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 symbol is passed through the
202 assembler and evaluated by the linker. Unlike normal strong symbols,
203 they are removed by the linker from the final linked image
204 (executable or dynamic library).
205 ``linker_private_weak``
206 Similar to "``linker_private``", but the symbol is weak. Note that
207 ``linker_private_weak`` symbols are subject to coalescing by the
208 linker. The symbols are removed by the linker from the final linked
209 image (executable or dynamic library).
211 Similar to private, but the value shows as a local symbol
212 (``STB_LOCAL`` in the case of ELF) in the object file. This
213 corresponds to the notion of the '``static``' keyword in C.
214 ``available_externally``
215 Globals with "``available_externally``" linkage are never emitted
216 into the object file corresponding to the LLVM module. They exist to
217 allow inlining and other optimizations to take place given knowledge
218 of the definition of the global, which is known to be somewhere
219 outside the module. Globals with ``available_externally`` linkage
220 are allowed to be discarded at will, and are otherwise the same as
221 ``linkonce_odr``. This linkage type is only allowed on definitions,
224 Globals with "``linkonce``" linkage are merged with other globals of
225 the same name when linkage occurs. This can be used to implement
226 some forms of inline functions, templates, or other code which must
227 be generated in each translation unit that uses it, but where the
228 body may be overridden with a more definitive definition later.
229 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
230 that ``linkonce`` linkage does not actually allow the optimizer to
231 inline the body of this function into callers because it doesn't
232 know if this definition of the function is the definitive definition
233 within the program or whether it will be overridden by a stronger
234 definition. To enable inlining and other optimizations, use
235 "``linkonce_odr``" linkage.
237 "``weak``" linkage has the same merging semantics as ``linkonce``
238 linkage, except that unreferenced globals with ``weak`` linkage may
239 not be discarded. This is used for globals that are declared "weak"
242 "``common``" linkage is most similar to "``weak``" linkage, but they
243 are used for tentative definitions in C, such as "``int X;``" at
244 global scope. Symbols with "``common``" linkage are merged in the
245 same way as ``weak symbols``, and they may not be deleted if
246 unreferenced. ``common`` symbols may not have an explicit section,
247 must have a zero initializer, and may not be marked
248 ':ref:`constant <globalvars>`'. Functions and aliases may not have
251 .. _linkage_appending:
254 "``appending``" linkage may only be applied to global variables of
255 pointer to array type. When two global variables with appending
256 linkage are linked together, the two global arrays are appended
257 together. This is the LLVM, typesafe, equivalent of having the
258 system linker append together "sections" with identical names when
261 The semantics of this linkage follow the ELF object file model: the
262 symbol is weak until linked, if not linked, the symbol becomes null
263 instead of being an undefined reference.
264 ``linkonce_odr``, ``weak_odr``
265 Some languages allow differing globals to be merged, such as two
266 functions with different semantics. Other languages, such as
267 ``C++``, ensure that only equivalent globals are ever merged (the
268 "one definition rule" --- "ODR"). Such languages can use the
269 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
270 global will only be merged with equivalent globals. These linkage
271 types are otherwise the same as their non-``odr`` versions.
273 If none of the above identifiers are used, the global is externally
274 visible, meaning that it participates in linkage and can be used to
275 resolve external symbol references.
277 It is illegal for a function *declaration* to have any linkage type
278 other than ``external`` or ``extern_weak``.
285 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
286 :ref:`invokes <i_invoke>` can all have an optional calling convention
287 specified for the call. The calling convention of any pair of dynamic
288 caller/callee must match, or the behavior of the program is undefined.
289 The following calling conventions are supported by LLVM, and more may be
292 "``ccc``" - The C calling convention
293 This calling convention (the default if no other calling convention
294 is specified) matches the target C calling conventions. This calling
295 convention supports varargs function calls and tolerates some
296 mismatch in the declared prototype and implemented declaration of
297 the function (as does normal C).
298 "``fastcc``" - The fast calling convention
299 This calling convention attempts to make calls as fast as possible
300 (e.g. by passing things in registers). This calling convention
301 allows the target to use whatever tricks it wants to produce fast
302 code for the target, without having to conform to an externally
303 specified ABI (Application Binary Interface). `Tail calls can only
304 be optimized when this, the GHC or the HiPE convention is
305 used. <CodeGenerator.html#id80>`_ This calling convention does not
306 support varargs and requires the prototype of all callees to exactly
307 match the prototype of the function definition.
308 "``coldcc``" - The cold calling convention
309 This calling convention attempts to make code in the caller as
310 efficient as possible under the assumption that the call is not
311 commonly executed. As such, these calls often preserve all registers
312 so that the call does not break any live ranges in the caller side.
313 This calling convention does not support varargs and requires the
314 prototype of all callees to exactly match the prototype of the
316 "``cc 10``" - GHC convention
317 This calling convention has been implemented specifically for use by
318 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
319 It passes everything in registers, going to extremes to achieve this
320 by disabling callee save registers. This calling convention should
321 not be used lightly but only for specific situations such as an
322 alternative to the *register pinning* performance technique often
323 used when implementing functional programming languages. At the
324 moment only X86 supports this convention and it has the following
327 - On *X86-32* only supports up to 4 bit type parameters. No
328 floating point types are supported.
329 - On *X86-64* only supports up to 10 bit type parameters and 6
330 floating point parameters.
332 This calling convention supports `tail call
333 optimization <CodeGenerator.html#id80>`_ but requires both the
334 caller and callee are using it.
335 "``cc 11``" - The HiPE calling convention
336 This calling convention has been implemented specifically for use by
337 the `High-Performance Erlang
338 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
339 native code compiler of the `Ericsson's Open Source Erlang/OTP
340 system <http://www.erlang.org/download.shtml>`_. It uses more
341 registers for argument passing than the ordinary C calling
342 convention and defines no callee-saved registers. The calling
343 convention properly supports `tail call
344 optimization <CodeGenerator.html#id80>`_ but requires that both the
345 caller and the callee use it. It uses a *register pinning*
346 mechanism, similar to GHC's convention, for keeping frequently
347 accessed runtime components pinned to specific hardware registers.
348 At the moment only X86 supports this convention (both 32 and 64
350 "``webkit_jscc``" - WebKit's JavaScript calling convention
351 This calling convention has been implemented for `WebKit FTL JIT
352 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
353 stack right to left (as cdecl does), and returns a value in the
354 platform's customary return register.
355 "``anyregcc``" - Dynamic calling convention for code patching
356 This is a special convention that supports patching an arbitrary code
357 sequence in place of a call site. This convention forces the call
358 arguments into registers but allows them to be dynamcially
359 allocated. This can currently only be used with calls to
360 llvm.experimental.patchpoint because only this intrinsic records
361 the location of its arguments in a side table. See :doc:`StackMaps`.
362 "``cc <n>``" - Numbered convention
363 Any calling convention may be specified by number, allowing
364 target-specific calling conventions to be used. Target specific
365 calling conventions start at 64.
367 More calling conventions can be added/defined on an as-needed basis, to
368 support Pascal conventions or any other well-known target-independent
371 .. _visibilitystyles:
376 All Global Variables and Functions have one of the following visibility
379 "``default``" - Default style
380 On targets that use the ELF object file format, default visibility
381 means that the declaration is visible to other modules and, in
382 shared libraries, means that the declared entity may be overridden.
383 On Darwin, default visibility means that the declaration is visible
384 to other modules. Default visibility corresponds to "external
385 linkage" in the language.
386 "``hidden``" - Hidden style
387 Two declarations of an object with hidden visibility refer to the
388 same object if they are in the same shared object. Usually, hidden
389 visibility indicates that the symbol will not be placed into the
390 dynamic symbol table, so no other module (executable or shared
391 library) can reference it directly.
392 "``protected``" - Protected style
393 On ELF, protected visibility indicates that the symbol will be
394 placed in the dynamic symbol table, but that references within the
395 defining module will bind to the local symbol. That is, the symbol
396 cannot be overridden by another module.
403 All Global Variables, Functions and Aliases can have one of the following
407 "``dllimport``" causes the compiler to reference a function or variable via
408 a global pointer to a pointer that is set up by the DLL exporting the
409 symbol. On Microsoft Windows targets, the pointer name is formed by
410 combining ``__imp_`` and the function or variable name.
412 "``dllexport``" causes the compiler to provide a global pointer to a pointer
413 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
414 Microsoft Windows targets, the pointer name is formed by combining
415 ``__imp_`` and the function or variable name. Since this storage class
416 exists for defining a dll interface, the compiler, assembler and linker know
417 it is externally referenced and must refrain from deleting the symbol.
422 LLVM IR allows you to specify name aliases for certain types. This can
423 make it easier to read the IR and make the IR more condensed
424 (particularly when recursive types are involved). An example of a name
429 %mytype = type { %mytype*, i32 }
431 You may give a name to any :ref:`type <typesystem>` except
432 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
433 expected with the syntax "%mytype".
435 Note that type names are aliases for the structural type that they
436 indicate, and that you can therefore specify multiple names for the same
437 type. This often leads to confusing behavior when dumping out a .ll
438 file. Since LLVM IR uses structural typing, the name is not part of the
439 type. When printing out LLVM IR, the printer will pick *one name* to
440 render all types of a particular shape. This means that if you have code
441 where two different source types end up having the same LLVM type, that
442 the dumper will sometimes print the "wrong" or unexpected type. This is
443 an important design point and isn't going to change.
450 Global variables define regions of memory allocated at compilation time
453 Global variables definitions must be initialized, may have an explicit section
454 to be placed in, and may have an optional explicit alignment specified.
456 Global variables in other translation units can also be declared, in which
457 case they don't have an initializer.
459 A variable may be defined as ``thread_local``, which means that it will
460 not be shared by threads (each thread will have a separated copy of the
461 variable). Not all targets support thread-local variables. Optionally, a
462 TLS model may be specified:
465 For variables that are only used within the current shared library.
467 For variables in modules that will not be loaded dynamically.
469 For variables defined in the executable and only used within it.
471 The models correspond to the ELF TLS models; see `ELF Handling For
472 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
473 more information on under which circumstances the different models may
474 be used. The target may choose a different TLS model if the specified
475 model is not supported, or if a better choice of model can be made.
477 A variable may be defined as a global ``constant``, which indicates that
478 the contents of the variable will **never** be modified (enabling better
479 optimization, allowing the global data to be placed in the read-only
480 section of an executable, etc). Note that variables that need runtime
481 initialization cannot be marked ``constant`` as there is a store to the
484 LLVM explicitly allows *declarations* of global variables to be marked
485 constant, even if the final definition of the global is not. This
486 capability can be used to enable slightly better optimization of the
487 program, but requires the language definition to guarantee that
488 optimizations based on the 'constantness' are valid for the translation
489 units that do not include the definition.
491 As SSA values, global variables define pointer values that are in scope
492 (i.e. they dominate) all basic blocks in the program. Global variables
493 always define a pointer to their "content" type because they describe a
494 region of memory, and all memory objects in LLVM are accessed through
497 Global variables can be marked with ``unnamed_addr`` which indicates
498 that the address is not significant, only the content. Constants marked
499 like this can be merged with other constants if they have the same
500 initializer. Note that a constant with significant address *can* be
501 merged with a ``unnamed_addr`` constant, the result being a constant
502 whose address is significant.
504 A global variable may be declared to reside in a target-specific
505 numbered address space. For targets that support them, address spaces
506 may affect how optimizations are performed and/or what target
507 instructions are used to access the variable. The default address space
508 is zero. The address space qualifier must precede any other attributes.
510 LLVM allows an explicit section to be specified for globals. If the
511 target supports it, it will emit globals to the section specified.
513 By default, global initializers are optimized by assuming that global
514 variables defined within the module are not modified from their
515 initial values before the start of the global initializer. This is
516 true even for variables potentially accessible from outside the
517 module, including those with external linkage or appearing in
518 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
519 by marking the variable with ``externally_initialized``.
521 An explicit alignment may be specified for a global, which must be a
522 power of 2. If not present, or if the alignment is set to zero, the
523 alignment of the global is set by the target to whatever it feels
524 convenient. If an explicit alignment is specified, the global is forced
525 to have exactly that alignment. Targets and optimizers are not allowed
526 to over-align the global if the global has an assigned section. In this
527 case, the extra alignment could be observable: for example, code could
528 assume that the globals are densely packed in their section and try to
529 iterate over them as an array, alignment padding would break this
532 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
536 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
537 [AddrSpace] [unnamed_addr] [ExternallyInitialized]
538 <global | constant> <Type>
539 [, section "name"] [, align <Alignment>]
541 For example, the following defines a global in a numbered address space
542 with an initializer, section, and alignment:
546 @G = addrspace(5) constant float 1.0, section "foo", align 4
548 The following example just declares a global variable
552 @G = external global i32
554 The following example defines a thread-local global with the
555 ``initialexec`` TLS model:
559 @G = thread_local(initialexec) global i32 0, align 4
561 .. _functionstructure:
566 LLVM function definitions consist of the "``define``" keyword, an
567 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
568 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
569 an optional :ref:`calling convention <callingconv>`,
570 an optional ``unnamed_addr`` attribute, a return type, an optional
571 :ref:`parameter attribute <paramattrs>` for the return type, a function
572 name, a (possibly empty) argument list (each with optional :ref:`parameter
573 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
574 an optional section, an optional alignment, an optional :ref:`garbage
575 collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
576 curly brace, a list of basic blocks, and a closing curly brace.
578 LLVM function declarations consist of the "``declare``" keyword, an
579 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
580 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
581 an optional :ref:`calling convention <callingconv>`,
582 an optional ``unnamed_addr`` attribute, a return type, an optional
583 :ref:`parameter attribute <paramattrs>` for the return type, a function
584 name, a possibly empty list of arguments, an optional alignment, an optional
585 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
587 A function definition contains a list of basic blocks, forming the CFG (Control
588 Flow Graph) for the function. Each basic block may optionally start with a label
589 (giving the basic block a symbol table entry), contains a list of instructions,
590 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
591 function return). If an explicit label is not provided, a block is assigned an
592 implicit numbered label, using the next value from the same counter as used for
593 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
594 entry block does not have an explicit label, it will be assigned label "%0",
595 then the first unnamed temporary in that block will be "%1", etc.
597 The first basic block in a function is special in two ways: it is
598 immediately executed on entrance to the function, and it is not allowed
599 to have predecessor basic blocks (i.e. there can not be any branches to
600 the entry block of a function). Because the block can have no
601 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
603 LLVM allows an explicit section to be specified for functions. If the
604 target supports it, it will emit functions to the section specified.
606 An explicit alignment may be specified for a function. If not present,
607 or if the alignment is set to zero, the alignment of the function is set
608 by the target to whatever it feels convenient. If an explicit alignment
609 is specified, the function is forced to have at least that much
610 alignment. All alignments must be a power of 2.
612 If the ``unnamed_addr`` attribute is given, the address is know to not
613 be significant and two identical functions can be merged.
617 define [linkage] [visibility] [DLLStorageClass]
619 <ResultType> @<FunctionName> ([argument list])
620 [fn Attrs] [section "name"] [align N]
621 [gc] [prefix Constant] { ... }
628 Aliases act as "second name" for the aliasee value (which can be either
629 function, global variable, another alias or bitcast of global value).
630 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
631 :ref:`visibility style <visibility>`, and an optional :ref:`DLL storage class
636 @<Name> = [Visibility] [DLLStorageClass] alias [Linkage] <AliaseeTy> @<Aliasee>
638 The linkage must be one of ``private``, ``linker_private``,
639 ``linker_private_weak``, ``internal``, ``linkonce``, ``weak``,
640 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
641 might not correctly handle dropping a weak symbol that is aliased by a non-weak
644 .. _namedmetadatastructure:
649 Named metadata is a collection of metadata. :ref:`Metadata
650 nodes <metadata>` (but not metadata strings) are the only valid
651 operands for a named metadata.
655 ; Some unnamed metadata nodes, which are referenced by the named metadata.
656 !0 = metadata !{metadata !"zero"}
657 !1 = metadata !{metadata !"one"}
658 !2 = metadata !{metadata !"two"}
660 !name = !{!0, !1, !2}
667 The return type and each parameter of a function type may have a set of
668 *parameter attributes* associated with them. Parameter attributes are
669 used to communicate additional information about the result or
670 parameters of a function. Parameter attributes are considered to be part
671 of the function, not of the function type, so functions with different
672 parameter attributes can have the same function type.
674 Parameter attributes are simple keywords that follow the type specified.
675 If multiple parameter attributes are needed, they are space separated.
680 declare i32 @printf(i8* noalias nocapture, ...)
681 declare i32 @atoi(i8 zeroext)
682 declare signext i8 @returns_signed_char()
684 Note that any attributes for the function result (``nounwind``,
685 ``readonly``) come immediately after the argument list.
687 Currently, only the following parameter attributes are defined:
690 This indicates to the code generator that the parameter or return
691 value should be zero-extended to the extent required by the target's
692 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
693 the caller (for a parameter) or the callee (for a return value).
695 This indicates to the code generator that the parameter or return
696 value should be sign-extended to the extent required by the target's
697 ABI (which is usually 32-bits) by the caller (for a parameter) or
698 the callee (for a return value).
700 This indicates that this parameter or return value should be treated
701 in a special target-dependent fashion during while emitting code for
702 a function call or return (usually, by putting it in a register as
703 opposed to memory, though some targets use it to distinguish between
704 two different kinds of registers). Use of this attribute is
707 This indicates that the pointer parameter should really be passed by
708 value to the function. The attribute implies that a hidden copy of
709 the pointee is made between the caller and the callee, so the callee
710 is unable to modify the value in the caller. This attribute is only
711 valid on LLVM pointer arguments. It is generally used to pass
712 structs and arrays by value, but is also valid on pointers to
713 scalars. The copy is considered to belong to the caller not the
714 callee (for example, ``readonly`` functions should not write to
715 ``byval`` parameters). This is not a valid attribute for return
718 The byval attribute also supports specifying an alignment with the
719 align attribute. It indicates the alignment of the stack slot to
720 form and the known alignment of the pointer specified to the call
721 site. If the alignment is not specified, then the code generator
722 makes a target-specific assumption.
728 .. Warning:: This feature is unstable and not fully implemented.
730 The ``inalloca`` argument attribute allows the caller to get the
731 address of an outgoing argument to a ``call`` or ``invoke`` before
732 it executes. It is similar to ``byval`` in that it is used to pass
733 arguments by value, but it guarantees that the argument will not be
736 To be :ref:`well formed <wellformed>`, the caller must pass in an
737 alloca value into an ``inalloca`` parameter, and an alloca may be
738 used as an ``inalloca`` argument at most once. The attribute can
739 only be applied to parameters that would be passed in memory and not
740 registers. The ``inalloca`` attribute cannot be used in conjunction
741 with other attributes that affect argument storage, like ``inreg``,
742 ``nest``, ``sret``, or ``byval``. The ``inalloca`` stack space is
743 considered to be clobbered by any call that uses it, so any
744 ``inalloca`` parameters cannot be marked ``readonly``.
746 Allocas passed with ``inalloca`` to a call must be in the opposite
747 order of the parameter list, meaning that the rightmost argument
748 must be allocated first. If a call has inalloca arguments, no other
749 allocas can occur between the first alloca used by the call and the
750 call site, unless they are are cleared by calls to
751 :ref:`llvm.stackrestore <int_stackrestore>`. Violating these rules
752 results in undefined behavior at runtime.
754 See :doc:`InAlloca` for more information on how to use this
758 This indicates that the pointer parameter specifies the address of a
759 structure that is the return value of the function in the source
760 program. This pointer must be guaranteed by the caller to be valid:
761 loads and stores to the structure may be assumed by the callee
762 not to trap and to be properly aligned. This may only be applied to
763 the first parameter. This is not a valid attribute for return
766 This indicates that pointer values :ref:`based <pointeraliasing>` on
767 the argument or return value do not alias pointer values which are
768 not *based* on it, ignoring certain "irrelevant" dependencies. For a
769 call to the parent function, dependencies between memory references
770 from before or after the call and from those during the call are
771 "irrelevant" to the ``noalias`` keyword for the arguments and return
772 value used in that call. The caller shares the responsibility with
773 the callee for ensuring that these requirements are met. For further
774 details, please see the discussion of the NoAlias response in `alias
775 analysis <AliasAnalysis.html#MustMayNo>`_.
777 Note that this definition of ``noalias`` is intentionally similar
778 to the definition of ``restrict`` in C99 for function arguments,
779 though it is slightly weaker.
781 For function return values, C99's ``restrict`` is not meaningful,
782 while LLVM's ``noalias`` is.
784 This indicates that the callee does not make any copies of the
785 pointer that outlive the callee itself. This is not a valid
786 attribute for return values.
791 This indicates that the pointer parameter can be excised using the
792 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
793 attribute for return values and can only be applied to one parameter.
796 This indicates that the function always returns the argument as its return
797 value. This is an optimization hint to the code generator when generating
798 the caller, allowing tail call optimization and omission of register saves
799 and restores in some cases; it is not checked or enforced when generating
800 the callee. The parameter and the function return type must be valid
801 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
802 valid attribute for return values and can only be applied to one parameter.
806 Garbage Collector Names
807 -----------------------
809 Each function may specify a garbage collector name, which is simply a
814 define void @f() gc "name" { ... }
816 The compiler declares the supported values of *name*. Specifying a
817 collector which will cause the compiler to alter its output in order to
818 support the named garbage collection algorithm.
825 Prefix data is data associated with a function which the code generator
826 will emit immediately before the function body. The purpose of this feature
827 is to allow frontends to associate language-specific runtime metadata with
828 specific functions and make it available through the function pointer while
829 still allowing the function pointer to be called. To access the data for a
830 given function, a program may bitcast the function pointer to a pointer to
831 the constant's type. This implies that the IR symbol points to the start
834 To maintain the semantics of ordinary function calls, the prefix data must
835 have a particular format. Specifically, it must begin with a sequence of
836 bytes which decode to a sequence of machine instructions, valid for the
837 module's target, which transfer control to the point immediately succeeding
838 the prefix data, without performing any other visible action. This allows
839 the inliner and other passes to reason about the semantics of the function
840 definition without needing to reason about the prefix data. Obviously this
841 makes the format of the prefix data highly target dependent.
843 Prefix data is laid out as if it were an initializer for a global variable
844 of the prefix data's type. No padding is automatically placed between the
845 prefix data and the function body. If padding is required, it must be part
848 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
849 which encodes the ``nop`` instruction:
853 define void @f() prefix i8 144 { ... }
855 Generally prefix data can be formed by encoding a relative branch instruction
856 which skips the metadata, as in this example of valid prefix data for the
857 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
861 %0 = type <{ i8, i8, i8* }>
863 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
865 A function may have prefix data but no body. This has similar semantics
866 to the ``available_externally`` linkage in that the data may be used by the
867 optimizers but will not be emitted in the object file.
874 Attribute groups are groups of attributes that are referenced by objects within
875 the IR. They are important for keeping ``.ll`` files readable, because a lot of
876 functions will use the same set of attributes. In the degenerative case of a
877 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
878 group will capture the important command line flags used to build that file.
880 An attribute group is a module-level object. To use an attribute group, an
881 object references the attribute group's ID (e.g. ``#37``). An object may refer
882 to more than one attribute group. In that situation, the attributes from the
883 different groups are merged.
885 Here is an example of attribute groups for a function that should always be
886 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
890 ; Target-independent attributes:
891 attributes #0 = { alwaysinline alignstack=4 }
893 ; Target-dependent attributes:
894 attributes #1 = { "no-sse" }
896 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
897 define void @f() #0 #1 { ... }
904 Function attributes are set to communicate additional information about
905 a function. Function attributes are considered to be part of the
906 function, not of the function type, so functions with different function
907 attributes can have the same function type.
909 Function attributes are simple keywords that follow the type specified.
910 If multiple attributes are needed, they are space separated. For
915 define void @f() noinline { ... }
916 define void @f() alwaysinline { ... }
917 define void @f() alwaysinline optsize { ... }
918 define void @f() optsize { ... }
921 This attribute indicates that, when emitting the prologue and
922 epilogue, the backend should forcibly align the stack pointer.
923 Specify the desired alignment, which must be a power of two, in
926 This attribute indicates that the inliner should attempt to inline
927 this function into callers whenever possible, ignoring any active
928 inlining size threshold for this caller.
