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 The ``inalloca`` argument attribute allows the caller to take the
773 address of outgoing stack arguments. An ``inalloca`` argument must
774 be a pointer to stack memory produced by an ``alloca`` instruction.
775 The alloca, or argument allocation, must also be tagged with the
776 inalloca keyword. Only the past argument may have the ``inalloca``
777 attribute, and that argument is guaranteed to be passed in memory.
779 An argument allocation may be used by a call at most once because
780 the call may deallocate it. The ``inalloca`` attribute cannot be
781 used in conjunction with other attributes that affect argument
782 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
783 ``inalloca`` attribute also disables LLVM's implicit lowering of
784 large aggregate return values, which means that frontend authors
785 must lower them with ``sret`` pointers.
787 When the call site is reached, the argument allocation must have
788 been the most recent stack allocation that is still live, or the
789 results are undefined. It is possible to allocate additional stack
790 space after an argument allocation and before its call site, but it
791 must be cleared off with :ref:`llvm.stackrestore
794 See :doc:`InAlloca` for more information on how to use this
798 This indicates that the pointer parameter specifies the address of a
799 structure that is the return value of the function in the source
800 program. This pointer must be guaranteed by the caller to be valid:
801 loads and stores to the structure may be assumed by the callee
802 not to trap and to be properly aligned. This may only be applied to
803 the first parameter. This is not a valid attribute for return
806 This indicates that pointer values :ref:`based <pointeraliasing>` on
807 the argument or return value do not alias pointer values which are
808 not *based* on it, ignoring certain "irrelevant" dependencies. For a
809 call to the parent function, dependencies between memory references
810 from before or after the call and from those during the call are
811 "irrelevant" to the ``noalias`` keyword for the arguments and return
812 value used in that call. The caller shares the responsibility with
813 the callee for ensuring that these requirements are met. For further
814 details, please see the discussion of the NoAlias response in `alias
815 analysis <AliasAnalysis.html#MustMayNo>`_.
817 Note that this definition of ``noalias`` is intentionally similar
818 to the definition of ``restrict`` in C99 for function arguments,
819 though it is slightly weaker.
821 For function return values, C99's ``restrict`` is not meaningful,
822 while LLVM's ``noalias`` is.
824 This indicates that the callee does not make any copies of the
825 pointer that outlive the callee itself. This is not a valid
826 attribute for return values.
831 This indicates that the pointer parameter can be excised using the
832 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
833 attribute for return values and can only be applied to one parameter.
836 This indicates that the function always returns the argument as its return
837 value. This is an optimization hint to the code generator when generating
838 the caller, allowing tail call optimization and omission of register saves
839 and restores in some cases; it is not checked or enforced when generating
840 the callee. The parameter and the function return type must be valid
841 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
842 valid attribute for return values and can only be applied to one parameter.
846 Garbage Collector Names
847 -----------------------
849 Each function may specify a garbage collector name, which is simply a
854 define void @f() gc "name" { ... }
856 The compiler declares the supported values of *name*. Specifying a
857 collector which will cause the compiler to alter its output in order to
858 support the named garbage collection algorithm.
865 Prefix data is data associated with a function which the code generator
866 will emit immediately before the function body. The purpose of this feature
867 is to allow frontends to associate language-specific runtime metadata with
868 specific functions and make it available through the function pointer while
869 still allowing the function pointer to be called. To access the data for a
870 given function, a program may bitcast the function pointer to a pointer to
871 the constant's type. This implies that the IR symbol points to the start
874 To maintain the semantics of ordinary function calls, the prefix data must
875 have a particular format. Specifically, it must begin with a sequence of
876 bytes which decode to a sequence of machine instructions, valid for the
877 module's target, which transfer control to the point immediately succeeding
878 the prefix data, without performing any other visible action. This allows
879 the inliner and other passes to reason about the semantics of the function
880 definition without needing to reason about the prefix data. Obviously this
881 makes the format of the prefix data highly target dependent.
883 Prefix data is laid out as if it were an initializer for a global variable
884 of the prefix data's type. No padding is automatically placed between the
885 prefix data and the function body. If padding is required, it must be part
888 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
889 which encodes the ``nop`` instruction:
893 define void @f() prefix i8 144 { ... }
895 Generally prefix data can be formed by encoding a relative branch instruction
896 which skips the metadata, as in this example of valid prefix data for the
897 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
901 %0 = type <{ i8, i8, i8* }>
903 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
905 A function may have prefix data but no body. This has similar semantics
906 to the ``available_externally`` linkage in that the data may be used by the
907 optimizers but will not be emitted in the object file.
914 Attribute groups are groups of attributes that are referenced by objects within
915 the IR. They are important for keeping ``.ll`` files readable, because a lot of
916 functions will use the same set of attributes. In the degenerative case of a
917 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
918 group will capture the important command line flags used to build that file.
920 An attribute group is a module-level object. To use an attribute group, an
921 object references the attribute group's ID (e.g. ``#37``). An object may refer
922 to more than one attribute group. In that situation, the attributes from the
923 different groups are merged.
925 Here is an example of attribute groups for a function that should always be
926 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
930 ; Target-independent attributes:
931 attributes #0 = { alwaysinline alignstack=4 }
933 ; Target-dependent attributes:
934 attributes #1 = { "no-sse" }
936 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
937 define void @f() #0 #1 { ... }
944 Function attributes are set to communicate additional information about
945 a function. Function attributes are considered to be part of the
946 function, not of the function type, so functions with different function
947 attributes can have the same function type.
949 Function attributes are simple keywords that follow the type specified.
950 If multiple attributes are needed, they are space separated. For
955 define void @f() noinline { ... }
956 define void @f() alwaysinline { ... }
957 define void @f() alwaysinline optsize { ... }
958 define void @f() optsize { ... }
961 This attribute indicates that, when emitting the prologue and
962 epilogue, the backend should forcibly align the stack pointer.
963 Specify the desired alignment, which must be a power of two, in
966 This attribute indicates that the inliner should attempt to inline
967 this function into callers whenever possible, ignoring any active
968 inlining size threshold for this caller.
970 This indicates that the callee function at a call site should be
971 recognized as a built-in function, even though the function's declaration
972 uses the ``nobuiltin`` attribute. This is only valid at call sites for
973 direct calls to functions which are declared with the ``nobuiltin``
976 This attribute indicates that this function is rarely called. When
977 computing edge weights, basic blocks post-dominated by a cold
978 function call are also considered to be cold; and, thus, given low
981 This attribute indicates that the source code contained a hint that
982 inlining this function is desirable (such as the "inline" keyword in
983 C/C++). It is just a hint; it imposes no requirements on the
986 This attribute suggests that optimization passes and code generator
987 passes make choices that keep the code size of this function as small
988 as possible and perform optimizations that may sacrifice runtime
989 performance in order to minimize the size of the generated code.
991 This attribute disables prologue / epilogue emission for the
992 function. This can have very system-specific consequences.
994 This indicates that the callee function at a call site is not recognized as
995 a built-in function. LLVM will retain the original call and not replace it
996 with equivalent code based on the semantics of the built-in function, unless
997 the call site uses the ``builtin`` attribute. This is valid at call sites
998 and on function declarations and definitions.
1000 This attribute indicates that calls to the function cannot be
1001 duplicated. A call to a ``noduplicate`` function may be moved
1002 within its parent function, but may not be duplicated within
1003 its parent function.
1005 A function containing a ``noduplicate`` call may still
1006 be an inlining candidate, provided that the call is not
1007 duplicated by inlining. That implies that the function has
1008 internal linkage and only has one call site, so the original
1009 call is dead after inlining.
1011 This attributes disables implicit floating point instructions.
1013 This attribute indicates that the inliner should never inline this
1014 function in any situation. This attribute may not be used together
1015 with the ``alwaysinline`` attribute.
1017 This attribute suppresses lazy symbol binding for the function. This
1018 may make calls to the function faster, at the cost of extra program
1019 startup time if the function is not called during program startup.
1021 This attribute indicates that the code generator should not use a
1022 red zone, even if the target-specific ABI normally permits it.
1024 This function attribute indicates that the function never returns
1025 normally. This produces undefined behavior at runtime if the
1026 function ever does dynamically return.
1028 This function attribute indicates that the function never returns
1029 with an unwind or exceptional control flow. If the function does
1030 unwind, its runtime behavior is undefined.
1032 This function attribute indicates that the function is not optimized
1033 by any optimization or code generator passes with the
1034 exception of interprocedural optimization passes.
1035 This attribute cannot be used together with the ``alwaysinline``
1036 attribute; this attribute is also incompatible
1037 with the ``minsize`` attribute and the ``optsize`` attribute.
1039 This attribute requires the ``noinline`` attribute to be specified on
1040 the function as well, so the function is never inlined into any caller.
1041 Only functions with the ``alwaysinline`` attribute are valid
1042 candidates for inlining into the body of this function.
1044 This attribute suggests that optimization passes and code generator
1045 passes make choices that keep the code size of this function low,
1046 and otherwise do optimizations specifically to reduce code size as
1047 long as they do not significantly impact runtime performance.
1049 On a function, this attribute indicates that the function computes its
1050 result (or decides to unwind an exception) based strictly on its arguments,
1051 without dereferencing any pointer arguments or otherwise accessing
1052 any mutable state (e.g. memory, control registers, etc) visible to
1053 caller functions. It does not write through any pointer arguments
1054 (including ``byval`` arguments) and never changes any state visible
1055 to callers. This means that it cannot unwind exceptions by calling
1056 the ``C++`` exception throwing methods.
1058 On an argument, this attribute indicates that the function does not
1059 dereference that pointer argument, even though it may read or write the
1060 memory that the pointer points to if accessed through other pointers.
1062 On a function, this attribute indicates that the function does not write
1063 through any pointer arguments (including ``byval`` arguments) or otherwise
1064 modify any state (e.g. memory, control registers, etc) visible to
1065 caller functions. It may dereference pointer arguments and read
1066 state that may be set in the caller. A readonly function always
1067 returns the same value (or unwinds an exception identically) when
1068 called with the same set of arguments and global state. It cannot
1069 unwind an exception by calling the ``C++`` exception throwing
1072 On an argument, this attribute indicates that the function does not write
1073 through this pointer argument, even though it may write to the memory that
1074 the pointer points to.
1076 This attribute indicates that this function can return twice. The C
1077 ``setjmp`` is an example of such a function. The compiler disables
1078 some optimizations (like tail calls) in the caller of these
1080 ``sanitize_address``
1081 This attribute indicates that AddressSanitizer checks
1082 (dynamic address safety analysis) are enabled for this function.
1084 This attribute indicates that MemorySanitizer checks (dynamic detection
1085 of accesses to uninitialized memory) are enabled for this function.
1087 This attribute indicates that ThreadSanitizer checks
1088 (dynamic thread safety analysis) are enabled for this function.
1090 This attribute indicates that the function should emit a stack
1091 smashing protector. It is in the form of a "canary" --- a random value
1092 placed on the stack before the local variables that's checked upon
1093 return from the function to see if it has been overwritten. A
1094 heuristic is used to determine if a function needs stack protectors
1095 or not. The heuristic used will enable protectors for functions with:
1097 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1098 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1099 - Calls to alloca() with variable sizes or constant sizes greater than
1100 ``ssp-buffer-size``.
1102 Variables that are identified as requiring a protector will be arranged
1103 on the stack such that they are adjacent to the stack protector guard.
1105 If a function that has an ``ssp`` attribute is inlined into a
1106 function that doesn't have an ``ssp`` attribute, then the resulting
1107 function will have an ``ssp`` attribute.
1109 This attribute indicates that the function should *always* emit a
1110 stack smashing protector. This overrides the ``ssp`` function
1113 Variables that are identified as requiring a protector will be arranged
1114 on the stack such that they are adjacent to the stack protector guard.
1115 The specific layout rules are:
1117 #. Large arrays and structures containing large arrays
1118 (``>= ssp-buffer-size``) are closest to the stack protector.
1119 #. Small arrays and structures containing small arrays
1120 (``< ssp-buffer-size``) are 2nd closest to the protector.
1121 #. Variables that have had their address taken are 3rd closest to the
1124 If a function that has an ``sspreq`` attribute is inlined into a
1125 function that doesn't have an ``sspreq`` attribute or which has an
1126 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1127 an ``sspreq`` attribute.
1129 This attribute indicates that the function should emit a stack smashing
1130 protector. This attribute causes a strong heuristic to be used when
1131 determining if a function needs stack protectors. The strong heuristic
1132 will enable protectors for functions with:
1134 - Arrays of any size and type
1135 - Aggregates containing an array of any size and type.
1136 - Calls to alloca().
1137 - Local variables that have had their address taken.
1139 Variables that are identified as requiring a protector will be arranged
1140 on the stack such that they are adjacent to the stack protector guard.
1141 The specific layout rules are:
1143 #. Large arrays and structures containing large arrays
1144 (``>= ssp-buffer-size``) are closest to the stack protector.
1145 #. Small arrays and structures containing small arrays
1146 (``< ssp-buffer-size``) are 2nd closest to the protector.
1147 #. Variables that have had their address taken are 3rd closest to the
1150 This overrides the ``ssp`` function attribute.
1152 If a function that has an ``sspstrong`` attribute is inlined into a
1153 function that doesn't have an ``sspstrong`` attribute, then the
1154 resulting function will have an ``sspstrong`` attribute.
1156 This attribute indicates that the ABI being targeted requires that
1157 an unwind table entry be produce for this function even if we can
1158 show that no exceptions passes by it. This is normally the case for
1159 the ELF x86-64 abi, but it can be disabled for some compilation
1164 Module-Level Inline Assembly
1165 ----------------------------
1167 Modules may contain "module-level inline asm" blocks, which corresponds
1168 to the GCC "file scope inline asm" blocks. These blocks are internally
1169 concatenated by LLVM and treated as a single unit, but may be separated
1170 in the ``.ll`` file if desired. The syntax is very simple:
1172 .. code-block:: llvm
1174 module asm "inline asm code goes here"
1175 module asm "more can go here"
1177 The strings can contain any character by escaping non-printable
1178 characters. The escape sequence used is simply "\\xx" where "xx" is the
1179 two digit hex code for the number.
1181 The inline asm code is simply printed to the machine code .s file when
1182 assembly code is generated.
1184 .. _langref_datalayout:
1189 A module may specify a target specific data layout string that specifies
1190 how data is to be laid out in memory. The syntax for the data layout is
1193 .. code-block:: llvm
1195 target datalayout = "layout specification"
1197 The *layout specification* consists of a list of specifications
1198 separated by the minus sign character ('-'). Each specification starts
1199 with a letter and may include other information after the letter to
1200 define some aspect of the data layout. The specifications accepted are
1204 Specifies that the target lays out data in big-endian form. That is,
1205 the bits with the most significance have the lowest address
1208 Specifies that the target lays out data in little-endian form. That
1209 is, the bits with the least significance have the lowest address
1212 Specifies the natural alignment of the stack in bits. Alignment
1213 promotion of stack variables is limited to the natural stack
1214 alignment to avoid dynamic stack realignment. The stack alignment
1215 must be a multiple of 8-bits. If omitted, the natural stack
1216 alignment defaults to "unspecified", which does not prevent any
1217 alignment promotions.
1218 ``p[n]:<size>:<abi>:<pref>``
1219 This specifies the *size* of a pointer and its ``<abi>`` and
1220 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1221 bits. The address space, ``n`` is optional, and if not specified,
1222 denotes the default address space 0. The value of ``n`` must be
1223 in the range [1,2^23).
1224 ``i<size>:<abi>:<pref>``
1225 This specifies the alignment for an integer type of a given bit
1226 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1227 ``v<size>:<abi>:<pref>``
1228 This specifies the alignment for a vector type of a given bit
1230 ``f<size>:<abi>:<pref>``
1231 This specifies the alignment for a floating point type of a given bit
1232 ``<size>``. Only values of ``<size>`` that are supported by the target
1233 will work. 32 (float) and 64 (double) are supported on all targets; 80
1234 or 128 (different flavors of long double) are also supported on some
1237 This specifies the alignment for an object of aggregate type.
1239 If present, specifies that llvm names are mangled in the output. The
1242 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1243 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1244 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1245 symbols get a ``_`` prefix.
1246 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1247 functions also get a suffix based on the frame size.
1248 ``n<size1>:<size2>:<size3>...``
1249 This specifies a set of native integer widths for the target CPU in
1250 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1251 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1252 this set are considered to support most general arithmetic operations
1255 On every specification that takes a ``<abi>:<pref>``, specifying the
1256 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1257 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1259 When constructing the data layout for a given target, LLVM starts with a
1260 default set of specifications which are then (possibly) overridden by
1261 the specifications in the ``datalayout`` keyword. The default
1262 specifications are given in this list:
1264 - ``E`` - big endian
1265 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1266 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1267 same as the default address space.
1268 - ``S0`` - natural stack alignment is unspecified
1269 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1270 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1271 - ``i16:16:16`` - i16 is 16-bit aligned
1272 - ``i32:32:32`` - i32 is 32-bit aligned
1273 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1274 alignment of 64-bits
1275 - ``f16:16:16`` - half is 16-bit aligned
1276 - ``f32:32:32`` - float is 32-bit aligned
1277 - ``f64:64:64`` - double is 64-bit aligned
1278 - ``f128:128:128`` - quad is 128-bit aligned
1279 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1280 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1281 - ``a:0:64`` - aggregates are 64-bit aligned
1283 When LLVM is determining the alignment for a given type, it uses the
1286 #. If the type sought is an exact match for one of the specifications,
1287 that specification is used.
1288 #. If no match is found, and the type sought is an integer type, then
1289 the smallest integer type that is larger than the bitwidth of the
1290 sought type is used. If none of the specifications are larger than
1291 the bitwidth then the largest integer type is used. For example,
1292 given the default specifications above, the i7 type will use the
1293 alignment of i8 (next largest) while both i65 and i256 will use the
1294 alignment of i64 (largest specified).
1295 #. If no match is found, and the type sought is a vector type, then the
1296 largest vector type that is smaller than the sought vector type will
1297 be used as a fall back. This happens because <128 x double> can be
1298 implemented in terms of 64 <2 x double>, for example.
1300 The function of the data layout string may not be what you expect.
1301 Notably, this is not a specification from the frontend of what alignment
1302 the code generator should use.
1304 Instead, if specified, the target data layout is required to match what
1305 the ultimate *code generator* expects. This string is used by the
1306 mid-level optimizers to improve code, and this only works if it matches
1307 what the ultimate code generator uses. If you would like to generate IR
1308 that does not embed this target-specific detail into the IR, then you
1309 don't have to specify the string. This will disable some optimizations
1310 that require precise layout information, but this also prevents those
1311 optimizations from introducing target specificity into the IR.
1318 A module may specify a target triple string that describes the target
1319 host. The syntax for the target triple is simply:
1321 .. code-block:: llvm
1323 target triple = "x86_64-apple-macosx10.7.0"
1325 The *target triple* string consists of a series of identifiers delimited
1326 by the minus sign character ('-'). The canonical forms are:
1330 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1331 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1333 This information is passed along to the backend so that it generates
1334 code for the proper architecture. It's possible to override this on the
1335 command line with the ``-mtriple`` command line option.
1337 .. _pointeraliasing:
1339 Pointer Aliasing Rules
1340 ----------------------
1342 Any memory access must be done through a pointer value associated with
1343 an address range of the memory access, otherwise the behavior is
1344 undefined. Pointer values are associated with address ranges according
1345 to the following rules:
1347 - A pointer value is associated with the addresses associated with any
1348 value it is *based* on.
1349 - An address of a global variable is associated with the address range
1350 of the variable's storage.
1351 - The result value of an allocation instruction is associated with the
1352 address range of the allocated storage.
1353 - A null pointer in the default address-space is associated with no
1355 - An integer constant other than zero or a pointer value returned from
1356 a function not defined within LLVM may be associated with address
1357 ranges allocated through mechanisms other than those provided by
1358 LLVM. Such ranges shall not overlap with any ranges of addresses
1359 allocated by mechanisms provided by LLVM.
1361 A pointer value is *based* on another pointer value according to the
1364 - A pointer value formed from a ``getelementptr`` operation is *based*
1365 on the first operand of the ``getelementptr``.
1366 - The result value of a ``bitcast`` is *based* on the operand of the
1368 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1369 values that contribute (directly or indirectly) to the computation of
1370 the pointer's value.
1371 - The "*based* on" relationship is transitive.
1373 Note that this definition of *"based"* is intentionally similar to the
1374 definition of *"based"* in C99, though it is slightly weaker.
1376 LLVM IR does not associate types with memory. The result type of a
1377 ``load`` merely indicates the size and alignment of the memory from
1378 which to load, as well as the interpretation of the value. The first
1379 operand type of a ``store`` similarly only indicates the size and
1380 alignment of the store.
1382 Consequently, type-based alias analysis, aka TBAA, aka
1383 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1384 :ref:`Metadata <metadata>` may be used to encode additional information
1385 which specialized optimization passes may use to implement type-based
1390 Volatile Memory Accesses
1391 ------------------------
1393 Certain memory accesses, such as :ref:`load <i_load>`'s,
1394 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1395 marked ``volatile``. The optimizers must not change the number of
1396 volatile operations or change their order of execution relative to other
1397 volatile operations. The optimizers *may* change the order of volatile
1398 operations relative to non-volatile operations. This is not Java's
1399 "volatile" and has no cross-thread synchronization behavior.
1401 IR-level volatile loads and stores cannot safely be optimized into
1402 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1403 flagged volatile. Likewise, the backend should never split or merge
1404 target-legal volatile load/store instructions.
1406 .. admonition:: Rationale
1408 Platforms may rely on volatile loads and stores of natively supported
1409 data width to be executed as single instruction. For example, in C
1410 this holds for an l-value of volatile primitive type with native
1411 hardware support, but not necessarily for aggregate types. The
1412 frontend upholds these expectations, which are intentionally
1413 unspecified in the IR. The rules above ensure that IR transformation
1414 do not violate the frontend's contract with the language.
1418 Memory Model for Concurrent Operations
1419 --------------------------------------
1421 The LLVM IR does not define any way to start parallel threads of
1422 execution or to register signal handlers. Nonetheless, there are
1423 platform-specific ways to create them, and we define LLVM IR's behavior
1424 in their presence. This model is inspired by the C++0x memory model.
1426 For a more informal introduction to this model, see the :doc:`Atomics`.
1428 We define a *happens-before* partial order as the least partial order
1431 - Is a superset of single-thread program order, and
1432 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1433 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1434 techniques, like pthread locks, thread creation, thread joining,
1435 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1436 Constraints <ordering>`).
