4054 :ref:`vector <t_vector>` versions of the values in which case the elements
4060 The two arguments to the '``srem``' instruction must be
4061 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4062 arguments must have identical types.
4067 This instruction returns the *remainder* of a division (where the result
4068 is either zero or has the same sign as the dividend, ``op1``), not the
4069 *modulo* operator (where the result is either zero or has the same sign
4070 as the divisor, ``op2``) of a value. For more information about the
4071 difference, see `The Math
4072 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4073 table of how this is implemented in various languages, please see
4075 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4077 Note that signed integer remainder and unsigned integer remainder are
4078 distinct operations; for unsigned integer remainder, use '``urem``'.
4080 Taking the remainder of a division by zero leads to undefined behavior.
4081 Overflow also leads to undefined behavior; this is a rare case, but can
4082 occur, for example, by taking the remainder of a 32-bit division of
4083 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4084 rule lets srem be implemented using instructions that return both the
4085 result of the division and the remainder.)
4090 .. code-block:: llvm
4092 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4096 '``frem``' Instruction
4097 ^^^^^^^^^^^^^^^^^^^^^^
4104 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4109 The '``frem``' instruction returns the remainder from the division of
4115 The two arguments to the '``frem``' instruction must be :ref:`floating
4116 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4117 Both arguments must have identical types.
4122 This instruction returns the *remainder* of a division. The remainder
4123 has the same sign as the dividend. This instruction can also take any
4124 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4125 to enable otherwise unsafe floating point optimizations:
4130 .. code-block:: llvm
4132 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4136 Bitwise Binary Operations
4137 -------------------------
4139 Bitwise binary operators are used to do various forms of bit-twiddling
4140 in a program. They are generally very efficient instructions and can
4141 commonly be strength reduced from other instructions. They require two
4142 operands of the same type, execute an operation on them, and produce a
4143 single value. The resulting value is the same type as its operands.
4145 '``shl``' Instruction
4146 ^^^^^^^^^^^^^^^^^^^^^
4153 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4154 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4155 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4156 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4161 The '``shl``' instruction returns the first operand shifted to the left
4162 a specified number of bits.
4167 Both arguments to the '``shl``' instruction must be the same
4168 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4169 '``op2``' is treated as an unsigned value.
4174 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4175 where ``n`` is the width of the result. If ``op2`` is (statically or
4176 dynamically) negative or equal to or larger than the number of bits in
4177 ``op1``, the result is undefined. If the arguments are vectors, each
4178 vector element of ``op1`` is shifted by the corresponding shift amount
4181 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4182 value <poisonvalues>` if it shifts out any non-zero bits. If the
4183 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4184 value <poisonvalues>` if it shifts out any bits that disagree with the
4185 resultant sign bit. As such, NUW/NSW have the same semantics as they
4186 would if the shift were expressed as a mul instruction with the same
4187 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4192 .. code-block:: llvm
4194 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4195 <result> = shl i32 4, 2 ; yields {i32}: 16
4196 <result> = shl i32 1, 10 ; yields {i32}: 1024
4197 <result> = shl i32 1, 32 ; undefined
4198 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4200 '``lshr``' Instruction
4201 ^^^^^^^^^^^^^^^^^^^^^^
4208 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4209 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4214 The '``lshr``' instruction (logical shift right) returns the first
4215 operand shifted to the right a specified number of bits with zero fill.
4220 Both arguments to the '``lshr``' instruction must be the same
4221 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4222 '``op2``' is treated as an unsigned value.
4227 This instruction always performs a logical shift right operation. The
4228 most significant bits of the result will be filled with zero bits after
4229 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4230 than the number of bits in ``op1``, the result is undefined. If the
4231 arguments are vectors, each vector element of ``op1`` is shifted by the
4232 corresponding shift amount in ``op2``.
4234 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4235 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4241 .. code-block:: llvm
4243 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4244 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4245 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4246 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4247 <result> = lshr i32 1, 32 ; undefined
4248 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4250 '``ashr``' Instruction
4251 ^^^^^^^^^^^^^^^^^^^^^^
4258 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4259 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4264 The '``ashr``' instruction (arithmetic shift right) returns the first
4265 operand shifted to the right a specified number of bits with sign
4271 Both arguments to the '``ashr``' instruction must be the same
4272 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4273 '``op2``' is treated as an unsigned value.
4278 This instruction always performs an arithmetic shift right operation,
4279 The most significant bits of the result will be filled with the sign bit
4280 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4281 than the number of bits in ``op1``, the result is undefined. If the
4282 arguments are vectors, each vector element of ``op1`` is shifted by the
4283 corresponding shift amount in ``op2``.
4285 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4286 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4292 .. code-block:: llvm
4294 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4295 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4296 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4297 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4298 <result> = ashr i32 1, 32 ; undefined
4299 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4301 '``and``' Instruction
4302 ^^^^^^^^^^^^^^^^^^^^^
4309 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4314 The '``and``' instruction returns the bitwise logical and of its two
4320 The two arguments to the '``and``' instruction must be
4321 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4322 arguments must have identical types.
4327 The truth table used for the '``and``' instruction is:
4344 .. code-block:: llvm
4346 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4347 <result> = and i32 15, 40 ; yields {i32}:result = 8
4348 <result> = and i32 4, 8 ; yields {i32}:result = 0
4350 '``or``' Instruction
4351 ^^^^^^^^^^^^^^^^^^^^
4358 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4363 The '``or``' instruction returns the bitwise logical inclusive or of its
4369 The two arguments to the '``or``' instruction must be
4370 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4371 arguments must have identical types.
4376 The truth table used for the '``or``' instruction is:
4395 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4396 <result> = or i32 15, 40 ; yields {i32}:result = 47
4397 <result> = or i32 4, 8 ; yields {i32}:result = 12
4399 '``xor``' Instruction
4400 ^^^^^^^^^^^^^^^^^^^^^
4407 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4412 The '``xor``' instruction returns the bitwise logical exclusive or of
4413 its two operands. The ``xor`` is used to implement the "one's
4414 complement" operation, which is the "~" operator in C.
4419 The two arguments to the '``xor``' instruction must be
4420 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4421 arguments must have identical types.
4426 The truth table used for the '``xor``' instruction is:
4443 .. code-block:: llvm
4445 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4446 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4447 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4448 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4453 LLVM supports several instructions to represent vector operations in a
4454 target-independent manner. These instructions cover the element-access
4455 and vector-specific operations needed to process vectors effectively.
4456 While LLVM does directly support these vector operations, many
4457 sophisticated algorithms will want to use target-specific intrinsics to
4458 take full advantage of a specific target.
4460 .. _i_extractelement:
4462 '``extractelement``' Instruction
4463 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4470 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4475 The '``extractelement``' instruction extracts a single scalar element
4476 from a vector at a specified index.
4481 The first operand of an '``extractelement``' instruction is a value of
4482 :ref:`vector <t_vector>` type. The second operand is an index indicating
4483 the position from which to extract the element. The index may be a
4489 The result is a scalar of the same type as the element type of ``val``.
4490 Its value is the value at position ``idx`` of ``val``. If ``idx``
4491 exceeds the length of ``val``, the results are undefined.
4496 .. code-block:: llvm
4498 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4500 .. _i_insertelement:
4502 '``insertelement``' Instruction
4503 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4510 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4515 The '``insertelement``' instruction inserts a scalar element into a
4516 vector at a specified index.
4521 The first operand of an '``insertelement``' instruction is a value of
4522 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4523 type must equal the element type of the first operand. The third operand
4524 is an index indicating the position at which to insert the value. The
4525 index may be a variable.
4530 The result is a vector of the same type as ``val``. Its element values
4531 are those of ``val`` except at position ``idx``, where it gets the value
4532 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4538 .. code-block:: llvm
4540 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4542 .. _i_shufflevector:
4544 '``shufflevector``' Instruction
4545 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4552 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4557 The '``shufflevector``' instruction constructs a permutation of elements
4558 from two input vectors, returning a vector with the same element type as
4559 the input and length that is the same as the shuffle mask.
4564 The first two operands of a '``shufflevector``' instruction are vectors
4565 with the same type. The third argument is a shuffle mask whose element
4566 type is always 'i32'. The result of the instruction is a vector whose
4567 length is the same as the shuffle mask and whose element type is the
4568 same as the element type of the first two operands.
4570 The shuffle mask operand is required to be a constant vector with either
4571 constant integer or undef values.
4576 The elements of the two input vectors are numbered from left to right
4577 across both of the vectors. The shuffle mask operand specifies, for each
4578 element of the result vector, which element of the two input vectors the
4579 result element gets. The element selector may be undef (meaning "don't
4580 care") and the second operand may be undef if performing a shuffle from
4586 .. code-block:: llvm
4588 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4589 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4590 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4591 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4592 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4593 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4594 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4595 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4597 Aggregate Operations
4598 --------------------
4600 LLVM supports several instructions for working with
4601 :ref:`aggregate <t_aggregate>` values.
4605 '``extractvalue``' Instruction
4606 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4613 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4618 The '``extractvalue``' instruction extracts the value of a member field
4619 from an :ref:`aggregate <t_aggregate>` value.
4624 The first operand of an '``extractvalue``' instruction is a value of
4625 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4626 constant indices to specify which value to extract in a similar manner
4627 as indices in a '``getelementptr``' instruction.
4629 The major differences to ``getelementptr`` indexing are:
4631 - Since the value being indexed is not a pointer, the first index is
4632 omitted and assumed to be zero.
4633 - At least one index must be specified.
4634 - Not only struct indices but also array indices must be in bounds.
4639 The result is the value at the position in the aggregate specified by
4645 .. code-block:: llvm
4647 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4651 '``insertvalue``' Instruction
4652 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4659 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4664 The '``insertvalue``' instruction inserts a value into a member field in
4665 an :ref:`aggregate <t_aggregate>` value.
4670 The first operand of an '``insertvalue``' instruction is a value of
4671 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4672 a first-class value to insert. The following operands are constant
4673 indices indicating the position at which to insert the value in a
4674 similar manner as indices in a '``extractvalue``' instruction. The value
4675 to insert must have the same type as the value identified by the
4681 The result is an aggregate of the same type as ``val``. Its value is
4682 that of ``val`` except that the value at the position specified by the
4683 indices is that of ``elt``.
4688 .. code-block:: llvm
4690 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4691 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4692 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4696 Memory Access and Addressing Operations
4697 ---------------------------------------
4699 A key design point of an SSA-based representation is how it represents
4700 memory. In LLVM, no memory locations are in SSA form, which makes things
4701 very simple. This section describes how to read, write, and allocate
4706 '``alloca``' Instruction
4707 ^^^^^^^^^^^^^^^^^^^^^^^^
4714 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields {type*}:result
4719 The '``alloca``' instruction allocates memory on the stack frame of the
4720 currently executing function, to be automatically released when this
4721 function returns to its caller. The object is always allocated in the
4722 generic address space (address space zero).
4727 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4728 bytes of memory on the runtime stack, returning a pointer of the
4729 appropriate type to the program. If "NumElements" is specified, it is
4730 the number of elements allocated, otherwise "NumElements" is defaulted
4731 to be one. If a constant alignment is specified, the value result of the
4732 allocation is guaranteed to be aligned to at least that boundary. If not
4733 specified, or if zero, the target can choose to align the allocation on
4734 any convenient boundary compatible with the type.
4736 '``type``' may be any sized type.
4741 Memory is allocated; a pointer is returned. The operation is undefined
4742 if there is insufficient stack space for the allocation. '``alloca``'d
4743 memory is automatically released when the function returns. The
4744 '``alloca``' instruction is commonly used to represent automatic
4745 variables that must have an address available. When the function returns
4746 (either with the ``ret`` or ``resume`` instructions), the memory is
4747 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4748 The order in which memory is allocated (ie., which way the stack grows)
4754 .. code-block:: llvm
4756 %ptr = alloca i32 ; yields {i32*}:ptr
4757 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4758 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4759 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4763 '``load``' Instruction
4764 ^^^^^^^^^^^^^^^^^^^^^^
4771 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4772 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4773 !<index> = !{ i32 1 }
4778 The '``load``' instruction is used to read from memory.
4783 The argument to the ``load`` instruction specifies the memory address
4784 from which to load. The pointer must point to a :ref:`first
4785 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4786 then the optimizer is not allowed to modify the number or order of
4787 execution of this ``load`` with other :ref:`volatile
4788 operations <volatile>`.
