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 The next two types of linkage are targeted for Microsoft Windows
278 platform only. They are designed to support importing (exporting)
279 symbols from (to) DLLs (Dynamic Link Libraries).
282 "``dllimport``" linkage causes the compiler to reference a function
283 or variable via a global pointer to a pointer that is set up by the
284 DLL exporting the symbol. On Microsoft Windows targets, the pointer
285 name is formed by combining ``__imp_`` and the function or variable
288 "``dllexport``" linkage causes the compiler to provide a global
289 pointer to a pointer in a DLL, so that it can be referenced with the
290 ``dllimport`` attribute. On Microsoft Windows targets, the pointer
291 name is formed by combining ``__imp_`` and the function or variable
292 name. Since this linkage exists for defining a dll interface, the
293 compiler, assembler and linker know it is externally referenced and
294 must refrain from deleting the symbol.
296 It is illegal for a function *declaration* to have any linkage type
297 other than ``external``, ``dllimport`` or ``extern_weak``.
304 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
305 :ref:`invokes <i_invoke>` can all have an optional calling convention
306 specified for the call. The calling convention of any pair of dynamic
307 caller/callee must match, or the behavior of the program is undefined.
308 The following calling conventions are supported by LLVM, and more may be
311 "``ccc``" - The C calling convention
312 This calling convention (the default if no other calling convention
313 is specified) matches the target C calling conventions. This calling
314 convention supports varargs function calls and tolerates some
315 mismatch in the declared prototype and implemented declaration of
316 the function (as does normal C).
317 "``fastcc``" - The fast calling convention
318 This calling convention attempts to make calls as fast as possible
319 (e.g. by passing things in registers). This calling convention
320 allows the target to use whatever tricks it wants to produce fast
321 code for the target, without having to conform to an externally
322 specified ABI (Application Binary Interface). `Tail calls can only
323 be optimized when this, the GHC or the HiPE convention is
324 used. <CodeGenerator.html#id80>`_ This calling convention does not
325 support varargs and requires the prototype of all callees to exactly
326 match the prototype of the function definition.
327 "``coldcc``" - The cold calling convention
328 This calling convention attempts to make code in the caller as
329 efficient as possible under the assumption that the call is not
330 commonly executed. As such, these calls often preserve all registers
331 so that the call does not break any live ranges in the caller side.
332 This calling convention does not support varargs and requires the
333 prototype of all callees to exactly match the prototype of the
335 "``cc 10``" - GHC convention
336 This calling convention has been implemented specifically for use by
337 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
338 It passes everything in registers, going to extremes to achieve this
339 by disabling callee save registers. This calling convention should
340 not be used lightly but only for specific situations such as an
341 alternative to the *register pinning* performance technique often
342 used when implementing functional programming languages. At the
343 moment only X86 supports this convention and it has the following
346 - On *X86-32* only supports up to 4 bit type parameters. No
347 floating point types are supported.
348 - On *X86-64* only supports up to 10 bit type parameters and 6
349 floating point parameters.
351 This calling convention supports `tail call
352 optimization <CodeGenerator.html#id80>`_ but requires both the
353 caller and callee are using it.
354 "``cc 11``" - The HiPE calling convention
355 This calling convention has been implemented specifically for use by
356 the `High-Performance Erlang
357 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
358 native code compiler of the `Ericsson's Open Source Erlang/OTP
359 system <http://www.erlang.org/download.shtml>`_. It uses more
360 registers for argument passing than the ordinary C calling
361 convention and defines no callee-saved registers. The calling
362 convention properly supports `tail call
363 optimization <CodeGenerator.html#id80>`_ but requires that both the
364 caller and the callee use it. It uses a *register pinning*
365 mechanism, similar to GHC's convention, for keeping frequently
366 accessed runtime components pinned to specific hardware registers.
367 At the moment only X86 supports this convention (both 32 and 64
369 "``cc <n>``" - Numbered convention
370 Any calling convention may be specified by number, allowing
371 target-specific calling conventions to be used. Target specific
372 calling conventions start at 64.
374 More calling conventions can be added/defined on an as-needed basis, to
375 support Pascal conventions or any other well-known target-independent
378 .. _visibilitystyles:
383 All Global Variables and Functions have one of the following visibility
386 "``default``" - Default style
387 On targets that use the ELF object file format, default visibility
388 means that the declaration is visible to other modules and, in
389 shared libraries, means that the declared entity may be overridden.
390 On Darwin, default visibility means that the declaration is visible
391 to other modules. Default visibility corresponds to "external
392 linkage" in the language.
393 "``hidden``" - Hidden style
394 Two declarations of an object with hidden visibility refer to the
395 same object if they are in the same shared object. Usually, hidden
396 visibility indicates that the symbol will not be placed into the
397 dynamic symbol table, so no other module (executable or shared
398 library) can reference it directly.
399 "``protected``" - Protected style
400 On ELF, protected visibility indicates that the symbol will be
401 placed in the dynamic symbol table, but that references within the
402 defining module will bind to the local symbol. That is, the symbol
403 cannot be overridden by another module.
410 LLVM IR allows you to specify name aliases for certain types. This can
411 make it easier to read the IR and make the IR more condensed
412 (particularly when recursive types are involved). An example of a name
417 %mytype = type { %mytype*, i32 }
419 You may give a name to any :ref:`type <typesystem>` except
420 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
421 expected with the syntax "%mytype".
423 Note that type names are aliases for the structural type that they
424 indicate, and that you can therefore specify multiple names for the same
425 type. This often leads to confusing behavior when dumping out a .ll
426 file. Since LLVM IR uses structural typing, the name is not part of the
427 type. When printing out LLVM IR, the printer will pick *one name* to
428 render all types of a particular shape. This means that if you have code
429 where two different source types end up having the same LLVM type, that
430 the dumper will sometimes print the "wrong" or unexpected type. This is
431 an important design point and isn't going to change.
438 Global variables define regions of memory allocated at compilation time
441 Global variables definitions must be initialized, may have an explicit section
442 to be placed in, and may have an optional explicit alignment specified.
444 Global variables in other translation units can also be declared, in which
445 case they don't have an initializer.
447 A variable may be defined as ``thread_local``, which means that it will
448 not be shared by threads (each thread will have a separated copy of the
449 variable). Not all targets support thread-local variables. Optionally, a
450 TLS model may be specified:
453 For variables that are only used within the current shared library.
455 For variables in modules that will not be loaded dynamically.
457 For variables defined in the executable and only used within it.
459 The models correspond to the ELF TLS models; see `ELF Handling For
460 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
461 more information on under which circumstances the different models may
462 be used. The target may choose a different TLS model if the specified
463 model is not supported, or if a better choice of model can be made.
465 A variable may be defined as a global ``constant``, which indicates that
466 the contents of the variable will **never** be modified (enabling better
467 optimization, allowing the global data to be placed in the read-only
468 section of an executable, etc). Note that variables that need runtime
469 initialization cannot be marked ``constant`` as there is a store to the
472 LLVM explicitly allows *declarations* of global variables to be marked
473 constant, even if the final definition of the global is not. This
474 capability can be used to enable slightly better optimization of the
475 program, but requires the language definition to guarantee that
476 optimizations based on the 'constantness' are valid for the translation
477 units that do not include the definition.
479 As SSA values, global variables define pointer values that are in scope
480 (i.e. they dominate) all basic blocks in the program. Global variables
481 always define a pointer to their "content" type because they describe a
482 region of memory, and all memory objects in LLVM are accessed through
485 Global variables can be marked with ``unnamed_addr`` which indicates
486 that the address is not significant, only the content. Constants marked
487 like this can be merged with other constants if they have the same
488 initializer. Note that a constant with significant address *can* be
489 merged with a ``unnamed_addr`` constant, the result being a constant
490 whose address is significant.
492 A global variable may be declared to reside in a target-specific
493 numbered address space. For targets that support them, address spaces
494 may affect how optimizations are performed and/or what target
495 instructions are used to access the variable. The default address space
496 is zero. The address space qualifier must precede any other attributes.
498 LLVM allows an explicit section to be specified for globals. If the
499 target supports it, it will emit globals to the section specified.
501 By default, global initializers are optimized by assuming that global
502 variables defined within the module are not modified from their
503 initial values before the start of the global initializer. This is
504 true even for variables potentially accessible from outside the
505 module, including those with external linkage or appearing in
506 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
507 by marking the variable with ``externally_initialized``.
509 An explicit alignment may be specified for a global, which must be a
510 power of 2. If not present, or if the alignment is set to zero, the
511 alignment of the global is set by the target to whatever it feels
512 convenient. If an explicit alignment is specified, the global is forced
513 to have exactly that alignment. Targets and optimizers are not allowed
514 to over-align the global if the global has an assigned section. In this
515 case, the extra alignment could be observable: for example, code could
516 assume that the globals are densely packed in their section and try to
517 iterate over them as an array, alignment padding would break this
520 For example, the following defines a global in a numbered address space
521 with an initializer, section, and alignment:
525 @G = addrspace(5) constant float 1.0, section "foo", align 4
527 The following example just declares a global variable
531 @G = external global i32
533 The following example defines a thread-local global with the
534 ``initialexec`` TLS model:
538 @G = thread_local(initialexec) global i32 0, align 4
540 .. _functionstructure:
545 LLVM function definitions consist of the "``define``" keyword, an
546 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
547 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
548 an optional ``unnamed_addr`` attribute, a return type, an optional
549 :ref:`parameter attribute <paramattrs>` for the return type, a function
550 name, a (possibly empty) argument list (each with optional :ref:`parameter
551 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
552 an optional section, an optional alignment, an optional :ref:`garbage
553 collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
554 curly brace, a list of basic blocks, and a closing curly brace.
556 LLVM function declarations consist of the "``declare``" keyword, an
557 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
558 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
559 an optional ``unnamed_addr`` attribute, a return type, an optional
560 :ref:`parameter attribute <paramattrs>` for the return type, a function
561 name, a possibly empty list of arguments, an optional alignment, an optional
562 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
564 A function definition contains a list of basic blocks, forming the CFG (Control
565 Flow Graph) for the function. Each basic block may optionally start with a label
566 (giving the basic block a symbol table entry), contains a list of instructions,
567 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
568 function return). If an explicit label is not provided, a block is assigned an
569 implicit numbered label, using the next value from the same counter as used for
570 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
571 entry block does not have an explicit label, it will be assigned label "%0",
572 then the first unnamed temporary in that block will be "%1", etc.
574 The first basic block in a function is special in two ways: it is
575 immediately executed on entrance to the function, and it is not allowed
576 to have predecessor basic blocks (i.e. there can not be any branches to
577 the entry block of a function). Because the block can have no
578 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
580 LLVM allows an explicit section to be specified for functions. If the
581 target supports it, it will emit functions to the section specified.
