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 take the
731 address of all stack-allocated arguments to a ``call`` or ``invoke``
732 before it executes. It is similar to ``byval`` in that it is used
733 to pass arguments by value, but it guarantees that the argument will
736 To be :ref:`well formed <wellformed>`, an alloca may be used as an
737 ``inalloca`` argument at most once. The attribute can only be
738 applied to the last parameter, and it guarantees that they are
739 passed in memory. The ``inalloca`` attribute cannot be used in
740 conjunction with other attributes that affect argument storage, like
741 ``inreg``, ``nest``, ``sret``, or ``byval``. The ``inalloca`` stack
742 space is considered to be clobbered by any call that uses it, so any
743 ``inalloca`` parameters cannot be marked ``readonly``.
745 When the call site is reached, the argument allocation must have
746 been the most recent stack allocation that is still live, or the
747 results are undefined. It is possible to allocate additional stack
748 space after an argument allocation and before its call site, but it
749 must be cleared off with :ref:`llvm.stackrestore
752 See :doc:`InAlloca` for more information on how to use this
756 This indicates that the pointer parameter specifies the address of a
757 structure that is the return value of the function in the source
758 program. This pointer must be guaranteed by the caller to be valid:
759 loads and stores to the structure may be assumed by the callee
760 not to trap and to be properly aligned. This may only be applied to
761 the first parameter. This is not a valid attribute for return
764 This indicates that pointer values :ref:`based <pointeraliasing>` on
765 the argument or return value do not alias pointer values which are
766 not *based* on it, ignoring certain "irrelevant" dependencies. For a
767 call to the parent function, dependencies between memory references
768 from before or after the call and from those during the call are
769 "irrelevant" to the ``noalias`` keyword for the arguments and return
770 value used in that call. The caller shares the responsibility with
771 the callee for ensuring that these requirements are met. For further
772 details, please see the discussion of the NoAlias response in `alias
773 analysis <AliasAnalysis.html#MustMayNo>`_.
775 Note that this definition of ``noalias`` is intentionally similar
776 to the definition of ``restrict`` in C99 for function arguments,
777 though it is slightly weaker.
779 For function return values, C99's ``restrict`` is not meaningful,
780 while LLVM's ``noalias`` is.
782 This indicates that the callee does not make any copies of the
783 pointer that outlive the callee itself. This is not a valid
784 attribute for return values.
789 This indicates that the pointer parameter can be excised using the
790 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
791 attribute for return values and can only be applied to one parameter.
794 This indicates that the function always returns the argument as its return
795 value. This is an optimization hint to the code generator when generating
796 the caller, allowing tail call optimization and omission of register saves
797 and restores in some cases; it is not checked or enforced when generating
798 the callee. The parameter and the function return type must be valid
799 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
800 valid attribute for return values and can only be applied to one parameter.
804 Garbage Collector Names
805 -----------------------
807 Each function may specify a garbage collector name, which is simply a
812 define void @f() gc "name" { ... }
814 The compiler declares the supported values of *name*. Specifying a
815 collector which will cause the compiler to alter its output in order to
816 support the named garbage collection algorithm.
823 Prefix data is data associated with a function which the code generator
824 will emit immediately before the function body. The purpose of this feature
825 is to allow frontends to associate language-specific runtime metadata with
826 specific functions and make it available through the function pointer while
827 still allowing the function pointer to be called. To access the data for a
828 given function, a program may bitcast the function pointer to a pointer to
829 the constant's type. This implies that the IR symbol points to the start
832 To maintain the semantics of ordinary function calls, the prefix data must
833 have a particular format. Specifically, it must begin with a sequence of
834 bytes which decode to a sequence of machine instructions, valid for the
835 module's target, which transfer control to the point immediately succeeding
836 the prefix data, without performing any other visible action. This allows
837 the inliner and other passes to reason about the semantics of the function
838 definition without needing to reason about the prefix data. Obviously this
839 makes the format of the prefix data highly target dependent.
841 Prefix data is laid out as if it were an initializer for a global variable
842 of the prefix data's type. No padding is automatically placed between the
843 prefix data and the function body. If padding is required, it must be part
846 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
847 which encodes the ``nop`` instruction:
851 define void @f() prefix i8 144 { ... }
853 Generally prefix data can be formed by encoding a relative branch instruction
854 which skips the metadata, as in this example of valid prefix data for the
855 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
859 %0 = type <{ i8, i8, i8* }>
861 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
863 A function may have prefix data but no body. This has similar semantics
864 to the ``available_externally`` linkage in that the data may be used by the
865 optimizers but will not be emitted in the object file.
872 Attribute groups are groups of attributes that are referenced by objects within
873 the IR. They are important for keeping ``.ll`` files readable, because a lot of
874 functions will use the same set of attributes. In the degenerative case of a
875 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
876 group will capture the important command line flags used to build that file.
878 An attribute group is a module-level object. To use an attribute group, an
879 object references the attribute group's ID (e.g. ``#37``). An object may refer
880 to more than one attribute group. In that situation, the attributes from the
881 different groups are merged.
883 Here is an example of attribute groups for a function that should always be
884 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
888 ; Target-independent attributes:
889 attributes #0 = { alwaysinline alignstack=4 }
891 ; Target-dependent attributes:
892 attributes #1 = { "no-sse" }
894 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
895 define void @f() #0 #1 { ... }
902 Function attributes are set to communicate additional information about
903 a function. Function attributes are considered to be part of the
904 function, not of the function type, so functions with different function
905 attributes can have the same function type.
907 Function attributes are simple keywords that follow the type specified.
908 If multiple attributes are needed, they are space separated. For
913 define void @f() noinline { ... }
914 define void @f() alwaysinline { ... }
915 define void @f() alwaysinline optsize { ... }
916 define void @f() optsize { ... }
919 This attribute indicates that, when emitting the prologue and
920 epilogue, the backend should forcibly align the stack pointer.
921 Specify the desired alignment, which must be a power of two, in
924 This attribute indicates that the inliner should attempt to inline
925 this function into callers whenever possible, ignoring any active
926 inlining size threshold for this caller.
928 This indicates that the callee function at a call site should be
929 recognized as a built-in function, even though the function's declaration
930 uses the ``nobuiltin`` attribute. This is only valid at call sites for
931 direct calls to functions which are declared with the ``nobuiltin``
934 This attribute indicates that this function is rarely called. When
935 computing edge weights, basic blocks post-dominated by a cold
936 function call are also considered to be cold; and, thus, given low
939 This attribute indicates that the source code contained a hint that
940 inlining this function is desirable (such as the "inline" keyword in
941 C/C++). It is just a hint; it imposes no requirements on the
944 This attribute suggests that optimization passes and code generator
945 passes make choices that keep the code size of this function as small
946 as possible and perform optimizations that may sacrifice runtime
947 performance in order to minimize the size of the generated code.
949 This attribute disables prologue / epilogue emission for the
950 function. This can have very system-specific consequences.
952 This indicates that the callee function at a call site is not recognized as
953 a built-in function. LLVM will retain the original call and not replace it
954 with equivalent code based on the semantics of the built-in function, unless
955 the call site uses the ``builtin`` attribute. This is valid at call sites
956 and on function declarations and definitions.
958 This attribute indicates that calls to the function cannot be
959 duplicated. A call to a ``noduplicate`` function may be moved
960 within its parent function, but may not be duplicated within
963 A function containing a ``noduplicate`` call may still
964 be an inlining candidate, provided that the call is not
965 duplicated by inlining. That implies that the function has
966 internal linkage and only has one call site, so the original
967 call is dead after inlining.
969 This attributes disables implicit floating point instructions.
971 This attribute indicates that the inliner should never inline this
972 function in any situation. This attribute may not be used together
973 with the ``alwaysinline`` attribute.
975 This attribute suppresses lazy symbol binding for the function. This
976 may make calls to the function faster, at the cost of extra program
977 startup time if the function is not called during program startup.
979 This attribute indicates that the code generator should not use a
980 red zone, even if the target-specific ABI normally permits it.
982 This function attribute indicates that the function never returns
983 normally. This produces undefined behavior at runtime if the
984 function ever does dynamically return.
986 This function attribute indicates that the function never returns
987 with an unwind or exceptional control flow. If the function does
988 unwind, its runtime behavior is undefined.
990 This function attribute indicates that the function is not optimized
991 by any optimization or code generator passes with the
992 exception of interprocedural optimization passes.
993 This attribute cannot be used together with the ``alwaysinline``
994 attribute; this attribute is also incompatible
995 with the ``minsize`` attribute and the ``optsize`` attribute.
997 This attribute requires the ``noinline`` attribute to be specified on
998 the function as well, so the function is never inlined into any caller.
999 Only functions with the ``alwaysinline`` attribute are valid
1000 candidates for inlining into the body of this function.
1002 This attribute suggests that optimization passes and code generator
1003 passes make choices that keep the code size of this function low,
1004 and otherwise do optimizations specifically to reduce code size as
1005 long as they do not significantly impact runtime performance.
1007 On a function, this attribute indicates that the function computes its
1008 result (or decides to unwind an exception) based strictly on its arguments,
1009 without dereferencing any pointer arguments or otherwise accessing
1010 any mutable state (e.g. memory, control registers, etc) visible to
1011 caller functions. It does not write through any pointer arguments
1012 (including ``byval`` arguments) and never changes any state visible
1013 to callers. This means that it cannot unwind exceptions by calling
1014 the ``C++`` exception throwing methods.
1016 On an argument, this attribute indicates that the function does not
1017 dereference that pointer argument, even though it may read or write the
1018 memory that the pointer points to if accessed through other pointers.
1020 On a function, this attribute indicates that the function does not write
1021 through any pointer arguments (including ``byval`` arguments) or otherwise
1022 modify any state (e.g. memory, control registers, etc) visible to
1023 caller functions. It may dereference pointer arguments and read
1024 state that may be set in the caller. A readonly function always
1025 returns the same value (or unwinds an exception identically) when
1026 called with the same set of arguments and global state. It cannot
1027 unwind an exception by calling the ``C++`` exception throwing
1030 On an argument, this attribute indicates that the function does not write
1031 through this pointer argument, even though it may write to the memory that
1032 the pointer points to.
1034 This attribute indicates that this function can return twice. The C
1035 ``setjmp`` is an example of such a function. The compiler disables
1036 some optimizations (like tail calls) in the caller of these
1038 ``sanitize_address``
1039 This attribute indicates that AddressSanitizer checks
1040 (dynamic address safety analysis) are enabled for this function.
1042 This attribute indicates that MemorySanitizer checks (dynamic detection
1043 of accesses to uninitialized memory) are enabled for this function.
1045 This attribute indicates that ThreadSanitizer checks
1046 (dynamic thread safety analysis) are enabled for this function.
1048 This attribute indicates that the function should emit a stack
1049 smashing protector. It is in the form of a "canary" --- a random value
1050 placed on the stack before the local variables that's checked upon
1051 return from the function to see if it has been overwritten. A
1052 heuristic is used to determine if a function needs stack protectors
1053 or not. The heuristic used will enable protectors for functions with:
1055 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1056 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1057 - Calls to alloca() with variable sizes or constant sizes greater than
1058 ``ssp-buffer-size``.
1060 If a function that has an ``ssp`` attribute is inlined into a
1061 function that doesn't have an ``ssp`` attribute, then the resulting
1062 function will have an ``ssp`` attribute.
1064 This attribute indicates that the function should *always* emit a
1065 stack smashing protector. This overrides the ``ssp`` function
1068 If a function that has an ``sspreq`` attribute is inlined into a
1069 function that doesn't have an ``sspreq`` attribute or which has an
1070 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1071 an ``sspreq`` attribute.
1073 This attribute indicates that the function should emit a stack smashing
1074 protector. This attribute causes a strong heuristic to be used when
1075 determining if a function needs stack protectors. The strong heuristic
1076 will enable protectors for functions with:
1078 - Arrays of any size and type
1079 - Aggregates containing an array of any size and type.
1080 - Calls to alloca().
1081 - Local variables that have had their address taken.
1083 This overrides the ``ssp`` function attribute.
1085 If a function that has an ``sspstrong`` attribute is inlined into a
1086 function that doesn't have an ``sspstrong`` attribute, then the
1087 resulting function will have an ``sspstrong`` attribute.
1089 This attribute indicates that the ABI being targeted requires that
1090 an unwind table entry be produce for this function even if we can
1091 show that no exceptions passes by it. This is normally the case for
1092 the ELF x86-64 abi, but it can be disabled for some compilation
1097 Module-Level Inline Assembly
1098 ----------------------------
1100 Modules may contain "module-level inline asm" blocks, which corresponds
1101 to the GCC "file scope inline asm" blocks. These blocks are internally
1102 concatenated by LLVM and treated as a single unit, but may be separated
1103 in the ``.ll`` file if desired. The syntax is very simple:
1105 .. code-block:: llvm
1107 module asm "inline asm code goes here"
1108 module asm "more can go here"
1110 The strings can contain any character by escaping non-printable
1111 characters. The escape sequence used is simply "\\xx" where "xx" is the
1112 two digit hex code for the number.
1114 The inline asm code is simply printed to the machine code .s file when
1115 assembly code is generated.
1117 .. _langref_datalayout:
1122 A module may specify a target specific data layout string that specifies
1123 how data is to be laid out in memory. The syntax for the data layout is
1126 .. code-block:: llvm
1128 target datalayout = "layout specification"
1130 The *layout specification* consists of a list of specifications
1131 separated by the minus sign character ('-'). Each specification starts
1132 with a letter and may include other information after the letter to
1133 define some aspect of the data layout. The specifications accepted are
1137 Specifies that the target lays out data in big-endian form. That is,
1138 the bits with the most significance have the lowest address
1141 Specifies that the target lays out data in little-endian form. That
1142 is, the bits with the least significance have the lowest address
1145 Specifies the natural alignment of the stack in bits. Alignment
1146 promotion of stack variables is limited to the natural stack
1147 alignment to avoid dynamic stack realignment. The stack alignment
1148 must be a multiple of 8-bits. If omitted, the natural stack
1149 alignment defaults to "unspecified", which does not prevent any
1150 alignment promotions.
1151 ``p[n]:<size>:<abi>:<pref>``
1152 This specifies the *size* of a pointer and its ``<abi>`` and
1153 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1154 bits. The address space, ``n`` is optional, and if not specified,
1155 denotes the default address space 0. The value of ``n`` must be
1156 in the range [1,2^23).
1157 ``i<size>:<abi>:<pref>``
1158 This specifies the alignment for an integer type of a given bit
1159 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1160 ``v<size>:<abi>:<pref>``
1161 This specifies the alignment for a vector type of a given bit
1163 ``f<size>:<abi>:<pref>``
1164 This specifies the alignment for a floating point type of a given bit
1165 ``<size>``. Only values of ``<size>`` that are supported by the target
1166 will work. 32 (float) and 64 (double) are supported on all targets; 80
1167 or 128 (different flavors of long double) are also supported on some
1170 This specifies the alignment for an object of aggregate type.
1172 If present, specifies that llvm names are mangled in the output. The
1175 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1176 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1177 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1178 symbols get a ``_`` prefix.
1179 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1180 functions also get a suffix based on the frame size.
1181 ``n<size1>:<size2>:<size3>...``
1182 This specifies a set of native integer widths for the target CPU in
1183 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1184 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1185 this set are considered to support most general arithmetic operations
1188 On every specification that takes a ``<abi>:<pref>``, specifying the
1189 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1190 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1192 When constructing the data layout for a given target, LLVM starts with a
1193 default set of specifications which are then (possibly) overridden by
1194 the specifications in the ``datalayout`` keyword. The default
1195 specifications are given in this list:
1197 - ``E`` - big endian
1198 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1199 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1200 same as the default address space.
1201 - ``S0`` - natural stack alignment is unspecified
1202 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1203 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1204 - ``i16:16:16`` - i16 is 16-bit aligned
1205 - ``i32:32:32`` - i32 is 32-bit aligned
1206 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1207 alignment of 64-bits
1208 - ``f16:16:16`` - half is 16-bit aligned
1209 - ``f32:32:32`` - float is 32-bit aligned
1210 - ``f64:64:64`` - double is 64-bit aligned
1211 - ``f128:128:128`` - quad is 128-bit aligned
1212 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1213 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1214 - ``a:0:64`` - aggregates are 64-bit aligned
1216 When LLVM is determining the alignment for a given type, it uses the
1219 #. If the type sought is an exact match for one of the specifications,
1220 that specification is used.
1221 #. If no match is found, and the type sought is an integer type, then
1222 the smallest integer type that is larger than the bitwidth of the
1223 sought type is used. If none of the specifications are larger than
1224 the bitwidth then the largest integer type is used. For example,
1225 given the default specifications above, the i7 type will use the
1226 alignment of i8 (next largest) while both i65 and i256 will use the
1227 alignment of i64 (largest specified).
1228 #. If no match is found, and the type sought is a vector type, then the
1229 largest vector type that is smaller than the sought vector type will
1230 be used as a fall back. This happens because <128 x double> can be
1231 implemented in terms of 64 <2 x double>, for example.
1233 The function of the data layout string may not be what you expect.
1234 Notably, this is not a specification from the frontend of what alignment
1235 the code generator should use.
1237 Instead, if specified, the target data layout is required to match what
1238 the ultimate *code generator* expects. This string is used by the
1239 mid-level optimizers to improve code, and this only works if it matches
1240 what the ultimate code generator uses. If you would like to generate IR
1241 that does not embed this target-specific detail into the IR, then you
1242 don't have to specify the string. This will disable some optimizations
1243 that require precise layout information, but this also prevents those
1244 optimizations from introducing target specificity into the IR.
1251 A module may specify a target triple string that describes the target
1252 host. The syntax for the target triple is simply:
1254 .. code-block:: llvm
1256 target triple = "x86_64-apple-macosx10.7.0"
1258 The *target triple* string consists of a series of identifiers delimited
1259 by the minus sign character ('-'). The canonical forms are:
1263 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1264 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1266 This information is passed along to the backend so that it generates
1267 code for the proper architecture. It's possible to override this on the
1268 command line with the ``-mtriple`` command line option.
1270 .. _pointeraliasing:
1272 Pointer Aliasing Rules
1273 ----------------------
1275 Any memory access must be done through a pointer value associated with
1276 an address range of the memory access, otherwise the behavior is
1277 undefined. Pointer values are associated with address ranges according
1278 to the following rules:
1280 - A pointer value is associated with the addresses associated with any
1281 value it is *based* on.
1282 - An address of a global variable is associated with the address range
1283 of the variable's storage.
1284 - The result value of an allocation instruction is associated with the
1285 address range of the allocated storage.
1286 - A null pointer in the default address-space is associated with no
1288 - An integer constant other than zero or a pointer value returned from
1289 a function not defined within LLVM may be associated with address
1290 ranges allocated through mechanisms other than those provided by
1291 LLVM. Such ranges shall not overlap with any ranges of addresses
1292 allocated by mechanisms provided by LLVM.
1294 A pointer value is *based* on another pointer value according to the
1297 - A pointer value formed from a ``getelementptr`` operation is *based*
1298 on the first operand of the ``getelementptr``.
1299 - The result value of a ``bitcast`` is *based* on the operand of the
1301 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1302 values that contribute (directly or indirectly) to the computation of
1303 the pointer's value.
1304 - The "*based* on" relationship is transitive.
1306 Note that this definition of *"based"* is intentionally similar to the
1307 definition of *"based"* in C99, though it is slightly weaker.
1309 LLVM IR does not associate types with memory. The result type of a
1310 ``load`` merely indicates the size and alignment of the memory from
1311 which to load, as well as the interpretation of the value. The first
1312 operand type of a ``store`` similarly only indicates the size and
1313 alignment of the store.
1315 Consequently, type-based alias analysis, aka TBAA, aka
1316 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1317 :ref:`Metadata <metadata>` may be used to encode additional information
1318 which specialized optimization passes may use to implement type-based
1323 Volatile Memory Accesses
1324 ------------------------
1326 Certain memory accesses, such as :ref:`load <i_load>`'s,
1327 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1328 marked ``volatile``. The optimizers must not change the number of
1329 volatile operations or change their order of execution relative to other
1330 volatile operations. The optimizers *may* change the order of volatile
1331 operations relative to non-volatile operations. This is not Java's
1332 "volatile" and has no cross-thread synchronization behavior.
1334 IR-level volatile loads and stores cannot safely be optimized into
1335 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1336 flagged volatile. Likewise, the backend should never split or merge
1337 target-legal volatile load/store instructions.
1339 .. admonition:: Rationale
1341 Platforms may rely on volatile loads and stores of natively supported
1342 data width to be executed as single instruction. For example, in C
1343 this holds for an l-value of volatile primitive type with native
1344 hardware support, but not necessarily for aggregate types. The
1345 frontend upholds these expectations, which are intentionally
1346 unspecified in the IR. The rules above ensure that IR transformation
1347 do not violate the frontend's contract with the language.
1351 Memory Model for Concurrent Operations
1352 --------------------------------------
1354 The LLVM IR does not define any way to start parallel threads of
1355 execution or to register signal handlers. Nonetheless, there are
1356 platform-specific ways to create them, and we define LLVM IR's behavior
1357 in their presence. This model is inspired by the C++0x memory model.
1359 For a more informal introduction to this model, see the :doc:`Atomics`.
1361 We define a *happens-before* partial order as the least partial order
1364 - Is a superset of single-thread program order, and
1365 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1366 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1367 techniques, like pthread locks, thread creation, thread joining,
1368 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1369 Constraints <ordering>`).
1371 Note that program order does not introduce *happens-before* edges
1372 between a thread and signals executing inside that thread.
1374 Every (defined) read operation (load instructions, memcpy, atomic
1375 loads/read-modify-writes, etc.) R reads a series of bytes written by
1376 (defined) write operations (store instructions, atomic
1377 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1378 section, initialized globals are considered to have a write of the
1379 initializer which is atomic and happens before any other read or write
1380 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1381 may see any write to the same byte, except:
1383 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1384 write\ :sub:`2` happens before R\ :sub:`byte`, then
1385 R\ :sub:`byte` does not see write\ :sub:`1`.
1386 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1387 R\ :sub:`byte` does not see write\ :sub:`3`.