4790 If the ``load`` is marked as ``atomic``, it takes an extra
4791 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4792 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4793 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4794 when they may see multiple atomic stores. The type of the pointee must
4795 be an integer type whose bit width is a power of two greater than or
4796 equal to eight and less than or equal to a target-specific size limit.
4797 ``align`` must be explicitly specified on atomic loads, and the load has
4798 undefined behavior if the alignment is not set to a value which is at
4799 least the size in bytes of the pointee. ``!nontemporal`` does not have
4800 any defined semantics for atomic loads.
4802 The optional constant ``align`` argument specifies the alignment of the
4803 operation (that is, the alignment of the memory address). A value of 0
4804 or an omitted ``align`` argument means that the operation has the ABI
4805 alignment for the target. It is the responsibility of the code emitter
4806 to ensure that the alignment information is correct. Overestimating the
4807 alignment results in undefined behavior. Underestimating the alignment
4808 may produce less efficient code. An alignment of 1 is always safe.
4810 The optional ``!nontemporal`` metadata must reference a single
4811 metadata name ``<index>`` corresponding to a metadata node with one
4812 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4813 metadata on the instruction tells the optimizer and code generator
4814 that this load is not expected to be reused in the cache. The code
4815 generator may select special instructions to save cache bandwidth, such
4816 as the ``MOVNT`` instruction on x86.
4818 The optional ``!invariant.load`` metadata must reference a single
4819 metadata name ``<index>`` corresponding to a metadata node with no
4820 entries. The existence of the ``!invariant.load`` metadata on the
4821 instruction tells the optimizer and code generator that this load
4822 address points to memory which does not change value during program
4823 execution. The optimizer may then move this load around, for example, by
4824 hoisting it out of loops using loop invariant code motion.
4829 The location of memory pointed to is loaded. If the value being loaded
4830 is of scalar type then the number of bytes read does not exceed the
4831 minimum number of bytes needed to hold all bits of the type. For
4832 example, loading an ``i24`` reads at most three bytes. When loading a
4833 value of a type like ``i20`` with a size that is not an integral number
4834 of bytes, the result is undefined if the value was not originally
4835 written using a store of the same type.
4840 .. code-block:: llvm
4842 %ptr = alloca i32 ; yields {i32*}:ptr
4843 store i32 3, i32* %ptr ; yields {void}
4844 %val = load i32* %ptr ; yields {i32}:val = i32 3
4848 '``store``' Instruction
4849 ^^^^^^^^^^^^^^^^^^^^^^^
4856 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4857 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4862 The '``store``' instruction is used to write to memory.
4867 There are two arguments to the ``store`` instruction: a value to store
4868 and an address at which to store it. The type of the ``<pointer>``
4869 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4870 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4871 then the optimizer is not allowed to modify the number or order of
4872 execution of this ``store`` with other :ref:`volatile
4873 operations <volatile>`.
4875 If the ``store`` is marked as ``atomic``, it takes an extra
4876 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4877 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4878 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4879 when they may see multiple atomic stores. The type of the pointee must
4880 be an integer type whose bit width is a power of two greater than or
4881 equal to eight and less than or equal to a target-specific size limit.
4882 ``align`` must be explicitly specified on atomic stores, and the store
4883 has undefined behavior if the alignment is not set to a value which is
4884 at least the size in bytes of the pointee. ``!nontemporal`` does not
4885 have any defined semantics for atomic stores.
4887 The optional constant ``align`` argument specifies the alignment of the
4888 operation (that is, the alignment of the memory address). A value of 0
4889 or an omitted ``align`` argument means that the operation has the ABI
4890 alignment for the target. It is the responsibility of the code emitter
4891 to ensure that the alignment information is correct. Overestimating the
4892 alignment results in undefined behavior. Underestimating the
4893 alignment may produce less efficient code. An alignment of 1 is always
4896 The optional ``!nontemporal`` metadata must reference a single metadata
4897 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4898 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4899 tells the optimizer and code generator that this load is not expected to
4900 be reused in the cache. The code generator may select special
4901 instructions to save cache bandwidth, such as the MOVNT instruction on
4907 The contents of memory are updated to contain ``<value>`` at the
4908 location specified by the ``<pointer>`` operand. If ``<value>`` is
4909 of scalar type then the number of bytes written does not exceed the
4910 minimum number of bytes needed to hold all bits of the type. For
4911 example, storing an ``i24`` writes at most three bytes. When writing a
4912 value of a type like ``i20`` with a size that is not an integral number
4913 of bytes, it is unspecified what happens to the extra bits that do not
4914 belong to the type, but they will typically be overwritten.
4919 .. code-block:: llvm
4921 %ptr = alloca i32 ; yields {i32*}:ptr
4922 store i32 3, i32* %ptr ; yields {void}
4923 %val = load i32* %ptr ; yields {i32}:val = i32 3
4927 '``fence``' Instruction
4928 ^^^^^^^^^^^^^^^^^^^^^^^
4935 fence [singlethread] <ordering> ; yields {void}
4940 The '``fence``' instruction is used to introduce happens-before edges
4946 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4947 defines what *synchronizes-with* edges they add. They can only be given
4948 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4953 A fence A which has (at least) ``release`` ordering semantics
4954 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4955 semantics if and only if there exist atomic operations X and Y, both
4956 operating on some atomic object M, such that A is sequenced before X, X
4957 modifies M (either directly or through some side effect of a sequence
4958 headed by X), Y is sequenced before B, and Y observes M. This provides a
4959 *happens-before* dependency between A and B. Rather than an explicit
4960 ``fence``, one (but not both) of the atomic operations X or Y might
4961 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4962 still *synchronize-with* the explicit ``fence`` and establish the
4963 *happens-before* edge.
4965 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4966 ``acquire`` and ``release`` semantics specified above, participates in
4967 the global program order of other ``seq_cst`` operations and/or fences.
4969 The optional ":ref:`singlethread <singlethread>`" argument specifies
4970 that the fence only synchronizes with other fences in the same thread.
4971 (This is useful for interacting with signal handlers.)
4976 .. code-block:: llvm
4978 fence acquire ; yields {void}
4979 fence singlethread seq_cst ; yields {void}
4983 '``cmpxchg``' Instruction
4984 ^^^^^^^^^^^^^^^^^^^^^^^^^
4991 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields {ty}
4996 The '``cmpxchg``' instruction is used to atomically modify memory. It
4997 loads a value in memory and compares it to a given value. If they are
4998 equal, it stores a new value into the memory.
5003 There are three arguments to the '``cmpxchg``' instruction: an address
5004 to operate on, a value to compare to the value currently be at that
5005 address, and a new value to place at that address if the compared values
5006 are equal. The type of '<cmp>' must be an integer type whose bit width
5007 is a power of two greater than or equal to eight and less than or equal
5008 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5009 type, and the type of '<pointer>' must be a pointer to that type. If the
5010 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5011 to modify the number or order of execution of this ``cmpxchg`` with
5012 other :ref:`volatile operations <volatile>`.
5014 The success and failure :ref:`ordering <ordering>` arguments specify how this
5015 ``cmpxchg`` synchronizes with other atomic operations. The both ordering
5016 parameters must be at least ``monotonic``, the ordering constraint on failure
5017 must be no stronger than that on success, and the failure ordering cannot be
5018 either ``release`` or ``acq_rel``.
5020 The optional "``singlethread``" argument declares that the ``cmpxchg``
5021 is only atomic with respect to code (usually signal handlers) running in
5022 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5023 respect to all other code in the system.
5025 The pointer passed into cmpxchg must have alignment greater than or
5026 equal to the size in memory of the operand.
5031 The contents of memory at the location specified by the '``<pointer>``'
5032 operand is read and compared to '``<cmp>``'; if the read value is the
5033 equal, '``<new>``' is written. The original value at the location is
5036 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5037 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5038 load with an ordering parameter determined the second ordering parameter.
5043 .. code-block:: llvm
5046 %orig = atomic load i32* %ptr unordered ; yields {i32}
5050 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5051 %squared = mul i32 %cmp, %cmp
5052 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields {i32}
5053 %success = icmp eq i32 %cmp, %old
5054 br i1 %success, label %done, label %loop
5061 '``atomicrmw``' Instruction
5062 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5069 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
5074 The '``atomicrmw``' instruction is used to atomically modify memory.
5079 There are three arguments to the '``atomicrmw``' instruction: an
5080 operation to apply, an address whose value to modify, an argument to the
5081 operation. The operation must be one of the following keywords:
5095 The type of '<value>' must be an integer type whose bit width is a power
5096 of two greater than or equal to eight and less than or equal to a
5097 target-specific size limit. The type of the '``<pointer>``' operand must
5098 be a pointer to that type. If the ``atomicrmw`` is marked as
5099 ``volatile``, then the optimizer is not allowed to modify the number or
5100 order of execution of this ``atomicrmw`` with other :ref:`volatile
5101 operations <volatile>`.
5106 The contents of memory at the location specified by the '``<pointer>``'
5107 operand are atomically read, modified, and written back. The original
5108 value at the location is returned. The modification is specified by the
5111 - xchg: ``*ptr = val``
5112 - add: ``*ptr = *ptr + val``
5113 - sub: ``*ptr = *ptr - val``
5114 - and: ``*ptr = *ptr & val``
5115 - nand: ``*ptr = ~(*ptr & val)``
5116 - or: ``*ptr = *ptr | val``
5117 - xor: ``*ptr = *ptr ^ val``
5118 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5119 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5120 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5122 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5128 .. code-block:: llvm
5130 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5132 .. _i_getelementptr:
5134 '``getelementptr``' Instruction
5135 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5142 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5143 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5144 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5149 The '``getelementptr``' instruction is used to get the address of a
5150 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5151 address calculation only and does not access memory.
5156 The first argument is always a pointer or a vector of pointers, and
5157 forms the basis of the calculation. The remaining arguments are indices
5158 that indicate which of the elements of the aggregate object are indexed.
5159 The interpretation of each index is dependent on the type being indexed
5160 into. The first index always indexes the pointer value given as the
5161 first argument, the second index indexes a value of the type pointed to
5162 (not necessarily the value directly pointed to, since the first index
5163 can be non-zero), etc. The first type indexed into must be a pointer
5164 value, subsequent types can be arrays, vectors, and structs. Note that
5165 subsequent types being indexed into can never be pointers, since that
5166 would require loading the pointer before continuing calculation.
5168 The type of each index argument depends on the type it is indexing into.
5169 When indexing into a (optionally packed) structure, only ``i32`` integer
5170 **constants** are allowed (when using a vector of indices they must all
5171 be the **same** ``i32`` integer constant). When indexing into an array,
5172 pointer or vector, integers of any width are allowed, and they are not
5173 required to be constant. These integers are treated as signed values
5176 For example, let's consider a C code fragment and how it gets compiled
5192 int *foo(struct ST *s) {
5193 return &s[1].Z.B[5][13];
5196 The LLVM code generated by Clang is:
5198 .. code-block:: llvm
5200 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5201 %struct.ST = type { i32, double, %struct.RT }
5203 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5205 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5212 In the example above, the first index is indexing into the
5213 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5214 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5215 indexes into the third element of the structure, yielding a
5216 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5217 structure. The third index indexes into the second element of the
5218 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5219 dimensions of the array are subscripted into, yielding an '``i32``'
5220 type. The '``getelementptr``' instruction returns a pointer to this
5221 element, thus computing a value of '``i32*``' type.