583 An explicit alignment may be specified for a function. If not present,
584 or if the alignment is set to zero, the alignment of the function is set
585 by the target to whatever it feels convenient. If an explicit alignment
586 is specified, the function is forced to have at least that much
587 alignment. All alignments must be a power of 2.
589 If the ``unnamed_addr`` attribute is given, the address is know to not
590 be significant and two identical functions can be merged.
594 define [linkage] [visibility]
596 <ResultType> @<FunctionName> ([argument list])
597 [fn Attrs] [section "name"] [align N]
598 [gc] [prefix Constant] { ... }
605 Aliases act as "second name" for the aliasee value (which can be either
606 function, global variable, another alias or bitcast of global value).
607 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
608 :ref:`visibility style <visibility>`.
612 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
614 The linkage must be one of ``private``, ``linker_private``,
615 ``linker_private_weak``, ``internal``, ``linkonce``, ``weak``,
616 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
617 might not correctly handle dropping a weak symbol that is aliased by a non-weak
620 .. _namedmetadatastructure:
625 Named metadata is a collection of metadata. :ref:`Metadata
626 nodes <metadata>` (but not metadata strings) are the only valid
627 operands for a named metadata.
631 ; Some unnamed metadata nodes, which are referenced by the named metadata.
632 !0 = metadata !{metadata !"zero"}
633 !1 = metadata !{metadata !"one"}
634 !2 = metadata !{metadata !"two"}
636 !name = !{!0, !1, !2}
643 The return type and each parameter of a function type may have a set of
644 *parameter attributes* associated with them. Parameter attributes are
645 used to communicate additional information about the result or
646 parameters of a function. Parameter attributes are considered to be part
647 of the function, not of the function type, so functions with different
648 parameter attributes can have the same function type.
650 Parameter attributes are simple keywords that follow the type specified.
651 If multiple parameter attributes are needed, they are space separated.
656 declare i32 @printf(i8* noalias nocapture, ...)
657 declare i32 @atoi(i8 zeroext)
658 declare signext i8 @returns_signed_char()
660 Note that any attributes for the function result (``nounwind``,
661 ``readonly``) come immediately after the argument list.
663 Currently, only the following parameter attributes are defined:
666 This indicates to the code generator that the parameter or return
667 value should be zero-extended to the extent required by the target's
668 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
669 the caller (for a parameter) or the callee (for a return value).
671 This indicates to the code generator that the parameter or return
672 value should be sign-extended to the extent required by the target's
673 ABI (which is usually 32-bits) by the caller (for a parameter) or
674 the callee (for a return value).
676 This indicates that this parameter or return value should be treated
677 in a special target-dependent fashion during while emitting code for
678 a function call or return (usually, by putting it in a register as
679 opposed to memory, though some targets use it to distinguish between
680 two different kinds of registers). Use of this attribute is
683 This indicates that the pointer parameter should really be passed by
684 value to the function. The attribute implies that a hidden copy of
685 the pointee is made between the caller and the callee, so the callee
686 is unable to modify the value in the caller. This attribute is only
687 valid on LLVM pointer arguments. It is generally used to pass
688 structs and arrays by value, but is also valid on pointers to
689 scalars. The copy is considered to belong to the caller not the
690 callee (for example, ``readonly`` functions should not write to
691 ``byval`` parameters). This is not a valid attribute for return
694 The byval attribute also supports specifying an alignment with the
695 align attribute. It indicates the alignment of the stack slot to
696 form and the known alignment of the pointer specified to the call
697 site. If the alignment is not specified, then the code generator
698 makes a target-specific assumption.
704 .. Warning:: This feature is unstable and not fully implemented.
706 The ``inalloca`` argument attribute allows the caller to get the
707 address of an outgoing argument to a ``call`` or ``invoke`` before
708 it executes. It is similar to ``byval`` in that it is used to pass
709 arguments by value, but it guarantees that the argument will not be
712 To be :ref:`well formed <wellformed>`, the caller must pass in an
713 alloca value into an ``inalloca`` parameter, and an alloca may be
714 used as an ``inalloca`` argument at most once. The attribute can
715 only be applied to parameters that would be passed in memory and not
716 registers. The ``inalloca`` attribute cannot be used in conjunction
717 with other attributes that affect argument storage, like ``inreg``,
718 ``nest``, ``sret``, or ``byval``. The ``inalloca`` stack space is
719 considered to be clobbered by any call that uses it, so any
720 ``inalloca`` parameters cannot be marked ``readonly``.
722 Allocas passed with ``inalloca`` to a call must be in the opposite
723 order of the parameter list, meaning that the rightmost argument
724 must be allocated first. If a call has inalloca arguments, no other
725 allocas can occur between the first alloca used by the call and the
726 call site, unless they are are cleared by calls to
727 :ref:`llvm.stackrestore <int_stackrestore>`. Violating these rules
728 results in undefined behavior at runtime.
730 See :doc:`InAlloca` for more information on how to use this
734 This indicates that the pointer parameter specifies the address of a
735 structure that is the return value of the function in the source
736 program. This pointer must be guaranteed by the caller to be valid:
737 loads and stores to the structure may be assumed by the callee
738 not to trap and to be properly aligned. This may only be applied to
739 the first parameter. This is not a valid attribute for return
742 This indicates that pointer values :ref:`based <pointeraliasing>` on
743 the argument or return value do not alias pointer values which are
744 not *based* on it, ignoring certain "irrelevant" dependencies. For a
745 call to the parent function, dependencies between memory references
746 from before or after the call and from those during the call are
747 "irrelevant" to the ``noalias`` keyword for the arguments and return
748 value used in that call. The caller shares the responsibility with
749 the callee for ensuring that these requirements are met. For further
750 details, please see the discussion of the NoAlias response in `alias
751 analysis <AliasAnalysis.html#MustMayNo>`_.
753 Note that this definition of ``noalias`` is intentionally similar
754 to the definition of ``restrict`` in C99 for function arguments,
755 though it is slightly weaker.
757 For function return values, C99's ``restrict`` is not meaningful,
758 while LLVM's ``noalias`` is.
760 This indicates that the callee does not make any copies of the
761 pointer that outlive the callee itself. This is not a valid
762 attribute for return values.
767 This indicates that the pointer parameter can be excised using the
768 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
769 attribute for return values and can only be applied to one parameter.
772 This indicates that the function always returns the argument as its return
773 value. This is an optimization hint to the code generator when generating
774 the caller, allowing tail call optimization and omission of register saves
775 and restores in some cases; it is not checked or enforced when generating
776 the callee. The parameter and the function return type must be valid
777 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
778 valid attribute for return values and can only be applied to one parameter.
782 Garbage Collector Names
783 -----------------------
785 Each function may specify a garbage collector name, which is simply a
790 define void @f() gc "name" { ... }
792 The compiler declares the supported values of *name*. Specifying a
793 collector which will cause the compiler to alter its output in order to
794 support the named garbage collection algorithm.
801 Prefix data is data associated with a function which the code generator
802 will emit immediately before the function body. The purpose of this feature
803 is to allow frontends to associate language-specific runtime metadata with
804 specific functions and make it available through the function pointer while
805 still allowing the function pointer to be called. To access the data for a
806 given function, a program may bitcast the function pointer to a pointer to
807 the constant's type. This implies that the IR symbol points to the start
810 To maintain the semantics of ordinary function calls, the prefix data must
811 have a particular format. Specifically, it must begin with a sequence of
812 bytes which decode to a sequence of machine instructions, valid for the
813 module's target, which transfer control to the point immediately succeeding
814 the prefix data, without performing any other visible action. This allows
815 the inliner and other passes to reason about the semantics of the function
816 definition without needing to reason about the prefix data. Obviously this
817 makes the format of the prefix data highly target dependent.
819 Prefix data is laid out as if it were an initializer for a global variable
820 of the prefix data's type. No padding is automatically placed between the
821 prefix data and the function body. If padding is required, it must be part
824 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
825 which encodes the ``nop`` instruction:
829 define void @f() prefix i8 144 { ... }
831 Generally prefix data can be formed by encoding a relative branch instruction
832 which skips the metadata, as in this example of valid prefix data for the
833 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
837 %0 = type <{ i8, i8, i8* }>
839 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
841 A function may have prefix data but no body. This has similar semantics
842 to the ``available_externally`` linkage in that the data may be used by the
843 optimizers but will not be emitted in the object file.
850 Attribute groups are groups of attributes that are referenced by objects within
851 the IR. They are important for keeping ``.ll`` files readable, because a lot of
852 functions will use the same set of attributes. In the degenerative case of a
853 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
854 group will capture the important command line flags used to build that file.
856 An attribute group is a module-level object. To use an attribute group, an
857 object references the attribute group's ID (e.g. ``#37``). An object may refer
858 to more than one attribute group. In that situation, the attributes from the
859 different groups are merged.
861 Here is an example of attribute groups for a function that should always be
862 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
866 ; Target-independent attributes:
867 attributes #0 = { alwaysinline alignstack=4 }
869 ; Target-dependent attributes:
870 attributes #1 = { "no-sse" }
872 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
873 define void @f() #0 #1 { ... }
880 Function attributes are set to communicate additional information about
881 a function. Function attributes are considered to be part of the
882 function, not of the function type, so functions with different function
883 attributes can have the same function type.
885 Function attributes are simple keywords that follow the type specified.
886 If multiple attributes are needed, they are space separated. For
891 define void @f() noinline { ... }
892 define void @f() alwaysinline { ... }
893 define void @f() alwaysinline optsize { ... }
894 define void @f() optsize { ... }
897 This attribute indicates that, when emitting the prologue and
898 epilogue, the backend should forcibly align the stack pointer.
899 Specify the desired alignment, which must be a power of two, in
902 This attribute indicates that the inliner should attempt to inline
903 this function into callers whenever possible, ignoring any active
904 inlining size threshold for this caller.
906 This indicates that the callee function at a call site should be
907 recognized as a built-in function, even though the function's declaration
908 uses the ``nobuiltin`` attribute. This is only valid at call sites for
909 direct calls to functions which are declared with the ``nobuiltin``
912 This attribute indicates that this function is rarely called. When
913 computing edge weights, basic blocks post-dominated by a cold
914 function call are also considered to be cold; and, thus, given low
917 This attribute indicates that the source code contained a hint that
918 inlining this function is desirable (such as the "inline" keyword in
919 C/C++). It is just a hint; it imposes no requirements on the
922 This attribute suggests that optimization passes and code generator
923 passes make choices that keep the code size of this function as small
924 as possible and perform optimizations that may sacrifice runtime
925 performance in order to minimize the size of the generated code.
927 This attribute disables prologue / epilogue emission for the
928 function. This can have very system-specific consequences.
930 This indicates that the callee function at a call site is not recognized as
931 a built-in function. LLVM will retain the original call and not replace it
932 with equivalent code based on the semantics of the built-in function, unless
933 the call site uses the ``builtin`` attribute. This is valid at call sites
934 and on function declarations and definitions.