930 This indicates that the callee function at a call site should be
931 recognized as a built-in function, even though the function's declaration
932 uses the ``nobuiltin`` attribute. This is only valid at call sites for
933 direct calls to functions which are declared with the ``nobuiltin``
936 This attribute indicates that this function is rarely called. When
937 computing edge weights, basic blocks post-dominated by a cold
938 function call are also considered to be cold; and, thus, given low
941 This attribute indicates that the source code contained a hint that
942 inlining this function is desirable (such as the "inline" keyword in
943 C/C++). It is just a hint; it imposes no requirements on the
946 This attribute suggests that optimization passes and code generator
947 passes make choices that keep the code size of this function as small
948 as possible and perform optimizations that may sacrifice runtime
949 performance in order to minimize the size of the generated code.
951 This attribute disables prologue / epilogue emission for the
952 function. This can have very system-specific consequences.
954 This indicates that the callee function at a call site is not recognized as
955 a built-in function. LLVM will retain the original call and not replace it
956 with equivalent code based on the semantics of the built-in function, unless
957 the call site uses the ``builtin`` attribute. This is valid at call sites
958 and on function declarations and definitions.
960 This attribute indicates that calls to the function cannot be
961 duplicated. A call to a ``noduplicate`` function may be moved
962 within its parent function, but may not be duplicated within
965 A function containing a ``noduplicate`` call may still
966 be an inlining candidate, provided that the call is not
967 duplicated by inlining. That implies that the function has
968 internal linkage and only has one call site, so the original
969 call is dead after inlining.
971 This attributes disables implicit floating point instructions.
973 This attribute indicates that the inliner should never inline this
974 function in any situation. This attribute may not be used together
975 with the ``alwaysinline`` attribute.
977 This attribute suppresses lazy symbol binding for the function. This
978 may make calls to the function faster, at the cost of extra program
979 startup time if the function is not called during program startup.
981 This attribute indicates that the code generator should not use a
982 red zone, even if the target-specific ABI normally permits it.
984 This function attribute indicates that the function never returns
985 normally. This produces undefined behavior at runtime if the
986 function ever does dynamically return.
988 This function attribute indicates that the function never returns
989 with an unwind or exceptional control flow. If the function does
990 unwind, its runtime behavior is undefined.
992 This function attribute indicates that the function is not optimized
993 by any optimization or code generator passes with the
994 exception of interprocedural optimization passes.
995 This attribute cannot be used together with the ``alwaysinline``
996 attribute; this attribute is also incompatible
997 with the ``minsize`` attribute and the ``optsize`` attribute.
999 This attribute requires the ``noinline`` attribute to be specified on
1000 the function as well, so the function is never inlined into any caller.
1001 Only functions with the ``alwaysinline`` attribute are valid
1002 candidates for inlining into the body of this function.
1004 This attribute suggests that optimization passes and code generator
1005 passes make choices that keep the code size of this function low,
1006 and otherwise do optimizations specifically to reduce code size as
1007 long as they do not significantly impact runtime performance.
1009 On a function, this attribute indicates that the function computes its
1010 result (or decides to unwind an exception) based strictly on its arguments,
1011 without dereferencing any pointer arguments or otherwise accessing
1012 any mutable state (e.g. memory, control registers, etc) visible to
1013 caller functions. It does not write through any pointer arguments
1014 (including ``byval`` arguments) and never changes any state visible
1015 to callers. This means that it cannot unwind exceptions by calling
1016 the ``C++`` exception throwing methods.
1018 On an argument, this attribute indicates that the function does not
1019 dereference that pointer argument, even though it may read or write the
1020 memory that the pointer points to if accessed through other pointers.
1022 On a function, this attribute indicates that the function does not write
1023 through any pointer arguments (including ``byval`` arguments) or otherwise
1024 modify any state (e.g. memory, control registers, etc) visible to
1025 caller functions. It may dereference pointer arguments and read
1026 state that may be set in the caller. A readonly function always
1027 returns the same value (or unwinds an exception identically) when
1028 called with the same set of arguments and global state. It cannot
1029 unwind an exception by calling the ``C++`` exception throwing
1032 On an argument, this attribute indicates that the function does not write
1033 through this pointer argument, even though it may write to the memory that
1034 the pointer points to.
1036 This attribute indicates that this function can return twice. The C
1037 ``setjmp`` is an example of such a function. The compiler disables
1038 some optimizations (like tail calls) in the caller of these
1040 ``sanitize_address``
1041 This attribute indicates that AddressSanitizer checks
1042 (dynamic address safety analysis) are enabled for this function.
1044 This attribute indicates that MemorySanitizer checks (dynamic detection
1045 of accesses to uninitialized memory) are enabled for this function.
1047 This attribute indicates that ThreadSanitizer checks
1048 (dynamic thread safety analysis) are enabled for this function.
1050 This attribute indicates that the function should emit a stack
1051 smashing protector. It is in the form of a "canary" --- a random value
1052 placed on the stack before the local variables that's checked upon
1053 return from the function to see if it has been overwritten. A
1054 heuristic is used to determine if a function needs stack protectors
1055 or not. The heuristic used will enable protectors for functions with:
1057 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1058 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1059 - Calls to alloca() with variable sizes or constant sizes greater than
1060 ``ssp-buffer-size``.
1062 If a function that has an ``ssp`` attribute is inlined into a
1063 function that doesn't have an ``ssp`` attribute, then the resulting
1064 function will have an ``ssp`` attribute.
1066 This attribute indicates that the function should *always* emit a
1067 stack smashing protector. This overrides the ``ssp`` function
1070 If a function that has an ``sspreq`` attribute is inlined into a
1071 function that doesn't have an ``sspreq`` attribute or which has an
1072 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1073 an ``sspreq`` attribute.
1075 This attribute indicates that the function should emit a stack smashing
1076 protector. This attribute causes a strong heuristic to be used when
1077 determining if a function needs stack protectors. The strong heuristic
1078 will enable protectors for functions with:
1080 - Arrays of any size and type
1081 - Aggregates containing an array of any size and type.
1082 - Calls to alloca().
1083 - Local variables that have had their address taken.
1085 This overrides the ``ssp`` function attribute.
1087 If a function that has an ``sspstrong`` attribute is inlined into a
1088 function that doesn't have an ``sspstrong`` attribute, then the
1089 resulting function will have an ``sspstrong`` attribute.
1091 This attribute indicates that the ABI being targeted requires that
1092 an unwind table entry be produce for this function even if we can
1093 show that no exceptions passes by it. This is normally the case for
1094 the ELF x86-64 abi, but it can be disabled for some compilation
1099 Module-Level Inline Assembly
1100 ----------------------------
1102 Modules may contain "module-level inline asm" blocks, which corresponds
1103 to the GCC "file scope inline asm" blocks. These blocks are internally
1104 concatenated by LLVM and treated as a single unit, but may be separated
1105 in the ``.ll`` file if desired. The syntax is very simple:
1107 .. code-block:: llvm
1109 module asm "inline asm code goes here"
1110 module asm "more can go here"
1112 The strings can contain any character by escaping non-printable
1113 characters. The escape sequence used is simply "\\xx" where "xx" is the
1114 two digit hex code for the number.
1116 The inline asm code is simply printed to the machine code .s file when
1117 assembly code is generated.
1119 .. _langref_datalayout:
1124 A module may specify a target specific data layout string that specifies
1125 how data is to be laid out in memory. The syntax for the data layout is
1128 .. code-block:: llvm
1130 target datalayout = "layout specification"
1132 The *layout specification* consists of a list of specifications
1133 separated by the minus sign character ('-'). Each specification starts
1134 with a letter and may include other information after the letter to
1135 define some aspect of the data layout. The specifications accepted are
1139 Specifies that the target lays out data in big-endian form. That is,
1140 the bits with the most significance have the lowest address
1143 Specifies that the target lays out data in little-endian form. That
1144 is, the bits with the least significance have the lowest address
1147 Specifies the natural alignment of the stack in bits. Alignment
1148 promotion of stack variables is limited to the natural stack
1149 alignment to avoid dynamic stack realignment. The stack alignment
1150 must be a multiple of 8-bits. If omitted, the natural stack
1151 alignment defaults to "unspecified", which does not prevent any
1152 alignment promotions.
1153 ``p[n]:<size>:<abi>:<pref>``
1154 This specifies the *size* of a pointer and its ``<abi>`` and
1155 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1156 bits. The address space, ``n`` is optional, and if not specified,
1157 denotes the default address space 0. The value of ``n`` must be
1158 in the range [1,2^23).
1159 ``i<size>:<abi>:<pref>``
1160 This specifies the alignment for an integer type of a given bit
1161 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1162 ``v<size>:<abi>:<pref>``
1163 This specifies the alignment for a vector type of a given bit
1165 ``f<size>:<abi>:<pref>``
1166 This specifies the alignment for a floating point type of a given bit
1167 ``<size>``. Only values of ``<size>`` that are supported by the target
1168 will work. 32 (float) and 64 (double) are supported on all targets; 80
1169 or 128 (different flavors of long double) are also supported on some
1172 This specifies the alignment for an object of aggregate type.
1174 If prerest, specifies that llvm names are mangled in the output. The
1176 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1177 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1178 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1179 symbols get a ``_`` prefix.
1180 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1181 functions also get a suffix based on the frame size.
1182 ``n<size1>:<size2>:<size3>...``
1183 This specifies a set of native integer widths for the target CPU in
1184 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1185 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1186 this set are considered to support most general arithmetic operations
1189 On every specification that takes a ``<abi>:<pref>``, specifying the
1190 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1191 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1193 When constructing the data layout for a given target, LLVM starts with a
1194 default set of specifications which are then (possibly) overridden by
1195 the specifications in the ``datalayout`` keyword. The default
1196 specifications are given in this list:
1198 - ``E`` - big endian
1199 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1200 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1201 same as the default address space.
1202 - ``S0`` - natural stack alignment is unspecified
1203 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1204 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1205 - ``i16:16:16`` - i16 is 16-bit aligned
1206 - ``i32:32:32`` - i32 is 32-bit aligned
1207 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1208 alignment of 64-bits
1209 - ``f16:16:16`` - half is 16-bit aligned
1210 - ``f32:32:32`` - float is 32-bit aligned
1211 - ``f64:64:64`` - double is 64-bit aligned
1212 - ``f128:128:128`` - quad is 128-bit aligned
1213 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1214 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1215 - ``a:0:64`` - aggregates are 64-bit aligned
1217 When LLVM is determining the alignment for a given type, it uses the
1220 #. If the type sought is an exact match for one of the specifications,
1221 that specification is used.
1222 #. If no match is found, and the type sought is an integer type, then
1223 the smallest integer type that is larger than the bitwidth of the
1224 sought type is used. If none of the specifications are larger than
1225 the bitwidth then the largest integer type is used. For example,
1226 given the default specifications above, the i7 type will use the
1227 alignment of i8 (next largest) while both i65 and i256 will use the
1228 alignment of i64 (largest specified).
1229 #. If no match is found, and the type sought is a vector type, then the
1230 largest vector type that is smaller than the sought vector type will
1231 be used as a fall back. This happens because <128 x double> can be
1232 implemented in terms of 64 <2 x double>, for example.
1234 The function of the data layout string may not be what you expect.
1235 Notably, this is not a specification from the frontend of what alignment
1236 the code generator should use.
1238 Instead, if specified, the target data layout is required to match what
1239 the ultimate *code generator* expects. This string is used by the
1240 mid-level optimizers to improve code, and this only works if it matches
1241 what the ultimate code generator uses. If you would like to generate IR
1242 that does not embed this target-specific detail into the IR, then you
1243 don't have to specify the string. This will disable some optimizations
1244 that require precise layout information, but this also prevents those
1245 optimizations from introducing target specificity into the IR.
1252 A module may specify a target triple string that describes the target
1253 host. The syntax for the target triple is simply:
1255 .. code-block:: llvm
1257 target triple = "x86_64-apple-macosx10.7.0"
1259 The *target triple* string consists of a series of identifiers delimited
1260 by the minus sign character ('-'). The canonical forms are:
1264 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1265 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1267 This information is passed along to the backend so that it generates
1268 code for the proper architecture. It's possible to override this on the
1269 command line with the ``-mtriple`` command line option.
1271 .. _pointeraliasing:
1273 Pointer Aliasing Rules
1274 ----------------------
1276 Any memory access must be done through a pointer value associated with
1277 an address range of the memory access, otherwise the behavior is
1278 undefined. Pointer values are associated with address ranges according
1279 to the following rules:
1281 - A pointer value is associated with the addresses associated with any
1282 value it is *based* on.
1283 - An address of a global variable is associated with the address range
1284 of the variable's storage.
1285 - The result value of an allocation instruction is associated with the
1286 address range of the allocated storage.
1287 - A null pointer in the default address-space is associated with no
1289 - An integer constant other than zero or a pointer value returned from
1290 a function not defined within LLVM may be associated with address
1291 ranges allocated through mechanisms other than those provided by
1292 LLVM. Such ranges shall not overlap with any ranges of addresses
1293 allocated by mechanisms provided by LLVM.
1295 A pointer value is *based* on another pointer value according to the
1298 - A pointer value formed from a ``getelementptr`` operation is *based*
1299 on the first operand of the ``getelementptr``.
1300 - The result value of a ``bitcast`` is *based* on the operand of the
1302 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1303 values that contribute (directly or indirectly) to the computation of
1304 the pointer's value.
1305 - The "*based* on" relationship is transitive.
1307 Note that this definition of *"based"* is intentionally similar to the
1308 definition of *"based"* in C99, though it is slightly weaker.
1310 LLVM IR does not associate types with memory. The result type of a
1311 ``load`` merely indicates the size and alignment of the memory from
1312 which to load, as well as the interpretation of the value. The first
1313 operand type of a ``store`` similarly only indicates the size and
1314 alignment of the store.
1316 Consequently, type-based alias analysis, aka TBAA, aka
1317 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1318 :ref:`Metadata <metadata>` may be used to encode additional information
1319 which specialized optimization passes may use to implement type-based
1324 Volatile Memory Accesses
1325 ------------------------
1327 Certain memory accesses, such as :ref:`load <i_load>`'s,
1328 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1329 marked ``volatile``. The optimizers must not change the number of
1330 volatile operations or change their order of execution relative to other
1331 volatile operations. The optimizers *may* change the order of volatile
1332 operations relative to non-volatile operations. This is not Java's
1333 "volatile" and has no cross-thread synchronization behavior.
1335 IR-level volatile loads and stores cannot safely be optimized into
1336 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1337 flagged volatile. Likewise, the backend should never split or merge
1338 target-legal volatile load/store instructions.
1340 .. admonition:: Rationale
1342 Platforms may rely on volatile loads and stores of natively supported
1343 data width to be executed as single instruction. For example, in C
1344 this holds for an l-value of volatile primitive type with native
1345 hardware support, but not necessarily for aggregate types. The
1346 frontend upholds these expectations, which are intentionally
1347 unspecified in the IR. The rules above ensure that IR transformation
1348 do not violate the frontend's contract with the language.
1352 Memory Model for Concurrent Operations
1353 --------------------------------------
1355 The LLVM IR does not define any way to start parallel threads of
1356 execution or to register signal handlers. Nonetheless, there are
1357 platform-specific ways to create them, and we define LLVM IR's behavior
1358 in their presence. This model is inspired by the C++0x memory model.
1360 For a more informal introduction to this model, see the :doc:`Atomics`.
1362 We define a *happens-before* partial order as the least partial order
1365 - Is a superset of single-thread program order, and
1366 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1367 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1368 techniques, like pthread locks, thread creation, thread joining,
1369 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1370 Constraints <ordering>`).
1372 Note that program order does not introduce *happens-before* edges
1373 between a thread and signals executing inside that thread.
1375 Every (defined) read operation (load instructions, memcpy, atomic
1376 loads/read-modify-writes, etc.) R reads a series of bytes written by
1377 (defined) write operations (store instructions, atomic
1378 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1379 section, initialized globals are considered to have a write of the
1380 initializer which is atomic and happens before any other read or write
1381 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1382 may see any write to the same byte, except:
1384 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1385 write\ :sub:`2` happens before R\ :sub:`byte`, then
1386 R\ :sub:`byte` does not see write\ :sub:`1`.
1387 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1388 R\ :sub:`byte` does not see write\ :sub:`3`.
1390 Given that definition, R\ :sub:`byte` is defined as follows:
1392 - If R is volatile, the result is target-dependent. (Volatile is
1393 supposed to give guarantees which can support ``sig_atomic_t`` in
1394 C/C++, and may be used for accesses to addresses which do not behave
1395 like normal memory. It does not generally provide cross-thread
1397 - Otherwise, if there is no write to the same byte that happens before
1398 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1399 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1400 R\ :sub:`byte` returns the value written by that write.
1401 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1402 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1403 Memory Ordering Constraints <ordering>` section for additional
1404 constraints on how the choice is made.
1405 - Otherwise R\ :sub:`byte` returns ``undef``.
1407 R returns the value composed of the series of bytes it read. This
1408 implies that some bytes within the value may be ``undef`` **without**
1409 the entire value being ``undef``. Note that this only defines the
1410 semantics of the operation; it doesn't mean that targets will emit more
1411 than one instruction to read the series of bytes.
1413 Note that in cases where none of the atomic intrinsics are used, this
1414 model places only one restriction on IR transformations on top of what
1415 is required for single-threaded execution: introducing a store to a byte
1416 which might not otherwise be stored is not allowed in general.
1417 (Specifically, in the case where another thread might write to and read
1418 from an address, introducing a store can change a load that may see
1419 exactly one write into a load that may see multiple writes.)
1423 Atomic Memory Ordering Constraints
1424 ----------------------------------
1426 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1427 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1428 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1429 an ordering parameter that determines which other atomic instructions on
1430 the same address they *synchronize with*. These semantics are borrowed
1431 from Java and C++0x, but are somewhat more colloquial. If these
1432 descriptions aren't precise enough, check those specs (see spec
1433 references in the :doc:`atomics guide <Atomics>`).
1434 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1435 differently since they don't take an address. See that instruction's
1436 documentation for details.
1438 For a simpler introduction to the ordering constraints, see the
1442 The set of values that can be read is governed by the happens-before
1443 partial order. A value cannot be read unless some operation wrote
1444 it. This is intended to provide a guarantee strong enough to model
1445 Java's non-volatile shared variables. This ordering cannot be
1446 specified for read-modify-write operations; it is not strong enough
1447 to make them atomic in any interesting way.
1449 In addition to the guarantees of ``unordered``, there is a single
1450 total order for modifications by ``monotonic`` operations on each
1451 address. All modification orders must be compatible with the
1452 happens-before order. There is no guarantee that the modification
1453 orders can be combined to a global total order for the whole program
1454 (and this often will not be possible). The read in an atomic
1455 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1456 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1457 order immediately before the value it writes. If one atomic read
1458 happens before another atomic read of the same address, the later
1459 read must see the same value or a later value in the address's
1460 modification order. This disallows reordering of ``monotonic`` (or
1461 stronger) operations on the same address. If an address is written
1462 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1463 read that address repeatedly, the other threads must eventually see
1464 the write. This corresponds to the C++0x/C1x
1465 ``memory_order_relaxed``.
1467 In addition to the guarantees of ``monotonic``, a
1468 *synchronizes-with* edge may be formed with a ``release`` operation.
1469 This is intended to model C++'s ``memory_order_acquire``.
1471 In addition to the guarantees of ``monotonic``, if this operation
1472 writes a value which is subsequently read by an ``acquire``
1473 operation, it *synchronizes-with* that operation. (This isn't a
1474 complete description; see the C++0x definition of a release
1475 sequence.) This corresponds to the C++0x/C1x
1476 ``memory_order_release``.
1477 ``acq_rel`` (acquire+release)
1478 Acts as both an ``acquire`` and ``release`` operation on its
1479 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1480 ``seq_cst`` (sequentially consistent)
1481 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1482 operation which only reads, ``release`` for an operation which only
1483 writes), there is a global total order on all
1484 sequentially-consistent operations on all addresses, which is
1485 consistent with the *happens-before* partial order and with the
1486 modification orders of all the affected addresses. Each
1487 sequentially-consistent read sees the last preceding write to the
1488 same address in this global order. This corresponds to the C++0x/C1x
1489 ``memory_order_seq_cst`` and Java volatile.
1493 If an atomic operation is marked ``singlethread``, it only *synchronizes
1494 with* or participates in modification and seq\_cst total orderings with
1495 other operations running in the same thread (for example, in signal
1503 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1504 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1505 :ref:`frem <i_frem>`) have the following flags that can set to enable
1506 otherwise unsafe floating point operations
1509 No NaNs - Allow optimizations to assume the arguments and result are not
1510 NaN. Such optimizations are required to retain defined behavior over
1511 NaNs, but the value of the result is undefined.
1514 No Infs - Allow optimizations to assume the arguments and result are not
1515 +/-Inf. Such optimizations are required to retain defined behavior over
1516 +/-Inf, but the value of the result is undefined.
1519 No Signed Zeros - Allow optimizations to treat the sign of a zero
1520 argument or result as insignificant.
1523 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1524 argument rather than perform division.
1527 Fast - Allow algebraically equivalent transformations that may
1528 dramatically change results in floating point (e.g. reassociate). This
1529 flag implies all the others.
1536 The LLVM type system is one of the most important features of the
1537 intermediate representation. Being typed enables a number of
1538 optimizations to be performed on the intermediate representation
1539 directly, without having to do extra analyses on the side before the
1540 transformation. A strong type system makes it easier to read the
1541 generated code and enables novel analyses and transformations that are
1542 not feasible to perform on normal three address code representations.
1552 The void type does not represent any value and has no size.
1570 The function type can be thought of as a function signature. It consists of a
1571 return type and a list of formal parameter types. The return type of a function
1572 type is a void type or first class type --- except for :ref:`label <t_label>`
1573 and :ref:`metadata <t_metadata>` types.
1579 <returntype> (<parameter list>)
1581 ...where '``<parameter list>``' is a comma-separated list of type
1582 specifiers. Optionally, the parameter list may include a type ``...``, which
1583 indicates that the function takes a variable number of arguments. Variable
1584 argument functions can access their arguments with the :ref:`variable argument
1585 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1586 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1590 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1591 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1592 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1593 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1594 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1595 | ``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. |
1596 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1597 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1598 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1605 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1606 Values of these types are the only ones which can be produced by
1614 These are the types that are valid in registers from CodeGen's perspective.
1623 The integer type is a very simple type that simply specifies an
1624 arbitrary bit width for the integer type desired. Any bit width from 1
1625 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1633 The number of bits the integer will occupy is specified by the ``N``
1639 +----------------+------------------------------------------------+
1640 | ``i1`` | a single-bit integer. |
1641 +----------------+------------------------------------------------+
1642 | ``i32`` | a 32-bit integer. |
1643 +----------------+------------------------------------------------+
1644 | ``i1942652`` | a really big integer of over 1 million bits. |
1645 +----------------+------------------------------------------------+
1649 Floating Point Types
1650 """"""""""""""""""""
1659 - 16-bit floating point value
1662 - 32-bit floating point value
1665 - 64-bit floating point value
1668 - 128-bit floating point value (112-bit mantissa)
1671 - 80-bit floating point value (X87)
1674 - 128-bit floating point value (two 64-bits)
1683 The x86mmx type represents a value held in an MMX register on an x86
1684 machine. The operations allowed on it are quite limited: parameters and
1685 return values, load and store, and bitcast. User-specified MMX
1686 instructions are represented as intrinsic or asm calls with arguments
1687 and/or results of this type. There are no arrays, vectors or constants
1704 The pointer type is used to specify memory locations. Pointers are
1705 commonly used to reference objects in memory.
1707 Pointer types may have an optional address space attribute defining the
1708 numbered address space where the pointed-to object resides. The default
1709 address space is number zero. The semantics of non-zero address spaces
1710 are target-specific.
1712 Note that LLVM does not permit pointers to void (``void*``) nor does it
1713 permit pointers to labels (``label*``). Use ``i8*`` instead.
1723 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1724 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1725 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1726 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1727 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1728 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1729 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1738 A vector type is a simple derived type that represents a vector of
1739 elements. Vector types are used when multiple primitive data are
1740 operated in parallel using a single instruction (SIMD). A vector type
1741 requires a size (number of elements) and an underlying primitive data
1742 type. Vector types are considered :ref:`first class <t_firstclass>`.
1748 < <# elements> x <elementtype> >
1750 The number of elements is a constant integer value larger than 0;
1751 elementtype may be any integer or floating point type, or a pointer to
1752 these types. Vectors of size zero are not allowed.