5223 Note that it is perfectly legal to index partially through a structure,
5224 returning a pointer to an inner element. Because of this, the LLVM code
5225 for the given testcase is equivalent to:
5227 .. code-block:: llvm
5229 define i32* @foo(%struct.ST* %s) {
5230 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5231 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5232 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5233 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5234 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5238 If the ``inbounds`` keyword is present, the result value of the
5239 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5240 pointer is not an *in bounds* address of an allocated object, or if any
5241 of the addresses that would be formed by successive addition of the
5242 offsets implied by the indices to the base address with infinitely
5243 precise signed arithmetic are not an *in bounds* address of that
5244 allocated object. The *in bounds* addresses for an allocated object are
5245 all the addresses that point into the object, plus the address one byte
5246 past the end. In cases where the base is a vector of pointers the
5247 ``inbounds`` keyword applies to each of the computations element-wise.
5249 If the ``inbounds`` keyword is not present, the offsets are added to the
5250 base address with silently-wrapping two's complement arithmetic. If the
5251 offsets have a different width from the pointer, they are sign-extended
5252 or truncated to the width of the pointer. The result value of the
5253 ``getelementptr`` may be outside the object pointed to by the base
5254 pointer. The result value may not necessarily be used to access memory
5255 though, even if it happens to point into allocated storage. See the
5256 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5259 The getelementptr instruction is often confusing. For some more insight
5260 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5265 .. code-block:: llvm
5267 ; yields [12 x i8]*:aptr
5268 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5270 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5272 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5274 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5276 In cases where the pointer argument is a vector of pointers, each index
5277 must be a vector with the same number of elements. For example:
5279 .. code-block:: llvm
5281 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5283 Conversion Operations
5284 ---------------------
5286 The instructions in this category are the conversion instructions
5287 (casting) which all take a single operand and a type. They perform
5288 various bit conversions on the operand.
5290 '``trunc .. to``' Instruction
5291 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5298 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5303 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5308 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5309 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5310 of the same number of integers. The bit size of the ``value`` must be
5311 larger than the bit size of the destination type, ``ty2``. Equal sized
5312 types are not allowed.
5317 The '``trunc``' instruction truncates the high order bits in ``value``
5318 and converts the remaining bits to ``ty2``. Since the source size must
5319 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5320 It will always truncate bits.
5325 .. code-block:: llvm
5327 %X = trunc i32 257 to i8 ; yields i8:1
5328 %Y = trunc i32 123 to i1 ; yields i1:true
5329 %Z = trunc i32 122 to i1 ; yields i1:false
5330 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5332 '``zext .. to``' Instruction
5333 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5340 <result> = zext <ty> <value> to <ty2> ; yields ty2
5345 The '``zext``' instruction zero extends its operand to type ``ty2``.
5350 The '``zext``' instruction takes a value to cast, and a type to cast it
5351 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5352 the same number of integers. The bit size of the ``value`` must be
5353 smaller than the bit size of the destination type, ``ty2``.
5358 The ``zext`` fills the high order bits of the ``value`` with zero bits
5359 until it reaches the size of the destination type, ``ty2``.
5361 When zero extending from i1, the result will always be either 0 or 1.
5366 .. code-block:: llvm
5368 %X = zext i32 257 to i64 ; yields i64:257
5369 %Y = zext i1 true to i32 ; yields i32:1
5370 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5372 '``sext .. to``' Instruction
5373 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5380 <result> = sext <ty> <value> to <ty2> ; yields ty2
5385 The '``sext``' sign extends ``value`` to the type ``ty2``.
5390 The '``sext``' instruction takes a value to cast, and a type to cast it
5391 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5392 the same number of integers. The bit size of the ``value`` must be
5393 smaller than the bit size of the destination type, ``ty2``.
5398 The '``sext``' instruction performs a sign extension by copying the sign
5399 bit (highest order bit) of the ``value`` until it reaches the bit size
5400 of the type ``ty2``.
5402 When sign extending from i1, the extension always results in -1 or 0.
5407 .. code-block:: llvm
5409 %X = sext i8 -1 to i16 ; yields i16 :65535
5410 %Y = sext i1 true to i32 ; yields i32:-1
5411 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5413 '``fptrunc .. to``' Instruction
5414 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5421 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5426 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5431 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5432 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5433 The size of ``value`` must be larger than the size of ``ty2``. This
5434 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5439 The '``fptrunc``' instruction truncates a ``value`` from a larger
5440 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5441 point <t_floating>` type. If the value cannot fit within the
5442 destination type, ``ty2``, then the results are undefined.
5447 .. code-block:: llvm
5449 %X = fptrunc double 123.0 to float ; yields float:123.0
5450 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5452 '``fpext .. to``' Instruction
5453 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5460 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5465 The '``fpext``' extends a floating point ``value`` to a larger floating
5471 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5472 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5473 to. The source type must be smaller than the destination type.
5478 The '``fpext``' instruction extends the ``value`` from a smaller
5479 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5480 point <t_floating>` type. The ``fpext`` cannot be used to make a
5481 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5482 *no-op cast* for a floating point cast.
5487 .. code-block:: llvm
5489 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5490 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5492 '``fptoui .. to``' Instruction
5493 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5500 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5505 The '``fptoui``' converts a floating point ``value`` to its unsigned
5506 integer equivalent of type ``ty2``.
5511 The '``fptoui``' instruction takes a value to cast, which must be a
5512 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5513 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5514 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5515 type with the same number of elements as ``ty``
5520 The '``fptoui``' instruction converts its :ref:`floating
5521 point <t_floating>` operand into the nearest (rounding towards zero)
5522 unsigned integer value. If the value cannot fit in ``ty2``, the results
5528 .. code-block:: llvm
5530 %X = fptoui double 123.0 to i32 ; yields i32:123
5531 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5532 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5534 '``fptosi .. to``' Instruction
5535 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5542 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5547 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5548 ``value`` to type ``ty2``.
5553 The '``fptosi``' instruction takes a value to cast, which must be a
5554 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5555 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5556 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5557 type with the same number of elements as ``ty``
5562 The '``fptosi``' instruction converts its :ref:`floating
5563 point <t_floating>` operand into the nearest (rounding towards zero)
5564 signed integer value. If the value cannot fit in ``ty2``, the results
5570 .. code-block:: llvm
5572 %X = fptosi double -123.0 to i32 ; yields i32:-123
5573 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5574 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5576 '``uitofp .. to``' Instruction
5577 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5584 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5589 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5590 and converts that value to the ``ty2`` type.
5595 The '``uitofp``' instruction takes a value to cast, which must be a
5596 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5597 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5598 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5599 type with the same number of elements as ``ty``
5604 The '``uitofp``' instruction interprets its operand as an unsigned
5605 integer quantity and converts it to the corresponding floating point
5606 value. If the value cannot fit in the floating point value, the results
5612 .. code-block:: llvm
5614 %X = uitofp i32 257 to float ; yields float:257.0
5615 %Y = uitofp i8 -1 to double ; yields double:255.0
5617 '``sitofp .. to``' Instruction
5618 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5625 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5630 The '``sitofp``' instruction regards ``value`` as a signed integer and
5631 converts that value to the ``ty2`` type.
5636 The '``sitofp``' instruction takes a value to cast, which must be a
5637 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5638 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5639 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5640 type with the same number of elements as ``ty``
5645 The '``sitofp``' instruction interprets its operand as a signed integer
5646 quantity and converts it to the corresponding floating point value. If
5647 the value cannot fit in the floating point value, the results are
5653 .. code-block:: llvm
5655 %X = sitofp i32 257 to float ; yields float:257.0
5656 %Y = sitofp i8 -1 to double ; yields double:-1.0
5660 '``ptrtoint .. to``' Instruction
5661 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5668 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5673 The '``ptrtoint``' instruction converts the pointer or a vector of
5674 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5679 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5680 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5681 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5682 a vector of integers type.
5687 The '``ptrtoint``' instruction converts ``value`` to integer type
5688 ``ty2`` by interpreting the pointer value as an integer and either
5689 truncating or zero extending that value to the size of the integer type.
5690 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5691 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5692 the same size, then nothing is done (*no-op cast*) other than a type
5698 .. code-block:: llvm
5700 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5701 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5702 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5706 '``inttoptr .. to``' Instruction
5707 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5714 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5719 The '``inttoptr``' instruction converts an integer ``value`` to a
5720 pointer type, ``ty2``.
5725 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5726 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5732 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5733 applying either a zero extension or a truncation depending on the size
5734 of the integer ``value``. If ``value`` is larger than the size of a
5735 pointer then a truncation is done. If ``value`` is smaller than the size
5736 of a pointer then a zero extension is done. If they are the same size,
5737 nothing is done (*no-op cast*).
5742 .. code-block:: llvm
5744 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5745 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5746 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5747 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5751 '``bitcast .. to``' Instruction
5752 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5759 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5764 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5770 The '``bitcast``' instruction takes a value to cast, which must be a
5771 non-aggregate first class value, and a type to cast it to, which must
5772 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5773 bit sizes of ``value`` and the destination type, ``ty2``, must be
5774 identical. If the source type is a pointer, the destination type must
5775 also be a pointer of the same size. This instruction supports bitwise
5776 conversion of vectors to integers and to vectors of other types (as
5777 long as they have the same size).
5782 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5783 is always a *no-op cast* because no bits change with this
5784 conversion. The conversion is done as if the ``value`` had been stored
5785 to memory and read back as type ``ty2``. Pointer (or vector of
5786 pointers) types may only be converted to other pointer (or vector of
5787 pointers) types with the same address space through this instruction.
5788 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5789 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5794 .. code-block:: llvm
5796 %X = bitcast i8 255 to i8 ; yields i8 :-1
5797 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5798 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5799 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5801 .. _i_addrspacecast:
5803 '``addrspacecast .. to``' Instruction
5804 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5811 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5816 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5817 address space ``n`` to type ``pty2`` in address space ``m``.
5822 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5823 to cast and a pointer type to cast it to, which must have a different
5829 The '``addrspacecast``' instruction converts the pointer value
5830 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5831 value modification, depending on the target and the address space
5832 pair. Pointer conversions within the same address space must be
5833 performed with the ``bitcast`` instruction. Note that if the address space
5834 conversion is legal then both result and operand refer to the same memory
5840 .. code-block:: llvm
5842 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5843 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5844 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5851 The instructions in this category are the "miscellaneous" instructions,
5852 which defy better classification.
5856 '``icmp``' Instruction
5857 ^^^^^^^^^^^^^^^^^^^^^^
5864 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5869 The '``icmp``' instruction returns a boolean value or a vector of
5870 boolean values based on comparison of its two integer, integer vector,
5871 pointer, or pointer vector operands.
5876 The '``icmp``' instruction takes three operands. The first operand is
5877 the condition code indicating the kind of comparison to perform. It is
5878 not a value, just a keyword. The possible condition code are:
5881 #. ``ne``: not equal
5882 #. ``ugt``: unsigned greater than
5883 #. ``uge``: unsigned greater or equal
5884 #. ``ult``: unsigned less than
5885 #. ``ule``: unsigned less or equal
5886 #. ``sgt``: signed greater than
5887 #. ``sge``: signed greater or equal
5888 #. ``slt``: signed less than
5889 #. ``sle``: signed less or equal
5891 The remaining two arguments must be :ref:`integer <t_integer>` or
5892 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5893 must also be identical types.
5898 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5899 code given as ``cond``. The comparison performed always yields either an
5900 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5902 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5903 otherwise. No sign interpretation is necessary or performed.
5904 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5905 otherwise. No sign interpretation is necessary or performed.
5906 #. ``ugt``: interprets the operands as unsigned values and yields
5907 ``true`` if ``op1`` is greater than ``op2``.
5908 #. ``uge``: interprets the operands as unsigned values and yields
5909 ``true`` if ``op1`` is greater than or equal to ``op2``.
5910 #. ``ult``: interprets the operands as unsigned values and yields
5911 ``true`` if ``op1`` is less than ``op2``.