936 This attribute indicates that calls to the function cannot be
937 duplicated. A call to a ``noduplicate`` function may be moved
938 within its parent function, but may not be duplicated within
941 A function containing a ``noduplicate`` call may still
942 be an inlining candidate, provided that the call is not
943 duplicated by inlining. That implies that the function has
944 internal linkage and only has one call site, so the original
945 call is dead after inlining.
947 This attributes disables implicit floating point instructions.
949 This attribute indicates that the inliner should never inline this
950 function in any situation. This attribute may not be used together
951 with the ``alwaysinline`` attribute.
953 This attribute suppresses lazy symbol binding for the function. This
954 may make calls to the function faster, at the cost of extra program
955 startup time if the function is not called during program startup.
957 This attribute indicates that the code generator should not use a
958 red zone, even if the target-specific ABI normally permits it.
960 This function attribute indicates that the function never returns
961 normally. This produces undefined behavior at runtime if the
962 function ever does dynamically return.
964 This function attribute indicates that the function never returns
965 with an unwind or exceptional control flow. If the function does
966 unwind, its runtime behavior is undefined.
968 This function attribute indicates that the function is not optimized
969 by any optimization or code generator passes with the
970 exception of interprocedural optimization passes.
971 This attribute cannot be used together with the ``alwaysinline``
972 attribute; this attribute is also incompatible
973 with the ``minsize`` attribute and the ``optsize`` attribute.
975 This attribute requires the ``noinline`` attribute to be specified on
976 the function as well, so the function is never inlined into any caller.
977 Only functions with the ``alwaysinline`` attribute are valid
978 candidates for inlining into the body of this function.
980 This attribute suggests that optimization passes and code generator
981 passes make choices that keep the code size of this function low,
982 and otherwise do optimizations specifically to reduce code size as
983 long as they do not significantly impact runtime performance.
985 On a function, this attribute indicates that the function computes its
986 result (or decides to unwind an exception) based strictly on its arguments,
987 without dereferencing any pointer arguments or otherwise accessing
988 any mutable state (e.g. memory, control registers, etc) visible to
989 caller functions. It does not write through any pointer arguments
990 (including ``byval`` arguments) and never changes any state visible
991 to callers. This means that it cannot unwind exceptions by calling
992 the ``C++`` exception throwing methods.
994 On an argument, this attribute indicates that the function does not
995 dereference that pointer argument, even though it may read or write the
996 memory that the pointer points to if accessed through other pointers.
998 On a function, this attribute indicates that the function does not write
999 through any pointer arguments (including ``byval`` arguments) or otherwise
1000 modify any state (e.g. memory, control registers, etc) visible to
1001 caller functions. It may dereference pointer arguments and read
1002 state that may be set in the caller. A readonly function always
1003 returns the same value (or unwinds an exception identically) when
1004 called with the same set of arguments and global state. It cannot
1005 unwind an exception by calling the ``C++`` exception throwing
1008 On an argument, this attribute indicates that the function does not write
1009 through this pointer argument, even though it may write to the memory that
1010 the pointer points to.
1012 This attribute indicates that this function can return twice. The C
1013 ``setjmp`` is an example of such a function. The compiler disables
1014 some optimizations (like tail calls) in the caller of these
1016 ``sanitize_address``
1017 This attribute indicates that AddressSanitizer checks
1018 (dynamic address safety analysis) are enabled for this function.
1020 This attribute indicates that MemorySanitizer checks (dynamic detection
1021 of accesses to uninitialized memory) are enabled for this function.
1023 This attribute indicates that ThreadSanitizer checks
1024 (dynamic thread safety analysis) are enabled for this function.
1026 This attribute indicates that the function should emit a stack
1027 smashing protector. It is in the form of a "canary" --- a random value
1028 placed on the stack before the local variables that's checked upon
1029 return from the function to see if it has been overwritten. A
1030 heuristic is used to determine if a function needs stack protectors
1031 or not. The heuristic used will enable protectors for functions with:
1033 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1034 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1035 - Calls to alloca() with variable sizes or constant sizes greater than
1036 ``ssp-buffer-size``.
1038 If a function that has an ``ssp`` attribute is inlined into a
1039 function that doesn't have an ``ssp`` attribute, then the resulting
1040 function will have an ``ssp`` attribute.
1042 This attribute indicates that the function should *always* emit a
1043 stack smashing protector. This overrides the ``ssp`` function
1046 If a function that has an ``sspreq`` attribute is inlined into a
1047 function that doesn't have an ``sspreq`` attribute or which has an
1048 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1049 an ``sspreq`` attribute.
1051 This attribute indicates that the function should emit a stack smashing
1052 protector. This attribute causes a strong heuristic to be used when
1053 determining if a function needs stack protectors. The strong heuristic
1054 will enable protectors for functions with:
1056 - Arrays of any size and type
1057 - Aggregates containing an array of any size and type.
1058 - Calls to alloca().
1059 - Local variables that have had their address taken.
1061 This overrides the ``ssp`` function attribute.
1063 If a function that has an ``sspstrong`` attribute is inlined into a
1064 function that doesn't have an ``sspstrong`` attribute, then the
1065 resulting function will have an ``sspstrong`` attribute.
1067 This attribute indicates that the ABI being targeted requires that
1068 an unwind table entry be produce for this function even if we can
1069 show that no exceptions passes by it. This is normally the case for
1070 the ELF x86-64 abi, but it can be disabled for some compilation
1075 Module-Level Inline Assembly
1076 ----------------------------
1078 Modules may contain "module-level inline asm" blocks, which corresponds
1079 to the GCC "file scope inline asm" blocks. These blocks are internally
1080 concatenated by LLVM and treated as a single unit, but may be separated
1081 in the ``.ll`` file if desired. The syntax is very simple:
1083 .. code-block:: llvm
1085 module asm "inline asm code goes here"
1086 module asm "more can go here"
1088 The strings can contain any character by escaping non-printable
1089 characters. The escape sequence used is simply "\\xx" where "xx" is the
1090 two digit hex code for the number.
1092 The inline asm code is simply printed to the machine code .s file when
1093 assembly code is generated.
1095 .. _langref_datalayout:
1100 A module may specify a target specific data layout string that specifies
1101 how data is to be laid out in memory. The syntax for the data layout is
1104 .. code-block:: llvm
1106 target datalayout = "layout specification"
1108 The *layout specification* consists of a list of specifications
1109 separated by the minus sign character ('-'). Each specification starts
1110 with a letter and may include other information after the letter to
1111 define some aspect of the data layout. The specifications accepted are
1115 Specifies that the target lays out data in big-endian form. That is,
1116 the bits with the most significance have the lowest address
1119 Specifies that the target lays out data in little-endian form. That
1120 is, the bits with the least significance have the lowest address
1123 Specifies the natural alignment of the stack in bits. Alignment
1124 promotion of stack variables is limited to the natural stack
1125 alignment to avoid dynamic stack realignment. The stack alignment
1126 must be a multiple of 8-bits. If omitted, the natural stack
1127 alignment defaults to "unspecified", which does not prevent any
1128 alignment promotions.
1129 ``p[n]:<size>:<abi>:<pref>``
1130 This specifies the *size* of a pointer and its ``<abi>`` and
1131 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1132 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1133 preceding ``:`` should be omitted too. The address space, ``n`` is
1134 optional, and if not specified, denotes the default address space 0.
1135 The value of ``n`` must be in the range [1,2^23).
1136 ``i<size>:<abi>:<pref>``
1137 This specifies the alignment for an integer type of a given bit
1138 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1139 ``v<size>:<abi>:<pref>``
1140 This specifies the alignment for a vector type of a given bit
1142 ``f<size>:<abi>:<pref>``
1143 This specifies the alignment for a floating point type of a given bit
1144 ``<size>``. Only values of ``<size>`` that are supported by the target
1145 will work. 32 (float) and 64 (double) are supported on all targets; 80
1146 or 128 (different flavors of long double) are also supported on some
1148 ``a<size>:<abi>:<pref>``
1149 This specifies the alignment for an aggregate type of a given bit
1151 ``s<size>:<abi>:<pref>``
1152 This specifies the alignment for a stack object of a given bit
1154 ``n<size1>:<size2>:<size3>...``
1155 This specifies a set of native integer widths for the target CPU in
1156 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1157 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1158 this set are considered to support most general arithmetic operations
1161 When constructing the data layout for a given target, LLVM starts with a
1162 default set of specifications which are then (possibly) overridden by
1163 the specifications in the ``datalayout`` keyword. The default
1164 specifications are given in this list:
1166 - ``E`` - big endian
1167 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1168 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1169 same as the default address space.
1170 - ``S0`` - natural stack alignment is unspecified
1171 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1172 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1173 - ``i16:16:16`` - i16 is 16-bit aligned
1174 - ``i32:32:32`` - i32 is 32-bit aligned
1175 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1176 alignment of 64-bits
1177 - ``f16:16:16`` - half is 16-bit aligned
1178 - ``f32:32:32`` - float is 32-bit aligned
1179 - ``f64:64:64`` - double is 64-bit aligned
1180 - ``f128:128:128`` - quad is 128-bit aligned
1181 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1182 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1183 - ``a:0:64`` - aggregates are 64-bit aligned
1185 When LLVM is determining the alignment for a given type, it uses the
1188 #. If the type sought is an exact match for one of the specifications,
1189 that specification is used.
1190 #. If no match is found, and the type sought is an integer type, then
1191 the smallest integer type that is larger than the bitwidth of the
1192 sought type is used. If none of the specifications are larger than
1193 the bitwidth then the largest integer type is used. For example,
1194 given the default specifications above, the i7 type will use the
1195 alignment of i8 (next largest) while both i65 and i256 will use the
1196 alignment of i64 (largest specified).
1197 #. If no match is found, and the type sought is a vector type, then the
1198 largest vector type that is smaller than the sought vector type will
1199 be used as a fall back. This happens because <128 x double> can be
1200 implemented in terms of 64 <2 x double>, for example.
1202 The function of the data layout string may not be what you expect.
1203 Notably, this is not a specification from the frontend of what alignment
1204 the code generator should use.
1206 Instead, if specified, the target data layout is required to match what
1207 the ultimate *code generator* expects. This string is used by the
1208 mid-level optimizers to improve code, and this only works if it matches
1209 what the ultimate code generator uses. If you would like to generate IR
1210 that does not embed this target-specific detail into the IR, then you
1211 don't have to specify the string. This will disable some optimizations
1212 that require precise layout information, but this also prevents those
1213 optimizations from introducing target specificity into the IR.
1220 A module may specify a target triple string that describes the target
1221 host. The syntax for the target triple is simply:
1223 .. code-block:: llvm
1225 target triple = "x86_64-apple-macosx10.7.0"
1227 The *target triple* string consists of a series of identifiers delimited
1228 by the minus sign character ('-'). The canonical forms are:
1232 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1233 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1235 This information is passed along to the backend so that it generates
1236 code for the proper architecture. It's possible to override this on the
1237 command line with the ``-mtriple`` command line option.