1756 +-------------------+--------------------------------------------------+
1757 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1758 +-------------------+--------------------------------------------------+
1759 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1760 +-------------------+--------------------------------------------------+
1761 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1762 +-------------------+--------------------------------------------------+
1763 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1764 +-------------------+--------------------------------------------------+
1773 The label type represents code labels.
1788 The metadata type represents embedded metadata. No derived types may be
1789 created from metadata except for :ref:`function <t_function>` arguments.
1802 Aggregate Types are a subset of derived types that can contain multiple
1803 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1804 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1814 The array type is a very simple derived type that arranges elements
1815 sequentially in memory. The array type requires a size (number of
1816 elements) and an underlying data type.
1822 [<# elements> x <elementtype>]
1824 The number of elements is a constant integer value; ``elementtype`` may
1825 be any type with a size.
1829 +------------------+--------------------------------------+
1830 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1831 +------------------+--------------------------------------+
1832 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1833 +------------------+--------------------------------------+
1834 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1835 +------------------+--------------------------------------+
1837 Here are some examples of multidimensional arrays:
1839 +-----------------------------+----------------------------------------------------------+
1840 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1841 +-----------------------------+----------------------------------------------------------+
1842 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1843 +-----------------------------+----------------------------------------------------------+
1844 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1845 +-----------------------------+----------------------------------------------------------+
1847 There is no restriction on indexing beyond the end of the array implied
1848 by a static type (though there are restrictions on indexing beyond the
1849 bounds of an allocated object in some cases). This means that
1850 single-dimension 'variable sized array' addressing can be implemented in
1851 LLVM with a zero length array type. An implementation of 'pascal style
1852 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1862 The structure type is used to represent a collection of data members
1863 together in memory. The elements of a structure may be any type that has
1866 Structures in memory are accessed using '``load``' and '``store``' by
1867 getting a pointer to a field with the '``getelementptr``' instruction.
1868 Structures in registers are accessed using the '``extractvalue``' and
1869 '``insertvalue``' instructions.
1871 Structures may optionally be "packed" structures, which indicate that
1872 the alignment of the struct is one byte, and that there is no padding
1873 between the elements. In non-packed structs, padding between field types
1874 is inserted as defined by the DataLayout string in the module, which is
1875 required to match what the underlying code generator expects.
1877 Structures can either be "literal" or "identified". A literal structure
1878 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1879 identified types are always defined at the top level with a name.
1880 Literal types are uniqued by their contents and can never be recursive
1881 or opaque since there is no way to write one. Identified types can be
1882 recursive, can be opaqued, and are never uniqued.
1888 %T1 = type { <type list> } ; Identified normal struct type
1889 %T2 = type <{ <type list> }> ; Identified packed struct type
1893 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1894 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1895 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1896 | ``{ 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``. |
1897 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1898 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1899 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1903 Opaque Structure Types
1904 """"""""""""""""""""""
1908 Opaque structure types are used to represent named structure types that
1909 do not have a body specified. This corresponds (for example) to the C
1910 notion of a forward declared structure.
1921 +--------------+-------------------+
1922 | ``opaque`` | An opaque type. |
1923 +--------------+-------------------+
1928 LLVM has several different basic types of constants. This section
1929 describes them all and their syntax.
1934 **Boolean constants**
1935 The two strings '``true``' and '``false``' are both valid constants
1937 **Integer constants**
1938 Standard integers (such as '4') are constants of the
1939 :ref:`integer <t_integer>` type. Negative numbers may be used with
1941 **Floating point constants**
1942 Floating point constants use standard decimal notation (e.g.
1943 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1944 hexadecimal notation (see below). The assembler requires the exact
1945 decimal value of a floating-point constant. For example, the
1946 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1947 decimal in binary. Floating point constants must have a :ref:`floating
1948 point <t_floating>` type.
1949 **Null pointer constants**
1950 The identifier '``null``' is recognized as a null pointer constant
1951 and must be of :ref:`pointer type <t_pointer>`.
1953 The one non-intuitive notation for constants is the hexadecimal form of
1954 floating point constants. For example, the form
1955 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1956 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1957 constants are required (and the only time that they are generated by the
1958 disassembler) is when a floating point constant must be emitted but it
1959 cannot be represented as a decimal floating point number in a reasonable
1960 number of digits. For example, NaN's, infinities, and other special
1961 values are represented in their IEEE hexadecimal format so that assembly
1962 and disassembly do not cause any bits to change in the constants.
1964 When using the hexadecimal form, constants of types half, float, and
1965 double are represented using the 16-digit form shown above (which
1966 matches the IEEE754 representation for double); half and float values
1967 must, however, be exactly representable as IEEE 754 half and single
1968 precision, respectively. Hexadecimal format is always used for long
1969 double, and there are three forms of long double. The 80-bit format used
1970 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1971 128-bit format used by PowerPC (two adjacent doubles) is represented by
1972 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1973 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1974 will only work if they match the long double format on your target.
1975 The IEEE 16-bit format (half precision) is represented by ``0xH``
1976 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1977 (sign bit at the left).
1979 There are no constants of type x86mmx.
1981 .. _complexconstants:
1986 Complex constants are a (potentially recursive) combination of simple
1987 constants and smaller complex constants.
1989 **Structure constants**
1990 Structure constants are represented with notation similar to
1991 structure type definitions (a comma separated list of elements,
1992 surrounded by braces (``{}``)). For example:
1993 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1994 "``@G = external global i32``". Structure constants must have
1995 :ref:`structure type <t_struct>`, and the number and types of elements
1996 must match those specified by the type.
1998 Array constants are represented with notation similar to array type
1999 definitions (a comma separated list of elements, surrounded by
2000 square brackets (``[]``)). For example:
2001 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2002 :ref:`array type <t_array>`, and the number and types of elements must
2003 match those specified by the type.
2004 **Vector constants**
2005 Vector constants are represented with notation similar to vector
2006 type definitions (a comma separated list of elements, surrounded by
2007 less-than/greater-than's (``<>``)). For example:
2008 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2009 must have :ref:`vector type <t_vector>`, and the number and types of
2010 elements must match those specified by the type.
2011 **Zero initialization**
2012 The string '``zeroinitializer``' can be used to zero initialize a
2013 value to zero of *any* type, including scalar and
2014 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2015 having to print large zero initializers (e.g. for large arrays) and
2016 is always exactly equivalent to using explicit zero initializers.
2018 A metadata node is a structure-like constant with :ref:`metadata
2019 type <t_metadata>`. For example:
2020 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2021 constants that are meant to be interpreted as part of the
2022 instruction stream, metadata is a place to attach additional
2023 information such as debug info.
2025 Global Variable and Function Addresses
2026 --------------------------------------
2028 The addresses of :ref:`global variables <globalvars>` and
2029 :ref:`functions <functionstructure>` are always implicitly valid
2030 (link-time) constants. These constants are explicitly referenced when
2031 the :ref:`identifier for the global <identifiers>` is used and always have
2032 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2035 .. code-block:: llvm
2039 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2046 The string '``undef``' can be used anywhere a constant is expected, and
2047 indicates that the user of the value may receive an unspecified
2048 bit-pattern. Undefined values may be of any type (other than '``label``'
2049 or '``void``') and be used anywhere a constant is permitted.
2051 Undefined values are useful because they indicate to the compiler that
2052 the program is well defined no matter what value is used. This gives the
2053 compiler more freedom to optimize. Here are some examples of
2054 (potentially surprising) transformations that are valid (in pseudo IR):
2056 .. code-block:: llvm
2066 This is safe because all of the output bits are affected by the undef
2067 bits. Any output bit can have a zero or one depending on the input bits.
2069 .. code-block:: llvm
2080 These logical operations have bits that are not always affected by the
2081 input. For example, if ``%X`` has a zero bit, then the output of the
2082 '``and``' operation will always be a zero for that bit, no matter what
2083 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2084 optimize or assume that the result of the '``and``' is '``undef``'.
2085 However, it is safe to assume that all bits of the '``undef``' could be
2086 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2087 all the bits of the '``undef``' operand to the '``or``' could be set,
2088 allowing the '``or``' to be folded to -1.
2090 .. code-block:: llvm
2092 %A = select undef, %X, %Y
2093 %B = select undef, 42, %Y
2094 %C = select %X, %Y, undef
2104 This set of examples shows that undefined '``select``' (and conditional
2105 branch) conditions can go *either way*, but they have to come from one
2106 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2107 both known to have a clear low bit, then ``%A`` would have to have a
2108 cleared low bit. However, in the ``%C`` example, the optimizer is
2109 allowed to assume that the '``undef``' operand could be the same as
2110 ``%Y``, allowing the whole '``select``' to be eliminated.
2112 .. code-block:: llvm
2114 %A = xor undef, undef
2131 This example points out that two '``undef``' operands are not
2132 necessarily the same. This can be surprising to people (and also matches
2133 C semantics) where they assume that "``X^X``" is always zero, even if
2134 ``X`` is undefined. This isn't true for a number of reasons, but the
2135 short answer is that an '``undef``' "variable" can arbitrarily change
2136 its value over its "live range". This is true because the variable
2137 doesn't actually *have a live range*. Instead, the value is logically
2138 read from arbitrary registers that happen to be around when needed, so
2139 the value is not necessarily consistent over time. In fact, ``%A`` and
2140 ``%C`` need to have the same semantics or the core LLVM "replace all
2141 uses with" concept would not hold.
2143 .. code-block:: llvm
2151 These examples show the crucial difference between an *undefined value*
2152 and *undefined behavior*. An undefined value (like '``undef``') is
2153 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2154 operation can be constant folded to '``undef``', because the '``undef``'
2155 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2156 However, in the second example, we can make a more aggressive
2157 assumption: because the ``undef`` is allowed to be an arbitrary value,
2158 we are allowed to assume that it could be zero. Since a divide by zero
2159 has *undefined behavior*, we are allowed to assume that the operation
2160 does not execute at all. This allows us to delete the divide and all
2161 code after it. Because the undefined operation "can't happen", the
2162 optimizer can assume that it occurs in dead code.
2164 .. code-block:: llvm
2166 a: store undef -> %X
2167 b: store %X -> undef
2172 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2173 value can be assumed to not have any effect; we can assume that the
2174 value is overwritten with bits that happen to match what was already
2175 there. However, a store *to* an undefined location could clobber
2176 arbitrary memory, therefore, it has undefined behavior.
2183 Poison values are similar to :ref:`undef values <undefvalues>`, however
2184 they also represent the fact that an instruction or constant expression
2185 which cannot evoke side effects has nevertheless detected a condition
2186 which results in undefined behavior.
2188 There is currently no way of representing a poison value in the IR; they
2189 only exist when produced by operations such as :ref:`add <i_add>` with
2192 Poison value behavior is defined in terms of value *dependence*:
2194 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2195 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2196 their dynamic predecessor basic block.
2197 - Function arguments depend on the corresponding actual argument values
2198 in the dynamic callers of their functions.
2199 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2200 instructions that dynamically transfer control back to them.
2201 - :ref:`Invoke <i_invoke>` instructions depend on the
2202 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2203 call instructions that dynamically transfer control back to them.
2204 - Non-volatile loads and stores depend on the most recent stores to all
2205 of the referenced memory addresses, following the order in the IR
2206 (including loads and stores implied by intrinsics such as
2207 :ref:`@llvm.memcpy <int_memcpy>`.)
2208 - An instruction with externally visible side effects depends on the
2209 most recent preceding instruction with externally visible side
2210 effects, following the order in the IR. (This includes :ref:`volatile
2211 operations <volatile>`.)
2212 - An instruction *control-depends* on a :ref:`terminator
2213 instruction <terminators>` if the terminator instruction has
2214 multiple successors and the instruction is always executed when
2215 control transfers to one of the successors, and may not be executed
2216 when control is transferred to another.
2217 - Additionally, an instruction also *control-depends* on a terminator
2218 instruction if the set of instructions it otherwise depends on would
2219 be different if the terminator had transferred control to a different
2221 - Dependence is transitive.
2223 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2224 with the additional affect that any instruction which has a *dependence*
2225 on a poison value has undefined behavior.
2227 Here are some examples:
2229 .. code-block:: llvm
2232 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2233 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2234 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2235 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2237 store i32 %poison, i32* @g ; Poison value stored to memory.
2238 %poison2 = load i32* @g ; Poison value loaded back from memory.
2240 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2242 %narrowaddr = bitcast i32* @g to i16*
2243 %wideaddr = bitcast i32* @g to i64*
2244 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2245 %poison4 = load i64* %wideaddr ; Returns a poison value.
2247 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2248 br i1 %cmp, label %true, label %end ; Branch to either destination.
2251 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2252 ; it has undefined behavior.
2256 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2257 ; Both edges into this PHI are
2258 ; control-dependent on %cmp, so this
2259 ; always results in a poison value.
2261 store volatile i32 0, i32* @g ; This would depend on the store in %true
2262 ; if %cmp is true, or the store in %entry
2263 ; otherwise, so this is undefined behavior.
2265 br i1 %cmp, label %second_true, label %second_end
2266 ; The same branch again, but this time the
2267 ; true block doesn't have side effects.
2274 store volatile i32 0, i32* @g ; This time, the instruction always depends
2275 ; on the store in %end. Also, it is
2276 ; control-equivalent to %end, so this is
2277 ; well-defined (ignoring earlier undefined
2278 ; behavior in this example).
2282 Addresses of Basic Blocks
2283 -------------------------
2285 ``blockaddress(@function, %block)``
2287 The '``blockaddress``' constant computes the address of the specified
2288 basic block in the specified function, and always has an ``i8*`` type.
2289 Taking the address of the entry block is illegal.
2291 This value only has defined behavior when used as an operand to the
2292 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2293 against null. Pointer equality tests between labels addresses results in
2294 undefined behavior --- though, again, comparison against null is ok, and
2295 no label is equal to the null pointer. This may be passed around as an
2296 opaque pointer sized value as long as the bits are not inspected. This
2297 allows ``ptrtoint`` and arithmetic to be performed on these values so
2298 long as the original value is reconstituted before the ``indirectbr``
2301 Finally, some targets may provide defined semantics when using the value
2302 as the operand to an inline assembly, but that is target specific.
2306 Constant Expressions
2307 --------------------
2309 Constant expressions are used to allow expressions involving other
2310 constants to be used as constants. Constant expressions may be of any
2311 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2312 that does not have side effects (e.g. load and call are not supported).
2313 The following is the syntax for constant expressions:
2315 ``trunc (CST to TYPE)``
2316 Truncate a constant to another type. The bit size of CST must be
2317 larger than the bit size of TYPE. Both types must be integers.
2318 ``zext (CST to TYPE)``
2319 Zero extend a constant to another type. The bit size of CST must be
2320 smaller than the bit size of TYPE. Both types must be integers.
2321 ``sext (CST to TYPE)``
2322 Sign extend a constant to another type. The bit size of CST must be
2323 smaller than the bit size of TYPE. Both types must be integers.
2324 ``fptrunc (CST to TYPE)``
2325 Truncate a floating point constant to another floating point type.
2326 The size of CST must be larger than the size of TYPE. Both types
2327 must be floating point.
2328 ``fpext (CST to TYPE)``
2329 Floating point extend a constant to another type. The size of CST
2330 must be smaller or equal to the size of TYPE. Both types must be
2332 ``fptoui (CST to TYPE)``
2333 Convert a floating point constant to the corresponding unsigned
2334 integer constant. TYPE must be a scalar or vector integer type. CST
2335 must be of scalar or vector floating point type. Both CST and TYPE
2336 must be scalars, or vectors of the same number of elements. If the
2337 value won't fit in the integer type, the results are undefined.
2338 ``fptosi (CST to TYPE)``
2339 Convert a floating point constant to the corresponding signed
2340 integer constant. TYPE must be a scalar or vector integer type. CST
2341 must be of scalar or vector floating point type. Both CST and TYPE
2342 must be scalars, or vectors of the same number of elements. If the
2343 value won't fit in the integer type, the results are undefined.
2344 ``uitofp (CST to TYPE)``
2345 Convert an unsigned integer constant to the corresponding floating
2346 point constant. TYPE must be a scalar or vector floating point type.
2347 CST must be of scalar or vector integer type. Both CST and TYPE must
2348 be scalars, or vectors of the same number of elements. If the value
2349 won't fit in the floating point type, the results are undefined.
2350 ``sitofp (CST to TYPE)``
2351 Convert a signed integer constant to the corresponding floating
2352 point constant. TYPE must be a scalar or vector floating point type.
2353 CST must be of scalar or vector integer type. Both CST and TYPE must
2354 be scalars, or vectors of the same number of elements. If the value
2355 won't fit in the floating point type, the results are undefined.
2356 ``ptrtoint (CST to TYPE)``
2357 Convert a pointer typed constant to the corresponding integer
2358 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2359 pointer type. The ``CST`` value is zero extended, truncated, or
2360 unchanged to make it fit in ``TYPE``.
2361 ``inttoptr (CST to TYPE)``
2362 Convert an integer constant to a pointer constant. TYPE must be a
2363 pointer type. CST must be of integer type. The CST value is zero
2364 extended, truncated, or unchanged to make it fit in a pointer size.
2365 This one is *really* dangerous!
2366 ``bitcast (CST to TYPE)``
2367 Convert a constant, CST, to another TYPE. The constraints of the
2368 operands are the same as those for the :ref:`bitcast
2369 instruction <i_bitcast>`.
2370 ``addrspacecast (CST to TYPE)``
2371 Convert a constant pointer or constant vector of pointer, CST, to another
2372 TYPE in a different address space. The constraints of the operands are the
2373 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2374 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2375 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2376 constants. As with the :ref:`getelementptr <i_getelementptr>`
2377 instruction, the index list may have zero or more indexes, which are
2378 required to make sense for the type of "CSTPTR".
2379 ``select (COND, VAL1, VAL2)``
2380 Perform the :ref:`select operation <i_select>` on constants.
2381 ``icmp COND (VAL1, VAL2)``
2382 Performs the :ref:`icmp operation <i_icmp>` on constants.
2383 ``fcmp COND (VAL1, VAL2)``
2384 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2385 ``extractelement (VAL, IDX)``
2386 Perform the :ref:`extractelement operation <i_extractelement>` on
2388 ``insertelement (VAL, ELT, IDX)``
2389 Perform the :ref:`insertelement operation <i_insertelement>` on
2391 ``shufflevector (VEC1, VEC2, IDXMASK)``
2392 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2394 ``extractvalue (VAL, IDX0, IDX1, ...)``
2395 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2396 constants. The index list is interpreted in a similar manner as
2397 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2398 least one index value must be specified.
2399 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2400 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2401 The index list is interpreted in a similar manner as indices in a
2402 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2403 value must be specified.
2404 ``OPCODE (LHS, RHS)``
2405 Perform the specified operation of the LHS and RHS constants. OPCODE
2406 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2407 binary <bitwiseops>` operations. The constraints on operands are
2408 the same as those for the corresponding instruction (e.g. no bitwise
2409 operations on floating point values are allowed).
2416 Inline Assembler Expressions
2417 ----------------------------
2419 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2420 Inline Assembly <moduleasm>`) through the use of a special value. This
2421 value represents the inline assembler as a string (containing the
2422 instructions to emit), a list of operand constraints (stored as a
2423 string), a flag that indicates whether or not the inline asm expression
2424 has side effects, and a flag indicating whether the function containing
2425 the asm needs to align its stack conservatively. An example inline
2426 assembler expression is:
2428 .. code-block:: llvm
2430 i32 (i32) asm "bswap $0", "=r,r"
2432 Inline assembler expressions may **only** be used as the callee operand
2433 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2434 Thus, typically we have:
2436 .. code-block:: llvm
2438 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2440 Inline asms with side effects not visible in the constraint list must be
2441 marked as having side effects. This is done through the use of the
2442 '``sideeffect``' keyword, like so:
2444 .. code-block:: llvm
2446 call void asm sideeffect "eieio", ""()
2448 In some cases inline asms will contain code that will not work unless
2449 the stack is aligned in some way, such as calls or SSE instructions on
2450 x86, yet will not contain code that does that alignment within the asm.
2451 The compiler should make conservative assumptions about what the asm
2452 might contain and should generate its usual stack alignment code in the
2453 prologue if the '``alignstack``' keyword is present:
2455 .. code-block:: llvm
2457 call void asm alignstack "eieio", ""()
2459 Inline asms also support using non-standard assembly dialects. The
2460 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2461 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2462 the only supported dialects. An example is:
2464 .. code-block:: llvm
2466 call void asm inteldialect "eieio", ""()
2468 If multiple keywords appear the '``sideeffect``' keyword must come
2469 first, the '``alignstack``' keyword second and the '``inteldialect``'
2475 The call instructions that wrap inline asm nodes may have a
2476 "``!srcloc``" MDNode attached to it that contains a list of constant
2477 integers. If present, the code generator will use the integer as the
2478 location cookie value when report errors through the ``LLVMContext``
2479 error reporting mechanisms. This allows a front-end to correlate backend
2480 errors that occur with inline asm back to the source code that produced
2483 .. code-block:: llvm
2485 call void asm sideeffect "something bad", ""(), !srcloc !42
2487 !42 = !{ i32 1234567 }
2489 It is up to the front-end to make sense of the magic numbers it places
2490 in the IR. If the MDNode contains multiple constants, the code generator
2491 will use the one that corresponds to the line of the asm that the error
2496 Metadata Nodes and Metadata Strings
2497 -----------------------------------
2499 LLVM IR allows metadata to be attached to instructions in the program
2500 that can convey extra information about the code to the optimizers and
2501 code generator. One example application of metadata is source-level
2502 debug information. There are two metadata primitives: strings and nodes.
2503 All metadata has the ``metadata`` type and is identified in syntax by a
2504 preceding exclamation point ('``!``').
2506 A metadata string is a string surrounded by double quotes. It can
2507 contain any character by escaping non-printable characters with
2508 "``\xx``" where "``xx``" is the two digit hex code. For example:
2511 Metadata nodes are represented with notation similar to structure
2512 constants (a comma separated list of elements, surrounded by braces and
2513 preceded by an exclamation point). Metadata nodes can have any values as
2514 their operand. For example:
2516 .. code-block:: llvm
2518 !{ metadata !"test\00", i32 10}
2520 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2521 metadata nodes, which can be looked up in the module symbol table. For
2524 .. code-block:: llvm
2526 !foo = metadata !{!4, !3}
2528 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2529 function is using two metadata arguments:
2531 .. code-block:: llvm
2533 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2535 Metadata can be attached with an instruction. Here metadata ``!21`` is
2536 attached to the ``add`` instruction using the ``!dbg`` identifier:
2538 .. code-block:: llvm
2540 %indvar.next = add i64 %indvar, 1, !dbg !21
2542 More information about specific metadata nodes recognized by the
2543 optimizers and code generator is found below.
2548 In LLVM IR, memory does not have types, so LLVM's own type system is not
2549 suitable for doing TBAA. Instead, metadata is added to the IR to
2550 describe a type system of a higher level language. This can be used to
2551 implement typical C/C++ TBAA, but it can also be used to implement
2552 custom alias analysis behavior for other languages.
2554 The current metadata format is very simple. TBAA metadata nodes have up
2555 to three fields, e.g.:
2557 .. code-block:: llvm
2559 !0 = metadata !{ metadata !"an example type tree" }
2560 !1 = metadata !{ metadata !"int", metadata !0 }
2561 !2 = metadata !{ metadata !"float", metadata !0 }
2562 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2564 The first field is an identity field. It can be any value, usually a
2565 metadata string, which uniquely identifies the type. The most important
2566 name in the tree is the name of the root node. Two trees with different
2567 root node names are entirely disjoint, even if they have leaves with
2570 The second field identifies the type's parent node in the tree, or is
2571 null or omitted for a root node. A type is considered to alias all of
2572 its descendants and all of its ancestors in the tree. Also, a type is
2573 considered to alias all types in other trees, so that bitcode produced
2574 from multiple front-ends is handled conservatively.
2576 If the third field is present, it's an integer which if equal to 1
2577 indicates that the type is "constant" (meaning
2578 ``pointsToConstantMemory`` should return true; see `other useful
2579 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2581 '``tbaa.struct``' Metadata
2582 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2584 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2585 aggregate assignment operations in C and similar languages, however it
2586 is defined to copy a contiguous region of memory, which is more than
2587 strictly necessary for aggregate types which contain holes due to
2588 padding. Also, it doesn't contain any TBAA information about the fields
2591 ``!tbaa.struct`` metadata can describe which memory subregions in a
2592 memcpy are padding and what the TBAA tags of the struct are.
2594 The current metadata format is very simple. ``!tbaa.struct`` metadata
2595 nodes are a list of operands which are in conceptual groups of three.