1389 Given that definition, R\ :sub:`byte` is defined as follows:
1391 - If R is volatile, the result is target-dependent. (Volatile is
1392 supposed to give guarantees which can support ``sig_atomic_t`` in
1393 C/C++, and may be used for accesses to addresses which do not behave
1394 like normal memory. It does not generally provide cross-thread
1396 - Otherwise, if there is no write to the same byte that happens before
1397 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1398 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1399 R\ :sub:`byte` returns the value written by that write.
1400 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1401 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1402 Memory Ordering Constraints <ordering>` section for additional
1403 constraints on how the choice is made.
1404 - Otherwise R\ :sub:`byte` returns ``undef``.
1406 R returns the value composed of the series of bytes it read. This
1407 implies that some bytes within the value may be ``undef`` **without**
1408 the entire value being ``undef``. Note that this only defines the
1409 semantics of the operation; it doesn't mean that targets will emit more
1410 than one instruction to read the series of bytes.
1412 Note that in cases where none of the atomic intrinsics are used, this
1413 model places only one restriction on IR transformations on top of what
1414 is required for single-threaded execution: introducing a store to a byte
1415 which might not otherwise be stored is not allowed in general.
1416 (Specifically, in the case where another thread might write to and read
1417 from an address, introducing a store can change a load that may see
1418 exactly one write into a load that may see multiple writes.)
1422 Atomic Memory Ordering Constraints
1423 ----------------------------------
1425 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1426 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1427 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1428 an ordering parameter that determines which other atomic instructions on
1429 the same address they *synchronize with*. These semantics are borrowed
1430 from Java and C++0x, but are somewhat more colloquial. If these
1431 descriptions aren't precise enough, check those specs (see spec
1432 references in the :doc:`atomics guide <Atomics>`).
1433 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1434 differently since they don't take an address. See that instruction's
1435 documentation for details.
1437 For a simpler introduction to the ordering constraints, see the
1441 The set of values that can be read is governed by the happens-before
1442 partial order. A value cannot be read unless some operation wrote
1443 it. This is intended to provide a guarantee strong enough to model
1444 Java's non-volatile shared variables. This ordering cannot be
1445 specified for read-modify-write operations; it is not strong enough
1446 to make them atomic in any interesting way.
1448 In addition to the guarantees of ``unordered``, there is a single
1449 total order for modifications by ``monotonic`` operations on each
1450 address. All modification orders must be compatible with the
1451 happens-before order. There is no guarantee that the modification
1452 orders can be combined to a global total order for the whole program
1453 (and this often will not be possible). The read in an atomic
1454 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1455 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1456 order immediately before the value it writes. If one atomic read
1457 happens before another atomic read of the same address, the later
1458 read must see the same value or a later value in the address's
1459 modification order. This disallows reordering of ``monotonic`` (or
1460 stronger) operations on the same address. If an address is written
1461 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1462 read that address repeatedly, the other threads must eventually see
1463 the write. This corresponds to the C++0x/C1x
1464 ``memory_order_relaxed``.
1466 In addition to the guarantees of ``monotonic``, a
1467 *synchronizes-with* edge may be formed with a ``release`` operation.
1468 This is intended to model C++'s ``memory_order_acquire``.
1470 In addition to the guarantees of ``monotonic``, if this operation
1471 writes a value which is subsequently read by an ``acquire``
1472 operation, it *synchronizes-with* that operation. (This isn't a
1473 complete description; see the C++0x definition of a release
1474 sequence.) This corresponds to the C++0x/C1x
1475 ``memory_order_release``.
1476 ``acq_rel`` (acquire+release)
1477 Acts as both an ``acquire`` and ``release`` operation on its
1478 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1479 ``seq_cst`` (sequentially consistent)
1480 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1481 operation which only reads, ``release`` for an operation which only
1482 writes), there is a global total order on all
1483 sequentially-consistent operations on all addresses, which is
1484 consistent with the *happens-before* partial order and with the
1485 modification orders of all the affected addresses. Each
1486 sequentially-consistent read sees the last preceding write to the
1487 same address in this global order. This corresponds to the C++0x/C1x
1488 ``memory_order_seq_cst`` and Java volatile.
1492 If an atomic operation is marked ``singlethread``, it only *synchronizes
1493 with* or participates in modification and seq\_cst total orderings with
1494 other operations running in the same thread (for example, in signal
1502 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1503 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1504 :ref:`frem <i_frem>`) have the following flags that can set to enable
1505 otherwise unsafe floating point operations
1508 No NaNs - Allow optimizations to assume the arguments and result are not
1509 NaN. Such optimizations are required to retain defined behavior over
1510 NaNs, but the value of the result is undefined.
1513 No Infs - Allow optimizations to assume the arguments and result are not
1514 +/-Inf. Such optimizations are required to retain defined behavior over
1515 +/-Inf, but the value of the result is undefined.
1518 No Signed Zeros - Allow optimizations to treat the sign of a zero
1519 argument or result as insignificant.
1522 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1523 argument rather than perform division.
1526 Fast - Allow algebraically equivalent transformations that may
1527 dramatically change results in floating point (e.g. reassociate). This
1528 flag implies all the others.
1535 The LLVM type system is one of the most important features of the
1536 intermediate representation. Being typed enables a number of
1537 optimizations to be performed on the intermediate representation
1538 directly, without having to do extra analyses on the side before the
1539 transformation. A strong type system makes it easier to read the
1540 generated code and enables novel analyses and transformations that are
1541 not feasible to perform on normal three address code representations.
1551 The void type does not represent any value and has no size.
1569 The function type can be thought of as a function signature. It consists of a
1570 return type and a list of formal parameter types. The return type of a function
1571 type is a void type or first class type --- except for :ref:`label <t_label>`
1572 and :ref:`metadata <t_metadata>` types.
1578 <returntype> (<parameter list>)
1580 ...where '``<parameter list>``' is a comma-separated list of type
1581 specifiers. Optionally, the parameter list may include a type ``...``, which
1582 indicates that the function takes a variable number of arguments. Variable
1583 argument functions can access their arguments with the :ref:`variable argument
1584 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1585 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1589 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1590 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1591 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1592 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1593 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1594 | ``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. |
1595 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1596 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1597 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1604 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1605 Values of these types are the only ones which can be produced by
1613 These are the types that are valid in registers from CodeGen's perspective.
1622 The integer type is a very simple type that simply specifies an
1623 arbitrary bit width for the integer type desired. Any bit width from 1
1624 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1632 The number of bits the integer will occupy is specified by the ``N``
1638 +----------------+------------------------------------------------+
1639 | ``i1`` | a single-bit integer. |
1640 +----------------+------------------------------------------------+
1641 | ``i32`` | a 32-bit integer. |
1642 +----------------+------------------------------------------------+
1643 | ``i1942652`` | a really big integer of over 1 million bits. |
1644 +----------------+------------------------------------------------+
1648 Floating Point Types
1649 """"""""""""""""""""
1658 - 16-bit floating point value
1661 - 32-bit floating point value
1664 - 64-bit floating point value
1667 - 128-bit floating point value (112-bit mantissa)
1670 - 80-bit floating point value (X87)
1673 - 128-bit floating point value (two 64-bits)
1682 The x86mmx type represents a value held in an MMX register on an x86
1683 machine. The operations allowed on it are quite limited: parameters and
1684 return values, load and store, and bitcast. User-specified MMX
1685 instructions are represented as intrinsic or asm calls with arguments
1686 and/or results of this type. There are no arrays, vectors or constants
1703 The pointer type is used to specify memory locations. Pointers are
1704 commonly used to reference objects in memory.
1706 Pointer types may have an optional address space attribute defining the
1707 numbered address space where the pointed-to object resides. The default
1708 address space is number zero. The semantics of non-zero address spaces
1709 are target-specific.
1711 Note that LLVM does not permit pointers to void (``void*``) nor does it
1712 permit pointers to labels (``label*``). Use ``i8*`` instead.
1722 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1723 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1724 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1725 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1726 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1727 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1728 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1737 A vector type is a simple derived type that represents a vector of
1738 elements. Vector types are used when multiple primitive data are
1739 operated in parallel using a single instruction (SIMD). A vector type
1740 requires a size (number of elements) and an underlying primitive data
1741 type. Vector types are considered :ref:`first class <t_firstclass>`.
1747 < <# elements> x <elementtype> >
1749 The number of elements is a constant integer value larger than 0;
1750 elementtype may be any integer or floating point type, or a pointer to
1751 these types. Vectors of size zero are not allowed.
1755 +-------------------+--------------------------------------------------+
1756 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1757 +-------------------+--------------------------------------------------+
1758 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1759 +-------------------+--------------------------------------------------+
1760 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1761 +-------------------+--------------------------------------------------+
1762 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1763 +-------------------+--------------------------------------------------+
1772 The label type represents code labels.
1787 The metadata type represents embedded metadata. No derived types may be
1788 created from metadata except for :ref:`function <t_function>` arguments.
1801 Aggregate Types are a subset of derived types that can contain multiple
1802 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1803 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1813 The array type is a very simple derived type that arranges elements
1814 sequentially in memory. The array type requires a size (number of
1815 elements) and an underlying data type.
1821 [<# elements> x <elementtype>]
1823 The number of elements is a constant integer value; ``elementtype`` may
1824 be any type with a size.
1828 +------------------+--------------------------------------+
1829 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1830 +------------------+--------------------------------------+
1831 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1832 +------------------+--------------------------------------+
1833 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1834 +------------------+--------------------------------------+
1836 Here are some examples of multidimensional arrays:
1838 +-----------------------------+----------------------------------------------------------+
1839 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1840 +-----------------------------+----------------------------------------------------------+
1841 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1842 +-----------------------------+----------------------------------------------------------+
1843 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1844 +-----------------------------+----------------------------------------------------------+
1846 There is no restriction on indexing beyond the end of the array implied
1847 by a static type (though there are restrictions on indexing beyond the
1848 bounds of an allocated object in some cases). This means that
1849 single-dimension 'variable sized array' addressing can be implemented in
1850 LLVM with a zero length array type. An implementation of 'pascal style
1851 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1861 The structure type is used to represent a collection of data members
1862 together in memory. The elements of a structure may be any type that has
1865 Structures in memory are accessed using '``load``' and '``store``' by
1866 getting a pointer to a field with the '``getelementptr``' instruction.
1867 Structures in registers are accessed using the '``extractvalue``' and
1868 '``insertvalue``' instructions.
1870 Structures may optionally be "packed" structures, which indicate that
1871 the alignment of the struct is one byte, and that there is no padding
1872 between the elements. In non-packed structs, padding between field types
1873 is inserted as defined by the DataLayout string in the module, which is
1874 required to match what the underlying code generator expects.
1876 Structures can either be "literal" or "identified". A literal structure
1877 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1878 identified types are always defined at the top level with a name.
1879 Literal types are uniqued by their contents and can never be recursive
1880 or opaque since there is no way to write one. Identified types can be
1881 recursive, can be opaqued, and are never uniqued.
1887 %T1 = type { <type list> } ; Identified normal struct type
1888 %T2 = type <{ <type list> }> ; Identified packed struct type
1892 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1893 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1894 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1895 | ``{ 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``. |
1896 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1897 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1898 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1902 Opaque Structure Types
1903 """"""""""""""""""""""
1907 Opaque structure types are used to represent named structure types that
1908 do not have a body specified. This corresponds (for example) to the C
1909 notion of a forward declared structure.
1920 +--------------+-------------------+
1921 | ``opaque`` | An opaque type. |
1922 +--------------+-------------------+
1927 LLVM has several different basic types of constants. This section
1928 describes them all and their syntax.
1933 **Boolean constants**
1934 The two strings '``true``' and '``false``' are both valid constants
1936 **Integer constants**
1937 Standard integers (such as '4') are constants of the
1938 :ref:`integer <t_integer>` type. Negative numbers may be used with
1940 **Floating point constants**
1941 Floating point constants use standard decimal notation (e.g.
1942 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1943 hexadecimal notation (see below). The assembler requires the exact
1944 decimal value of a floating-point constant. For example, the
1945 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1946 decimal in binary. Floating point constants must have a :ref:`floating
1947 point <t_floating>` type.
1948 **Null pointer constants**
1949 The identifier '``null``' is recognized as a null pointer constant
1950 and must be of :ref:`pointer type <t_pointer>`.
1952 The one non-intuitive notation for constants is the hexadecimal form of
1953 floating point constants. For example, the form
1954 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1955 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1956 constants are required (and the only time that they are generated by the
1957 disassembler) is when a floating point constant must be emitted but it
1958 cannot be represented as a decimal floating point number in a reasonable
1959 number of digits. For example, NaN's, infinities, and other special
1960 values are represented in their IEEE hexadecimal format so that assembly
1961 and disassembly do not cause any bits to change in the constants.
1963 When using the hexadecimal form, constants of types half, float, and
1964 double are represented using the 16-digit form shown above (which
1965 matches the IEEE754 representation for double); half and float values
1966 must, however, be exactly representable as IEEE 754 half and single
1967 precision, respectively. Hexadecimal format is always used for long
1968 double, and there are three forms of long double. The 80-bit format used
1969 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1970 128-bit format used by PowerPC (two adjacent doubles) is represented by
1971 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1972 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1973 will only work if they match the long double format on your target.
1974 The IEEE 16-bit format (half precision) is represented by ``0xH``
1975 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1976 (sign bit at the left).
1978 There are no constants of type x86mmx.
1980 .. _complexconstants:
1985 Complex constants are a (potentially recursive) combination of simple
1986 constants and smaller complex constants.
1988 **Structure constants**
1989 Structure constants are represented with notation similar to
1990 structure type definitions (a comma separated list of elements,
1991 surrounded by braces (``{}``)). For example:
1992 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1993 "``@G = external global i32``". Structure constants must have
1994 :ref:`structure type <t_struct>`, and the number and types of elements
1995 must match those specified by the type.
1997 Array constants are represented with notation similar to array type
1998 definitions (a comma separated list of elements, surrounded by
1999 square brackets (``[]``)). For example:
2000 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2001 :ref:`array type <t_array>`, and the number and types of elements must
2002 match those specified by the type.
2003 **Vector constants**
2004 Vector constants are represented with notation similar to vector
2005 type definitions (a comma separated list of elements, surrounded by
2006 less-than/greater-than's (``<>``)). For example:
2007 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2008 must have :ref:`vector type <t_vector>`, and the number and types of
2009 elements must match those specified by the type.
2010 **Zero initialization**
2011 The string '``zeroinitializer``' can be used to zero initialize a
2012 value to zero of *any* type, including scalar and
2013 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2014 having to print large zero initializers (e.g. for large arrays) and
2015 is always exactly equivalent to using explicit zero initializers.
2017 A metadata node is a structure-like constant with :ref:`metadata
2018 type <t_metadata>`. For example:
2019 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2020 constants that are meant to be interpreted as part of the
2021 instruction stream, metadata is a place to attach additional
2022 information such as debug info.
2024 Global Variable and Function Addresses
2025 --------------------------------------
2027 The addresses of :ref:`global variables <globalvars>` and
2028 :ref:`functions <functionstructure>` are always implicitly valid
2029 (link-time) constants. These constants are explicitly referenced when
2030 the :ref:`identifier for the global <identifiers>` is used and always have
2031 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2034 .. code-block:: llvm
2038 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2045 The string '``undef``' can be used anywhere a constant is expected, and
2046 indicates that the user of the value may receive an unspecified
2047 bit-pattern. Undefined values may be of any type (other than '``label``'
2048 or '``void``') and be used anywhere a constant is permitted.
2050 Undefined values are useful because they indicate to the compiler that
2051 the program is well defined no matter what value is used. This gives the
2052 compiler more freedom to optimize. Here are some examples of
2053 (potentially surprising) transformations that are valid (in pseudo IR):
2055 .. code-block:: llvm
2065 This is safe because all of the output bits are affected by the undef
2066 bits. Any output bit can have a zero or one depending on the input bits.
2068 .. code-block:: llvm
2079 These logical operations have bits that are not always affected by the
2080 input. For example, if ``%X`` has a zero bit, then the output of the
2081 '``and``' operation will always be a zero for that bit, no matter what
2082 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2083 optimize or assume that the result of the '``and``' is '``undef``'.
2084 However, it is safe to assume that all bits of the '``undef``' could be
2085 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2086 all the bits of the '``undef``' operand to the '``or``' could be set,
2087 allowing the '``or``' to be folded to -1.
2089 .. code-block:: llvm
2091 %A = select undef, %X, %Y
2092 %B = select undef, 42, %Y
2093 %C = select %X, %Y, undef
2103 This set of examples shows that undefined '``select``' (and conditional
2104 branch) conditions can go *either way*, but they have to come from one
2105 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2106 both known to have a clear low bit, then ``%A`` would have to have a
2107 cleared low bit. However, in the ``%C`` example, the optimizer is
2108 allowed to assume that the '``undef``' operand could be the same as
2109 ``%Y``, allowing the whole '``select``' to be eliminated.
2111 .. code-block:: llvm
2113 %A = xor undef, undef
2130 This example points out that two '``undef``' operands are not
2131 necessarily the same. This can be surprising to people (and also matches
2132 C semantics) where they assume that "``X^X``" is always zero, even if
2133 ``X`` is undefined. This isn't true for a number of reasons, but the
2134 short answer is that an '``undef``' "variable" can arbitrarily change
2135 its value over its "live range". This is true because the variable
2136 doesn't actually *have a live range*. Instead, the value is logically
2137 read from arbitrary registers that happen to be around when needed, so
2138 the value is not necessarily consistent over time. In fact, ``%A`` and
2139 ``%C`` need to have the same semantics or the core LLVM "replace all
2140 uses with" concept would not hold.
2142 .. code-block:: llvm
2150 These examples show the crucial difference between an *undefined value*
2151 and *undefined behavior*. An undefined value (like '``undef``') is
2152 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2153 operation can be constant folded to '``undef``', because the '``undef``'
2154 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2155 However, in the second example, we can make a more aggressive
2156 assumption: because the ``undef`` is allowed to be an arbitrary value,
2157 we are allowed to assume that it could be zero. Since a divide by zero
2158 has *undefined behavior*, we are allowed to assume that the operation
2159 does not execute at all. This allows us to delete the divide and all
2160 code after it. Because the undefined operation "can't happen", the
2161 optimizer can assume that it occurs in dead code.
2163 .. code-block:: llvm
2165 a: store undef -> %X
2166 b: store %X -> undef
2171 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2172 value can be assumed to not have any effect; we can assume that the
2173 value is overwritten with bits that happen to match what was already
2174 there. However, a store *to* an undefined location could clobber
2175 arbitrary memory, therefore, it has undefined behavior.
2182 Poison values are similar to :ref:`undef values <undefvalues>`, however
2183 they also represent the fact that an instruction or constant expression
2184 which cannot evoke side effects has nevertheless detected a condition
2185 which results in undefined behavior.
2187 There is currently no way of representing a poison value in the IR; they
2188 only exist when produced by operations such as :ref:`add <i_add>` with
2191 Poison value behavior is defined in terms of value *dependence*:
2193 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2194 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2195 their dynamic predecessor basic block.
2196 - Function arguments depend on the corresponding actual argument values
2197 in the dynamic callers of their functions.
2198 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2199 instructions that dynamically transfer control back to them.
2200 - :ref:`Invoke <i_invoke>` instructions depend on the
2201 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2202 call instructions that dynamically transfer control back to them.
2203 - Non-volatile loads and stores depend on the most recent stores to all
2204 of the referenced memory addresses, following the order in the IR
2205 (including loads and stores implied by intrinsics such as
2206 :ref:`@llvm.memcpy <int_memcpy>`.)
2207 - An instruction with externally visible side effects depends on the
2208 most recent preceding instruction with externally visible side
2209 effects, following the order in the IR. (This includes :ref:`volatile
2210 operations <volatile>`.)
2211 - An instruction *control-depends* on a :ref:`terminator
2212 instruction <terminators>` if the terminator instruction has
2213 multiple successors and the instruction is always executed when
2214 control transfers to one of the successors, and may not be executed
2215 when control is transferred to another.
2216 - Additionally, an instruction also *control-depends* on a terminator
2217 instruction if the set of instructions it otherwise depends on would
2218 be different if the terminator had transferred control to a different
2220 - Dependence is transitive.
2222 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2223 with the additional affect that any instruction which has a *dependence*
2224 on a poison value has undefined behavior.
2226 Here are some examples:
2228 .. code-block:: llvm
2231 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2232 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2233 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2234 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2236 store i32 %poison, i32* @g ; Poison value stored to memory.
2237 %poison2 = load i32* @g ; Poison value loaded back from memory.
2239 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2241 %narrowaddr = bitcast i32* @g to i16*
2242 %wideaddr = bitcast i32* @g to i64*
2243 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2244 %poison4 = load i64* %wideaddr ; Returns a poison value.
2246 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2247 br i1 %cmp, label %true, label %end ; Branch to either destination.
2250 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2251 ; it has undefined behavior.
2255 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2256 ; Both edges into this PHI are
2257 ; control-dependent on %cmp, so this
2258 ; always results in a poison value.
2260 store volatile i32 0, i32* @g ; This would depend on the store in %true
2261 ; if %cmp is true, or the store in %entry
2262 ; otherwise, so this is undefined behavior.
2264 br i1 %cmp, label %second_true, label %second_end
2265 ; The same branch again, but this time the
2266 ; true block doesn't have side effects.
2273 store volatile i32 0, i32* @g ; This time, the instruction always depends
2274 ; on the store in %end. Also, it is
2275 ; control-equivalent to %end, so this is
2276 ; well-defined (ignoring earlier undefined
2277 ; behavior in this example).
2281 Addresses of Basic Blocks
2282 -------------------------
2284 ``blockaddress(@function, %block)``
2286 The '``blockaddress``' constant computes the address of the specified
2287 basic block in the specified function, and always has an ``i8*`` type.
2288 Taking the address of the entry block is illegal.
2290 This value only has defined behavior when used as an operand to the
2291 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2292 against null. Pointer equality tests between labels addresses results in
2293 undefined behavior --- though, again, comparison against null is ok, and
2294 no label is equal to the null pointer. This may be passed around as an
2295 opaque pointer sized value as long as the bits are not inspected. This
2296 allows ``ptrtoint`` and arithmetic to be performed on these values so
2297 long as the original value is reconstituted before the ``indirectbr``
2300 Finally, some targets may provide defined semantics when using the value
2301 as the operand to an inline assembly, but that is target specific.
2305 Constant Expressions
2306 --------------------
2308 Constant expressions are used to allow expressions involving other
2309 constants to be used as constants. Constant expressions may be of any
2310 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2311 that does not have side effects (e.g. load and call are not supported).