5912 #. ``ule``: interprets the operands as unsigned values and yields
5913 ``true`` if ``op1`` is less than or equal to ``op2``.
5914 #. ``sgt``: interprets the operands as signed values and yields ``true``
5915 if ``op1`` is greater than ``op2``.
5916 #. ``sge``: interprets the operands as signed values and yields ``true``
5917 if ``op1`` is greater than or equal to ``op2``.
5918 #. ``slt``: interprets the operands as signed values and yields ``true``
5919 if ``op1`` is less than ``op2``.
5920 #. ``sle``: interprets the operands as signed values and yields ``true``
5921 if ``op1`` is less than or equal to ``op2``.
5923 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5924 are compared as if they were integers.
5926 If the operands are integer vectors, then they are compared element by
5927 element. The result is an ``i1`` vector with the same number of elements
5928 as the values being compared. Otherwise, the result is an ``i1``.
5933 .. code-block:: llvm
5935 <result> = icmp eq i32 4, 5 ; yields: result=false
5936 <result> = icmp ne float* %X, %X ; yields: result=false
5937 <result> = icmp ult i16 4, 5 ; yields: result=true
5938 <result> = icmp sgt i16 4, 5 ; yields: result=false
5939 <result> = icmp ule i16 -4, 5 ; yields: result=false
5940 <result> = icmp sge i16 4, 5 ; yields: result=false
5942 Note that the code generator does not yet support vector types with the
5943 ``icmp`` instruction.
5947 '``fcmp``' Instruction
5948 ^^^^^^^^^^^^^^^^^^^^^^
5955 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5960 The '``fcmp``' instruction returns a boolean value or vector of boolean
5961 values based on comparison of its operands.
5963 If the operands are floating point scalars, then the result type is a
5964 boolean (:ref:`i1 <t_integer>`).
5966 If the operands are floating point vectors, then the result type is a
5967 vector of boolean with the same number of elements as the operands being
5973 The '``fcmp``' instruction takes three operands. The first operand is
5974 the condition code indicating the kind of comparison to perform. It is
5975 not a value, just a keyword. The possible condition code are:
5977 #. ``false``: no comparison, always returns false
5978 #. ``oeq``: ordered and equal
5979 #. ``ogt``: ordered and greater than
5980 #. ``oge``: ordered and greater than or equal
5981 #. ``olt``: ordered and less than
5982 #. ``ole``: ordered and less than or equal
5983 #. ``one``: ordered and not equal
5984 #. ``ord``: ordered (no nans)
5985 #. ``ueq``: unordered or equal
5986 #. ``ugt``: unordered or greater than
5987 #. ``uge``: unordered or greater than or equal
5988 #. ``ult``: unordered or less than
5989 #. ``ule``: unordered or less than or equal
5990 #. ``une``: unordered or not equal
5991 #. ``uno``: unordered (either nans)
5992 #. ``true``: no comparison, always returns true
5994 *Ordered* means that neither operand is a QNAN while *unordered* means
5995 that either operand may be a QNAN.
5997 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5998 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5999 type. They must have identical types.
6004 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6005 condition code given as ``cond``. If the operands are vectors, then the
6006 vectors are compared element by element. Each comparison performed
6007 always yields an :ref:`i1 <t_integer>` result, as follows:
6009 #. ``false``: always yields ``false``, regardless of operands.
6010 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6011 is equal to ``op2``.
6012 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6013 is greater than ``op2``.
6014 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6015 is greater than or equal to ``op2``.
6016 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6017 is less than ``op2``.
6018 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6019 is less than or equal to ``op2``.
6020 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6021 is not equal to ``op2``.
6022 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6023 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6025 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6026 greater than ``op2``.
6027 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6028 greater than or equal to ``op2``.
6029 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6031 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6032 less than or equal to ``op2``.
6033 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6034 not equal to ``op2``.
6035 #. ``uno``: yields ``true`` if either operand is a QNAN.
6036 #. ``true``: always yields ``true``, regardless of operands.
6041 .. code-block:: llvm
6043 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6044 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6045 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6046 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6048 Note that the code generator does not yet support vector types with the
6049 ``fcmp`` instruction.
6053 '``phi``' Instruction
6054 ^^^^^^^^^^^^^^^^^^^^^
6061 <result> = phi <ty> [ <val0>, <label0>], ...
6066 The '``phi``' instruction is used to implement the φ node in the SSA
6067 graph representing the function.
6072 The type of the incoming values is specified with the first type field.
6073 After this, the '``phi``' instruction takes a list of pairs as
6074 arguments, with one pair for each predecessor basic block of the current
6075 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6076 the value arguments to the PHI node. Only labels may be used as the
6079 There must be no non-phi instructions between the start of a basic block
6080 and the PHI instructions: i.e. PHI instructions must be first in a basic
6083 For the purposes of the SSA form, the use of each incoming value is
6084 deemed to occur on the edge from the corresponding predecessor block to
6085 the current block (but after any definition of an '``invoke``'
6086 instruction's return value on the same edge).
6091 At runtime, the '``phi``' instruction logically takes on the value
6092 specified by the pair corresponding to the predecessor basic block that
6093 executed just prior to the current block.
6098 .. code-block:: llvm
6100 Loop: ; Infinite loop that counts from 0 on up...
6101 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6102 %nextindvar = add i32 %indvar, 1
6107 '``select``' Instruction
6108 ^^^^^^^^^^^^^^^^^^^^^^^^
6115 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6117 selty is either i1 or {<N x i1>}
6122 The '``select``' instruction is used to choose one value based on a
6123 condition, without branching.
6128 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6129 values indicating the condition, and two values of the same :ref:`first
6130 class <t_firstclass>` type. If the val1/val2 are vectors and the
6131 condition is a scalar, then entire vectors are selected, not individual
6137 If the condition is an i1 and it evaluates to 1, the instruction returns
6138 the first value argument; otherwise, it returns the second value
6141 If the condition is a vector of i1, then the value arguments must be
6142 vectors of the same size, and the selection is done element by element.
6147 .. code-block:: llvm
6149 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6153 '``call``' Instruction
6154 ^^^^^^^^^^^^^^^^^^^^^^
6161 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6166 The '``call``' instruction represents a simple function call.
6171 This instruction requires several arguments:
6173 #. The optional "tail" marker indicates that the callee function does
6174 not access any allocas or varargs in the caller. Note that calls may
6175 be marked "tail" even if they do not occur before a
6176 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6177 function call is eligible for tail call optimization, but `might not
6178 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6179 The code generator may optimize calls marked "tail" with either 1)
6180 automatic `sibling call
6181 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6182 callee have matching signatures, or 2) forced tail call optimization
6183 when the following extra requirements are met:
6185 - Caller and callee both have the calling convention ``fastcc``.
6186 - The call is in tail position (ret immediately follows call and ret
6187 uses value of call or is void).
6188 - Option ``-tailcallopt`` is enabled, or
6189 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6190 - `Platform specific constraints are
6191 met. <CodeGenerator.html#tailcallopt>`_
6193 #. The optional "cconv" marker indicates which :ref:`calling
6194 convention <callingconv>` the call should use. If none is
6195 specified, the call defaults to using C calling conventions. The
6196 calling convention of the call must match the calling convention of
6197 the target function, or else the behavior is undefined.
6198 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6199 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6201 #. '``ty``': the type of the call instruction itself which is also the
6202 type of the return value. Functions that return no value are marked
6204 #. '``fnty``': shall be the signature of the pointer to function value
6205 being invoked. The argument types must match the types implied by
6206 this signature. This type can be omitted if the function is not
6207 varargs and if the function type does not return a pointer to a
6209 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6210 be invoked. In most cases, this is a direct function invocation, but
6211 indirect ``call``'s are just as possible, calling an arbitrary pointer
6213 #. '``function args``': argument list whose types match the function
6214 signature argument types and parameter attributes. All arguments must
6215 be of :ref:`first class <t_firstclass>` type. If the function signature
6216 indicates the function accepts a variable number of arguments, the
6217 extra arguments can be specified.
6218 #. The optional :ref:`function attributes <fnattrs>` list. Only
6219 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6220 attributes are valid here.
6225 The '``call``' instruction is used to cause control flow to transfer to
6226 a specified function, with its incoming arguments bound to the specified
6227 values. Upon a '``ret``' instruction in the called function, control
6228 flow continues with the instruction after the function call, and the
6229 return value of the function is bound to the result argument.
6234 .. code-block:: llvm
6236 %retval = call i32 @test(i32 %argc)
6237 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6238 %X = tail call i32 @foo() ; yields i32
6239 %Y = tail call fastcc i32 @foo() ; yields i32
6240 call void %foo(i8 97 signext)
6242 %struct.A = type { i32, i8 }
6243 %r = call %struct.A @foo() ; yields { 32, i8 }
6244 %gr = extractvalue %struct.A %r, 0 ; yields i32
6245 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6246 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6247 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6249 llvm treats calls to some functions with names and arguments that match
6250 the standard C99 library as being the C99 library functions, and may
6251 perform optimizations or generate code for them under that assumption.
6252 This is something we'd like to change in the future to provide better
6253 support for freestanding environments and non-C-based languages.
6257 '``va_arg``' Instruction
6258 ^^^^^^^^^^^^^^^^^^^^^^^^
6265 <resultval> = va_arg <va_list*> <arglist>, <argty>
6270 The '``va_arg``' instruction is used to access arguments passed through
6271 the "variable argument" area of a function call. It is used to implement
6272 the ``va_arg`` macro in C.
6277 This instruction takes a ``va_list*`` value and the type of the
6278 argument. It returns a value of the specified argument type and
6279 increments the ``va_list`` to point to the next argument. The actual
6280 type of ``va_list`` is target specific.
6285 The '``va_arg``' instruction loads an argument of the specified type
6286 from the specified ``va_list`` and causes the ``va_list`` to point to
6287 the next argument. For more information, see the variable argument
6288 handling :ref:`Intrinsic Functions <int_varargs>`.
6290 It is legal for this instruction to be called in a function which does
6291 not take a variable number of arguments, for example, the ``vfprintf``
6294 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6295 function <intrinsics>` because it takes a type as an argument.
6300 See the :ref:`variable argument processing <int_varargs>` section.
6302 Note that the code generator does not yet fully support va\_arg on many
6303 targets. Also, it does not currently support va\_arg with aggregate
6304 types on any target.
6308 '``landingpad``' Instruction
6309 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6316 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6317 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6319 <clause> := catch <type> <value>
6320 <clause> := filter <array constant type> <array constant>
6325 The '``landingpad``' instruction is used by `LLVM's exception handling
6326 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6327 is a landing pad --- one where the exception lands, and corresponds to the
6328 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6329 defines values supplied by the personality function (``pers_fn``) upon
6330 re-entry to the function. The ``resultval`` has the type ``resultty``.
6335 This instruction takes a ``pers_fn`` value. This is the personality
6336 function associated with the unwinding mechanism. The optional
6337 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6339 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6340 contains the global variable representing the "type" that may be caught
6341 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6342 clause takes an array constant as its argument. Use
6343 "``[0 x i8**] undef``" for a filter which cannot throw. The
6344 '``landingpad``' instruction must contain *at least* one ``clause`` or
6345 the ``cleanup`` flag.
6350 The '``landingpad``' instruction defines the values which are set by the
6351 personality function (``pers_fn``) upon re-entry to the function, and
6352 therefore the "result type" of the ``landingpad`` instruction. As with
6353 calling conventions, how the personality function results are
6354 represented in LLVM IR is target specific.
6356 The clauses are applied in order from top to bottom. If two
6357 ``landingpad`` instructions are merged together through inlining, the
6358 clauses from the calling function are appended to the list of clauses.
6359 When the call stack is being unwound due to an exception being thrown,
6360 the exception is compared against each ``clause`` in turn. If it doesn't
6361 match any of the clauses, and the ``cleanup`` flag is not set, then
6362 unwinding continues further up the call stack.
6364 The ``landingpad`` instruction has several restrictions:
6366 - A landing pad block is a basic block which is the unwind destination
6367 of an '``invoke``' instruction.