2596 For each group of three, the first operand gives the byte offset of a
2597 field in bytes, the second gives its size in bytes, and the third gives
2600 .. code-block:: llvm
2602 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2604 This describes a struct with two fields. The first is at offset 0 bytes
2605 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2606 and has size 4 bytes and has tbaa tag !2.
2608 Note that the fields need not be contiguous. In this example, there is a
2609 4 byte gap between the two fields. This gap represents padding which
2610 does not carry useful data and need not be preserved.
2612 '``fpmath``' Metadata
2613 ^^^^^^^^^^^^^^^^^^^^^
2615 ``fpmath`` metadata may be attached to any instruction of floating point
2616 type. It can be used to express the maximum acceptable error in the
2617 result of that instruction, in ULPs, thus potentially allowing the
2618 compiler to use a more efficient but less accurate method of computing
2619 it. ULP is defined as follows:
2621 If ``x`` is a real number that lies between two finite consecutive
2622 floating-point numbers ``a`` and ``b``, without being equal to one
2623 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2624 distance between the two non-equal finite floating-point numbers
2625 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2627 The metadata node shall consist of a single positive floating point
2628 number representing the maximum relative error, for example:
2630 .. code-block:: llvm
2632 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2634 '``range``' Metadata
2635 ^^^^^^^^^^^^^^^^^^^^
2637 ``range`` metadata may be attached only to loads of integer types. It
2638 expresses the possible ranges the loaded value is in. The ranges are
2639 represented with a flattened list of integers. The loaded value is known
2640 to be in the union of the ranges defined by each consecutive pair. Each
2641 pair has the following properties:
2643 - The type must match the type loaded by the instruction.
2644 - The pair ``a,b`` represents the range ``[a,b)``.
2645 - Both ``a`` and ``b`` are constants.
2646 - The range is allowed to wrap.
2647 - The range should not represent the full or empty set. That is,
2650 In addition, the pairs must be in signed order of the lower bound and
2651 they must be non-contiguous.
2655 .. code-block:: llvm
2657 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2658 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2659 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2660 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2662 !0 = metadata !{ i8 0, i8 2 }
2663 !1 = metadata !{ i8 255, i8 2 }
2664 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2665 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2670 It is sometimes useful to attach information to loop constructs. Currently,
2671 loop metadata is implemented as metadata attached to the branch instruction
2672 in the loop latch block. This type of metadata refer to a metadata node that is
2673 guaranteed to be separate for each loop. The loop identifier metadata is
2674 specified with the name ``llvm.loop``.
2676 The loop identifier metadata is implemented using a metadata that refers to
2677 itself to avoid merging it with any other identifier metadata, e.g.,
2678 during module linkage or function inlining. That is, each loop should refer
2679 to their own identification metadata even if they reside in separate functions.
2680 The following example contains loop identifier metadata for two separate loop
2683 .. code-block:: llvm
2685 !0 = metadata !{ metadata !0 }
2686 !1 = metadata !{ metadata !1 }
2688 The loop identifier metadata can be used to specify additional per-loop
2689 metadata. Any operands after the first operand can be treated as user-defined
2690 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2691 by the loop vectorizer to indicate how many times to unroll the loop:
2693 .. code-block:: llvm
2695 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2697 !0 = metadata !{ metadata !0, metadata !1 }
2698 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2703 Metadata types used to annotate memory accesses with information helpful
2704 for optimizations are prefixed with ``llvm.mem``.
2706 '``llvm.mem.parallel_loop_access``' Metadata
2707 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2709 For a loop to be parallel, in addition to using
2710 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2711 also all of the memory accessing instructions in the loop body need to be
2712 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2713 is at least one memory accessing instruction not marked with the metadata,
2714 the loop must be considered a sequential loop. This causes parallel loops to be
2715 converted to sequential loops due to optimization passes that are unaware of
2716 the parallel semantics and that insert new memory instructions to the loop
2719 Example of a loop that is considered parallel due to its correct use of
2720 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2721 metadata types that refer to the same loop identifier metadata.
2723 .. code-block:: llvm
2727 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2729 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2731 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2735 !0 = metadata !{ metadata !0 }
2737 It is also possible to have nested parallel loops. In that case the
2738 memory accesses refer to a list of loop identifier metadata nodes instead of
2739 the loop identifier metadata node directly:
2741 .. code-block:: llvm
2748 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2750 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2752 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2756 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2758 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2760 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2762 outer.for.end: ; preds = %for.body
2764 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2765 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2766 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2768 '``llvm.vectorizer``'
2769 ^^^^^^^^^^^^^^^^^^^^^
2771 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2772 vectorization parameters such as vectorization factor and unroll factor.
2774 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2775 loop identification metadata.
2777 '``llvm.vectorizer.unroll``' Metadata
2778 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2780 This metadata instructs the loop vectorizer to unroll the specified
2781 loop exactly ``N`` times.
2783 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2784 operand is an integer specifying the unroll factor. For example:
2786 .. code-block:: llvm
2788 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2790 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2793 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2794 determined automatically.
2796 '``llvm.vectorizer.width``' Metadata
2797 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2799 This metadata sets the target width of the vectorizer to ``N``. Without
2800 this metadata, the vectorizer will choose a width automatically.
2801 Regardless of this metadata, the vectorizer will only vectorize loops if
2802 it believes it is valid to do so.
2804 The first operand is the string ``llvm.vectorizer.width`` and the second
2805 operand is an integer specifying the width. For example:
2807 .. code-block:: llvm
2809 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2811 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2814 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2817 Module Flags Metadata
2818 =====================
2820 Information about the module as a whole is difficult to convey to LLVM's
2821 subsystems. The LLVM IR isn't sufficient to transmit this information.
2822 The ``llvm.module.flags`` named metadata exists in order to facilitate
2823 this. These flags are in the form of key / value pairs --- much like a
2824 dictionary --- making it easy for any subsystem who cares about a flag to
2827 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2828 Each triplet has the following form:
2830 - The first element is a *behavior* flag, which specifies the behavior
2831 when two (or more) modules are merged together, and it encounters two
2832 (or more) metadata with the same ID. The supported behaviors are
2834 - The second element is a metadata string that is a unique ID for the
2835 metadata. Each module may only have one flag entry for each unique ID (not
2836 including entries with the **Require** behavior).
2837 - The third element is the value of the flag.
2839 When two (or more) modules are merged together, the resulting
2840 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2841 each unique metadata ID string, there will be exactly one entry in the merged
2842 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2843 be determined by the merge behavior flag, as described below. The only exception
2844 is that entries with the *Require* behavior are always preserved.
2846 The following behaviors are supported:
2857 Emits an error if two values disagree, otherwise the resulting value
2858 is that of the operands.
2862 Emits a warning if two values disagree. The result value will be the
2863 operand for the flag from the first module being linked.
2867 Adds a requirement that another module flag be present and have a
2868 specified value after linking is performed. The value must be a
2869 metadata pair, where the first element of the pair is the ID of the
2870 module flag to be restricted, and the second element of the pair is
2871 the value the module flag should be restricted to. This behavior can
2872 be used to restrict the allowable results (via triggering of an
2873 error) of linking IDs with the **Override** behavior.
2877 Uses the specified value, regardless of the behavior or value of the
2878 other module. If both modules specify **Override**, but the values
2879 differ, an error will be emitted.
2883 Appends the two values, which are required to be metadata nodes.
2887 Appends the two values, which are required to be metadata
2888 nodes. However, duplicate entries in the second list are dropped
2889 during the append operation.
2891 It is an error for a particular unique flag ID to have multiple behaviors,
2892 except in the case of **Require** (which adds restrictions on another metadata
2893 value) or **Override**.
2895 An example of module flags:
2897 .. code-block:: llvm
2899 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2900 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2901 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2902 !3 = metadata !{ i32 3, metadata !"qux",
2904 metadata !"foo", i32 1
2907 !llvm.module.flags = !{ !0, !1, !2, !3 }
2909 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2910 if two or more ``!"foo"`` flags are seen is to emit an error if their
2911 values are not equal.
2913 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2914 behavior if two or more ``!"bar"`` flags are seen is to use the value
2917 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2918 behavior if two or more ``!"qux"`` flags are seen is to emit a
2919 warning if their values are not equal.
2921 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2925 metadata !{ metadata !"foo", i32 1 }
2927 The behavior is to emit an error if the ``llvm.module.flags`` does not
2928 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2931 Objective-C Garbage Collection Module Flags Metadata
2932 ----------------------------------------------------
2934 On the Mach-O platform, Objective-C stores metadata about garbage
2935 collection in a special section called "image info". The metadata
2936 consists of a version number and a bitmask specifying what types of
2937 garbage collection are supported (if any) by the file. If two or more
2938 modules are linked together their garbage collection metadata needs to
2939 be merged rather than appended together.
2941 The Objective-C garbage collection module flags metadata consists of the
2942 following key-value pairs:
2951 * - ``Objective-C Version``
2952 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2954 * - ``Objective-C Image Info Version``
2955 - **[Required]** --- The version of the image info section. Currently
2958 * - ``Objective-C Image Info Section``
2959 - **[Required]** --- The section to place the metadata. Valid values are
2960 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2961 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2962 Objective-C ABI version 2.
2964 * - ``Objective-C Garbage Collection``
2965 - **[Required]** --- Specifies whether garbage collection is supported or
2966 not. Valid values are 0, for no garbage collection, and 2, for garbage
2967 collection supported.
2969 * - ``Objective-C GC Only``
2970 - **[Optional]** --- Specifies that only garbage collection is supported.
2971 If present, its value must be 6. This flag requires that the
2972 ``Objective-C Garbage Collection`` flag have the value 2.
2974 Some important flag interactions:
2976 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2977 merged with a module with ``Objective-C Garbage Collection`` set to
2978 2, then the resulting module has the
2979 ``Objective-C Garbage Collection`` flag set to 0.
2980 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2981 merged with a module with ``Objective-C GC Only`` set to 6.
2983 Automatic Linker Flags Module Flags Metadata
2984 --------------------------------------------
2986 Some targets support embedding flags to the linker inside individual object
2987 files. Typically this is used in conjunction with language extensions which
2988 allow source files to explicitly declare the libraries they depend on, and have
2989 these automatically be transmitted to the linker via object files.
2991 These flags are encoded in the IR using metadata in the module flags section,
2992 using the ``Linker Options`` key. The merge behavior for this flag is required
2993 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2994 node which should be a list of other metadata nodes, each of which should be a
2995 list of metadata strings defining linker options.
2997 For example, the following metadata section specifies two separate sets of
2998 linker options, presumably to link against ``libz`` and the ``Cocoa``
3001 !0 = metadata !{ i32 6, metadata !"Linker Options",
3003 metadata !{ metadata !"-lz" },
3004 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3005 !llvm.module.flags = !{ !0 }
3007 The metadata encoding as lists of lists of options, as opposed to a collapsed
3008 list of options, is chosen so that the IR encoding can use multiple option
3009 strings to specify e.g., a single library, while still having that specifier be
3010 preserved as an atomic element that can be recognized by a target specific
3011 assembly writer or object file emitter.
3013 Each individual option is required to be either a valid option for the target's
3014 linker, or an option that is reserved by the target specific assembly writer or
3015 object file emitter. No other aspect of these options is defined by the IR.
3017 .. _intrinsicglobalvariables:
3019 Intrinsic Global Variables
3020 ==========================
3022 LLVM has a number of "magic" global variables that contain data that
3023 affect code generation or other IR semantics. These are documented here.
3024 All globals of this sort should have a section specified as
3025 "``llvm.metadata``". This section and all globals that start with
3026 "``llvm.``" are reserved for use by LLVM.
3030 The '``llvm.used``' Global Variable
3031 -----------------------------------
3033 The ``@llvm.used`` global is an array which has
3034 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3035 pointers to named global variables, functions and aliases which may optionally
3036 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3039 .. code-block:: llvm
3044 @llvm.used = appending global [2 x i8*] [
3046 i8* bitcast (i32* @Y to i8*)
3047 ], section "llvm.metadata"
3049 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3050 and linker are required to treat the symbol as if there is a reference to the
3051 symbol that it cannot see (which is why they have to be named). For example, if
3052 a variable has internal linkage and no references other than that from the
3053 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3054 references from inline asms and other things the compiler cannot "see", and
3055 corresponds to "``attribute((used))``" in GNU C.
3057 On some targets, the code generator must emit a directive to the
3058 assembler or object file to prevent the assembler and linker from
3059 molesting the symbol.
3061 .. _gv_llvmcompilerused:
3063 The '``llvm.compiler.used``' Global Variable
3064 --------------------------------------------
3066 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3067 directive, except that it only prevents the compiler from touching the
3068 symbol. On targets that support it, this allows an intelligent linker to
3069 optimize references to the symbol without being impeded as it would be
3072 This is a rare construct that should only be used in rare circumstances,
3073 and should not be exposed to source languages.
3075 .. _gv_llvmglobalctors:
3077 The '``llvm.global_ctors``' Global Variable
3078 -------------------------------------------
3080 .. code-block:: llvm
3082 %0 = type { i32, void ()* }
3083 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3085 The ``@llvm.global_ctors`` array contains a list of constructor
3086 functions and associated priorities. The functions referenced by this
3087 array will be called in ascending order of priority (i.e. lowest first)
3088 when the module is loaded. The order of functions with the same priority
3091 .. _llvmglobaldtors:
3093 The '``llvm.global_dtors``' Global Variable
3094 -------------------------------------------
3096 .. code-block:: llvm
3098 %0 = type { i32, void ()* }
3099 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3101 The ``@llvm.global_dtors`` array contains a list of destructor functions
3102 and associated priorities. The functions referenced by this array will
3103 be called in descending order of priority (i.e. highest first) when the
3104 module is loaded. The order of functions with the same priority is not
3107 Instruction Reference
3108 =====================
3110 The LLVM instruction set consists of several different classifications
3111 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3112 instructions <binaryops>`, :ref:`bitwise binary
3113 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3114 :ref:`other instructions <otherops>`.
3118 Terminator Instructions
3119 -----------------------
3121 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3122 program ends with a "Terminator" instruction, which indicates which
3123 block should be executed after the current block is finished. These
3124 terminator instructions typically yield a '``void``' value: they produce
3125 control flow, not values (the one exception being the
3126 ':ref:`invoke <i_invoke>`' instruction).
3128 The terminator instructions are: ':ref:`ret <i_ret>`',
3129 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3130 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3131 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3135 '``ret``' Instruction
3136 ^^^^^^^^^^^^^^^^^^^^^
3143 ret <type> <value> ; Return a value from a non-void function
3144 ret void ; Return from void function
3149 The '``ret``' instruction is used to return control flow (and optionally
3150 a value) from a function back to the caller.
3152 There are two forms of the '``ret``' instruction: one that returns a
3153 value and then causes control flow, and one that just causes control
3159 The '``ret``' instruction optionally accepts a single argument, the
3160 return value. The type of the return value must be a ':ref:`first
3161 class <t_firstclass>`' type.
3163 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3164 return type and contains a '``ret``' instruction with no return value or
3165 a return value with a type that does not match its type, or if it has a
3166 void return type and contains a '``ret``' instruction with a return
3172 When the '``ret``' instruction is executed, control flow returns back to
3173 the calling function's context. If the caller is a
3174 ":ref:`call <i_call>`" instruction, execution continues at the
3175 instruction after the call. If the caller was an
3176 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3177 beginning of the "normal" destination block. If the instruction returns
3178 a value, that value shall set the call or invoke instruction's return
3184 .. code-block:: llvm
3186 ret i32 5 ; Return an integer value of 5
3187 ret void ; Return from a void function
3188 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3192 '``br``' Instruction
3193 ^^^^^^^^^^^^^^^^^^^^
3200 br i1 <cond>, label <iftrue>, label <iffalse>
3201 br label <dest> ; Unconditional branch
3206 The '``br``' instruction is used to cause control flow to transfer to a
3207 different basic block in the current function. There are two forms of
3208 this instruction, corresponding to a conditional branch and an
3209 unconditional branch.
3214 The conditional branch form of the '``br``' instruction takes a single
3215 '``i1``' value and two '``label``' values. The unconditional form of the
3216 '``br``' instruction takes a single '``label``' value as a target.
3221 Upon execution of a conditional '``br``' instruction, the '``i1``'
3222 argument is evaluated. If the value is ``true``, control flows to the
3223 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3224 to the '``iffalse``' ``label`` argument.
3229 .. code-block:: llvm
3232 %cond = icmp eq i32 %a, %b
3233 br i1 %cond, label %IfEqual, label %IfUnequal
3241 '``switch``' Instruction
3242 ^^^^^^^^^^^^^^^^^^^^^^^^
3249 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3254 The '``switch``' instruction is used to transfer control flow to one of
3255 several different places. It is a generalization of the '``br``'
3256 instruction, allowing a branch to occur to one of many possible
3262 The '``switch``' instruction uses three parameters: an integer
3263 comparison value '``value``', a default '``label``' destination, and an
3264 array of pairs of comparison value constants and '``label``'s. The table
3265 is not allowed to contain duplicate constant entries.
3270 The ``switch`` instruction specifies a table of values and destinations.
3271 When the '``switch``' instruction is executed, this table is searched
3272 for the given value. If the value is found, control flow is transferred
3273 to the corresponding destination; otherwise, control flow is transferred
3274 to the default destination.
3279 Depending on properties of the target machine and the particular
3280 ``switch`` instruction, this instruction may be code generated in
3281 different ways. For example, it could be generated as a series of
3282 chained conditional branches or with a lookup table.
3287 .. code-block:: llvm
3289 ; Emulate a conditional br instruction
3290 %Val = zext i1 %value to i32
3291 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3293 ; Emulate an unconditional br instruction
3294 switch i32 0, label %dest [ ]
3296 ; Implement a jump table:
3297 switch i32 %val, label %otherwise [ i32 0, label %onzero
3299 i32 2, label %ontwo ]
3303 '``indirectbr``' Instruction
3304 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3311 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3316 The '``indirectbr``' instruction implements an indirect branch to a
3317 label within the current function, whose address is specified by
3318 "``address``". Address must be derived from a
3319 :ref:`blockaddress <blockaddress>` constant.
3324 The '``address``' argument is the address of the label to jump to. The
3325 rest of the arguments indicate the full set of possible destinations
3326 that the address may point to. Blocks are allowed to occur multiple
3327 times in the destination list, though this isn't particularly useful.
3329 This destination list is required so that dataflow analysis has an
3330 accurate understanding of the CFG.
3335 Control transfers to the block specified in the address argument. All
3336 possible destination blocks must be listed in the label list, otherwise
3337 this instruction has undefined behavior. This implies that jumps to
3338 labels defined in other functions have undefined behavior as well.
3343 This is typically implemented with a jump through a register.
3348 .. code-block:: llvm
3350 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3354 '``invoke``' Instruction
3355 ^^^^^^^^^^^^^^^^^^^^^^^^
3362 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3363 to label <normal label> unwind label <exception label>
3368 The '``invoke``' instruction causes control to transfer to a specified
3369 function, with the possibility of control flow transfer to either the
3370 '``normal``' label or the '``exception``' label. If the callee function
3371 returns with the "``ret``" instruction, control flow will return to the
3372 "normal" label. If the callee (or any indirect callees) returns via the
3373 ":ref:`resume <i_resume>`" instruction or other exception handling
3374 mechanism, control is interrupted and continued at the dynamically
3375 nearest "exception" label.
3377 The '``exception``' label is a `landing
3378 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3379 '``exception``' label is required to have the
3380 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3381 information about the behavior of the program after unwinding happens,
3382 as its first non-PHI instruction. The restrictions on the
3383 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3384 instruction, so that the important information contained within the
3385 "``landingpad``" instruction can't be lost through normal code motion.
3390 This instruction requires several arguments:
3392 #. The optional "cconv" marker indicates which :ref:`calling
3393 convention <callingconv>` the call should use. If none is
3394 specified, the call defaults to using C calling conventions.
3395 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3396 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3398 #. '``ptr to function ty``': shall be the signature of the pointer to
3399 function value being invoked. In most cases, this is a direct
3400 function invocation, but indirect ``invoke``'s are just as possible,
3401 branching off an arbitrary pointer to function value.
3402 #. '``function ptr val``': An LLVM value containing a pointer to a
3403 function to be invoked.
3404 #. '``function args``': argument list whose types match the function
3405 signature argument types and parameter attributes. All arguments must
3406 be of :ref:`first class <t_firstclass>` type. If the function signature
3407 indicates the function accepts a variable number of arguments, the
3408 extra arguments can be specified.
3409 #. '``normal label``': the label reached when the called function
3410 executes a '``ret``' instruction.
3411 #. '``exception label``': the label reached when a callee returns via
3412 the :ref:`resume <i_resume>` instruction or other exception handling
3414 #. The optional :ref:`function attributes <fnattrs>` list. Only
3415 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3416 attributes are valid here.
3421 This instruction is designed to operate as a standard '``call``'
3422 instruction in most regards. The primary difference is that it
3423 establishes an association with a label, which is used by the runtime
3424 library to unwind the stack.
3426 This instruction is used in languages with destructors to ensure that
3427 proper cleanup is performed in the case of either a ``longjmp`` or a
3428 thrown exception. Additionally, this is important for implementation of
3429 '``catch``' clauses in high-level languages that support them.
3431 For the purposes of the SSA form, the definition of the value returned
3432 by the '``invoke``' instruction is deemed to occur on the edge from the
3433 current block to the "normal" label. If the callee unwinds then no
3434 return value is available.
3439 .. code-block:: llvm
3441 %retval = invoke i32 @Test(i32 15) to label %Continue
3442 unwind label %TestCleanup ; {i32}:retval set
3443 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3444 unwind label %TestCleanup ; {i32}:retval set
3448 '``resume``' Instruction
3449 ^^^^^^^^^^^^^^^^^^^^^^^^
3456 resume <type> <value>
3461 The '``resume``' instruction is a terminator instruction that has no
3467 The '``resume``' instruction requires one argument, which must have the
3468 same type as the result of any '``landingpad``' instruction in the same
3474 The '``resume``' instruction resumes propagation of an existing
3475 (in-flight) exception whose unwinding was interrupted with a
3476 :ref:`landingpad <i_landingpad>` instruction.
3481 .. code-block:: llvm
3483 resume { i8*, i32 } %exn
3487 '``unreachable``' Instruction
3488 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3500 The '``unreachable``' instruction has no defined semantics. This
3501 instruction is used to inform the optimizer that a particular portion of
3502 the code is not reachable. This can be used to indicate that the code
3503 after a no-return function cannot be reached, and other facts.
3508 The '``unreachable``' instruction has no defined semantics.
3515 Binary operators are used to do most of the computation in a program.
3516 They require two operands of the same type, execute an operation on
3517 them, and produce a single value. The operands might represent multiple
3518 data, as is the case with the :ref:`vector <t_vector>` data type. The
3519 result value has the same type as its operands.
3521 There are several different binary operators:
3525 '``add``' Instruction
3526 ^^^^^^^^^^^^^^^^^^^^^
3533 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3534 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3535 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3536 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3541 The '``add``' instruction returns the sum of its two operands.
3546 The two arguments to the '``add``' instruction must be
3547 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3548 arguments must have identical types.
3553 The value produced is the integer sum of the two operands.
3555 If the sum has unsigned overflow, the result returned is the
3556 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3559 Because LLVM integers use a two's complement representation, this
3560 instruction is appropriate for both signed and unsigned integers.
3562 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3563 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3564 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3565 unsigned and/or signed overflow, respectively, occurs.
3570 .. code-block:: llvm
3572 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3576 '``fadd``' Instruction
3577 ^^^^^^^^^^^^^^^^^^^^^^
3584 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3589 The '``fadd``' instruction returns the sum of its two operands.
3594 The two arguments to the '``fadd``' instruction must be :ref:`floating
3595 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3596 Both arguments must have identical types.
3601 The value produced is the floating point sum of the two operands. This
3602 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3603 which are optimization hints to enable otherwise unsafe floating point
3609 .. code-block:: llvm
3611 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3613 '``sub``' Instruction
3614 ^^^^^^^^^^^^^^^^^^^^^
3621 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3622 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3623 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3624 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3629 The '``sub``' instruction returns the difference of its two operands.
3631 Note that the '``sub``' instruction is used to represent the '``neg``'
3632 instruction present in most other intermediate representations.
3637 The two arguments to the '``sub``' instruction must be
3638 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3639 arguments must have identical types.
3644 The value produced is the integer difference of the two operands.
3646 If the difference has unsigned overflow, the result returned is the
3647 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3650 Because LLVM integers use a two's complement representation, this
3651 instruction is appropriate for both signed and unsigned integers.