2312 The following is the syntax for constant expressions:
2314 ``trunc (CST to TYPE)``
2315 Truncate a constant to another type. The bit size of CST must be
2316 larger than the bit size of TYPE. Both types must be integers.
2317 ``zext (CST to TYPE)``
2318 Zero extend a constant to another type. The bit size of CST must be
2319 smaller than the bit size of TYPE. Both types must be integers.
2320 ``sext (CST to TYPE)``
2321 Sign extend a constant to another type. The bit size of CST must be
2322 smaller than the bit size of TYPE. Both types must be integers.
2323 ``fptrunc (CST to TYPE)``
2324 Truncate a floating point constant to another floating point type.
2325 The size of CST must be larger than the size of TYPE. Both types
2326 must be floating point.
2327 ``fpext (CST to TYPE)``
2328 Floating point extend a constant to another type. The size of CST
2329 must be smaller or equal to the size of TYPE. Both types must be
2331 ``fptoui (CST to TYPE)``
2332 Convert a floating point constant to the corresponding unsigned
2333 integer constant. TYPE must be a scalar or vector integer type. CST
2334 must be of scalar or vector floating point type. Both CST and TYPE
2335 must be scalars, or vectors of the same number of elements. If the
2336 value won't fit in the integer type, the results are undefined.
2337 ``fptosi (CST to TYPE)``
2338 Convert a floating point constant to the corresponding signed
2339 integer constant. TYPE must be a scalar or vector integer type. CST
2340 must be of scalar or vector floating point type. Both CST and TYPE
2341 must be scalars, or vectors of the same number of elements. If the
2342 value won't fit in the integer type, the results are undefined.
2343 ``uitofp (CST to TYPE)``
2344 Convert an unsigned integer constant to the corresponding floating
2345 point constant. TYPE must be a scalar or vector floating point type.
2346 CST must be of scalar or vector integer type. Both CST and TYPE must
2347 be scalars, or vectors of the same number of elements. If the value
2348 won't fit in the floating point type, the results are undefined.
2349 ``sitofp (CST to TYPE)``
2350 Convert a signed integer constant to the corresponding floating
2351 point constant. TYPE must be a scalar or vector floating point type.
2352 CST must be of scalar or vector integer type. Both CST and TYPE must
2353 be scalars, or vectors of the same number of elements. If the value
2354 won't fit in the floating point type, the results are undefined.
2355 ``ptrtoint (CST to TYPE)``
2356 Convert a pointer typed constant to the corresponding integer
2357 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2358 pointer type. The ``CST`` value is zero extended, truncated, or
2359 unchanged to make it fit in ``TYPE``.
2360 ``inttoptr (CST to TYPE)``
2361 Convert an integer constant to a pointer constant. TYPE must be a
2362 pointer type. CST must be of integer type. The CST value is zero
2363 extended, truncated, or unchanged to make it fit in a pointer size.
2364 This one is *really* dangerous!
2365 ``bitcast (CST to TYPE)``
2366 Convert a constant, CST, to another TYPE. The constraints of the
2367 operands are the same as those for the :ref:`bitcast
2368 instruction <i_bitcast>`.
2369 ``addrspacecast (CST to TYPE)``
2370 Convert a constant pointer or constant vector of pointer, CST, to another
2371 TYPE in a different address space. The constraints of the operands are the
2372 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2373 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2374 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2375 constants. As with the :ref:`getelementptr <i_getelementptr>`
2376 instruction, the index list may have zero or more indexes, which are
2377 required to make sense for the type of "CSTPTR".
2378 ``select (COND, VAL1, VAL2)``
2379 Perform the :ref:`select operation <i_select>` on constants.
2380 ``icmp COND (VAL1, VAL2)``
2381 Performs the :ref:`icmp operation <i_icmp>` on constants.
2382 ``fcmp COND (VAL1, VAL2)``
2383 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2384 ``extractelement (VAL, IDX)``
2385 Perform the :ref:`extractelement operation <i_extractelement>` on
2387 ``insertelement (VAL, ELT, IDX)``
2388 Perform the :ref:`insertelement operation <i_insertelement>` on
2390 ``shufflevector (VEC1, VEC2, IDXMASK)``
2391 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2393 ``extractvalue (VAL, IDX0, IDX1, ...)``
2394 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2395 constants. The index list is interpreted in a similar manner as
2396 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2397 least one index value must be specified.
2398 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2399 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2400 The index list is interpreted in a similar manner as indices in a
2401 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2402 value must be specified.
2403 ``OPCODE (LHS, RHS)``
2404 Perform the specified operation of the LHS and RHS constants. OPCODE
2405 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2406 binary <bitwiseops>` operations. The constraints on operands are
2407 the same as those for the corresponding instruction (e.g. no bitwise
2408 operations on floating point values are allowed).
2415 Inline Assembler Expressions
2416 ----------------------------
2418 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2419 Inline Assembly <moduleasm>`) through the use of a special value. This
2420 value represents the inline assembler as a string (containing the
2421 instructions to emit), a list of operand constraints (stored as a
2422 string), a flag that indicates whether or not the inline asm expression
2423 has side effects, and a flag indicating whether the function containing
2424 the asm needs to align its stack conservatively. An example inline
2425 assembler expression is:
2427 .. code-block:: llvm
2429 i32 (i32) asm "bswap $0", "=r,r"
2431 Inline assembler expressions may **only** be used as the callee operand
2432 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2433 Thus, typically we have:
2435 .. code-block:: llvm
2437 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2439 Inline asms with side effects not visible in the constraint list must be
2440 marked as having side effects. This is done through the use of the
2441 '``sideeffect``' keyword, like so:
2443 .. code-block:: llvm
2445 call void asm sideeffect "eieio", ""()
2447 In some cases inline asms will contain code that will not work unless
2448 the stack is aligned in some way, such as calls or SSE instructions on
2449 x86, yet will not contain code that does that alignment within the asm.
2450 The compiler should make conservative assumptions about what the asm
2451 might contain and should generate its usual stack alignment code in the
2452 prologue if the '``alignstack``' keyword is present:
2454 .. code-block:: llvm
2456 call void asm alignstack "eieio", ""()
2458 Inline asms also support using non-standard assembly dialects. The
2459 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2460 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2461 the only supported dialects. An example is:
2463 .. code-block:: llvm
2465 call void asm inteldialect "eieio", ""()
2467 If multiple keywords appear the '``sideeffect``' keyword must come
2468 first, the '``alignstack``' keyword second and the '``inteldialect``'
2474 The call instructions that wrap inline asm nodes may have a
2475 "``!srcloc``" MDNode attached to it that contains a list of constant
2476 integers. If present, the code generator will use the integer as the
2477 location cookie value when report errors through the ``LLVMContext``
2478 error reporting mechanisms. This allows a front-end to correlate backend
2479 errors that occur with inline asm back to the source code that produced
2482 .. code-block:: llvm
2484 call void asm sideeffect "something bad", ""(), !srcloc !42
2486 !42 = !{ i32 1234567 }
2488 It is up to the front-end to make sense of the magic numbers it places
2489 in the IR. If the MDNode contains multiple constants, the code generator
2490 will use the one that corresponds to the line of the asm that the error
2495 Metadata Nodes and Metadata Strings
2496 -----------------------------------
2498 LLVM IR allows metadata to be attached to instructions in the program
2499 that can convey extra information about the code to the optimizers and
2500 code generator. One example application of metadata is source-level
2501 debug information. There are two metadata primitives: strings and nodes.
2502 All metadata has the ``metadata`` type and is identified in syntax by a
2503 preceding exclamation point ('``!``').
2505 A metadata string is a string surrounded by double quotes. It can
2506 contain any character by escaping non-printable characters with
2507 "``\xx``" where "``xx``" is the two digit hex code. For example:
2510 Metadata nodes are represented with notation similar to structure
2511 constants (a comma separated list of elements, surrounded by braces and
2512 preceded by an exclamation point). Metadata nodes can have any values as
2513 their operand. For example:
2515 .. code-block:: llvm
2517 !{ metadata !"test\00", i32 10}
2519 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2520 metadata nodes, which can be looked up in the module symbol table. For
2523 .. code-block:: llvm
2525 !foo = metadata !{!4, !3}
2527 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2528 function is using two metadata arguments:
2530 .. code-block:: llvm
2532 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2534 Metadata can be attached with an instruction. Here metadata ``!21`` is
2535 attached to the ``add`` instruction using the ``!dbg`` identifier:
2537 .. code-block:: llvm
2539 %indvar.next = add i64 %indvar, 1, !dbg !21
2541 More information about specific metadata nodes recognized by the
2542 optimizers and code generator is found below.
2547 In LLVM IR, memory does not have types, so LLVM's own type system is not
2548 suitable for doing TBAA. Instead, metadata is added to the IR to
2549 describe a type system of a higher level language. This can be used to
2550 implement typical C/C++ TBAA, but it can also be used to implement
2551 custom alias analysis behavior for other languages.
2553 The current metadata format is very simple. TBAA metadata nodes have up
2554 to three fields, e.g.:
2556 .. code-block:: llvm
2558 !0 = metadata !{ metadata !"an example type tree" }
2559 !1 = metadata !{ metadata !"int", metadata !0 }
2560 !2 = metadata !{ metadata !"float", metadata !0 }
2561 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2563 The first field is an identity field. It can be any value, usually a
2564 metadata string, which uniquely identifies the type. The most important
2565 name in the tree is the name of the root node. Two trees with different
2566 root node names are entirely disjoint, even if they have leaves with
2569 The second field identifies the type's parent node in the tree, or is
2570 null or omitted for a root node. A type is considered to alias all of
2571 its descendants and all of its ancestors in the tree. Also, a type is
2572 considered to alias all types in other trees, so that bitcode produced
2573 from multiple front-ends is handled conservatively.
2575 If the third field is present, it's an integer which if equal to 1
2576 indicates that the type is "constant" (meaning
2577 ``pointsToConstantMemory`` should return true; see `other useful
2578 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2580 '``tbaa.struct``' Metadata
2581 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2583 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2584 aggregate assignment operations in C and similar languages, however it
2585 is defined to copy a contiguous region of memory, which is more than
2586 strictly necessary for aggregate types which contain holes due to
2587 padding. Also, it doesn't contain any TBAA information about the fields
2590 ``!tbaa.struct`` metadata can describe which memory subregions in a
2591 memcpy are padding and what the TBAA tags of the struct are.
2593 The current metadata format is very simple. ``!tbaa.struct`` metadata
2594 nodes are a list of operands which are in conceptual groups of three.
2595 For each group of three, the first operand gives the byte offset of a
2596 field in bytes, the second gives its size in bytes, and the third gives
2599 .. code-block:: llvm
2601 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2603 This describes a struct with two fields. The first is at offset 0 bytes
2604 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2605 and has size 4 bytes and has tbaa tag !2.
2607 Note that the fields need not be contiguous. In this example, there is a
2608 4 byte gap between the two fields. This gap represents padding which
2609 does not carry useful data and need not be preserved.
2611 '``fpmath``' Metadata
2612 ^^^^^^^^^^^^^^^^^^^^^
2614 ``fpmath`` metadata may be attached to any instruction of floating point
2615 type. It can be used to express the maximum acceptable error in the
2616 result of that instruction, in ULPs, thus potentially allowing the
2617 compiler to use a more efficient but less accurate method of computing
2618 it. ULP is defined as follows:
2620 If ``x`` is a real number that lies between two finite consecutive
2621 floating-point numbers ``a`` and ``b``, without being equal to one
2622 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2623 distance between the two non-equal finite floating-point numbers
2624 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2626 The metadata node shall consist of a single positive floating point
2627 number representing the maximum relative error, for example:
2629 .. code-block:: llvm
2631 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2633 '``range``' Metadata
2634 ^^^^^^^^^^^^^^^^^^^^
2636 ``range`` metadata may be attached only to loads of integer types. It
2637 expresses the possible ranges the loaded value is in. The ranges are
2638 represented with a flattened list of integers. The loaded value is known
2639 to be in the union of the ranges defined by each consecutive pair. Each
2640 pair has the following properties:
2642 - The type must match the type loaded by the instruction.
2643 - The pair ``a,b`` represents the range ``[a,b)``.
2644 - Both ``a`` and ``b`` are constants.
2645 - The range is allowed to wrap.
2646 - The range should not represent the full or empty set. That is,
2649 In addition, the pairs must be in signed order of the lower bound and
2650 they must be non-contiguous.
2654 .. code-block:: llvm
2656 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2657 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2658 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2659 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2661 !0 = metadata !{ i8 0, i8 2 }
2662 !1 = metadata !{ i8 255, i8 2 }
2663 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2664 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2669 It is sometimes useful to attach information to loop constructs. Currently,
2670 loop metadata is implemented as metadata attached to the branch instruction
2671 in the loop latch block. This type of metadata refer to a metadata node that is
2672 guaranteed to be separate for each loop. The loop identifier metadata is
2673 specified with the name ``llvm.loop``.
2675 The loop identifier metadata is implemented using a metadata that refers to
2676 itself to avoid merging it with any other identifier metadata, e.g.,
2677 during module linkage or function inlining. That is, each loop should refer
2678 to their own identification metadata even if they reside in separate functions.
2679 The following example contains loop identifier metadata for two separate loop
2682 .. code-block:: llvm
2684 !0 = metadata !{ metadata !0 }
2685 !1 = metadata !{ metadata !1 }
2687 The loop identifier metadata can be used to specify additional per-loop
2688 metadata. Any operands after the first operand can be treated as user-defined
2689 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2690 by the loop vectorizer to indicate how many times to unroll the loop:
2692 .. code-block:: llvm
2694 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2696 !0 = metadata !{ metadata !0, metadata !1 }
2697 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2702 Metadata types used to annotate memory accesses with information helpful
2703 for optimizations are prefixed with ``llvm.mem``.
2705 '``llvm.mem.parallel_loop_access``' Metadata
2706 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2708 For a loop to be parallel, in addition to using
2709 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2710 also all of the memory accessing instructions in the loop body need to be
2711 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2712 is at least one memory accessing instruction not marked with the metadata,
2713 the loop must be considered a sequential loop. This causes parallel loops to be
2714 converted to sequential loops due to optimization passes that are unaware of
2715 the parallel semantics and that insert new memory instructions to the loop
2718 Example of a loop that is considered parallel due to its correct use of
2719 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2720 metadata types that refer to the same loop identifier metadata.
2722 .. code-block:: llvm
2726 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2728 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2730 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2734 !0 = metadata !{ metadata !0 }
2736 It is also possible to have nested parallel loops. In that case the
2737 memory accesses refer to a list of loop identifier metadata nodes instead of
2738 the loop identifier metadata node directly:
2740 .. code-block:: llvm
2747 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2749 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2751 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2755 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2757 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2759 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2761 outer.for.end: ; preds = %for.body
2763 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2764 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2765 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2767 '``llvm.vectorizer``'
2768 ^^^^^^^^^^^^^^^^^^^^^
2770 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2771 vectorization parameters such as vectorization factor and unroll factor.
2773 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2774 loop identification metadata.
2776 '``llvm.vectorizer.unroll``' Metadata
2777 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2779 This metadata instructs the loop vectorizer to unroll the specified
2780 loop exactly ``N`` times.
2782 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2783 operand is an integer specifying the unroll factor. For example:
2785 .. code-block:: llvm
2787 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2789 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2792 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2793 determined automatically.
2795 '``llvm.vectorizer.width``' Metadata
2796 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2798 This metadata sets the target width of the vectorizer to ``N``. Without
2799 this metadata, the vectorizer will choose a width automatically.
2800 Regardless of this metadata, the vectorizer will only vectorize loops if
2801 it believes it is valid to do so.
2803 The first operand is the string ``llvm.vectorizer.width`` and the second
2804 operand is an integer specifying the width. For example:
2806 .. code-block:: llvm
2808 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2810 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2813 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2816 Module Flags Metadata
2817 =====================
2819 Information about the module as a whole is difficult to convey to LLVM's
2820 subsystems. The LLVM IR isn't sufficient to transmit this information.
2821 The ``llvm.module.flags`` named metadata exists in order to facilitate
2822 this. These flags are in the form of key / value pairs --- much like a
2823 dictionary --- making it easy for any subsystem who cares about a flag to
2826 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2827 Each triplet has the following form:
2829 - The first element is a *behavior* flag, which specifies the behavior
2830 when two (or more) modules are merged together, and it encounters two
2831 (or more) metadata with the same ID. The supported behaviors are
2833 - The second element is a metadata string that is a unique ID for the
2834 metadata. Each module may only have one flag entry for each unique ID (not
2835 including entries with the **Require** behavior).
2836 - The third element is the value of the flag.
2838 When two (or more) modules are merged together, the resulting
2839 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2840 each unique metadata ID string, there will be exactly one entry in the merged
2841 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2842 be determined by the merge behavior flag, as described below. The only exception
2843 is that entries with the *Require* behavior are always preserved.
2845 The following behaviors are supported:
2856 Emits an error if two values disagree, otherwise the resulting value
2857 is that of the operands.
2861 Emits a warning if two values disagree. The result value will be the
2862 operand for the flag from the first module being linked.
2866 Adds a requirement that another module flag be present and have a
2867 specified value after linking is performed. The value must be a
2868 metadata pair, where the first element of the pair is the ID of the
2869 module flag to be restricted, and the second element of the pair is
2870 the value the module flag should be restricted to. This behavior can
2871 be used to restrict the allowable results (via triggering of an
2872 error) of linking IDs with the **Override** behavior.
2876 Uses the specified value, regardless of the behavior or value of the
2877 other module. If both modules specify **Override**, but the values
2878 differ, an error will be emitted.
2882 Appends the two values, which are required to be metadata nodes.
2886 Appends the two values, which are required to be metadata
2887 nodes. However, duplicate entries in the second list are dropped
2888 during the append operation.
2890 It is an error for a particular unique flag ID to have multiple behaviors,
2891 except in the case of **Require** (which adds restrictions on another metadata
2892 value) or **Override**.
2894 An example of module flags:
2896 .. code-block:: llvm
2898 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2899 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2900 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2901 !3 = metadata !{ i32 3, metadata !"qux",
2903 metadata !"foo", i32 1
2906 !llvm.module.flags = !{ !0, !1, !2, !3 }
2908 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2909 if two or more ``!"foo"`` flags are seen is to emit an error if their
2910 values are not equal.
2912 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2913 behavior if two or more ``!"bar"`` flags are seen is to use the value
2916 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2917 behavior if two or more ``!"qux"`` flags are seen is to emit a
2918 warning if their values are not equal.
2920 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2924 metadata !{ metadata !"foo", i32 1 }
2926 The behavior is to emit an error if the ``llvm.module.flags`` does not
2927 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2930 Objective-C Garbage Collection Module Flags Metadata
2931 ----------------------------------------------------
2933 On the Mach-O platform, Objective-C stores metadata about garbage
2934 collection in a special section called "image info". The metadata
2935 consists of a version number and a bitmask specifying what types of
2936 garbage collection are supported (if any) by the file. If two or more
2937 modules are linked together their garbage collection metadata needs to
2938 be merged rather than appended together.
2940 The Objective-C garbage collection module flags metadata consists of the
2941 following key-value pairs:
2950 * - ``Objective-C Version``
2951 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2953 * - ``Objective-C Image Info Version``
2954 - **[Required]** --- The version of the image info section. Currently
2957 * - ``Objective-C Image Info Section``
2958 - **[Required]** --- The section to place the metadata. Valid values are
2959 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2960 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2961 Objective-C ABI version 2.
2963 * - ``Objective-C Garbage Collection``
2964 - **[Required]** --- Specifies whether garbage collection is supported or
2965 not. Valid values are 0, for no garbage collection, and 2, for garbage
2966 collection supported.
2968 * - ``Objective-C GC Only``
2969 - **[Optional]** --- Specifies that only garbage collection is supported.
2970 If present, its value must be 6. This flag requires that the
2971 ``Objective-C Garbage Collection`` flag have the value 2.
2973 Some important flag interactions:
2975 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2976 merged with a module with ``Objective-C Garbage Collection`` set to
2977 2, then the resulting module has the
2978 ``Objective-C Garbage Collection`` flag set to 0.
2979 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2980 merged with a module with ``Objective-C GC Only`` set to 6.
2982 Automatic Linker Flags Module Flags Metadata
2983 --------------------------------------------
2985 Some targets support embedding flags to the linker inside individual object
2986 files. Typically this is used in conjunction with language extensions which
2987 allow source files to explicitly declare the libraries they depend on, and have
2988 these automatically be transmitted to the linker via object files.
2990 These flags are encoded in the IR using metadata in the module flags section,
2991 using the ``Linker Options`` key. The merge behavior for this flag is required
2992 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2993 node which should be a list of other metadata nodes, each of which should be a
2994 list of metadata strings defining linker options.
2996 For example, the following metadata section specifies two separate sets of
2997 linker options, presumably to link against ``libz`` and the ``Cocoa``
3000 !0 = metadata !{ i32 6, metadata !"Linker Options",
3002 metadata !{ metadata !"-lz" },
3003 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3004 !llvm.module.flags = !{ !0 }
3006 The metadata encoding as lists of lists of options, as opposed to a collapsed
3007 list of options, is chosen so that the IR encoding can use multiple option
3008 strings to specify e.g., a single library, while still having that specifier be
3009 preserved as an atomic element that can be recognized by a target specific
3010 assembly writer or object file emitter.
3012 Each individual option is required to be either a valid option for the target's
3013 linker, or an option that is reserved by the target specific assembly writer or
3014 object file emitter. No other aspect of these options is defined by the IR.
3016 .. _intrinsicglobalvariables:
3018 Intrinsic Global Variables
3019 ==========================
3021 LLVM has a number of "magic" global variables that contain data that
3022 affect code generation or other IR semantics. These are documented here.
3023 All globals of this sort should have a section specified as
3024 "``llvm.metadata``". This section and all globals that start with
3025 "``llvm.``" are reserved for use by LLVM.