6368 - A landing pad block must have a '``landingpad``' instruction as its
6369 first non-PHI instruction.
6370 - There can be only one '``landingpad``' instruction within the landing
6372 - A basic block that is not a landing pad block may not include a
6373 '``landingpad``' instruction.
6374 - All '``landingpad``' instructions in a function must have the same
6375 personality function.
6380 .. code-block:: llvm
6382 ;; A landing pad which can catch an integer.
6383 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6385 ;; A landing pad that is a cleanup.
6386 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6388 ;; A landing pad which can catch an integer and can only throw a double.
6389 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6391 filter [1 x i8**] [@_ZTId]
6398 LLVM supports the notion of an "intrinsic function". These functions
6399 have well known names and semantics and are required to follow certain
6400 restrictions. Overall, these intrinsics represent an extension mechanism
6401 for the LLVM language that does not require changing all of the
6402 transformations in LLVM when adding to the language (or the bitcode
6403 reader/writer, the parser, etc...).
6405 Intrinsic function names must all start with an "``llvm.``" prefix. This
6406 prefix is reserved in LLVM for intrinsic names; thus, function names may
6407 not begin with this prefix. Intrinsic functions must always be external
6408 functions: you cannot define the body of intrinsic functions. Intrinsic
6409 functions may only be used in call or invoke instructions: it is illegal
6410 to take the address of an intrinsic function. Additionally, because
6411 intrinsic functions are part of the LLVM language, it is required if any
6412 are added that they be documented here.
6414 Some intrinsic functions can be overloaded, i.e., the intrinsic
6415 represents a family of functions that perform the same operation but on
6416 different data types. Because LLVM can represent over 8 million
6417 different integer types, overloading is used commonly to allow an
6418 intrinsic function to operate on any integer type. One or more of the
6419 argument types or the result type can be overloaded to accept any
6420 integer type. Argument types may also be defined as exactly matching a
6421 previous argument's type or the result type. This allows an intrinsic
6422 function which accepts multiple arguments, but needs all of them to be
6423 of the same type, to only be overloaded with respect to a single
6424 argument or the result.
6426 Overloaded intrinsics will have the names of its overloaded argument
6427 types encoded into its function name, each preceded by a period. Only
6428 those types which are overloaded result in a name suffix. Arguments
6429 whose type is matched against another type do not. For example, the
6430 ``llvm.ctpop`` function can take an integer of any width and returns an
6431 integer of exactly the same integer width. This leads to a family of
6432 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6433 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6434 overloaded, and only one type suffix is required. Because the argument's
6435 type is matched against the return type, it does not require its own
6438 To learn how to add an intrinsic function, please see the `Extending
6439 LLVM Guide <ExtendingLLVM.html>`_.
6443 Variable Argument Handling Intrinsics
6444 -------------------------------------
6446 Variable argument support is defined in LLVM with the
6447 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6448 functions. These functions are related to the similarly named macros
6449 defined in the ``<stdarg.h>`` header file.
6451 All of these functions operate on arguments that use a target-specific
6452 value type "``va_list``". The LLVM assembly language reference manual
6453 does not define what this type is, so all transformations should be
6454 prepared to handle these functions regardless of the type used.
6456 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6457 variable argument handling intrinsic functions are used.
6459 .. code-block:: llvm
6461 define i32 @test(i32 %X, ...) {
6462 ; Initialize variable argument processing
6464 %ap2 = bitcast i8** %ap to i8*
6465 call void @llvm.va_start(i8* %ap2)
6467 ; Read a single integer argument
6468 %tmp = va_arg i8** %ap, i32
6470 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6472 %aq2 = bitcast i8** %aq to i8*
6473 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6474 call void @llvm.va_end(i8* %aq2)
6476 ; Stop processing of arguments.
6477 call void @llvm.va_end(i8* %ap2)
6481 declare void @llvm.va_start(i8*)
6482 declare void @llvm.va_copy(i8*, i8*)
6483 declare void @llvm.va_end(i8*)
6487 '``llvm.va_start``' Intrinsic
6488 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6495 declare void @llvm.va_start(i8* <arglist>)
6500 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6501 subsequent use by ``va_arg``.
6506 The argument is a pointer to a ``va_list`` element to initialize.
6511 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6512 available in C. In a target-dependent way, it initializes the
6513 ``va_list`` element to which the argument points, so that the next call
6514 to ``va_arg`` will produce the first variable argument passed to the
6515 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6516 to know the last argument of the function as the compiler can figure
6519 '``llvm.va_end``' Intrinsic
6520 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6527 declare void @llvm.va_end(i8* <arglist>)
6532 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6533 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6538 The argument is a pointer to a ``va_list`` to destroy.
6543 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6544 available in C. In a target-dependent way, it destroys the ``va_list``
6545 element to which the argument points. Calls to
6546 :ref:`llvm.va_start <int_va_start>` and
6547 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6552 '``llvm.va_copy``' Intrinsic
6553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6560 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6565 The '``llvm.va_copy``' intrinsic copies the current argument position
6566 from the source argument list to the destination argument list.
6571 The first argument is a pointer to a ``va_list`` element to initialize.
6572 The second argument is a pointer to a ``va_list`` element to copy from.
6577 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6578 available in C. In a target-dependent way, it copies the source
6579 ``va_list`` element into the destination ``va_list`` element. This
6580 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6581 arbitrarily complex and require, for example, memory allocation.
6583 Accurate Garbage Collection Intrinsics
6584 --------------------------------------
6586 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6587 (GC) requires the implementation and generation of these intrinsics.
6588 These intrinsics allow identification of :ref:`GC roots on the
6589 stack <int_gcroot>`, as well as garbage collector implementations that
6590 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6591 Front-ends for type-safe garbage collected languages should generate
6592 these intrinsics to make use of the LLVM garbage collectors. For more
6593 details, see `Accurate Garbage Collection with
6594 LLVM <GarbageCollection.html>`_.
6596 The garbage collection intrinsics only operate on objects in the generic
6597 address space (address space zero).
6601 '``llvm.gcroot``' Intrinsic
6602 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6609 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6614 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6615 the code generator, and allows some metadata to be associated with it.
6620 The first argument specifies the address of a stack object that contains
6621 the root pointer. The second pointer (which must be either a constant or
6622 a global value address) contains the meta-data to be associated with the
6628 At runtime, a call to this intrinsic stores a null pointer into the
6629 "ptrloc" location. At compile-time, the code generator generates
6630 information to allow the runtime to find the pointer at GC safe points.
6631 The '``llvm.gcroot``' intrinsic may only be used in a function which
6632 :ref:`specifies a GC algorithm <gc>`.
6636 '``llvm.gcread``' Intrinsic
6637 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6644 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6649 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6650 locations, allowing garbage collector implementations that require read
6656 The second argument is the address to read from, which should be an
6657 address allocated from the garbage collector. The first object is a
6658 pointer to the start of the referenced object, if needed by the language
6659 runtime (otherwise null).
6664 The '``llvm.gcread``' intrinsic has the same semantics as a load
6665 instruction, but may be replaced with substantially more complex code by
6666 the garbage collector runtime, as needed. The '``llvm.gcread``'
6667 intrinsic may only be used in a function which :ref:`specifies a GC
6672 '``llvm.gcwrite``' Intrinsic
6673 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6680 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6685 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6686 locations, allowing garbage collector implementations that require write
6687 barriers (such as generational or reference counting collectors).
6692 The first argument is the reference to store, the second is the start of
6693 the object to store it to, and the third is the address of the field of
6694 Obj to store to. If the runtime does not require a pointer to the
6695 object, Obj may be null.
6700 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6701 instruction, but may be replaced with substantially more complex code by
6702 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6703 intrinsic may only be used in a function which :ref:`specifies a GC
6706 Code Generator Intrinsics
6707 -------------------------
6709 These intrinsics are provided by LLVM to expose special features that
6710 may only be implemented with code generator support.
6712 '``llvm.returnaddress``' Intrinsic
6713 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6720 declare i8 *@llvm.returnaddress(i32 <level>)
6725 The '``llvm.returnaddress``' intrinsic attempts to compute a
6726 target-specific value indicating the return address of the current
6727 function or one of its callers.
6732 The argument to this intrinsic indicates which function to return the
6733 address for. Zero indicates the calling function, one indicates its
6734 caller, etc. The argument is **required** to be a constant integer
6740 The '``llvm.returnaddress``' intrinsic either returns a pointer
6741 indicating the return address of the specified call frame, or zero if it
6742 cannot be identified. The value returned by this intrinsic is likely to
6743 be incorrect or 0 for arguments other than zero, so it should only be
6744 used for debugging purposes.
6746 Note that calling this intrinsic does not prevent function inlining or
6747 other aggressive transformations, so the value returned may not be that
6748 of the obvious source-language caller.
6750 '``llvm.frameaddress``' Intrinsic
6751 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6758 declare i8* @llvm.frameaddress(i32 <level>)
6763 The '``llvm.frameaddress``' intrinsic attempts to return the
6764 target-specific frame pointer value for the specified stack frame.
6769 The argument to this intrinsic indicates which function to return the
6770 frame pointer for. Zero indicates the calling function, one indicates
6771 its caller, etc. The argument is **required** to be a constant integer
6777 The '``llvm.frameaddress``' intrinsic either returns a pointer
6778 indicating the frame address of the specified call frame, or zero if it
6779 cannot be identified. The value returned by this intrinsic is likely to
6780 be incorrect or 0 for arguments other than zero, so it should only be
6781 used for debugging purposes.
6783 Note that calling this intrinsic does not prevent function inlining or
6784 other aggressive transformations, so the value returned may not be that
6785 of the obvious source-language caller.
6789 '``llvm.stacksave``' Intrinsic
6790 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6797 declare i8* @llvm.stacksave()
6802 The '``llvm.stacksave``' intrinsic is used to remember the current state
6803 of the function stack, for use with
6804 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6805 implementing language features like scoped automatic variable sized
6811 This intrinsic returns a opaque pointer value that can be passed to
6812 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6813 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6814 ``llvm.stacksave``, it effectively restores the state of the stack to
6815 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6816 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6817 were allocated after the ``llvm.stacksave`` was executed.
6819 .. _int_stackrestore:
6821 '``llvm.stackrestore``' Intrinsic
6822 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6829 declare void @llvm.stackrestore(i8* %ptr)
6834 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6835 the function stack to the state it was in when the corresponding
6836 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6837 useful for implementing language features like scoped automatic variable
6838 sized arrays in C99.
6843 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6845 '``llvm.prefetch``' Intrinsic
6846 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6853 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6858 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6859 insert a prefetch instruction if supported; otherwise, it is a noop.
6860 Prefetches have no effect on the behavior of the program but can change
6861 its performance characteristics.
6866 ``address`` is the address to be prefetched, ``rw`` is the specifier
6867 determining if the fetch should be for a read (0) or write (1), and
6868 ``locality`` is a temporal locality specifier ranging from (0) - no
6869 locality, to (3) - extremely local keep in cache. The ``cache type``
6870 specifies whether the prefetch is performed on the data (1) or
6871 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6872 arguments must be constant integers.
6877 This intrinsic does not modify the behavior of the program. In
6878 particular, prefetches cannot trap and do not produce a value. On
6879 targets that support this intrinsic, the prefetch can provide hints to
6880 the processor cache for better performance.
6882 '``llvm.pcmarker``' Intrinsic
6883 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6890 declare void @llvm.pcmarker(i32 <id>)
6895 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6896 Counter (PC) in a region of code to simulators and other tools. The
6897 method is target specific, but it is expected that the marker will use
6898 exported symbols to transmit the PC of the marker. The marker makes no
6899 guarantees that it will remain with any specific instruction after
6900 optimizations. It is possible that the presence of a marker will inhibit
6901 optimizations. The intended use is to be inserted after optimizations to
6902 allow correlations of simulation runs.
6907 ``id`` is a numerical id identifying the marker.