3653 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3654 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3655 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3656 unsigned and/or signed overflow, respectively, occurs.
3661 .. code-block:: llvm
3663 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3664 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3668 '``fsub``' Instruction
3669 ^^^^^^^^^^^^^^^^^^^^^^
3676 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3681 The '``fsub``' instruction returns the difference of its two operands.
3683 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3684 instruction present in most other intermediate representations.
3689 The two arguments to the '``fsub``' instruction must be :ref:`floating
3690 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3691 Both arguments must have identical types.
3696 The value produced is the floating point difference of the two operands.
3697 This instruction can also take any number of :ref:`fast-math
3698 flags <fastmath>`, which are optimization hints to enable otherwise
3699 unsafe floating point optimizations:
3704 .. code-block:: llvm
3706 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3707 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3709 '``mul``' Instruction
3710 ^^^^^^^^^^^^^^^^^^^^^
3717 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3718 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3719 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3720 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3725 The '``mul``' instruction returns the product of its two operands.
3730 The two arguments to the '``mul``' instruction must be
3731 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3732 arguments must have identical types.
3737 The value produced is the integer product of the two operands.
3739 If the result of the multiplication has unsigned overflow, the result
3740 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3741 bit width of the result.
3743 Because LLVM integers use a two's complement representation, and the
3744 result is the same width as the operands, this instruction returns the
3745 correct result for both signed and unsigned integers. If a full product
3746 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3747 sign-extended or zero-extended as appropriate to the width of the full
3750 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3751 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3752 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3753 unsigned and/or signed overflow, respectively, occurs.
3758 .. code-block:: llvm
3760 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3764 '``fmul``' Instruction
3765 ^^^^^^^^^^^^^^^^^^^^^^
3772 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3777 The '``fmul``' instruction returns the product of its two operands.
3782 The two arguments to the '``fmul``' instruction must be :ref:`floating
3783 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3784 Both arguments must have identical types.
3789 The value produced is the floating point product of the two operands.
3790 This instruction can also take any number of :ref:`fast-math
3791 flags <fastmath>`, which are optimization hints to enable otherwise
3792 unsafe floating point optimizations:
3797 .. code-block:: llvm
3799 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3801 '``udiv``' Instruction
3802 ^^^^^^^^^^^^^^^^^^^^^^
3809 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3810 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3815 The '``udiv``' instruction returns the quotient of its two operands.
3820 The two arguments to the '``udiv``' instruction must be
3821 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3822 arguments must have identical types.
3827 The value produced is the unsigned integer quotient of the two operands.
3829 Note that unsigned integer division and signed integer division are
3830 distinct operations; for signed integer division, use '``sdiv``'.
3832 Division by zero leads to undefined behavior.
3834 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3835 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3836 such, "((a udiv exact b) mul b) == a").
3841 .. code-block:: llvm
3843 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3845 '``sdiv``' Instruction
3846 ^^^^^^^^^^^^^^^^^^^^^^
3853 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3854 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3859 The '``sdiv``' instruction returns the quotient of its two operands.
3864 The two arguments to the '``sdiv``' instruction must be
3865 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3866 arguments must have identical types.
3871 The value produced is the signed integer quotient of the two operands
3872 rounded towards zero.
3874 Note that signed integer division and unsigned integer division are
3875 distinct operations; for unsigned integer division, use '``udiv``'.
3877 Division by zero leads to undefined behavior. Overflow also leads to
3878 undefined behavior; this is a rare case, but can occur, for example, by
3879 doing a 32-bit division of -2147483648 by -1.
3881 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3882 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3887 .. code-block:: llvm
3889 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3893 '``fdiv``' Instruction
3894 ^^^^^^^^^^^^^^^^^^^^^^
3901 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3906 The '``fdiv``' instruction returns the quotient of its two operands.
3911 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3912 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3913 Both arguments must have identical types.
3918 The value produced is the floating point quotient of the two operands.
3919 This instruction can also take any number of :ref:`fast-math
3920 flags <fastmath>`, which are optimization hints to enable otherwise
3921 unsafe floating point optimizations:
3926 .. code-block:: llvm
3928 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3930 '``urem``' Instruction
3931 ^^^^^^^^^^^^^^^^^^^^^^
3938 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3943 The '``urem``' instruction returns the remainder from the unsigned
3944 division of its two arguments.
3949 The two arguments to the '``urem``' instruction must be
3950 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3951 arguments must have identical types.
3956 This instruction returns the unsigned integer *remainder* of a division.
3957 This instruction always performs an unsigned division to get the
3960 Note that unsigned integer remainder and signed integer remainder are
3961 distinct operations; for signed integer remainder, use '``srem``'.
3963 Taking the remainder of a division by zero leads to undefined behavior.
3968 .. code-block:: llvm
3970 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3972 '``srem``' Instruction
3973 ^^^^^^^^^^^^^^^^^^^^^^
3980 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3985 The '``srem``' instruction returns the remainder from the signed
3986 division of its two operands. This instruction can also take
3987 :ref:`vector <t_vector>` versions of the values in which case the elements
3993 The two arguments to the '``srem``' instruction must be
3994 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3995 arguments must have identical types.
4000 This instruction returns the *remainder* of a division (where the result
4001 is either zero or has the same sign as the dividend, ``op1``), not the
4002 *modulo* operator (where the result is either zero or has the same sign
4003 as the divisor, ``op2``) of a value. For more information about the
4004 difference, see `The Math
4005 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4006 table of how this is implemented in various languages, please see
4008 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4010 Note that signed integer remainder and unsigned integer remainder are
4011 distinct operations; for unsigned integer remainder, use '``urem``'.
4013 Taking the remainder of a division by zero leads to undefined behavior.
4014 Overflow also leads to undefined behavior; this is a rare case, but can
4015 occur, for example, by taking the remainder of a 32-bit division of
4016 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4017 rule lets srem be implemented using instructions that return both the
4018 result of the division and the remainder.)
4023 .. code-block:: llvm
4025 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4029 '``frem``' Instruction
4030 ^^^^^^^^^^^^^^^^^^^^^^
4037 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4042 The '``frem``' instruction returns the remainder from the division of
4048 The two arguments to the '``frem``' instruction must be :ref:`floating
4049 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4050 Both arguments must have identical types.
4055 This instruction returns the *remainder* of a division. The remainder
4056 has the same sign as the dividend. This instruction can also take any
4057 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4058 to enable otherwise unsafe floating point optimizations:
4063 .. code-block:: llvm
4065 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4069 Bitwise Binary Operations
4070 -------------------------
4072 Bitwise binary operators are used to do various forms of bit-twiddling
4073 in a program. They are generally very efficient instructions and can
4074 commonly be strength reduced from other instructions. They require two
4075 operands of the same type, execute an operation on them, and produce a
4076 single value. The resulting value is the same type as its operands.
4078 '``shl``' Instruction
4079 ^^^^^^^^^^^^^^^^^^^^^
4086 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4087 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4088 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4089 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4094 The '``shl``' instruction returns the first operand shifted to the left
4095 a specified number of bits.
4100 Both arguments to the '``shl``' instruction must be the same
4101 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4102 '``op2``' is treated as an unsigned value.
4107 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4108 where ``n`` is the width of the result. If ``op2`` is (statically or
4109 dynamically) negative or equal to or larger than the number of bits in
4110 ``op1``, the result is undefined. If the arguments are vectors, each
4111 vector element of ``op1`` is shifted by the corresponding shift amount
4114 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4115 value <poisonvalues>` if it shifts out any non-zero bits. If the
4116 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4117 value <poisonvalues>` if it shifts out any bits that disagree with the
4118 resultant sign bit. As such, NUW/NSW have the same semantics as they
4119 would if the shift were expressed as a mul instruction with the same
4120 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4125 .. code-block:: llvm
4127 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4128 <result> = shl i32 4, 2 ; yields {i32}: 16
4129 <result> = shl i32 1, 10 ; yields {i32}: 1024
4130 <result> = shl i32 1, 32 ; undefined
4131 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4133 '``lshr``' Instruction
4134 ^^^^^^^^^^^^^^^^^^^^^^
4141 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4142 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4147 The '``lshr``' instruction (logical shift right) returns the first
4148 operand shifted to the right a specified number of bits with zero fill.
4153 Both arguments to the '``lshr``' instruction must be the same
4154 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4155 '``op2``' is treated as an unsigned value.
4160 This instruction always performs a logical shift right operation. The
4161 most significant bits of the result will be filled with zero bits after
4162 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4163 than the number of bits in ``op1``, the result is undefined. If the
4164 arguments are vectors, each vector element of ``op1`` is shifted by the
4165 corresponding shift amount in ``op2``.
4167 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4168 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4174 .. code-block:: llvm
4176 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4177 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4178 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4179 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4180 <result> = lshr i32 1, 32 ; undefined
4181 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4183 '``ashr``' Instruction
4184 ^^^^^^^^^^^^^^^^^^^^^^
4191 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4192 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4197 The '``ashr``' instruction (arithmetic shift right) returns the first
4198 operand shifted to the right a specified number of bits with sign
4204 Both arguments to the '``ashr``' instruction must be the same
4205 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4206 '``op2``' is treated as an unsigned value.
4211 This instruction always performs an arithmetic shift right operation,
4212 The most significant bits of the result will be filled with the sign bit
4213 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4214 than the number of bits in ``op1``, the result is undefined. If the
4215 arguments are vectors, each vector element of ``op1`` is shifted by the
4216 corresponding shift amount in ``op2``.
4218 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4219 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4225 .. code-block:: llvm
4227 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4228 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4229 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4230 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4231 <result> = ashr i32 1, 32 ; undefined
4232 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4234 '``and``' Instruction
4235 ^^^^^^^^^^^^^^^^^^^^^
4242 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4247 The '``and``' instruction returns the bitwise logical and of its two
4253 The two arguments to the '``and``' instruction must be
4254 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4255 arguments must have identical types.
4260 The truth table used for the '``and``' instruction is:
4277 .. code-block:: llvm
4279 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4280 <result> = and i32 15, 40 ; yields {i32}:result = 8
4281 <result> = and i32 4, 8 ; yields {i32}:result = 0
4283 '``or``' Instruction
4284 ^^^^^^^^^^^^^^^^^^^^
4291 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4296 The '``or``' instruction returns the bitwise logical inclusive or of its
4302 The two arguments to the '``or``' instruction must be
4303 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4304 arguments must have identical types.
4309 The truth table used for the '``or``' instruction is:
4328 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4329 <result> = or i32 15, 40 ; yields {i32}:result = 47
4330 <result> = or i32 4, 8 ; yields {i32}:result = 12
4332 '``xor``' Instruction
4333 ^^^^^^^^^^^^^^^^^^^^^
4340 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4345 The '``xor``' instruction returns the bitwise logical exclusive or of
4346 its two operands. The ``xor`` is used to implement the "one's
4347 complement" operation, which is the "~" operator in C.
4352 The two arguments to the '``xor``' instruction must be
4353 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4354 arguments must have identical types.
4359 The truth table used for the '``xor``' instruction is:
4376 .. code-block:: llvm
4378 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4379 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4380 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4381 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4386 LLVM supports several instructions to represent vector operations in a
4387 target-independent manner. These instructions cover the element-access
4388 and vector-specific operations needed to process vectors effectively.
4389 While LLVM does directly support these vector operations, many
4390 sophisticated algorithms will want to use target-specific intrinsics to
4391 take full advantage of a specific target.
4393 .. _i_extractelement:
4395 '``extractelement``' Instruction
4396 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4403 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4408 The '``extractelement``' instruction extracts a single scalar element
4409 from a vector at a specified index.
4414 The first operand of an '``extractelement``' instruction is a value of
4415 :ref:`vector <t_vector>` type. The second operand is an index indicating
4416 the position from which to extract the element. The index may be a
4422 The result is a scalar of the same type as the element type of ``val``.
4423 Its value is the value at position ``idx`` of ``val``. If ``idx``
4424 exceeds the length of ``val``, the results are undefined.
4429 .. code-block:: llvm
4431 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4433 .. _i_insertelement:
4435 '``insertelement``' Instruction
4436 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4443 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4448 The '``insertelement``' instruction inserts a scalar element into a
4449 vector at a specified index.
4454 The first operand of an '``insertelement``' instruction is a value of
4455 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4456 type must equal the element type of the first operand. The third operand
4457 is an index indicating the position at which to insert the value. The
4458 index may be a variable.
4463 The result is a vector of the same type as ``val``. Its element values
4464 are those of ``val`` except at position ``idx``, where it gets the value
4465 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4471 .. code-block:: llvm
4473 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4475 .. _i_shufflevector:
4477 '``shufflevector``' Instruction
4478 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4485 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4490 The '``shufflevector``' instruction constructs a permutation of elements
4491 from two input vectors, returning a vector with the same element type as
4492 the input and length that is the same as the shuffle mask.
4497 The first two operands of a '``shufflevector``' instruction are vectors
4498 with the same type. The third argument is a shuffle mask whose element
4499 type is always 'i32'. The result of the instruction is a vector whose
4500 length is the same as the shuffle mask and whose element type is the
4501 same as the element type of the first two operands.
4503 The shuffle mask operand is required to be a constant vector with either
4504 constant integer or undef values.
4509 The elements of the two input vectors are numbered from left to right
4510 across both of the vectors. The shuffle mask operand specifies, for each
4511 element of the result vector, which element of the two input vectors the
4512 result element gets. The element selector may be undef (meaning "don't
4513 care") and the second operand may be undef if performing a shuffle from
4519 .. code-block:: llvm
4521 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4522 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4523 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4524 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4525 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4526 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4527 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4528 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4530 Aggregate Operations
4531 --------------------
4533 LLVM supports several instructions for working with
4534 :ref:`aggregate <t_aggregate>` values.
4538 '``extractvalue``' Instruction
4539 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4546 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4551 The '``extractvalue``' instruction extracts the value of a member field
4552 from an :ref:`aggregate <t_aggregate>` value.
4557 The first operand of an '``extractvalue``' instruction is a value of
4558 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4559 constant indices to specify which value to extract in a similar manner
4560 as indices in a '``getelementptr``' instruction.
4562 The major differences to ``getelementptr`` indexing are:
4564 - Since the value being indexed is not a pointer, the first index is
4565 omitted and assumed to be zero.
4566 - At least one index must be specified.
4567 - Not only struct indices but also array indices must be in bounds.
4572 The result is the value at the position in the aggregate specified by
4578 .. code-block:: llvm
4580 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4584 '``insertvalue``' Instruction
4585 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4592 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4597 The '``insertvalue``' instruction inserts a value into a member field in
4598 an :ref:`aggregate <t_aggregate>` value.
4603 The first operand of an '``insertvalue``' instruction is a value of
4604 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4605 a first-class value to insert. The following operands are constant
4606 indices indicating the position at which to insert the value in a
4607 similar manner as indices in a '``extractvalue``' instruction. The value
4608 to insert must have the same type as the value identified by the
4614 The result is an aggregate of the same type as ``val``. Its value is
4615 that of ``val`` except that the value at the position specified by the
4616 indices is that of ``elt``.
4621 .. code-block:: llvm
4623 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4624 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4625 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4629 Memory Access and Addressing Operations
4630 ---------------------------------------
4632 A key design point of an SSA-based representation is how it represents
4633 memory. In LLVM, no memory locations are in SSA form, which makes things
4634 very simple. This section describes how to read, write, and allocate
4639 '``alloca``' Instruction
4640 ^^^^^^^^^^^^^^^^^^^^^^^^
4647 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4652 The '``alloca``' instruction allocates memory on the stack frame of the
4653 currently executing function, to be automatically released when this
4654 function returns to its caller. The object is always allocated in the
4655 generic address space (address space zero).
4660 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4661 bytes of memory on the runtime stack, returning a pointer of the
4662 appropriate type to the program. If "NumElements" is specified, it is
4663 the number of elements allocated, otherwise "NumElements" is defaulted
4664 to be one. If a constant alignment is specified, the value result of the
4665 allocation is guaranteed to be aligned to at least that boundary. If not
4666 specified, or if zero, the target can choose to align the allocation on
4667 any convenient boundary compatible with the type.
4669 '``type``' may be any sized type.
4674 Memory is allocated; a pointer is returned. The operation is undefined
4675 if there is insufficient stack space for the allocation. '``alloca``'d
4676 memory is automatically released when the function returns. The
4677 '``alloca``' instruction is commonly used to represent automatic
4678 variables that must have an address available. When the function returns
4679 (either with the ``ret`` or ``resume`` instructions), the memory is
4680 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4681 The order in which memory is allocated (ie., which way the stack grows)
4687 .. code-block:: llvm
4689 %ptr = alloca i32 ; yields {i32*}:ptr
4690 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4691 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4692 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4696 '``load``' Instruction
4697 ^^^^^^^^^^^^^^^^^^^^^^
4704 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4705 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4706 !<index> = !{ i32 1 }
4711 The '``load``' instruction is used to read from memory.
4716 The argument to the ``load`` instruction specifies the memory address
4717 from which to load. The pointer must point to a :ref:`first
4718 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4719 then the optimizer is not allowed to modify the number or order of
4720 execution of this ``load`` with other :ref:`volatile
4721 operations <volatile>`.
4723 If the ``load`` is marked as ``atomic``, it takes an extra
4724 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4725 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4726 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4727 when they may see multiple atomic stores. The type of the pointee must
4728 be an integer type whose bit width is a power of two greater than or
4729 equal to eight and less than or equal to a target-specific size limit.
4730 ``align`` must be explicitly specified on atomic loads, and the load has
4731 undefined behavior if the alignment is not set to a value which is at
4732 least the size in bytes of the pointee. ``!nontemporal`` does not have
4733 any defined semantics for atomic loads.
4735 The optional constant ``align`` argument specifies the alignment of the
4736 operation (that is, the alignment of the memory address). A value of 0
4737 or an omitted ``align`` argument means that the operation has the ABI
4738 alignment for the target. It is the responsibility of the code emitter
4739 to ensure that the alignment information is correct. Overestimating the
4740 alignment results in undefined behavior. Underestimating the alignment
4741 may produce less efficient code. An alignment of 1 is always safe.
4743 The optional ``!nontemporal`` metadata must reference a single
4744 metadata name ``<index>`` corresponding to a metadata node with one
4745 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4746 metadata on the instruction tells the optimizer and code generator
4747 that this load is not expected to be reused in the cache. The code
4748 generator may select special instructions to save cache bandwidth, such
4749 as the ``MOVNT`` instruction on x86.
4751 The optional ``!invariant.load`` metadata must reference a single
4752 metadata name ``<index>`` corresponding to a metadata node with no
4753 entries. The existence of the ``!invariant.load`` metadata on the
4754 instruction tells the optimizer and code generator that this load
4755 address points to memory which does not change value during program
4756 execution. The optimizer may then move this load around, for example, by
4757 hoisting it out of loops using loop invariant code motion.
4762 The location of memory pointed to is loaded. If the value being loaded
4763 is of scalar type then the number of bytes read does not exceed the
4764 minimum number of bytes needed to hold all bits of the type. For
4765 example, loading an ``i24`` reads at most three bytes. When loading a
4766 value of a type like ``i20`` with a size that is not an integral number
4767 of bytes, the result is undefined if the value was not originally
4768 written using a store of the same type.
4773 .. code-block:: llvm
4775 %ptr = alloca i32 ; yields {i32*}:ptr
4776 store i32 3, i32* %ptr ; yields {void}
4777 %val = load i32* %ptr ; yields {i32}:val = i32 3
4781 '``store``' Instruction
4782 ^^^^^^^^^^^^^^^^^^^^^^^
4789 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4790 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4795 The '``store``' instruction is used to write to memory.
4800 There are two arguments to the ``store`` instruction: a value to store
4801 and an address at which to store it. The type of the ``<pointer>``
4802 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4803 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4804 then the optimizer is not allowed to modify the number or order of
4805 execution of this ``store`` with other :ref:`volatile
4806 operations <volatile>`.
4808 If the ``store`` is marked as ``atomic``, it takes an extra
4809 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4810 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4811 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4812 when they may see multiple atomic stores. The type of the pointee must
4813 be an integer type whose bit width is a power of two greater than or
4814 equal to eight and less than or equal to a target-specific size limit.
4815 ``align`` must be explicitly specified on atomic stores, and the store
4816 has undefined behavior if the alignment is not set to a value which is
4817 at least the size in bytes of the pointee. ``!nontemporal`` does not
4818 have any defined semantics for atomic stores.
4820 The optional constant ``align`` argument specifies the alignment of the
4821 operation (that is, the alignment of the memory address). A value of 0
4822 or an omitted ``align`` argument means that the operation has the ABI
4823 alignment for the target. It is the responsibility of the code emitter
4824 to ensure that the alignment information is correct. Overestimating the
4825 alignment results in undefined behavior. Underestimating the
4826 alignment may produce less efficient code. An alignment of 1 is always
4829 The optional ``!nontemporal`` metadata must reference a single metadata
4830 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4831 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4832 tells the optimizer and code generator that this load is not expected to
4833 be reused in the cache. The code generator may select special
4834 instructions to save cache bandwidth, such as the MOVNT instruction on
4840 The contents of memory are updated to contain ``<value>`` at the
4841 location specified by the ``<pointer>`` operand. If ``<value>`` is
4842 of scalar type then the number of bytes written does not exceed the
4843 minimum number of bytes needed to hold all bits of the type. For
4844 example, storing an ``i24`` writes at most three bytes. When writing a
4845 value of a type like ``i20`` with a size that is not an integral number
4846 of bytes, it is unspecified what happens to the extra bits that do not
4847 belong to the type, but they will typically be overwritten.
4852 .. code-block:: llvm
4854 %ptr = alloca i32 ; yields {i32*}:ptr
4855 store i32 3, i32* %ptr ; yields {void}
4856 %val = load i32* %ptr ; yields {i32}:val = i32 3
4860 '``fence``' Instruction
4861 ^^^^^^^^^^^^^^^^^^^^^^^
4868 fence [singlethread] <ordering> ; yields {void}
4873 The '``fence``' instruction is used to introduce happens-before edges
4879 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4880 defines what *synchronizes-with* edges they add. They can only be given
4881 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4886 A fence A which has (at least) ``release`` ordering semantics
4887 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4888 semantics if and only if there exist atomic operations X and Y, both
4889 operating on some atomic object M, such that A is sequenced before X, X
4890 modifies M (either directly or through some side effect of a sequence
4891 headed by X), Y is sequenced before B, and Y observes M. This provides a
4892 *happens-before* dependency between A and B. Rather than an explicit
4893 ``fence``, one (but not both) of the atomic operations X or Y might
4894 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4895 still *synchronize-with* the explicit ``fence`` and establish the
4896 *happens-before* edge.
4898 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4899 ``acquire`` and ``release`` semantics specified above, participates in
4900 the global program order of other ``seq_cst`` operations and/or fences.
4902 The optional ":ref:`singlethread <singlethread>`" argument specifies
4903 that the fence only synchronizes with other fences in the same thread.
4904 (This is useful for interacting with signal handlers.)
4909 .. code-block:: llvm
4911 fence acquire ; yields {void}
4912 fence singlethread seq_cst ; yields {void}
4916 '``cmpxchg``' Instruction
4917 ^^^^^^^^^^^^^^^^^^^^^^^^^
4924 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4929 The '``cmpxchg``' instruction is used to atomically modify memory. It
4930 loads a value in memory and compares it to a given value. If they are
4931 equal, it stores a new value into the memory.
4936 There are three arguments to the '``cmpxchg``' instruction: an address
4937 to operate on, a value to compare to the value currently be at that
4938 address, and a new value to place at that address if the compared values
4939 are equal. The type of '<cmp>' must be an integer type whose bit width
4940 is a power of two greater than or equal to eight and less than or equal
4941 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4942 type, and the type of '<pointer>' must be a pointer to that type. If the
4943 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4944 to modify the number or order of execution of this ``cmpxchg`` with
4945 other :ref:`volatile operations <volatile>`.
4947 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4948 synchronizes with other atomic operations.
4950 The optional "``singlethread``" argument declares that the ``cmpxchg``
4951 is only atomic with respect to code (usually signal handlers) running in
4952 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4953 respect to all other code in the system.
4955 The pointer passed into cmpxchg must have alignment greater than or
4956 equal to the size in memory of the operand.
4961 The contents of memory at the location specified by the '``<pointer>``'
4962 operand is read and compared to '``<cmp>``'; if the read value is the
4963 equal, '``<new>``' is written. The original value at the location is
4966 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4967 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4968 atomic load with an ordering parameter determined by dropping any
4969 ``release`` part of the ``cmpxchg``'s ordering.