3029 The '``llvm.used``' Global Variable
3030 -----------------------------------
3032 The ``@llvm.used`` global is an array which has
3033 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3034 pointers to named global variables, functions and aliases which may optionally
3035 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3038 .. code-block:: llvm
3043 @llvm.used = appending global [2 x i8*] [
3045 i8* bitcast (i32* @Y to i8*)
3046 ], section "llvm.metadata"
3048 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3049 and linker are required to treat the symbol as if there is a reference to the
3050 symbol that it cannot see (which is why they have to be named). For example, if
3051 a variable has internal linkage and no references other than that from the
3052 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3053 references from inline asms and other things the compiler cannot "see", and
3054 corresponds to "``attribute((used))``" in GNU C.
3056 On some targets, the code generator must emit a directive to the
3057 assembler or object file to prevent the assembler and linker from
3058 molesting the symbol.
3060 .. _gv_llvmcompilerused:
3062 The '``llvm.compiler.used``' Global Variable
3063 --------------------------------------------
3065 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3066 directive, except that it only prevents the compiler from touching the
3067 symbol. On targets that support it, this allows an intelligent linker to
3068 optimize references to the symbol without being impeded as it would be
3071 This is a rare construct that should only be used in rare circumstances,
3072 and should not be exposed to source languages.
3074 .. _gv_llvmglobalctors:
3076 The '``llvm.global_ctors``' Global Variable
3077 -------------------------------------------
3079 .. code-block:: llvm
3081 %0 = type { i32, void ()* }
3082 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3084 The ``@llvm.global_ctors`` array contains a list of constructor
3085 functions and associated priorities. The functions referenced by this
3086 array will be called in ascending order of priority (i.e. lowest first)
3087 when the module is loaded. The order of functions with the same priority
3090 .. _llvmglobaldtors:
3092 The '``llvm.global_dtors``' Global Variable
3093 -------------------------------------------
3095 .. code-block:: llvm
3097 %0 = type { i32, void ()* }
3098 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3100 The ``@llvm.global_dtors`` array contains a list of destructor functions
3101 and associated priorities. The functions referenced by this array will
3102 be called in descending order of priority (i.e. highest first) when the
3103 module is loaded. The order of functions with the same priority is not
3106 Instruction Reference
3107 =====================
3109 The LLVM instruction set consists of several different classifications
3110 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3111 instructions <binaryops>`, :ref:`bitwise binary
3112 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3113 :ref:`other instructions <otherops>`.
3117 Terminator Instructions
3118 -----------------------
3120 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3121 program ends with a "Terminator" instruction, which indicates which
3122 block should be executed after the current block is finished. These
3123 terminator instructions typically yield a '``void``' value: they produce
3124 control flow, not values (the one exception being the
3125 ':ref:`invoke <i_invoke>`' instruction).
3127 The terminator instructions are: ':ref:`ret <i_ret>`',
3128 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3129 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3130 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3134 '``ret``' Instruction
3135 ^^^^^^^^^^^^^^^^^^^^^
3142 ret <type> <value> ; Return a value from a non-void function
3143 ret void ; Return from void function
3148 The '``ret``' instruction is used to return control flow (and optionally
3149 a value) from a function back to the caller.
3151 There are two forms of the '``ret``' instruction: one that returns a
3152 value and then causes control flow, and one that just causes control
3158 The '``ret``' instruction optionally accepts a single argument, the
3159 return value. The type of the return value must be a ':ref:`first
3160 class <t_firstclass>`' type.
3162 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3163 return type and contains a '``ret``' instruction with no return value or
3164 a return value with a type that does not match its type, or if it has a
3165 void return type and contains a '``ret``' instruction with a return
3171 When the '``ret``' instruction is executed, control flow returns back to
3172 the calling function's context. If the caller is a
3173 ":ref:`call <i_call>`" instruction, execution continues at the
3174 instruction after the call. If the caller was an
3175 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3176 beginning of the "normal" destination block. If the instruction returns
3177 a value, that value shall set the call or invoke instruction's return
3183 .. code-block:: llvm
3185 ret i32 5 ; Return an integer value of 5
3186 ret void ; Return from a void function
3187 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3191 '``br``' Instruction
3192 ^^^^^^^^^^^^^^^^^^^^
3199 br i1 <cond>, label <iftrue>, label <iffalse>
3200 br label <dest> ; Unconditional branch
3205 The '``br``' instruction is used to cause control flow to transfer to a
3206 different basic block in the current function. There are two forms of
3207 this instruction, corresponding to a conditional branch and an
3208 unconditional branch.
3213 The conditional branch form of the '``br``' instruction takes a single
3214 '``i1``' value and two '``label``' values. The unconditional form of the
3215 '``br``' instruction takes a single '``label``' value as a target.
3220 Upon execution of a conditional '``br``' instruction, the '``i1``'
3221 argument is evaluated. If the value is ``true``, control flows to the
3222 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3223 to the '``iffalse``' ``label`` argument.
3228 .. code-block:: llvm
3231 %cond = icmp eq i32 %a, %b
3232 br i1 %cond, label %IfEqual, label %IfUnequal
3240 '``switch``' Instruction
3241 ^^^^^^^^^^^^^^^^^^^^^^^^
3248 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3253 The '``switch``' instruction is used to transfer control flow to one of
3254 several different places. It is a generalization of the '``br``'
3255 instruction, allowing a branch to occur to one of many possible
3261 The '``switch``' instruction uses three parameters: an integer
3262 comparison value '``value``', a default '``label``' destination, and an
3263 array of pairs of comparison value constants and '``label``'s. The table
3264 is not allowed to contain duplicate constant entries.
3269 The ``switch`` instruction specifies a table of values and destinations.
3270 When the '``switch``' instruction is executed, this table is searched
3271 for the given value. If the value is found, control flow is transferred
3272 to the corresponding destination; otherwise, control flow is transferred
3273 to the default destination.
3278 Depending on properties of the target machine and the particular
3279 ``switch`` instruction, this instruction may be code generated in
3280 different ways. For example, it could be generated as a series of
3281 chained conditional branches or with a lookup table.
3286 .. code-block:: llvm
3288 ; Emulate a conditional br instruction
3289 %Val = zext i1 %value to i32
3290 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3292 ; Emulate an unconditional br instruction
3293 switch i32 0, label %dest [ ]
3295 ; Implement a jump table:
3296 switch i32 %val, label %otherwise [ i32 0, label %onzero
3298 i32 2, label %ontwo ]
3302 '``indirectbr``' Instruction
3303 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3310 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3315 The '``indirectbr``' instruction implements an indirect branch to a
3316 label within the current function, whose address is specified by
3317 "``address``". Address must be derived from a
3318 :ref:`blockaddress <blockaddress>` constant.
3323 The '``address``' argument is the address of the label to jump to. The
3324 rest of the arguments indicate the full set of possible destinations
3325 that the address may point to. Blocks are allowed to occur multiple
3326 times in the destination list, though this isn't particularly useful.
3328 This destination list is required so that dataflow analysis has an
3329 accurate understanding of the CFG.
3334 Control transfers to the block specified in the address argument. All
3335 possible destination blocks must be listed in the label list, otherwise
3336 this instruction has undefined behavior. This implies that jumps to
3337 labels defined in other functions have undefined behavior as well.
3342 This is typically implemented with a jump through a register.
3347 .. code-block:: llvm
3349 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3353 '``invoke``' Instruction
3354 ^^^^^^^^^^^^^^^^^^^^^^^^
3361 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3362 to label <normal label> unwind label <exception label>
3367 The '``invoke``' instruction causes control to transfer to a specified
3368 function, with the possibility of control flow transfer to either the
3369 '``normal``' label or the '``exception``' label. If the callee function
3370 returns with the "``ret``" instruction, control flow will return to the
3371 "normal" label. If the callee (or any indirect callees) returns via the
3372 ":ref:`resume <i_resume>`" instruction or other exception handling
3373 mechanism, control is interrupted and continued at the dynamically
3374 nearest "exception" label.
3376 The '``exception``' label is a `landing
3377 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3378 '``exception``' label is required to have the
3379 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3380 information about the behavior of the program after unwinding happens,
3381 as its first non-PHI instruction. The restrictions on the
3382 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3383 instruction, so that the important information contained within the
3384 "``landingpad``" instruction can't be lost through normal code motion.
3389 This instruction requires several arguments:
3391 #. The optional "cconv" marker indicates which :ref:`calling
3392 convention <callingconv>` the call should use. If none is
3393 specified, the call defaults to using C calling conventions.
3394 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3395 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3397 #. '``ptr to function ty``': shall be the signature of the pointer to
3398 function value being invoked. In most cases, this is a direct
3399 function invocation, but indirect ``invoke``'s are just as possible,
3400 branching off an arbitrary pointer to function value.
3401 #. '``function ptr val``': An LLVM value containing a pointer to a
3402 function to be invoked.
3403 #. '``function args``': argument list whose types match the function
3404 signature argument types and parameter attributes. All arguments must
3405 be of :ref:`first class <t_firstclass>` type. If the function signature
3406 indicates the function accepts a variable number of arguments, the
3407 extra arguments can be specified.
3408 #. '``normal label``': the label reached when the called function
3409 executes a '``ret``' instruction.
3410 #. '``exception label``': the label reached when a callee returns via
3411 the :ref:`resume <i_resume>` instruction or other exception handling
3413 #. The optional :ref:`function attributes <fnattrs>` list. Only
3414 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3415 attributes are valid here.
3420 This instruction is designed to operate as a standard '``call``'
3421 instruction in most regards. The primary difference is that it
3422 establishes an association with a label, which is used by the runtime
3423 library to unwind the stack.
3425 This instruction is used in languages with destructors to ensure that
3426 proper cleanup is performed in the case of either a ``longjmp`` or a
3427 thrown exception. Additionally, this is important for implementation of
3428 '``catch``' clauses in high-level languages that support them.
3430 For the purposes of the SSA form, the definition of the value returned
3431 by the '``invoke``' instruction is deemed to occur on the edge from the
3432 current block to the "normal" label. If the callee unwinds then no
3433 return value is available.
3438 .. code-block:: llvm
3440 %retval = invoke i32 @Test(i32 15) to label %Continue
3441 unwind label %TestCleanup ; {i32}:retval set
3442 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3443 unwind label %TestCleanup ; {i32}:retval set
3447 '``resume``' Instruction
3448 ^^^^^^^^^^^^^^^^^^^^^^^^
3455 resume <type> <value>
3460 The '``resume``' instruction is a terminator instruction that has no
3466 The '``resume``' instruction requires one argument, which must have the
3467 same type as the result of any '``landingpad``' instruction in the same
3473 The '``resume``' instruction resumes propagation of an existing
3474 (in-flight) exception whose unwinding was interrupted with a
3475 :ref:`landingpad <i_landingpad>` instruction.
3480 .. code-block:: llvm
3482 resume { i8*, i32 } %exn
3486 '``unreachable``' Instruction
3487 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3499 The '``unreachable``' instruction has no defined semantics. This
3500 instruction is used to inform the optimizer that a particular portion of
3501 the code is not reachable. This can be used to indicate that the code
3502 after a no-return function cannot be reached, and other facts.
3507 The '``unreachable``' instruction has no defined semantics.
3514 Binary operators are used to do most of the computation in a program.
3515 They require two operands of the same type, execute an operation on
3516 them, and produce a single value. The operands might represent multiple
3517 data, as is the case with the :ref:`vector <t_vector>` data type. The
3518 result value has the same type as its operands.
3520 There are several different binary operators:
3524 '``add``' Instruction
3525 ^^^^^^^^^^^^^^^^^^^^^
3532 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3533 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3534 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3535 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3540 The '``add``' instruction returns the sum of its two operands.
3545 The two arguments to the '``add``' instruction must be
3546 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3547 arguments must have identical types.
3552 The value produced is the integer sum of the two operands.
3554 If the sum has unsigned overflow, the result returned is the
3555 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3558 Because LLVM integers use a two's complement representation, this
3559 instruction is appropriate for both signed and unsigned integers.
3561 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3562 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3563 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3564 unsigned and/or signed overflow, respectively, occurs.
3569 .. code-block:: llvm
3571 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3575 '``fadd``' Instruction
3576 ^^^^^^^^^^^^^^^^^^^^^^
3583 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3588 The '``fadd``' instruction returns the sum of its two operands.
3593 The two arguments to the '``fadd``' instruction must be :ref:`floating
3594 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3595 Both arguments must have identical types.
3600 The value produced is the floating point sum of the two operands. This
3601 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3602 which are optimization hints to enable otherwise unsafe floating point
3608 .. code-block:: llvm
3610 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3612 '``sub``' Instruction
3613 ^^^^^^^^^^^^^^^^^^^^^
3620 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3621 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3622 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3623 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3628 The '``sub``' instruction returns the difference of its two operands.
3630 Note that the '``sub``' instruction is used to represent the '``neg``'
3631 instruction present in most other intermediate representations.
3636 The two arguments to the '``sub``' instruction must be
3637 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3638 arguments must have identical types.
3643 The value produced is the integer difference of the two operands.
3645 If the difference has unsigned overflow, the result returned is the
3646 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3649 Because LLVM integers use a two's complement representation, this
3650 instruction is appropriate for both signed and unsigned integers.
3652 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3653 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3654 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3655 unsigned and/or signed overflow, respectively, occurs.
3660 .. code-block:: llvm
3662 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3663 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3667 '``fsub``' Instruction
3668 ^^^^^^^^^^^^^^^^^^^^^^
3675 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3680 The '``fsub``' instruction returns the difference of its two operands.
3682 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3683 instruction present in most other intermediate representations.
3688 The two arguments to the '``fsub``' instruction must be :ref:`floating
3689 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3690 Both arguments must have identical types.
3695 The value produced is the floating point difference of the two operands.
3696 This instruction can also take any number of :ref:`fast-math
3697 flags <fastmath>`, which are optimization hints to enable otherwise
3698 unsafe floating point optimizations:
3703 .. code-block:: llvm
3705 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3706 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3708 '``mul``' Instruction
3709 ^^^^^^^^^^^^^^^^^^^^^
3716 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3717 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3718 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3719 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3724 The '``mul``' instruction returns the product of its two operands.
3729 The two arguments to the '``mul``' instruction must be
3730 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3731 arguments must have identical types.
3736 The value produced is the integer product of the two operands.
3738 If the result of the multiplication has unsigned overflow, the result
3739 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3740 bit width of the result.
3742 Because LLVM integers use a two's complement representation, and the
3743 result is the same width as the operands, this instruction returns the
3744 correct result for both signed and unsigned integers. If a full product
3745 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3746 sign-extended or zero-extended as appropriate to the width of the full
3749 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3750 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3751 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3752 unsigned and/or signed overflow, respectively, occurs.
3757 .. code-block:: llvm
3759 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3763 '``fmul``' Instruction
3764 ^^^^^^^^^^^^^^^^^^^^^^
3771 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3776 The '``fmul``' instruction returns the product of its two operands.
3781 The two arguments to the '``fmul``' instruction must be :ref:`floating
3782 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3783 Both arguments must have identical types.
3788 The value produced is the floating point product of the two operands.
3789 This instruction can also take any number of :ref:`fast-math
3790 flags <fastmath>`, which are optimization hints to enable otherwise
3791 unsafe floating point optimizations:
3796 .. code-block:: llvm
3798 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3800 '``udiv``' Instruction
3801 ^^^^^^^^^^^^^^^^^^^^^^
3808 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3809 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3814 The '``udiv``' instruction returns the quotient of its two operands.
3819 The two arguments to the '``udiv``' instruction must be
3820 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3821 arguments must have identical types.
3826 The value produced is the unsigned integer quotient of the two operands.
3828 Note that unsigned integer division and signed integer division are
3829 distinct operations; for signed integer division, use '``sdiv``'.
3831 Division by zero leads to undefined behavior.
3833 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3834 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3835 such, "((a udiv exact b) mul b) == a").
3840 .. code-block:: llvm
3842 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3844 '``sdiv``' Instruction
3845 ^^^^^^^^^^^^^^^^^^^^^^
3852 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3853 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3858 The '``sdiv``' instruction returns the quotient of its two operands.
3863 The two arguments to the '``sdiv``' instruction must be
3864 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3865 arguments must have identical types.
3870 The value produced is the signed integer quotient of the two operands
3871 rounded towards zero.
3873 Note that signed integer division and unsigned integer division are
3874 distinct operations; for unsigned integer division, use '``udiv``'.
3876 Division by zero leads to undefined behavior. Overflow also leads to
3877 undefined behavior; this is a rare case, but can occur, for example, by
3878 doing a 32-bit division of -2147483648 by -1.
3880 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3881 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3886 .. code-block:: llvm
3888 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3892 '``fdiv``' Instruction
3893 ^^^^^^^^^^^^^^^^^^^^^^
3900 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3905 The '``fdiv``' instruction returns the quotient of its two operands.
3910 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3911 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3912 Both arguments must have identical types.
3917 The value produced is the floating point quotient of the two operands.
3918 This instruction can also take any number of :ref:`fast-math
3919 flags <fastmath>`, which are optimization hints to enable otherwise
3920 unsafe floating point optimizations:
3925 .. code-block:: llvm
3927 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3929 '``urem``' Instruction
3930 ^^^^^^^^^^^^^^^^^^^^^^
3937 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3942 The '``urem``' instruction returns the remainder from the unsigned
3943 division of its two arguments.
3948 The two arguments to the '``urem``' instruction must be
3949 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3950 arguments must have identical types.
3955 This instruction returns the unsigned integer *remainder* of a division.
3956 This instruction always performs an unsigned division to get the
3959 Note that unsigned integer remainder and signed integer remainder are
3960 distinct operations; for signed integer remainder, use '``srem``'.
3962 Taking the remainder of a division by zero leads to undefined behavior.
3967 .. code-block:: llvm
3969 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3971 '``srem``' Instruction
3972 ^^^^^^^^^^^^^^^^^^^^^^
3979 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3984 The '``srem``' instruction returns the remainder from the signed
3985 division of its two operands. This instruction can also take
3986 :ref:`vector <t_vector>` versions of the values in which case the elements
3992 The two arguments to the '``srem``' instruction must be
3993 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3994 arguments must have identical types.
3999 This instruction returns the *remainder* of a division (where the result
4000 is either zero or has the same sign as the dividend, ``op1``), not the
4001 *modulo* operator (where the result is either zero or has the same sign
4002 as the divisor, ``op2``) of a value. For more information about the
4003 difference, see `The Math
4004 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4005 table of how this is implemented in various languages, please see
4007 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4009 Note that signed integer remainder and unsigned integer remainder are
4010 distinct operations; for unsigned integer remainder, use '``urem``'.
4012 Taking the remainder of a division by zero leads to undefined behavior.
4013 Overflow also leads to undefined behavior; this is a rare case, but can
4014 occur, for example, by taking the remainder of a 32-bit division of
4015 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4016 rule lets srem be implemented using instructions that return both the
4017 result of the division and the remainder.)
4022 .. code-block:: llvm
4024 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4028 '``frem``' Instruction
4029 ^^^^^^^^^^^^^^^^^^^^^^
4036 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4041 The '``frem``' instruction returns the remainder from the division of
4047 The two arguments to the '``frem``' instruction must be :ref:`floating
4048 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4049 Both arguments must have identical types.
4054 This instruction returns the *remainder* of a division. The remainder
4055 has the same sign as the dividend. This instruction can also take any
4056 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4057 to enable otherwise unsafe floating point optimizations:
4062 .. code-block:: llvm
4064 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4068 Bitwise Binary Operations
4069 -------------------------
4071 Bitwise binary operators are used to do various forms of bit-twiddling
4072 in a program. They are generally very efficient instructions and can
4073 commonly be strength reduced from other instructions. They require two
4074 operands of the same type, execute an operation on them, and produce a
4075 single value. The resulting value is the same type as its operands.
4077 '``shl``' Instruction
4078 ^^^^^^^^^^^^^^^^^^^^^
4085 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4086 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4087 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4088 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4093 The '``shl``' instruction returns the first operand shifted to the left
4094 a specified number of bits.
4099 Both arguments to the '``shl``' instruction must be the same
4100 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4101 '``op2``' is treated as an unsigned value.
4106 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4107 where ``n`` is the width of the result. If ``op2`` is (statically or
4108 dynamically) negative or equal to or larger than the number of bits in
4109 ``op1``, the result is undefined. If the arguments are vectors, each
4110 vector element of ``op1`` is shifted by the corresponding shift amount
4113 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4114 value <poisonvalues>` if it shifts out any non-zero bits. If the
4115 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4116 value <poisonvalues>` if it shifts out any bits that disagree with the
4117 resultant sign bit. As such, NUW/NSW have the same semantics as they
4118 would if the shift were expressed as a mul instruction with the same
4119 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4124 .. code-block:: llvm
4126 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4127 <result> = shl i32 4, 2 ; yields {i32}: 16
4128 <result> = shl i32 1, 10 ; yields {i32}: 1024
4129 <result> = shl i32 1, 32 ; undefined
4130 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4132 '``lshr``' Instruction
4133 ^^^^^^^^^^^^^^^^^^^^^^
4140 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4141 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4146 The '``lshr``' instruction (logical shift right) returns the first
4147 operand shifted to the right a specified number of bits with zero fill.
4152 Both arguments to the '``lshr``' instruction must be the same
4153 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4154 '``op2``' is treated as an unsigned value.
4159 This instruction always performs a logical shift right operation. The
4160 most significant bits of the result will be filled with zero bits after
4161 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4162 than the number of bits in ``op1``, the result is undefined. If the
4163 arguments are vectors, each vector element of ``op1`` is shifted by the
4164 corresponding shift amount in ``op2``.
4166 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4167 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4173 .. code-block:: llvm
4175 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4176 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4177 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4178 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4179 <result> = lshr i32 1, 32 ; undefined
4180 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4182 '``ashr``' Instruction
4183 ^^^^^^^^^^^^^^^^^^^^^^
4190 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4191 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4196 The '``ashr``' instruction (arithmetic shift right) returns the first
4197 operand shifted to the right a specified number of bits with sign
4203 Both arguments to the '``ashr``' instruction must be the same
4204 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4205 '``op2``' is treated as an unsigned value.
4210 This instruction always performs an arithmetic shift right operation,
4211 The most significant bits of the result will be filled with the sign bit
4212 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4213 than the number of bits in ``op1``, the result is undefined. If the
4214 arguments are vectors, each vector element of ``op1`` is shifted by the
4215 corresponding shift amount in ``op2``.
4217 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4218 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4224 .. code-block:: llvm
4226 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4227 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4228 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4229 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4230 <result> = ashr i32 1, 32 ; undefined
4231 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4233 '``and``' Instruction
4234 ^^^^^^^^^^^^^^^^^^^^^
4241 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4246 The '``and``' instruction returns the bitwise logical and of its two
4252 The two arguments to the '``and``' instruction must be
4253 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4254 arguments must have identical types.