1438 Note that program order does not introduce *happens-before* edges
1439 between a thread and signals executing inside that thread.
1441 Every (defined) read operation (load instructions, memcpy, atomic
1442 loads/read-modify-writes, etc.) R reads a series of bytes written by
1443 (defined) write operations (store instructions, atomic
1444 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1445 section, initialized globals are considered to have a write of the
1446 initializer which is atomic and happens before any other read or write
1447 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1448 may see any write to the same byte, except:
1450 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1451 write\ :sub:`2` happens before R\ :sub:`byte`, then
1452 R\ :sub:`byte` does not see write\ :sub:`1`.
1453 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1454 R\ :sub:`byte` does not see write\ :sub:`3`.
1456 Given that definition, R\ :sub:`byte` is defined as follows:
1458 - If R is volatile, the result is target-dependent. (Volatile is
1459 supposed to give guarantees which can support ``sig_atomic_t`` in
1460 C/C++, and may be used for accesses to addresses which do not behave
1461 like normal memory. It does not generally provide cross-thread
1463 - Otherwise, if there is no write to the same byte that happens before
1464 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1465 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1466 R\ :sub:`byte` returns the value written by that write.
1467 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1468 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1469 Memory Ordering Constraints <ordering>` section for additional
1470 constraints on how the choice is made.
1471 - Otherwise R\ :sub:`byte` returns ``undef``.
1473 R returns the value composed of the series of bytes it read. This
1474 implies that some bytes within the value may be ``undef`` **without**
1475 the entire value being ``undef``. Note that this only defines the
1476 semantics of the operation; it doesn't mean that targets will emit more
1477 than one instruction to read the series of bytes.
1479 Note that in cases where none of the atomic intrinsics are used, this
1480 model places only one restriction on IR transformations on top of what
1481 is required for single-threaded execution: introducing a store to a byte
1482 which might not otherwise be stored is not allowed in general.
1483 (Specifically, in the case where another thread might write to and read
1484 from an address, introducing a store can change a load that may see
1485 exactly one write into a load that may see multiple writes.)
1489 Atomic Memory Ordering Constraints
1490 ----------------------------------
1492 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1493 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1494 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1495 ordering parameters that determine which other atomic instructions on
1496 the same address they *synchronize with*. These semantics are borrowed
1497 from Java and C++0x, but are somewhat more colloquial. If these
1498 descriptions aren't precise enough, check those specs (see spec
1499 references in the :doc:`atomics guide <Atomics>`).
1500 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1501 differently since they don't take an address. See that instruction's
1502 documentation for details.
1504 For a simpler introduction to the ordering constraints, see the
1508 The set of values that can be read is governed by the happens-before
1509 partial order. A value cannot be read unless some operation wrote
1510 it. This is intended to provide a guarantee strong enough to model
1511 Java's non-volatile shared variables. This ordering cannot be
1512 specified for read-modify-write operations; it is not strong enough
1513 to make them atomic in any interesting way.
1515 In addition to the guarantees of ``unordered``, there is a single
1516 total order for modifications by ``monotonic`` operations on each
1517 address. All modification orders must be compatible with the
1518 happens-before order. There is no guarantee that the modification
1519 orders can be combined to a global total order for the whole program
1520 (and this often will not be possible). The read in an atomic
1521 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1522 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1523 order immediately before the value it writes. If one atomic read
1524 happens before another atomic read of the same address, the later
1525 read must see the same value or a later value in the address's
1526 modification order. This disallows reordering of ``monotonic`` (or
1527 stronger) operations on the same address. If an address is written
1528 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1529 read that address repeatedly, the other threads must eventually see
1530 the write. This corresponds to the C++0x/C1x
1531 ``memory_order_relaxed``.
1533 In addition to the guarantees of ``monotonic``, a
1534 *synchronizes-with* edge may be formed with a ``release`` operation.
1535 This is intended to model C++'s ``memory_order_acquire``.
1537 In addition to the guarantees of ``monotonic``, if this operation
1538 writes a value which is subsequently read by an ``acquire``
1539 operation, it *synchronizes-with* that operation. (This isn't a
1540 complete description; see the C++0x definition of a release
1541 sequence.) This corresponds to the C++0x/C1x
1542 ``memory_order_release``.
1543 ``acq_rel`` (acquire+release)
1544 Acts as both an ``acquire`` and ``release`` operation on its
1545 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1546 ``seq_cst`` (sequentially consistent)
1547 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1548 operation which only reads, ``release`` for an operation which only
1549 writes), there is a global total order on all
1550 sequentially-consistent operations on all addresses, which is
1551 consistent with the *happens-before* partial order and with the
1552 modification orders of all the affected addresses. Each
1553 sequentially-consistent read sees the last preceding write to the
1554 same address in this global order. This corresponds to the C++0x/C1x
1555 ``memory_order_seq_cst`` and Java volatile.
1559 If an atomic operation is marked ``singlethread``, it only *synchronizes
1560 with* or participates in modification and seq\_cst total orderings with
1561 other operations running in the same thread (for example, in signal
1569 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1570 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1571 :ref:`frem <i_frem>`) have the following flags that can set to enable
1572 otherwise unsafe floating point operations
1575 No NaNs - Allow optimizations to assume the arguments and result are not
1576 NaN. Such optimizations are required to retain defined behavior over
1577 NaNs, but the value of the result is undefined.
1580 No Infs - Allow optimizations to assume the arguments and result are not
1581 +/-Inf. Such optimizations are required to retain defined behavior over
1582 +/-Inf, but the value of the result is undefined.
1585 No Signed Zeros - Allow optimizations to treat the sign of a zero
1586 argument or result as insignificant.
1589 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1590 argument rather than perform division.
1593 Fast - Allow algebraically equivalent transformations that may
1594 dramatically change results in floating point (e.g. reassociate). This
1595 flag implies all the others.
1602 The LLVM type system is one of the most important features of the
1603 intermediate representation. Being typed enables a number of
1604 optimizations to be performed on the intermediate representation
1605 directly, without having to do extra analyses on the side before the
1606 transformation. A strong type system makes it easier to read the
1607 generated code and enables novel analyses and transformations that are
1608 not feasible to perform on normal three address code representations.
1618 The void type does not represent any value and has no size.
1636 The function type can be thought of as a function signature. It consists of a
1637 return type and a list of formal parameter types. The return type of a function
1638 type is a void type or first class type --- except for :ref:`label <t_label>`
1639 and :ref:`metadata <t_metadata>` types.
1645 <returntype> (<parameter list>)
1647 ...where '``<parameter list>``' is a comma-separated list of type
1648 specifiers. Optionally, the parameter list may include a type ``...``, which
1649 indicates that the function takes a variable number of arguments. Variable
1650 argument functions can access their arguments with the :ref:`variable argument
1651 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1652 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1656 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1657 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1658 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1659 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1660 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1661 | ``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. |
1662 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1663 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1664 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1671 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1672 Values of these types are the only ones which can be produced by
1680 These are the types that are valid in registers from CodeGen's perspective.
1689 The integer type is a very simple type that simply specifies an
1690 arbitrary bit width for the integer type desired. Any bit width from 1
1691 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1699 The number of bits the integer will occupy is specified by the ``N``
1705 +----------------+------------------------------------------------+
1706 | ``i1`` | a single-bit integer. |
1707 +----------------+------------------------------------------------+
1708 | ``i32`` | a 32-bit integer. |
1709 +----------------+------------------------------------------------+
1710 | ``i1942652`` | a really big integer of over 1 million bits. |
1711 +----------------+------------------------------------------------+
1715 Floating Point Types
1716 """"""""""""""""""""
1725 - 16-bit floating point value
1728 - 32-bit floating point value
1731 - 64-bit floating point value
1734 - 128-bit floating point value (112-bit mantissa)
1737 - 80-bit floating point value (X87)
1740 - 128-bit floating point value (two 64-bits)
1747 The x86_mmx type represents a value held in an MMX register on an x86
1748 machine. The operations allowed on it are quite limited: parameters and
1749 return values, load and store, and bitcast. User-specified MMX
1750 instructions are represented as intrinsic or asm calls with arguments
1751 and/or results of this type. There are no arrays, vectors or constants
1768 The pointer type is used to specify memory locations. Pointers are
1769 commonly used to reference objects in memory.
1771 Pointer types may have an optional address space attribute defining the
1772 numbered address space where the pointed-to object resides. The default
1773 address space is number zero. The semantics of non-zero address spaces
1774 are target-specific.
1776 Note that LLVM does not permit pointers to void (``void*``) nor does it
1777 permit pointers to labels (``label*``). Use ``i8*`` instead.
1787 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1788 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1789 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1790 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1791 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1792 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1793 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1802 A vector type is a simple derived type that represents a vector of
1803 elements. Vector types are used when multiple primitive data are
1804 operated in parallel using a single instruction (SIMD). A vector type
1805 requires a size (number of elements) and an underlying primitive data
1806 type. Vector types are considered :ref:`first class <t_firstclass>`.
1812 < <# elements> x <elementtype> >
1814 The number of elements is a constant integer value larger than 0;
1815 elementtype may be any integer or floating point type, or a pointer to
1816 these types. Vectors of size zero are not allowed.
1820 +-------------------+--------------------------------------------------+
1821 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1822 +-------------------+--------------------------------------------------+
1823 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1824 +-------------------+--------------------------------------------------+
1825 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1826 +-------------------+--------------------------------------------------+
1827 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1828 +-------------------+--------------------------------------------------+
1837 The label type represents code labels.
1852 The metadata type represents embedded metadata. No derived types may be
1853 created from metadata except for :ref:`function <t_function>` arguments.
1866 Aggregate Types are a subset of derived types that can contain multiple
1867 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1868 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1878 The array type is a very simple derived type that arranges elements
1879 sequentially in memory. The array type requires a size (number of
1880 elements) and an underlying data type.
1886 [<# elements> x <elementtype>]
1888 The number of elements is a constant integer value; ``elementtype`` may
1889 be any type with a size.
1893 +------------------+--------------------------------------+
1894 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1895 +------------------+--------------------------------------+
1896 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1897 +------------------+--------------------------------------+
1898 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1899 +------------------+--------------------------------------+
1901 Here are some examples of multidimensional arrays:
1903 +-----------------------------+----------------------------------------------------------+
1904 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1905 +-----------------------------+----------------------------------------------------------+
1906 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1907 +-----------------------------+----------------------------------------------------------+
1908 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1909 +-----------------------------+----------------------------------------------------------+
1911 There is no restriction on indexing beyond the end of the array implied
1912 by a static type (though there are restrictions on indexing beyond the
1913 bounds of an allocated object in some cases). This means that
1914 single-dimension 'variable sized array' addressing can be implemented in
1915 LLVM with a zero length array type. An implementation of 'pascal style
1916 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1926 The structure type is used to represent a collection of data members
1927 together in memory. The elements of a structure may be any type that has
1930 Structures in memory are accessed using '``load``' and '``store``' by
1931 getting a pointer to a field with the '``getelementptr``' instruction.
1932 Structures in registers are accessed using the '``extractvalue``' and
1933 '``insertvalue``' instructions.
1935 Structures may optionally be "packed" structures, which indicate that
1936 the alignment of the struct is one byte, and that there is no padding
1937 between the elements. In non-packed structs, padding between field types
1938 is inserted as defined by the DataLayout string in the module, which is
1939 required to match what the underlying code generator expects.
1941 Structures can either be "literal" or "identified". A literal structure
1942 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1943 identified types are always defined at the top level with a name.
1944 Literal types are uniqued by their contents and can never be recursive
1945 or opaque since there is no way to write one. Identified types can be
1946 recursive, can be opaqued, and are never uniqued.
1952 %T1 = type { <type list> } ; Identified normal struct type
1953 %T2 = type <{ <type list> }> ; Identified packed struct type
1957 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1958 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1959 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1960 | ``{ 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``. |
1961 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1962 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1963 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1967 Opaque Structure Types
1968 """"""""""""""""""""""
1972 Opaque structure types are used to represent named structure types that
1973 do not have a body specified. This corresponds (for example) to the C
1974 notion of a forward declared structure.
1985 +--------------+-------------------+
1986 | ``opaque`` | An opaque type. |
1987 +--------------+-------------------+
1992 LLVM has several different basic types of constants. This section
1993 describes them all and their syntax.
1998 **Boolean constants**
1999 The two strings '``true``' and '``false``' are both valid constants
2001 **Integer constants**
2002 Standard integers (such as '4') are constants of the
2003 :ref:`integer <t_integer>` type. Negative numbers may be used with
2005 **Floating point constants**
2006 Floating point constants use standard decimal notation (e.g.
2007 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2008 hexadecimal notation (see below). The assembler requires the exact
2009 decimal value of a floating-point constant. For example, the
2010 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2011 decimal in binary. Floating point constants must have a :ref:`floating
2012 point <t_floating>` type.
2013 **Null pointer constants**
2014 The identifier '``null``' is recognized as a null pointer constant
2015 and must be of :ref:`pointer type <t_pointer>`.
2017 The one non-intuitive notation for constants is the hexadecimal form of
2018 floating point constants. For example, the form
2019 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2020 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2021 constants are required (and the only time that they are generated by the
2022 disassembler) is when a floating point constant must be emitted but it
2023 cannot be represented as a decimal floating point number in a reasonable
2024 number of digits. For example, NaN's, infinities, and other special
2025 values are represented in their IEEE hexadecimal format so that assembly
2026 and disassembly do not cause any bits to change in the constants.
2028 When using the hexadecimal form, constants of types half, float, and
2029 double are represented using the 16-digit form shown above (which
2030 matches the IEEE754 representation for double); half and float values
2031 must, however, be exactly representable as IEEE 754 half and single
2032 precision, respectively. Hexadecimal format is always used for long
2033 double, and there are three forms of long double. The 80-bit format used
2034 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2035 128-bit format used by PowerPC (two adjacent doubles) is represented by
2036 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2037 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2038 will only work if they match the long double format on your target.
2039 The IEEE 16-bit format (half precision) is represented by ``0xH``
2040 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2041 (sign bit at the left).
2043 There are no constants of type x86_mmx.
2045 .. _complexconstants:
2050 Complex constants are a (potentially recursive) combination of simple
2051 constants and smaller complex constants.
2053 **Structure constants**
2054 Structure constants are represented with notation similar to
2055 structure type definitions (a comma separated list of elements,
2056 surrounded by braces (``{}``)). For example:
2057 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2058 "``@G = external global i32``". Structure constants must have
2059 :ref:`structure type <t_struct>`, and the number and types of elements
2060 must match those specified by the type.
2062 Array constants are represented with notation similar to array type
2063 definitions (a comma separated list of elements, surrounded by
2064 square brackets (``[]``)). For example:
2065 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2066 :ref:`array type <t_array>`, and the number and types of elements must
2067 match those specified by the type.
2068 **Vector constants**
2069 Vector constants are represented with notation similar to vector
2070 type definitions (a comma separated list of elements, surrounded by
2071 less-than/greater-than's (``<>``)). For example:
2072 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2073 must have :ref:`vector type <t_vector>`, and the number and types of
2074 elements must match those specified by the type.
2075 **Zero initialization**
2076 The string '``zeroinitializer``' can be used to zero initialize a
2077 value to zero of *any* type, including scalar and
2078 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2079 having to print large zero initializers (e.g. for large arrays) and
2080 is always exactly equivalent to using explicit zero initializers.
2082 A metadata node is a structure-like constant with :ref:`metadata
2083 type <t_metadata>`. For example:
2084 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2085 constants that are meant to be interpreted as part of the
2086 instruction stream, metadata is a place to attach additional
2087 information such as debug info.
2089 Global Variable and Function Addresses
2090 --------------------------------------
2092 The addresses of :ref:`global variables <globalvars>` and
2093 :ref:`functions <functionstructure>` are always implicitly valid
2094 (link-time) constants. These constants are explicitly referenced when
2095 the :ref:`identifier for the global <identifiers>` is used and always have
2096 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2099 .. code-block:: llvm
2103 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2110 The string '``undef``' can be used anywhere a constant is expected, and
2111 indicates that the user of the value may receive an unspecified
2112 bit-pattern. Undefined values may be of any type (other than '``label``'
2113 or '``void``') and be used anywhere a constant is permitted.
2115 Undefined values are useful because they indicate to the compiler that
2116 the program is well defined no matter what value is used. This gives the
2117 compiler more freedom to optimize. Here are some examples of
2118 (potentially surprising) transformations that are valid (in pseudo IR):
2120 .. code-block:: llvm
2130 This is safe because all of the output bits are affected by the undef
2131 bits. Any output bit can have a zero or one depending on the input bits.
2133 .. code-block:: llvm
2144 These logical operations have bits that are not always affected by the
2145 input. For example, if ``%X`` has a zero bit, then the output of the
2146 '``and``' operation will always be a zero for that bit, no matter what
2147 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2148 optimize or assume that the result of the '``and``' is '``undef``'.
2149 However, it is safe to assume that all bits of the '``undef``' could be
2150 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2151 all the bits of the '``undef``' operand to the '``or``' could be set,
2152 allowing the '``or``' to be folded to -1.
2154 .. code-block:: llvm
2156 %A = select undef, %X, %Y
2157 %B = select undef, 42, %Y
2158 %C = select %X, %Y, undef
2168 This set of examples shows that undefined '``select``' (and conditional
2169 branch) conditions can go *either way*, but they have to come from one
2170 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2171 both known to have a clear low bit, then ``%A`` would have to have a
2172 cleared low bit. However, in the ``%C`` example, the optimizer is
2173 allowed to assume that the '``undef``' operand could be the same as
2174 ``%Y``, allowing the whole '``select``' to be eliminated.
2176 .. code-block:: llvm
2178 %A = xor undef, undef
2195 This example points out that two '``undef``' operands are not
2196 necessarily the same. This can be surprising to people (and also matches
2197 C semantics) where they assume that "``X^X``" is always zero, even if
2198 ``X`` is undefined. This isn't true for a number of reasons, but the
2199 short answer is that an '``undef``' "variable" can arbitrarily change
2200 its value over its "live range". This is true because the variable
2201 doesn't actually *have a live range*. Instead, the value is logically
2202 read from arbitrary registers that happen to be around when needed, so
2203 the value is not necessarily consistent over time. In fact, ``%A`` and
2204 ``%C`` need to have the same semantics or the core LLVM "replace all
2205 uses with" concept would not hold.
2207 .. code-block:: llvm
2215 These examples show the crucial difference between an *undefined value*
2216 and *undefined behavior*. An undefined value (like '``undef``') is
2217 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2218 operation can be constant folded to '``undef``', because the '``undef``'
2219 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2220 However, in the second example, we can make a more aggressive
2221 assumption: because the ``undef`` is allowed to be an arbitrary value,
2222 we are allowed to assume that it could be zero. Since a divide by zero
2223 has *undefined behavior*, we are allowed to assume that the operation
2224 does not execute at all. This allows us to delete the divide and all
2225 code after it. Because the undefined operation "can't happen", the
2226 optimizer can assume that it occurs in dead code.
2228 .. code-block:: llvm
2230 a: store undef -> %X
2231 b: store %X -> undef
2236 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2237 value can be assumed to not have any effect; we can assume that the
2238 value is overwritten with bits that happen to match what was already
2239 there. However, a store *to* an undefined location could clobber
2240 arbitrary memory, therefore, it has undefined behavior.
2247 Poison values are similar to :ref:`undef values <undefvalues>`, however
2248 they also represent the fact that an instruction or constant expression
2249 which cannot evoke side effects has nevertheless detected a condition
2250 which results in undefined behavior.
2252 There is currently no way of representing a poison value in the IR; they
2253 only exist when produced by operations such as :ref:`add <i_add>` with
2256 Poison value behavior is defined in terms of value *dependence*:
2258 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2259 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2260 their dynamic predecessor basic block.
2261 - Function arguments depend on the corresponding actual argument values
2262 in the dynamic callers of their functions.
2263 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2264 instructions that dynamically transfer control back to them.
2265 - :ref:`Invoke <i_invoke>` instructions depend on the
2266 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2267 call instructions that dynamically transfer control back to them.
2268 - Non-volatile loads and stores depend on the most recent stores to all
2269 of the referenced memory addresses, following the order in the IR
2270 (including loads and stores implied by intrinsics such as
2271 :ref:`@llvm.memcpy <int_memcpy>`.)
2272 - An instruction with externally visible side effects depends on the
2273 most recent preceding instruction with externally visible side
2274 effects, following the order in the IR. (This includes :ref:`volatile
2275 operations <volatile>`.)
2276 - An instruction *control-depends* on a :ref:`terminator
2277 instruction <terminators>` if the terminator instruction has
2278 multiple successors and the instruction is always executed when
2279 control transfers to one of the successors, and may not be executed
2280 when control is transferred to another.
2281 - Additionally, an instruction also *control-depends* on a terminator
2282 instruction if the set of instructions it otherwise depends on would
2283 be different if the terminator had transferred control to a different
2285 - Dependence is transitive.
2287 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2288 with the additional affect that any instruction which has a *dependence*
2289 on a poison value has undefined behavior.
2291 Here are some examples:
2293 .. code-block:: llvm
2296 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2297 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2298 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2299 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2301 store i32 %poison, i32* @g ; Poison value stored to memory.
2302 %poison2 = load i32* @g ; Poison value loaded back from memory.
2304 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2306 %narrowaddr = bitcast i32* @g to i16*
2307 %wideaddr = bitcast i32* @g to i64*
2308 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2309 %poison4 = load i64* %wideaddr ; Returns a poison value.
2311 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2312 br i1 %cmp, label %true, label %end ; Branch to either destination.
2315 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2316 ; it has undefined behavior.
2320 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2321 ; Both edges into this PHI are
2322 ; control-dependent on %cmp, so this
2323 ; always results in a poison value.
2325 store volatile i32 0, i32* @g ; This would depend on the store in %true
2326 ; if %cmp is true, or the store in %entry
2327 ; otherwise, so this is undefined behavior.
2329 br i1 %cmp, label %second_true, label %second_end
2330 ; The same branch again, but this time the
2331 ; true block doesn't have side effects.
2338 store volatile i32 0, i32* @g ; This time, the instruction always depends
2339 ; on the store in %end. Also, it is
2340 ; control-equivalent to %end, so this is
2341 ; well-defined (ignoring earlier undefined
2342 ; behavior in this example).
2346 Addresses of Basic Blocks
2347 -------------------------
2349 ``blockaddress(@function, %block)``
2351 The '``blockaddress``' constant computes the address of the specified
2352 basic block in the specified function, and always has an ``i8*`` type.