4974 .. code-block:: llvm
4977 %orig = atomic load i32* %ptr unordered ; yields {i32}
4981 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4982 %squared = mul i32 %cmp, %cmp
4983 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4984 %success = icmp eq i32 %cmp, %old
4985 br i1 %success, label %done, label %loop
4992 '``atomicrmw``' Instruction
4993 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5000 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
5005 The '``atomicrmw``' instruction is used to atomically modify memory.
5010 There are three arguments to the '``atomicrmw``' instruction: an
5011 operation to apply, an address whose value to modify, an argument to the
5012 operation. The operation must be one of the following keywords:
5026 The type of '<value>' must be an integer type whose bit width is a power
5027 of two greater than or equal to eight and less than or equal to a
5028 target-specific size limit. The type of the '``<pointer>``' operand must
5029 be a pointer to that type. If the ``atomicrmw`` is marked as
5030 ``volatile``, then the optimizer is not allowed to modify the number or
5031 order of execution of this ``atomicrmw`` with other :ref:`volatile
5032 operations <volatile>`.
5037 The contents of memory at the location specified by the '``<pointer>``'
5038 operand are atomically read, modified, and written back. The original
5039 value at the location is returned. The modification is specified by the
5042 - xchg: ``*ptr = val``
5043 - add: ``*ptr = *ptr + val``
5044 - sub: ``*ptr = *ptr - val``
5045 - and: ``*ptr = *ptr & val``
5046 - nand: ``*ptr = ~(*ptr & val)``
5047 - or: ``*ptr = *ptr | val``
5048 - xor: ``*ptr = *ptr ^ val``
5049 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5050 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5051 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5053 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5059 .. code-block:: llvm
5061 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5063 .. _i_getelementptr:
5065 '``getelementptr``' Instruction
5066 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5073 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5074 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5075 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5080 The '``getelementptr``' instruction is used to get the address of a
5081 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5082 address calculation only and does not access memory.
5087 The first argument is always a pointer or a vector of pointers, and
5088 forms the basis of the calculation. The remaining arguments are indices
5089 that indicate which of the elements of the aggregate object are indexed.
5090 The interpretation of each index is dependent on the type being indexed
5091 into. The first index always indexes the pointer value given as the
5092 first argument, the second index indexes a value of the type pointed to
5093 (not necessarily the value directly pointed to, since the first index
5094 can be non-zero), etc. The first type indexed into must be a pointer
5095 value, subsequent types can be arrays, vectors, and structs. Note that
5096 subsequent types being indexed into can never be pointers, since that
5097 would require loading the pointer before continuing calculation.
5099 The type of each index argument depends on the type it is indexing into.
5100 When indexing into a (optionally packed) structure, only ``i32`` integer
5101 **constants** are allowed (when using a vector of indices they must all
5102 be the **same** ``i32`` integer constant). When indexing into an array,
5103 pointer or vector, integers of any width are allowed, and they are not
5104 required to be constant. These integers are treated as signed values
5107 For example, let's consider a C code fragment and how it gets compiled
5123 int *foo(struct ST *s) {
5124 return &s[1].Z.B[5][13];
5127 The LLVM code generated by Clang is:
5129 .. code-block:: llvm
5131 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5132 %struct.ST = type { i32, double, %struct.RT }
5134 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5136 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5143 In the example above, the first index is indexing into the
5144 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5145 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5146 indexes into the third element of the structure, yielding a
5147 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5148 structure. The third index indexes into the second element of the
5149 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5150 dimensions of the array are subscripted into, yielding an '``i32``'
5151 type. The '``getelementptr``' instruction returns a pointer to this
5152 element, thus computing a value of '``i32*``' type.
5154 Note that it is perfectly legal to index partially through a structure,
5155 returning a pointer to an inner element. Because of this, the LLVM code
5156 for the given testcase is equivalent to:
5158 .. code-block:: llvm
5160 define i32* @foo(%struct.ST* %s) {
5161 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5162 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5163 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5164 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5165 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5169 If the ``inbounds`` keyword is present, the result value of the
5170 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5171 pointer is not an *in bounds* address of an allocated object, or if any
5172 of the addresses that would be formed by successive addition of the
5173 offsets implied by the indices to the base address with infinitely
5174 precise signed arithmetic are not an *in bounds* address of that
5175 allocated object. The *in bounds* addresses for an allocated object are
5176 all the addresses that point into the object, plus the address one byte
5177 past the end. In cases where the base is a vector of pointers the
5178 ``inbounds`` keyword applies to each of the computations element-wise.
5180 If the ``inbounds`` keyword is not present, the offsets are added to the
5181 base address with silently-wrapping two's complement arithmetic. If the
5182 offsets have a different width from the pointer, they are sign-extended
5183 or truncated to the width of the pointer. The result value of the
5184 ``getelementptr`` may be outside the object pointed to by the base
5185 pointer. The result value may not necessarily be used to access memory
5186 though, even if it happens to point into allocated storage. See the
5187 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5190 The getelementptr instruction is often confusing. For some more insight
5191 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5196 .. code-block:: llvm
5198 ; yields [12 x i8]*:aptr
5199 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5201 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5203 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5205 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5207 In cases where the pointer argument is a vector of pointers, each index
5208 must be a vector with the same number of elements. For example:
5210 .. code-block:: llvm
5212 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5214 Conversion Operations
5215 ---------------------
5217 The instructions in this category are the conversion instructions
5218 (casting) which all take a single operand and a type. They perform
5219 various bit conversions on the operand.
5221 '``trunc .. to``' Instruction
5222 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5229 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5234 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5239 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5240 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5241 of the same number of integers. The bit size of the ``value`` must be
5242 larger than the bit size of the destination type, ``ty2``. Equal sized
5243 types are not allowed.
5248 The '``trunc``' instruction truncates the high order bits in ``value``
5249 and converts the remaining bits to ``ty2``. Since the source size must
5250 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5251 It will always truncate bits.
5256 .. code-block:: llvm
5258 %X = trunc i32 257 to i8 ; yields i8:1
5259 %Y = trunc i32 123 to i1 ; yields i1:true
5260 %Z = trunc i32 122 to i1 ; yields i1:false
5261 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5263 '``zext .. to``' Instruction
5264 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5271 <result> = zext <ty> <value> to <ty2> ; yields ty2
5276 The '``zext``' instruction zero extends its operand to type ``ty2``.
5281 The '``zext``' instruction takes a value to cast, and a type to cast it
5282 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5283 the same number of integers. The bit size of the ``value`` must be
5284 smaller than the bit size of the destination type, ``ty2``.
5289 The ``zext`` fills the high order bits of the ``value`` with zero bits
5290 until it reaches the size of the destination type, ``ty2``.
5292 When zero extending from i1, the result will always be either 0 or 1.
5297 .. code-block:: llvm
5299 %X = zext i32 257 to i64 ; yields i64:257
5300 %Y = zext i1 true to i32 ; yields i32:1
5301 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5303 '``sext .. to``' Instruction
5304 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5311 <result> = sext <ty> <value> to <ty2> ; yields ty2
5316 The '``sext``' sign extends ``value`` to the type ``ty2``.
5321 The '``sext``' instruction takes a value to cast, and a type to cast it
5322 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5323 the same number of integers. The bit size of the ``value`` must be
5324 smaller than the bit size of the destination type, ``ty2``.
5329 The '``sext``' instruction performs a sign extension by copying the sign
5330 bit (highest order bit) of the ``value`` until it reaches the bit size
5331 of the type ``ty2``.
5333 When sign extending from i1, the extension always results in -1 or 0.
5338 .. code-block:: llvm
5340 %X = sext i8 -1 to i16 ; yields i16 :65535
5341 %Y = sext i1 true to i32 ; yields i32:-1
5342 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5344 '``fptrunc .. to``' Instruction
5345 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5352 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5357 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5362 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5363 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5364 The size of ``value`` must be larger than the size of ``ty2``. This
5365 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5370 The '``fptrunc``' instruction truncates a ``value`` from a larger
5371 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5372 point <t_floating>` type. If the value cannot fit within the
5373 destination type, ``ty2``, then the results are undefined.
5378 .. code-block:: llvm
5380 %X = fptrunc double 123.0 to float ; yields float:123.0
5381 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5383 '``fpext .. to``' Instruction
5384 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5391 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5396 The '``fpext``' extends a floating point ``value`` to a larger floating
5402 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5403 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5404 to. The source type must be smaller than the destination type.
5409 The '``fpext``' instruction extends the ``value`` from a smaller
5410 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5411 point <t_floating>` type. The ``fpext`` cannot be used to make a
5412 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5413 *no-op cast* for a floating point cast.
5418 .. code-block:: llvm
5420 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5421 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5423 '``fptoui .. to``' Instruction
5424 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5431 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5436 The '``fptoui``' converts a floating point ``value`` to its unsigned
5437 integer equivalent of type ``ty2``.
5442 The '``fptoui``' instruction takes a value to cast, which must be a
5443 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5444 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5445 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5446 type with the same number of elements as ``ty``
5451 The '``fptoui``' instruction converts its :ref:`floating
5452 point <t_floating>` operand into the nearest (rounding towards zero)
5453 unsigned integer value. If the value cannot fit in ``ty2``, the results
5459 .. code-block:: llvm
5461 %X = fptoui double 123.0 to i32 ; yields i32:123
5462 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5463 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5465 '``fptosi .. to``' Instruction
5466 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5473 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5478 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5479 ``value`` to type ``ty2``.
5484 The '``fptosi``' instruction takes a value to cast, which must be a
5485 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5486 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5487 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5488 type with the same number of elements as ``ty``
5493 The '``fptosi``' instruction converts its :ref:`floating
5494 point <t_floating>` operand into the nearest (rounding towards zero)
5495 signed integer value. If the value cannot fit in ``ty2``, the results
5501 .. code-block:: llvm
5503 %X = fptosi double -123.0 to i32 ; yields i32:-123
5504 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5505 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5507 '``uitofp .. to``' Instruction
5508 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5515 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5520 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5521 and converts that value to the ``ty2`` type.
5526 The '``uitofp``' instruction takes a value to cast, which must be a
5527 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5528 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5529 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5530 type with the same number of elements as ``ty``
5535 The '``uitofp``' instruction interprets its operand as an unsigned
5536 integer quantity and converts it to the corresponding floating point
5537 value. If the value cannot fit in the floating point value, the results
5543 .. code-block:: llvm
5545 %X = uitofp i32 257 to float ; yields float:257.0
5546 %Y = uitofp i8 -1 to double ; yields double:255.0
5548 '``sitofp .. to``' Instruction
5549 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5556 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5561 The '``sitofp``' instruction regards ``value`` as a signed integer and
5562 converts that value to the ``ty2`` type.
5567 The '``sitofp``' instruction takes a value to cast, which must be a
5568 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5569 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5570 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5571 type with the same number of elements as ``ty``
5576 The '``sitofp``' instruction interprets its operand as a signed integer
5577 quantity and converts it to the corresponding floating point value. If
5578 the value cannot fit in the floating point value, the results are
5584 .. code-block:: llvm
5586 %X = sitofp i32 257 to float ; yields float:257.0
5587 %Y = sitofp i8 -1 to double ; yields double:-1.0
5591 '``ptrtoint .. to``' Instruction
5592 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5599 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5604 The '``ptrtoint``' instruction converts the pointer or a vector of
5605 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5610 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5611 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5612 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5613 a vector of integers type.
5618 The '``ptrtoint``' instruction converts ``value`` to integer type
5619 ``ty2`` by interpreting the pointer value as an integer and either
5620 truncating or zero extending that value to the size of the integer type.
5621 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5622 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5623 the same size, then nothing is done (*no-op cast*) other than a type
5629 .. code-block:: llvm
5631 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5632 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5633 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5637 '``inttoptr .. to``' Instruction
5638 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5645 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5650 The '``inttoptr``' instruction converts an integer ``value`` to a
5651 pointer type, ``ty2``.
5656 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5657 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5663 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5664 applying either a zero extension or a truncation depending on the size
5665 of the integer ``value``. If ``value`` is larger than the size of a
5666 pointer then a truncation is done. If ``value`` is smaller than the size
5667 of a pointer then a zero extension is done. If they are the same size,
5668 nothing is done (*no-op cast*).
5673 .. code-block:: llvm
5675 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5676 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5677 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5678 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5682 '``bitcast .. to``' Instruction
5683 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5690 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5695 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5701 The '``bitcast``' instruction takes a value to cast, which must be a
5702 non-aggregate first class value, and a type to cast it to, which must
5703 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5704 bit sizes of ``value`` and the destination type, ``ty2``, must be
5705 identical. If the source type is a pointer, the destination type must
5706 also be a pointer of the same size. This instruction supports bitwise
5707 conversion of vectors to integers and to vectors of other types (as
5708 long as they have the same size).
5713 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5714 is always a *no-op cast* because no bits change with this
5715 conversion. The conversion is done as if the ``value`` had been stored
5716 to memory and read back as type ``ty2``. Pointer (or vector of
5717 pointers) types may only be converted to other pointer (or vector of
5718 pointers) types with the same address space through this instruction.
5719 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5720 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5725 .. code-block:: llvm
5727 %X = bitcast i8 255 to i8 ; yields i8 :-1
5728 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5729 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5730 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5732 .. _i_addrspacecast:
5734 '``addrspacecast .. to``' Instruction
5735 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5742 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5747 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5748 address space ``n`` to type ``pty2`` in address space ``m``.
5753 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5754 to cast and a pointer type to cast it to, which must have a different
5760 The '``addrspacecast``' instruction converts the pointer value
5761 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5762 value modification, depending on the target and the address space
5763 pair. Pointer conversions within the same address space must be
5764 performed with the ``bitcast`` instruction. Note that if the address space
5765 conversion is legal then both result and operand refer to the same memory
5771 .. code-block:: llvm
5773 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5774 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5775 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5782 The instructions in this category are the "miscellaneous" instructions,
5783 which defy better classification.
5787 '``icmp``' Instruction
5788 ^^^^^^^^^^^^^^^^^^^^^^
5795 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5800 The '``icmp``' instruction returns a boolean value or a vector of
5801 boolean values based on comparison of its two integer, integer vector,
5802 pointer, or pointer vector operands.
5807 The '``icmp``' instruction takes three operands. The first operand is
5808 the condition code indicating the kind of comparison to perform. It is
5809 not a value, just a keyword. The possible condition code are:
5812 #. ``ne``: not equal
5813 #. ``ugt``: unsigned greater than
5814 #. ``uge``: unsigned greater or equal
5815 #. ``ult``: unsigned less than
5816 #. ``ule``: unsigned less or equal
5817 #. ``sgt``: signed greater than
5818 #. ``sge``: signed greater or equal
5819 #. ``slt``: signed less than
5820 #. ``sle``: signed less or equal
5822 The remaining two arguments must be :ref:`integer <t_integer>` or
5823 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5824 must also be identical types.
5829 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5830 code given as ``cond``. The comparison performed always yields either an
5831 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5833 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5834 otherwise. No sign interpretation is necessary or performed.
5835 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5836 otherwise. No sign interpretation is necessary or performed.
5837 #. ``ugt``: interprets the operands as unsigned values and yields
5838 ``true`` if ``op1`` is greater than ``op2``.
5839 #. ``uge``: interprets the operands as unsigned values and yields
5840 ``true`` if ``op1`` is greater than or equal to ``op2``.
5841 #. ``ult``: interprets the operands as unsigned values and yields
5842 ``true`` if ``op1`` is less than ``op2``.
5843 #. ``ule``: interprets the operands as unsigned values and yields
5844 ``true`` if ``op1`` is less than or equal to ``op2``.
5845 #. ``sgt``: interprets the operands as signed values and yields ``true``
5846 if ``op1`` is greater than ``op2``.
5847 #. ``sge``: interprets the operands as signed values and yields ``true``
5848 if ``op1`` is greater than or equal to ``op2``.
5849 #. ``slt``: interprets the operands as signed values and yields ``true``
5850 if ``op1`` is less than ``op2``.
5851 #. ``sle``: interprets the operands as signed values and yields ``true``
5852 if ``op1`` is less than or equal to ``op2``.
5854 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5855 are compared as if they were integers.
5857 If the operands are integer vectors, then they are compared element by
5858 element. The result is an ``i1`` vector with the same number of elements
5859 as the values being compared. Otherwise, the result is an ``i1``.
5864 .. code-block:: llvm
5866 <result> = icmp eq i32 4, 5 ; yields: result=false
5867 <result> = icmp ne float* %X, %X ; yields: result=false
5868 <result> = icmp ult i16 4, 5 ; yields: result=true
5869 <result> = icmp sgt i16 4, 5 ; yields: result=false
5870 <result> = icmp ule i16 -4, 5 ; yields: result=false
5871 <result> = icmp sge i16 4, 5 ; yields: result=false
5873 Note that the code generator does not yet support vector types with the
5874 ``icmp`` instruction.
5878 '``fcmp``' Instruction
5879 ^^^^^^^^^^^^^^^^^^^^^^
5886 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5891 The '``fcmp``' instruction returns a boolean value or vector of boolean
5892 values based on comparison of its operands.
5894 If the operands are floating point scalars, then the result type is a
5895 boolean (:ref:`i1 <t_integer>`).
5897 If the operands are floating point vectors, then the result type is a
5898 vector of boolean with the same number of elements as the operands being
5904 The '``fcmp``' instruction takes three operands. The first operand is
5905 the condition code indicating the kind of comparison to perform. It is
5906 not a value, just a keyword. The possible condition code are:
5908 #. ``false``: no comparison, always returns false
5909 #. ``oeq``: ordered and equal
5910 #. ``ogt``: ordered and greater than
5911 #. ``oge``: ordered and greater than or equal
5912 #. ``olt``: ordered and less than
5913 #. ``ole``: ordered and less than or equal
5914 #. ``one``: ordered and not equal
5915 #. ``ord``: ordered (no nans)
5916 #. ``ueq``: unordered or equal
5917 #. ``ugt``: unordered or greater than
5918 #. ``uge``: unordered or greater than or equal
5919 #. ``ult``: unordered or less than
5920 #. ``ule``: unordered or less than or equal
5921 #. ``une``: unordered or not equal
5922 #. ``uno``: unordered (either nans)
5923 #. ``true``: no comparison, always returns true
5925 *Ordered* means that neither operand is a QNAN while *unordered* means
5926 that either operand may be a QNAN.
5928 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5929 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5930 type. They must have identical types.
5935 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5936 condition code given as ``cond``. If the operands are vectors, then the
5937 vectors are compared element by element. Each comparison performed
5938 always yields an :ref:`i1 <t_integer>` result, as follows:
5940 #. ``false``: always yields ``false``, regardless of operands.
5941 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5942 is equal to ``op2``.
5943 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5944 is greater than ``op2``.
5945 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5946 is greater than or equal to ``op2``.
5947 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5948 is less than ``op2``.
5949 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5950 is less than or equal to ``op2``.
5951 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5952 is not equal to ``op2``.
5953 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5954 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5956 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5957 greater than ``op2``.
5958 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5959 greater than or equal to ``op2``.
5960 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5962 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5963 less than or equal to ``op2``.
5964 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5965 not equal to ``op2``.
5966 #. ``uno``: yields ``true`` if either operand is a QNAN.
5967 #. ``true``: always yields ``true``, regardless of operands.
5972 .. code-block:: llvm
5974 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5975 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5976 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5977 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5979 Note that the code generator does not yet support vector types with the
5980 ``fcmp`` instruction.
5984 '``phi``' Instruction
5985 ^^^^^^^^^^^^^^^^^^^^^
5992 <result> = phi <ty> [ <val0>, <label0>], ...
5997 The '``phi``' instruction is used to implement the φ node in the SSA
5998 graph representing the function.
6003 The type of the incoming values is specified with the first type field.
6004 After this, the '``phi``' instruction takes a list of pairs as
6005 arguments, with one pair for each predecessor basic block of the current
6006 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6007 the value arguments to the PHI node. Only labels may be used as the
6010 There must be no non-phi instructions between the start of a basic block
6011 and the PHI instructions: i.e. PHI instructions must be first in a basic
6014 For the purposes of the SSA form, the use of each incoming value is
6015 deemed to occur on the edge from the corresponding predecessor block to
6016 the current block (but after any definition of an '``invoke``'
6017 instruction's return value on the same edge).
6022 At runtime, the '``phi``' instruction logically takes on the value
6023 specified by the pair corresponding to the predecessor basic block that
6024 executed just prior to the current block.
6029 .. code-block:: llvm
6031 Loop: ; Infinite loop that counts from 0 on up...
6032 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6033 %nextindvar = add i32 %indvar, 1
6038 '``select``' Instruction
6039 ^^^^^^^^^^^^^^^^^^^^^^^^
6046 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6048 selty is either i1 or {<N x i1>}
6053 The '``select``' instruction is used to choose one value based on a
6054 condition, without branching.
6059 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6060 values indicating the condition, and two values of the same :ref:`first
6061 class <t_firstclass>` type. If the val1/val2 are vectors and the
6062 condition is a scalar, then entire vectors are selected, not individual
6068 If the condition is an i1 and it evaluates to 1, the instruction returns
6069 the first value argument; otherwise, it returns the second value
6072 If the condition is a vector of i1, then the value arguments must be
6073 vectors of the same size, and the selection is done element by element.
6078 .. code-block:: llvm
6080 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6084 '``call``' Instruction
6085 ^^^^^^^^^^^^^^^^^^^^^^
6092 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6097 The '``call``' instruction represents a simple function call.
6102 This instruction requires several arguments:
6104 #. The optional "tail" marker indicates that the callee function does
6105 not access any allocas or varargs in the caller. Note that calls may
6106 be marked "tail" even if they do not occur before a
6107 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6108 function call is eligible for tail call optimization, but `might not
6109 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6110 The code generator may optimize calls marked "tail" with either 1)
6111 automatic `sibling call
6112 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6113 callee have matching signatures, or 2) forced tail call optimization
6114 when the following extra requirements are met:
6116 - Caller and callee both have the calling convention ``fastcc``.
6117 - The call is in tail position (ret immediately follows call and ret
6118 uses value of call or is void).
6119 - Option ``-tailcallopt`` is enabled, or
6120 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6121 - `Platform specific constraints are
6122 met. <CodeGenerator.html#tailcallopt>`_
6124 #. The optional "cconv" marker indicates which :ref:`calling
6125 convention <callingconv>` the call should use. If none is
6126 specified, the call defaults to using C calling conventions. The
6127 calling convention of the call must match the calling convention of
6128 the target function, or else the behavior is undefined.
6129 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6130 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6132 #. '``ty``': the type of the call instruction itself which is also the
6133 type of the return value. Functions that return no value are marked
6135 #. '``fnty``': shall be the signature of the pointer to function value
6136 being invoked. The argument types must match the types implied by
6137 this signature. This type can be omitted if the function is not
6138 varargs and if the function type does not return a pointer to a
6140 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6141 be invoked. In most cases, this is a direct function invocation, but
6142 indirect ``call``'s are just as possible, calling an arbitrary pointer
6144 #. '``function args``': argument list whose types match the function
6145 signature argument types and parameter attributes. All arguments must
6146 be of :ref:`first class <t_firstclass>` type. If the function signature
6147 indicates the function accepts a variable number of arguments, the
6148 extra arguments can be specified.
6149 #. The optional :ref:`function attributes <fnattrs>` list. Only
6150 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6151 attributes are valid here.
6156 The '``call``' instruction is used to cause control flow to transfer to
6157 a specified function, with its incoming arguments bound to the specified
6158 values. Upon a '``ret``' instruction in the called function, control
6159 flow continues with the instruction after the function call, and the
6160 return value of the function is bound to the result argument.
6165 .. code-block:: llvm
6167 %retval = call i32 @test(i32 %argc)
6168 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6169 %X = tail call i32 @foo() ; yields i32
6170 %Y = tail call fastcc i32 @foo() ; yields i32
6171 call void %foo(i8 97 signext)
6173 %struct.A = type { i32, i8 }
6174 %r = call %struct.A @foo() ; yields { 32, i8 }
6175 %gr = extractvalue %struct.A %r, 0 ; yields i32
6176 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6177 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6178 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6180 llvm treats calls to some functions with names and arguments that match
6181 the standard C99 library as being the C99 library functions, and may
6182 perform optimizations or generate code for them under that assumption.
6183 This is something we'd like to change in the future to provide better
6184 support for freestanding environments and non-C-based languages.
6188 '``va_arg``' Instruction
6189 ^^^^^^^^^^^^^^^^^^^^^^^^
6196 <resultval> = va_arg <va_list*> <arglist>, <argty>
6201 The '``va_arg``' instruction is used to access arguments passed through
6202 the "variable argument" area of a function call. It is used to implement
6203 the ``va_arg`` macro in C.