1239 .. _pointeraliasing:
1241 Pointer Aliasing Rules
1242 ----------------------
1244 Any memory access must be done through a pointer value associated with
1245 an address range of the memory access, otherwise the behavior is
1246 undefined. Pointer values are associated with address ranges according
1247 to the following rules:
1249 - A pointer value is associated with the addresses associated with any
1250 value it is *based* on.
1251 - An address of a global variable is associated with the address range
1252 of the variable's storage.
1253 - The result value of an allocation instruction is associated with the
1254 address range of the allocated storage.
1255 - A null pointer in the default address-space is associated with no
1257 - An integer constant other than zero or a pointer value returned from
1258 a function not defined within LLVM may be associated with address
1259 ranges allocated through mechanisms other than those provided by
1260 LLVM. Such ranges shall not overlap with any ranges of addresses
1261 allocated by mechanisms provided by LLVM.
1263 A pointer value is *based* on another pointer value according to the
1266 - A pointer value formed from a ``getelementptr`` operation is *based*
1267 on the first operand of the ``getelementptr``.
1268 - The result value of a ``bitcast`` is *based* on the operand of the
1270 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1271 values that contribute (directly or indirectly) to the computation of
1272 the pointer's value.
1273 - The "*based* on" relationship is transitive.
1275 Note that this definition of *"based"* is intentionally similar to the
1276 definition of *"based"* in C99, though it is slightly weaker.
1278 LLVM IR does not associate types with memory. The result type of a
1279 ``load`` merely indicates the size and alignment of the memory from
1280 which to load, as well as the interpretation of the value. The first
1281 operand type of a ``store`` similarly only indicates the size and
1282 alignment of the store.
1284 Consequently, type-based alias analysis, aka TBAA, aka
1285 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1286 :ref:`Metadata <metadata>` may be used to encode additional information
1287 which specialized optimization passes may use to implement type-based
1292 Volatile Memory Accesses
1293 ------------------------
1295 Certain memory accesses, such as :ref:`load <i_load>`'s,
1296 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1297 marked ``volatile``. The optimizers must not change the number of
1298 volatile operations or change their order of execution relative to other
1299 volatile operations. The optimizers *may* change the order of volatile
1300 operations relative to non-volatile operations. This is not Java's
1301 "volatile" and has no cross-thread synchronization behavior.
1303 IR-level volatile loads and stores cannot safely be optimized into
1304 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1305 flagged volatile. Likewise, the backend should never split or merge
1306 target-legal volatile load/store instructions.
1308 .. admonition:: Rationale
1310 Platforms may rely on volatile loads and stores of natively supported
1311 data width to be executed as single instruction. For example, in C
1312 this holds for an l-value of volatile primitive type with native
1313 hardware support, but not necessarily for aggregate types. The
1314 frontend upholds these expectations, which are intentionally
1315 unspecified in the IR. The rules above ensure that IR transformation
1316 do not violate the frontend's contract with the language.
1320 Memory Model for Concurrent Operations
1321 --------------------------------------
1323 The LLVM IR does not define any way to start parallel threads of
1324 execution or to register signal handlers. Nonetheless, there are
1325 platform-specific ways to create them, and we define LLVM IR's behavior
1326 in their presence. This model is inspired by the C++0x memory model.
1328 For a more informal introduction to this model, see the :doc:`Atomics`.
1330 We define a *happens-before* partial order as the least partial order
1333 - Is a superset of single-thread program order, and
1334 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1335 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1336 techniques, like pthread locks, thread creation, thread joining,
1337 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1338 Constraints <ordering>`).
1340 Note that program order does not introduce *happens-before* edges
1341 between a thread and signals executing inside that thread.
1343 Every (defined) read operation (load instructions, memcpy, atomic
1344 loads/read-modify-writes, etc.) R reads a series of bytes written by
1345 (defined) write operations (store instructions, atomic
1346 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1347 section, initialized globals are considered to have a write of the
1348 initializer which is atomic and happens before any other read or write
1349 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1350 may see any write to the same byte, except:
1352 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1353 write\ :sub:`2` happens before R\ :sub:`byte`, then
1354 R\ :sub:`byte` does not see write\ :sub:`1`.
1355 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1356 R\ :sub:`byte` does not see write\ :sub:`3`.
1358 Given that definition, R\ :sub:`byte` is defined as follows:
1360 - If R is volatile, the result is target-dependent. (Volatile is
1361 supposed to give guarantees which can support ``sig_atomic_t`` in
1362 C/C++, and may be used for accesses to addresses which do not behave
1363 like normal memory. It does not generally provide cross-thread
1365 - Otherwise, if there is no write to the same byte that happens before
1366 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1367 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1368 R\ :sub:`byte` returns the value written by that write.
1369 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1370 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1371 Memory Ordering Constraints <ordering>` section for additional
1372 constraints on how the choice is made.
1373 - Otherwise R\ :sub:`byte` returns ``undef``.
1375 R returns the value composed of the series of bytes it read. This
1376 implies that some bytes within the value may be ``undef`` **without**
1377 the entire value being ``undef``. Note that this only defines the
1378 semantics of the operation; it doesn't mean that targets will emit more
1379 than one instruction to read the series of bytes.
1381 Note that in cases where none of the atomic intrinsics are used, this
1382 model places only one restriction on IR transformations on top of what
1383 is required for single-threaded execution: introducing a store to a byte
1384 which might not otherwise be stored is not allowed in general.
1385 (Specifically, in the case where another thread might write to and read
1386 from an address, introducing a store can change a load that may see
1387 exactly one write into a load that may see multiple writes.)
1391 Atomic Memory Ordering Constraints
1392 ----------------------------------
1394 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1395 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1396 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1397 an ordering parameter that determines which other atomic instructions on
1398 the same address they *synchronize with*. These semantics are borrowed
1399 from Java and C++0x, but are somewhat more colloquial. If these
1400 descriptions aren't precise enough, check those specs (see spec
1401 references in the :doc:`atomics guide <Atomics>`).
1402 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1403 differently since they don't take an address. See that instruction's
1404 documentation for details.
1406 For a simpler introduction to the ordering constraints, see the
1410 The set of values that can be read is governed by the happens-before
1411 partial order. A value cannot be read unless some operation wrote
1412 it. This is intended to provide a guarantee strong enough to model
1413 Java's non-volatile shared variables. This ordering cannot be
1414 specified for read-modify-write operations; it is not strong enough
1415 to make them atomic in any interesting way.
1417 In addition to the guarantees of ``unordered``, there is a single
1418 total order for modifications by ``monotonic`` operations on each
1419 address. All modification orders must be compatible with the
1420 happens-before order. There is no guarantee that the modification
1421 orders can be combined to a global total order for the whole program
1422 (and this often will not be possible). The read in an atomic
1423 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1424 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1425 order immediately before the value it writes. If one atomic read
1426 happens before another atomic read of the same address, the later
1427 read must see the same value or a later value in the address's
1428 modification order. This disallows reordering of ``monotonic`` (or
1429 stronger) operations on the same address. If an address is written
1430 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1431 read that address repeatedly, the other threads must eventually see
1432 the write. This corresponds to the C++0x/C1x
1433 ``memory_order_relaxed``.
1435 In addition to the guarantees of ``monotonic``, a
1436 *synchronizes-with* edge may be formed with a ``release`` operation.
1437 This is intended to model C++'s ``memory_order_acquire``.
1439 In addition to the guarantees of ``monotonic``, if this operation
1440 writes a value which is subsequently read by an ``acquire``
1441 operation, it *synchronizes-with* that operation. (This isn't a
1442 complete description; see the C++0x definition of a release
1443 sequence.) This corresponds to the C++0x/C1x
1444 ``memory_order_release``.
1445 ``acq_rel`` (acquire+release)
1446 Acts as both an ``acquire`` and ``release`` operation on its
1447 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1448 ``seq_cst`` (sequentially consistent)
1449 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1450 operation which only reads, ``release`` for an operation which only
1451 writes), there is a global total order on all
1452 sequentially-consistent operations on all addresses, which is
1453 consistent with the *happens-before* partial order and with the
1454 modification orders of all the affected addresses. Each
1455 sequentially-consistent read sees the last preceding write to the
1456 same address in this global order. This corresponds to the C++0x/C1x
1457 ``memory_order_seq_cst`` and Java volatile.
1461 If an atomic operation is marked ``singlethread``, it only *synchronizes
1462 with* or participates in modification and seq\_cst total orderings with
1463 other operations running in the same thread (for example, in signal
1471 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1472 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1473 :ref:`frem <i_frem>`) have the following flags that can set to enable
1474 otherwise unsafe floating point operations
1477 No NaNs - Allow optimizations to assume the arguments and result are not
1478 NaN. Such optimizations are required to retain defined behavior over
1479 NaNs, but the value of the result is undefined.
1482 No Infs - Allow optimizations to assume the arguments and result are not
1483 +/-Inf. Such optimizations are required to retain defined behavior over
1484 +/-Inf, but the value of the result is undefined.
1487 No Signed Zeros - Allow optimizations to treat the sign of a zero
1488 argument or result as insignificant.
1491 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1492 argument rather than perform division.
1495 Fast - Allow algebraically equivalent transformations that may
1496 dramatically change results in floating point (e.g. reassociate). This
1497 flag implies all the others.
1504 The LLVM type system is one of the most important features of the
1505 intermediate representation. Being typed enables a number of
1506 optimizations to be performed on the intermediate representation
1507 directly, without having to do extra analyses on the side before the
1508 transformation. A strong type system makes it easier to read the
1509 generated code and enables novel analyses and transformations that are
1510 not feasible to perform on normal three address code representations.
1520 The void type does not represent any value and has no size.
1538 The function type can be thought of as a function signature. It consists of a
1539 return type and a list of formal parameter types. The return type of a function
1540 type is a void type or first class type --- except for :ref:`label <t_label>`
1541 and :ref:`metadata <t_metadata>` types.