2353 Taking the address of the entry block is illegal.
2355 This value only has defined behavior when used as an operand to the
2356 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2357 against null. Pointer equality tests between labels addresses results in
2358 undefined behavior --- though, again, comparison against null is ok, and
2359 no label is equal to the null pointer. This may be passed around as an
2360 opaque pointer sized value as long as the bits are not inspected. This
2361 allows ``ptrtoint`` and arithmetic to be performed on these values so
2362 long as the original value is reconstituted before the ``indirectbr``
2365 Finally, some targets may provide defined semantics when using the value
2366 as the operand to an inline assembly, but that is target specific.
2370 Constant Expressions
2371 --------------------
2373 Constant expressions are used to allow expressions involving other
2374 constants to be used as constants. Constant expressions may be of any
2375 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2376 that does not have side effects (e.g. load and call are not supported).
2377 The following is the syntax for constant expressions:
2379 ``trunc (CST to TYPE)``
2380 Truncate a constant to another type. The bit size of CST must be
2381 larger than the bit size of TYPE. Both types must be integers.
2382 ``zext (CST to TYPE)``
2383 Zero extend a constant to another type. The bit size of CST must be
2384 smaller than the bit size of TYPE. Both types must be integers.
2385 ``sext (CST to TYPE)``
2386 Sign extend a constant to another type. The bit size of CST must be
2387 smaller than the bit size of TYPE. Both types must be integers.
2388 ``fptrunc (CST to TYPE)``
2389 Truncate a floating point constant to another floating point type.
2390 The size of CST must be larger than the size of TYPE. Both types
2391 must be floating point.
2392 ``fpext (CST to TYPE)``
2393 Floating point extend a constant to another type. The size of CST
2394 must be smaller or equal to the size of TYPE. Both types must be
2396 ``fptoui (CST to TYPE)``
2397 Convert a floating point constant to the corresponding unsigned
2398 integer constant. TYPE must be a scalar or vector integer type. CST
2399 must be of scalar or vector floating point type. Both CST and TYPE
2400 must be scalars, or vectors of the same number of elements. If the
2401 value won't fit in the integer type, the results are undefined.
2402 ``fptosi (CST to TYPE)``
2403 Convert a floating point constant to the corresponding signed
2404 integer constant. TYPE must be a scalar or vector integer type. CST
2405 must be of scalar or vector floating point type. Both CST and TYPE
2406 must be scalars, or vectors of the same number of elements. If the
2407 value won't fit in the integer type, the results are undefined.
2408 ``uitofp (CST to TYPE)``
2409 Convert an unsigned integer constant to the corresponding floating
2410 point constant. TYPE must be a scalar or vector floating point type.
2411 CST must be of scalar or vector integer type. Both CST and TYPE must
2412 be scalars, or vectors of the same number of elements. If the value
2413 won't fit in the floating point type, the results are undefined.
2414 ``sitofp (CST to TYPE)``
2415 Convert a signed integer constant to the corresponding floating
2416 point constant. TYPE must be a scalar or vector floating point type.
2417 CST must be of scalar or vector integer type. Both CST and TYPE must
2418 be scalars, or vectors of the same number of elements. If the value
2419 won't fit in the floating point type, the results are undefined.
2420 ``ptrtoint (CST to TYPE)``
2421 Convert a pointer typed constant to the corresponding integer
2422 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2423 pointer type. The ``CST`` value is zero extended, truncated, or
2424 unchanged to make it fit in ``TYPE``.
2425 ``inttoptr (CST to TYPE)``
2426 Convert an integer constant to a pointer constant. TYPE must be a
2427 pointer type. CST must be of integer type. The CST value is zero
2428 extended, truncated, or unchanged to make it fit in a pointer size.
2429 This one is *really* dangerous!
2430 ``bitcast (CST to TYPE)``
2431 Convert a constant, CST, to another TYPE. The constraints of the
2432 operands are the same as those for the :ref:`bitcast
2433 instruction <i_bitcast>`.
2434 ``addrspacecast (CST to TYPE)``
2435 Convert a constant pointer or constant vector of pointer, CST, to another
2436 TYPE in a different address space. The constraints of the operands are the
2437 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2438 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2439 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2440 constants. As with the :ref:`getelementptr <i_getelementptr>`
2441 instruction, the index list may have zero or more indexes, which are
2442 required to make sense for the type of "CSTPTR".
2443 ``select (COND, VAL1, VAL2)``
2444 Perform the :ref:`select operation <i_select>` on constants.
2445 ``icmp COND (VAL1, VAL2)``
2446 Performs the :ref:`icmp operation <i_icmp>` on constants.
2447 ``fcmp COND (VAL1, VAL2)``
2448 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2449 ``extractelement (VAL, IDX)``
2450 Perform the :ref:`extractelement operation <i_extractelement>` on
2452 ``insertelement (VAL, ELT, IDX)``
2453 Perform the :ref:`insertelement operation <i_insertelement>` on
2455 ``shufflevector (VEC1, VEC2, IDXMASK)``
2456 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2458 ``extractvalue (VAL, IDX0, IDX1, ...)``
2459 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2460 constants. The index list is interpreted in a similar manner as
2461 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2462 least one index value must be specified.
2463 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2464 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2465 The index list is interpreted in a similar manner as indices in a
2466 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2467 value must be specified.
2468 ``OPCODE (LHS, RHS)``
2469 Perform the specified operation of the LHS and RHS constants. OPCODE
2470 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2471 binary <bitwiseops>` operations. The constraints on operands are
2472 the same as those for the corresponding instruction (e.g. no bitwise
2473 operations on floating point values are allowed).
2480 Inline Assembler Expressions
2481 ----------------------------
2483 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2484 Inline Assembly <moduleasm>`) through the use of a special value. This
2485 value represents the inline assembler as a string (containing the
2486 instructions to emit), a list of operand constraints (stored as a
2487 string), a flag that indicates whether or not the inline asm expression
2488 has side effects, and a flag indicating whether the function containing
2489 the asm needs to align its stack conservatively. An example inline
2490 assembler expression is:
2492 .. code-block:: llvm
2494 i32 (i32) asm "bswap $0", "=r,r"
2496 Inline assembler expressions may **only** be used as the callee operand
2497 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2498 Thus, typically we have:
2500 .. code-block:: llvm
2502 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2504 Inline asms with side effects not visible in the constraint list must be
2505 marked as having side effects. This is done through the use of the
2506 '``sideeffect``' keyword, like so:
2508 .. code-block:: llvm
2510 call void asm sideeffect "eieio", ""()
2512 In some cases inline asms will contain code that will not work unless
2513 the stack is aligned in some way, such as calls or SSE instructions on
2514 x86, yet will not contain code that does that alignment within the asm.
2515 The compiler should make conservative assumptions about what the asm
2516 might contain and should generate its usual stack alignment code in the
2517 prologue if the '``alignstack``' keyword is present:
2519 .. code-block:: llvm
2521 call void asm alignstack "eieio", ""()
2523 Inline asms also support using non-standard assembly dialects. The
2524 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2525 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2526 the only supported dialects. An example is:
2528 .. code-block:: llvm
2530 call void asm inteldialect "eieio", ""()
2532 If multiple keywords appear the '``sideeffect``' keyword must come
2533 first, the '``alignstack``' keyword second and the '``inteldialect``'
2539 The call instructions that wrap inline asm nodes may have a
2540 "``!srcloc``" MDNode attached to it that contains a list of constant
2541 integers. If present, the code generator will use the integer as the
2542 location cookie value when report errors through the ``LLVMContext``
2543 error reporting mechanisms. This allows a front-end to correlate backend
2544 errors that occur with inline asm back to the source code that produced
2547 .. code-block:: llvm
2549 call void asm sideeffect "something bad", ""(), !srcloc !42
2551 !42 = !{ i32 1234567 }
2553 It is up to the front-end to make sense of the magic numbers it places
2554 in the IR. If the MDNode contains multiple constants, the code generator
2555 will use the one that corresponds to the line of the asm that the error
2560 Metadata Nodes and Metadata Strings
2561 -----------------------------------
2563 LLVM IR allows metadata to be attached to instructions in the program
2564 that can convey extra information about the code to the optimizers and
2565 code generator. One example application of metadata is source-level
2566 debug information. There are two metadata primitives: strings and nodes.
2567 All metadata has the ``metadata`` type and is identified in syntax by a
2568 preceding exclamation point ('``!``').
2570 A metadata string is a string surrounded by double quotes. It can
2571 contain any character by escaping non-printable characters with
2572 "``\xx``" where "``xx``" is the two digit hex code. For example:
2575 Metadata nodes are represented with notation similar to structure
2576 constants (a comma separated list of elements, surrounded by braces and
2577 preceded by an exclamation point). Metadata nodes can have any values as
2578 their operand. For example:
2580 .. code-block:: llvm
2582 !{ metadata !"test\00", i32 10}
2584 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2585 metadata nodes, which can be looked up in the module symbol table. For
2588 .. code-block:: llvm
2590 !foo = metadata !{!4, !3}
2592 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2593 function is using two metadata arguments:
2595 .. code-block:: llvm
2597 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2599 Metadata can be attached with an instruction. Here metadata ``!21`` is
2600 attached to the ``add`` instruction using the ``!dbg`` identifier:
2602 .. code-block:: llvm
2604 %indvar.next = add i64 %indvar, 1, !dbg !21
2606 More information about specific metadata nodes recognized by the
2607 optimizers and code generator is found below.
2612 In LLVM IR, memory does not have types, so LLVM's own type system is not
2613 suitable for doing TBAA. Instead, metadata is added to the IR to
2614 describe a type system of a higher level language. This can be used to
2615 implement typical C/C++ TBAA, but it can also be used to implement
2616 custom alias analysis behavior for other languages.
2618 The current metadata format is very simple. TBAA metadata nodes have up
2619 to three fields, e.g.:
2621 .. code-block:: llvm
2623 !0 = metadata !{ metadata !"an example type tree" }
2624 !1 = metadata !{ metadata !"int", metadata !0 }
2625 !2 = metadata !{ metadata !"float", metadata !0 }
2626 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2628 The first field is an identity field. It can be any value, usually a
2629 metadata string, which uniquely identifies the type. The most important
2630 name in the tree is the name of the root node. Two trees with different
2631 root node names are entirely disjoint, even if they have leaves with
2634 The second field identifies the type's parent node in the tree, or is
2635 null or omitted for a root node. A type is considered to alias all of
2636 its descendants and all of its ancestors in the tree. Also, a type is
2637 considered to alias all types in other trees, so that bitcode produced
2638 from multiple front-ends is handled conservatively.
2640 If the third field is present, it's an integer which if equal to 1
2641 indicates that the type is "constant" (meaning
2642 ``pointsToConstantMemory`` should return true; see `other useful
2643 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2645 '``tbaa.struct``' Metadata
2646 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2648 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2649 aggregate assignment operations in C and similar languages, however it
2650 is defined to copy a contiguous region of memory, which is more than
2651 strictly necessary for aggregate types which contain holes due to
2652 padding. Also, it doesn't contain any TBAA information about the fields
2655 ``!tbaa.struct`` metadata can describe which memory subregions in a
2656 memcpy are padding and what the TBAA tags of the struct are.
2658 The current metadata format is very simple. ``!tbaa.struct`` metadata
2659 nodes are a list of operands which are in conceptual groups of three.
2660 For each group of three, the first operand gives the byte offset of a
2661 field in bytes, the second gives its size in bytes, and the third gives
2664 .. code-block:: llvm
2666 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2668 This describes a struct with two fields. The first is at offset 0 bytes
2669 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2670 and has size 4 bytes and has tbaa tag !2.
2672 Note that the fields need not be contiguous. In this example, there is a
2673 4 byte gap between the two fields. This gap represents padding which
2674 does not carry useful data and need not be preserved.
2676 '``fpmath``' Metadata
2677 ^^^^^^^^^^^^^^^^^^^^^
2679 ``fpmath`` metadata may be attached to any instruction of floating point
2680 type. It can be used to express the maximum acceptable error in the
2681 result of that instruction, in ULPs, thus potentially allowing the
2682 compiler to use a more efficient but less accurate method of computing
2683 it. ULP is defined as follows:
2685 If ``x`` is a real number that lies between two finite consecutive
2686 floating-point numbers ``a`` and ``b``, without being equal to one
2687 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2688 distance between the two non-equal finite floating-point numbers
2689 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2691 The metadata node shall consist of a single positive floating point
2692 number representing the maximum relative error, for example:
2694 .. code-block:: llvm
2696 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2698 '``range``' Metadata
2699 ^^^^^^^^^^^^^^^^^^^^
2701 ``range`` metadata may be attached only to loads of integer types. It
2702 expresses the possible ranges the loaded value is in. The ranges are
2703 represented with a flattened list of integers. The loaded value is known
2704 to be in the union of the ranges defined by each consecutive pair. Each
2705 pair has the following properties:
2707 - The type must match the type loaded by the instruction.
2708 - The pair ``a,b`` represents the range ``[a,b)``.
2709 - Both ``a`` and ``b`` are constants.
2710 - The range is allowed to wrap.
2711 - The range should not represent the full or empty set. That is,
2714 In addition, the pairs must be in signed order of the lower bound and
2715 they must be non-contiguous.
2719 .. code-block:: llvm
2721 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2722 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2723 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2724 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2726 !0 = metadata !{ i8 0, i8 2 }
2727 !1 = metadata !{ i8 255, i8 2 }
2728 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2729 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2734 It is sometimes useful to attach information to loop constructs. Currently,
2735 loop metadata is implemented as metadata attached to the branch instruction
2736 in the loop latch block. This type of metadata refer to a metadata node that is
2737 guaranteed to be separate for each loop. The loop identifier metadata is
2738 specified with the name ``llvm.loop``.
2740 The loop identifier metadata is implemented using a metadata that refers to
2741 itself to avoid merging it with any other identifier metadata, e.g.,
2742 during module linkage or function inlining. That is, each loop should refer
2743 to their own identification metadata even if they reside in separate functions.
2744 The following example contains loop identifier metadata for two separate loop
2747 .. code-block:: llvm
2749 !0 = metadata !{ metadata !0 }
2750 !1 = metadata !{ metadata !1 }
2752 The loop identifier metadata can be used to specify additional per-loop
2753 metadata. Any operands after the first operand can be treated as user-defined
2754 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2755 by the loop vectorizer to indicate how many times to unroll the loop:
2757 .. code-block:: llvm
2759 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2761 !0 = metadata !{ metadata !0, metadata !1 }
2762 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2767 Metadata types used to annotate memory accesses with information helpful
2768 for optimizations are prefixed with ``llvm.mem``.
2770 '``llvm.mem.parallel_loop_access``' Metadata
2771 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2773 For a loop to be parallel, in addition to using
2774 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2775 also all of the memory accessing instructions in the loop body need to be
2776 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2777 is at least one memory accessing instruction not marked with the metadata,
2778 the loop must be considered a sequential loop. This causes parallel loops to be
2779 converted to sequential loops due to optimization passes that are unaware of
2780 the parallel semantics and that insert new memory instructions to the loop
2783 Example of a loop that is considered parallel due to its correct use of
2784 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2785 metadata types that refer to the same loop identifier metadata.
2787 .. code-block:: llvm
2791 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
2793 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2795 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2799 !0 = metadata !{ metadata !0 }
2801 It is also possible to have nested parallel loops. In that case the
2802 memory accesses refer to a list of loop identifier metadata nodes instead of
2803 the loop identifier metadata node directly:
2805 .. code-block:: llvm
2809 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
2811 br label %inner.for.body
2815 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2817 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
2819 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2823 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
2825 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2827 outer.for.end: ; preds = %for.body
2829 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2830 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2831 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2833 '``llvm.vectorizer``'
2834 ^^^^^^^^^^^^^^^^^^^^^
2836 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2837 vectorization parameters such as vectorization factor and unroll factor.
2839 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2840 loop identification metadata.
2842 '``llvm.vectorizer.unroll``' Metadata
2843 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2845 This metadata instructs the loop vectorizer to unroll the specified
2846 loop exactly ``N`` times.
2848 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2849 operand is an integer specifying the unroll factor. For example:
2851 .. code-block:: llvm
2853 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2855 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2858 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2859 determined automatically.
2861 '``llvm.vectorizer.width``' Metadata
2862 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2864 This metadata sets the target width of the vectorizer to ``N``. Without
2865 this metadata, the vectorizer will choose a width automatically.
2866 Regardless of this metadata, the vectorizer will only vectorize loops if
2867 it believes it is valid to do so.
2869 The first operand is the string ``llvm.vectorizer.width`` and the second
2870 operand is an integer specifying the width. For example:
2872 .. code-block:: llvm
2874 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2876 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2879 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2882 Module Flags Metadata
2883 =====================
2885 Information about the module as a whole is difficult to convey to LLVM's
2886 subsystems. The LLVM IR isn't sufficient to transmit this information.
2887 The ``llvm.module.flags`` named metadata exists in order to facilitate
2888 this. These flags are in the form of key / value pairs --- much like a
2889 dictionary --- making it easy for any subsystem who cares about a flag to
2892 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2893 Each triplet has the following form:
2895 - The first element is a *behavior* flag, which specifies the behavior
2896 when two (or more) modules are merged together, and it encounters two
2897 (or more) metadata with the same ID. The supported behaviors are
2899 - The second element is a metadata string that is a unique ID for the
2900 metadata. Each module may only have one flag entry for each unique ID (not
2901 including entries with the **Require** behavior).
2902 - The third element is the value of the flag.
2904 When two (or more) modules are merged together, the resulting
2905 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2906 each unique metadata ID string, there will be exactly one entry in the merged
2907 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2908 be determined by the merge behavior flag, as described below. The only exception
2909 is that entries with the *Require* behavior are always preserved.
2911 The following behaviors are supported:
2922 Emits an error if two values disagree, otherwise the resulting value
2923 is that of the operands.
2927 Emits a warning if two values disagree. The result value will be the
2928 operand for the flag from the first module being linked.
2932 Adds a requirement that another module flag be present and have a
2933 specified value after linking is performed. The value must be a
2934 metadata pair, where the first element of the pair is the ID of the
2935 module flag to be restricted, and the second element of the pair is
2936 the value the module flag should be restricted to. This behavior can
2937 be used to restrict the allowable results (via triggering of an
2938 error) of linking IDs with the **Override** behavior.
2942 Uses the specified value, regardless of the behavior or value of the
2943 other module. If both modules specify **Override**, but the values
2944 differ, an error will be emitted.
2948 Appends the two values, which are required to be metadata nodes.
2952 Appends the two values, which are required to be metadata
2953 nodes. However, duplicate entries in the second list are dropped
2954 during the append operation.
2956 It is an error for a particular unique flag ID to have multiple behaviors,
2957 except in the case of **Require** (which adds restrictions on another metadata
2958 value) or **Override**.
2960 An example of module flags:
2962 .. code-block:: llvm
2964 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2965 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2966 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2967 !3 = metadata !{ i32 3, metadata !"qux",
2969 metadata !"foo", i32 1
2972 !llvm.module.flags = !{ !0, !1, !2, !3 }
2974 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2975 if two or more ``!"foo"`` flags are seen is to emit an error if their
2976 values are not equal.
2978 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2979 behavior if two or more ``!"bar"`` flags are seen is to use the value
2982 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2983 behavior if two or more ``!"qux"`` flags are seen is to emit a
2984 warning if their values are not equal.
2986 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2990 metadata !{ metadata !"foo", i32 1 }
2992 The behavior is to emit an error if the ``llvm.module.flags`` does not
2993 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2996 Objective-C Garbage Collection Module Flags Metadata
2997 ----------------------------------------------------
2999 On the Mach-O platform, Objective-C stores metadata about garbage
3000 collection in a special section called "image info". The metadata
3001 consists of a version number and a bitmask specifying what types of
3002 garbage collection are supported (if any) by the file. If two or more
3003 modules are linked together their garbage collection metadata needs to
3004 be merged rather than appended together.
3006 The Objective-C garbage collection module flags metadata consists of the
3007 following key-value pairs:
3016 * - ``Objective-C Version``
3017 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3019 * - ``Objective-C Image Info Version``
3020 - **[Required]** --- The version of the image info section. Currently
3023 * - ``Objective-C Image Info Section``
3024 - **[Required]** --- The section to place the metadata. Valid values are
3025 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3026 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3027 Objective-C ABI version 2.
3029 * - ``Objective-C Garbage Collection``
3030 - **[Required]** --- Specifies whether garbage collection is supported or
3031 not. Valid values are 0, for no garbage collection, and 2, for garbage
3032 collection supported.
3034 * - ``Objective-C GC Only``
3035 - **[Optional]** --- Specifies that only garbage collection is supported.
3036 If present, its value must be 6. This flag requires that the
3037 ``Objective-C Garbage Collection`` flag have the value 2.
3039 Some important flag interactions:
3041 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3042 merged with a module with ``Objective-C Garbage Collection`` set to
3043 2, then the resulting module has the
3044 ``Objective-C Garbage Collection`` flag set to 0.
3045 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3046 merged with a module with ``Objective-C GC Only`` set to 6.
3048 Automatic Linker Flags Module Flags Metadata
3049 --------------------------------------------
3051 Some targets support embedding flags to the linker inside individual object
3052 files. Typically this is used in conjunction with language extensions which
3053 allow source files to explicitly declare the libraries they depend on, and have
3054 these automatically be transmitted to the linker via object files.
3056 These flags are encoded in the IR using metadata in the module flags section,
3057 using the ``Linker Options`` key. The merge behavior for this flag is required
3058 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3059 node which should be a list of other metadata nodes, each of which should be a
3060 list of metadata strings defining linker options.
3062 For example, the following metadata section specifies two separate sets of
3063 linker options, presumably to link against ``libz`` and the ``Cocoa``
3066 !0 = metadata !{ i32 6, metadata !"Linker Options",
3068 metadata !{ metadata !"-lz" },
3069 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3070 !llvm.module.flags = !{ !0 }
3072 The metadata encoding as lists of lists of options, as opposed to a collapsed
3073 list of options, is chosen so that the IR encoding can use multiple option
3074 strings to specify e.g., a single library, while still having that specifier be
3075 preserved as an atomic element that can be recognized by a target specific
3076 assembly writer or object file emitter.
3078 Each individual option is required to be either a valid option for the target's
3079 linker, or an option that is reserved by the target specific assembly writer or
3080 object file emitter. No other aspect of these options is defined by the IR.
3082 .. _intrinsicglobalvariables:
3084 Intrinsic Global Variables
3085 ==========================
3087 LLVM has a number of "magic" global variables that contain data that
3088 affect code generation or other IR semantics. These are documented here.
3089 All globals of this sort should have a section specified as
3090 "``llvm.metadata``". This section and all globals that start with
3091 "``llvm.``" are reserved for use by LLVM.