4259 The truth table used for the '``and``' instruction is:
4276 .. code-block:: llvm
4278 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4279 <result> = and i32 15, 40 ; yields {i32}:result = 8
4280 <result> = and i32 4, 8 ; yields {i32}:result = 0
4282 '``or``' Instruction
4283 ^^^^^^^^^^^^^^^^^^^^
4290 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4295 The '``or``' instruction returns the bitwise logical inclusive or of its
4301 The two arguments to the '``or``' instruction must be
4302 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4303 arguments must have identical types.
4308 The truth table used for the '``or``' instruction is:
4327 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4328 <result> = or i32 15, 40 ; yields {i32}:result = 47
4329 <result> = or i32 4, 8 ; yields {i32}:result = 12
4331 '``xor``' Instruction
4332 ^^^^^^^^^^^^^^^^^^^^^
4339 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4344 The '``xor``' instruction returns the bitwise logical exclusive or of
4345 its two operands. The ``xor`` is used to implement the "one's
4346 complement" operation, which is the "~" operator in C.
4351 The two arguments to the '``xor``' instruction must be
4352 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4353 arguments must have identical types.
4358 The truth table used for the '``xor``' instruction is:
4375 .. code-block:: llvm
4377 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4378 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4379 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4380 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4385 LLVM supports several instructions to represent vector operations in a
4386 target-independent manner. These instructions cover the element-access
4387 and vector-specific operations needed to process vectors effectively.
4388 While LLVM does directly support these vector operations, many
4389 sophisticated algorithms will want to use target-specific intrinsics to
4390 take full advantage of a specific target.
4392 .. _i_extractelement:
4394 '``extractelement``' Instruction
4395 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4402 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4407 The '``extractelement``' instruction extracts a single scalar element
4408 from a vector at a specified index.
4413 The first operand of an '``extractelement``' instruction is a value of
4414 :ref:`vector <t_vector>` type. The second operand is an index indicating
4415 the position from which to extract the element. The index may be a
4421 The result is a scalar of the same type as the element type of ``val``.
4422 Its value is the value at position ``idx`` of ``val``. If ``idx``
4423 exceeds the length of ``val``, the results are undefined.
4428 .. code-block:: llvm
4430 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4432 .. _i_insertelement:
4434 '``insertelement``' Instruction
4435 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4442 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4447 The '``insertelement``' instruction inserts a scalar element into a
4448 vector at a specified index.
4453 The first operand of an '``insertelement``' instruction is a value of
4454 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4455 type must equal the element type of the first operand. The third operand
4456 is an index indicating the position at which to insert the value. The
4457 index may be a variable.
4462 The result is a vector of the same type as ``val``. Its element values
4463 are those of ``val`` except at position ``idx``, where it gets the value
4464 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4470 .. code-block:: llvm
4472 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4474 .. _i_shufflevector:
4476 '``shufflevector``' Instruction
4477 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4484 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4489 The '``shufflevector``' instruction constructs a permutation of elements
4490 from two input vectors, returning a vector with the same element type as
4491 the input and length that is the same as the shuffle mask.
4496 The first two operands of a '``shufflevector``' instruction are vectors
4497 with the same type. The third argument is a shuffle mask whose element
4498 type is always 'i32'. The result of the instruction is a vector whose
4499 length is the same as the shuffle mask and whose element type is the
4500 same as the element type of the first two operands.
4502 The shuffle mask operand is required to be a constant vector with either
4503 constant integer or undef values.
4508 The elements of the two input vectors are numbered from left to right
4509 across both of the vectors. The shuffle mask operand specifies, for each
4510 element of the result vector, which element of the two input vectors the
4511 result element gets. The element selector may be undef (meaning "don't
4512 care") and the second operand may be undef if performing a shuffle from
4518 .. code-block:: llvm
4520 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4521 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4522 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4523 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4524 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4525 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4526 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4527 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4529 Aggregate Operations
4530 --------------------
4532 LLVM supports several instructions for working with
4533 :ref:`aggregate <t_aggregate>` values.
4537 '``extractvalue``' Instruction
4538 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4545 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4550 The '``extractvalue``' instruction extracts the value of a member field
4551 from an :ref:`aggregate <t_aggregate>` value.
4556 The first operand of an '``extractvalue``' instruction is a value of
4557 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4558 constant indices to specify which value to extract in a similar manner
4559 as indices in a '``getelementptr``' instruction.
4561 The major differences to ``getelementptr`` indexing are:
4563 - Since the value being indexed is not a pointer, the first index is
4564 omitted and assumed to be zero.
4565 - At least one index must be specified.
4566 - Not only struct indices but also array indices must be in bounds.
4571 The result is the value at the position in the aggregate specified by
4577 .. code-block:: llvm
4579 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4583 '``insertvalue``' Instruction
4584 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4591 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4596 The '``insertvalue``' instruction inserts a value into a member field in
4597 an :ref:`aggregate <t_aggregate>` value.
4602 The first operand of an '``insertvalue``' instruction is a value of
4603 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4604 a first-class value to insert. The following operands are constant
4605 indices indicating the position at which to insert the value in a
4606 similar manner as indices in a '``extractvalue``' instruction. The value
4607 to insert must have the same type as the value identified by the
4613 The result is an aggregate of the same type as ``val``. Its value is
4614 that of ``val`` except that the value at the position specified by the
4615 indices is that of ``elt``.
4620 .. code-block:: llvm
4622 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4623 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4624 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4628 Memory Access and Addressing Operations
4629 ---------------------------------------
4631 A key design point of an SSA-based representation is how it represents
4632 memory. In LLVM, no memory locations are in SSA form, which makes things
4633 very simple. This section describes how to read, write, and allocate
4638 '``alloca``' Instruction
4639 ^^^^^^^^^^^^^^^^^^^^^^^^
4646 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4651 The '``alloca``' instruction allocates memory on the stack frame of the
4652 currently executing function, to be automatically released when this
4653 function returns to its caller. The object is always allocated in the
4654 generic address space (address space zero).
4659 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4660 bytes of memory on the runtime stack, returning a pointer of the
4661 appropriate type to the program. If "NumElements" is specified, it is
4662 the number of elements allocated, otherwise "NumElements" is defaulted
4663 to be one. If a constant alignment is specified, the value result of the
4664 allocation is guaranteed to be aligned to at least that boundary. If not
4665 specified, or if zero, the target can choose to align the allocation on
4666 any convenient boundary compatible with the type.
4668 '``type``' may be any sized type.
4673 Memory is allocated; a pointer is returned. The operation is undefined
4674 if there is insufficient stack space for the allocation. '``alloca``'d
4675 memory is automatically released when the function returns. The
4676 '``alloca``' instruction is commonly used to represent automatic
4677 variables that must have an address available. When the function returns
4678 (either with the ``ret`` or ``resume`` instructions), the memory is
4679 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4680 The order in which memory is allocated (ie., which way the stack grows)
4686 .. code-block:: llvm
4688 %ptr = alloca i32 ; yields {i32*}:ptr
4689 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4690 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4691 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4695 '``load``' Instruction
4696 ^^^^^^^^^^^^^^^^^^^^^^
4703 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4704 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4705 !<index> = !{ i32 1 }
4710 The '``load``' instruction is used to read from memory.
4715 The argument to the ``load`` instruction specifies the memory address
4716 from which to load. The pointer must point to a :ref:`first
4717 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4718 then the optimizer is not allowed to modify the number or order of
4719 execution of this ``load`` with other :ref:`volatile
4720 operations <volatile>`.
4722 If the ``load`` is marked as ``atomic``, it takes an extra
4723 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4724 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4725 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4726 when they may see multiple atomic stores. The type of the pointee must
4727 be an integer type whose bit width is a power of two greater than or
4728 equal to eight and less than or equal to a target-specific size limit.
4729 ``align`` must be explicitly specified on atomic loads, and the load has
4730 undefined behavior if the alignment is not set to a value which is at
4731 least the size in bytes of the pointee. ``!nontemporal`` does not have
4732 any defined semantics for atomic loads.
4734 The optional constant ``align`` argument specifies the alignment of the
4735 operation (that is, the alignment of the memory address). A value of 0
4736 or an omitted ``align`` argument means that the operation has the ABI
4737 alignment for the target. It is the responsibility of the code emitter
4738 to ensure that the alignment information is correct. Overestimating the
4739 alignment results in undefined behavior. Underestimating the alignment
4740 may produce less efficient code. An alignment of 1 is always safe.
4742 The optional ``!nontemporal`` metadata must reference a single
4743 metadata name ``<index>`` corresponding to a metadata node with one
4744 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4745 metadata on the instruction tells the optimizer and code generator
4746 that this load is not expected to be reused in the cache. The code
4747 generator may select special instructions to save cache bandwidth, such
4748 as the ``MOVNT`` instruction on x86.
4750 The optional ``!invariant.load`` metadata must reference a single
4751 metadata name ``<index>`` corresponding to a metadata node with no
4752 entries. The existence of the ``!invariant.load`` metadata on the
4753 instruction tells the optimizer and code generator that this load
4754 address points to memory which does not change value during program
4755 execution. The optimizer may then move this load around, for example, by
4756 hoisting it out of loops using loop invariant code motion.
4761 The location of memory pointed to is loaded. If the value being loaded
4762 is of scalar type then the number of bytes read does not exceed the
4763 minimum number of bytes needed to hold all bits of the type. For
4764 example, loading an ``i24`` reads at most three bytes. When loading a
4765 value of a type like ``i20`` with a size that is not an integral number
4766 of bytes, the result is undefined if the value was not originally
4767 written using a store of the same type.
4772 .. code-block:: llvm
4774 %ptr = alloca i32 ; yields {i32*}:ptr
4775 store i32 3, i32* %ptr ; yields {void}
4776 %val = load i32* %ptr ; yields {i32}:val = i32 3
4780 '``store``' Instruction
4781 ^^^^^^^^^^^^^^^^^^^^^^^
4788 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4789 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4794 The '``store``' instruction is used to write to memory.
4799 There are two arguments to the ``store`` instruction: a value to store
4800 and an address at which to store it. The type of the ``<pointer>``
4801 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4802 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4803 then the optimizer is not allowed to modify the number or order of
4804 execution of this ``store`` with other :ref:`volatile
4805 operations <volatile>`.
4807 If the ``store`` is marked as ``atomic``, it takes an extra
4808 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4809 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4810 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4811 when they may see multiple atomic stores. The type of the pointee must
4812 be an integer type whose bit width is a power of two greater than or
4813 equal to eight and less than or equal to a target-specific size limit.
4814 ``align`` must be explicitly specified on atomic stores, and the store
4815 has undefined behavior if the alignment is not set to a value which is
4816 at least the size in bytes of the pointee. ``!nontemporal`` does not
4817 have any defined semantics for atomic stores.
4819 The optional constant ``align`` argument specifies the alignment of the
4820 operation (that is, the alignment of the memory address). A value of 0
4821 or an omitted ``align`` argument means that the operation has the ABI
4822 alignment for the target. It is the responsibility of the code emitter
4823 to ensure that the alignment information is correct. Overestimating the
4824 alignment results in undefined behavior. Underestimating the
4825 alignment may produce less efficient code. An alignment of 1 is always
4828 The optional ``!nontemporal`` metadata must reference a single metadata
4829 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4830 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4831 tells the optimizer and code generator that this load is not expected to
4832 be reused in the cache. The code generator may select special
4833 instructions to save cache bandwidth, such as the MOVNT instruction on
4839 The contents of memory are updated to contain ``<value>`` at the
4840 location specified by the ``<pointer>`` operand. If ``<value>`` is
4841 of scalar type then the number of bytes written does not exceed the
4842 minimum number of bytes needed to hold all bits of the type. For
4843 example, storing an ``i24`` writes at most three bytes. When writing a
4844 value of a type like ``i20`` with a size that is not an integral number
4845 of bytes, it is unspecified what happens to the extra bits that do not
4846 belong to the type, but they will typically be overwritten.
4851 .. code-block:: llvm
4853 %ptr = alloca i32 ; yields {i32*}:ptr
4854 store i32 3, i32* %ptr ; yields {void}
4855 %val = load i32* %ptr ; yields {i32}:val = i32 3
4859 '``fence``' Instruction
4860 ^^^^^^^^^^^^^^^^^^^^^^^
4867 fence [singlethread] <ordering> ; yields {void}
4872 The '``fence``' instruction is used to introduce happens-before edges
4878 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4879 defines what *synchronizes-with* edges they add. They can only be given
4880 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4885 A fence A which has (at least) ``release`` ordering semantics
4886 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4887 semantics if and only if there exist atomic operations X and Y, both
4888 operating on some atomic object M, such that A is sequenced before X, X
4889 modifies M (either directly or through some side effect of a sequence
4890 headed by X), Y is sequenced before B, and Y observes M. This provides a
4891 *happens-before* dependency between A and B. Rather than an explicit
4892 ``fence``, one (but not both) of the atomic operations X or Y might
4893 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4894 still *synchronize-with* the explicit ``fence`` and establish the
4895 *happens-before* edge.
4897 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4898 ``acquire`` and ``release`` semantics specified above, participates in
4899 the global program order of other ``seq_cst`` operations and/or fences.
4901 The optional ":ref:`singlethread <singlethread>`" argument specifies
4902 that the fence only synchronizes with other fences in the same thread.
4903 (This is useful for interacting with signal handlers.)
4908 .. code-block:: llvm
4910 fence acquire ; yields {void}
4911 fence singlethread seq_cst ; yields {void}
4915 '``cmpxchg``' Instruction
4916 ^^^^^^^^^^^^^^^^^^^^^^^^^
4923 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4928 The '``cmpxchg``' instruction is used to atomically modify memory. It
4929 loads a value in memory and compares it to a given value. If they are
4930 equal, it stores a new value into the memory.
4935 There are three arguments to the '``cmpxchg``' instruction: an address
4936 to operate on, a value to compare to the value currently be at that
4937 address, and a new value to place at that address if the compared values
4938 are equal. The type of '<cmp>' must be an integer type whose bit width
4939 is a power of two greater than or equal to eight and less than or equal
4940 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4941 type, and the type of '<pointer>' must be a pointer to that type. If the
4942 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4943 to modify the number or order of execution of this ``cmpxchg`` with
4944 other :ref:`volatile operations <volatile>`.
4946 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4947 synchronizes with other atomic operations.
4949 The optional "``singlethread``" argument declares that the ``cmpxchg``
4950 is only atomic with respect to code (usually signal handlers) running in
4951 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4952 respect to all other code in the system.
4954 The pointer passed into cmpxchg must have alignment greater than or
4955 equal to the size in memory of the operand.
4960 The contents of memory at the location specified by the '``<pointer>``'
4961 operand is read and compared to '``<cmp>``'; if the read value is the
4962 equal, '``<new>``' is written. The original value at the location is
4965 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4966 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4967 atomic load with an ordering parameter determined by dropping any
4968 ``release`` part of the ``cmpxchg``'s ordering.
4973 .. code-block:: llvm
4976 %orig = atomic load i32* %ptr unordered ; yields {i32}
4980 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4981 %squared = mul i32 %cmp, %cmp
4982 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4983 %success = icmp eq i32 %cmp, %old
4984 br i1 %success, label %done, label %loop
4991 '``atomicrmw``' Instruction
4992 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4999 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
5004 The '``atomicrmw``' instruction is used to atomically modify memory.
5009 There are three arguments to the '``atomicrmw``' instruction: an
5010 operation to apply, an address whose value to modify, an argument to the
5011 operation. The operation must be one of the following keywords:
5025 The type of '<value>' must be an integer type whose bit width is a power
5026 of two greater than or equal to eight and less than or equal to a
5027 target-specific size limit. The type of the '``<pointer>``' operand must
5028 be a pointer to that type. If the ``atomicrmw`` is marked as
5029 ``volatile``, then the optimizer is not allowed to modify the number or
5030 order of execution of this ``atomicrmw`` with other :ref:`volatile
5031 operations <volatile>`.
5036 The contents of memory at the location specified by the '``<pointer>``'
5037 operand are atomically read, modified, and written back. The original
5038 value at the location is returned. The modification is specified by the
5041 - xchg: ``*ptr = val``
5042 - add: ``*ptr = *ptr + val``
5043 - sub: ``*ptr = *ptr - val``
5044 - and: ``*ptr = *ptr & val``
5045 - nand: ``*ptr = ~(*ptr & val)``
5046 - or: ``*ptr = *ptr | val``
5047 - xor: ``*ptr = *ptr ^ val``
5048 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5049 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5050 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5052 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5058 .. code-block:: llvm
5060 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5062 .. _i_getelementptr:
5064 '``getelementptr``' Instruction
5065 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5072 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5073 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5074 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5079 The '``getelementptr``' instruction is used to get the address of a
5080 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5081 address calculation only and does not access memory.
5086 The first argument is always a pointer or a vector of pointers, and
5087 forms the basis of the calculation. The remaining arguments are indices
5088 that indicate which of the elements of the aggregate object are indexed.
5089 The interpretation of each index is dependent on the type being indexed
5090 into. The first index always indexes the pointer value given as the
5091 first argument, the second index indexes a value of the type pointed to
5092 (not necessarily the value directly pointed to, since the first index
5093 can be non-zero), etc. The first type indexed into must be a pointer
5094 value, subsequent types can be arrays, vectors, and structs. Note that
5095 subsequent types being indexed into can never be pointers, since that
5096 would require loading the pointer before continuing calculation.
5098 The type of each index argument depends on the type it is indexing into.
5099 When indexing into a (optionally packed) structure, only ``i32`` integer
5100 **constants** are allowed (when using a vector of indices they must all
5101 be the **same** ``i32`` integer constant). When indexing into an array,
5102 pointer or vector, integers of any width are allowed, and they are not
5103 required to be constant. These integers are treated as signed values
5106 For example, let's consider a C code fragment and how it gets compiled
5122 int *foo(struct ST *s) {
5123 return &s[1].Z.B[5][13];
5126 The LLVM code generated by Clang is:
5128 .. code-block:: llvm
5130 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5131 %struct.ST = type { i32, double, %struct.RT }
5133 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5135 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5142 In the example above, the first index is indexing into the
5143 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5144 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5145 indexes into the third element of the structure, yielding a
5146 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5147 structure. The third index indexes into the second element of the
5148 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5149 dimensions of the array are subscripted into, yielding an '``i32``'
5150 type. The '``getelementptr``' instruction returns a pointer to this
5151 element, thus computing a value of '``i32*``' type.
5153 Note that it is perfectly legal to index partially through a structure,
5154 returning a pointer to an inner element. Because of this, the LLVM code
5155 for the given testcase is equivalent to:
5157 .. code-block:: llvm
5159 define i32* @foo(%struct.ST* %s) {
5160 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5161 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5162 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5163 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5164 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5168 If the ``inbounds`` keyword is present, the result value of the
5169 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5170 pointer is not an *in bounds* address of an allocated object, or if any
5171 of the addresses that would be formed by successive addition of the
5172 offsets implied by the indices to the base address with infinitely
5173 precise signed arithmetic are not an *in bounds* address of that
5174 allocated object. The *in bounds* addresses for an allocated object are
5175 all the addresses that point into the object, plus the address one byte
5176 past the end. In cases where the base is a vector of pointers the
5177 ``inbounds`` keyword applies to each of the computations element-wise.
5179 If the ``inbounds`` keyword is not present, the offsets are added to the
5180 base address with silently-wrapping two's complement arithmetic. If the
5181 offsets have a different width from the pointer, they are sign-extended
5182 or truncated to the width of the pointer. The result value of the
5183 ``getelementptr`` may be outside the object pointed to by the base
5184 pointer. The result value may not necessarily be used to access memory
5185 though, even if it happens to point into allocated storage. See the
5186 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5189 The getelementptr instruction is often confusing. For some more insight
5190 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5195 .. code-block:: llvm
5197 ; yields [12 x i8]*:aptr
5198 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5200 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5202 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5204 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5206 In cases where the pointer argument is a vector of pointers, each index
5207 must be a vector with the same number of elements. For example:
5209 .. code-block:: llvm
5211 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5213 Conversion Operations
5214 ---------------------
5216 The instructions in this category are the conversion instructions
5217 (casting) which all take a single operand and a type. They perform
5218 various bit conversions on the operand.
5220 '``trunc .. to``' Instruction
5221 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5228 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5233 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5238 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5239 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5240 of the same number of integers. The bit size of the ``value`` must be
5241 larger than the bit size of the destination type, ``ty2``. Equal sized
5242 types are not allowed.
5247 The '``trunc``' instruction truncates the high order bits in ``value``
5248 and converts the remaining bits to ``ty2``. Since the source size must
5249 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5250 It will always truncate bits.
5255 .. code-block:: llvm
5257 %X = trunc i32 257 to i8 ; yields i8:1
5258 %Y = trunc i32 123 to i1 ; yields i1:true
5259 %Z = trunc i32 122 to i1 ; yields i1:false
5260 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5262 '``zext .. to``' Instruction
5263 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5270 <result> = zext <ty> <value> to <ty2> ; yields ty2
5275 The '``zext``' instruction zero extends its operand to type ``ty2``.
5280 The '``zext``' instruction takes a value to cast, and a type to cast it
5281 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5282 the same number of integers. The bit size of the ``value`` must be
5283 smaller than the bit size of the destination type, ``ty2``.
5288 The ``zext`` fills the high order bits of the ``value`` with zero bits
5289 until it reaches the size of the destination type, ``ty2``.
5291 When zero extending from i1, the result will always be either 0 or 1.
5296 .. code-block:: llvm
5298 %X = zext i32 257 to i64 ; yields i64:257
5299 %Y = zext i1 true to i32 ; yields i32:1
5300 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5302 '``sext .. to``' Instruction
5303 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5310 <result> = sext <ty> <value> to <ty2> ; yields ty2
5315 The '``sext``' sign extends ``value`` to the type ``ty2``.
5320 The '``sext``' instruction takes a value to cast, and a type to cast it
5321 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5322 the same number of integers. The bit size of the ``value`` must be
5323 smaller than the bit size of the destination type, ``ty2``.
5328 The '``sext``' instruction performs a sign extension by copying the sign
5329 bit (highest order bit) of the ``value`` until it reaches the bit size
5330 of the type ``ty2``.