1547 <returntype> (<parameter list>)
1549 ...where '``<parameter list>``' is a comma-separated list of type
1550 specifiers. Optionally, the parameter list may include a type ``...``, which
1551 indicates that the function takes a variable number of arguments. Variable
1552 argument functions can access their arguments with the :ref:`variable argument
1553 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1554 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1558 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1559 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1560 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1561 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1562 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1563 | ``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. |
1564 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1565 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1566 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1573 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1574 Values of these types are the only ones which can be produced by
1582 These are the types that are valid in registers from CodeGen's perspective.
1591 The integer type is a very simple type that simply specifies an
1592 arbitrary bit width for the integer type desired. Any bit width from 1
1593 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1601 The number of bits the integer will occupy is specified by the ``N``
1607 +----------------+------------------------------------------------+
1608 | ``i1`` | a single-bit integer. |
1609 +----------------+------------------------------------------------+
1610 | ``i32`` | a 32-bit integer. |
1611 +----------------+------------------------------------------------+
1612 | ``i1942652`` | a really big integer of over 1 million bits. |
1613 +----------------+------------------------------------------------+
1617 Floating Point Types
1618 """"""""""""""""""""
1627 - 16-bit floating point value
1630 - 32-bit floating point value
1633 - 64-bit floating point value
1636 - 128-bit floating point value (112-bit mantissa)
1639 - 80-bit floating point value (X87)
1642 - 128-bit floating point value (two 64-bits)
1651 The x86mmx type represents a value held in an MMX register on an x86
1652 machine. The operations allowed on it are quite limited: parameters and
1653 return values, load and store, and bitcast. User-specified MMX
1654 instructions are represented as intrinsic or asm calls with arguments
1655 and/or results of this type. There are no arrays, vectors or constants
1672 The pointer type is used to specify memory locations. Pointers are
1673 commonly used to reference objects in memory.
1675 Pointer types may have an optional address space attribute defining the
1676 numbered address space where the pointed-to object resides. The default
1677 address space is number zero. The semantics of non-zero address spaces
1678 are target-specific.
1680 Note that LLVM does not permit pointers to void (``void*``) nor does it
1681 permit pointers to labels (``label*``). Use ``i8*`` instead.
1691 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1692 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1693 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1694 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1695 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1696 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1697 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1706 A vector type is a simple derived type that represents a vector of
1707 elements. Vector types are used when multiple primitive data are
1708 operated in parallel using a single instruction (SIMD). A vector type
1709 requires a size (number of elements) and an underlying primitive data
1710 type. Vector types are considered :ref:`first class <t_firstclass>`.
1716 < <# elements> x <elementtype> >
1718 The number of elements is a constant integer value larger than 0;
1719 elementtype may be any integer or floating point type, or a pointer to
1720 these types. Vectors of size zero are not allowed.
1724 +-------------------+--------------------------------------------------+
1725 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1726 +-------------------+--------------------------------------------------+
1727 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1728 +-------------------+--------------------------------------------------+
1729 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1730 +-------------------+--------------------------------------------------+
1731 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1732 +-------------------+--------------------------------------------------+
1741 The label type represents code labels.
1756 The metadata type represents embedded metadata. No derived types may be
1757 created from metadata except for :ref:`function <t_function>` arguments.
1770 Aggregate Types are a subset of derived types that can contain multiple
1771 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1772 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1782 The array type is a very simple derived type that arranges elements
1783 sequentially in memory. The array type requires a size (number of
1784 elements) and an underlying data type.
1790 [<# elements> x <elementtype>]
1792 The number of elements is a constant integer value; ``elementtype`` may
1793 be any type with a size.
1797 +------------------+--------------------------------------+
1798 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1799 +------------------+--------------------------------------+
1800 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1801 +------------------+--------------------------------------+
1802 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1803 +------------------+--------------------------------------+
1805 Here are some examples of multidimensional arrays:
1807 +-----------------------------+----------------------------------------------------------+
1808 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1809 +-----------------------------+----------------------------------------------------------+
1810 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1811 +-----------------------------+----------------------------------------------------------+
1812 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1813 +-----------------------------+----------------------------------------------------------+
1815 There is no restriction on indexing beyond the end of the array implied
1816 by a static type (though there are restrictions on indexing beyond the
1817 bounds of an allocated object in some cases). This means that
1818 single-dimension 'variable sized array' addressing can be implemented in
1819 LLVM with a zero length array type. An implementation of 'pascal style
1820 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1830 The structure type is used to represent a collection of data members
1831 together in memory. The elements of a structure may be any type that has
1834 Structures in memory are accessed using '``load``' and '``store``' by
1835 getting a pointer to a field with the '``getelementptr``' instruction.
1836 Structures in registers are accessed using the '``extractvalue``' and
1837 '``insertvalue``' instructions.
1839 Structures may optionally be "packed" structures, which indicate that
1840 the alignment of the struct is one byte, and that there is no padding
1841 between the elements. In non-packed structs, padding between field types
1842 is inserted as defined by the DataLayout string in the module, which is
1843 required to match what the underlying code generator expects.
1845 Structures can either be "literal" or "identified". A literal structure
1846 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1847 identified types are always defined at the top level with a name.
1848 Literal types are uniqued by their contents and can never be recursive
1849 or opaque since there is no way to write one. Identified types can be
1850 recursive, can be opaqued, and are never uniqued.
1856 %T1 = type { <type list> } ; Identified normal struct type
1857 %T2 = type <{ <type list> }> ; Identified packed struct type
1861 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1862 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1863 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1864 | ``{ 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``. |
1865 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1866 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1867 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1871 Opaque Structure Types
1872 """"""""""""""""""""""
1876 Opaque structure types are used to represent named structure types that
1877 do not have a body specified. This corresponds (for example) to the C
1878 notion of a forward declared structure.
1889 +--------------+-------------------+
1890 | ``opaque`` | An opaque type. |
1891 +--------------+-------------------+
1896 LLVM has several different basic types of constants. This section
1897 describes them all and their syntax.
1902 **Boolean constants**
1903 The two strings '``true``' and '``false``' are both valid constants
1905 **Integer constants**
1906 Standard integers (such as '4') are constants of the
1907 :ref:`integer <t_integer>` type. Negative numbers may be used with
1909 **Floating point constants**
1910 Floating point constants use standard decimal notation (e.g.
1911 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1912 hexadecimal notation (see below). The assembler requires the exact
1913 decimal value of a floating-point constant. For example, the
1914 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1915 decimal in binary. Floating point constants must have a :ref:`floating
1916 point <t_floating>` type.
1917 **Null pointer constants**
1918 The identifier '``null``' is recognized as a null pointer constant
1919 and must be of :ref:`pointer type <t_pointer>`.
1921 The one non-intuitive notation for constants is the hexadecimal form of
1922 floating point constants. For example, the form
1923 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1924 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1925 constants are required (and the only time that they are generated by the
1926 disassembler) is when a floating point constant must be emitted but it
1927 cannot be represented as a decimal floating point number in a reasonable
1928 number of digits. For example, NaN's, infinities, and other special
1929 values are represented in their IEEE hexadecimal format so that assembly
1930 and disassembly do not cause any bits to change in the constants.
1932 When using the hexadecimal form, constants of types half, float, and
1933 double are represented using the 16-digit form shown above (which
1934 matches the IEEE754 representation for double); half and float values
1935 must, however, be exactly representable as IEEE 754 half and single
1936 precision, respectively. Hexadecimal format is always used for long
1937 double, and there are three forms of long double. The 80-bit format used
1938 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1939 128-bit format used by PowerPC (two adjacent doubles) is represented by
1940 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1941 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1942 will only work if they match the long double format on your target.
1943 The IEEE 16-bit format (half precision) is represented by ``0xH``
1944 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1945 (sign bit at the left).
1947 There are no constants of type x86mmx.
1949 .. _complexconstants:
1954 Complex constants are a (potentially recursive) combination of simple
1955 constants and smaller complex constants.
1957 **Structure constants**
1958 Structure constants are represented with notation similar to
1959 structure type definitions (a comma separated list of elements,
1960 surrounded by braces (``{}``)). For example:
1961 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1962 "``@G = external global i32``". Structure constants must have
1963 :ref:`structure type <t_struct>`, and the number and types of elements
1964 must match those specified by the type.
1966 Array constants are represented with notation similar to array type
1967 definitions (a comma separated list of elements, surrounded by
1968 square brackets (``[]``)). For example:
1969 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1970 :ref:`array type <t_array>`, and the number and types of elements must
1971 match those specified by the type.
1972 **Vector constants**
1973 Vector constants are represented with notation similar to vector
1974 type definitions (a comma separated list of elements, surrounded by
1975 less-than/greater-than's (``<>``)). For example:
1976 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1977 must have :ref:`vector type <t_vector>`, and the number and types of
1978 elements must match those specified by the type.
1979 **Zero initialization**
1980 The string '``zeroinitializer``' can be used to zero initialize a
1981 value to zero of *any* type, including scalar and
1982 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1983 having to print large zero initializers (e.g. for large arrays) and
1984 is always exactly equivalent to using explicit zero initializers.
1986 A metadata node is a structure-like constant with :ref:`metadata
1987 type <t_metadata>`. For example:
1988 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1989 constants that are meant to be interpreted as part of the
1990 instruction stream, metadata is a place to attach additional
1991 information such as debug info.
1993 Global Variable and Function Addresses
1994 --------------------------------------
1996 The addresses of :ref:`global variables <globalvars>` and
1997 :ref:`functions <functionstructure>` are always implicitly valid
1998 (link-time) constants. These constants are explicitly referenced when
1999 the :ref:`identifier for the global <identifiers>` is used and always have
2000 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2003 .. code-block:: llvm
2007 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2014 The string '``undef``' can be used anywhere a constant is expected, and
2015 indicates that the user of the value may receive an unspecified
2016 bit-pattern. Undefined values may be of any type (other than '``label``'
2017 or '``void``') and be used anywhere a constant is permitted.
2019 Undefined values are useful because they indicate to the compiler that
2020 the program is well defined no matter what value is used. This gives the
2021 compiler more freedom to optimize. Here are some examples of
2022 (potentially surprising) transformations that are valid (in pseudo IR):
2024 .. code-block:: llvm
2034 This is safe because all of the output bits are affected by the undef
2035 bits. Any output bit can have a zero or one depending on the input bits.
2037 .. code-block:: llvm
2048 These logical operations have bits that are not always affected by the
2049 input. For example, if ``%X`` has a zero bit, then the output of the
2050 '``and``' operation will always be a zero for that bit, no matter what
2051 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2052 optimize or assume that the result of the '``and``' is '``undef``'.
2053 However, it is safe to assume that all bits of the '``undef``' could be
2054 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2055 all the bits of the '``undef``' operand to the '``or``' could be set,
2056 allowing the '``or``' to be folded to -1.
2058 .. code-block:: llvm
2060 %A = select undef, %X, %Y
2061 %B = select undef, 42, %Y
2062 %C = select %X, %Y, undef
2072 This set of examples shows that undefined '``select``' (and conditional
2073 branch) conditions can go *either way*, but they have to come from one
2074 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2075 both known to have a clear low bit, then ``%A`` would have to have a
2076 cleared low bit. However, in the ``%C`` example, the optimizer is
2077 allowed to assume that the '``undef``' operand could be the same as
2078 ``%Y``, allowing the whole '``select``' to be eliminated.