3095 The '``llvm.used``' Global Variable
3096 -----------------------------------
3098 The ``@llvm.used`` global is an array which has
3099 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3100 pointers to named global variables, functions and aliases which may optionally
3101 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3104 .. code-block:: llvm
3109 @llvm.used = appending global [2 x i8*] [
3111 i8* bitcast (i32* @Y to i8*)
3112 ], section "llvm.metadata"
3114 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3115 and linker are required to treat the symbol as if there is a reference to the
3116 symbol that it cannot see (which is why they have to be named). For example, if
3117 a variable has internal linkage and no references other than that from the
3118 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3119 references from inline asms and other things the compiler cannot "see", and
3120 corresponds to "``attribute((used))``" in GNU C.
3122 On some targets, the code generator must emit a directive to the
3123 assembler or object file to prevent the assembler and linker from
3124 molesting the symbol.
3126 .. _gv_llvmcompilerused:
3128 The '``llvm.compiler.used``' Global Variable
3129 --------------------------------------------
3131 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3132 directive, except that it only prevents the compiler from touching the
3133 symbol. On targets that support it, this allows an intelligent linker to
3134 optimize references to the symbol without being impeded as it would be
3137 This is a rare construct that should only be used in rare circumstances,
3138 and should not be exposed to source languages.
3140 .. _gv_llvmglobalctors:
3142 The '``llvm.global_ctors``' Global Variable
3143 -------------------------------------------
3145 .. code-block:: llvm
3147 %0 = type { i32, void ()* }
3148 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3150 The ``@llvm.global_ctors`` array contains a list of constructor
3151 functions and associated priorities. The functions referenced by this
3152 array will be called in ascending order of priority (i.e. lowest first)
3153 when the module is loaded. The order of functions with the same priority
3156 .. _llvmglobaldtors:
3158 The '``llvm.global_dtors``' Global Variable
3159 -------------------------------------------
3161 .. code-block:: llvm
3163 %0 = type { i32, void ()* }
3164 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3166 The ``@llvm.global_dtors`` array contains a list of destructor functions
3167 and associated priorities. The functions referenced by this array will
3168 be called in descending order of priority (i.e. highest first) when the
3169 module is loaded. The order of functions with the same priority is not
3172 Instruction Reference
3173 =====================
3175 The LLVM instruction set consists of several different classifications
3176 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3177 instructions <binaryops>`, :ref:`bitwise binary
3178 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3179 :ref:`other instructions <otherops>`.
3183 Terminator Instructions
3184 -----------------------
3186 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3187 program ends with a "Terminator" instruction, which indicates which
3188 block should be executed after the current block is finished. These
3189 terminator instructions typically yield a '``void``' value: they produce
3190 control flow, not values (the one exception being the
3191 ':ref:`invoke <i_invoke>`' instruction).
3193 The terminator instructions are: ':ref:`ret <i_ret>`',
3194 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3195 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3196 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3200 '``ret``' Instruction
3201 ^^^^^^^^^^^^^^^^^^^^^
3208 ret <type> <value> ; Return a value from a non-void function
3209 ret void ; Return from void function
3214 The '``ret``' instruction is used to return control flow (and optionally
3215 a value) from a function back to the caller.
3217 There are two forms of the '``ret``' instruction: one that returns a
3218 value and then causes control flow, and one that just causes control
3224 The '``ret``' instruction optionally accepts a single argument, the
3225 return value. The type of the return value must be a ':ref:`first
3226 class <t_firstclass>`' type.
3228 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3229 return type and contains a '``ret``' instruction with no return value or
3230 a return value with a type that does not match its type, or if it has a
3231 void return type and contains a '``ret``' instruction with a return
3237 When the '``ret``' instruction is executed, control flow returns back to
3238 the calling function's context. If the caller is a
3239 ":ref:`call <i_call>`" instruction, execution continues at the
3240 instruction after the call. If the caller was an
3241 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3242 beginning of the "normal" destination block. If the instruction returns
3243 a value, that value shall set the call or invoke instruction's return
3249 .. code-block:: llvm
3251 ret i32 5 ; Return an integer value of 5
3252 ret void ; Return from a void function
3253 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3257 '``br``' Instruction
3258 ^^^^^^^^^^^^^^^^^^^^
3265 br i1 <cond>, label <iftrue>, label <iffalse>
3266 br label <dest> ; Unconditional branch
3271 The '``br``' instruction is used to cause control flow to transfer to a
3272 different basic block in the current function. There are two forms of
3273 this instruction, corresponding to a conditional branch and an
3274 unconditional branch.
3279 The conditional branch form of the '``br``' instruction takes a single
3280 '``i1``' value and two '``label``' values. The unconditional form of the
3281 '``br``' instruction takes a single '``label``' value as a target.
3286 Upon execution of a conditional '``br``' instruction, the '``i1``'
3287 argument is evaluated. If the value is ``true``, control flows to the
3288 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3289 to the '``iffalse``' ``label`` argument.
3294 .. code-block:: llvm
3297 %cond = icmp eq i32 %a, %b
3298 br i1 %cond, label %IfEqual, label %IfUnequal
3306 '``switch``' Instruction
3307 ^^^^^^^^^^^^^^^^^^^^^^^^
3314 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3319 The '``switch``' instruction is used to transfer control flow to one of
3320 several different places. It is a generalization of the '``br``'
3321 instruction, allowing a branch to occur to one of many possible
3327 The '``switch``' instruction uses three parameters: an integer
3328 comparison value '``value``', a default '``label``' destination, and an
3329 array of pairs of comparison value constants and '``label``'s. The table
3330 is not allowed to contain duplicate constant entries.
3335 The ``switch`` instruction specifies a table of values and destinations.
3336 When the '``switch``' instruction is executed, this table is searched
3337 for the given value. If the value is found, control flow is transferred
3338 to the corresponding destination; otherwise, control flow is transferred
3339 to the default destination.
3344 Depending on properties of the target machine and the particular
3345 ``switch`` instruction, this instruction may be code generated in
3346 different ways. For example, it could be generated as a series of
3347 chained conditional branches or with a lookup table.
3352 .. code-block:: llvm
3354 ; Emulate a conditional br instruction
3355 %Val = zext i1 %value to i32
3356 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3358 ; Emulate an unconditional br instruction
3359 switch i32 0, label %dest [ ]
3361 ; Implement a jump table:
3362 switch i32 %val, label %otherwise [ i32 0, label %onzero
3364 i32 2, label %ontwo ]
3368 '``indirectbr``' Instruction
3369 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3376 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3381 The '``indirectbr``' instruction implements an indirect branch to a
3382 label within the current function, whose address is specified by
3383 "``address``". Address must be derived from a
3384 :ref:`blockaddress <blockaddress>` constant.
3389 The '``address``' argument is the address of the label to jump to. The
3390 rest of the arguments indicate the full set of possible destinations
3391 that the address may point to. Blocks are allowed to occur multiple
3392 times in the destination list, though this isn't particularly useful.
3394 This destination list is required so that dataflow analysis has an
3395 accurate understanding of the CFG.
3400 Control transfers to the block specified in the address argument. All
3401 possible destination blocks must be listed in the label list, otherwise
3402 this instruction has undefined behavior. This implies that jumps to
3403 labels defined in other functions have undefined behavior as well.
3408 This is typically implemented with a jump through a register.
3413 .. code-block:: llvm
3415 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3419 '``invoke``' Instruction
3420 ^^^^^^^^^^^^^^^^^^^^^^^^
3427 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3428 to label <normal label> unwind label <exception label>
3433 The '``invoke``' instruction causes control to transfer to a specified
3434 function, with the possibility of control flow transfer to either the
3435 '``normal``' label or the '``exception``' label. If the callee function
3436 returns with the "``ret``" instruction, control flow will return to the
3437 "normal" label. If the callee (or any indirect callees) returns via the
3438 ":ref:`resume <i_resume>`" instruction or other exception handling
3439 mechanism, control is interrupted and continued at the dynamically
3440 nearest "exception" label.
3442 The '``exception``' label is a `landing
3443 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3444 '``exception``' label is required to have the
3445 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3446 information about the behavior of the program after unwinding happens,
3447 as its first non-PHI instruction. The restrictions on the
3448 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3449 instruction, so that the important information contained within the
3450 "``landingpad``" instruction can't be lost through normal code motion.
3455 This instruction requires several arguments:
3457 #. The optional "cconv" marker indicates which :ref:`calling
3458 convention <callingconv>` the call should use. If none is
3459 specified, the call defaults to using C calling conventions.
3460 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3461 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3463 #. '``ptr to function ty``': shall be the signature of the pointer to
3464 function value being invoked. In most cases, this is a direct
3465 function invocation, but indirect ``invoke``'s are just as possible,
3466 branching off an arbitrary pointer to function value.
3467 #. '``function ptr val``': An LLVM value containing a pointer to a
3468 function to be invoked.
3469 #. '``function args``': argument list whose types match the function
3470 signature argument types and parameter attributes. All arguments must
3471 be of :ref:`first class <t_firstclass>` type. If the function signature
3472 indicates the function accepts a variable number of arguments, the
3473 extra arguments can be specified.
3474 #. '``normal label``': the label reached when the called function
3475 executes a '``ret``' instruction.
3476 #. '``exception label``': the label reached when a callee returns via
3477 the :ref:`resume <i_resume>` instruction or other exception handling
3479 #. The optional :ref:`function attributes <fnattrs>` list. Only
3480 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3481 attributes are valid here.
3486 This instruction is designed to operate as a standard '``call``'
3487 instruction in most regards. The primary difference is that it
3488 establishes an association with a label, which is used by the runtime
3489 library to unwind the stack.
3491 This instruction is used in languages with destructors to ensure that
3492 proper cleanup is performed in the case of either a ``longjmp`` or a
3493 thrown exception. Additionally, this is important for implementation of
3494 '``catch``' clauses in high-level languages that support them.
3496 For the purposes of the SSA form, the definition of the value returned
3497 by the '``invoke``' instruction is deemed to occur on the edge from the
3498 current block to the "normal" label. If the callee unwinds then no
3499 return value is available.
3504 .. code-block:: llvm
3506 %retval = invoke i32 @Test(i32 15) to label %Continue
3507 unwind label %TestCleanup ; {i32}:retval set
3508 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3509 unwind label %TestCleanup ; {i32}:retval set
3513 '``resume``' Instruction
3514 ^^^^^^^^^^^^^^^^^^^^^^^^
3521 resume <type> <value>
3526 The '``resume``' instruction is a terminator instruction that has no
3532 The '``resume``' instruction requires one argument, which must have the
3533 same type as the result of any '``landingpad``' instruction in the same
3539 The '``resume``' instruction resumes propagation of an existing
3540 (in-flight) exception whose unwinding was interrupted with a
3541 :ref:`landingpad <i_landingpad>` instruction.
3546 .. code-block:: llvm
3548 resume { i8*, i32 } %exn
3552 '``unreachable``' Instruction
3553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3565 The '``unreachable``' instruction has no defined semantics. This
3566 instruction is used to inform the optimizer that a particular portion of
3567 the code is not reachable. This can be used to indicate that the code
3568 after a no-return function cannot be reached, and other facts.
3573 The '``unreachable``' instruction has no defined semantics.
3580 Binary operators are used to do most of the computation in a program.
3581 They require two operands of the same type, execute an operation on
3582 them, and produce a single value. The operands might represent multiple
3583 data, as is the case with the :ref:`vector <t_vector>` data type. The
3584 result value has the same type as its operands.
3586 There are several different binary operators:
3590 '``add``' Instruction
3591 ^^^^^^^^^^^^^^^^^^^^^
3598 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3599 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3600 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3601 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3606 The '``add``' instruction returns the sum of its two operands.
3611 The two arguments to the '``add``' instruction must be
3612 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3613 arguments must have identical types.
3618 The value produced is the integer sum of the two operands.
3620 If the sum has unsigned overflow, the result returned is the
3621 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3624 Because LLVM integers use a two's complement representation, this
3625 instruction is appropriate for both signed and unsigned integers.
3627 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3628 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3629 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3630 unsigned and/or signed overflow, respectively, occurs.
3635 .. code-block:: llvm
3637 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3641 '``fadd``' Instruction
3642 ^^^^^^^^^^^^^^^^^^^^^^
3649 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3654 The '``fadd``' instruction returns the sum of its two operands.
3659 The two arguments to the '``fadd``' instruction must be :ref:`floating
3660 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3661 Both arguments must have identical types.
3666 The value produced is the floating point sum of the two operands. This
3667 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3668 which are optimization hints to enable otherwise unsafe floating point
3674 .. code-block:: llvm
3676 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3678 '``sub``' Instruction
3679 ^^^^^^^^^^^^^^^^^^^^^
3686 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3687 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3688 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3689 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3694 The '``sub``' instruction returns the difference of its two operands.
3696 Note that the '``sub``' instruction is used to represent the '``neg``'
3697 instruction present in most other intermediate representations.
3702 The two arguments to the '``sub``' instruction must be
3703 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3704 arguments must have identical types.
3709 The value produced is the integer difference of the two operands.
3711 If the difference has unsigned overflow, the result returned is the
3712 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3715 Because LLVM integers use a two's complement representation, this
3716 instruction is appropriate for both signed and unsigned integers.
3718 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3719 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3720 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3721 unsigned and/or signed overflow, respectively, occurs.
3726 .. code-block:: llvm
3728 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3729 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3733 '``fsub``' Instruction
3734 ^^^^^^^^^^^^^^^^^^^^^^
3741 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3746 The '``fsub``' instruction returns the difference of its two operands.
3748 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3749 instruction present in most other intermediate representations.
3754 The two arguments to the '``fsub``' instruction must be :ref:`floating
3755 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3756 Both arguments must have identical types.
3761 The value produced is the floating point difference of the two operands.
3762 This instruction can also take any number of :ref:`fast-math
3763 flags <fastmath>`, which are optimization hints to enable otherwise
3764 unsafe floating point optimizations:
3769 .. code-block:: llvm
3771 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3772 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3774 '``mul``' Instruction
3775 ^^^^^^^^^^^^^^^^^^^^^
3782 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3783 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3784 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3785 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3790 The '``mul``' instruction returns the product of its two operands.
3795 The two arguments to the '``mul``' instruction must be
3796 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3797 arguments must have identical types.
3802 The value produced is the integer product of the two operands.
3804 If the result of the multiplication has unsigned overflow, the result
3805 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3806 bit width of the result.
3808 Because LLVM integers use a two's complement representation, and the
3809 result is the same width as the operands, this instruction returns the
3810 correct result for both signed and unsigned integers. If a full product
3811 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3812 sign-extended or zero-extended as appropriate to the width of the full
3815 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3816 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3817 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3818 unsigned and/or signed overflow, respectively, occurs.
3823 .. code-block:: llvm
3825 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3829 '``fmul``' Instruction
3830 ^^^^^^^^^^^^^^^^^^^^^^
3837 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3842 The '``fmul``' instruction returns the product of its two operands.
3847 The two arguments to the '``fmul``' instruction must be :ref:`floating
3848 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3849 Both arguments must have identical types.
3854 The value produced is the floating point product of the two operands.
3855 This instruction can also take any number of :ref:`fast-math
3856 flags <fastmath>`, which are optimization hints to enable otherwise
3857 unsafe floating point optimizations:
3862 .. code-block:: llvm
3864 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3866 '``udiv``' Instruction
3867 ^^^^^^^^^^^^^^^^^^^^^^
3874 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3875 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3880 The '``udiv``' instruction returns the quotient of its two operands.
3885 The two arguments to the '``udiv``' instruction must be
3886 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3887 arguments must have identical types.
3892 The value produced is the unsigned integer quotient of the two operands.
3894 Note that unsigned integer division and signed integer division are
3895 distinct operations; for signed integer division, use '``sdiv``'.
3897 Division by zero leads to undefined behavior.
3899 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3900 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3901 such, "((a udiv exact b) mul b) == a").
3906 .. code-block:: llvm
3908 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3910 '``sdiv``' Instruction
3911 ^^^^^^^^^^^^^^^^^^^^^^
3918 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3919 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3924 The '``sdiv``' instruction returns the quotient of its two operands.
3929 The two arguments to the '``sdiv``' instruction must be
3930 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3931 arguments must have identical types.
3936 The value produced is the signed integer quotient of the two operands
3937 rounded towards zero.
3939 Note that signed integer division and unsigned integer division are
3940 distinct operations; for unsigned integer division, use '``udiv``'.
3942 Division by zero leads to undefined behavior. Overflow also leads to
3943 undefined behavior; this is a rare case, but can occur, for example, by
3944 doing a 32-bit division of -2147483648 by -1.
3946 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3947 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3952 .. code-block:: llvm
3954 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3958 '``fdiv``' Instruction
3959 ^^^^^^^^^^^^^^^^^^^^^^
3966 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3971 The '``fdiv``' instruction returns the quotient of its two operands.
3976 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3977 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3978 Both arguments must have identical types.
3983 The value produced is the floating point quotient of the two operands.
3984 This instruction can also take any number of :ref:`fast-math
3985 flags <fastmath>`, which are optimization hints to enable otherwise
3986 unsafe floating point optimizations:
3991 .. code-block:: llvm
3993 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3995 '``urem``' Instruction
3996 ^^^^^^^^^^^^^^^^^^^^^^
4003 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
4008 The '``urem``' instruction returns the remainder from the unsigned
4009 division of its two arguments.
4014 The two arguments to the '``urem``' instruction must be
4015 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4016 arguments must have identical types.
4021 This instruction returns the unsigned integer *remainder* of a division.
4022 This instruction always performs an unsigned division to get the
4025 Note that unsigned integer remainder and signed integer remainder are
4026 distinct operations; for signed integer remainder, use '``srem``'.
4028 Taking the remainder of a division by zero leads to undefined behavior.
4033 .. code-block:: llvm
4035 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
4037 '``srem``' Instruction
4038 ^^^^^^^^^^^^^^^^^^^^^^
4045 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
4050 The '``srem``' instruction returns the remainder from the signed
4051 division of its two operands. This instruction can also take
4052 :ref:`vector <t_vector>` versions of the values in which case the elements
4058 The two arguments to the '``srem``' instruction must be
4059 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4060 arguments must have identical types.
4065 This instruction returns the *remainder* of a division (where the result
4066 is either zero or has the same sign as the dividend, ``op1``), not the
4067 *modulo* operator (where the result is either zero or has the same sign
4068 as the divisor, ``op2``) of a value. For more information about the
4069 difference, see `The Math
4070 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4071 table of how this is implemented in various languages, please see
4073 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4075 Note that signed integer remainder and unsigned integer remainder are
4076 distinct operations; for unsigned integer remainder, use '``urem``'.
4078 Taking the remainder of a division by zero leads to undefined behavior.
4079 Overflow also leads to undefined behavior; this is a rare case, but can
4080 occur, for example, by taking the remainder of a 32-bit division of
4081 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4082 rule lets srem be implemented using instructions that return both the
4083 result of the division and the remainder.)
4088 .. code-block:: llvm
4090 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4094 '``frem``' Instruction
4095 ^^^^^^^^^^^^^^^^^^^^^^
4102 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4107 The '``frem``' instruction returns the remainder from the division of
4113 The two arguments to the '``frem``' instruction must be :ref:`floating
4114 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4115 Both arguments must have identical types.
4120 This instruction returns the *remainder* of a division. The remainder
4121 has the same sign as the dividend. This instruction can also take any
4122 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4123 to enable otherwise unsafe floating point optimizations:
4128 .. code-block:: llvm
4130 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4134 Bitwise Binary Operations
4135 -------------------------
4137 Bitwise binary operators are used to do various forms of bit-twiddling
4138 in a program. They are generally very efficient instructions and can
4139 commonly be strength reduced from other instructions. They require two
4140 operands of the same type, execute an operation on them, and produce a
4141 single value. The resulting value is the same type as its operands.
4143 '``shl``' Instruction
4144 ^^^^^^^^^^^^^^^^^^^^^
4151 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4152 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4153 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4154 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4159 The '``shl``' instruction returns the first operand shifted to the left
4160 a specified number of bits.
4165 Both arguments to the '``shl``' instruction must be the same
4166 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4167 '``op2``' is treated as an unsigned value.
4172 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4173 where ``n`` is the width of the result. If ``op2`` is (statically or
4174 dynamically) negative or equal to or larger than the number of bits in
4175 ``op1``, the result is undefined. If the arguments are vectors, each
4176 vector element of ``op1`` is shifted by the corresponding shift amount
4179 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4180 value <poisonvalues>` if it shifts out any non-zero bits. If the
4181 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4182 value <poisonvalues>` if it shifts out any bits that disagree with the
4183 resultant sign bit. As such, NUW/NSW have the same semantics as they
4184 would if the shift were expressed as a mul instruction with the same
4185 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4190 .. code-block:: llvm
4192 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4193 <result> = shl i32 4, 2 ; yields {i32}: 16
4194 <result> = shl i32 1, 10 ; yields {i32}: 1024
4195 <result> = shl i32 1, 32 ; undefined
4196 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4198 '``lshr``' Instruction
4199 ^^^^^^^^^^^^^^^^^^^^^^
4206 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4207 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4212 The '``lshr``' instruction (logical shift right) returns the first
4213 operand shifted to the right a specified number of bits with zero fill.
4218 Both arguments to the '``lshr``' instruction must be the same
4219 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4220 '``op2``' is treated as an unsigned value.
4225 This instruction always performs a logical shift right operation. The
4226 most significant bits of the result will be filled with zero bits after
4227 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4228 than the number of bits in ``op1``, the result is undefined. If the
4229 arguments are vectors, each vector element of ``op1`` is shifted by the
4230 corresponding shift amount in ``op2``.
4232 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4233 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4239 .. code-block:: llvm
4241 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4242 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4243 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4244 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4245 <result> = lshr i32 1, 32 ; undefined
4246 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4248 '``ashr``' Instruction
4249 ^^^^^^^^^^^^^^^^^^^^^^
4256 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4257 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4262 The '``ashr``' instruction (arithmetic shift right) returns the first
4263 operand shifted to the right a specified number of bits with sign
4269 Both arguments to the '``ashr``' instruction must be the same
4270 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4271 '``op2``' is treated as an unsigned value.
4276 This instruction always performs an arithmetic shift right operation,
4277 The most significant bits of the result will be filled with the sign bit
4278 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4279 than the number of bits in ``op1``, the result is undefined. If the
4280 arguments are vectors, each vector element of ``op1`` is shifted by the
4281 corresponding shift amount in ``op2``.