6912 This intrinsic does not modify the behavior of the program. Backends
6913 that do not support this intrinsic may ignore it.
6915 '``llvm.readcyclecounter``' Intrinsic
6916 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6923 declare i64 @llvm.readcyclecounter()
6928 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6929 counter register (or similar low latency, high accuracy clocks) on those
6930 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6931 should map to RPCC. As the backing counters overflow quickly (on the
6932 order of 9 seconds on alpha), this should only be used for small
6938 When directly supported, reading the cycle counter should not modify any
6939 memory. Implementations are allowed to either return a application
6940 specific value or a system wide value. On backends without support, this
6941 is lowered to a constant 0.
6943 Note that runtime support may be conditional on the privilege-level code is
6944 running at and the host platform.
6946 Standard C Library Intrinsics
6947 -----------------------------
6949 LLVM provides intrinsics for a few important standard C library
6950 functions. These intrinsics allow source-language front-ends to pass
6951 information about the alignment of the pointer arguments to the code
6952 generator, providing opportunity for more efficient code generation.
6956 '``llvm.memcpy``' Intrinsic
6957 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6962 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6963 integer bit width and for different address spaces. Not all targets
6964 support all bit widths however.
6968 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6969 i32 <len>, i32 <align>, i1 <isvolatile>)
6970 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6971 i64 <len>, i32 <align>, i1 <isvolatile>)
6976 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6977 source location to the destination location.
6979 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6980 intrinsics do not return a value, takes extra alignment/isvolatile
6981 arguments and the pointers can be in specified address spaces.
6986 The first argument is a pointer to the destination, the second is a
6987 pointer to the source. The third argument is an integer argument
6988 specifying the number of bytes to copy, the fourth argument is the
6989 alignment of the source and destination locations, and the fifth is a
6990 boolean indicating a volatile access.
6992 If the call to this intrinsic has an alignment value that is not 0 or 1,
6993 then the caller guarantees that both the source and destination pointers
6994 are aligned to that boundary.
6996 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6997 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6998 very cleanly specified and it is unwise to depend on it.
7003 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7004 source location to the destination location, which are not allowed to
7005 overlap. It copies "len" bytes of memory over. If the argument is known
7006 to be aligned to some boundary, this can be specified as the fourth
7007 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
7009 '``llvm.memmove``' Intrinsic
7010 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7015 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7016 bit width and for different address space. Not all targets support all
7021 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7022 i32 <len>, i32 <align>, i1 <isvolatile>)
7023 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7024 i64 <len>, i32 <align>, i1 <isvolatile>)
7029 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7030 source location to the destination location. It is similar to the
7031 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7034 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7035 intrinsics do not return a value, takes extra alignment/isvolatile
7036 arguments and the pointers can be in specified address spaces.
7041 The first argument is a pointer to the destination, the second is a
7042 pointer to the source. The third argument is an integer argument
7043 specifying the number of bytes to copy, the fourth argument is the
7044 alignment of the source and destination locations, and the fifth is a
7045 boolean indicating a volatile access.
7047 If the call to this intrinsic has an alignment value that is not 0 or 1,
7048 then the caller guarantees that the source and destination pointers are
7049 aligned to that boundary.
7051 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7052 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7053 not very cleanly specified and it is unwise to depend on it.
7058 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7059 source location to the destination location, which may overlap. It
7060 copies "len" bytes of memory over. If the argument is known to be
7061 aligned to some boundary, this can be specified as the fourth argument,
7062 otherwise it should be set to 0 or 1 (both meaning no alignment).
7064 '``llvm.memset.*``' Intrinsics
7065 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7070 This is an overloaded intrinsic. You can use llvm.memset on any integer
7071 bit width and for different address spaces. However, not all targets
7072 support all bit widths.
7076 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7077 i32 <len>, i32 <align>, i1 <isvolatile>)
7078 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7079 i64 <len>, i32 <align>, i1 <isvolatile>)
7084 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7085 particular byte value.
7087 Note that, unlike the standard libc function, the ``llvm.memset``
7088 intrinsic does not return a value and takes extra alignment/volatile
7089 arguments. Also, the destination can be in an arbitrary address space.
7094 The first argument is a pointer to the destination to fill, the second
7095 is the byte value with which to fill it, the third argument is an
7096 integer argument specifying the number of bytes to fill, and the fourth
7097 argument is the known alignment of the destination location.
7099 If the call to this intrinsic has an alignment value that is not 0 or 1,
7100 then the caller guarantees that the destination pointer is aligned to
7103 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7104 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7105 very cleanly specified and it is unwise to depend on it.
7110 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7111 at the destination location. If the argument is known to be aligned to
7112 some boundary, this can be specified as the fourth argument, otherwise
7113 it should be set to 0 or 1 (both meaning no alignment).
7115 '``llvm.sqrt.*``' Intrinsic
7116 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7121 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7122 floating point or vector of floating point type. Not all targets support
7127 declare float @llvm.sqrt.f32(float %Val)
7128 declare double @llvm.sqrt.f64(double %Val)
7129 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7130 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7131 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7136 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7137 returning the same value as the libm '``sqrt``' functions would. Unlike
7138 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7139 negative numbers other than -0.0 (which allows for better optimization,
7140 because there is no need to worry about errno being set).
7141 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7146 The argument and return value are floating point numbers of the same
7152 This function returns the sqrt of the specified operand if it is a
7153 nonnegative floating point number.
7155 '``llvm.powi.*``' Intrinsic
7156 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7161 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7162 floating point or vector of floating point type. Not all targets support
7167 declare float @llvm.powi.f32(float %Val, i32 %power)
7168 declare double @llvm.powi.f64(double %Val, i32 %power)
7169 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7170 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7171 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7176 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7177 specified (positive or negative) power. The order of evaluation of
7178 multiplications is not defined. When a vector of floating point type is
7179 used, the second argument remains a scalar integer value.
7184 The second argument is an integer power, and the first is a value to
7185 raise to that power.
7190 This function returns the first value raised to the second power with an
7191 unspecified sequence of rounding operations.
7193 '``llvm.sin.*``' Intrinsic
7194 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7199 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7200 floating point or vector of floating point type. Not all targets support
7205 declare float @llvm.sin.f32(float %Val)
7206 declare double @llvm.sin.f64(double %Val)
7207 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7208 declare fp128 @llvm.sin.f128(fp128 %Val)
7209 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7214 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7219 The argument and return value are floating point numbers of the same
7225 This function returns the sine of the specified operand, returning the
7226 same values as the libm ``sin`` functions would, and handles error
7227 conditions in the same way.
7229 '``llvm.cos.*``' Intrinsic
7230 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7235 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7236 floating point or vector of floating point type. Not all targets support
7241 declare float @llvm.cos.f32(float %Val)
7242 declare double @llvm.cos.f64(double %Val)
7243 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7244 declare fp128 @llvm.cos.f128(fp128 %Val)
7245 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7250 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7255 The argument and return value are floating point numbers of the same
7261 This function returns the cosine of the specified operand, returning the
7262 same values as the libm ``cos`` functions would, and handles error
7263 conditions in the same way.
7265 '``llvm.pow.*``' Intrinsic
7266 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7271 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7272 floating point or vector of floating point type. Not all targets support
7277 declare float @llvm.pow.f32(float %Val, float %Power)
7278 declare double @llvm.pow.f64(double %Val, double %Power)
7279 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7280 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7281 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7286 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7287 specified (positive or negative) power.
7292 The second argument is a floating point power, and the first is a value
7293 to raise to that power.
7298 This function returns the first value raised to the second power,
7299 returning the same values as the libm ``pow`` functions would, and
7300 handles error conditions in the same way.
7302 '``llvm.exp.*``' Intrinsic
7303 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7308 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7309 floating point or vector of floating point type. Not all targets support
7314 declare float @llvm.exp.f32(float %Val)
7315 declare double @llvm.exp.f64(double %Val)
7316 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7317 declare fp128 @llvm.exp.f128(fp128 %Val)
7318 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7323 The '``llvm.exp.*``' intrinsics perform the exp function.
7328 The argument and return value are floating point numbers of the same
7334 This function returns the same values as the libm ``exp`` functions
7335 would, and handles error conditions in the same way.
7337 '``llvm.exp2.*``' Intrinsic
7338 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7343 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7344 floating point or vector of floating point type. Not all targets support
7349 declare float @llvm.exp2.f32(float %Val)
7350 declare double @llvm.exp2.f64(double %Val)
7351 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7352 declare fp128 @llvm.exp2.f128(fp128 %Val)
7353 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7358 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7363 The argument and return value are floating point numbers of the same
7369 This function returns the same values as the libm ``exp2`` functions
7370 would, and handles error conditions in the same way.
7372 '``llvm.log.*``' Intrinsic
7373 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7378 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7379 floating point or vector of floating point type. Not all targets support
7384 declare float @llvm.log.f32(float %Val)
7385 declare double @llvm.log.f64(double %Val)
7386 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7387 declare fp128 @llvm.log.f128(fp128 %Val)
7388 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7393 The '``llvm.log.*``' intrinsics perform the log function.
7398 The argument and return value are floating point numbers of the same
7404 This function returns the same values as the libm ``log`` functions
7405 would, and handles error conditions in the same way.
7407 '``llvm.log10.*``' Intrinsic
7408 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7413 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7414 floating point or vector of floating point type. Not all targets support
7419 declare float @llvm.log10.f32(float %Val)
7420 declare double @llvm.log10.f64(double %Val)
7421 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7422 declare fp128 @llvm.log10.f128(fp128 %Val)
7423 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7428 The '``llvm.log10.*``' intrinsics perform the log10 function.
7433 The argument and return value are floating point numbers of the same
7439 This function returns the same values as the libm ``log10`` functions
7440 would, and handles error conditions in the same way.
7442 '``llvm.log2.*``' Intrinsic
7443 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7448 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7449 floating point or vector of floating point type. Not all targets support
7454 declare float @llvm.log2.f32(float %Val)
7455 declare double @llvm.log2.f64(double %Val)
7456 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7457 declare fp128 @llvm.log2.f128(fp128 %Val)
7458 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7463 The '``llvm.log2.*``' intrinsics perform the log2 function.
7468 The argument and return value are floating point numbers of the same
7474 This function returns the same values as the libm ``log2`` functions
7475 would, and handles error conditions in the same way.
7477 '``llvm.fma.*``' Intrinsic
7478 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7483 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7484 floating point or vector of floating point type. Not all targets support
7489 declare float @llvm.fma.f32(float %a, float %b, float %c)
7490 declare double @llvm.fma.f64(double %a, double %b, double %c)
7491 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7492 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7493 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7498 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7504 The argument and return value are floating point numbers of the same
7510 This function returns the same values as the libm ``fma`` functions
7511 would, and does not set errno.
7513 '``llvm.fabs.*``' Intrinsic
7514 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7519 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7520 floating point or vector of floating point type. Not all targets support
7525 declare float @llvm.fabs.f32(float %Val)
7526 declare double @llvm.fabs.f64(double %Val)
7527 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7528 declare fp128 @llvm.fabs.f128(fp128 %Val)
7529 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7534 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7540 The argument and return value are floating point numbers of the same
7546 This function returns the same values as the libm ``fabs`` functions
7547 would, and handles error conditions in the same way.
7549 '``llvm.copysign.*``' Intrinsic
7550 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7555 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7556 floating point or vector of floating point type. Not all targets support
7561 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7562 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7563 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7564 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7565 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7570 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7571 first operand and the sign of the second operand.
7576 The arguments and return value are floating point numbers of the same
7582 This function returns the same values as the libm ``copysign``
7583 functions would, and handles error conditions in the same way.
7585 '``llvm.floor.*``' Intrinsic
7586 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7591 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7592 floating point or vector of floating point type. Not all targets support
7597 declare float @llvm.floor.f32(float %Val)
7598 declare double @llvm.floor.f64(double %Val)
7599 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7600 declare fp128 @llvm.floor.f128(fp128 %Val)
7601 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7606 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7611 The argument and return value are floating point numbers of the same
7617 This function returns the same values as the libm ``floor`` functions
7618 would, and handles error conditions in the same way.