2080 .. code-block:: llvm
2082 %A = xor undef, undef
2099 This example points out that two '``undef``' operands are not
2100 necessarily the same. This can be surprising to people (and also matches
2101 C semantics) where they assume that "``X^X``" is always zero, even if
2102 ``X`` is undefined. This isn't true for a number of reasons, but the
2103 short answer is that an '``undef``' "variable" can arbitrarily change
2104 its value over its "live range". This is true because the variable
2105 doesn't actually *have a live range*. Instead, the value is logically
2106 read from arbitrary registers that happen to be around when needed, so
2107 the value is not necessarily consistent over time. In fact, ``%A`` and
2108 ``%C`` need to have the same semantics or the core LLVM "replace all
2109 uses with" concept would not hold.
2111 .. code-block:: llvm
2119 These examples show the crucial difference between an *undefined value*
2120 and *undefined behavior*. An undefined value (like '``undef``') is
2121 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2122 operation can be constant folded to '``undef``', because the '``undef``'
2123 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2124 However, in the second example, we can make a more aggressive
2125 assumption: because the ``undef`` is allowed to be an arbitrary value,
2126 we are allowed to assume that it could be zero. Since a divide by zero
2127 has *undefined behavior*, we are allowed to assume that the operation
2128 does not execute at all. This allows us to delete the divide and all
2129 code after it. Because the undefined operation "can't happen", the
2130 optimizer can assume that it occurs in dead code.
2132 .. code-block:: llvm
2134 a: store undef -> %X
2135 b: store %X -> undef
2140 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2141 value can be assumed to not have any effect; we can assume that the
2142 value is overwritten with bits that happen to match what was already
2143 there. However, a store *to* an undefined location could clobber
2144 arbitrary memory, therefore, it has undefined behavior.
2151 Poison values are similar to :ref:`undef values <undefvalues>`, however
2152 they also represent the fact that an instruction or constant expression
2153 which cannot evoke side effects has nevertheless detected a condition
2154 which results in undefined behavior.
2156 There is currently no way of representing a poison value in the IR; they
2157 only exist when produced by operations such as :ref:`add <i_add>` with
2160 Poison value behavior is defined in terms of value *dependence*:
2162 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2163 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2164 their dynamic predecessor basic block.
2165 - Function arguments depend on the corresponding actual argument values
2166 in the dynamic callers of their functions.
2167 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2168 instructions that dynamically transfer control back to them.
2169 - :ref:`Invoke <i_invoke>` instructions depend on the
2170 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2171 call instructions that dynamically transfer control back to them.
2172 - Non-volatile loads and stores depend on the most recent stores to all
2173 of the referenced memory addresses, following the order in the IR
2174 (including loads and stores implied by intrinsics such as
2175 :ref:`@llvm.memcpy <int_memcpy>`.)
2176 - An instruction with externally visible side effects depends on the
2177 most recent preceding instruction with externally visible side
2178 effects, following the order in the IR. (This includes :ref:`volatile
2179 operations <volatile>`.)
2180 - An instruction *control-depends* on a :ref:`terminator
2181 instruction <terminators>` if the terminator instruction has
2182 multiple successors and the instruction is always executed when
2183 control transfers to one of the successors, and may not be executed
2184 when control is transferred to another.
2185 - Additionally, an instruction also *control-depends* on a terminator
2186 instruction if the set of instructions it otherwise depends on would
2187 be different if the terminator had transferred control to a different
2189 - Dependence is transitive.
2191 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2192 with the additional affect that any instruction which has a *dependence*
2193 on a poison value has undefined behavior.
2195 Here are some examples:
2197 .. code-block:: llvm
2200 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2201 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2202 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2203 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2205 store i32 %poison, i32* @g ; Poison value stored to memory.
2206 %poison2 = load i32* @g ; Poison value loaded back from memory.
2208 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2210 %narrowaddr = bitcast i32* @g to i16*
2211 %wideaddr = bitcast i32* @g to i64*
2212 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2213 %poison4 = load i64* %wideaddr ; Returns a poison value.
2215 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2216 br i1 %cmp, label %true, label %end ; Branch to either destination.
2219 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2220 ; it has undefined behavior.
2224 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2225 ; Both edges into this PHI are
2226 ; control-dependent on %cmp, so this
2227 ; always results in a poison value.
2229 store volatile i32 0, i32* @g ; This would depend on the store in %true
2230 ; if %cmp is true, or the store in %entry
2231 ; otherwise, so this is undefined behavior.
2233 br i1 %cmp, label %second_true, label %second_end
2234 ; The same branch again, but this time the
2235 ; true block doesn't have side effects.
2242 store volatile i32 0, i32* @g ; This time, the instruction always depends
2243 ; on the store in %end. Also, it is
2244 ; control-equivalent to %end, so this is
2245 ; well-defined (ignoring earlier undefined
2246 ; behavior in this example).
2250 Addresses of Basic Blocks
2251 -------------------------
2253 ``blockaddress(@function, %block)``
2255 The '``blockaddress``' constant computes the address of the specified
2256 basic block in the specified function, and always has an ``i8*`` type.
2257 Taking the address of the entry block is illegal.
2259 This value only has defined behavior when used as an operand to the
2260 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2261 against null. Pointer equality tests between labels addresses results in
2262 undefined behavior --- though, again, comparison against null is ok, and
2263 no label is equal to the null pointer. This may be passed around as an
2264 opaque pointer sized value as long as the bits are not inspected. This
2265 allows ``ptrtoint`` and arithmetic to be performed on these values so
2266 long as the original value is reconstituted before the ``indirectbr``
2269 Finally, some targets may provide defined semantics when using the value
2270 as the operand to an inline assembly, but that is target specific.
2274 Constant Expressions
2275 --------------------
2277 Constant expressions are used to allow expressions involving other
2278 constants to be used as constants. Constant expressions may be of any
2279 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2280 that does not have side effects (e.g. load and call are not supported).
2281 The following is the syntax for constant expressions:
2283 ``trunc (CST to TYPE)``
2284 Truncate a constant to another type. The bit size of CST must be
2285 larger than the bit size of TYPE. Both types must be integers.
2286 ``zext (CST to TYPE)``
2287 Zero extend a constant to another type. The bit size of CST must be
2288 smaller than the bit size of TYPE. Both types must be integers.
2289 ``sext (CST to TYPE)``
2290 Sign extend a constant to another type. The bit size of CST must be
2291 smaller than the bit size of TYPE. Both types must be integers.
2292 ``fptrunc (CST to TYPE)``
2293 Truncate a floating point constant to another floating point type.
2294 The size of CST must be larger than the size of TYPE. Both types
2295 must be floating point.
2296 ``fpext (CST to TYPE)``
2297 Floating point extend a constant to another type. The size of CST
2298 must be smaller or equal to the size of TYPE. Both types must be
2300 ``fptoui (CST to TYPE)``
2301 Convert a floating point constant to the corresponding unsigned
2302 integer constant. TYPE must be a scalar or vector integer type. CST
2303 must be of scalar or vector floating point type. Both CST and TYPE
2304 must be scalars, or vectors of the same number of elements. If the
2305 value won't fit in the integer type, the results are undefined.
2306 ``fptosi (CST to TYPE)``
2307 Convert a floating point constant to the corresponding signed
2308 integer constant. TYPE must be a scalar or vector integer type. CST
2309 must be of scalar or vector floating point type. Both CST and TYPE
2310 must be scalars, or vectors of the same number of elements. If the
2311 value won't fit in the integer type, the results are undefined.
2312 ``uitofp (CST to TYPE)``
2313 Convert an unsigned integer constant to the corresponding floating
2314 point constant. TYPE must be a scalar or vector floating point type.
2315 CST must be of scalar or vector integer type. Both CST and TYPE must
2316 be scalars, or vectors of the same number of elements. If the value
2317 won't fit in the floating point type, the results are undefined.
2318 ``sitofp (CST to TYPE)``
2319 Convert a signed integer constant to the corresponding floating
2320 point constant. TYPE must be a scalar or vector floating point type.
2321 CST must be of scalar or vector integer type. Both CST and TYPE must
2322 be scalars, or vectors of the same number of elements. If the value
2323 won't fit in the floating point type, the results are undefined.
2324 ``ptrtoint (CST to TYPE)``
2325 Convert a pointer typed constant to the corresponding integer
2326 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2327 pointer type. The ``CST`` value is zero extended, truncated, or
2328 unchanged to make it fit in ``TYPE``.
2329 ``inttoptr (CST to TYPE)``
2330 Convert an integer constant to a pointer constant. TYPE must be a
2331 pointer type. CST must be of integer type. The CST value is zero
2332 extended, truncated, or unchanged to make it fit in a pointer size.
2333 This one is *really* dangerous!
2334 ``bitcast (CST to TYPE)``
2335 Convert a constant, CST, to another TYPE. The constraints of the
2336 operands are the same as those for the :ref:`bitcast
2337 instruction <i_bitcast>`.
2338 ``addrspacecast (CST to TYPE)``
2339 Convert a constant pointer or constant vector of pointer, CST, to another
2340 TYPE in a different address space. The constraints of the operands are the
2341 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2342 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2343 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2344 constants. As with the :ref:`getelementptr <i_getelementptr>`
2345 instruction, the index list may have zero or more indexes, which are
2346 required to make sense for the type of "CSTPTR".
2347 ``select (COND, VAL1, VAL2)``
2348 Perform the :ref:`select operation <i_select>` on constants.
2349 ``icmp COND (VAL1, VAL2)``
2350 Performs the :ref:`icmp operation <i_icmp>` on constants.
2351 ``fcmp COND (VAL1, VAL2)``
2352 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2353 ``extractelement (VAL, IDX)``
2354 Perform the :ref:`extractelement operation <i_extractelement>` on
2356 ``insertelement (VAL, ELT, IDX)``
2357 Perform the :ref:`insertelement operation <i_insertelement>` on
2359 ``shufflevector (VEC1, VEC2, IDXMASK)``
2360 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2362 ``extractvalue (VAL, IDX0, IDX1, ...)``
2363 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2364 constants. The index list is interpreted in a similar manner as
2365 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2366 least one index value must be specified.
2367 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2368 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2369 The index list is interpreted in a similar manner as indices in a
2370 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2371 value must be specified.
2372 ``OPCODE (LHS, RHS)``
2373 Perform the specified operation of the LHS and RHS constants. OPCODE
2374 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2375 binary <bitwiseops>` operations. The constraints on operands are
2376 the same as those for the corresponding instruction (e.g. no bitwise
2377 operations on floating point values are allowed).