5332 When sign extending from i1, the extension always results in -1 or 0.
5337 .. code-block:: llvm
5339 %X = sext i8 -1 to i16 ; yields i16 :65535
5340 %Y = sext i1 true to i32 ; yields i32:-1
5341 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5343 '``fptrunc .. to``' Instruction
5344 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5351 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5356 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5361 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5362 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5363 The size of ``value`` must be larger than the size of ``ty2``. This
5364 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5369 The '``fptrunc``' instruction truncates a ``value`` from a larger
5370 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5371 point <t_floating>` type. If the value cannot fit within the
5372 destination type, ``ty2``, then the results are undefined.
5377 .. code-block:: llvm
5379 %X = fptrunc double 123.0 to float ; yields float:123.0
5380 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5382 '``fpext .. to``' Instruction
5383 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5390 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5395 The '``fpext``' extends a floating point ``value`` to a larger floating
5401 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5402 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5403 to. The source type must be smaller than the destination type.
5408 The '``fpext``' instruction extends the ``value`` from a smaller
5409 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5410 point <t_floating>` type. The ``fpext`` cannot be used to make a
5411 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5412 *no-op cast* for a floating point cast.
5417 .. code-block:: llvm
5419 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5420 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5422 '``fptoui .. to``' Instruction
5423 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5430 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5435 The '``fptoui``' converts a floating point ``value`` to its unsigned
5436 integer equivalent of type ``ty2``.
5441 The '``fptoui``' instruction takes a value to cast, which must be a
5442 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5443 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5444 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5445 type with the same number of elements as ``ty``
5450 The '``fptoui``' instruction converts its :ref:`floating
5451 point <t_floating>` operand into the nearest (rounding towards zero)
5452 unsigned integer value. If the value cannot fit in ``ty2``, the results
5458 .. code-block:: llvm
5460 %X = fptoui double 123.0 to i32 ; yields i32:123
5461 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5462 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5464 '``fptosi .. to``' Instruction
5465 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5472 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5477 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5478 ``value`` to type ``ty2``.
5483 The '``fptosi``' instruction takes a value to cast, which must be a
5484 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5485 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5486 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5487 type with the same number of elements as ``ty``
5492 The '``fptosi``' instruction converts its :ref:`floating
5493 point <t_floating>` operand into the nearest (rounding towards zero)
5494 signed integer value. If the value cannot fit in ``ty2``, the results
5500 .. code-block:: llvm
5502 %X = fptosi double -123.0 to i32 ; yields i32:-123
5503 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5504 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5506 '``uitofp .. to``' Instruction
5507 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5514 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5519 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5520 and converts that value to the ``ty2`` type.
5525 The '``uitofp``' instruction takes a value to cast, which must be a
5526 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5527 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5528 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5529 type with the same number of elements as ``ty``
5534 The '``uitofp``' instruction interprets its operand as an unsigned
5535 integer quantity and converts it to the corresponding floating point
5536 value. If the value cannot fit in the floating point value, the results
5542 .. code-block:: llvm
5544 %X = uitofp i32 257 to float ; yields float:257.0
5545 %Y = uitofp i8 -1 to double ; yields double:255.0
5547 '``sitofp .. to``' Instruction
5548 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5555 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5560 The '``sitofp``' instruction regards ``value`` as a signed integer and
5561 converts that value to the ``ty2`` type.
5566 The '``sitofp``' instruction takes a value to cast, which must be a
5567 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5568 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5569 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5570 type with the same number of elements as ``ty``
5575 The '``sitofp``' instruction interprets its operand as a signed integer
5576 quantity and converts it to the corresponding floating point value. If
5577 the value cannot fit in the floating point value, the results are
5583 .. code-block:: llvm
5585 %X = sitofp i32 257 to float ; yields float:257.0
5586 %Y = sitofp i8 -1 to double ; yields double:-1.0
5590 '``ptrtoint .. to``' Instruction
5591 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5598 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5603 The '``ptrtoint``' instruction converts the pointer or a vector of
5604 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5609 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5610 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5611 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5612 a vector of integers type.
5617 The '``ptrtoint``' instruction converts ``value`` to integer type
5618 ``ty2`` by interpreting the pointer value as an integer and either
5619 truncating or zero extending that value to the size of the integer type.
5620 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5621 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5622 the same size, then nothing is done (*no-op cast*) other than a type
5628 .. code-block:: llvm
5630 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5631 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5632 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5636 '``inttoptr .. to``' Instruction
5637 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5644 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5649 The '``inttoptr``' instruction converts an integer ``value`` to a
5650 pointer type, ``ty2``.
5655 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5656 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5662 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5663 applying either a zero extension or a truncation depending on the size
5664 of the integer ``value``. If ``value`` is larger than the size of a
5665 pointer then a truncation is done. If ``value`` is smaller than the size
5666 of a pointer then a zero extension is done. If they are the same size,
5667 nothing is done (*no-op cast*).
5672 .. code-block:: llvm
5674 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5675 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5676 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5677 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5681 '``bitcast .. to``' Instruction
5682 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5689 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5694 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5700 The '``bitcast``' instruction takes a value to cast, which must be a
5701 non-aggregate first class value, and a type to cast it to, which must
5702 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5703 bit sizes of ``value`` and the destination type, ``ty2``, must be
5704 identical. If the source type is a pointer, the destination type must
5705 also be a pointer of the same size. This instruction supports bitwise
5706 conversion of vectors to integers and to vectors of other types (as
5707 long as they have the same size).
5712 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5713 is always a *no-op cast* because no bits change with this
5714 conversion. The conversion is done as if the ``value`` had been stored
5715 to memory and read back as type ``ty2``. Pointer (or vector of
5716 pointers) types may only be converted to other pointer (or vector of
5717 pointers) types with the same address space through this instruction.
5718 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5719 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5724 .. code-block:: llvm
5726 %X = bitcast i8 255 to i8 ; yields i8 :-1
5727 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5728 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5729 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5731 .. _i_addrspacecast:
5733 '``addrspacecast .. to``' Instruction
5734 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5741 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5746 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5747 address space ``n`` to type ``pty2`` in address space ``m``.
5752 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5753 to cast and a pointer type to cast it to, which must have a different
5759 The '``addrspacecast``' instruction converts the pointer value
5760 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5761 value modification, depending on the target and the address space
5762 pair. Pointer conversions within the same address space must be
5763 performed with the ``bitcast`` instruction. Note that if the address space
5764 conversion is legal then both result and operand refer to the same memory
5770 .. code-block:: llvm
5772 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5773 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5774 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5781 The instructions in this category are the "miscellaneous" instructions,
5782 which defy better classification.
5786 '``icmp``' Instruction
5787 ^^^^^^^^^^^^^^^^^^^^^^
5794 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5799 The '``icmp``' instruction returns a boolean value or a vector of
5800 boolean values based on comparison of its two integer, integer vector,
5801 pointer, or pointer vector operands.
5806 The '``icmp``' instruction takes three operands. The first operand is
5807 the condition code indicating the kind of comparison to perform. It is
5808 not a value, just a keyword. The possible condition code are:
5811 #. ``ne``: not equal
5812 #. ``ugt``: unsigned greater than
5813 #. ``uge``: unsigned greater or equal
5814 #. ``ult``: unsigned less than
5815 #. ``ule``: unsigned less or equal
5816 #. ``sgt``: signed greater than
5817 #. ``sge``: signed greater or equal
5818 #. ``slt``: signed less than
5819 #. ``sle``: signed less or equal
5821 The remaining two arguments must be :ref:`integer <t_integer>` or
5822 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5823 must also be identical types.
5828 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5829 code given as ``cond``. The comparison performed always yields either an
5830 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5832 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5833 otherwise. No sign interpretation is necessary or performed.
5834 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5835 otherwise. No sign interpretation is necessary or performed.
5836 #. ``ugt``: interprets the operands as unsigned values and yields
5837 ``true`` if ``op1`` is greater than ``op2``.
5838 #. ``uge``: interprets the operands as unsigned values and yields
5839 ``true`` if ``op1`` is greater than or equal to ``op2``.
5840 #. ``ult``: interprets the operands as unsigned values and yields
5841 ``true`` if ``op1`` is less than ``op2``.
5842 #. ``ule``: interprets the operands as unsigned values and yields
5843 ``true`` if ``op1`` is less than or equal to ``op2``.
5844 #. ``sgt``: interprets the operands as signed values and yields ``true``
5845 if ``op1`` is greater than ``op2``.
5846 #. ``sge``: interprets the operands as signed values and yields ``true``
5847 if ``op1`` is greater than or equal to ``op2``.
5848 #. ``slt``: interprets the operands as signed values and yields ``true``
5849 if ``op1`` is less than ``op2``.
5850 #. ``sle``: interprets the operands as signed values and yields ``true``
5851 if ``op1`` is less than or equal to ``op2``.
5853 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5854 are compared as if they were integers.
5856 If the operands are integer vectors, then they are compared element by
5857 element. The result is an ``i1`` vector with the same number of elements
5858 as the values being compared. Otherwise, the result is an ``i1``.
5863 .. code-block:: llvm
5865 <result> = icmp eq i32 4, 5 ; yields: result=false
5866 <result> = icmp ne float* %X, %X ; yields: result=false
5867 <result> = icmp ult i16 4, 5 ; yields: result=true
5868 <result> = icmp sgt i16 4, 5 ; yields: result=false
5869 <result> = icmp ule i16 -4, 5 ; yields: result=false
5870 <result> = icmp sge i16 4, 5 ; yields: result=false
5872 Note that the code generator does not yet support vector types with the
5873 ``icmp`` instruction.
5877 '``fcmp``' Instruction
5878 ^^^^^^^^^^^^^^^^^^^^^^
5885 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5890 The '``fcmp``' instruction returns a boolean value or vector of boolean
5891 values based on comparison of its operands.
5893 If the operands are floating point scalars, then the result type is a
5894 boolean (:ref:`i1 <t_integer>`).
5896 If the operands are floating point vectors, then the result type is a
5897 vector of boolean with the same number of elements as the operands being
5903 The '``fcmp``' instruction takes three operands. The first operand is
5904 the condition code indicating the kind of comparison to perform. It is
5905 not a value, just a keyword. The possible condition code are:
5907 #. ``false``: no comparison, always returns false
5908 #. ``oeq``: ordered and equal
5909 #. ``ogt``: ordered and greater than
5910 #. ``oge``: ordered and greater than or equal
5911 #. ``olt``: ordered and less than
5912 #. ``ole``: ordered and less than or equal
5913 #. ``one``: ordered and not equal
5914 #. ``ord``: ordered (no nans)
5915 #. ``ueq``: unordered or equal
5916 #. ``ugt``: unordered or greater than
5917 #. ``uge``: unordered or greater than or equal
5918 #. ``ult``: unordered or less than
5919 #. ``ule``: unordered or less than or equal
5920 #. ``une``: unordered or not equal
5921 #. ``uno``: unordered (either nans)
5922 #. ``true``: no comparison, always returns true
5924 *Ordered* means that neither operand is a QNAN while *unordered* means
5925 that either operand may be a QNAN.
5927 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5928 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5929 type. They must have identical types.
5934 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5935 condition code given as ``cond``. If the operands are vectors, then the
5936 vectors are compared element by element. Each comparison performed
5937 always yields an :ref:`i1 <t_integer>` result, as follows:
5939 #. ``false``: always yields ``false``, regardless of operands.
5940 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5941 is equal to ``op2``.
5942 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5943 is greater than ``op2``.
5944 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5945 is greater than or equal to ``op2``.
5946 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5947 is less than ``op2``.
5948 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5949 is less than or equal to ``op2``.
5950 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5951 is not equal to ``op2``.
5952 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5953 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5955 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5956 greater than ``op2``.
5957 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5958 greater than or equal to ``op2``.
5959 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5961 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5962 less than or equal to ``op2``.
5963 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5964 not equal to ``op2``.
5965 #. ``uno``: yields ``true`` if either operand is a QNAN.
5966 #. ``true``: always yields ``true``, regardless of operands.
5971 .. code-block:: llvm
5973 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5974 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5975 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5976 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5978 Note that the code generator does not yet support vector types with the
5979 ``fcmp`` instruction.
5983 '``phi``' Instruction
5984 ^^^^^^^^^^^^^^^^^^^^^
5991 <result> = phi <ty> [ <val0>, <label0>], ...
5996 The '``phi``' instruction is used to implement the φ node in the SSA
5997 graph representing the function.
6002 The type of the incoming values is specified with the first type field.
6003 After this, the '``phi``' instruction takes a list of pairs as
6004 arguments, with one pair for each predecessor basic block of the current
6005 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6006 the value arguments to the PHI node. Only labels may be used as the
6009 There must be no non-phi instructions between the start of a basic block
6010 and the PHI instructions: i.e. PHI instructions must be first in a basic
6013 For the purposes of the SSA form, the use of each incoming value is
6014 deemed to occur on the edge from the corresponding predecessor block to
6015 the current block (but after any definition of an '``invoke``'
6016 instruction's return value on the same edge).
6021 At runtime, the '``phi``' instruction logically takes on the value
6022 specified by the pair corresponding to the predecessor basic block that
6023 executed just prior to the current block.
6028 .. code-block:: llvm
6030 Loop: ; Infinite loop that counts from 0 on up...
6031 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6032 %nextindvar = add i32 %indvar, 1
6037 '``select``' Instruction
6038 ^^^^^^^^^^^^^^^^^^^^^^^^
6045 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6047 selty is either i1 or {<N x i1>}
6052 The '``select``' instruction is used to choose one value based on a
6053 condition, without branching.
6058 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6059 values indicating the condition, and two values of the same :ref:`first
6060 class <t_firstclass>` type. If the val1/val2 are vectors and the
6061 condition is a scalar, then entire vectors are selected, not individual
6067 If the condition is an i1 and it evaluates to 1, the instruction returns
6068 the first value argument; otherwise, it returns the second value
6071 If the condition is a vector of i1, then the value arguments must be
6072 vectors of the same size, and the selection is done element by element.
6077 .. code-block:: llvm
6079 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6083 '``call``' Instruction
6084 ^^^^^^^^^^^^^^^^^^^^^^
6091 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6096 The '``call``' instruction represents a simple function call.
6101 This instruction requires several arguments:
6103 #. The optional "tail" marker indicates that the callee function does
6104 not access any allocas or varargs in the caller. Note that calls may
6105 be marked "tail" even if they do not occur before a
6106 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6107 function call is eligible for tail call optimization, but `might not
6108 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6109 The code generator may optimize calls marked "tail" with either 1)
6110 automatic `sibling call
6111 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6112 callee have matching signatures, or 2) forced tail call optimization
6113 when the following extra requirements are met:
6115 - Caller and callee both have the calling convention ``fastcc``.
6116 - The call is in tail position (ret immediately follows call and ret
6117 uses value of call or is void).
6118 - Option ``-tailcallopt`` is enabled, or
6119 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6120 - `Platform specific constraints are
6121 met. <CodeGenerator.html#tailcallopt>`_
6123 #. The optional "cconv" marker indicates which :ref:`calling
6124 convention <callingconv>` the call should use. If none is
6125 specified, the call defaults to using C calling conventions. The
6126 calling convention of the call must match the calling convention of
6127 the target function, or else the behavior is undefined.
6128 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6129 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6131 #. '``ty``': the type of the call instruction itself which is also the
6132 type of the return value. Functions that return no value are marked
6134 #. '``fnty``': shall be the signature of the pointer to function value
6135 being invoked. The argument types must match the types implied by
6136 this signature. This type can be omitted if the function is not
6137 varargs and if the function type does not return a pointer to a
6139 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6140 be invoked. In most cases, this is a direct function invocation, but
6141 indirect ``call``'s are just as possible, calling an arbitrary pointer
6143 #. '``function args``': argument list whose types match the function
6144 signature argument types and parameter attributes. All arguments must
6145 be of :ref:`first class <t_firstclass>` type. If the function signature
6146 indicates the function accepts a variable number of arguments, the
6147 extra arguments can be specified.
6148 #. The optional :ref:`function attributes <fnattrs>` list. Only
6149 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6150 attributes are valid here.
6155 The '``call``' instruction is used to cause control flow to transfer to
6156 a specified function, with its incoming arguments bound to the specified
6157 values. Upon a '``ret``' instruction in the called function, control
6158 flow continues with the instruction after the function call, and the
6159 return value of the function is bound to the result argument.
6164 .. code-block:: llvm
6166 %retval = call i32 @test(i32 %argc)
6167 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6168 %X = tail call i32 @foo() ; yields i32
6169 %Y = tail call fastcc i32 @foo() ; yields i32
6170 call void %foo(i8 97 signext)
6172 %struct.A = type { i32, i8 }
6173 %r = call %struct.A @foo() ; yields { 32, i8 }
6174 %gr = extractvalue %struct.A %r, 0 ; yields i32
6175 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6176 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6177 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6179 llvm treats calls to some functions with names and arguments that match
6180 the standard C99 library as being the C99 library functions, and may
6181 perform optimizations or generate code for them under that assumption.
6182 This is something we'd like to change in the future to provide better
6183 support for freestanding environments and non-C-based languages.
6187 '``va_arg``' Instruction
6188 ^^^^^^^^^^^^^^^^^^^^^^^^
6195 <resultval> = va_arg <va_list*> <arglist>, <argty>
6200 The '``va_arg``' instruction is used to access arguments passed through
6201 the "variable argument" area of a function call. It is used to implement
6202 the ``va_arg`` macro in C.
6207 This instruction takes a ``va_list*`` value and the type of the
6208 argument. It returns a value of the specified argument type and
6209 increments the ``va_list`` to point to the next argument. The actual
6210 type of ``va_list`` is target specific.
6215 The '``va_arg``' instruction loads an argument of the specified type
6216 from the specified ``va_list`` and causes the ``va_list`` to point to
6217 the next argument. For more information, see the variable argument
6218 handling :ref:`Intrinsic Functions <int_varargs>`.
6220 It is legal for this instruction to be called in a function which does
6221 not take a variable number of arguments, for example, the ``vfprintf``
6224 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6225 function <intrinsics>` because it takes a type as an argument.
6230 See the :ref:`variable argument processing <int_varargs>` section.
6232 Note that the code generator does not yet fully support va\_arg on many
6233 targets. Also, it does not currently support va\_arg with aggregate
6234 types on any target.
6238 '``landingpad``' Instruction
6239 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6246 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6247 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6249 <clause> := catch <type> <value>
6250 <clause> := filter <array constant type> <array constant>
6255 The '``landingpad``' instruction is used by `LLVM's exception handling
6256 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6257 is a landing pad --- one where the exception lands, and corresponds to the
6258 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6259 defines values supplied by the personality function (``pers_fn``) upon
6260 re-entry to the function. The ``resultval`` has the type ``resultty``.
6265 This instruction takes a ``pers_fn`` value. This is the personality
6266 function associated with the unwinding mechanism. The optional
6267 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6269 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6270 contains the global variable representing the "type" that may be caught
6271 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6272 clause takes an array constant as its argument. Use
6273 "``[0 x i8**] undef``" for a filter which cannot throw. The
6274 '``landingpad``' instruction must contain *at least* one ``clause`` or
6275 the ``cleanup`` flag.
6280 The '``landingpad``' instruction defines the values which are set by the
6281 personality function (``pers_fn``) upon re-entry to the function, and
6282 therefore the "result type" of the ``landingpad`` instruction. As with
6283 calling conventions, how the personality function results are
6284 represented in LLVM IR is target specific.
6286 The clauses are applied in order from top to bottom. If two
6287 ``landingpad`` instructions are merged together through inlining, the
6288 clauses from the calling function are appended to the list of clauses.
6289 When the call stack is being unwound due to an exception being thrown,
6290 the exception is compared against each ``clause`` in turn. If it doesn't
6291 match any of the clauses, and the ``cleanup`` flag is not set, then
6292 unwinding continues further up the call stack.
6294 The ``landingpad`` instruction has several restrictions:
6296 - A landing pad block is a basic block which is the unwind destination
6297 of an '``invoke``' instruction.
6298 - A landing pad block must have a '``landingpad``' instruction as its
6299 first non-PHI instruction.
6300 - There can be only one '``landingpad``' instruction within the landing
6302 - A basic block that is not a landing pad block may not include a
6303 '``landingpad``' instruction.
6304 - All '``landingpad``' instructions in a function must have the same
6305 personality function.
6310 .. code-block:: llvm
6312 ;; A landing pad which can catch an integer.
6313 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6315 ;; A landing pad that is a cleanup.
6316 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6318 ;; A landing pad which can catch an integer and can only throw a double.
6319 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6321 filter [1 x i8**] [@_ZTId]
6328 LLVM supports the notion of an "intrinsic function". These functions
6329 have well known names and semantics and are required to follow certain
6330 restrictions. Overall, these intrinsics represent an extension mechanism
6331 for the LLVM language that does not require changing all of the
6332 transformations in LLVM when adding to the language (or the bitcode
6333 reader/writer, the parser, etc...).
6335 Intrinsic function names must all start with an "``llvm.``" prefix. This
6336 prefix is reserved in LLVM for intrinsic names; thus, function names may
6337 not begin with this prefix. Intrinsic functions must always be external
6338 functions: you cannot define the body of intrinsic functions. Intrinsic
6339 functions may only be used in call or invoke instructions: it is illegal
6340 to take the address of an intrinsic function. Additionally, because
6341 intrinsic functions are part of the LLVM language, it is required if any
6342 are added that they be documented here.
6344 Some intrinsic functions can be overloaded, i.e., the intrinsic
6345 represents a family of functions that perform the same operation but on
6346 different data types. Because LLVM can represent over 8 million
6347 different integer types, overloading is used commonly to allow an
6348 intrinsic function to operate on any integer type. One or more of the
6349 argument types or the result type can be overloaded to accept any
6350 integer type. Argument types may also be defined as exactly matching a
6351 previous argument's type or the result type. This allows an intrinsic
6352 function which accepts multiple arguments, but needs all of them to be
6353 of the same type, to only be overloaded with respect to a single
6354 argument or the result.