6208 This instruction takes a ``va_list*`` value and the type of the
6209 argument. It returns a value of the specified argument type and
6210 increments the ``va_list`` to point to the next argument. The actual
6211 type of ``va_list`` is target specific.
6216 The '``va_arg``' instruction loads an argument of the specified type
6217 from the specified ``va_list`` and causes the ``va_list`` to point to
6218 the next argument. For more information, see the variable argument
6219 handling :ref:`Intrinsic Functions <int_varargs>`.
6221 It is legal for this instruction to be called in a function which does
6222 not take a variable number of arguments, for example, the ``vfprintf``
6225 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6226 function <intrinsics>` because it takes a type as an argument.
6231 See the :ref:`variable argument processing <int_varargs>` section.
6233 Note that the code generator does not yet fully support va\_arg on many
6234 targets. Also, it does not currently support va\_arg with aggregate
6235 types on any target.
6239 '``landingpad``' Instruction
6240 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6247 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6248 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6250 <clause> := catch <type> <value>
6251 <clause> := filter <array constant type> <array constant>
6256 The '``landingpad``' instruction is used by `LLVM's exception handling
6257 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6258 is a landing pad --- one where the exception lands, and corresponds to the
6259 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6260 defines values supplied by the personality function (``pers_fn``) upon
6261 re-entry to the function. The ``resultval`` has the type ``resultty``.
6266 This instruction takes a ``pers_fn`` value. This is the personality
6267 function associated with the unwinding mechanism. The optional
6268 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6270 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6271 contains the global variable representing the "type" that may be caught
6272 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6273 clause takes an array constant as its argument. Use
6274 "``[0 x i8**] undef``" for a filter which cannot throw. The
6275 '``landingpad``' instruction must contain *at least* one ``clause`` or
6276 the ``cleanup`` flag.
6281 The '``landingpad``' instruction defines the values which are set by the
6282 personality function (``pers_fn``) upon re-entry to the function, and
6283 therefore the "result type" of the ``landingpad`` instruction. As with
6284 calling conventions, how the personality function results are
6285 represented in LLVM IR is target specific.
6287 The clauses are applied in order from top to bottom. If two
6288 ``landingpad`` instructions are merged together through inlining, the
6289 clauses from the calling function are appended to the list of clauses.
6290 When the call stack is being unwound due to an exception being thrown,
6291 the exception is compared against each ``clause`` in turn. If it doesn't
6292 match any of the clauses, and the ``cleanup`` flag is not set, then
6293 unwinding continues further up the call stack.
6295 The ``landingpad`` instruction has several restrictions:
6297 - A landing pad block is a basic block which is the unwind destination
6298 of an '``invoke``' instruction.
6299 - A landing pad block must have a '``landingpad``' instruction as its
6300 first non-PHI instruction.
6301 - There can be only one '``landingpad``' instruction within the landing
6303 - A basic block that is not a landing pad block may not include a
6304 '``landingpad``' instruction.
6305 - All '``landingpad``' instructions in a function must have the same
6306 personality function.
6311 .. code-block:: llvm
6313 ;; A landing pad which can catch an integer.
6314 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6316 ;; A landing pad that is a cleanup.
6317 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6319 ;; A landing pad which can catch an integer and can only throw a double.
6320 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6322 filter [1 x i8**] [@_ZTId]
6329 LLVM supports the notion of an "intrinsic function". These functions
6330 have well known names and semantics and are required to follow certain
6331 restrictions. Overall, these intrinsics represent an extension mechanism
6332 for the LLVM language that does not require changing all of the
6333 transformations in LLVM when adding to the language (or the bitcode
6334 reader/writer, the parser, etc...).
6336 Intrinsic function names must all start with an "``llvm.``" prefix. This
6337 prefix is reserved in LLVM for intrinsic names; thus, function names may
6338 not begin with this prefix. Intrinsic functions must always be external
6339 functions: you cannot define the body of intrinsic functions. Intrinsic
6340 functions may only be used in call or invoke instructions: it is illegal
6341 to take the address of an intrinsic function. Additionally, because
6342 intrinsic functions are part of the LLVM language, it is required if any
6343 are added that they be documented here.
6345 Some intrinsic functions can be overloaded, i.e., the intrinsic
6346 represents a family of functions that perform the same operation but on
6347 different data types. Because LLVM can represent over 8 million
6348 different integer types, overloading is used commonly to allow an
6349 intrinsic function to operate on any integer type. One or more of the
6350 argument types or the result type can be overloaded to accept any
6351 integer type. Argument types may also be defined as exactly matching a
6352 previous argument's type or the result type. This allows an intrinsic
6353 function which accepts multiple arguments, but needs all of them to be
6354 of the same type, to only be overloaded with respect to a single
6355 argument or the result.
6357 Overloaded intrinsics will have the names of its overloaded argument
6358 types encoded into its function name, each preceded by a period. Only
6359 those types which are overloaded result in a name suffix. Arguments
6360 whose type is matched against another type do not. For example, the
6361 ``llvm.ctpop`` function can take an integer of any width and returns an
6362 integer of exactly the same integer width. This leads to a family of
6363 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6364 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6365 overloaded, and only one type suffix is required. Because the argument's
6366 type is matched against the return type, it does not require its own
6369 To learn how to add an intrinsic function, please see the `Extending
6370 LLVM Guide <ExtendingLLVM.html>`_.
6374 Variable Argument Handling Intrinsics
6375 -------------------------------------
6377 Variable argument support is defined in LLVM with the
6378 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6379 functions. These functions are related to the similarly named macros
6380 defined in the ``<stdarg.h>`` header file.
6382 All of these functions operate on arguments that use a target-specific
6383 value type "``va_list``". The LLVM assembly language reference manual
6384 does not define what this type is, so all transformations should be
6385 prepared to handle these functions regardless of the type used.
6387 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6388 variable argument handling intrinsic functions are used.
6390 .. code-block:: llvm
6392 define i32 @test(i32 %X, ...) {
6393 ; Initialize variable argument processing
6395 %ap2 = bitcast i8** %ap to i8*
6396 call void @llvm.va_start(i8* %ap2)
6398 ; Read a single integer argument
6399 %tmp = va_arg i8** %ap, i32
6401 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6403 %aq2 = bitcast i8** %aq to i8*
6404 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6405 call void @llvm.va_end(i8* %aq2)
6407 ; Stop processing of arguments.
6408 call void @llvm.va_end(i8* %ap2)
6412 declare void @llvm.va_start(i8*)
6413 declare void @llvm.va_copy(i8*, i8*)
6414 declare void @llvm.va_end(i8*)
6418 '``llvm.va_start``' Intrinsic
6419 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6426 declare void @llvm.va_start(i8* <arglist>)
6431 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6432 subsequent use by ``va_arg``.
6437 The argument is a pointer to a ``va_list`` element to initialize.
6442 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6443 available in C. In a target-dependent way, it initializes the
6444 ``va_list`` element to which the argument points, so that the next call
6445 to ``va_arg`` will produce the first variable argument passed to the
6446 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6447 to know the last argument of the function as the compiler can figure
6450 '``llvm.va_end``' Intrinsic
6451 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6458 declare void @llvm.va_end(i8* <arglist>)
6463 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6464 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6469 The argument is a pointer to a ``va_list`` to destroy.
6474 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6475 available in C. In a target-dependent way, it destroys the ``va_list``
6476 element to which the argument points. Calls to
6477 :ref:`llvm.va_start <int_va_start>` and
6478 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6483 '``llvm.va_copy``' Intrinsic
6484 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6491 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6496 The '``llvm.va_copy``' intrinsic copies the current argument position
6497 from the source argument list to the destination argument list.
6502 The first argument is a pointer to a ``va_list`` element to initialize.
6503 The second argument is a pointer to a ``va_list`` element to copy from.
6508 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6509 available in C. In a target-dependent way, it copies the source
6510 ``va_list`` element into the destination ``va_list`` element. This
6511 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6512 arbitrarily complex and require, for example, memory allocation.
6514 Accurate Garbage Collection Intrinsics
6515 --------------------------------------
6517 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6518 (GC) requires the implementation and generation of these intrinsics.
6519 These intrinsics allow identification of :ref:`GC roots on the
6520 stack <int_gcroot>`, as well as garbage collector implementations that
6521 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6522 Front-ends for type-safe garbage collected languages should generate
6523 these intrinsics to make use of the LLVM garbage collectors. For more
6524 details, see `Accurate Garbage Collection with
6525 LLVM <GarbageCollection.html>`_.
6527 The garbage collection intrinsics only operate on objects in the generic
6528 address space (address space zero).
6532 '``llvm.gcroot``' Intrinsic
6533 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6540 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6545 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6546 the code generator, and allows some metadata to be associated with it.
6551 The first argument specifies the address of a stack object that contains
6552 the root pointer. The second pointer (which must be either a constant or
6553 a global value address) contains the meta-data to be associated with the
6559 At runtime, a call to this intrinsic stores a null pointer into the
6560 "ptrloc" location. At compile-time, the code generator generates
6561 information to allow the runtime to find the pointer at GC safe points.
6562 The '``llvm.gcroot``' intrinsic may only be used in a function which
6563 :ref:`specifies a GC algorithm <gc>`.
6567 '``llvm.gcread``' Intrinsic
6568 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6575 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6580 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6581 locations, allowing garbage collector implementations that require read
6587 The second argument is the address to read from, which should be an
6588 address allocated from the garbage collector. The first object is a
6589 pointer to the start of the referenced object, if needed by the language
6590 runtime (otherwise null).
6595 The '``llvm.gcread``' intrinsic has the same semantics as a load
6596 instruction, but may be replaced with substantially more complex code by
6597 the garbage collector runtime, as needed. The '``llvm.gcread``'
6598 intrinsic may only be used in a function which :ref:`specifies a GC
6603 '``llvm.gcwrite``' Intrinsic
6604 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6611 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6616 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6617 locations, allowing garbage collector implementations that require write
6618 barriers (such as generational or reference counting collectors).
6623 The first argument is the reference to store, the second is the start of
6624 the object to store it to, and the third is the address of the field of
6625 Obj to store to. If the runtime does not require a pointer to the
6626 object, Obj may be null.
6631 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6632 instruction, but may be replaced with substantially more complex code by
6633 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6634 intrinsic may only be used in a function which :ref:`specifies a GC
6637 Code Generator Intrinsics
6638 -------------------------
6640 These intrinsics are provided by LLVM to expose special features that
6641 may only be implemented with code generator support.
6643 '``llvm.returnaddress``' Intrinsic
6644 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6651 declare i8 *@llvm.returnaddress(i32 <level>)
6656 The '``llvm.returnaddress``' intrinsic attempts to compute a
6657 target-specific value indicating the return address of the current
6658 function or one of its callers.
6663 The argument to this intrinsic indicates which function to return the
6664 address for. Zero indicates the calling function, one indicates its
6665 caller, etc. The argument is **required** to be a constant integer
6671 The '``llvm.returnaddress``' intrinsic either returns a pointer
6672 indicating the return address of the specified call frame, or zero if it
6673 cannot be identified. The value returned by this intrinsic is likely to
6674 be incorrect or 0 for arguments other than zero, so it should only be
6675 used for debugging purposes.
6677 Note that calling this intrinsic does not prevent function inlining or
6678 other aggressive transformations, so the value returned may not be that
6679 of the obvious source-language caller.
6681 '``llvm.frameaddress``' Intrinsic
6682 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6689 declare i8* @llvm.frameaddress(i32 <level>)
6694 The '``llvm.frameaddress``' intrinsic attempts to return the
6695 target-specific frame pointer value for the specified stack frame.
6700 The argument to this intrinsic indicates which function to return the
6701 frame pointer for. Zero indicates the calling function, one indicates
6702 its caller, etc. The argument is **required** to be a constant integer
6708 The '``llvm.frameaddress``' intrinsic either returns a pointer
6709 indicating the frame address of the specified call frame, or zero if it
6710 cannot be identified. The value returned by this intrinsic is likely to
6711 be incorrect or 0 for arguments other than zero, so it should only be
6712 used for debugging purposes.
6714 Note that calling this intrinsic does not prevent function inlining or
6715 other aggressive transformations, so the value returned may not be that
6716 of the obvious source-language caller.
6720 '``llvm.stacksave``' Intrinsic
6721 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6728 declare i8* @llvm.stacksave()
6733 The '``llvm.stacksave``' intrinsic is used to remember the current state
6734 of the function stack, for use with
6735 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6736 implementing language features like scoped automatic variable sized
6742 This intrinsic returns a opaque pointer value that can be passed to
6743 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6744 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6745 ``llvm.stacksave``, it effectively restores the state of the stack to
6746 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6747 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6748 were allocated after the ``llvm.stacksave`` was executed.
6750 .. _int_stackrestore:
6752 '``llvm.stackrestore``' Intrinsic
6753 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6760 declare void @llvm.stackrestore(i8* %ptr)
6765 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6766 the function stack to the state it was in when the corresponding
6767 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6768 useful for implementing language features like scoped automatic variable
6769 sized arrays in C99.
6774 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6776 '``llvm.prefetch``' Intrinsic
6777 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6784 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6789 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6790 insert a prefetch instruction if supported; otherwise, it is a noop.
6791 Prefetches have no effect on the behavior of the program but can change
6792 its performance characteristics.
6797 ``address`` is the address to be prefetched, ``rw`` is the specifier
6798 determining if the fetch should be for a read (0) or write (1), and
6799 ``locality`` is a temporal locality specifier ranging from (0) - no
6800 locality, to (3) - extremely local keep in cache. The ``cache type``
6801 specifies whether the prefetch is performed on the data (1) or
6802 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6803 arguments must be constant integers.
6808 This intrinsic does not modify the behavior of the program. In
6809 particular, prefetches cannot trap and do not produce a value. On
6810 targets that support this intrinsic, the prefetch can provide hints to
6811 the processor cache for better performance.
6813 '``llvm.pcmarker``' Intrinsic
6814 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6821 declare void @llvm.pcmarker(i32 <id>)
6826 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6827 Counter (PC) in a region of code to simulators and other tools. The
6828 method is target specific, but it is expected that the marker will use
6829 exported symbols to transmit the PC of the marker. The marker makes no
6830 guarantees that it will remain with any specific instruction after
6831 optimizations. It is possible that the presence of a marker will inhibit
6832 optimizations. The intended use is to be inserted after optimizations to
6833 allow correlations of simulation runs.
6838 ``id`` is a numerical id identifying the marker.
6843 This intrinsic does not modify the behavior of the program. Backends
6844 that do not support this intrinsic may ignore it.
6846 '``llvm.readcyclecounter``' Intrinsic
6847 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6854 declare i64 @llvm.readcyclecounter()
6859 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6860 counter register (or similar low latency, high accuracy clocks) on those
6861 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6862 should map to RPCC. As the backing counters overflow quickly (on the
6863 order of 9 seconds on alpha), this should only be used for small
6869 When directly supported, reading the cycle counter should not modify any
6870 memory. Implementations are allowed to either return a application
6871 specific value or a system wide value. On backends without support, this
6872 is lowered to a constant 0.
6874 Note that runtime support may be conditional on the privilege-level code is
6875 running at and the host platform.
6877 Standard C Library Intrinsics
6878 -----------------------------
6880 LLVM provides intrinsics for a few important standard C library
6881 functions. These intrinsics allow source-language front-ends to pass
6882 information about the alignment of the pointer arguments to the code
6883 generator, providing opportunity for more efficient code generation.
6887 '``llvm.memcpy``' Intrinsic
6888 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6893 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6894 integer bit width and for different address spaces. Not all targets
6895 support all bit widths however.
6899 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6900 i32 <len>, i32 <align>, i1 <isvolatile>)
6901 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6902 i64 <len>, i32 <align>, i1 <isvolatile>)
6907 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6908 source location to the destination location.
6910 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6911 intrinsics do not return a value, takes extra alignment/isvolatile
6912 arguments and the pointers can be in specified address spaces.
6917 The first argument is a pointer to the destination, the second is a
6918 pointer to the source. The third argument is an integer argument
6919 specifying the number of bytes to copy, the fourth argument is the
6920 alignment of the source and destination locations, and the fifth is a
6921 boolean indicating a volatile access.
6923 If the call to this intrinsic has an alignment value that is not 0 or 1,
6924 then the caller guarantees that both the source and destination pointers
6925 are aligned to that boundary.
6927 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6928 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6929 very cleanly specified and it is unwise to depend on it.
6934 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6935 source location to the destination location, which are not allowed to
6936 overlap. It copies "len" bytes of memory over. If the argument is known
6937 to be aligned to some boundary, this can be specified as the fourth
6938 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
6940 '``llvm.memmove``' Intrinsic
6941 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6946 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6947 bit width and for different address space. Not all targets support all
6952 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6953 i32 <len>, i32 <align>, i1 <isvolatile>)
6954 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6955 i64 <len>, i32 <align>, i1 <isvolatile>)
6960 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6961 source location to the destination location. It is similar to the
6962 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6965 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6966 intrinsics do not return a value, takes extra alignment/isvolatile
6967 arguments and the pointers can be in specified address spaces.
6972 The first argument is a pointer to the destination, the second is a
6973 pointer to the source. The third argument is an integer argument
6974 specifying the number of bytes to copy, the fourth argument is the
6975 alignment of the source and destination locations, and the fifth is a
6976 boolean indicating a volatile access.
6978 If the call to this intrinsic has an alignment value that is not 0 or 1,
6979 then the caller guarantees that the source and destination pointers are
6980 aligned to that boundary.
6982 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6983 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6984 not very cleanly specified and it is unwise to depend on it.
6989 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6990 source location to the destination location, which may overlap. It
6991 copies "len" bytes of memory over. If the argument is known to be
6992 aligned to some boundary, this can be specified as the fourth argument,
6993 otherwise it should be set to 0 or 1 (both meaning no alignment).
6995 '``llvm.memset.*``' Intrinsics
6996 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7001 This is an overloaded intrinsic. You can use llvm.memset on any integer
7002 bit width and for different address spaces. However, not all targets
7003 support all bit widths.
7007 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7008 i32 <len>, i32 <align>, i1 <isvolatile>)
7009 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7010 i64 <len>, i32 <align>, i1 <isvolatile>)
7015 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7016 particular byte value.
7018 Note that, unlike the standard libc function, the ``llvm.memset``
7019 intrinsic does not return a value and takes extra alignment/volatile
7020 arguments. Also, the destination can be in an arbitrary address space.
7025 The first argument is a pointer to the destination to fill, the second
7026 is the byte value with which to fill it, the third argument is an
7027 integer argument specifying the number of bytes to fill, and the fourth
7028 argument is the known alignment of the destination location.
7030 If the call to this intrinsic has an alignment value that is not 0 or 1,
7031 then the caller guarantees that the destination pointer is aligned to
7034 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7035 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7036 very cleanly specified and it is unwise to depend on it.
7041 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7042 at the destination location. If the argument is known to be aligned to
7043 some boundary, this can be specified as the fourth argument, otherwise
7044 it should be set to 0 or 1 (both meaning no alignment).
7046 '``llvm.sqrt.*``' Intrinsic
7047 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7052 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7053 floating point or vector of floating point type. Not all targets support
7058 declare float @llvm.sqrt.f32(float %Val)
7059 declare double @llvm.sqrt.f64(double %Val)
7060 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7061 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7062 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7067 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7068 returning the same value as the libm '``sqrt``' functions would. Unlike
7069 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7070 negative numbers other than -0.0 (which allows for better optimization,
7071 because there is no need to worry about errno being set).
7072 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7077 The argument and return value are floating point numbers of the same
7083 This function returns the sqrt of the specified operand if it is a
7084 nonnegative floating point number.
7086 '``llvm.powi.*``' Intrinsic
7087 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7092 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7093 floating point or vector of floating point type. Not all targets support
7098 declare float @llvm.powi.f32(float %Val, i32 %power)
7099 declare double @llvm.powi.f64(double %Val, i32 %power)
7100 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7101 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7102 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7107 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7108 specified (positive or negative) power. The order of evaluation of
7109 multiplications is not defined. When a vector of floating point type is
7110 used, the second argument remains a scalar integer value.
7115 The second argument is an integer power, and the first is a value to
7116 raise to that power.
7121 This function returns the first value raised to the second power with an
7122 unspecified sequence of rounding operations.
7124 '``llvm.sin.*``' Intrinsic
7125 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7130 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7131 floating point or vector of floating point type. Not all targets support
7136 declare float @llvm.sin.f32(float %Val)
7137 declare double @llvm.sin.f64(double %Val)
7138 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7139 declare fp128 @llvm.sin.f128(fp128 %Val)
7140 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7145 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7150 The argument and return value are floating point numbers of the same
7156 This function returns the sine of the specified operand, returning the
7157 same values as the libm ``sin`` functions would, and handles error
7158 conditions in the same way.
7160 '``llvm.cos.*``' Intrinsic
7161 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7166 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7167 floating point or vector of floating point type. Not all targets support
7172 declare float @llvm.cos.f32(float %Val)
7173 declare double @llvm.cos.f64(double %Val)
7174 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7175 declare fp128 @llvm.cos.f128(fp128 %Val)
7176 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7181 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7186 The argument and return value are floating point numbers of the same
7192 This function returns the cosine of the specified operand, returning the
7193 same values as the libm ``cos`` functions would, and handles error
7194 conditions in the same way.
7196 '``llvm.pow.*``' Intrinsic
7197 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7202 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7203 floating point or vector of floating point type. Not all targets support
7208 declare float @llvm.pow.f32(float %Val, float %Power)
7209 declare double @llvm.pow.f64(double %Val, double %Power)
7210 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7211 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7212 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7217 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7218 specified (positive or negative) power.
7223 The second argument is a floating point power, and the first is a value
7224 to raise to that power.
7229 This function returns the first value raised to the second power,
7230 returning the same values as the libm ``pow`` functions would, and
7231 handles error conditions in the same way.
7233 '``llvm.exp.*``' Intrinsic
7234 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7239 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7240 floating point or vector of floating point type. Not all targets support
7245 declare float @llvm.exp.f32(float %Val)
7246 declare double @llvm.exp.f64(double %Val)
7247 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7248 declare fp128 @llvm.exp.f128(fp128 %Val)
7249 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7254 The '``llvm.exp.*``' intrinsics perform the exp function.
7259 The argument and return value are floating point numbers of the same
7265 This function returns the same values as the libm ``exp`` functions
7266 would, and handles error conditions in the same way.
7268 '``llvm.exp2.*``' Intrinsic
7269 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7274 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7275 floating point or vector of floating point type. Not all targets support
7280 declare float @llvm.exp2.f32(float %Val)
7281 declare double @llvm.exp2.f64(double %Val)
7282 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7283 declare fp128 @llvm.exp2.f128(fp128 %Val)
7284 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7289 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7294 The argument and return value are floating point numbers of the same
7300 This function returns the same values as the libm ``exp2`` functions
7301 would, and handles error conditions in the same way.
7303 '``llvm.log.*``' Intrinsic
7304 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7309 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7310 floating point or vector of floating point type. Not all targets support
7315 declare float @llvm.log.f32(float %Val)
7316 declare double @llvm.log.f64(double %Val)
7317 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7318 declare fp128 @llvm.log.f128(fp128 %Val)
7319 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7324 The '``llvm.log.*``' intrinsics perform the log function.
7329 The argument and return value are floating point numbers of the same
7335 This function returns the same values as the libm ``log`` functions
7336 would, and handles error conditions in the same way.
7338 '``llvm.log10.*``' Intrinsic
7339 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7344 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7345 floating point or vector of floating point type. Not all targets support
7350 declare float @llvm.log10.f32(float %Val)
7351 declare double @llvm.log10.f64(double %Val)
7352 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7353 declare fp128 @llvm.log10.f128(fp128 %Val)
7354 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7359 The '``llvm.log10.*``' intrinsics perform the log10 function.
7364 The argument and return value are floating point numbers of the same
7370 This function returns the same values as the libm ``log10`` functions
7371 would, and handles error conditions in the same way.
7373 '``llvm.log2.*``' Intrinsic
7374 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7379 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7380 floating point or vector of floating point type. Not all targets support
7385 declare float @llvm.log2.f32(float %Val)
7386 declare double @llvm.log2.f64(double %Val)
7387 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7388 declare fp128 @llvm.log2.f128(fp128 %Val)
7389 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7394 The '``llvm.log2.*``' intrinsics perform the log2 function.
7399 The argument and return value are floating point numbers of the same
7405 This function returns the same values as the libm ``log2`` functions
7406 would, and handles error conditions in the same way.