7620 '``llvm.ceil.*``' Intrinsic
7621 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7626 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7627 floating point or vector of floating point type. Not all targets support
7632 declare float @llvm.ceil.f32(float %Val)
7633 declare double @llvm.ceil.f64(double %Val)
7634 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7635 declare fp128 @llvm.ceil.f128(fp128 %Val)
7636 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7641 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7646 The argument and return value are floating point numbers of the same
7652 This function returns the same values as the libm ``ceil`` functions
7653 would, and handles error conditions in the same way.
7655 '``llvm.trunc.*``' Intrinsic
7656 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7661 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7662 floating point or vector of floating point type. Not all targets support
7667 declare float @llvm.trunc.f32(float %Val)
7668 declare double @llvm.trunc.f64(double %Val)
7669 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7670 declare fp128 @llvm.trunc.f128(fp128 %Val)
7671 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7676 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7677 nearest integer not larger in magnitude than the operand.
7682 The argument and return value are floating point numbers of the same
7688 This function returns the same values as the libm ``trunc`` functions
7689 would, and handles error conditions in the same way.
7691 '``llvm.rint.*``' Intrinsic
7692 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7697 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7698 floating point or vector of floating point type. Not all targets support
7703 declare float @llvm.rint.f32(float %Val)
7704 declare double @llvm.rint.f64(double %Val)
7705 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7706 declare fp128 @llvm.rint.f128(fp128 %Val)
7707 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7712 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7713 nearest integer. It may raise an inexact floating-point exception if the
7714 operand isn't an integer.
7719 The argument and return value are floating point numbers of the same
7725 This function returns the same values as the libm ``rint`` functions
7726 would, and handles error conditions in the same way.
7728 '``llvm.nearbyint.*``' Intrinsic
7729 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7734 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7735 floating point or vector of floating point type. Not all targets support
7740 declare float @llvm.nearbyint.f32(float %Val)
7741 declare double @llvm.nearbyint.f64(double %Val)
7742 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7743 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7744 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7749 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7755 The argument and return value are floating point numbers of the same
7761 This function returns the same values as the libm ``nearbyint``
7762 functions would, and handles error conditions in the same way.
7764 '``llvm.round.*``' Intrinsic
7765 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7770 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7771 floating point or vector of floating point type. Not all targets support
7776 declare float @llvm.round.f32(float %Val)
7777 declare double @llvm.round.f64(double %Val)
7778 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7779 declare fp128 @llvm.round.f128(fp128 %Val)
7780 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7785 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7791 The argument and return value are floating point numbers of the same
7797 This function returns the same values as the libm ``round``
7798 functions would, and handles error conditions in the same way.
7800 Bit Manipulation Intrinsics
7801 ---------------------------
7803 LLVM provides intrinsics for a few important bit manipulation
7804 operations. These allow efficient code generation for some algorithms.
7806 '``llvm.bswap.*``' Intrinsics
7807 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7812 This is an overloaded intrinsic function. You can use bswap on any
7813 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7817 declare i16 @llvm.bswap.i16(i16 <id>)
7818 declare i32 @llvm.bswap.i32(i32 <id>)
7819 declare i64 @llvm.bswap.i64(i64 <id>)
7824 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7825 values with an even number of bytes (positive multiple of 16 bits).
7826 These are useful for performing operations on data that is not in the
7827 target's native byte order.
7832 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7833 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7834 intrinsic returns an i32 value that has the four bytes of the input i32
7835 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7836 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7837 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7838 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7841 '``llvm.ctpop.*``' Intrinsic
7842 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7847 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7848 bit width, or on any vector with integer elements. Not all targets
7849 support all bit widths or vector types, however.
7853 declare i8 @llvm.ctpop.i8(i8 <src>)
7854 declare i16 @llvm.ctpop.i16(i16 <src>)
7855 declare i32 @llvm.ctpop.i32(i32 <src>)
7856 declare i64 @llvm.ctpop.i64(i64 <src>)
7857 declare i256 @llvm.ctpop.i256(i256 <src>)
7858 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7863 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7869 The only argument is the value to be counted. The argument may be of any
7870 integer type, or a vector with integer elements. The return type must
7871 match the argument type.
7876 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7877 each element of a vector.
7879 '``llvm.ctlz.*``' Intrinsic
7880 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7885 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7886 integer bit width, or any vector whose elements are integers. Not all
7887 targets support all bit widths or vector types, however.
7891 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7892 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7893 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7894 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7895 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7896 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7901 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7902 leading zeros in a variable.
7907 The first argument is the value to be counted. This argument may be of
7908 any integer type, or a vectory with integer element type. The return
7909 type must match the first argument type.
7911 The second argument must be a constant and is a flag to indicate whether
7912 the intrinsic should ensure that a zero as the first argument produces a
7913 defined result. Historically some architectures did not provide a
7914 defined result for zero values as efficiently, and many algorithms are
7915 now predicated on avoiding zero-value inputs.
7920 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7921 zeros in a variable, or within each element of the vector. If
7922 ``src == 0`` then the result is the size in bits of the type of ``src``
7923 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7924 ``llvm.ctlz(i32 2) = 30``.
7926 '``llvm.cttz.*``' Intrinsic
7927 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7932 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7933 integer bit width, or any vector of integer elements. Not all targets
7934 support all bit widths or vector types, however.
7938 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7939 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7940 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7941 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7942 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7943 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7948 The '``llvm.cttz``' family of intrinsic functions counts the number of
7954 The first argument is the value to be counted. This argument may be of
7955 any integer type, or a vectory with integer element type. The return
7956 type must match the first argument type.
7958 The second argument must be a constant and is a flag to indicate whether
7959 the intrinsic should ensure that a zero as the first argument produces a
7960 defined result. Historically some architectures did not provide a
7961 defined result for zero values as efficiently, and many algorithms are
7962 now predicated on avoiding zero-value inputs.
7967 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7968 zeros in a variable, or within each element of a vector. If ``src == 0``
7969 then the result is the size in bits of the type of ``src`` if
7970 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7971 ``llvm.cttz(2) = 1``.
7973 Arithmetic with Overflow Intrinsics
7974 -----------------------------------
7976 LLVM provides intrinsics for some arithmetic with overflow operations.
7978 '``llvm.sadd.with.overflow.*``' Intrinsics
7979 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7984 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7985 on any integer bit width.
7989 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7990 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7991 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7996 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7997 a signed addition of the two arguments, and indicate whether an overflow
7998 occurred during the signed summation.
8003 The arguments (%a and %b) and the first element of the result structure
8004 may be of integer types of any bit width, but they must have the same
8005 bit width. The second element of the result structure must be of type
8006 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8012 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8013 a signed addition of the two variables. They return a structure --- the
8014 first element of which is the signed summation, and the second element
8015 of which is a bit specifying if the signed summation resulted in an
8021 .. code-block:: llvm
8023 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8024 %sum = extractvalue {i32, i1} %res, 0
8025 %obit = extractvalue {i32, i1} %res, 1
8026 br i1 %obit, label %overflow, label %normal
8028 '``llvm.uadd.with.overflow.*``' Intrinsics
8029 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8034 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8035 on any integer bit width.
8039 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8040 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8041 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8046 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8047 an unsigned addition of the two arguments, and indicate whether a carry
8048 occurred during the unsigned summation.
8053 The arguments (%a and %b) and the first element of the result structure
8054 may be of integer types of any bit width, but they must have the same
8055 bit width. The second element of the result structure must be of type
8056 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8062 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8063 an unsigned addition of the two arguments. They return a structure --- the
8064 first element of which is the sum, and the second element of which is a
8065 bit specifying if the unsigned summation resulted in a carry.
8070 .. code-block:: llvm
8072 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8073 %sum = extractvalue {i32, i1} %res, 0
8074 %obit = extractvalue {i32, i1} %res, 1
8075 br i1 %obit, label %carry, label %normal
8077 '``llvm.ssub.with.overflow.*``' Intrinsics
8078 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8083 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8084 on any integer bit width.
8088 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8089 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8090 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8095 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8096 a signed subtraction of the two arguments, and indicate whether an
8097 overflow occurred during the signed subtraction.
8102 The arguments (%a and %b) and the first element of the result structure
8103 may be of integer types of any bit width, but they must have the same
8104 bit width. The second element of the result structure must be of type
8105 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8111 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8112 a signed subtraction of the two arguments. They return a structure --- the
8113 first element of which is the subtraction, and the second element of
8114 which is a bit specifying if the signed subtraction resulted in an
8120 .. code-block:: llvm
8122 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8123 %sum = extractvalue {i32, i1} %res, 0
8124 %obit = extractvalue {i32, i1} %res, 1
8125 br i1 %obit, label %overflow, label %normal
8127 '``llvm.usub.with.overflow.*``' Intrinsics
8128 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8133 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8134 on any integer bit width.
8138 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8139 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8140 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8145 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8146 an unsigned subtraction of the two arguments, and indicate whether an
8147 overflow occurred during the unsigned subtraction.
8152 The arguments (%a and %b) and the first element of the result structure
8153 may be of integer types of any bit width, but they must have the same
8154 bit width. The second element of the result structure must be of type
8155 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8161 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8162 an unsigned subtraction of the two arguments. They return a structure ---
8163 the first element of which is the subtraction, and the second element of
8164 which is a bit specifying if the unsigned subtraction resulted in an
8170 .. code-block:: llvm
8172 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8173 %sum = extractvalue {i32, i1} %res, 0
8174 %obit = extractvalue {i32, i1} %res, 1
8175 br i1 %obit, label %overflow, label %normal
8177 '``llvm.smul.with.overflow.*``' Intrinsics
8178 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8183 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8184 on any integer bit width.
8188 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8189 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8190 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8195 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8196 a signed multiplication of the two arguments, and indicate whether an
8197 overflow occurred during the signed multiplication.
8202 The arguments (%a and %b) and the first element of the result structure
8203 may be of integer types of any bit width, but they must have the same
8204 bit width. The second element of the result structure must be of type
8205 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8211 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8212 a signed multiplication of the two arguments. They return a structure ---
8213 the first element of which is the multiplication, and the second element
8214 of which is a bit specifying if the signed multiplication resulted in an
8220 .. code-block:: llvm
8222 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8223 %sum = extractvalue {i32, i1} %res, 0
8224 %obit = extractvalue {i32, i1} %res, 1
8225 br i1 %obit, label %overflow, label %normal
8227 '``llvm.umul.with.overflow.*``' Intrinsics
8228 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8233 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8234 on any integer bit width.
8238 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8239 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8240 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8245 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8246 a unsigned multiplication of the two arguments, and indicate whether an
8247 overflow occurred during the unsigned multiplication.
8252 The arguments (%a and %b) and the first element of the result structure
8253 may be of integer types of any bit width, but they must have the same
8254 bit width. The second element of the result structure must be of type
8255 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8261 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8262 an unsigned multiplication of the two arguments. They return a structure ---
8263 the first element of which is the multiplication, and the second
8264 element of which is a bit specifying if the unsigned multiplication
8265 resulted in an overflow.
8270 .. code-block:: llvm
8272 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8273 %sum = extractvalue {i32, i1} %res, 0
8274 %obit = extractvalue {i32, i1} %res, 1
8275 br i1 %obit, label %overflow, label %normal
8277 Specialised Arithmetic Intrinsics
8278 ---------------------------------
8280 '``llvm.fmuladd.*``' Intrinsic
8281 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8288 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8289 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8294 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8295 expressions that can be fused if the code generator determines that (a) the
8296 target instruction set has support for a fused operation, and (b) that the
8297 fused operation is more efficient than the equivalent, separate pair of mul
8298 and add instructions.
8303 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8304 multiplicands, a and b, and an addend c.