6356 Overloaded intrinsics will have the names of its overloaded argument
6357 types encoded into its function name, each preceded by a period. Only
6358 those types which are overloaded result in a name suffix. Arguments
6359 whose type is matched against another type do not. For example, the
6360 ``llvm.ctpop`` function can take an integer of any width and returns an
6361 integer of exactly the same integer width. This leads to a family of
6362 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6363 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6364 overloaded, and only one type suffix is required. Because the argument's
6365 type is matched against the return type, it does not require its own
6368 To learn how to add an intrinsic function, please see the `Extending
6369 LLVM Guide <ExtendingLLVM.html>`_.
6373 Variable Argument Handling Intrinsics
6374 -------------------------------------
6376 Variable argument support is defined in LLVM with the
6377 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6378 functions. These functions are related to the similarly named macros
6379 defined in the ``<stdarg.h>`` header file.
6381 All of these functions operate on arguments that use a target-specific
6382 value type "``va_list``". The LLVM assembly language reference manual
6383 does not define what this type is, so all transformations should be
6384 prepared to handle these functions regardless of the type used.
6386 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6387 variable argument handling intrinsic functions are used.
6389 .. code-block:: llvm
6391 define i32 @test(i32 %X, ...) {
6392 ; Initialize variable argument processing
6394 %ap2 = bitcast i8** %ap to i8*
6395 call void @llvm.va_start(i8* %ap2)
6397 ; Read a single integer argument
6398 %tmp = va_arg i8** %ap, i32
6400 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6402 %aq2 = bitcast i8** %aq to i8*
6403 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6404 call void @llvm.va_end(i8* %aq2)
6406 ; Stop processing of arguments.
6407 call void @llvm.va_end(i8* %ap2)
6411 declare void @llvm.va_start(i8*)
6412 declare void @llvm.va_copy(i8*, i8*)
6413 declare void @llvm.va_end(i8*)
6417 '``llvm.va_start``' Intrinsic
6418 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6425 declare void @llvm.va_start(i8* <arglist>)
6430 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6431 subsequent use by ``va_arg``.
6436 The argument is a pointer to a ``va_list`` element to initialize.
6441 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6442 available in C. In a target-dependent way, it initializes the
6443 ``va_list`` element to which the argument points, so that the next call
6444 to ``va_arg`` will produce the first variable argument passed to the
6445 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6446 to know the last argument of the function as the compiler can figure
6449 '``llvm.va_end``' Intrinsic
6450 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6457 declare void @llvm.va_end(i8* <arglist>)
6462 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6463 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6468 The argument is a pointer to a ``va_list`` to destroy.
6473 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6474 available in C. In a target-dependent way, it destroys the ``va_list``
6475 element to which the argument points. Calls to
6476 :ref:`llvm.va_start <int_va_start>` and
6477 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6482 '``llvm.va_copy``' Intrinsic
6483 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6490 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6495 The '``llvm.va_copy``' intrinsic copies the current argument position
6496 from the source argument list to the destination argument list.
6501 The first argument is a pointer to a ``va_list`` element to initialize.
6502 The second argument is a pointer to a ``va_list`` element to copy from.
6507 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6508 available in C. In a target-dependent way, it copies the source
6509 ``va_list`` element into the destination ``va_list`` element. This
6510 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6511 arbitrarily complex and require, for example, memory allocation.
6513 Accurate Garbage Collection Intrinsics
6514 --------------------------------------
6516 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6517 (GC) requires the implementation and generation of these intrinsics.
6518 These intrinsics allow identification of :ref:`GC roots on the
6519 stack <int_gcroot>`, as well as garbage collector implementations that
6520 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6521 Front-ends for type-safe garbage collected languages should generate
6522 these intrinsics to make use of the LLVM garbage collectors. For more
6523 details, see `Accurate Garbage Collection with
6524 LLVM <GarbageCollection.html>`_.
6526 The garbage collection intrinsics only operate on objects in the generic
6527 address space (address space zero).
6531 '``llvm.gcroot``' Intrinsic
6532 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6539 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6544 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6545 the code generator, and allows some metadata to be associated with it.
6550 The first argument specifies the address of a stack object that contains
6551 the root pointer. The second pointer (which must be either a constant or
6552 a global value address) contains the meta-data to be associated with the
6558 At runtime, a call to this intrinsic stores a null pointer into the
6559 "ptrloc" location. At compile-time, the code generator generates
6560 information to allow the runtime to find the pointer at GC safe points.
6561 The '``llvm.gcroot``' intrinsic may only be used in a function which
6562 :ref:`specifies a GC algorithm <gc>`.
6566 '``llvm.gcread``' Intrinsic
6567 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6574 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6579 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6580 locations, allowing garbage collector implementations that require read
6586 The second argument is the address to read from, which should be an
6587 address allocated from the garbage collector. The first object is a
6588 pointer to the start of the referenced object, if needed by the language
6589 runtime (otherwise null).
6594 The '``llvm.gcread``' intrinsic has the same semantics as a load
6595 instruction, but may be replaced with substantially more complex code by
6596 the garbage collector runtime, as needed. The '``llvm.gcread``'
6597 intrinsic may only be used in a function which :ref:`specifies a GC
6602 '``llvm.gcwrite``' Intrinsic
6603 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6610 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6615 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6616 locations, allowing garbage collector implementations that require write
6617 barriers (such as generational or reference counting collectors).
6622 The first argument is the reference to store, the second is the start of
6623 the object to store it to, and the third is the address of the field of
6624 Obj to store to. If the runtime does not require a pointer to the
6625 object, Obj may be null.
6630 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6631 instruction, but may be replaced with substantially more complex code by
6632 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6633 intrinsic may only be used in a function which :ref:`specifies a GC
6636 Code Generator Intrinsics
6637 -------------------------
6639 These intrinsics are provided by LLVM to expose special features that
6640 may only be implemented with code generator support.
6642 '``llvm.returnaddress``' Intrinsic
6643 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6650 declare i8 *@llvm.returnaddress(i32 <level>)
6655 The '``llvm.returnaddress``' intrinsic attempts to compute a
6656 target-specific value indicating the return address of the current
6657 function or one of its callers.
6662 The argument to this intrinsic indicates which function to return the
6663 address for. Zero indicates the calling function, one indicates its
6664 caller, etc. The argument is **required** to be a constant integer
6670 The '``llvm.returnaddress``' intrinsic either returns a pointer
6671 indicating the return address of the specified call frame, or zero if it
6672 cannot be identified. The value returned by this intrinsic is likely to
6673 be incorrect or 0 for arguments other than zero, so it should only be
6674 used for debugging purposes.
6676 Note that calling this intrinsic does not prevent function inlining or
6677 other aggressive transformations, so the value returned may not be that
6678 of the obvious source-language caller.
6680 '``llvm.frameaddress``' Intrinsic
6681 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6688 declare i8* @llvm.frameaddress(i32 <level>)
6693 The '``llvm.frameaddress``' intrinsic attempts to return the
6694 target-specific frame pointer value for the specified stack frame.
6699 The argument to this intrinsic indicates which function to return the
6700 frame pointer for. Zero indicates the calling function, one indicates
6701 its caller, etc. The argument is **required** to be a constant integer
6707 The '``llvm.frameaddress``' intrinsic either returns a pointer
6708 indicating the frame address of the specified call frame, or zero if it
6709 cannot be identified. The value returned by this intrinsic is likely to
6710 be incorrect or 0 for arguments other than zero, so it should only be
6711 used for debugging purposes.
6713 Note that calling this intrinsic does not prevent function inlining or
6714 other aggressive transformations, so the value returned may not be that
6715 of the obvious source-language caller.
6719 '``llvm.stacksave``' Intrinsic
6720 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6727 declare i8* @llvm.stacksave()
6732 The '``llvm.stacksave``' intrinsic is used to remember the current state
6733 of the function stack, for use with
6734 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6735 implementing language features like scoped automatic variable sized
6741 This intrinsic returns a opaque pointer value that can be passed to
6742 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6743 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6744 ``llvm.stacksave``, it effectively restores the state of the stack to
6745 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6746 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6747 were allocated after the ``llvm.stacksave`` was executed.
6749 .. _int_stackrestore:
6751 '``llvm.stackrestore``' Intrinsic
6752 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6759 declare void @llvm.stackrestore(i8* %ptr)
6764 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6765 the function stack to the state it was in when the corresponding
6766 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6767 useful for implementing language features like scoped automatic variable
6768 sized arrays in C99.
6773 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6775 '``llvm.prefetch``' Intrinsic
6776 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6783 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6788 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6789 insert a prefetch instruction if supported; otherwise, it is a noop.
6790 Prefetches have no effect on the behavior of the program but can change
6791 its performance characteristics.
6796 ``address`` is the address to be prefetched, ``rw`` is the specifier
6797 determining if the fetch should be for a read (0) or write (1), and
6798 ``locality`` is a temporal locality specifier ranging from (0) - no
6799 locality, to (3) - extremely local keep in cache. The ``cache type``
6800 specifies whether the prefetch is performed on the data (1) or
6801 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6802 arguments must be constant integers.
6807 This intrinsic does not modify the behavior of the program. In
6808 particular, prefetches cannot trap and do not produce a value. On
6809 targets that support this intrinsic, the prefetch can provide hints to
6810 the processor cache for better performance.
6812 '``llvm.pcmarker``' Intrinsic
6813 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6820 declare void @llvm.pcmarker(i32 <id>)
6825 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6826 Counter (PC) in a region of code to simulators and other tools. The
6827 method is target specific, but it is expected that the marker will use
6828 exported symbols to transmit the PC of the marker. The marker makes no
6829 guarantees that it will remain with any specific instruction after
6830 optimizations. It is possible that the presence of a marker will inhibit
6831 optimizations. The intended use is to be inserted after optimizations to
6832 allow correlations of simulation runs.
6837 ``id`` is a numerical id identifying the marker.
6842 This intrinsic does not modify the behavior of the program. Backends
6843 that do not support this intrinsic may ignore it.
6845 '``llvm.readcyclecounter``' Intrinsic
6846 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6853 declare i64 @llvm.readcyclecounter()
6858 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6859 counter register (or similar low latency, high accuracy clocks) on those
6860 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6861 should map to RPCC. As the backing counters overflow quickly (on the
6862 order of 9 seconds on alpha), this should only be used for small
6868 When directly supported, reading the cycle counter should not modify any
6869 memory. Implementations are allowed to either return a application
6870 specific value or a system wide value. On backends without support, this
6871 is lowered to a constant 0.
6873 Note that runtime support may be conditional on the privilege-level code is
6874 running at and the host platform.
6876 Standard C Library Intrinsics
6877 -----------------------------
6879 LLVM provides intrinsics for a few important standard C library
6880 functions. These intrinsics allow source-language front-ends to pass
6881 information about the alignment of the pointer arguments to the code
6882 generator, providing opportunity for more efficient code generation.
6886 '``llvm.memcpy``' Intrinsic
6887 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6892 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6893 integer bit width and for different address spaces. Not all targets
6894 support all bit widths however.
6898 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6899 i32 <len>, i32 <align>, i1 <isvolatile>)
6900 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6901 i64 <len>, i32 <align>, i1 <isvolatile>)
6906 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6907 source location to the destination location.
6909 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6910 intrinsics do not return a value, takes extra alignment/isvolatile
6911 arguments and the pointers can be in specified address spaces.
6916 The first argument is a pointer to the destination, the second is a
6917 pointer to the source. The third argument is an integer argument
6918 specifying the number of bytes to copy, the fourth argument is the
6919 alignment of the source and destination locations, and the fifth is a
6920 boolean indicating a volatile access.
6922 If the call to this intrinsic has an alignment value that is not 0 or 1,
6923 then the caller guarantees that both the source and destination pointers
6924 are aligned to that boundary.
6926 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6927 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6928 very cleanly specified and it is unwise to depend on it.
6933 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6934 source location to the destination location, which are not allowed to
6935 overlap. It copies "len" bytes of memory over. If the argument is known
6936 to be aligned to some boundary, this can be specified as the fourth
6937 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
6939 '``llvm.memmove``' Intrinsic
6940 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6945 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6946 bit width and for different address space. Not all targets support all
6951 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6952 i32 <len>, i32 <align>, i1 <isvolatile>)
6953 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6954 i64 <len>, i32 <align>, i1 <isvolatile>)
6959 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6960 source location to the destination location. It is similar to the
6961 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6964 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6965 intrinsics do not return a value, takes extra alignment/isvolatile
6966 arguments and the pointers can be in specified address spaces.
6971 The first argument is a pointer to the destination, the second is a
6972 pointer to the source. The third argument is an integer argument
6973 specifying the number of bytes to copy, the fourth argument is the
6974 alignment of the source and destination locations, and the fifth is a
6975 boolean indicating a volatile access.
6977 If the call to this intrinsic has an alignment value that is not 0 or 1,
6978 then the caller guarantees that the source and destination pointers are
6979 aligned to that boundary.
6981 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6982 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6983 not very cleanly specified and it is unwise to depend on it.
6988 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6989 source location to the destination location, which may overlap. It
6990 copies "len" bytes of memory over. If the argument is known to be
6991 aligned to some boundary, this can be specified as the fourth argument,
6992 otherwise it should be set to 0 or 1 (both meaning no alignment).
6994 '``llvm.memset.*``' Intrinsics
6995 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7000 This is an overloaded intrinsic. You can use llvm.memset on any integer
7001 bit width and for different address spaces. However, not all targets
7002 support all bit widths.
7006 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7007 i32 <len>, i32 <align>, i1 <isvolatile>)
7008 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7009 i64 <len>, i32 <align>, i1 <isvolatile>)
7014 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7015 particular byte value.
7017 Note that, unlike the standard libc function, the ``llvm.memset``
7018 intrinsic does not return a value and takes extra alignment/volatile
7019 arguments. Also, the destination can be in an arbitrary address space.
7024 The first argument is a pointer to the destination to fill, the second
7025 is the byte value with which to fill it, the third argument is an
7026 integer argument specifying the number of bytes to fill, and the fourth
7027 argument is the known alignment of the destination location.
7029 If the call to this intrinsic has an alignment value that is not 0 or 1,
7030 then the caller guarantees that the destination pointer is aligned to
7033 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7034 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7035 very cleanly specified and it is unwise to depend on it.
7040 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7041 at the destination location. If the argument is known to be aligned to
7042 some boundary, this can be specified as the fourth argument, otherwise
7043 it should be set to 0 or 1 (both meaning no alignment).
7045 '``llvm.sqrt.*``' Intrinsic
7046 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7051 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7052 floating point or vector of floating point type. Not all targets support
7057 declare float @llvm.sqrt.f32(float %Val)
7058 declare double @llvm.sqrt.f64(double %Val)
7059 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7060 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7061 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7066 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7067 returning the same value as the libm '``sqrt``' functions would. Unlike
7068 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7069 negative numbers other than -0.0 (which allows for better optimization,
7070 because there is no need to worry about errno being set).
7071 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7076 The argument and return value are floating point numbers of the same
7082 This function returns the sqrt of the specified operand if it is a
7083 nonnegative floating point number.
7085 '``llvm.powi.*``' Intrinsic
7086 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7091 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7092 floating point or vector of floating point type. Not all targets support
7097 declare float @llvm.powi.f32(float %Val, i32 %power)
7098 declare double @llvm.powi.f64(double %Val, i32 %power)
7099 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7100 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7101 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7106 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7107 specified (positive or negative) power. The order of evaluation of
7108 multiplications is not defined. When a vector of floating point type is
7109 used, the second argument remains a scalar integer value.
7114 The second argument is an integer power, and the first is a value to
7115 raise to that power.
7120 This function returns the first value raised to the second power with an
7121 unspecified sequence of rounding operations.
7123 '``llvm.sin.*``' Intrinsic
7124 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7129 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7130 floating point or vector of floating point type. Not all targets support
7135 declare float @llvm.sin.f32(float %Val)
7136 declare double @llvm.sin.f64(double %Val)
7137 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7138 declare fp128 @llvm.sin.f128(fp128 %Val)
7139 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7144 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7149 The argument and return value are floating point numbers of the same
7155 This function returns the sine of the specified operand, returning the
7156 same values as the libm ``sin`` functions would, and handles error
7157 conditions in the same way.
7159 '``llvm.cos.*``' Intrinsic
7160 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7165 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7166 floating point or vector of floating point type. Not all targets support
7171 declare float @llvm.cos.f32(float %Val)
7172 declare double @llvm.cos.f64(double %Val)
7173 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7174 declare fp128 @llvm.cos.f128(fp128 %Val)
7175 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7180 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7185 The argument and return value are floating point numbers of the same
7191 This function returns the cosine of the specified operand, returning the
7192 same values as the libm ``cos`` functions would, and handles error
7193 conditions in the same way.
7195 '``llvm.pow.*``' Intrinsic
7196 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7201 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7202 floating point or vector of floating point type. Not all targets support
7207 declare float @llvm.pow.f32(float %Val, float %Power)
7208 declare double @llvm.pow.f64(double %Val, double %Power)
7209 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7210 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7211 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7216 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7217 specified (positive or negative) power.
7222 The second argument is a floating point power, and the first is a value
7223 to raise to that power.
7228 This function returns the first value raised to the second power,
7229 returning the same values as the libm ``pow`` functions would, and
7230 handles error conditions in the same way.
7232 '``llvm.exp.*``' Intrinsic
7233 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7238 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7239 floating point or vector of floating point type. Not all targets support
7244 declare float @llvm.exp.f32(float %Val)
7245 declare double @llvm.exp.f64(double %Val)
7246 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7247 declare fp128 @llvm.exp.f128(fp128 %Val)
7248 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7253 The '``llvm.exp.*``' intrinsics perform the exp function.
7258 The argument and return value are floating point numbers of the same
7264 This function returns the same values as the libm ``exp`` functions
7265 would, and handles error conditions in the same way.
7267 '``llvm.exp2.*``' Intrinsic
7268 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7273 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7274 floating point or vector of floating point type. Not all targets support
7279 declare float @llvm.exp2.f32(float %Val)
7280 declare double @llvm.exp2.f64(double %Val)
7281 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7282 declare fp128 @llvm.exp2.f128(fp128 %Val)
7283 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7288 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7293 The argument and return value are floating point numbers of the same
7299 This function returns the same values as the libm ``exp2`` functions
7300 would, and handles error conditions in the same way.
7302 '``llvm.log.*``' Intrinsic
7303 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7308 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7309 floating point or vector of floating point type. Not all targets support
7314 declare float @llvm.log.f32(float %Val)
7315 declare double @llvm.log.f64(double %Val)
7316 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7317 declare fp128 @llvm.log.f128(fp128 %Val)
7318 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7323 The '``llvm.log.*``' intrinsics perform the log function.
7328 The argument and return value are floating point numbers of the same
7334 This function returns the same values as the libm ``log`` functions
7335 would, and handles error conditions in the same way.
7337 '``llvm.log10.*``' Intrinsic
7338 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7343 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7344 floating point or vector of floating point type. Not all targets support
7349 declare float @llvm.log10.f32(float %Val)
7350 declare double @llvm.log10.f64(double %Val)
7351 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7352 declare fp128 @llvm.log10.f128(fp128 %Val)
7353 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7358 The '``llvm.log10.*``' intrinsics perform the log10 function.
7363 The argument and return value are floating point numbers of the same
7369 This function returns the same values as the libm ``log10`` functions
7370 would, and handles error conditions in the same way.
7372 '``llvm.log2.*``' Intrinsic
7373 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7378 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7379 floating point or vector of floating point type. Not all targets support
7384 declare float @llvm.log2.f32(float %Val)
7385 declare double @llvm.log2.f64(double %Val)
7386 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7387 declare fp128 @llvm.log2.f128(fp128 %Val)
7388 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7393 The '``llvm.log2.*``' intrinsics perform the log2 function.
7398 The argument and return value are floating point numbers of the same
7404 This function returns the same values as the libm ``log2`` functions
7405 would, and handles error conditions in the same way.
7407 '``llvm.fma.*``' Intrinsic
7408 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7413 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7414 floating point or vector of floating point type. Not all targets support
7419 declare float @llvm.fma.f32(float %a, float %b, float %c)
7420 declare double @llvm.fma.f64(double %a, double %b, double %c)
7421 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7422 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7423 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7428 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7434 The argument and return value are floating point numbers of the same
7440 This function returns the same values as the libm ``fma`` functions
7443 '``llvm.fabs.*``' Intrinsic
7444 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7449 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7450 floating point or vector of floating point type. Not all targets support
7455 declare float @llvm.fabs.f32(float %Val)
7456 declare double @llvm.fabs.f64(double %Val)
7457 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7458 declare fp128 @llvm.fabs.f128(fp128 %Val)
7459 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7464 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7470 The argument and return value are floating point numbers of the same
7476 This function returns the same values as the libm ``fabs`` functions
7477 would, and handles error conditions in the same way.
7479 '``llvm.copysign.*``' Intrinsic
7480 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7485 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7486 floating point or vector of floating point type. Not all targets support
7491 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7492 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7493 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7494 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7495 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7500 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7501 first operand and the sign of the second operand.
7506 The arguments and return value are floating point numbers of the same
7512 This function returns the same values as the libm ``copysign``
7513 functions would, and handles error conditions in the same way.
7515 '``llvm.floor.*``' Intrinsic
7516 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7521 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7522 floating point or vector of floating point type. Not all targets support
7527 declare float @llvm.floor.f32(float %Val)
7528 declare double @llvm.floor.f64(double %Val)
7529 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7530 declare fp128 @llvm.floor.f128(fp128 %Val)
7531 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7536 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7541 The argument and return value are floating point numbers of the same
7547 This function returns the same values as the libm ``floor`` functions
7548 would, and handles error conditions in the same way.
7550 '``llvm.ceil.*``' Intrinsic
7551 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7556 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7557 floating point or vector of floating point type. Not all targets support
7562 declare float @llvm.ceil.f32(float %Val)
7563 declare double @llvm.ceil.f64(double %Val)
7564 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7565 declare fp128 @llvm.ceil.f128(fp128 %Val)
7566 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7571 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7576 The argument and return value are floating point numbers of the same
7582 This function returns the same values as the libm ``ceil`` functions
7583 would, and handles error conditions in the same way.
7585 '``llvm.trunc.*``' Intrinsic
7586 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7591 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7592 floating point or vector of floating point type. Not all targets support
7597 declare float @llvm.trunc.f32(float %Val)
7598 declare double @llvm.trunc.f64(double %Val)
7599 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7600 declare fp128 @llvm.trunc.f128(fp128 %Val)
7601 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7606 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7607 nearest integer not larger in magnitude than the operand.
7612 The argument and return value are floating point numbers of the same
7618 This function returns the same values as the libm ``trunc`` functions
7619 would, and handles error conditions in the same way.