2384 Inline Assembler Expressions
2385 ----------------------------
2387 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2388 Inline Assembly <moduleasm>`) through the use of a special value. This
2389 value represents the inline assembler as a string (containing the
2390 instructions to emit), a list of operand constraints (stored as a
2391 string), a flag that indicates whether or not the inline asm expression
2392 has side effects, and a flag indicating whether the function containing
2393 the asm needs to align its stack conservatively. An example inline
2394 assembler expression is:
2396 .. code-block:: llvm
2398 i32 (i32) asm "bswap $0", "=r,r"
2400 Inline assembler expressions may **only** be used as the callee operand
2401 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2402 Thus, typically we have:
2404 .. code-block:: llvm
2406 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2408 Inline asms with side effects not visible in the constraint list must be
2409 marked as having side effects. This is done through the use of the
2410 '``sideeffect``' keyword, like so:
2412 .. code-block:: llvm
2414 call void asm sideeffect "eieio", ""()
2416 In some cases inline asms will contain code that will not work unless
2417 the stack is aligned in some way, such as calls or SSE instructions on
2418 x86, yet will not contain code that does that alignment within the asm.
2419 The compiler should make conservative assumptions about what the asm
2420 might contain and should generate its usual stack alignment code in the
2421 prologue if the '``alignstack``' keyword is present:
2423 .. code-block:: llvm
2425 call void asm alignstack "eieio", ""()
2427 Inline asms also support using non-standard assembly dialects. The
2428 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2429 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2430 the only supported dialects. An example is:
2432 .. code-block:: llvm
2434 call void asm inteldialect "eieio", ""()
2436 If multiple keywords appear the '``sideeffect``' keyword must come
2437 first, the '``alignstack``' keyword second and the '``inteldialect``'
2443 The call instructions that wrap inline asm nodes may have a
2444 "``!srcloc``" MDNode attached to it that contains a list of constant
2445 integers. If present, the code generator will use the integer as the
2446 location cookie value when report errors through the ``LLVMContext``
2447 error reporting mechanisms. This allows a front-end to correlate backend
2448 errors that occur with inline asm back to the source code that produced
2451 .. code-block:: llvm
2453 call void asm sideeffect "something bad", ""(), !srcloc !42
2455 !42 = !{ i32 1234567 }
2457 It is up to the front-end to make sense of the magic numbers it places
2458 in the IR. If the MDNode contains multiple constants, the code generator
2459 will use the one that corresponds to the line of the asm that the error
2464 Metadata Nodes and Metadata Strings
2465 -----------------------------------
2467 LLVM IR allows metadata to be attached to instructions in the program
2468 that can convey extra information about the code to the optimizers and
2469 code generator. One example application of metadata is source-level
2470 debug information. There are two metadata primitives: strings and nodes.
2471 All metadata has the ``metadata`` type and is identified in syntax by a
2472 preceding exclamation point ('``!``').
2474 A metadata string is a string surrounded by double quotes. It can
2475 contain any character by escaping non-printable characters with
2476 "``\xx``" where "``xx``" is the two digit hex code. For example:
2479 Metadata nodes are represented with notation similar to structure
2480 constants (a comma separated list of elements, surrounded by braces and
2481 preceded by an exclamation point). Metadata nodes can have any values as
2482 their operand. For example:
2484 .. code-block:: llvm
2486 !{ metadata !"test\00", i32 10}
2488 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2489 metadata nodes, which can be looked up in the module symbol table. For
2492 .. code-block:: llvm
2494 !foo = metadata !{!4, !3}
2496 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2497 function is using two metadata arguments:
2499 .. code-block:: llvm
2501 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2503 Metadata can be attached with an instruction. Here metadata ``!21`` is
2504 attached to the ``add`` instruction using the ``!dbg`` identifier:
2506 .. code-block:: llvm
2508 %indvar.next = add i64 %indvar, 1, !dbg !21
2510 More information about specific metadata nodes recognized by the
2511 optimizers and code generator is found below.
2516 In LLVM IR, memory does not have types, so LLVM's own type system is not
2517 suitable for doing TBAA. Instead, metadata is added to the IR to
2518 describe a type system of a higher level language. This can be used to
2519 implement typical C/C++ TBAA, but it can also be used to implement
2520 custom alias analysis behavior for other languages.
2522 The current metadata format is very simple. TBAA metadata nodes have up
2523 to three fields, e.g.:
2525 .. code-block:: llvm
2527 !0 = metadata !{ metadata !"an example type tree" }
2528 !1 = metadata !{ metadata !"int", metadata !0 }
2529 !2 = metadata !{ metadata !"float", metadata !0 }
2530 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2532 The first field is an identity field. It can be any value, usually a
2533 metadata string, which uniquely identifies the type. The most important
2534 name in the tree is the name of the root node. Two trees with different
2535 root node names are entirely disjoint, even if they have leaves with
2538 The second field identifies the type's parent node in the tree, or is
2539 null or omitted for a root node. A type is considered to alias all of
2540 its descendants and all of its ancestors in the tree. Also, a type is
2541 considered to alias all types in other trees, so that bitcode produced
2542 from multiple front-ends is handled conservatively.
2544 If the third field is present, it's an integer which if equal to 1
2545 indicates that the type is "constant" (meaning
2546 ``pointsToConstantMemory`` should return true; see `other useful
2547 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2549 '``tbaa.struct``' Metadata
2550 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2552 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2553 aggregate assignment operations in C and similar languages, however it
2554 is defined to copy a contiguous region of memory, which is more than
2555 strictly necessary for aggregate types which contain holes due to
2556 padding. Also, it doesn't contain any TBAA information about the fields
2559 ``!tbaa.struct`` metadata can describe which memory subregions in a
2560 memcpy are padding and what the TBAA tags of the struct are.
2562 The current metadata format is very simple. ``!tbaa.struct`` metadata
2563 nodes are a list of operands which are in conceptual groups of three.
2564 For each group of three, the first operand gives the byte offset of a
2565 field in bytes, the second gives its size in bytes, and the third gives
2568 .. code-block:: llvm
2570 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2572 This describes a struct with two fields. The first is at offset 0 bytes
2573 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2574 and has size 4 bytes and has tbaa tag !2.
2576 Note that the fields need not be contiguous. In this example, there is a
2577 4 byte gap between the two fields. This gap represents padding which
2578 does not carry useful data and need not be preserved.
2580 '``fpmath``' Metadata
2581 ^^^^^^^^^^^^^^^^^^^^^
2583 ``fpmath`` metadata may be attached to any instruction of floating point
2584 type. It can be used to express the maximum acceptable error in the
2585 result of that instruction, in ULPs, thus potentially allowing the
2586 compiler to use a more efficient but less accurate method of computing
2587 it. ULP is defined as follows:
2589 If ``x`` is a real number that lies between two finite consecutive
2590 floating-point numbers ``a`` and ``b``, without being equal to one
2591 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2592 distance between the two non-equal finite floating-point numbers
2593 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2595 The metadata node shall consist of a single positive floating point
2596 number representing the maximum relative error, for example:
2598 .. code-block:: llvm
2600 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2602 '``range``' Metadata
2603 ^^^^^^^^^^^^^^^^^^^^
2605 ``range`` metadata may be attached only to loads of integer types. It
2606 expresses the possible ranges the loaded value is in. The ranges are
2607 represented with a flattened list of integers. The loaded value is known
2608 to be in the union of the ranges defined by each consecutive pair. Each
2609 pair has the following properties:
2611 - The type must match the type loaded by the instruction.
2612 - The pair ``a,b`` represents the range ``[a,b)``.
2613 - Both ``a`` and ``b`` are constants.
2614 - The range is allowed to wrap.
2615 - The range should not represent the full or empty set. That is,
2618 In addition, the pairs must be in signed order of the lower bound and
2619 they must be non-contiguous.
2623 .. code-block:: llvm
2625 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2626 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2627 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2628 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2630 !0 = metadata !{ i8 0, i8 2 }
2631 !1 = metadata !{ i8 255, i8 2 }
2632 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2633 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2638 It is sometimes useful to attach information to loop constructs. Currently,
2639 loop metadata is implemented as metadata attached to the branch instruction
2640 in the loop latch block. This type of metadata refer to a metadata node that is
2641 guaranteed to be separate for each loop. The loop identifier metadata is
2642 specified with the name ``llvm.loop``.
2644 The loop identifier metadata is implemented using a metadata that refers to
2645 itself to avoid merging it with any other identifier metadata, e.g.,
2646 during module linkage or function inlining. That is, each loop should refer
2647 to their own identification metadata even if they reside in separate functions.
2648 The following example contains loop identifier metadata for two separate loop
2651 .. code-block:: llvm
2653 !0 = metadata !{ metadata !0 }
2654 !1 = metadata !{ metadata !1 }
2656 The loop identifier metadata can be used to specify additional per-loop
2657 metadata. Any operands after the first operand can be treated as user-defined
2658 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2659 by the loop vectorizer to indicate how many times to unroll the loop:
2661 .. code-block:: llvm
2663 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2665 !0 = metadata !{ metadata !0, metadata !1 }
2666 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2671 Metadata types used to annotate memory accesses with information helpful
2672 for optimizations are prefixed with ``llvm.mem``.
2674 '``llvm.mem.parallel_loop_access``' Metadata
2675 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2677 For a loop to be parallel, in addition to using
2678 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2679 also all of the memory accessing instructions in the loop body need to be
2680 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2681 is at least one memory accessing instruction not marked with the metadata,
2682 the loop must be considered a sequential loop. This causes parallel loops to be
2683 converted to sequential loops due to optimization passes that are unaware of
2684 the parallel semantics and that insert new memory instructions to the loop
2687 Example of a loop that is considered parallel due to its correct use of
2688 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2689 metadata types that refer to the same loop identifier metadata.
2691 .. code-block:: llvm
2695 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2697 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2699 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2703 !0 = metadata !{ metadata !0 }
2705 It is also possible to have nested parallel loops. In that case the
2706 memory accesses refer to a list of loop identifier metadata nodes instead of
2707 the loop identifier metadata node directly:
2709 .. code-block:: llvm
2716 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2718 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2720 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2724 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2726 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2728 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2730 outer.for.end: ; preds = %for.body
2732 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2733 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2734 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2736 '``llvm.vectorizer``'
2737 ^^^^^^^^^^^^^^^^^^^^^
2739 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2740 vectorization parameters such as vectorization factor and unroll factor.
2742 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2743 loop identification metadata.
2745 '``llvm.vectorizer.unroll``' Metadata
2746 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2748 This metadata instructs the loop vectorizer to unroll the specified
2749 loop exactly ``N`` times.
2751 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2752 operand is an integer specifying the unroll factor. For example:
2754 .. code-block:: llvm
2756 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2758 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2761 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2762 determined automatically.
2764 '``llvm.vectorizer.width``' Metadata
2765 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2767 This metadata sets the target width of the vectorizer to ``N``. Without
2768 this metadata, the vectorizer will choose a width automatically.
2769 Regardless of this metadata, the vectorizer will only vectorize loops if
2770 it believes it is valid to do so.
2772 The first operand is the string ``llvm.vectorizer.width`` and the second
2773 operand is an integer specifying the width. For example:
2775 .. code-block:: llvm
2777 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2779 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2782 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2785 Module Flags Metadata
2786 =====================
2788 Information about the module as a whole is difficult to convey to LLVM's
2789 subsystems. The LLVM IR isn't sufficient to transmit this information.