8313 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8315 is equivalent to the expression a \* b + c, except that rounding will
8316 not be performed between the multiplication and addition steps if the
8317 code generator fuses the operations. Fusion is not guaranteed, even if
8318 the target platform supports it. If a fused multiply-add is required the
8319 corresponding llvm.fma.\* intrinsic function should be used
8320 instead. This never sets errno, just as '``llvm.fma.*``'.
8325 .. code-block:: llvm
8327 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8329 Half Precision Floating Point Intrinsics
8330 ----------------------------------------
8332 For most target platforms, half precision floating point is a
8333 storage-only format. This means that it is a dense encoding (in memory)
8334 but does not support computation in the format.
8336 This means that code must first load the half-precision floating point
8337 value as an i16, then convert it to float with
8338 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8339 then be performed on the float value (including extending to double
8340 etc). To store the value back to memory, it is first converted to float
8341 if needed, then converted to i16 with
8342 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8345 .. _int_convert_to_fp16:
8347 '``llvm.convert.to.fp16``' Intrinsic
8348 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8355 declare i16 @llvm.convert.to.fp16(f32 %a)
8360 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8361 from single precision floating point format to half precision floating
8367 The intrinsic function contains single argument - the value to be
8373 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8374 from single precision floating point format to half precision floating
8375 point format. The return value is an ``i16`` which contains the
8381 .. code-block:: llvm
8383 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8384 store i16 %res, i16* @x, align 2
8386 .. _int_convert_from_fp16:
8388 '``llvm.convert.from.fp16``' Intrinsic
8389 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8396 declare f32 @llvm.convert.from.fp16(i16 %a)
8401 The '``llvm.convert.from.fp16``' intrinsic function performs a
8402 conversion from half precision floating point format to single precision
8403 floating point format.
8408 The intrinsic function contains single argument - the value to be
8414 The '``llvm.convert.from.fp16``' intrinsic function performs a
8415 conversion from half single precision floating point format to single
8416 precision floating point format. The input half-float value is
8417 represented by an ``i16`` value.
8422 .. code-block:: llvm
8424 %a = load i16* @x, align 2
8425 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8430 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8431 prefix), are described in the `LLVM Source Level
8432 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8435 Exception Handling Intrinsics
8436 -----------------------------
8438 The LLVM exception handling intrinsics (which all start with
8439 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8440 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8444 Trampoline Intrinsics
8445 ---------------------
8447 These intrinsics make it possible to excise one parameter, marked with
8448 the :ref:`nest <nest>` attribute, from a function. The result is a
8449 callable function pointer lacking the nest parameter - the caller does
8450 not need to provide a value for it. Instead, the value to use is stored
8451 in advance in a "trampoline", a block of memory usually allocated on the
8452 stack, which also contains code to splice the nest value into the
8453 argument list. This is used to implement the GCC nested function address
8456 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8457 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8458 It can be created as follows:
8460 .. code-block:: llvm
8462 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8463 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8464 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8465 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8466 %fp = bitcast i8* %p to i32 (i32, i32)*
8468 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8469 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8473 '``llvm.init.trampoline``' Intrinsic
8474 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8481 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8486 This fills the memory pointed to by ``tramp`` with executable code,
8487 turning it into a trampoline.
8492 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8493 pointers. The ``tramp`` argument must point to a sufficiently large and
8494 sufficiently aligned block of memory; this memory is written to by the
8495 intrinsic. Note that the size and the alignment are target-specific -
8496 LLVM currently provides no portable way of determining them, so a
8497 front-end that generates this intrinsic needs to have some
8498 target-specific knowledge. The ``func`` argument must hold a function
8499 bitcast to an ``i8*``.
8504 The block of memory pointed to by ``tramp`` is filled with target
8505 dependent code, turning it into a function. Then ``tramp`` needs to be
8506 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8507 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8508 function's signature is the same as that of ``func`` with any arguments
8509 marked with the ``nest`` attribute removed. At most one such ``nest``
8510 argument is allowed, and it must be of pointer type. Calling the new
8511 function is equivalent to calling ``func`` with the same argument list,
8512 but with ``nval`` used for the missing ``nest`` argument. If, after
8513 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8514 modified, then the effect of any later call to the returned function
8515 pointer is undefined.
8519 '``llvm.adjust.trampoline``' Intrinsic
8520 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8527 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8532 This performs any required machine-specific adjustment to the address of
8533 a trampoline (passed as ``tramp``).
8538 ``tramp`` must point to a block of memory which already has trampoline
8539 code filled in by a previous call to
8540 :ref:`llvm.init.trampoline <int_it>`.
8545 On some architectures the address of the code to be executed needs to be
8546 different to the address where the trampoline is actually stored. This
8547 intrinsic returns the executable address corresponding to ``tramp``
8548 after performing the required machine specific adjustments. The pointer
8549 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8554 This class of intrinsics exists to information about the lifetime of
8555 memory objects and ranges where variables are immutable.
8559 '``llvm.lifetime.start``' Intrinsic
8560 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8567 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8572 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8578 The first argument is a constant integer representing the size of the
8579 object, or -1 if it is variable sized. The second argument is a pointer
8585 This intrinsic indicates that before this point in the code, the value
8586 of the memory pointed to by ``ptr`` is dead. This means that it is known
8587 to never be used and has an undefined value. A load from the pointer
8588 that precedes this intrinsic can be replaced with ``'undef'``.
8592 '``llvm.lifetime.end``' Intrinsic
8593 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8600 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8605 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8611 The first argument is a constant integer representing the size of the
8612 object, or -1 if it is variable sized. The second argument is a pointer
8618 This intrinsic indicates that after this point in the code, the value of
8619 the memory pointed to by ``ptr`` is dead. This means that it is known to
8620 never be used and has an undefined value. Any stores into the memory
8621 object following this intrinsic may be removed as dead.
8623 '``llvm.invariant.start``' Intrinsic
8624 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8631 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8636 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8637 a memory object will not change.
8642 The first argument is a constant integer representing the size of the
8643 object, or -1 if it is variable sized. The second argument is a pointer
8649 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8650 the return value, the referenced memory location is constant and
8653 '``llvm.invariant.end``' Intrinsic
8654 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8661 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8666 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8667 memory object are mutable.
8672 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8673 The second argument is a constant integer representing the size of the
8674 object, or -1 if it is variable sized and the third argument is a
8675 pointer to the object.
8680 This intrinsic indicates that the memory is mutable again.
8685 This class of intrinsics is designed to be generic and has no specific
8688 '``llvm.var.annotation``' Intrinsic
8689 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8696 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8701 The '``llvm.var.annotation``' intrinsic.
8706 The first argument is a pointer to a value, the second is a pointer to a
8707 global string, the third is a pointer to a global string which is the
8708 source file name, and the last argument is the line number.
8713 This intrinsic allows annotation of local variables with arbitrary
8714 strings. This can be useful for special purpose optimizations that want
8715 to look for these annotations. These have no other defined use; they are
8716 ignored by code generation and optimization.
8718 '``llvm.ptr.annotation.*``' Intrinsic
8719 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8724 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8725 pointer to an integer of any width. *NOTE* you must specify an address space for
8726 the pointer. The identifier for the default address space is the integer
8731 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8732 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8733 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8734 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8735 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8740 The '``llvm.ptr.annotation``' intrinsic.
8745 The first argument is a pointer to an integer value of arbitrary bitwidth
8746 (result of some expression), the second is a pointer to a global string, the
8747 third is a pointer to a global string which is the source file name, and the
8748 last argument is the line number. It returns the value of the first argument.
8753 This intrinsic allows annotation of a pointer to an integer with arbitrary
8754 strings. This can be useful for special purpose optimizations that want to look
8755 for these annotations. These have no other defined use; they are ignored by code
8756 generation and optimization.
8758 '``llvm.annotation.*``' Intrinsic
8759 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8764 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8765 any integer bit width.
8769 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8770 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8771 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8772 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8773 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8778 The '``llvm.annotation``' intrinsic.
8783 The first argument is an integer value (result of some expression), the
8784 second is a pointer to a global string, the third is a pointer to a
8785 global string which is the source file name, and the last argument is
8786 the line number. It returns the value of the first argument.
8791 This intrinsic allows annotations to be put on arbitrary expressions
8792 with arbitrary strings. This can be useful for special purpose
8793 optimizations that want to look for these annotations. These have no
8794 other defined use; they are ignored by code generation and optimization.
8796 '``llvm.trap``' Intrinsic
8797 ^^^^^^^^^^^^^^^^^^^^^^^^^
8804 declare void @llvm.trap() noreturn nounwind
8809 The '``llvm.trap``' intrinsic.
8819 This intrinsic is lowered to the target dependent trap instruction. If
8820 the target does not have a trap instruction, this intrinsic will be
8821 lowered to a call of the ``abort()`` function.
8823 '``llvm.debugtrap``' Intrinsic
8824 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8831 declare void @llvm.debugtrap() nounwind
8836 The '``llvm.debugtrap``' intrinsic.
8846 This intrinsic is lowered to code which is intended to cause an
8847 execution trap with the intention of requesting the attention of a
8850 '``llvm.stackprotector``' Intrinsic
8851 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8858 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8863 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8864 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8865 is placed on the stack before local variables.
8870 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8871 The first argument is the value loaded from the stack guard
8872 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8873 enough space to hold the value of the guard.
8878 This intrinsic causes the prologue/epilogue inserter to force the position of
8879 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8880 to ensure that if a local variable on the stack is overwritten, it will destroy
8881 the value of the guard. When the function exits, the guard on the stack is
8882 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8883 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8884 calling the ``__stack_chk_fail()`` function.
8886 '``llvm.stackprotectorcheck``' Intrinsic
8887 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8894 declare void @llvm.stackprotectorcheck(i8** <guard>)
8899 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8900 created stack protector and if they are not equal calls the
8901 ``__stack_chk_fail()`` function.
8906 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8907 the variable ``@__stack_chk_guard``.
8912 This intrinsic is provided to perform the stack protector check by comparing
8913 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8914 values do not match call the ``__stack_chk_fail()`` function.
8916 The reason to provide this as an IR level intrinsic instead of implementing it
8917 via other IR operations is that in order to perform this operation at the IR
8918 level without an intrinsic, one would need to create additional basic blocks to
8919 handle the success/failure cases. This makes it difficult to stop the stack
8920 protector check from disrupting sibling tail calls in Codegen. With this
8921 intrinsic, we are able to generate the stack protector basic blocks late in
8922 codegen after the tail call decision has occurred.
8924 '``llvm.objectsize``' Intrinsic
8925 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8932 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8933 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8938 The ``llvm.objectsize`` intrinsic is designed to provide information to
8939 the optimizers to determine at compile time whether a) an operation
8940 (like memcpy) will overflow a buffer that corresponds to an object, or
8941 b) that a runtime check for overflow isn't necessary. An object in this
8942 context means an allocation of a specific class, structure, array, or
8948 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8949 argument is a pointer to or into the ``object``. The second argument is
8950 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8951 or -1 (if false) when the object size is unknown. The second argument
8952 only accepts constants.
8957 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8958 the size of the object concerned. If the size cannot be determined at
8959 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8960 on the ``min`` argument).
8962 '``llvm.expect``' Intrinsic
8963 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8968 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
8973 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
8974 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8975 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8980 The ``llvm.expect`` intrinsic provides information about expected (the
8981 most probable) value of ``val``, which can be used by optimizers.
8986 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8987 a value. The second argument is an expected value, this needs to be a
8988 constant value, variables are not allowed.
8993 This intrinsic is lowered to the ``val``.
8995 '``llvm.donothing``' Intrinsic
8996 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9003 declare void @llvm.donothing() nounwind readnone
9008 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
9009 only intrinsic that can be called with an invoke instruction.
9019 This intrinsic does nothing, and it's removed by optimizers and ignored
9022 Stack Map Intrinsics
9023 --------------------
9025 LLVM provides experimental intrinsics to support runtime patching
9026 mechanisms commonly desired in dynamic language JITs. These intrinsics
9027 are described in :doc:`StackMaps`.