7621 '``llvm.rint.*``' Intrinsic
7622 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7627 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7628 floating point or vector of floating point type. Not all targets support
7633 declare float @llvm.rint.f32(float %Val)
7634 declare double @llvm.rint.f64(double %Val)
7635 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7636 declare fp128 @llvm.rint.f128(fp128 %Val)
7637 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7642 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7643 nearest integer. It may raise an inexact floating-point exception if the
7644 operand isn't an integer.
7649 The argument and return value are floating point numbers of the same
7655 This function returns the same values as the libm ``rint`` functions
7656 would, and handles error conditions in the same way.
7658 '``llvm.nearbyint.*``' Intrinsic
7659 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7664 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7665 floating point or vector of floating point type. Not all targets support
7670 declare float @llvm.nearbyint.f32(float %Val)
7671 declare double @llvm.nearbyint.f64(double %Val)
7672 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7673 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7674 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7679 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7685 The argument and return value are floating point numbers of the same
7691 This function returns the same values as the libm ``nearbyint``
7692 functions would, and handles error conditions in the same way.
7694 '``llvm.round.*``' Intrinsic
7695 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7700 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7701 floating point or vector of floating point type. Not all targets support
7706 declare float @llvm.round.f32(float %Val)
7707 declare double @llvm.round.f64(double %Val)
7708 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7709 declare fp128 @llvm.round.f128(fp128 %Val)
7710 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7715 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7721 The argument and return value are floating point numbers of the same
7727 This function returns the same values as the libm ``round``
7728 functions would, and handles error conditions in the same way.
7730 Bit Manipulation Intrinsics
7731 ---------------------------
7733 LLVM provides intrinsics for a few important bit manipulation
7734 operations. These allow efficient code generation for some algorithms.
7736 '``llvm.bswap.*``' Intrinsics
7737 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7742 This is an overloaded intrinsic function. You can use bswap on any
7743 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7747 declare i16 @llvm.bswap.i16(i16 <id>)
7748 declare i32 @llvm.bswap.i32(i32 <id>)
7749 declare i64 @llvm.bswap.i64(i64 <id>)
7754 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7755 values with an even number of bytes (positive multiple of 16 bits).
7756 These are useful for performing operations on data that is not in the
7757 target's native byte order.
7762 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7763 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7764 intrinsic returns an i32 value that has the four bytes of the input i32
7765 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7766 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7767 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7768 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7771 '``llvm.ctpop.*``' Intrinsic
7772 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7777 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7778 bit width, or on any vector with integer elements. Not all targets
7779 support all bit widths or vector types, however.
7783 declare i8 @llvm.ctpop.i8(i8 <src>)
7784 declare i16 @llvm.ctpop.i16(i16 <src>)
7785 declare i32 @llvm.ctpop.i32(i32 <src>)
7786 declare i64 @llvm.ctpop.i64(i64 <src>)
7787 declare i256 @llvm.ctpop.i256(i256 <src>)
7788 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7793 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7799 The only argument is the value to be counted. The argument may be of any
7800 integer type, or a vector with integer elements. The return type must
7801 match the argument type.
7806 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7807 each element of a vector.
7809 '``llvm.ctlz.*``' Intrinsic
7810 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7815 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7816 integer bit width, or any vector whose elements are integers. Not all
7817 targets support all bit widths or vector types, however.
7821 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7822 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7823 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7824 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7825 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7826 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7831 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7832 leading zeros in a variable.
7837 The first argument is the value to be counted. This argument may be of
7838 any integer type, or a vectory with integer element type. The return
7839 type must match the first argument type.
7841 The second argument must be a constant and is a flag to indicate whether
7842 the intrinsic should ensure that a zero as the first argument produces a
7843 defined result. Historically some architectures did not provide a
7844 defined result for zero values as efficiently, and many algorithms are
7845 now predicated on avoiding zero-value inputs.
7850 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7851 zeros in a variable, or within each element of the vector. If
7852 ``src == 0`` then the result is the size in bits of the type of ``src``
7853 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7854 ``llvm.ctlz(i32 2) = 30``.
7856 '``llvm.cttz.*``' Intrinsic
7857 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7862 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7863 integer bit width, or any vector of integer elements. Not all targets
7864 support all bit widths or vector types, however.
7868 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7869 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7870 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7871 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7872 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7873 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7878 The '``llvm.cttz``' family of intrinsic functions counts the number of
7884 The first argument is the value to be counted. This argument may be of
7885 any integer type, or a vectory with integer element type. The return
7886 type must match the first argument type.
7888 The second argument must be a constant and is a flag to indicate whether
7889 the intrinsic should ensure that a zero as the first argument produces a
7890 defined result. Historically some architectures did not provide a
7891 defined result for zero values as efficiently, and many algorithms are
7892 now predicated on avoiding zero-value inputs.
7897 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7898 zeros in a variable, or within each element of a vector. If ``src == 0``
7899 then the result is the size in bits of the type of ``src`` if
7900 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7901 ``llvm.cttz(2) = 1``.
7903 Arithmetic with Overflow Intrinsics
7904 -----------------------------------
7906 LLVM provides intrinsics for some arithmetic with overflow operations.
7908 '``llvm.sadd.with.overflow.*``' Intrinsics
7909 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7914 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7915 on any integer bit width.
7919 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7920 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7921 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7926 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7927 a signed addition of the two arguments, and indicate whether an overflow
7928 occurred during the signed summation.
7933 The arguments (%a and %b) and the first element of the result structure
7934 may be of integer types of any bit width, but they must have the same
7935 bit width. The second element of the result structure must be of type
7936 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7942 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7943 a signed addition of the two variables. They return a structure --- the
7944 first element of which is the signed summation, and the second element
7945 of which is a bit specifying if the signed summation resulted in an
7951 .. code-block:: llvm
7953 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7954 %sum = extractvalue {i32, i1} %res, 0
7955 %obit = extractvalue {i32, i1} %res, 1
7956 br i1 %obit, label %overflow, label %normal
7958 '``llvm.uadd.with.overflow.*``' Intrinsics
7959 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7964 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7965 on any integer bit width.
7969 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7970 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7971 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7976 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7977 an unsigned addition of the two arguments, and indicate whether a carry
7978 occurred during the unsigned summation.
7983 The arguments (%a and %b) and the first element of the result structure
7984 may be of integer types of any bit width, but they must have the same
7985 bit width. The second element of the result structure must be of type
7986 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7992 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7993 an unsigned addition of the two arguments. They return a structure --- the
7994 first element of which is the sum, and the second element of which is a
7995 bit specifying if the unsigned summation resulted in a carry.
8000 .. code-block:: llvm
8002 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8003 %sum = extractvalue {i32, i1} %res, 0
8004 %obit = extractvalue {i32, i1} %res, 1
8005 br i1 %obit, label %carry, label %normal
8007 '``llvm.ssub.with.overflow.*``' Intrinsics
8008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8013 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8014 on any integer bit width.
8018 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8019 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8020 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8025 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8026 a signed subtraction of the two arguments, and indicate whether an
8027 overflow occurred during the signed subtraction.
8032 The arguments (%a and %b) and the first element of the result structure
8033 may be of integer types of any bit width, but they must have the same
8034 bit width. The second element of the result structure must be of type
8035 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8041 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8042 a signed subtraction of the two arguments. They return a structure --- the
8043 first element of which is the subtraction, and the second element of
8044 which is a bit specifying if the signed subtraction resulted in an
8050 .. code-block:: llvm
8052 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8053 %sum = extractvalue {i32, i1} %res, 0
8054 %obit = extractvalue {i32, i1} %res, 1
8055 br i1 %obit, label %overflow, label %normal
8057 '``llvm.usub.with.overflow.*``' Intrinsics
8058 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8063 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8064 on any integer bit width.
8068 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8069 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8070 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8075 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8076 an unsigned subtraction of the two arguments, and indicate whether an
8077 overflow occurred during the unsigned subtraction.
8082 The arguments (%a and %b) and the first element of the result structure
8083 may be of integer types of any bit width, but they must have the same
8084 bit width. The second element of the result structure must be of type
8085 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8091 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8092 an unsigned subtraction of the two arguments. They return a structure ---
8093 the first element of which is the subtraction, and the second element of
8094 which is a bit specifying if the unsigned subtraction resulted in an
8100 .. code-block:: llvm
8102 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8103 %sum = extractvalue {i32, i1} %res, 0
8104 %obit = extractvalue {i32, i1} %res, 1
8105 br i1 %obit, label %overflow, label %normal
8107 '``llvm.smul.with.overflow.*``' Intrinsics
8108 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8113 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8114 on any integer bit width.
8118 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8119 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8120 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8125 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8126 a signed multiplication of the two arguments, and indicate whether an
8127 overflow occurred during the signed multiplication.
8132 The arguments (%a and %b) and the first element of the result structure
8133 may be of integer types of any bit width, but they must have the same
8134 bit width. The second element of the result structure must be of type
8135 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8141 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8142 a signed multiplication of the two arguments. They return a structure ---
8143 the first element of which is the multiplication, and the second element
8144 of which is a bit specifying if the signed multiplication resulted in an
8150 .. code-block:: llvm
8152 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8153 %sum = extractvalue {i32, i1} %res, 0
8154 %obit = extractvalue {i32, i1} %res, 1
8155 br i1 %obit, label %overflow, label %normal
8157 '``llvm.umul.with.overflow.*``' Intrinsics
8158 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8163 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8164 on any integer bit width.
8168 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8169 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8170 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8175 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8176 a unsigned multiplication of the two arguments, and indicate whether an
8177 overflow occurred during the unsigned multiplication.
8182 The arguments (%a and %b) and the first element of the result structure
8183 may be of integer types of any bit width, but they must have the same
8184 bit width. The second element of the result structure must be of type
8185 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8191 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8192 an unsigned multiplication of the two arguments. They return a structure ---
8193 the first element of which is the multiplication, and the second
8194 element of which is a bit specifying if the unsigned multiplication
8195 resulted in an overflow.
8200 .. code-block:: llvm
8202 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8203 %sum = extractvalue {i32, i1} %res, 0
8204 %obit = extractvalue {i32, i1} %res, 1
8205 br i1 %obit, label %overflow, label %normal
8207 Specialised Arithmetic Intrinsics
8208 ---------------------------------
8210 '``llvm.fmuladd.*``' Intrinsic
8211 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8218 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8219 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8224 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8225 expressions that can be fused if the code generator determines that (a) the
8226 target instruction set has support for a fused operation, and (b) that the
8227 fused operation is more efficient than the equivalent, separate pair of mul
8228 and add instructions.
8233 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8234 multiplicands, a and b, and an addend c.
8243 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8245 is equivalent to the expression a \* b + c, except that rounding will
8246 not be performed between the multiplication and addition steps if the
8247 code generator fuses the operations. Fusion is not guaranteed, even if
8248 the target platform supports it. If a fused multiply-add is required the
8249 corresponding llvm.fma.\* intrinsic function should be used instead.
8254 .. code-block:: llvm
8256 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8258 Half Precision Floating Point Intrinsics
8259 ----------------------------------------
8261 For most target platforms, half precision floating point is a
8262 storage-only format. This means that it is a dense encoding (in memory)
8263 but does not support computation in the format.
8265 This means that code must first load the half-precision floating point
8266 value as an i16, then convert it to float with
8267 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8268 then be performed on the float value (including extending to double
8269 etc). To store the value back to memory, it is first converted to float
8270 if needed, then converted to i16 with
8271 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8274 .. _int_convert_to_fp16:
8276 '``llvm.convert.to.fp16``' Intrinsic
8277 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8284 declare i16 @llvm.convert.to.fp16(f32 %a)
8289 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8290 from single precision floating point format to half precision floating
8296 The intrinsic function contains single argument - the value to be
8302 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8303 from single precision floating point format to half precision floating
8304 point format. The return value is an ``i16`` which contains the
8310 .. code-block:: llvm
8312 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8313 store i16 %res, i16* @x, align 2
8315 .. _int_convert_from_fp16:
8317 '``llvm.convert.from.fp16``' Intrinsic
8318 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8325 declare f32 @llvm.convert.from.fp16(i16 %a)
8330 The '``llvm.convert.from.fp16``' intrinsic function performs a
8331 conversion from half precision floating point format to single precision
8332 floating point format.
8337 The intrinsic function contains single argument - the value to be
8343 The '``llvm.convert.from.fp16``' intrinsic function performs a
8344 conversion from half single precision floating point format to single
8345 precision floating point format. The input half-float value is
8346 represented by an ``i16`` value.
8351 .. code-block:: llvm
8353 %a = load i16* @x, align 2
8354 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8359 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8360 prefix), are described in the `LLVM Source Level
8361 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8364 Exception Handling Intrinsics
8365 -----------------------------
8367 The LLVM exception handling intrinsics (which all start with
8368 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8369 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8373 Trampoline Intrinsics
8374 ---------------------
8376 These intrinsics make it possible to excise one parameter, marked with
8377 the :ref:`nest <nest>` attribute, from a function. The result is a
8378 callable function pointer lacking the nest parameter - the caller does
8379 not need to provide a value for it. Instead, the value to use is stored
8380 in advance in a "trampoline", a block of memory usually allocated on the
8381 stack, which also contains code to splice the nest value into the
8382 argument list. This is used to implement the GCC nested function address
8385 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8386 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8387 It can be created as follows:
8389 .. code-block:: llvm
8391 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8392 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8393 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8394 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8395 %fp = bitcast i8* %p to i32 (i32, i32)*
8397 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8398 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8402 '``llvm.init.trampoline``' Intrinsic
8403 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8410 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8415 This fills the memory pointed to by ``tramp`` with executable code,
8416 turning it into a trampoline.
8421 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8422 pointers. The ``tramp`` argument must point to a sufficiently large and
8423 sufficiently aligned block of memory; this memory is written to by the
8424 intrinsic. Note that the size and the alignment are target-specific -
8425 LLVM currently provides no portable way of determining them, so a
8426 front-end that generates this intrinsic needs to have some
8427 target-specific knowledge. The ``func`` argument must hold a function
8428 bitcast to an ``i8*``.
8433 The block of memory pointed to by ``tramp`` is filled with target
8434 dependent code, turning it into a function. Then ``tramp`` needs to be
8435 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8436 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8437 function's signature is the same as that of ``func`` with any arguments
8438 marked with the ``nest`` attribute removed. At most one such ``nest``
8439 argument is allowed, and it must be of pointer type. Calling the new
8440 function is equivalent to calling ``func`` with the same argument list,
8441 but with ``nval`` used for the missing ``nest`` argument. If, after
8442 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8443 modified, then the effect of any later call to the returned function
8444 pointer is undefined.
8448 '``llvm.adjust.trampoline``' Intrinsic
8449 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8456 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8461 This performs any required machine-specific adjustment to the address of
8462 a trampoline (passed as ``tramp``).
8467 ``tramp`` must point to a block of memory which already has trampoline
8468 code filled in by a previous call to
8469 :ref:`llvm.init.trampoline <int_it>`.
8474 On some architectures the address of the code to be executed needs to be
8475 different to the address where the trampoline is actually stored. This
8476 intrinsic returns the executable address corresponding to ``tramp``
8477 after performing the required machine specific adjustments. The pointer
8478 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8483 This class of intrinsics exists to information about the lifetime of
8484 memory objects and ranges where variables are immutable.
8488 '``llvm.lifetime.start``' Intrinsic
8489 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8496 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8501 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8507 The first argument is a constant integer representing the size of the
8508 object, or -1 if it is variable sized. The second argument is a pointer
8514 This intrinsic indicates that before this point in the code, the value
8515 of the memory pointed to by ``ptr`` is dead. This means that it is known
8516 to never be used and has an undefined value. A load from the pointer
8517 that precedes this intrinsic can be replaced with ``'undef'``.
8521 '``llvm.lifetime.end``' Intrinsic
8522 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8529 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8534 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8540 The first argument is a constant integer representing the size of the
8541 object, or -1 if it is variable sized. The second argument is a pointer
8547 This intrinsic indicates that after this point in the code, the value of
8548 the memory pointed to by ``ptr`` is dead. This means that it is known to
8549 never be used and has an undefined value. Any stores into the memory
8550 object following this intrinsic may be removed as dead.
8552 '``llvm.invariant.start``' Intrinsic
8553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8560 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8565 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8566 a memory object will not change.
8571 The first argument is a constant integer representing the size of the
8572 object, or -1 if it is variable sized. The second argument is a pointer
8578 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8579 the return value, the referenced memory location is constant and
8582 '``llvm.invariant.end``' Intrinsic
8583 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8590 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8595 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8596 memory object are mutable.
8601 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8602 The second argument is a constant integer representing the size of the
8603 object, or -1 if it is variable sized and the third argument is a
8604 pointer to the object.
8609 This intrinsic indicates that the memory is mutable again.
8614 This class of intrinsics is designed to be generic and has no specific
8617 '``llvm.var.annotation``' Intrinsic
8618 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8625 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8630 The '``llvm.var.annotation``' intrinsic.
8635 The first argument is a pointer to a value, the second is a pointer to a
8636 global string, the third is a pointer to a global string which is the
8637 source file name, and the last argument is the line number.
8642 This intrinsic allows annotation of local variables with arbitrary
8643 strings. This can be useful for special purpose optimizations that want
8644 to look for these annotations. These have no other defined use; they are
8645 ignored by code generation and optimization.
8647 '``llvm.ptr.annotation.*``' Intrinsic
8648 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8653 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8654 pointer to an integer of any width. *NOTE* you must specify an address space for
8655 the pointer. The identifier for the default address space is the integer
8660 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8661 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8662 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8663 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8664 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8669 The '``llvm.ptr.annotation``' intrinsic.
8674 The first argument is a pointer to an integer value of arbitrary bitwidth
8675 (result of some expression), the second is a pointer to a global string, the
8676 third is a pointer to a global string which is the source file name, and the
8677 last argument is the line number. It returns the value of the first argument.
8682 This intrinsic allows annotation of a pointer to an integer with arbitrary
8683 strings. This can be useful for special purpose optimizations that want to look
8684 for these annotations. These have no other defined use; they are ignored by code
8685 generation and optimization.
8687 '``llvm.annotation.*``' Intrinsic
8688 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8693 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8694 any integer bit width.
8698 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8699 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8700 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8701 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8702 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8707 The '``llvm.annotation``' intrinsic.
8712 The first argument is an integer value (result of some expression), the
8713 second is a pointer to a global string, the third is a pointer to a
8714 global string which is the source file name, and the last argument is
8715 the line number. It returns the value of the first argument.
8720 This intrinsic allows annotations to be put on arbitrary expressions
8721 with arbitrary strings. This can be useful for special purpose
8722 optimizations that want to look for these annotations. These have no
8723 other defined use; they are ignored by code generation and optimization.
8725 '``llvm.trap``' Intrinsic
8726 ^^^^^^^^^^^^^^^^^^^^^^^^^
8733 declare void @llvm.trap() noreturn nounwind
8738 The '``llvm.trap``' intrinsic.
8748 This intrinsic is lowered to the target dependent trap instruction. If
8749 the target does not have a trap instruction, this intrinsic will be
8750 lowered to a call of the ``abort()`` function.
8752 '``llvm.debugtrap``' Intrinsic
8753 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8760 declare void @llvm.debugtrap() nounwind
8765 The '``llvm.debugtrap``' intrinsic.
8775 This intrinsic is lowered to code which is intended to cause an
8776 execution trap with the intention of requesting the attention of a
8779 '``llvm.stackprotector``' Intrinsic
8780 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8787 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8792 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8793 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8794 is placed on the stack before local variables.
8799 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8800 The first argument is the value loaded from the stack guard
8801 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8802 enough space to hold the value of the guard.
8807 This intrinsic causes the prologue/epilogue inserter to force the position of
8808 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8809 to ensure that if a local variable on the stack is overwritten, it will destroy
8810 the value of the guard. When the function exits, the guard on the stack is
8811 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8812 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8813 calling the ``__stack_chk_fail()`` function.
8815 '``llvm.stackprotectorcheck``' Intrinsic
8816 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8823 declare void @llvm.stackprotectorcheck(i8** <guard>)
8828 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8829 created stack protector and if they are not equal calls the
8830 ``__stack_chk_fail()`` function.
8835 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8836 the variable ``@__stack_chk_guard``.
8841 This intrinsic is provided to perform the stack protector check by comparing
8842 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8843 values do not match call the ``__stack_chk_fail()`` function.
8845 The reason to provide this as an IR level intrinsic instead of implementing it
8846 via other IR operations is that in order to perform this operation at the IR
8847 level without an intrinsic, one would need to create additional basic blocks to
8848 handle the success/failure cases. This makes it difficult to stop the stack
8849 protector check from disrupting sibling tail calls in Codegen. With this
8850 intrinsic, we are able to generate the stack protector basic blocks late in
8851 codegen after the tail call decision has occurred.
8853 '``llvm.objectsize``' Intrinsic
8854 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8861 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8862 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8867 The ``llvm.objectsize`` intrinsic is designed to provide information to
8868 the optimizers to determine at compile time whether a) an operation
8869 (like memcpy) will overflow a buffer that corresponds to an object, or
8870 b) that a runtime check for overflow isn't necessary. An object in this
8871 context means an allocation of a specific class, structure, array, or
8877 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8878 argument is a pointer to or into the ``object``. The second argument is
8879 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8880 or -1 (if false) when the object size is unknown. The second argument
8881 only accepts constants.
8886 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8887 the size of the object concerned. If the size cannot be determined at
8888 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8889 on the ``min`` argument).
8891 '``llvm.expect``' Intrinsic
8892 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8899 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8900 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8905 The ``llvm.expect`` intrinsic provides information about expected (the
8906 most probable) value of ``val``, which can be used by optimizers.
8911 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8912 a value. The second argument is an expected value, this needs to be a
8913 constant value, variables are not allowed.
8918 This intrinsic is lowered to the ``val``.
8920 '``llvm.donothing``' Intrinsic
8921 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8928 declare void @llvm.donothing() nounwind readnone
8933 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8934 only intrinsic that can be called with an invoke instruction.
8944 This intrinsic does nothing, and it's removed by optimizers and ignored
8947 Stack Map Intrinsics
8948 --------------------
8950 LLVM provides experimental intrinsics to support runtime patching
8951 mechanisms commonly desired in dynamic language JITs. These intrinsics
8952 are described in :doc:`StackMaps`.