4283 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4284 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4290 .. code-block:: llvm
4292 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4293 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4294 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4295 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4296 <result> = ashr i32 1, 32 ; undefined
4297 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4299 '``and``' Instruction
4300 ^^^^^^^^^^^^^^^^^^^^^
4307 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4312 The '``and``' instruction returns the bitwise logical and of its two
4318 The two arguments to the '``and``' instruction must be
4319 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4320 arguments must have identical types.
4325 The truth table used for the '``and``' instruction is:
4342 .. code-block:: llvm
4344 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4345 <result> = and i32 15, 40 ; yields {i32}:result = 8
4346 <result> = and i32 4, 8 ; yields {i32}:result = 0
4348 '``or``' Instruction
4349 ^^^^^^^^^^^^^^^^^^^^
4356 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4361 The '``or``' instruction returns the bitwise logical inclusive or of its
4367 The two arguments to the '``or``' instruction must be
4368 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4369 arguments must have identical types.
4374 The truth table used for the '``or``' instruction is:
4393 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4394 <result> = or i32 15, 40 ; yields {i32}:result = 47
4395 <result> = or i32 4, 8 ; yields {i32}:result = 12
4397 '``xor``' Instruction
4398 ^^^^^^^^^^^^^^^^^^^^^
4405 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4410 The '``xor``' instruction returns the bitwise logical exclusive or of
4411 its two operands. The ``xor`` is used to implement the "one's
4412 complement" operation, which is the "~" operator in C.
4417 The two arguments to the '``xor``' instruction must be
4418 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4419 arguments must have identical types.
4424 The truth table used for the '``xor``' instruction is:
4441 .. code-block:: llvm
4443 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4444 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4445 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4446 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4451 LLVM supports several instructions to represent vector operations in a
4452 target-independent manner. These instructions cover the element-access
4453 and vector-specific operations needed to process vectors effectively.
4454 While LLVM does directly support these vector operations, many
4455 sophisticated algorithms will want to use target-specific intrinsics to
4456 take full advantage of a specific target.
4458 .. _i_extractelement:
4460 '``extractelement``' Instruction
4461 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4468 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4473 The '``extractelement``' instruction extracts a single scalar element
4474 from a vector at a specified index.
4479 The first operand of an '``extractelement``' instruction is a value of
4480 :ref:`vector <t_vector>` type. The second operand is an index indicating
4481 the position from which to extract the element. The index may be a
4487 The result is a scalar of the same type as the element type of ``val``.
4488 Its value is the value at position ``idx`` of ``val``. If ``idx``
4489 exceeds the length of ``val``, the results are undefined.
4494 .. code-block:: llvm
4496 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4498 .. _i_insertelement:
4500 '``insertelement``' Instruction
4501 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4508 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4513 The '``insertelement``' instruction inserts a scalar element into a
4514 vector at a specified index.
4519 The first operand of an '``insertelement``' instruction is a value of
4520 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4521 type must equal the element type of the first operand. The third operand
4522 is an index indicating the position at which to insert the value. The
4523 index may be a variable.
4528 The result is a vector of the same type as ``val``. Its element values
4529 are those of ``val`` except at position ``idx``, where it gets the value
4530 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4536 .. code-block:: llvm
4538 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4540 .. _i_shufflevector:
4542 '``shufflevector``' Instruction
4543 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4550 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4555 The '``shufflevector``' instruction constructs a permutation of elements
4556 from two input vectors, returning a vector with the same element type as
4557 the input and length that is the same as the shuffle mask.
4562 The first two operands of a '``shufflevector``' instruction are vectors
4563 with the same type. The third argument is a shuffle mask whose element
4564 type is always 'i32'. The result of the instruction is a vector whose
4565 length is the same as the shuffle mask and whose element type is the
4566 same as the element type of the first two operands.
4568 The shuffle mask operand is required to be a constant vector with either
4569 constant integer or undef values.
4574 The elements of the two input vectors are numbered from left to right
4575 across both of the vectors. The shuffle mask operand specifies, for each
4576 element of the result vector, which element of the two input vectors the
4577 result element gets. The element selector may be undef (meaning "don't
4578 care") and the second operand may be undef if performing a shuffle from
4584 .. code-block:: llvm
4586 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4587 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4588 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4589 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4590 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4591 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4592 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4593 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4595 Aggregate Operations
4596 --------------------
4598 LLVM supports several instructions for working with
4599 :ref:`aggregate <t_aggregate>` values.
4603 '``extractvalue``' Instruction
4604 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4611 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4616 The '``extractvalue``' instruction extracts the value of a member field
4617 from an :ref:`aggregate <t_aggregate>` value.
4622 The first operand of an '``extractvalue``' instruction is a value of
4623 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4624 constant indices to specify which value to extract in a similar manner
4625 as indices in a '``getelementptr``' instruction.
4627 The major differences to ``getelementptr`` indexing are:
4629 - Since the value being indexed is not a pointer, the first index is
4630 omitted and assumed to be zero.
4631 - At least one index must be specified.
4632 - Not only struct indices but also array indices must be in bounds.
4637 The result is the value at the position in the aggregate specified by
4643 .. code-block:: llvm
4645 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4649 '``insertvalue``' Instruction
4650 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4657 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4662 The '``insertvalue``' instruction inserts a value into a member field in
4663 an :ref:`aggregate <t_aggregate>` value.
4668 The first operand of an '``insertvalue``' instruction is a value of
4669 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4670 a first-class value to insert. The following operands are constant
4671 indices indicating the position at which to insert the value in a
4672 similar manner as indices in a '``extractvalue``' instruction. The value
4673 to insert must have the same type as the value identified by the
4679 The result is an aggregate of the same type as ``val``. Its value is
4680 that of ``val`` except that the value at the position specified by the
4681 indices is that of ``elt``.
4686 .. code-block:: llvm
4688 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4689 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4690 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4694 Memory Access and Addressing Operations
4695 ---------------------------------------
4697 A key design point of an SSA-based representation is how it represents
4698 memory. In LLVM, no memory locations are in SSA form, which makes things
4699 very simple. This section describes how to read, write, and allocate
4704 '``alloca``' Instruction
4705 ^^^^^^^^^^^^^^^^^^^^^^^^
4712 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields {type*}:result
4717 The '``alloca``' instruction allocates memory on the stack frame of the
4718 currently executing function, to be automatically released when this
4719 function returns to its caller. The object is always allocated in the
4720 generic address space (address space zero).
4725 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4726 bytes of memory on the runtime stack, returning a pointer of the
4727 appropriate type to the program. If "NumElements" is specified, it is
4728 the number of elements allocated, otherwise "NumElements" is defaulted
4729 to be one. If a constant alignment is specified, the value result of the
4730 allocation is guaranteed to be aligned to at least that boundary. If not
4731 specified, or if zero, the target can choose to align the allocation on
4732 any convenient boundary compatible with the type.
4734 '``type``' may be any sized type.
4739 Memory is allocated; a pointer is returned. The operation is undefined
4740 if there is insufficient stack space for the allocation. '``alloca``'d
4741 memory is automatically released when the function returns. The
4742 '``alloca``' instruction is commonly used to represent automatic
4743 variables that must have an address available. When the function returns
4744 (either with the ``ret`` or ``resume`` instructions), the memory is
4745 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4746 The order in which memory is allocated (ie., which way the stack grows)
4752 .. code-block:: llvm
4754 %ptr = alloca i32 ; yields {i32*}:ptr
4755 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4756 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4757 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4761 '``load``' Instruction
4762 ^^^^^^^^^^^^^^^^^^^^^^
4769 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4770 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4771 !<index> = !{ i32 1 }
4776 The '``load``' instruction is used to read from memory.
4781 The argument to the ``load`` instruction specifies the memory address
4782 from which to load. The pointer must point to a :ref:`first
4783 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4784 then the optimizer is not allowed to modify the number or order of
4785 execution of this ``load`` with other :ref:`volatile
4786 operations <volatile>`.
4788 If the ``load`` is marked as ``atomic``, it takes an extra
4789 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4790 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4791 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4792 when they may see multiple atomic stores. The type of the pointee must
4793 be an integer type whose bit width is a power of two greater than or
4794 equal to eight and less than or equal to a target-specific size limit.
4795 ``align`` must be explicitly specified on atomic loads, and the load has
4796 undefined behavior if the alignment is not set to a value which is at
4797 least the size in bytes of the pointee. ``!nontemporal`` does not have
4798 any defined semantics for atomic loads.
4800 The optional constant ``align`` argument specifies the alignment of the
4801 operation (that is, the alignment of the memory address). A value of 0
4802 or an omitted ``align`` argument means that the operation has the ABI
4803 alignment for the target. It is the responsibility of the code emitter
4804 to ensure that the alignment information is correct. Overestimating the
4805 alignment results in undefined behavior. Underestimating the alignment
4806 may produce less efficient code. An alignment of 1 is always safe.
4808 The optional ``!nontemporal`` metadata must reference a single
4809 metadata name ``<index>`` corresponding to a metadata node with one
4810 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4811 metadata on the instruction tells the optimizer and code generator
4812 that this load is not expected to be reused in the cache. The code
4813 generator may select special instructions to save cache bandwidth, such
4814 as the ``MOVNT`` instruction on x86.
4816 The optional ``!invariant.load`` metadata must reference a single
4817 metadata name ``<index>`` corresponding to a metadata node with no
4818 entries. The existence of the ``!invariant.load`` metadata on the
4819 instruction tells the optimizer and code generator that this load
4820 address points to memory which does not change value during program
4821 execution. The optimizer may then move this load around, for example, by
4822 hoisting it out of loops using loop invariant code motion.
4827 The location of memory pointed to is loaded. If the value being loaded
4828 is of scalar type then the number of bytes read does not exceed the
4829 minimum number of bytes needed to hold all bits of the type. For
4830 example, loading an ``i24`` reads at most three bytes. When loading a
4831 value of a type like ``i20`` with a size that is not an integral number
4832 of bytes, the result is undefined if the value was not originally
4833 written using a store of the same type.
4838 .. code-block:: llvm
4840 %ptr = alloca i32 ; yields {i32*}:ptr
4841 store i32 3, i32* %ptr ; yields {void}
4842 %val = load i32* %ptr ; yields {i32}:val = i32 3
4846 '``store``' Instruction
4847 ^^^^^^^^^^^^^^^^^^^^^^^
4854 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4855 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4860 The '``store``' instruction is used to write to memory.
4865 There are two arguments to the ``store`` instruction: a value to store
4866 and an address at which to store it. The type of the ``<pointer>``
4867 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4868 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4869 then the optimizer is not allowed to modify the number or order of
4870 execution of this ``store`` with other :ref:`volatile
4871 operations <volatile>`.
4873 If the ``store`` is marked as ``atomic``, it takes an extra
4874 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4875 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4876 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4877 when they may see multiple atomic stores. The type of the pointee must
4878 be an integer type whose bit width is a power of two greater than or
4879 equal to eight and less than or equal to a target-specific size limit.
4880 ``align`` must be explicitly specified on atomic stores, and the store
4881 has undefined behavior if the alignment is not set to a value which is
4882 at least the size in bytes of the pointee. ``!nontemporal`` does not
4883 have any defined semantics for atomic stores.
4885 The optional constant ``align`` argument specifies the alignment of the
4886 operation (that is, the alignment of the memory address). A value of 0
4887 or an omitted ``align`` argument means that the operation has the ABI
4888 alignment for the target. It is the responsibility of the code emitter
4889 to ensure that the alignment information is correct. Overestimating the
4890 alignment results in undefined behavior. Underestimating the
4891 alignment may produce less efficient code. An alignment of 1 is always
4894 The optional ``!nontemporal`` metadata must reference a single metadata
4895 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4896 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4897 tells the optimizer and code generator that this load is not expected to
4898 be reused in the cache. The code generator may select special
4899 instructions to save cache bandwidth, such as the MOVNT instruction on
4905 The contents of memory are updated to contain ``<value>`` at the
4906 location specified by the ``<pointer>`` operand. If ``<value>`` is
4907 of scalar type then the number of bytes written does not exceed the
4908 minimum number of bytes needed to hold all bits of the type. For
4909 example, storing an ``i24`` writes at most three bytes. When writing a
4910 value of a type like ``i20`` with a size that is not an integral number
4911 of bytes, it is unspecified what happens to the extra bits that do not
4912 belong to the type, but they will typically be overwritten.
4917 .. code-block:: llvm
4919 %ptr = alloca i32 ; yields {i32*}:ptr
4920 store i32 3, i32* %ptr ; yields {void}
4921 %val = load i32* %ptr ; yields {i32}:val = i32 3
4925 '``fence``' Instruction
4926 ^^^^^^^^^^^^^^^^^^^^^^^
4933 fence [singlethread] <ordering> ; yields {void}
4938 The '``fence``' instruction is used to introduce happens-before edges
4944 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4945 defines what *synchronizes-with* edges they add. They can only be given
4946 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4951 A fence A which has (at least) ``release`` ordering semantics
4952 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4953 semantics if and only if there exist atomic operations X and Y, both
4954 operating on some atomic object M, such that A is sequenced before X, X
4955 modifies M (either directly or through some side effect of a sequence
4956 headed by X), Y is sequenced before B, and Y observes M. This provides a
4957 *happens-before* dependency between A and B. Rather than an explicit
4958 ``fence``, one (but not both) of the atomic operations X or Y might
4959 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4960 still *synchronize-with* the explicit ``fence`` and establish the
4961 *happens-before* edge.
4963 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4964 ``acquire`` and ``release`` semantics specified above, participates in
4965 the global program order of other ``seq_cst`` operations and/or fences.
4967 The optional ":ref:`singlethread <singlethread>`" argument specifies
4968 that the fence only synchronizes with other fences in the same thread.
4969 (This is useful for interacting with signal handlers.)
4974 .. code-block:: llvm
4976 fence acquire ; yields {void}
4977 fence singlethread seq_cst ; yields {void}
4981 '``cmpxchg``' Instruction
4982 ^^^^^^^^^^^^^^^^^^^^^^^^^
4989 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields {ty}
4994 The '``cmpxchg``' instruction is used to atomically modify memory. It
4995 loads a value in memory and compares it to a given value. If they are
4996 equal, it stores a new value into the memory.
5001 There are three arguments to the '``cmpxchg``' instruction: an address
5002 to operate on, a value to compare to the value currently be at that
5003 address, and a new value to place at that address if the compared values
5004 are equal. The type of '<cmp>' must be an integer type whose bit width
5005 is a power of two greater than or equal to eight and less than or equal
5006 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5007 type, and the type of '<pointer>' must be a pointer to that type. If the
5008 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5009 to modify the number or order of execution of this ``cmpxchg`` with
5010 other :ref:`volatile operations <volatile>`.
5012 The success and failure :ref:`ordering <ordering>` arguments specify how this
5013 ``cmpxchg`` synchronizes with other atomic operations. The both ordering
5014 parameters must be at least ``monotonic``, the ordering constraint on failure
5015 must be no stronger than that on success, and the failure ordering cannot be
5016 either ``release`` or ``acq_rel``.
5018 The optional "``singlethread``" argument declares that the ``cmpxchg``
5019 is only atomic with respect to code (usually signal handlers) running in
5020 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5021 respect to all other code in the system.
5023 The pointer passed into cmpxchg must have alignment greater than or
5024 equal to the size in memory of the operand.
5029 The contents of memory at the location specified by the '``<pointer>``'
5030 operand is read and compared to '``<cmp>``'; if the read value is the
5031 equal, '``<new>``' is written. The original value at the location is
5034 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5035 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5036 load with an ordering parameter determined the second ordering parameter.
5041 .. code-block:: llvm
5044 %orig = atomic load i32* %ptr unordered ; yields {i32}
5048 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5049 %squared = mul i32 %cmp, %cmp
5050 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields {i32}
5051 %success = icmp eq i32 %cmp, %old
5052 br i1 %success, label %done, label %loop
5059 '``atomicrmw``' Instruction
5060 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5067 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
5072 The '``atomicrmw``' instruction is used to atomically modify memory.
5077 There are three arguments to the '``atomicrmw``' instruction: an
5078 operation to apply, an address whose value to modify, an argument to the
5079 operation. The operation must be one of the following keywords:
5093 The type of '<value>' must be an integer type whose bit width is a power
5094 of two greater than or equal to eight and less than or equal to a
5095 target-specific size limit. The type of the '``<pointer>``' operand must
5096 be a pointer to that type. If the ``atomicrmw`` is marked as
5097 ``volatile``, then the optimizer is not allowed to modify the number or
5098 order of execution of this ``atomicrmw`` with other :ref:`volatile
5099 operations <volatile>`.
5104 The contents of memory at the location specified by the '``<pointer>``'
5105 operand are atomically read, modified, and written back. The original
5106 value at the location is returned. The modification is specified by the
5109 - xchg: ``*ptr = val``
5110 - add: ``*ptr = *ptr + val``
5111 - sub: ``*ptr = *ptr - val``
5112 - and: ``*ptr = *ptr & val``
5113 - nand: ``*ptr = ~(*ptr & val)``
5114 - or: ``*ptr = *ptr | val``
5115 - xor: ``*ptr = *ptr ^ val``
5116 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5117 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5118 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5120 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5126 .. code-block:: llvm
5128 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5130 .. _i_getelementptr:
5132 '``getelementptr``' Instruction
5133 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5140 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5141 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5142 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5147 The '``getelementptr``' instruction is used to get the address of a
5148 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5149 address calculation only and does not access memory.
5154 The first argument is always a pointer or a vector of pointers, and
5155 forms the basis of the calculation. The remaining arguments are indices
5156 that indicate which of the elements of the aggregate object are indexed.
5157 The interpretation of each index is dependent on the type being indexed
5158 into. The first index always indexes the pointer value given as the
5159 first argument, the second index indexes a value of the type pointed to
5160 (not necessarily the value directly pointed to, since the first index
5161 can be non-zero), etc. The first type indexed into must be a pointer
5162 value, subsequent types can be arrays, vectors, and structs. Note that
5163 subsequent types being indexed into can never be pointers, since that
5164 would require loading the pointer before continuing calculation.
5166 The type of each index argument depends on the type it is indexing into.
5167 When indexing into a (optionally packed) structure, only ``i32`` integer
5168 **constants** are allowed (when using a vector of indices they must all
5169 be the **same** ``i32`` integer constant). When indexing into an array,
5170 pointer or vector, integers of any width are allowed, and they are not
5171 required to be constant. These integers are treated as signed values
5174 For example, let's consider a C code fragment and how it gets compiled
5190 int *foo(struct ST *s) {
5191 return &s[1].Z.B[5][13];
5194 The LLVM code generated by Clang is:
5196 .. code-block:: llvm
5198 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5199 %struct.ST = type { i32, double, %struct.RT }
5201 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5203 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5210 In the example above, the first index is indexing into the
5211 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5212 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5213 indexes into the third element of the structure, yielding a
5214 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5215 structure. The third index indexes into the second element of the
5216 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5217 dimensions of the array are subscripted into, yielding an '``i32``'
5218 type. The '``getelementptr``' instruction returns a pointer to this
5219 element, thus computing a value of '``i32*``' type.
5221 Note that it is perfectly legal to index partially through a structure,
5222 returning a pointer to an inner element. Because of this, the LLVM code
5223 for the given testcase is equivalent to:
5225 .. code-block:: llvm
5227 define i32* @foo(%struct.ST* %s) {
5228 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5229 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5230 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5231 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5232 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5236 If the ``inbounds`` keyword is present, the result value of the
5237 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5238 pointer is not an *in bounds* address of an allocated object, or if any
5239 of the addresses that would be formed by successive addition of the
5240 offsets implied by the indices to the base address with infinitely
5241 precise signed arithmetic are not an *in bounds* address of that
5242 allocated object. The *in bounds* addresses for an allocated object are
5243 all the addresses that point into the object, plus the address one byte
5244 past the end. In cases where the base is a vector of pointers the
5245 ``inbounds`` keyword applies to each of the computations element-wise.
5247 If the ``inbounds`` keyword is not present, the offsets are added to the
5248 base address with silently-wrapping two's complement arithmetic. If the
5249 offsets have a different width from the pointer, they are sign-extended
5250 or truncated to the width of the pointer. The result value of the
5251 ``getelementptr`` may be outside the object pointed to by the base
5252 pointer. The result value may not necessarily be used to access memory
5253 though, even if it happens to point into allocated storage. See the
5254 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5257 The getelementptr instruction is often confusing. For some more insight
5258 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5263 .. code-block:: llvm
5265 ; yields [12 x i8]*:aptr
5266 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5268 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5270 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5272 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5274 In cases where the pointer argument is a vector of pointers, each index
5275 must be a vector with the same number of elements. For example:
5277 .. code-block:: llvm
5279 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5281 Conversion Operations
5282 ---------------------
5284 The instructions in this category are the conversion instructions
5285 (casting) which all take a single operand and a type. They perform
5286 various bit conversions on the operand.
5288 '``trunc .. to``' Instruction
5289 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5296 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5301 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5306 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5307 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5308 of the same number of integers. The bit size of the ``value`` must be
5309 larger than the bit size of the destination type, ``ty2``. Equal sized
5310 types are not allowed.
5315 The '``trunc``' instruction truncates the high order bits in ``value``
5316 and converts the remaining bits to ``ty2``. Since the source size must
5317 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5318 It will always truncate bits.
5323 .. code-block:: llvm
5325 %X = trunc i32 257 to i8 ; yields i8:1
5326 %Y = trunc i32 123 to i1 ; yields i1:true
5327 %Z = trunc i32 122 to i1 ; yields i1:false
5328 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5330 '``zext .. to``' Instruction
5331 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5338 <result> = zext <ty> <value> to <ty2> ; yields ty2
5343 The '``zext``' instruction zero extends its operand to type ``ty2``.
5348 The '``zext``' instruction takes a value to cast, and a type to cast it
5349 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5350 the same number of integers. The bit size of the ``value`` must be
5351 smaller than the bit size of the destination type, ``ty2``.
5356 The ``zext`` fills the high order bits of the ``value`` with zero bits
5357 until it reaches the size of the destination type, ``ty2``.
5359 When zero extending from i1, the result will always be either 0 or 1.
5364 .. code-block:: llvm
5366 %X = zext i32 257 to i64 ; yields i64:257
5367 %Y = zext i1 true to i32 ; yields i32:1
5368 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5370 '``sext .. to``' Instruction
5371 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5378 <result> = sext <ty> <value> to <ty2> ; yields ty2
5383 The '``sext``' sign extends ``value`` to the type ``ty2``.
5388 The '``sext``' instruction takes a value to cast, and a type to cast it
5389 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5390 the same number of integers. The bit size of the ``value`` must be
5391 smaller than the bit size of the destination type, ``ty2``.