7408 '``llvm.fma.*``' Intrinsic
7409 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7414 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7415 floating point or vector of floating point type. Not all targets support
7420 declare float @llvm.fma.f32(float %a, float %b, float %c)
7421 declare double @llvm.fma.f64(double %a, double %b, double %c)
7422 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7423 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7424 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7429 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7435 The argument and return value are floating point numbers of the same
7441 This function returns the same values as the libm ``fma`` functions
7444 '``llvm.fabs.*``' Intrinsic
7445 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7450 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7451 floating point or vector of floating point type. Not all targets support
7456 declare float @llvm.fabs.f32(float %Val)
7457 declare double @llvm.fabs.f64(double %Val)
7458 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7459 declare fp128 @llvm.fabs.f128(fp128 %Val)
7460 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7465 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7471 The argument and return value are floating point numbers of the same
7477 This function returns the same values as the libm ``fabs`` functions
7478 would, and handles error conditions in the same way.
7480 '``llvm.copysign.*``' Intrinsic
7481 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7486 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7487 floating point or vector of floating point type. Not all targets support
7492 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7493 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7494 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7495 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7496 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7501 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7502 first operand and the sign of the second operand.
7507 The arguments and return value are floating point numbers of the same
7513 This function returns the same values as the libm ``copysign``
7514 functions would, and handles error conditions in the same way.
7516 '``llvm.floor.*``' Intrinsic
7517 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7522 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7523 floating point or vector of floating point type. Not all targets support
7528 declare float @llvm.floor.f32(float %Val)
7529 declare double @llvm.floor.f64(double %Val)
7530 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7531 declare fp128 @llvm.floor.f128(fp128 %Val)
7532 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7537 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7542 The argument and return value are floating point numbers of the same
7548 This function returns the same values as the libm ``floor`` functions
7549 would, and handles error conditions in the same way.
7551 '``llvm.ceil.*``' Intrinsic
7552 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7557 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7558 floating point or vector of floating point type. Not all targets support
7563 declare float @llvm.ceil.f32(float %Val)
7564 declare double @llvm.ceil.f64(double %Val)
7565 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7566 declare fp128 @llvm.ceil.f128(fp128 %Val)
7567 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7572 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7577 The argument and return value are floating point numbers of the same
7583 This function returns the same values as the libm ``ceil`` functions
7584 would, and handles error conditions in the same way.
7586 '``llvm.trunc.*``' Intrinsic
7587 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7592 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7593 floating point or vector of floating point type. Not all targets support
7598 declare float @llvm.trunc.f32(float %Val)
7599 declare double @llvm.trunc.f64(double %Val)
7600 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7601 declare fp128 @llvm.trunc.f128(fp128 %Val)
7602 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7607 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7608 nearest integer not larger in magnitude than the operand.
7613 The argument and return value are floating point numbers of the same
7619 This function returns the same values as the libm ``trunc`` functions
7620 would, and handles error conditions in the same way.
7622 '``llvm.rint.*``' Intrinsic
7623 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7628 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7629 floating point or vector of floating point type. Not all targets support
7634 declare float @llvm.rint.f32(float %Val)
7635 declare double @llvm.rint.f64(double %Val)
7636 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7637 declare fp128 @llvm.rint.f128(fp128 %Val)
7638 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7643 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7644 nearest integer. It may raise an inexact floating-point exception if the
7645 operand isn't an integer.
7650 The argument and return value are floating point numbers of the same
7656 This function returns the same values as the libm ``rint`` functions
7657 would, and handles error conditions in the same way.
7659 '``llvm.nearbyint.*``' Intrinsic
7660 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7665 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7666 floating point or vector of floating point type. Not all targets support
7671 declare float @llvm.nearbyint.f32(float %Val)
7672 declare double @llvm.nearbyint.f64(double %Val)
7673 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7674 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7675 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7680 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7686 The argument and return value are floating point numbers of the same
7692 This function returns the same values as the libm ``nearbyint``
7693 functions would, and handles error conditions in the same way.
7695 '``llvm.round.*``' Intrinsic
7696 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7701 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7702 floating point or vector of floating point type. Not all targets support
7707 declare float @llvm.round.f32(float %Val)
7708 declare double @llvm.round.f64(double %Val)
7709 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7710 declare fp128 @llvm.round.f128(fp128 %Val)
7711 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7716 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7722 The argument and return value are floating point numbers of the same
7728 This function returns the same values as the libm ``round``
7729 functions would, and handles error conditions in the same way.
7731 Bit Manipulation Intrinsics
7732 ---------------------------
7734 LLVM provides intrinsics for a few important bit manipulation
7735 operations. These allow efficient code generation for some algorithms.
7737 '``llvm.bswap.*``' Intrinsics
7738 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7743 This is an overloaded intrinsic function. You can use bswap on any
7744 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7748 declare i16 @llvm.bswap.i16(i16 <id>)
7749 declare i32 @llvm.bswap.i32(i32 <id>)
7750 declare i64 @llvm.bswap.i64(i64 <id>)
7755 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7756 values with an even number of bytes (positive multiple of 16 bits).
7757 These are useful for performing operations on data that is not in the
7758 target's native byte order.
7763 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7764 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7765 intrinsic returns an i32 value that has the four bytes of the input i32
7766 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7767 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7768 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7769 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7772 '``llvm.ctpop.*``' Intrinsic
7773 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7778 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7779 bit width, or on any vector with integer elements. Not all targets
7780 support all bit widths or vector types, however.
7784 declare i8 @llvm.ctpop.i8(i8 <src>)
7785 declare i16 @llvm.ctpop.i16(i16 <src>)
7786 declare i32 @llvm.ctpop.i32(i32 <src>)
7787 declare i64 @llvm.ctpop.i64(i64 <src>)
7788 declare i256 @llvm.ctpop.i256(i256 <src>)
7789 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7794 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7800 The only argument is the value to be counted. The argument may be of any
7801 integer type, or a vector with integer elements. The return type must
7802 match the argument type.
7807 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7808 each element of a vector.
7810 '``llvm.ctlz.*``' Intrinsic
7811 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7816 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7817 integer bit width, or any vector whose elements are integers. Not all
7818 targets support all bit widths or vector types, however.
7822 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7823 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7824 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7825 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7826 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7827 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7832 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7833 leading zeros in a variable.
7838 The first argument is the value to be counted. This argument may be of
7839 any integer type, or a vectory with integer element type. The return
7840 type must match the first argument type.
7842 The second argument must be a constant and is a flag to indicate whether
7843 the intrinsic should ensure that a zero as the first argument produces a
7844 defined result. Historically some architectures did not provide a
7845 defined result for zero values as efficiently, and many algorithms are
7846 now predicated on avoiding zero-value inputs.
7851 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7852 zeros in a variable, or within each element of the vector. If
7853 ``src == 0`` then the result is the size in bits of the type of ``src``
7854 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7855 ``llvm.ctlz(i32 2) = 30``.
7857 '``llvm.cttz.*``' Intrinsic
7858 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7863 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7864 integer bit width, or any vector of integer elements. Not all targets
7865 support all bit widths or vector types, however.
7869 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7870 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7871 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7872 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7873 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7874 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7879 The '``llvm.cttz``' family of intrinsic functions counts the number of
7885 The first argument is the value to be counted. This argument may be of
7886 any integer type, or a vectory with integer element type. The return
7887 type must match the first argument type.
7889 The second argument must be a constant and is a flag to indicate whether
7890 the intrinsic should ensure that a zero as the first argument produces a
7891 defined result. Historically some architectures did not provide a
7892 defined result for zero values as efficiently, and many algorithms are
7893 now predicated on avoiding zero-value inputs.
7898 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7899 zeros in a variable, or within each element of a vector. If ``src == 0``
7900 then the result is the size in bits of the type of ``src`` if
7901 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7902 ``llvm.cttz(2) = 1``.
7904 Arithmetic with Overflow Intrinsics
7905 -----------------------------------
7907 LLVM provides intrinsics for some arithmetic with overflow operations.
7909 '``llvm.sadd.with.overflow.*``' Intrinsics
7910 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7915 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7916 on any integer bit width.
7920 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7921 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7922 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7927 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7928 a signed addition of the two arguments, and indicate whether an overflow
7929 occurred during the signed summation.
7934 The arguments (%a and %b) and the first element of the result structure
7935 may be of integer types of any bit width, but they must have the same
7936 bit width. The second element of the result structure must be of type
7937 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7943 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7944 a signed addition of the two variables. They return a structure --- the
7945 first element of which is the signed summation, and the second element
7946 of which is a bit specifying if the signed summation resulted in an
7952 .. code-block:: llvm
7954 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7955 %sum = extractvalue {i32, i1} %res, 0
7956 %obit = extractvalue {i32, i1} %res, 1
7957 br i1 %obit, label %overflow, label %normal
7959 '``llvm.uadd.with.overflow.*``' Intrinsics
7960 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7965 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7966 on any integer bit width.
7970 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7971 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7972 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7977 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7978 an unsigned addition of the two arguments, and indicate whether a carry
7979 occurred during the unsigned summation.
7984 The arguments (%a and %b) and the first element of the result structure
7985 may be of integer types of any bit width, but they must have the same
7986 bit width. The second element of the result structure must be of type
7987 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7993 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7994 an unsigned addition of the two arguments. They return a structure --- the
7995 first element of which is the sum, and the second element of which is a
7996 bit specifying if the unsigned summation resulted in a carry.
8001 .. code-block:: llvm
8003 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8004 %sum = extractvalue {i32, i1} %res, 0
8005 %obit = extractvalue {i32, i1} %res, 1
8006 br i1 %obit, label %carry, label %normal
8008 '``llvm.ssub.with.overflow.*``' Intrinsics
8009 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8014 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8015 on any integer bit width.
8019 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8020 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8021 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8026 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8027 a signed subtraction of the two arguments, and indicate whether an
8028 overflow occurred during the signed subtraction.
8033 The arguments (%a and %b) and the first element of the result structure
8034 may be of integer types of any bit width, but they must have the same
8035 bit width. The second element of the result structure must be of type
8036 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8042 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8043 a signed subtraction of the two arguments. They return a structure --- the
8044 first element of which is the subtraction, and the second element of
8045 which is a bit specifying if the signed subtraction resulted in an
8051 .. code-block:: llvm
8053 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8054 %sum = extractvalue {i32, i1} %res, 0
8055 %obit = extractvalue {i32, i1} %res, 1
8056 br i1 %obit, label %overflow, label %normal
8058 '``llvm.usub.with.overflow.*``' Intrinsics
8059 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8064 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8065 on any integer bit width.
8069 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8070 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8071 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8076 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8077 an unsigned subtraction of the two arguments, and indicate whether an
8078 overflow occurred during the unsigned subtraction.
8083 The arguments (%a and %b) and the first element of the result structure
8084 may be of integer types of any bit width, but they must have the same
8085 bit width. The second element of the result structure must be of type
8086 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8092 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8093 an unsigned subtraction of the two arguments. They return a structure ---
8094 the first element of which is the subtraction, and the second element of
8095 which is a bit specifying if the unsigned subtraction resulted in an
8101 .. code-block:: llvm
8103 %res = call {i32, i1} @llvm.usub.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 %overflow, label %normal
8108 '``llvm.smul.with.overflow.*``' Intrinsics
8109 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8114 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8115 on any integer bit width.
8119 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8120 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8121 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8126 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8127 a signed multiplication of the two arguments, and indicate whether an
8128 overflow occurred during the signed multiplication.
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.smul.with.overflow``' family of intrinsic functions perform
8143 a signed multiplication of the two arguments. They return a structure ---
8144 the first element of which is the multiplication, and the second element
8145 of which is a bit specifying if the signed multiplication resulted in an
8151 .. code-block:: llvm
8153 %res = call {i32, i1} @llvm.smul.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.umul.with.overflow.*``' Intrinsics
8159 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8164 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8165 on any integer bit width.
8169 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8170 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8171 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8176 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8177 a unsigned multiplication of the two arguments, and indicate whether an
8178 overflow occurred during the unsigned multiplication.
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.umul.with.overflow``' family of intrinsic functions perform
8193 an unsigned multiplication of the two arguments. They return a structure ---
8194 the first element of which is the multiplication, and the second
8195 element of which is a bit specifying if the unsigned multiplication
8196 resulted in an overflow.
8201 .. code-block:: llvm
8203 %res = call {i32, i1} @llvm.umul.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 Specialised Arithmetic Intrinsics
8209 ---------------------------------
8211 '``llvm.fmuladd.*``' Intrinsic
8212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8219 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8220 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8225 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8226 expressions that can be fused if the code generator determines that (a) the
8227 target instruction set has support for a fused operation, and (b) that the
8228 fused operation is more efficient than the equivalent, separate pair of mul
8229 and add instructions.
8234 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8235 multiplicands, a and b, and an addend c.
8244 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8246 is equivalent to the expression a \* b + c, except that rounding will
8247 not be performed between the multiplication and addition steps if the
8248 code generator fuses the operations. Fusion is not guaranteed, even if
8249 the target platform supports it. If a fused multiply-add is required the
8250 corresponding llvm.fma.\* intrinsic function should be used instead.
8255 .. code-block:: llvm
8257 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8259 Half Precision Floating Point Intrinsics
8260 ----------------------------------------
8262 For most target platforms, half precision floating point is a
8263 storage-only format. This means that it is a dense encoding (in memory)
8264 but does not support computation in the format.
8266 This means that code must first load the half-precision floating point
8267 value as an i16, then convert it to float with
8268 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8269 then be performed on the float value (including extending to double
8270 etc). To store the value back to memory, it is first converted to float
8271 if needed, then converted to i16 with
8272 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8275 .. _int_convert_to_fp16:
8277 '``llvm.convert.to.fp16``' Intrinsic
8278 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8285 declare i16 @llvm.convert.to.fp16(f32 %a)
8290 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8291 from single precision floating point format to half precision floating
8297 The intrinsic function contains single argument - the value to be
8303 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8304 from single precision floating point format to half precision floating
8305 point format. The return value is an ``i16`` which contains the
8311 .. code-block:: llvm
8313 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8314 store i16 %res, i16* @x, align 2
8316 .. _int_convert_from_fp16:
8318 '``llvm.convert.from.fp16``' Intrinsic
8319 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8326 declare f32 @llvm.convert.from.fp16(i16 %a)
8331 The '``llvm.convert.from.fp16``' intrinsic function performs a
8332 conversion from half precision floating point format to single precision
8333 floating point format.
8338 The intrinsic function contains single argument - the value to be
8344 The '``llvm.convert.from.fp16``' intrinsic function performs a
8345 conversion from half single precision floating point format to single
8346 precision floating point format. The input half-float value is
8347 represented by an ``i16`` value.
8352 .. code-block:: llvm
8354 %a = load i16* @x, align 2
8355 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8360 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8361 prefix), are described in the `LLVM Source Level
8362 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8365 Exception Handling Intrinsics
8366 -----------------------------
8368 The LLVM exception handling intrinsics (which all start with
8369 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8370 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8374 Trampoline Intrinsics
8375 ---------------------
8377 These intrinsics make it possible to excise one parameter, marked with
8378 the :ref:`nest <nest>` attribute, from a function. The result is a
8379 callable function pointer lacking the nest parameter - the caller does
8380 not need to provide a value for it. Instead, the value to use is stored
8381 in advance in a "trampoline", a block of memory usually allocated on the
8382 stack, which also contains code to splice the nest value into the
8383 argument list. This is used to implement the GCC nested function address
8386 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8387 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8388 It can be created as follows:
8390 .. code-block:: llvm
8392 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8393 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8394 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8395 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8396 %fp = bitcast i8* %p to i32 (i32, i32)*
8398 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8399 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8403 '``llvm.init.trampoline``' Intrinsic
8404 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8411 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8416 This fills the memory pointed to by ``tramp`` with executable code,
8417 turning it into a trampoline.
8422 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8423 pointers. The ``tramp`` argument must point to a sufficiently large and
8424 sufficiently aligned block of memory; this memory is written to by the
8425 intrinsic. Note that the size and the alignment are target-specific -
8426 LLVM currently provides no portable way of determining them, so a
8427 front-end that generates this intrinsic needs to have some
8428 target-specific knowledge. The ``func`` argument must hold a function
8429 bitcast to an ``i8*``.
8434 The block of memory pointed to by ``tramp`` is filled with target
8435 dependent code, turning it into a function. Then ``tramp`` needs to be
8436 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8437 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8438 function's signature is the same as that of ``func`` with any arguments
8439 marked with the ``nest`` attribute removed. At most one such ``nest``
8440 argument is allowed, and it must be of pointer type. Calling the new
8441 function is equivalent to calling ``func`` with the same argument list,
8442 but with ``nval`` used for the missing ``nest`` argument. If, after
8443 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8444 modified, then the effect of any later call to the returned function
8445 pointer is undefined.
8449 '``llvm.adjust.trampoline``' Intrinsic
8450 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8457 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8462 This performs any required machine-specific adjustment to the address of
8463 a trampoline (passed as ``tramp``).
8468 ``tramp`` must point to a block of memory which already has trampoline
8469 code filled in by a previous call to
8470 :ref:`llvm.init.trampoline <int_it>`.
8475 On some architectures the address of the code to be executed needs to be
8476 different to the address where the trampoline is actually stored. This
8477 intrinsic returns the executable address corresponding to ``tramp``
8478 after performing the required machine specific adjustments. The pointer
8479 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8484 This class of intrinsics exists to information about the lifetime of
8485 memory objects and ranges where variables are immutable.
8489 '``llvm.lifetime.start``' Intrinsic
8490 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8497 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8502 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8508 The first argument is a constant integer representing the size of the
8509 object, or -1 if it is variable sized. The second argument is a pointer
8515 This intrinsic indicates that before this point in the code, the value
8516 of the memory pointed to by ``ptr`` is dead. This means that it is known
8517 to never be used and has an undefined value. A load from the pointer
8518 that precedes this intrinsic can be replaced with ``'undef'``.
8522 '``llvm.lifetime.end``' Intrinsic
8523 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8530 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8535 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8541 The first argument is a constant integer representing the size of the
8542 object, or -1 if it is variable sized. The second argument is a pointer
8548 This intrinsic indicates that after this point in the code, the value of
8549 the memory pointed to by ``ptr`` is dead. This means that it is known to
8550 never be used and has an undefined value. Any stores into the memory
8551 object following this intrinsic may be removed as dead.
8553 '``llvm.invariant.start``' Intrinsic
8554 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8561 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8566 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8567 a memory object will not change.
8572 The first argument is a constant integer representing the size of the
8573 object, or -1 if it is variable sized. The second argument is a pointer
8579 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8580 the return value, the referenced memory location is constant and
8583 '``llvm.invariant.end``' Intrinsic
8584 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8591 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8596 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8597 memory object are mutable.
8602 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8603 The second argument is a constant integer representing the size of the
8604 object, or -1 if it is variable sized and the third argument is a
8605 pointer to the object.
8610 This intrinsic indicates that the memory is mutable again.
8615 This class of intrinsics is designed to be generic and has no specific
8618 '``llvm.var.annotation``' Intrinsic
8619 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8626 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8631 The '``llvm.var.annotation``' intrinsic.
8636 The first argument is a pointer to a value, the second is a pointer to a
8637 global string, the third is a pointer to a global string which is the
8638 source file name, and the last argument is the line number.
8643 This intrinsic allows annotation of local variables with arbitrary
8644 strings. This can be useful for special purpose optimizations that want
8645 to look for these annotations. These have no other defined use; they are
8646 ignored by code generation and optimization.
8648 '``llvm.ptr.annotation.*``' Intrinsic
8649 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8654 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8655 pointer to an integer of any width. *NOTE* you must specify an address space for
8656 the pointer. The identifier for the default address space is the integer
8661 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8662 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8663 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8664 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8665 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8670 The '``llvm.ptr.annotation``' intrinsic.
8675 The first argument is a pointer to an integer value of arbitrary bitwidth
8676 (result of some expression), the second is a pointer to a global string, the
8677 third is a pointer to a global string which is the source file name, and the
8678 last argument is the line number. It returns the value of the first argument.
8683 This intrinsic allows annotation of a pointer to an integer with arbitrary
8684 strings. This can be useful for special purpose optimizations that want to look
8685 for these annotations. These have no other defined use; they are ignored by code
8686 generation and optimization.
8688 '``llvm.annotation.*``' Intrinsic
8689 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8694 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8695 any integer bit width.
8699 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8700 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8701 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8702 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8703 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8708 The '``llvm.annotation``' intrinsic.
8713 The first argument is an integer value (result of some expression), the
8714 second is a pointer to a global string, the third is a pointer to a
8715 global string which is the source file name, and the last argument is
8716 the line number. It returns the value of the first argument.
8721 This intrinsic allows annotations to be put on arbitrary expressions
8722 with arbitrary strings. This can be useful for special purpose
8723 optimizations that want to look for these annotations. These have no
8724 other defined use; they are ignored by code generation and optimization.
8726 '``llvm.trap``' Intrinsic
8727 ^^^^^^^^^^^^^^^^^^^^^^^^^
8734 declare void @llvm.trap() noreturn nounwind
8739 The '``llvm.trap``' intrinsic.
8749 This intrinsic is lowered to the target dependent trap instruction. If
8750 the target does not have a trap instruction, this intrinsic will be
8751 lowered to a call of the ``abort()`` function.
8753 '``llvm.debugtrap``' Intrinsic
8754 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8761 declare void @llvm.debugtrap() nounwind
8766 The '``llvm.debugtrap``' intrinsic.
8776 This intrinsic is lowered to code which is intended to cause an
8777 execution trap with the intention of requesting the attention of a
8780 '``llvm.stackprotector``' Intrinsic
8781 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8788 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8793 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8794 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8795 is placed on the stack before local variables.
8800 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8801 The first argument is the value loaded from the stack guard
8802 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8803 enough space to hold the value of the guard.
8808 This intrinsic causes the prologue/epilogue inserter to force the position of
8809 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8810 to ensure that if a local variable on the stack is overwritten, it will destroy
8811 the value of the guard. When the function exits, the guard on the stack is
8812 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8813 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8814 calling the ``__stack_chk_fail()`` function.
8816 '``llvm.stackprotectorcheck``' Intrinsic
8817 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8824 declare void @llvm.stackprotectorcheck(i8** <guard>)
8829 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8830 created stack protector and if they are not equal calls the
8831 ``__stack_chk_fail()`` function.
8836 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8837 the variable ``@__stack_chk_guard``.
8842 This intrinsic is provided to perform the stack protector check by comparing
8843 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8844 values do not match call the ``__stack_chk_fail()`` function.
8846 The reason to provide this as an IR level intrinsic instead of implementing it
8847 via other IR operations is that in order to perform this operation at the IR
8848 level without an intrinsic, one would need to create additional basic blocks to
8849 handle the success/failure cases. This makes it difficult to stop the stack
8850 protector check from disrupting sibling tail calls in Codegen. With this
8851 intrinsic, we are able to generate the stack protector basic blocks late in
8852 codegen after the tail call decision has occurred.
8854 '``llvm.objectsize``' Intrinsic
8855 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8862 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8863 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8868 The ``llvm.objectsize`` intrinsic is designed to provide information to
8869 the optimizers to determine at compile time whether a) an operation
8870 (like memcpy) will overflow a buffer that corresponds to an object, or
8871 b) that a runtime check for overflow isn't necessary. An object in this
8872 context means an allocation of a specific class, structure, array, or
8878 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8879 argument is a pointer to or into the ``object``. The second argument is
8880 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8881 or -1 (if false) when the object size is unknown. The second argument
8882 only accepts constants.
8887 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8888 the size of the object concerned. If the size cannot be determined at
8889 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8890 on the ``min`` argument).
8892 '``llvm.expect``' Intrinsic
8893 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8900 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8901 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8906 The ``llvm.expect`` intrinsic provides information about expected (the
8907 most probable) value of ``val``, which can be used by optimizers.
8912 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8913 a value. The second argument is an expected value, this needs to be a
8914 constant value, variables are not allowed.
8919 This intrinsic is lowered to the ``val``.
8921 '``llvm.donothing``' Intrinsic
8922 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8929 declare void @llvm.donothing() nounwind readnone
8934 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8935 only intrinsic that can be called with an invoke instruction.
8945 This intrinsic does nothing, and it's removed by optimizers and ignored
8948 Stack Map Intrinsics
8949 --------------------
8951 LLVM provides experimental intrinsics to support runtime patching
8952 mechanisms commonly desired in dynamic language JITs. These intrinsics
8953 are described in :doc:`StackMaps`.