5396 The '``sext``' instruction performs a sign extension by copying the sign
5397 bit (highest order bit) of the ``value`` until it reaches the bit size
5398 of the type ``ty2``.
5400 When sign extending from i1, the extension always results in -1 or 0.
5405 .. code-block:: llvm
5407 %X = sext i8 -1 to i16 ; yields i16 :65535
5408 %Y = sext i1 true to i32 ; yields i32:-1
5409 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5411 '``fptrunc .. to``' Instruction
5412 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5419 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5424 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5429 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5430 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5431 The size of ``value`` must be larger than the size of ``ty2``. This
5432 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5437 The '``fptrunc``' instruction truncates a ``value`` from a larger
5438 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5439 point <t_floating>` type. If the value cannot fit within the
5440 destination type, ``ty2``, then the results are undefined.
5445 .. code-block:: llvm
5447 %X = fptrunc double 123.0 to float ; yields float:123.0
5448 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5450 '``fpext .. to``' Instruction
5451 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5458 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5463 The '``fpext``' extends a floating point ``value`` to a larger floating
5469 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5470 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5471 to. The source type must be smaller than the destination type.
5476 The '``fpext``' instruction extends the ``value`` from a smaller
5477 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5478 point <t_floating>` type. The ``fpext`` cannot be used to make a
5479 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5480 *no-op cast* for a floating point cast.
5485 .. code-block:: llvm
5487 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5488 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5490 '``fptoui .. to``' Instruction
5491 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5498 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5503 The '``fptoui``' converts a floating point ``value`` to its unsigned
5504 integer equivalent of type ``ty2``.
5509 The '``fptoui``' instruction takes a value to cast, which must be a
5510 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5511 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5512 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5513 type with the same number of elements as ``ty``
5518 The '``fptoui``' instruction converts its :ref:`floating
5519 point <t_floating>` operand into the nearest (rounding towards zero)
5520 unsigned integer value. If the value cannot fit in ``ty2``, the results
5526 .. code-block:: llvm
5528 %X = fptoui double 123.0 to i32 ; yields i32:123
5529 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5530 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5532 '``fptosi .. to``' Instruction
5533 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5540 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5545 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5546 ``value`` to type ``ty2``.
5551 The '``fptosi``' instruction takes a value to cast, which must be a
5552 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5553 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5554 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5555 type with the same number of elements as ``ty``
5560 The '``fptosi``' instruction converts its :ref:`floating
5561 point <t_floating>` operand into the nearest (rounding towards zero)
5562 signed integer value. If the value cannot fit in ``ty2``, the results
5568 .. code-block:: llvm
5570 %X = fptosi double -123.0 to i32 ; yields i32:-123
5571 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5572 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5574 '``uitofp .. to``' Instruction
5575 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5582 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5587 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5588 and converts that value to the ``ty2`` type.
5593 The '``uitofp``' instruction takes a value to cast, which must be a
5594 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5595 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5596 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5597 type with the same number of elements as ``ty``
5602 The '``uitofp``' instruction interprets its operand as an unsigned
5603 integer quantity and converts it to the corresponding floating point
5604 value. If the value cannot fit in the floating point value, the results
5610 .. code-block:: llvm
5612 %X = uitofp i32 257 to float ; yields float:257.0
5613 %Y = uitofp i8 -1 to double ; yields double:255.0
5615 '``sitofp .. to``' Instruction
5616 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5623 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5628 The '``sitofp``' instruction regards ``value`` as a signed integer and
5629 converts that value to the ``ty2`` type.
5634 The '``sitofp``' instruction takes a value to cast, which must be a
5635 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5636 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5637 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5638 type with the same number of elements as ``ty``
5643 The '``sitofp``' instruction interprets its operand as a signed integer
5644 quantity and converts it to the corresponding floating point value. If
5645 the value cannot fit in the floating point value, the results are
5651 .. code-block:: llvm
5653 %X = sitofp i32 257 to float ; yields float:257.0
5654 %Y = sitofp i8 -1 to double ; yields double:-1.0
5658 '``ptrtoint .. to``' Instruction
5659 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5666 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5671 The '``ptrtoint``' instruction converts the pointer or a vector of
5672 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5677 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5678 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5679 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5680 a vector of integers type.
5685 The '``ptrtoint``' instruction converts ``value`` to integer type
5686 ``ty2`` by interpreting the pointer value as an integer and either
5687 truncating or zero extending that value to the size of the integer type.
5688 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5689 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5690 the same size, then nothing is done (*no-op cast*) other than a type
5696 .. code-block:: llvm
5698 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5699 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5700 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5704 '``inttoptr .. to``' Instruction
5705 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5712 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5717 The '``inttoptr``' instruction converts an integer ``value`` to a
5718 pointer type, ``ty2``.
5723 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5724 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5730 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5731 applying either a zero extension or a truncation depending on the size
5732 of the integer ``value``. If ``value`` is larger than the size of a
5733 pointer then a truncation is done. If ``value`` is smaller than the size
5734 of a pointer then a zero extension is done. If they are the same size,
5735 nothing is done (*no-op cast*).
5740 .. code-block:: llvm
5742 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5743 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5744 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5745 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5749 '``bitcast .. to``' Instruction
5750 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5757 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5762 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5768 The '``bitcast``' instruction takes a value to cast, which must be a
5769 non-aggregate first class value, and a type to cast it to, which must
5770 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5771 bit sizes of ``value`` and the destination type, ``ty2``, must be
5772 identical. If the source type is a pointer, the destination type must
5773 also be a pointer of the same size. This instruction supports bitwise
5774 conversion of vectors to integers and to vectors of other types (as
5775 long as they have the same size).
5780 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5781 is always a *no-op cast* because no bits change with this
5782 conversion. The conversion is done as if the ``value`` had been stored
5783 to memory and read back as type ``ty2``. Pointer (or vector of
5784 pointers) types may only be converted to other pointer (or vector of
5785 pointers) types with the same address space through this instruction.
5786 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5787 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5792 .. code-block:: llvm
5794 %X = bitcast i8 255 to i8 ; yields i8 :-1
5795 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5796 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5797 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5799 .. _i_addrspacecast:
5801 '``addrspacecast .. to``' Instruction
5802 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5809 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5814 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5815 address space ``n`` to type ``pty2`` in address space ``m``.
5820 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5821 to cast and a pointer type to cast it to, which must have a different
5827 The '``addrspacecast``' instruction converts the pointer value
5828 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5829 value modification, depending on the target and the address space
5830 pair. Pointer conversions within the same address space must be
5831 performed with the ``bitcast`` instruction. Note that if the address space
5832 conversion is legal then both result and operand refer to the same memory
5838 .. code-block:: llvm
5840 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5841 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5842 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5849 The instructions in this category are the "miscellaneous" instructions,
5850 which defy better classification.
5854 '``icmp``' Instruction
5855 ^^^^^^^^^^^^^^^^^^^^^^
5862 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5867 The '``icmp``' instruction returns a boolean value or a vector of
5868 boolean values based on comparison of its two integer, integer vector,
5869 pointer, or pointer vector operands.
5874 The '``icmp``' instruction takes three operands. The first operand is
5875 the condition code indicating the kind of comparison to perform. It is
5876 not a value, just a keyword. The possible condition code are:
5879 #. ``ne``: not equal
5880 #. ``ugt``: unsigned greater than
5881 #. ``uge``: unsigned greater or equal
5882 #. ``ult``: unsigned less than
5883 #. ``ule``: unsigned less or equal
5884 #. ``sgt``: signed greater than
5885 #. ``sge``: signed greater or equal
5886 #. ``slt``: signed less than
5887 #. ``sle``: signed less or equal
5889 The remaining two arguments must be :ref:`integer <t_integer>` or
5890 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5891 must also be identical types.
5896 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5897 code given as ``cond``. The comparison performed always yields either an
5898 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5900 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5901 otherwise. No sign interpretation is necessary or performed.
5902 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5903 otherwise. No sign interpretation is necessary or performed.
5904 #. ``ugt``: interprets the operands as unsigned values and yields
5905 ``true`` if ``op1`` is greater than ``op2``.
5906 #. ``uge``: interprets the operands as unsigned values and yields
5907 ``true`` if ``op1`` is greater than or equal to ``op2``.
5908 #. ``ult``: interprets the operands as unsigned values and yields
5909 ``true`` if ``op1`` is less than ``op2``.
5910 #. ``ule``: interprets the operands as unsigned values and yields
5911 ``true`` if ``op1`` is less than or equal to ``op2``.
5912 #. ``sgt``: interprets the operands as signed values and yields ``true``
5913 if ``op1`` is greater than ``op2``.
5914 #. ``sge``: interprets the operands as signed values and yields ``true``
5915 if ``op1`` is greater than or equal to ``op2``.
5916 #. ``slt``: interprets the operands as signed values and yields ``true``
5917 if ``op1`` is less than ``op2``.
5918 #. ``sle``: interprets the operands as signed values and yields ``true``
5919 if ``op1`` is less than or equal to ``op2``.
5921 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5922 are compared as if they were integers.
5924 If the operands are integer vectors, then they are compared element by
5925 element. The result is an ``i1`` vector with the same number of elements
5926 as the values being compared. Otherwise, the result is an ``i1``.
5931 .. code-block:: llvm
5933 <result> = icmp eq i32 4, 5 ; yields: result=false
5934 <result> = icmp ne float* %X, %X ; yields: result=false
5935 <result> = icmp ult i16 4, 5 ; yields: result=true
5936 <result> = icmp sgt i16 4, 5 ; yields: result=false
5937 <result> = icmp ule i16 -4, 5 ; yields: result=false
5938 <result> = icmp sge i16 4, 5 ; yields: result=false
5940 Note that the code generator does not yet support vector types with the
5941 ``icmp`` instruction.
5945 '``fcmp``' Instruction
5946 ^^^^^^^^^^^^^^^^^^^^^^
5953 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5958 The '``fcmp``' instruction returns a boolean value or vector of boolean
5959 values based on comparison of its operands.
5961 If the operands are floating point scalars, then the result type is a
5962 boolean (:ref:`i1 <t_integer>`).
5964 If the operands are floating point vectors, then the result type is a
5965 vector of boolean with the same number of elements as the operands being
5971 The '``fcmp``' instruction takes three operands. The first operand is
5972 the condition code indicating the kind of comparison to perform. It is
5973 not a value, just a keyword. The possible condition code are:
5975 #. ``false``: no comparison, always returns false
5976 #. ``oeq``: ordered and equal
5977 #. ``ogt``: ordered and greater than
5978 #. ``oge``: ordered and greater than or equal
5979 #. ``olt``: ordered and less than
5980 #. ``ole``: ordered and less than or equal
5981 #. ``one``: ordered and not equal
5982 #. ``ord``: ordered (no nans)
5983 #. ``ueq``: unordered or equal
5984 #. ``ugt``: unordered or greater than
5985 #. ``uge``: unordered or greater than or equal
5986 #. ``ult``: unordered or less than
5987 #. ``ule``: unordered or less than or equal
5988 #. ``une``: unordered or not equal
5989 #. ``uno``: unordered (either nans)
5990 #. ``true``: no comparison, always returns true
5992 *Ordered* means that neither operand is a QNAN while *unordered* means
5993 that either operand may be a QNAN.
5995 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5996 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5997 type. They must have identical types.
6002 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6003 condition code given as ``cond``. If the operands are vectors, then the
6004 vectors are compared element by element. Each comparison performed
6005 always yields an :ref:`i1 <t_integer>` result, as follows:
6007 #. ``false``: always yields ``false``, regardless of operands.
6008 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6009 is equal to ``op2``.
6010 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6011 is greater than ``op2``.
6012 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6013 is greater than or equal to ``op2``.
6014 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6015 is less than ``op2``.
6016 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6017 is less than or equal to ``op2``.
6018 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6019 is not equal to ``op2``.
6020 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6021 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6023 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6024 greater than ``op2``.
6025 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6026 greater than or equal to ``op2``.
6027 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6029 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6030 less than or equal to ``op2``.
6031 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6032 not equal to ``op2``.
6033 #. ``uno``: yields ``true`` if either operand is a QNAN.
6034 #. ``true``: always yields ``true``, regardless of operands.
6039 .. code-block:: llvm
6041 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6042 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6043 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6044 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6046 Note that the code generator does not yet support vector types with the
6047 ``fcmp`` instruction.
6051 '``phi``' Instruction
6052 ^^^^^^^^^^^^^^^^^^^^^
6059 <result> = phi <ty> [ <val0>, <label0>], ...
6064 The '``phi``' instruction is used to implement the φ node in the SSA
6065 graph representing the function.
6070 The type of the incoming values is specified with the first type field.
6071 After this, the '``phi``' instruction takes a list of pairs as
6072 arguments, with one pair for each predecessor basic block of the current
6073 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6074 the value arguments to the PHI node. Only labels may be used as the
6077 There must be no non-phi instructions between the start of a basic block
6078 and the PHI instructions: i.e. PHI instructions must be first in a basic
6081 For the purposes of the SSA form, the use of each incoming value is
6082 deemed to occur on the edge from the corresponding predecessor block to
6083 the current block (but after any definition of an '``invoke``'
6084 instruction's return value on the same edge).
6089 At runtime, the '``phi``' instruction logically takes on the value
6090 specified by the pair corresponding to the predecessor basic block that
6091 executed just prior to the current block.
6096 .. code-block:: llvm
6098 Loop: ; Infinite loop that counts from 0 on up...
6099 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6100 %nextindvar = add i32 %indvar, 1
6105 '``select``' Instruction
6106 ^^^^^^^^^^^^^^^^^^^^^^^^
6113 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6115 selty is either i1 or {<N x i1>}
6120 The '``select``' instruction is used to choose one value based on a
6121 condition, without IR-level branching.
6126 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6127 values indicating the condition, and two values of the same :ref:`first
6128 class <t_firstclass>` type. If the val1/val2 are vectors and the
6129 condition is a scalar, then entire vectors are selected, not individual
6135 If the condition is an i1 and it evaluates to 1, the instruction returns
6136 the first value argument; otherwise, it returns the second value
6139 If the condition is a vector of i1, then the value arguments must be
6140 vectors of the same size, and the selection is done element by element.
6145 .. code-block:: llvm
6147 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6151 '``call``' Instruction
6152 ^^^^^^^^^^^^^^^^^^^^^^
6159 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6164 The '``call``' instruction represents a simple function call.
6169 This instruction requires several arguments:
6171 #. The optional "tail" marker indicates that the callee function does
6172 not access any allocas or varargs in the caller. Note that calls may
6173 be marked "tail" even if they do not occur before a
6174 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6175 function call is eligible for tail call optimization, but `might not
6176 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6177 The code generator may optimize calls marked "tail" with either 1)
6178 automatic `sibling call
6179 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6180 callee have matching signatures, or 2) forced tail call optimization
6181 when the following extra requirements are met:
6183 - Caller and callee both have the calling convention ``fastcc``.
6184 - The call is in tail position (ret immediately follows call and ret
6185 uses value of call or is void).
6186 - Option ``-tailcallopt`` is enabled, or
6187 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6188 - `Platform specific constraints are
6189 met. <CodeGenerator.html#tailcallopt>`_
6191 #. The optional "cconv" marker indicates which :ref:`calling
6192 convention <callingconv>` the call should use. If none is
6193 specified, the call defaults to using C calling conventions. The
6194 calling convention of the call must match the calling convention of
6195 the target function, or else the behavior is undefined.
6196 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6197 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6199 #. '``ty``': the type of the call instruction itself which is also the
6200 type of the return value. Functions that return no value are marked
6202 #. '``fnty``': shall be the signature of the pointer to function value
6203 being invoked. The argument types must match the types implied by
6204 this signature. This type can be omitted if the function is not
6205 varargs and if the function type does not return a pointer to a
6207 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6208 be invoked. In most cases, this is a direct function invocation, but
6209 indirect ``call``'s are just as possible, calling an arbitrary pointer
6211 #. '``function args``': argument list whose types match the function
6212 signature argument types and parameter attributes. All arguments must
6213 be of :ref:`first class <t_firstclass>` type. If the function signature
6214 indicates the function accepts a variable number of arguments, the
6215 extra arguments can be specified.
6216 #. The optional :ref:`function attributes <fnattrs>` list. Only
6217 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6218 attributes are valid here.
6223 The '``call``' instruction is used to cause control flow to transfer to
6224 a specified function, with its incoming arguments bound to the specified
6225 values. Upon a '``ret``' instruction in the called function, control
6226 flow continues with the instruction after the function call, and the
6227 return value of the function is bound to the result argument.
6232 .. code-block:: llvm
6234 %retval = call i32 @test(i32 %argc)
6235 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6236 %X = tail call i32 @foo() ; yields i32
6237 %Y = tail call fastcc i32 @foo() ; yields i32
6238 call void %foo(i8 97 signext)
6240 %struct.A = type { i32, i8 }
6241 %r = call %struct.A @foo() ; yields { 32, i8 }
6242 %gr = extractvalue %struct.A %r, 0 ; yields i32
6243 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6244 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6245 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6247 llvm treats calls to some functions with names and arguments that match
6248 the standard C99 library as being the C99 library functions, and may
6249 perform optimizations or generate code for them under that assumption.
6250 This is something we'd like to change in the future to provide better
6251 support for freestanding environments and non-C-based languages.
6255 '``va_arg``' Instruction
6256 ^^^^^^^^^^^^^^^^^^^^^^^^
6263 <resultval> = va_arg <va_list*> <arglist>, <argty>
6268 The '``va_arg``' instruction is used to access arguments passed through
6269 the "variable argument" area of a function call. It is used to implement
6270 the ``va_arg`` macro in C.
6275 This instruction takes a ``va_list*`` value and the type of the
6276 argument. It returns a value of the specified argument type and
6277 increments the ``va_list`` to point to the next argument. The actual
6278 type of ``va_list`` is target specific.
6283 The '``va_arg``' instruction loads an argument of the specified type
6284 from the specified ``va_list`` and causes the ``va_list`` to point to
6285 the next argument. For more information, see the variable argument
6286 handling :ref:`Intrinsic Functions <int_varargs>`.
6288 It is legal for this instruction to be called in a function which does
6289 not take a variable number of arguments, for example, the ``vfprintf``
6292 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6293 function <intrinsics>` because it takes a type as an argument.
6298 See the :ref:`variable argument processing <int_varargs>` section.
6300 Note that the code generator does not yet fully support va\_arg on many
6301 targets. Also, it does not currently support va\_arg with aggregate
6302 types on any target.
6306 '``landingpad``' Instruction
6307 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6314 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6315 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6317 <clause> := catch <type> <value>
6318 <clause> := filter <array constant type> <array constant>
6323 The '``landingpad``' instruction is used by `LLVM's exception handling
6324 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6325 is a landing pad --- one where the exception lands, and corresponds to the
6326 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6327 defines values supplied by the personality function (``pers_fn``) upon
6328 re-entry to the function. The ``resultval`` has the type ``resultty``.
6333 This instruction takes a ``pers_fn`` value. This is the personality
6334 function associated with the unwinding mechanism. The optional
6335 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6337 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6338 contains the global variable representing the "type" that may be caught
6339 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6340 clause takes an array constant as its argument. Use
6341 "``[0 x i8**] undef``" for a filter which cannot throw. The
6342 '``landingpad``' instruction must contain *at least* one ``clause`` or
6343 the ``cleanup`` flag.
6348 The '``landingpad``' instruction defines the values which are set by the
6349 personality function (``pers_fn``) upon re-entry to the function, and
6350 therefore the "result type" of the ``landingpad`` instruction. As with
6351 calling conventions, how the personality function results are
6352 represented in LLVM IR is target specific.
6354 The clauses are applied in order from top to bottom. If two
6355 ``landingpad`` instructions are merged together through inlining, the
6356 clauses from the calling function are appended to the list of clauses.
6357 When the call stack is being unwound due to an exception being thrown,
6358 the exception is compared against each ``clause`` in turn. If it doesn't
6359 match any of the clauses, and the ``cleanup`` flag is not set, then
6360 unwinding continues further up the call stack.
6362 The ``landingpad`` instruction has several restrictions:
6364 - A landing pad block is a basic block which is the unwind destination
6365 of an '``invoke``' instruction.
6366 - A landing pad block must have a '``landingpad``' instruction as its
6367 first non-PHI instruction.
6368 - There can be only one '``landingpad``' instruction within the landing
6370 - A basic block that is not a landing pad block may not include a
6371 '``landingpad``' instruction.
6372 - All '``landingpad``' instructions in a function must have the same
6373 personality function.
6378 .. code-block:: llvm
6380 ;; A landing pad which can catch an integer.
6381 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6383 ;; A landing pad that is a cleanup.
6384 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6386 ;; A landing pad which can catch an integer and can only throw a double.
6387 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6389 filter [1 x i8**] [@_ZTId]
6396 LLVM supports the notion of an "intrinsic function". These functions
6397 have well known names and semantics and are required to follow certain
6398 restrictions. Overall, these intrinsics represent an extension mechanism
6399 for the LLVM language that does not require changing all of the
6400 transformations in LLVM when adding to the language (or the bitcode
6401 reader/writer, the parser, etc...).
6403 Intrinsic function names must all start with an "``llvm.``" prefix. This
6404 prefix is reserved in LLVM for intrinsic names; thus, function names may
6405 not begin with this prefix. Intrinsic functions must always be external
6406 functions: you cannot define the body of intrinsic functions. Intrinsic
6407 functions may only be used in call or invoke instructions: it is illegal
6408 to take the address of an intrinsic function. Additionally, because
6409 intrinsic functions are part of the LLVM language, it is required if any
6410 are added that they be documented here.
6412 Some intrinsic functions can be overloaded, i.e., the intrinsic
6413 represents a family of functions that perform the same operation but on
6414 different data types. Because LLVM can represent over 8 million
6415 different integer types, overloading is used commonly to allow an
6416 intrinsic function to operate on any integer type. One or more of the
6417 argument types or the result type can be overloaded to accept any
6418 integer type. Argument types may also be defined as exactly matching a
6419 previous argument's type or the result type. This allows an intrinsic
6420 function which accepts multiple arguments, but needs all of them to be
6421 of the same type, to only be overloaded with respect to a single
6422 argument or the result.
6424 Overloaded intrinsics will have the names of its overloaded argument
6425 types encoded into its function name, each preceded by a period. Only
6426 those types which are overloaded result in a name suffix. Arguments
6427 whose type is matched against another type do not. For example, the
6428 ``llvm.ctpop`` function can take an integer of any width and returns an
6429 integer of exactly the same integer width. This leads to a family of
6430 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6431 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6432 overloaded, and only one type suffix is required. Because the argument's
6433 type is matched against the return type, it does not require its own
6436 To learn how to add an intrinsic function, please see the `Extending
6437 LLVM Guide <ExtendingLLVM.html>`_.
6441 Variable Argument Handling Intrinsics
6442 -------------------------------------
6444 Variable argument support is defined in LLVM with the
6445 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6446 functions. These functions are related to the similarly named macros
6447 defined in the ``<stdarg.h>`` header file.
6449 All of these functions operate on arguments that use a target-specific
6450 value type "``va_list``". The LLVM assembly language reference manual
6451 does not define what this type is, so all transformations should be
6452 prepared to handle these functions regardless of the type used.
6454 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6455 variable argument handling intrinsic functions are used.
6457 .. code-block:: llvm
6459 define i32 @test(i32 %X, ...) {
6460 ; Initialize variable argument processing
6462 %ap2 = bitcast i8** %ap to i8*
6463 call void @llvm.va_start(i8* %ap2)
6465 ; Read a single integer argument
6466 %tmp = va_arg i8** %ap, i32
6468 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6470 %aq2 = bitcast i8** %aq to i8*
6471 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6472 call void @llvm.va_end(i8* %aq2)
6474 ; Stop processing of arguments.
6475 call void @llvm.va_end(i8* %ap2)
6479 declare void @llvm.va_start(i8*)
6480 declare void @llvm.va_copy(i8*, i8*)
6481 declare void @llvm.va_end(i8*)
6485 '``llvm.va_start``' Intrinsic
6486 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6493 declare void @llvm.va_start(i8* <arglist>)
6498 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6499 subsequent use by ``va_arg``.
6504 The argument is a pointer to a ``va_list`` element to initialize.
6509 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6510 available in C. In a target-dependent way, it initializes the
6511 ``va_list`` element to which the argument points, so that the next call
6512 to ``va_arg`` will produce the first variable argument passed to the
6513 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6514 to know the last argument of the function as the compiler can figure
6517 '``llvm.va_end``' Intrinsic
6518 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6525 declare void @llvm.va_end(i8* <arglist>)
6530 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6531 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6536 The argument is a pointer to a ``va_list`` to destroy.
6541 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6542 available in C. In a target-dependent way, it destroys the ``va_list``
6543 element to which the argument points. Calls to
6544 :ref:`llvm.va_start <int_va_start>` and
6545 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6550 '``llvm.va_copy``' Intrinsic
6551 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6558 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6563 The '``llvm.va_copy``' intrinsic copies the current argument position
6564 from the source argument list to the destination argument list.
6569 The first argument is a pointer to a ``va_list`` element to initialize.
6570 The second argument is a pointer to a ``va_list`` element to copy from.
6575 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6576 available in C. In a target-dependent way, it copies the source
6577 ``va_list`` element into the destination ``va_list`` element. This
6578 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6579 arbitrarily complex and require, for example, memory allocation.
6581 Accurate Garbage Collection Intrinsics
6582 --------------------------------------
6584 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6585 (GC) requires the implementation and generation of these intrinsics.
6586 These intrinsics allow identification of :ref:`GC roots on the
6587 stack <int_gcroot>`, as well as garbage collector implementations that
6588 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6589 Front-ends for type-safe garbage collected languages should generate
6590 these intrinsics to make use of the LLVM garbage collectors. For more
6591 details, see `Accurate Garbage Collection with
6592 LLVM <GarbageCollection.html>`_.
6594 The garbage collection intrinsics only operate on objects in the generic
6595 address space (address space zero).
6599 '``llvm.gcroot``' Intrinsic
6600 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6607 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6612 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6613 the code generator, and allows some metadata to be associated with it.
6618 The first argument specifies the address of a stack object that contains
6619 the root pointer. The second pointer (which must be either a constant or
6620 a global value address) contains the meta-data to be associated with the
6626 At runtime, a call to this intrinsic stores a null pointer into the
6627 "ptrloc" location. At compile-time, the code generator generates
6628 information to allow the runtime to find the pointer at GC safe points.
6629 The '``llvm.gcroot``' intrinsic may only be used in a function which
6630 :ref:`specifies a GC algorithm <gc>`.
6634 '``llvm.gcread``' Intrinsic
6635 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6642 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6647 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6648 locations, allowing garbage collector implementations that require read
6654 The second argument is the address to read from, which should be an
6655 address allocated from the garbage collector. The first object is a
6656 pointer to the start of the referenced object, if needed by the language
6657 runtime (otherwise null).
6662 The '``llvm.gcread``' intrinsic has the same semantics as a load
6663 instruction, but may be replaced with substantially more complex code by
6664 the garbage collector runtime, as needed. The '``llvm.gcread``'
6665 intrinsic may only be used in a function which :ref:`specifies a GC
6670 '``llvm.gcwrite``' Intrinsic
6671 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6678 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6683 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6684 locations, allowing garbage collector implementations that require write
6685 barriers (such as generational or reference counting collectors).
6690 The first argument is the reference to store, the second is the start of
6691 the object to store it to, and the third is the address of the field of
6692 Obj to store to. If the runtime does not require a pointer to the
6693 object, Obj may be null.
6698 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6699 instruction, but may be replaced with substantially more complex code by
6700 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6701 intrinsic may only be used in a function which :ref:`specifies a GC
6704 Code Generator Intrinsics
6705 -------------------------
6707 These intrinsics are provided by LLVM to expose special features that
6708 may only be implemented with code generator support.
6710 '``llvm.returnaddress``' Intrinsic
6711 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6718 declare i8 *@llvm.returnaddress(i32 <level>)
6723 The '``llvm.returnaddress``' intrinsic attempts to compute a
6724 target-specific value indicating the return address of the current
6725 function or one of its callers.
6730 The argument to this intrinsic indicates which function to return the
6731 address for. Zero indicates the calling function, one indicates its
6732 caller, etc. The argument is **required** to be a constant integer
6738 The '``llvm.returnaddress``' intrinsic either returns a pointer
6739 indicating the return address of the specified call frame, or zero if it
6740 cannot be identified. The value returned by this intrinsic is likely to
6741 be incorrect or 0 for arguments other than zero, so it should only be
6742 used for debugging purposes.
6744 Note that calling this intrinsic does not prevent function inlining or
6745 other aggressive transformations, so the value returned may not be that
6746 of the obvious source-language caller.
6748 '``llvm.frameaddress``' Intrinsic
6749 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6756 declare i8* @llvm.frameaddress(i32 <level>)
6761 The '``llvm.frameaddress``' intrinsic attempts to return the
6762 target-specific frame pointer value for the specified stack frame.
6767 The argument to this intrinsic indicates which function to return the
6768 frame pointer for. Zero indicates the calling function, one indicates
6769 its caller, etc. The argument is **required** to be a constant integer
6775 The '``llvm.frameaddress``' intrinsic either returns a pointer
6776 indicating the frame address of the specified call frame, or zero if it
6777 cannot be identified. The value returned by this intrinsic is likely to
6778 be incorrect or 0 for arguments other than zero, so it should only be
6779 used for debugging purposes.
6781 Note that calling this intrinsic does not prevent function inlining or
6782 other aggressive transformations, so the value returned may not be that
6783 of the obvious source-language caller.
6787 '``llvm.stacksave``' Intrinsic
6788 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6795 declare i8* @llvm.stacksave()
6800 The '``llvm.stacksave``' intrinsic is used to remember the current state
6801 of the function stack, for use with
6802 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6803 implementing language features like scoped automatic variable sized
6809 This intrinsic returns a opaque pointer value that can be passed to
6810 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6811 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6812 ``llvm.stacksave``, it effectively restores the state of the stack to
6813 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6814 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6815 were allocated after the ``llvm.stacksave`` was executed.
6817 .. _int_stackrestore:
6819 '``llvm.stackrestore``' Intrinsic
6820 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6827 declare void @llvm.stackrestore(i8* %ptr)
6832 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6833 the function stack to the state it was in when the corresponding
6834 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6835 useful for implementing language features like scoped automatic variable
6836 sized arrays in C99.
6841 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6843 '``llvm.prefetch``' Intrinsic
6844 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6851 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6856 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6857 insert a prefetch instruction if supported; otherwise, it is a noop.
6858 Prefetches have no effect on the behavior of the program but can change
6859 its performance characteristics.
6864 ``address`` is the address to be prefetched, ``rw`` is the specifier
6865 determining if the fetch should be for a read (0) or write (1), and
6866 ``locality`` is a temporal locality specifier ranging from (0) - no
6867 locality, to (3) - extremely local keep in cache. The ``cache type``
6868 specifies whether the prefetch is performed on the data (1) or
6869 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6870 arguments must be constant integers.
6875 This intrinsic does not modify the behavior of the program. In
6876 particular, prefetches cannot trap and do not produce a value. On
6877 targets that support this intrinsic, the prefetch can provide hints to
6878 the processor cache for better performance.
6880 '``llvm.pcmarker``' Intrinsic
6881 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6888 declare void @llvm.pcmarker(i32 <id>)
6893 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6894 Counter (PC) in a region of code to simulators and other tools. The
6895 method is target specific, but it is expected that the marker will use
6896 exported symbols to transmit the PC of the marker. The marker makes no
6897 guarantees that it will remain with any specific instruction after
6898 optimizations. It is possible that the presence of a marker will inhibit
6899 optimizations. The intended use is to be inserted after optimizations to
6900 allow correlations of simulation runs.
6905 ``id`` is a numerical id identifying the marker.
6910 This intrinsic does not modify the behavior of the program. Backends
6911 that do not support this intrinsic may ignore it.
6913 '``llvm.readcyclecounter``' Intrinsic
6914 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6921 declare i64 @llvm.readcyclecounter()
6926 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6927 counter register (or similar low latency, high accuracy clocks) on those
6928 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6929 should map to RPCC. As the backing counters overflow quickly (on the
6930 order of 9 seconds on alpha), this should only be used for small
6936 When directly supported, reading the cycle counter should not modify any
6937 memory. Implementations are allowed to either return a application
6938 specific value or a system wide value. On backends without support, this
6939 is lowered to a constant 0.
6941 Note that runtime support may be conditional on the privilege-level code is
6942 running at and the host platform.
6944 '``llvm.clear_cache``' Intrinsic
6945 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6952 declare void @llvm.clear_cache(i8*, i8*)
6957 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
6958 in the specified range to the execution unit of the processor. On
6959 targets with non-unified instruction and data cache, the implementation
6960 flushes the instruction cache.
6965 On platforms with coherent instruction and data caches (e.g. x86), this
6966 intrinsic is a nop. On platforms with non-coherent instruction and data
6967 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropiate
6968 instructions or a system call, if cache flushing requires special
6971 The default behavior is to emit a call to ``__clear_cache'' from the run
6974 This instrinsic does *not* empty the instruction pipeline. Modifications
6975 of the current function are outside the scope of the intrinsic.
6977 Standard C Library Intrinsics
6978 -----------------------------
6980 LLVM provides intrinsics for a few important standard C library
6981 functions. These intrinsics allow source-language front-ends to pass
6982 information about the alignment of the pointer arguments to the code
6983 generator, providing opportunity for more efficient code generation.
6987 '``llvm.memcpy``' Intrinsic
6988 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6993 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6994 integer bit width and for different address spaces. Not all targets
6995 support all bit widths however.
6999 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7000 i32 <len>, i32 <align>, i1 <isvolatile>)
7001 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7002 i64 <len>, i32 <align>, i1 <isvolatile>)
7007 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7008 source location to the destination location.
7010 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
7011 intrinsics do not return a value, takes extra alignment/isvolatile
7012 arguments and the pointers can be in specified address spaces.
7017 The first argument is a pointer to the destination, the second is a
7018 pointer to the source. The third argument is an integer argument
7019 specifying the number of bytes to copy, the fourth argument is the
7020 alignment of the source and destination locations, and the fifth is a
7021 boolean indicating a volatile access.
7023 If the call to this intrinsic has an alignment value that is not 0 or 1,
7024 then the caller guarantees that both the source and destination pointers
7025 are aligned to that boundary.
7027 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
7028 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7029 very cleanly specified and it is unwise to depend on it.
7034 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7035 source location to the destination location, which are not allowed to
7036 overlap. It copies "len" bytes of memory over. If the argument is known
7037 to be aligned to some boundary, this can be specified as the fourth
7038 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
7040 '``llvm.memmove``' Intrinsic
7041 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7046 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7047 bit width and for different address space. Not all targets support all
7052 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7053 i32 <len>, i32 <align>, i1 <isvolatile>)
7054 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7055 i64 <len>, i32 <align>, i1 <isvolatile>)
7060 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7061 source location to the destination location. It is similar to the
7062 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7065 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7066 intrinsics do not return a value, takes extra alignment/isvolatile
7067 arguments and the pointers can be in specified address spaces.
7072 The first argument is a pointer to the destination, the second is a
7073 pointer to the source. The third argument is an integer argument
7074 specifying the number of bytes to copy, the fourth argument is the
7075 alignment of the source and destination locations, and the fifth is a
7076 boolean indicating a volatile access.
7078 If the call to this intrinsic has an alignment value that is not 0 or 1,
7079 then the caller guarantees that the source and destination pointers are
7080 aligned to that boundary.
7082 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7083 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7084 not very cleanly specified and it is unwise to depend on it.
7089 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7090 source location to the destination location, which may overlap. It
7091 copies "len" bytes of memory over. If the argument is known to be
7092 aligned to some boundary, this can be specified as the fourth argument,
7093 otherwise it should be set to 0 or 1 (both meaning no alignment).
7095 '``llvm.memset.*``' Intrinsics
7096 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7101 This is an overloaded intrinsic. You can use llvm.memset on any integer
7102 bit width and for different address spaces. However, not all targets
7103 support all bit widths.
7107 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7108 i32 <len>, i32 <align>, i1 <isvolatile>)
7109 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7110 i64 <len>, i32 <align>, i1 <isvolatile>)
7115 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7116 particular byte value.
7118 Note that, unlike the standard libc function, the ``llvm.memset``
7119 intrinsic does not return a value and takes extra alignment/volatile
7120 arguments. Also, the destination can be in an arbitrary address space.
7125 The first argument is a pointer to the destination to fill, the second
7126 is the byte value with which to fill it, the third argument is an
7127 integer argument specifying the number of bytes to fill, and the fourth
7128 argument is the known alignment of the destination location.
7130 If the call to this intrinsic has an alignment value that is not 0 or 1,
7131 then the caller guarantees that the destination pointer is aligned to
7134 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7135 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7136 very cleanly specified and it is unwise to depend on it.
7141 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7142 at the destination location. If the argument is known to be aligned to
7143 some boundary, this can be specified as the fourth argument, otherwise
7144 it should be set to 0 or 1 (both meaning no alignment).
7146 '``llvm.sqrt.*``' Intrinsic
7147 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7152 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7153 floating point or vector of floating point type. Not all targets support
7158 declare float @llvm.sqrt.f32(float %Val)
7159 declare double @llvm.sqrt.f64(double %Val)
7160 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7161 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7162 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7167 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7168 returning the same value as the libm '``sqrt``' functions would. Unlike
7169 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7170 negative numbers other than -0.0 (which allows for better optimization,
7171 because there is no need to worry about errno being set).
7172 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7177 The argument and return value are floating point numbers of the same
7183 This function returns the sqrt of the specified operand if it is a
7184 nonnegative floating point number.
7186 '``llvm.powi.*``' Intrinsic
7187 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7192 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7193 floating point or vector of floating point type. Not all targets support
7198 declare float @llvm.powi.f32(float %Val, i32 %power)
7199 declare double @llvm.powi.f64(double %Val, i32 %power)
7200 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7201 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7202 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7207 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7208 specified (positive or negative) power. The order of evaluation of
7209 multiplications is not defined. When a vector of floating point type is
7210 used, the second argument remains a scalar integer value.
7215 The second argument is an integer power, and the first is a value to
7216 raise to that power.
7221 This function returns the first value raised to the second power with an
7222 unspecified sequence of rounding operations.
7224 '``llvm.sin.*``' Intrinsic
7225 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7230 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7231 floating point or vector of floating point type. Not all targets support
7236 declare float @llvm.sin.f32(float %Val)
7237 declare double @llvm.sin.f64(double %Val)
7238 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7239 declare fp128 @llvm.sin.f128(fp128 %Val)
7240 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7245 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7250 The argument and return value are floating point numbers of the same
7256 This function returns the sine of the specified operand, returning the
7257 same values as the libm ``sin`` functions would, and handles error
7258 conditions in the same way.
7260 '``llvm.cos.*``' Intrinsic
7261 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7266 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7267 floating point or vector of floating point type. Not all targets support
7272 declare float @llvm.cos.f32(float %Val)
7273 declare double @llvm.cos.f64(double %Val)
7274 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7275 declare fp128 @llvm.cos.f128(fp128 %Val)
7276 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7281 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7286 The argument and return value are floating point numbers of the same
7292 This function returns the cosine of the specified operand, returning the
7293 same values as the libm ``cos`` functions would, and handles error
7294 conditions in the same way.
7296 '``llvm.pow.*``' Intrinsic
7297 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7302 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7303 floating point or vector of floating point type. Not all targets support
7308 declare float @llvm.pow.f32(float %Val, float %Power)
7309 declare double @llvm.pow.f64(double %Val, double %Power)
7310 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7311 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7312 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7317 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7318 specified (positive or negative) power.
7323 The second argument is a floating point power, and the first is a value
7324 to raise to that power.
7329 This function returns the first value raised to the second power,
7330 returning the same values as the libm ``pow`` functions would, and
7331 handles error conditions in the same way.
7333 '``llvm.exp.*``' Intrinsic
7334 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7339 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7340 floating point or vector of floating point type. Not all targets support
7345 declare float @llvm.exp.f32(float %Val)
7346 declare double @llvm.exp.f64(double %Val)
7347 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7348 declare fp128 @llvm.exp.f128(fp128 %Val)
7349 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7354 The '``llvm.exp.*``' intrinsics perform the exp function.
7359 The argument and return value are floating point numbers of the same
7365 This function returns the same values as the libm ``exp`` functions
7366 would, and handles error conditions in the same way.
7368 '``llvm.exp2.*``' Intrinsic
7369 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7374 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7375 floating point or vector of floating point type. Not all targets support
7380 declare float @llvm.exp2.f32(float %Val)
7381 declare double @llvm.exp2.f64(double %Val)
7382 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7383 declare fp128 @llvm.exp2.f128(fp128 %Val)
7384 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7389 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7394 The argument and return value are floating point numbers of the same
7400 This function returns the same values as the libm ``exp2`` functions
7401 would, and handles error conditions in the same way.
7403 '``llvm.log.*``' Intrinsic
7404 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7409 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7410 floating point or vector of floating point type. Not all targets support
7415 declare float @llvm.log.f32(float %Val)
7416 declare double @llvm.log.f64(double %Val)
7417 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7418 declare fp128 @llvm.log.f128(fp128 %Val)
7419 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7424 The '``llvm.log.*``' intrinsics perform the log function.
7429 The argument and return value are floating point numbers of the same
7435 This function returns the same values as the libm ``log`` functions
7436 would, and handles error conditions in the same way.
7438 '``llvm.log10.*``' Intrinsic
7439 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7444 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7445 floating point or vector of floating point type. Not all targets support
7450 declare float @llvm.log10.f32(float %Val)
7451 declare double @llvm.log10.f64(double %Val)
7452 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7453 declare fp128 @llvm.log10.f128(fp128 %Val)
7454 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7459 The '``llvm.log10.*``' intrinsics perform the log10 function.
7464 The argument and return value are floating point numbers of the same
7470 This function returns the same values as the libm ``log10`` functions
7471 would, and handles error conditions in the same way.
7473 '``llvm.log2.*``' Intrinsic
7474 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7479 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7480 floating point or vector of floating point type. Not all targets support
7485 declare float @llvm.log2.f32(float %Val)
7486 declare double @llvm.log2.f64(double %Val)
7487 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7488 declare fp128 @llvm.log2.f128(fp128 %Val)
7489 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7494 The '``llvm.log2.*``' intrinsics perform the log2 function.
7499 The argument and return value are floating point numbers of the same
7505 This function returns the same values as the libm ``log2`` functions
7506 would, and handles error conditions in the same way.
7508 '``llvm.fma.*``' Intrinsic
7509 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7514 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7515 floating point or vector of floating point type. Not all targets support
7520 declare float @llvm.fma.f32(float %a, float %b, float %c)
7521 declare double @llvm.fma.f64(double %a, double %b, double %c)
7522 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7523 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7524 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7529 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7535 The argument and return value are floating point numbers of the same
7541 This function returns the same values as the libm ``fma`` functions
7542 would, and does not set errno.
7544 '``llvm.fabs.*``' Intrinsic
7545 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7550 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7551 floating point or vector of floating point type. Not all targets support
7556 declare float @llvm.fabs.f32(float %Val)
7557 declare double @llvm.fabs.f64(double %Val)
7558 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7559 declare fp128 @llvm.fabs.f128(fp128 %Val)
7560 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7565 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7571 The argument and return value are floating point numbers of the same
7577 This function returns the same values as the libm ``fabs`` functions
7578 would, and handles error conditions in the same way.
7580 '``llvm.copysign.*``' Intrinsic
7581 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7586 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7587 floating point or vector of floating point type. Not all targets support
7592 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7593 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7594 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7595 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7596 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7601 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7602 first operand and the sign of the second operand.
7607 The arguments and return value are floating point numbers of the same
7613 This function returns the same values as the libm ``copysign``
7614 functions would, and handles error conditions in the same way.
7616 '``llvm.floor.*``' Intrinsic
7617 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7622 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7623 floating point or vector of floating point type. Not all targets support
7628 declare float @llvm.floor.f32(float %Val)
7629 declare double @llvm.floor.f64(double %Val)
7630 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7631 declare fp128 @llvm.floor.f128(fp128 %Val)
7632 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7637 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7642 The argument and return value are floating point numbers of the same
7648 This function returns the same values as the libm ``floor`` functions
7649 would, and handles error conditions in the same way.
7651 '``llvm.ceil.*``' Intrinsic
7652 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7657 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7658 floating point or vector of floating point type. Not all targets support
7663 declare float @llvm.ceil.f32(float %Val)
7664 declare double @llvm.ceil.f64(double %Val)
7665 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7666 declare fp128 @llvm.ceil.f128(fp128 %Val)
7667 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7672 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7677 The argument and return value are floating point numbers of the same
7683 This function returns the same values as the libm ``ceil`` functions
7684 would, and handles error conditions in the same way.
7686 '``llvm.trunc.*``' Intrinsic
7687 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7692 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7693 floating point or vector of floating point type. Not all targets support
7698 declare float @llvm.trunc.f32(float %Val)
7699 declare double @llvm.trunc.f64(double %Val)
7700 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7701 declare fp128 @llvm.trunc.f128(fp128 %Val)
7702 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7707 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7708 nearest integer not larger in magnitude than the operand.
7713 The argument and return value are floating point numbers of the same
7719 This function returns the same values as the libm ``trunc`` functions
7720 would, and handles error conditions in the same way.
7722 '``llvm.rint.*``' Intrinsic
7723 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7728 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7729 floating point or vector of floating point type. Not all targets support
7734 declare float @llvm.rint.f32(float %Val)
7735 declare double @llvm.rint.f64(double %Val)
7736 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7737 declare fp128 @llvm.rint.f128(fp128 %Val)
7738 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7743 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7744 nearest integer. It may raise an inexact floating-point exception if the
7745 operand isn't an integer.
7750 The argument and return value are floating point numbers of the same
7756 This function returns the same values as the libm ``rint`` functions
7757 would, and handles error conditions in the same way.
7759 '``llvm.nearbyint.*``' Intrinsic
7760 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7765 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7766 floating point or vector of floating point type. Not all targets support
7771 declare float @llvm.nearbyint.f32(float %Val)
7772 declare double @llvm.nearbyint.f64(double %Val)
7773 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7774 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7775 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7780 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7786 The argument and return value are floating point numbers of the same
7792 This function returns the same values as the libm ``nearbyint``
7793 functions would, and handles error conditions in the same way.
7795 '``llvm.round.*``' Intrinsic
7796 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7801 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7802 floating point or vector of floating point type. Not all targets support
7807 declare float @llvm.round.f32(float %Val)
7808 declare double @llvm.round.f64(double %Val)
7809 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7810 declare fp128 @llvm.round.f128(fp128 %Val)
7811 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7816 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7822 The argument and return value are floating point numbers of the same
7828 This function returns the same values as the libm ``round``
7829 functions would, and handles error conditions in the same way.
7831 Bit Manipulation Intrinsics
7832 ---------------------------
7834 LLVM provides intrinsics for a few important bit manipulation
7835 operations. These allow efficient code generation for some algorithms.
7837 '``llvm.bswap.*``' Intrinsics
7838 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7843 This is an overloaded intrinsic function. You can use bswap on any
7844 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7848 declare i16 @llvm.bswap.i16(i16 <id>)
7849 declare i32 @llvm.bswap.i32(i32 <id>)
7850 declare i64 @llvm.bswap.i64(i64 <id>)
7855 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7856 values with an even number of bytes (positive multiple of 16 bits).
7857 These are useful for performing operations on data that is not in the
7858 target's native byte order.
7863 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7864 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7865 intrinsic returns an i32 value that has the four bytes of the input i32
7866 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7867 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7868 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7869 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7872 '``llvm.ctpop.*``' Intrinsic
7873 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7878 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7879 bit width, or on any vector with integer elements. Not all targets
7880 support all bit widths or vector types, however.
7884 declare i8 @llvm.ctpop.i8(i8 <src>)
7885 declare i16 @llvm.ctpop.i16(i16 <src>)
7886 declare i32 @llvm.ctpop.i32(i32 <src>)
7887 declare i64 @llvm.ctpop.i64(i64 <src>)
7888 declare i256 @llvm.ctpop.i256(i256 <src>)
7889 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7894 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7900 The only argument is the value to be counted. The argument may be of any
7901 integer type, or a vector with integer elements. The return type must
7902 match the argument type.
7907 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7908 each element of a vector.
7910 '``llvm.ctlz.*``' Intrinsic
7911 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7916 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7917 integer bit width, or any vector whose elements are integers. Not all
7918 targets support all bit widths or vector types, however.
7922 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7923 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7924 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7925 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7926 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7927 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7932 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7933 leading zeros in a variable.
7938 The first argument is the value to be counted. This argument may be of
7939 any integer type, or a vectory with integer element type. The return
7940 type must match the first argument type.
7942 The second argument must be a constant and is a flag to indicate whether
7943 the intrinsic should ensure that a zero as the first argument produces a
7944 defined result. Historically some architectures did not provide a
7945 defined result for zero values as efficiently, and many algorithms are
7946 now predicated on avoiding zero-value inputs.
7951 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7952 zeros in a variable, or within each element of the vector. If
7953 ``src == 0`` then the result is the size in bits of the type of ``src``
7954 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7955 ``llvm.ctlz(i32 2) = 30``.
7957 '``llvm.cttz.*``' Intrinsic
7958 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7963 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7964 integer bit width, or any vector of integer elements. Not all targets
7965 support all bit widths or vector types, however.
7969 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7970 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7971 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7972 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7973 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7974 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7979 The '``llvm.cttz``' family of intrinsic functions counts the number of
7985 The first argument is the value to be counted. This argument may be of
7986 any integer type, or a vectory with integer element type. The return
7987 type must match the first argument type.
7989 The second argument must be a constant and is a flag to indicate whether
7990 the intrinsic should ensure that a zero as the first argument produces a
7991 defined result. Historically some architectures did not provide a
7992 defined result for zero values as efficiently, and many algorithms are
7993 now predicated on avoiding zero-value inputs.
7998 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7999 zeros in a variable, or within each element of a vector. If ``src == 0``
8000 then the result is the size in bits of the type of ``src`` if
8001 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8002 ``llvm.cttz(2) = 1``.
8004 Arithmetic with Overflow Intrinsics
8005 -----------------------------------
8007 LLVM provides intrinsics for some arithmetic with overflow operations.
8009 '``llvm.sadd.with.overflow.*``' Intrinsics
8010 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8015 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
8016 on any integer bit width.
8020 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8021 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8022 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8027 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8028 a signed addition of the two arguments, and indicate whether an overflow
8029 occurred during the signed summation.
8034 The arguments (%a and %b) and the first element of the result structure
8035 may be of integer types of any bit width, but they must have the same
8036 bit width. The second element of the result structure must be of type
8037 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8043 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8044 a signed addition of the two variables. They return a structure --- the
8045 first element of which is the signed summation, and the second element
8046 of which is a bit specifying if the signed summation resulted in an
8052 .. code-block:: llvm
8054 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8055 %sum = extractvalue {i32, i1} %res, 0
8056 %obit = extractvalue {i32, i1} %res, 1
8057 br i1 %obit, label %overflow, label %normal
8059 '``llvm.uadd.with.overflow.*``' Intrinsics
8060 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8065 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8066 on any integer bit width.
8070 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8071 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8072 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8077 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8078 an unsigned addition of the two arguments, and indicate whether a carry
8079 occurred during the unsigned summation.
8084 The arguments (%a and %b) and the first element of the result structure
8085 may be of integer types of any bit width, but they must have the same
8086 bit width. The second element of the result structure must be of type
8087 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8093 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8094 an unsigned addition of the two arguments. They return a structure --- the
8095 first element of which is the sum, and the second element of which is a
8096 bit specifying if the unsigned summation resulted in a carry.
8101 .. code-block:: llvm
8103 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8104 %sum = extractvalue {i32, i1} %res, 0
8105 %obit = extractvalue {i32, i1} %res, 1
8106 br i1 %obit, label %carry, label %normal
8108 '``llvm.ssub.with.overflow.*``' Intrinsics
8109 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8114 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8115 on any integer bit width.
8119 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8120 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8121 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8126 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8127 a signed subtraction of the two arguments, and indicate whether an
8128 overflow occurred during the signed subtraction.
8133 The arguments (%a and %b) and the first element of the result structure
8134 may be of integer types of any bit width, but they must have the same
8135 bit width. The second element of the result structure must be of type
8136 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8142 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8143 a signed subtraction of the two arguments. They return a structure --- the
8144 first element of which is the subtraction, and the second element of
8145 which is a bit specifying if the signed subtraction resulted in an
8151 .. code-block:: llvm
8153 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8154 %sum = extractvalue {i32, i1} %res, 0
8155 %obit = extractvalue {i32, i1} %res, 1
8156 br i1 %obit, label %overflow, label %normal
8158 '``llvm.usub.with.overflow.*``' Intrinsics
8159 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8164 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8165 on any integer bit width.
8169 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8170 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8171 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8176 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8177 an unsigned subtraction of the two arguments, and indicate whether an
8178 overflow occurred during the unsigned subtraction.
8183 The arguments (%a and %b) and the first element of the result structure
8184 may be of integer types of any bit width, but they must have the same
8185 bit width. The second element of the result structure must be of type
8186 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8192 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8193 an unsigned subtraction of the two arguments. They return a structure ---
8194 the first element of which is the subtraction, and the second element of
8195 which is a bit specifying if the unsigned subtraction resulted in an
8201 .. code-block:: llvm
8203 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8204 %sum = extractvalue {i32, i1} %res, 0
8205 %obit = extractvalue {i32, i1} %res, 1
8206 br i1 %obit, label %overflow, label %normal
8208 '``llvm.smul.with.overflow.*``' Intrinsics
8209 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8214 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8215 on any integer bit width.
8219 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8220 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8221 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8226 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8227 a signed multiplication of the two arguments, and indicate whether an
8228 overflow occurred during the signed multiplication.
8233 The arguments (%a and %b) and the first element of the result structure
8234 may be of integer types of any bit width, but they must have the same
8235 bit width. The second element of the result structure must be of type
8236 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8242 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8243 a signed multiplication of the two arguments. They return a structure ---
8244 the first element of which is the multiplication, and the second element
8245 of which is a bit specifying if the signed multiplication resulted in an
8251 .. code-block:: llvm
8253 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8254 %sum = extractvalue {i32, i1} %res, 0
8255 %obit = extractvalue {i32, i1} %res, 1
8256 br i1 %obit, label %overflow, label %normal
8258 '``llvm.umul.with.overflow.*``' Intrinsics
8259 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8264 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8265 on any integer bit width.
8269 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8270 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8271 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8276 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8277 a unsigned multiplication of the two arguments, and indicate whether an
8278 overflow occurred during the unsigned multiplication.
8283 The arguments (%a and %b) and the first element of the result structure
8284 may be of integer types of any bit width, but they must have the same
8285 bit width. The second element of the result structure must be of type
8286 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8292 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8293 an unsigned multiplication of the two arguments. They return a structure ---
8294 the first element of which is the multiplication, and the second
8295 element of which is a bit specifying if the unsigned multiplication
8296 resulted in an overflow.
8301 .. code-block:: llvm
8303 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8304 %sum = extractvalue {i32, i1} %res, 0
8305 %obit = extractvalue {i32, i1} %res, 1
8306 br i1 %obit, label %overflow, label %normal
8308 Specialised Arithmetic Intrinsics
8309 ---------------------------------
8311 '``llvm.fmuladd.*``' Intrinsic
8312 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8319 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8320 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8325 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8326 expressions that can be fused if the code generator determines that (a) the
8327 target instruction set has support for a fused operation, and (b) that the
8328 fused operation is more efficient than the equivalent, separate pair of mul
8329 and add instructions.
8334 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8335 multiplicands, a and b, and an addend c.
8344 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8346 is equivalent to the expression a \* b + c, except that rounding will
8347 not be performed between the multiplication and addition steps if the
8348 code generator fuses the operations. Fusion is not guaranteed, even if
8349 the target platform supports it. If a fused multiply-add is required the
8350 corresponding llvm.fma.\* intrinsic function should be used
8351 instead. This never sets errno, just as '``llvm.fma.*``'.
8356 .. code-block:: llvm
8358 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8360 Half Precision Floating Point Intrinsics
8361 ----------------------------------------
8363 For most target platforms, half precision floating point is a
8364 storage-only format. This means that it is a dense encoding (in memory)
8365 but does not support computation in the format.
8367 This means that code must first load the half-precision floating point
8368 value as an i16, then convert it to float with
8369 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8370 then be performed on the float value (including extending to double
8371 etc). To store the value back to memory, it is first converted to float
8372 if needed, then converted to i16 with
8373 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8376 .. _int_convert_to_fp16:
8378 '``llvm.convert.to.fp16``' Intrinsic
8379 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8386 declare i16 @llvm.convert.to.fp16(f32 %a)
8391 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8392 from single precision floating point format to half precision floating
8398 The intrinsic function contains single argument - the value to be
8404 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8405 from single precision floating point format to half precision floating
8406 point format. The return value is an ``i16`` which contains the
8412 .. code-block:: llvm
8414 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8415 store i16 %res, i16* @x, align 2
8417 .. _int_convert_from_fp16:
8419 '``llvm.convert.from.fp16``' Intrinsic
8420 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8427 declare f32 @llvm.convert.from.fp16(i16 %a)
8432 The '``llvm.convert.from.fp16``' intrinsic function performs a
8433 conversion from half precision floating point format to single precision
8434 floating point format.
8439 The intrinsic function contains single argument - the value to be
8445 The '``llvm.convert.from.fp16``' intrinsic function performs a
8446 conversion from half single precision floating point format to single
8447 precision floating point format. The input half-float value is
8448 represented by an ``i16`` value.
8453 .. code-block:: llvm
8455 %a = load i16* @x, align 2
8456 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8461 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8462 prefix), are described in the `LLVM Source Level
8463 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8466 Exception Handling Intrinsics
8467 -----------------------------
8469 The LLVM exception handling intrinsics (which all start with
8470 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8471 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8475 Trampoline Intrinsics
8476 ---------------------
8478 These intrinsics make it possible to excise one parameter, marked with
8479 the :ref:`nest <nest>` attribute, from a function. The result is a
8480 callable function pointer lacking the nest parameter - the caller does
8481 not need to provide a value for it. Instead, the value to use is stored
8482 in advance in a "trampoline", a block of memory usually allocated on the
8483 stack, which also contains code to splice the nest value into the
8484 argument list. This is used to implement the GCC nested function address
8487 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8488 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8489 It can be created as follows:
8491 .. code-block:: llvm
8493 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8494 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8495 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8496 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8497 %fp = bitcast i8* %p to i32 (i32, i32)*
8499 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8500 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8504 '``llvm.init.trampoline``' Intrinsic
8505 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8512 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8517 This fills the memory pointed to by ``tramp`` with executable code,
8518 turning it into a trampoline.
8523 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8524 pointers. The ``tramp`` argument must point to a sufficiently large and
8525 sufficiently aligned block of memory; this memory is written to by the
8526 intrinsic. Note that the size and the alignment are target-specific -
8527 LLVM currently provides no portable way of determining them, so a
8528 front-end that generates this intrinsic needs to have some
8529 target-specific knowledge. The ``func`` argument must hold a function
8530 bitcast to an ``i8*``.
8535 The block of memory pointed to by ``tramp`` is filled with target
8536 dependent code, turning it into a function. Then ``tramp`` needs to be
8537 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8538 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8539 function's signature is the same as that of ``func`` with any arguments
8540 marked with the ``nest`` attribute removed. At most one such ``nest``
8541 argument is allowed, and it must be of pointer type. Calling the new
8542 function is equivalent to calling ``func`` with the same argument list,
8543 but with ``nval`` used for the missing ``nest`` argument. If, after
8544 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8545 modified, then the effect of any later call to the returned function
8546 pointer is undefined.
8550 '``llvm.adjust.trampoline``' Intrinsic
8551 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8558 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8563 This performs any required machine-specific adjustment to the address of
8564 a trampoline (passed as ``tramp``).
8569 ``tramp`` must point to a block of memory which already has trampoline
8570 code filled in by a previous call to
8571 :ref:`llvm.init.trampoline <int_it>`.
8576 On some architectures the address of the code to be executed needs to be
8577 different to the address where the trampoline is actually stored. This
8578 intrinsic returns the executable address corresponding to ``tramp``
8579 after performing the required machine specific adjustments. The pointer
8580 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8585 This class of intrinsics exists to information about the lifetime of
8586 memory objects and ranges where variables are immutable.
8590 '``llvm.lifetime.start``' Intrinsic
8591 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8598 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8603 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8609 The first argument is a constant integer representing the size of the
8610 object, or -1 if it is variable sized. The second argument is a pointer
8616 This intrinsic indicates that before this point in the code, the value
8617 of the memory pointed to by ``ptr`` is dead. This means that it is known
8618 to never be used and has an undefined value. A load from the pointer
8619 that precedes this intrinsic can be replaced with ``'undef'``.
8623 '``llvm.lifetime.end``' Intrinsic
8624 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8631 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8636 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
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 after this point in the code, the value of
8650 the memory pointed to by ``ptr`` is dead. This means that it is known to
8651 never be used and has an undefined value. Any stores into the memory
8652 object following this intrinsic may be removed as dead.
8654 '``llvm.invariant.start``' Intrinsic
8655 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8662 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8667 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8668 a memory object will not change.
8673 The first argument is a constant integer representing the size of the
8674 object, or -1 if it is variable sized. The second argument is a pointer
8680 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8681 the return value, the referenced memory location is constant and
8684 '``llvm.invariant.end``' Intrinsic
8685 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8692 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8697 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8698 memory object are mutable.
8703 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8704 The second argument is a constant integer representing the size of the
8705 object, or -1 if it is variable sized and the third argument is a
8706 pointer to the object.
8711 This intrinsic indicates that the memory is mutable again.
8716 This class of intrinsics is designed to be generic and has no specific
8719 '``llvm.var.annotation``' Intrinsic
8720 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8727 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8732 The '``llvm.var.annotation``' intrinsic.
8737 The first argument is a pointer to a value, the second is a pointer to a
8738 global string, the third is a pointer to a global string which is the
8739 source file name, and the last argument is the line number.
8744 This intrinsic allows annotation of local variables with arbitrary
8745 strings. This can be useful for special purpose optimizations that want
8746 to look for these annotations. These have no other defined use; they are
8747 ignored by code generation and optimization.
8749 '``llvm.ptr.annotation.*``' Intrinsic
8750 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8755 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8756 pointer to an integer of any width. *NOTE* you must specify an address space for
8757 the pointer. The identifier for the default address space is the integer
8762 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8763 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8764 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8765 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8766 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8771 The '``llvm.ptr.annotation``' intrinsic.
8776 The first argument is a pointer to an integer value of arbitrary bitwidth
8777 (result of some expression), the second is a pointer to a global string, the
8778 third is a pointer to a global string which is the source file name, and the
8779 last argument is the line number. It returns the value of the first argument.
8784 This intrinsic allows annotation of a pointer to an integer with arbitrary
8785 strings. This can be useful for special purpose optimizations that want to look
8786 for these annotations. These have no other defined use; they are ignored by code
8787 generation and optimization.
8789 '``llvm.annotation.*``' Intrinsic
8790 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8795 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8796 any integer bit width.
8800 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8801 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8802 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8803 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8804 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8809 The '``llvm.annotation``' intrinsic.
8814 The first argument is an integer value (result of some expression), the
8815 second is a pointer to a global string, the third is a pointer to a
8816 global string which is the source file name, and the last argument is
8817 the line number. It returns the value of the first argument.
8822 This intrinsic allows annotations to be put on arbitrary expressions
8823 with arbitrary strings. This can be useful for special purpose
8824 optimizations that want to look for these annotations. These have no
8825 other defined use; they are ignored by code generation and optimization.
8827 '``llvm.trap``' Intrinsic
8828 ^^^^^^^^^^^^^^^^^^^^^^^^^
8835 declare void @llvm.trap() noreturn nounwind
8840 The '``llvm.trap``' intrinsic.
8850 This intrinsic is lowered to the target dependent trap instruction. If
8851 the target does not have a trap instruction, this intrinsic will be
8852 lowered to a call of the ``abort()`` function.
8854 '``llvm.debugtrap``' Intrinsic
8855 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8862 declare void @llvm.debugtrap() nounwind
8867 The '``llvm.debugtrap``' intrinsic.
8877 This intrinsic is lowered to code which is intended to cause an
8878 execution trap with the intention of requesting the attention of a
8881 '``llvm.stackprotector``' Intrinsic
8882 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8889 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8894 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8895 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8896 is placed on the stack before local variables.
8901 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8902 The first argument is the value loaded from the stack guard
8903 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8904 enough space to hold the value of the guard.
8909 This intrinsic causes the prologue/epilogue inserter to force the position of
8910 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8911 to ensure that if a local variable on the stack is overwritten, it will destroy
8912 the value of the guard. When the function exits, the guard on the stack is
8913 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8914 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8915 calling the ``__stack_chk_fail()`` function.
8917 '``llvm.stackprotectorcheck``' Intrinsic
8918 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8925 declare void @llvm.stackprotectorcheck(i8** <guard>)
8930 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8931 created stack protector and if they are not equal calls the
8932 ``__stack_chk_fail()`` function.
8937 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8938 the variable ``@__stack_chk_guard``.
8943 This intrinsic is provided to perform the stack protector check by comparing
8944 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8945 values do not match call the ``__stack_chk_fail()`` function.
8947 The reason to provide this as an IR level intrinsic instead of implementing it
8948 via other IR operations is that in order to perform this operation at the IR
8949 level without an intrinsic, one would need to create additional basic blocks to
8950 handle the success/failure cases. This makes it difficult to stop the stack
8951 protector check from disrupting sibling tail calls in Codegen. With this
8952 intrinsic, we are able to generate the stack protector basic blocks late in
8953 codegen after the tail call decision has occurred.
8955 '``llvm.objectsize``' Intrinsic
8956 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8963 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8964 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8969 The ``llvm.objectsize`` intrinsic is designed to provide information to
8970 the optimizers to determine at compile time whether a) an operation
8971 (like memcpy) will overflow a buffer that corresponds to an object, or
8972 b) that a runtime check for overflow isn't necessary. An object in this
8973 context means an allocation of a specific class, structure, array, or
8979 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8980 argument is a pointer to or into the ``object``. The second argument is
8981 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8982 or -1 (if false) when the object size is unknown. The second argument
8983 only accepts constants.
8988 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8989 the size of the object concerned. If the size cannot be determined at
8990 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8991 on the ``min`` argument).
8993 '``llvm.expect``' Intrinsic
8994 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8999 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
9004 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
9005 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9006 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9011 The ``llvm.expect`` intrinsic provides information about expected (the
9012 most probable) value of ``val``, which can be used by optimizers.
9017 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
9018 a value. The second argument is an expected value, this needs to be a
9019 constant value, variables are not allowed.
9024 This intrinsic is lowered to the ``val``.
9026 '``llvm.donothing``' Intrinsic
9027 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9034 declare void @llvm.donothing() nounwind readnone
9039 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
9040 only intrinsic that can be called with an invoke instruction.
9050 This intrinsic does nothing, and it's removed by optimizers and ignored
9053 Stack Map Intrinsics
9054 --------------------
9056 LLVM provides experimental intrinsics to support runtime patching
9057 mechanisms commonly desired in dynamic language JITs. These intrinsics
9058 are described in :doc:`StackMaps`.