2790 The ``llvm.module.flags`` named metadata exists in order to facilitate
2791 this. These flags are in the form of key / value pairs --- much like a
2792 dictionary --- making it easy for any subsystem who cares about a flag to
2795 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2796 Each triplet has the following form:
2798 - The first element is a *behavior* flag, which specifies the behavior
2799 when two (or more) modules are merged together, and it encounters two
2800 (or more) metadata with the same ID. The supported behaviors are
2802 - The second element is a metadata string that is a unique ID for the
2803 metadata. Each module may only have one flag entry for each unique ID (not
2804 including entries with the **Require** behavior).
2805 - The third element is the value of the flag.
2807 When two (or more) modules are merged together, the resulting
2808 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2809 each unique metadata ID string, there will be exactly one entry in the merged
2810 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2811 be determined by the merge behavior flag, as described below. The only exception
2812 is that entries with the *Require* behavior are always preserved.
2814 The following behaviors are supported:
2825 Emits an error if two values disagree, otherwise the resulting value
2826 is that of the operands.
2830 Emits a warning if two values disagree. The result value will be the
2831 operand for the flag from the first module being linked.
2835 Adds a requirement that another module flag be present and have a
2836 specified value after linking is performed. The value must be a
2837 metadata pair, where the first element of the pair is the ID of the
2838 module flag to be restricted, and the second element of the pair is
2839 the value the module flag should be restricted to. This behavior can
2840 be used to restrict the allowable results (via triggering of an
2841 error) of linking IDs with the **Override** behavior.
2845 Uses the specified value, regardless of the behavior or value of the
2846 other module. If both modules specify **Override**, but the values
2847 differ, an error will be emitted.
2851 Appends the two values, which are required to be metadata nodes.
2855 Appends the two values, which are required to be metadata
2856 nodes. However, duplicate entries in the second list are dropped
2857 during the append operation.
2859 It is an error for a particular unique flag ID to have multiple behaviors,
2860 except in the case of **Require** (which adds restrictions on another metadata
2861 value) or **Override**.
2863 An example of module flags:
2865 .. code-block:: llvm
2867 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2868 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2869 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2870 !3 = metadata !{ i32 3, metadata !"qux",
2872 metadata !"foo", i32 1
2875 !llvm.module.flags = !{ !0, !1, !2, !3 }
2877 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2878 if two or more ``!"foo"`` flags are seen is to emit an error if their
2879 values are not equal.
2881 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2882 behavior if two or more ``!"bar"`` flags are seen is to use the value
2885 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2886 behavior if two or more ``!"qux"`` flags are seen is to emit a
2887 warning if their values are not equal.
2889 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2893 metadata !{ metadata !"foo", i32 1 }
2895 The behavior is to emit an error if the ``llvm.module.flags`` does not
2896 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2899 Objective-C Garbage Collection Module Flags Metadata
2900 ----------------------------------------------------
2902 On the Mach-O platform, Objective-C stores metadata about garbage
2903 collection in a special section called "image info". The metadata
2904 consists of a version number and a bitmask specifying what types of
2905 garbage collection are supported (if any) by the file. If two or more
2906 modules are linked together their garbage collection metadata needs to
2907 be merged rather than appended together.
2909 The Objective-C garbage collection module flags metadata consists of the
2910 following key-value pairs:
2919 * - ``Objective-C Version``
2920 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2922 * - ``Objective-C Image Info Version``
2923 - **[Required]** --- The version of the image info section. Currently
2926 * - ``Objective-C Image Info Section``
2927 - **[Required]** --- The section to place the metadata. Valid values are
2928 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2929 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2930 Objective-C ABI version 2.
2932 * - ``Objective-C Garbage Collection``
2933 - **[Required]** --- Specifies whether garbage collection is supported or
2934 not. Valid values are 0, for no garbage collection, and 2, for garbage
2935 collection supported.
2937 * - ``Objective-C GC Only``
2938 - **[Optional]** --- Specifies that only garbage collection is supported.
2939 If present, its value must be 6. This flag requires that the
2940 ``Objective-C Garbage Collection`` flag have the value 2.
2942 Some important flag interactions:
2944 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2945 merged with a module with ``Objective-C Garbage Collection`` set to
2946 2, then the resulting module has the
2947 ``Objective-C Garbage Collection`` flag set to 0.
2948 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2949 merged with a module with ``Objective-C GC Only`` set to 6.
2951 Automatic Linker Flags Module Flags Metadata
2952 --------------------------------------------
2954 Some targets support embedding flags to the linker inside individual object
2955 files. Typically this is used in conjunction with language extensions which
2956 allow source files to explicitly declare the libraries they depend on, and have
2957 these automatically be transmitted to the linker via object files.
2959 These flags are encoded in the IR using metadata in the module flags section,
2960 using the ``Linker Options`` key. The merge behavior for this flag is required
2961 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2962 node which should be a list of other metadata nodes, each of which should be a
2963 list of metadata strings defining linker options.
2965 For example, the following metadata section specifies two separate sets of
2966 linker options, presumably to link against ``libz`` and the ``Cocoa``
2969 !0 = metadata !{ i32 6, metadata !"Linker Options",
2971 metadata !{ metadata !"-lz" },
2972 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2973 !llvm.module.flags = !{ !0 }
2975 The metadata encoding as lists of lists of options, as opposed to a collapsed
2976 list of options, is chosen so that the IR encoding can use multiple option
2977 strings to specify e.g., a single library, while still having that specifier be
2978 preserved as an atomic element that can be recognized by a target specific
2979 assembly writer or object file emitter.
2981 Each individual option is required to be either a valid option for the target's
2982 linker, or an option that is reserved by the target specific assembly writer or
2983 object file emitter. No other aspect of these options is defined by the IR.
2985 .. _intrinsicglobalvariables:
2987 Intrinsic Global Variables
2988 ==========================
2990 LLVM has a number of "magic" global variables that contain data that
2991 affect code generation or other IR semantics. These are documented here.
2992 All globals of this sort should have a section specified as
2993 "``llvm.metadata``". This section and all globals that start with
2994 "``llvm.``" are reserved for use by LLVM.
2998 The '``llvm.used``' Global Variable
2999 -----------------------------------
3001 The ``@llvm.used`` global is an array which has
3002 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3003 pointers to named global variables, functions and aliases which may optionally
3004 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3007 .. code-block:: llvm
3012 @llvm.used = appending global [2 x i8*] [
3014 i8* bitcast (i32* @Y to i8*)
3015 ], section "llvm.metadata"
3017 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3018 and linker are required to treat the symbol as if there is a reference to the
3019 symbol that it cannot see (which is why they have to be named). For example, if
3020 a variable has internal linkage and no references other than that from the
3021 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3022 references from inline asms and other things the compiler cannot "see", and
3023 corresponds to "``attribute((used))``" in GNU C.
3025 On some targets, the code generator must emit a directive to the
3026 assembler or object file to prevent the assembler and linker from
3027 molesting the symbol.
3029 .. _gv_llvmcompilerused:
3031 The '``llvm.compiler.used``' Global Variable
3032 --------------------------------------------
3034 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3035 directive, except that it only prevents the compiler from touching the
3036 symbol. On targets that support it, this allows an intelligent linker to
3037 optimize references to the symbol without being impeded as it would be
3040 This is a rare construct that should only be used in rare circumstances,
3041 and should not be exposed to source languages.
3043 .. _gv_llvmglobalctors:
3045 The '``llvm.global_ctors``' Global Variable
3046 -------------------------------------------
3048 .. code-block:: llvm
3050 %0 = type { i32, void ()* }
3051 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3053 The ``@llvm.global_ctors`` array contains a list of constructor
3054 functions and associated priorities. The functions referenced by this
3055 array will be called in ascending order of priority (i.e. lowest first)
3056 when the module is loaded. The order of functions with the same priority
3059 .. _llvmglobaldtors:
3061 The '``llvm.global_dtors``' Global Variable
3062 -------------------------------------------
3064 .. code-block:: llvm
3066 %0 = type { i32, void ()* }
3067 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3069 The ``@llvm.global_dtors`` array contains a list of destructor functions
3070 and associated priorities. The functions referenced by this array will
3071 be called in descending order of priority (i.e. highest first) when the
3072 module is loaded. The order of functions with the same priority is not
3075 Instruction Reference
3076 =====================
3078 The LLVM instruction set consists of several different classifications
3079 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3080 instructions <binaryops>`, :ref:`bitwise binary
3081 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3082 :ref:`other instructions <otherops>`.
3086 Terminator Instructions
3087 -----------------------
3089 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3090 program ends with a "Terminator" instruction, which indicates which
3091 block should be executed after the current block is finished. These
3092 terminator instructions typically yield a '``void``' value: they produce
3093 control flow, not values (the one exception being the
3094 ':ref:`invoke <i_invoke>`' instruction).
3096 The terminator instructions are: ':ref:`ret <i_ret>`',
3097 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3098 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3099 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3103 '``ret``' Instruction
3104 ^^^^^^^^^^^^^^^^^^^^^
3111 ret <type> <value> ; Return a value from a non-void function
3112 ret void ; Return from void function
3117 The '``ret``' instruction is used to return control flow (and optionally
3118 a value) from a function back to the caller.
3120 There are two forms of the '``ret``' instruction: one that returns a
3121 value and then causes control flow, and one that just causes control
3127 The '``ret``' instruction optionally accepts a single argument, the
3128 return value. The type of the return value must be a ':ref:`first
3129 class <t_firstclass>`' type.
3131 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3132 return type and contains a '``ret``' instruction with no return value or
3133 a return value with a type that does not match its type, or if it has a
3134 void return type and contains a '``ret``' instruction with a return
3140 When the '``ret``' instruction is executed, control flow returns back to
3141 the calling function's context. If the caller is a
3142 ":ref:`call <i_call>`" instruction, execution continues at the
3143 instruction after the call. If the caller was an
3144 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3145 beginning of the "normal" destination block. If the instruction returns
3146 a value, that value shall set the call or invoke instruction's return
3152 .. code-block:: llvm
3154 ret i32 5 ; Return an integer value of 5
3155 ret void ; Return from a void function
3156 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3160 '``br``' Instruction
3161 ^^^^^^^^^^^^^^^^^^^^
3168 br i1 <cond>, label <iftrue>, label <iffalse>
3169 br label <dest> ; Unconditional branch
3174 The '``br``' instruction is used to cause control flow to transfer to a
3175 different basic block in the current function. There are two forms of
3176 this instruction, corresponding to a conditional branch and an
3177 unconditional branch.
3182 The conditional branch form of the '``br``' instruction takes a single
3183 '``i1``' value and two '``label``' values. The unconditional form of the
3184 '``br``' instruction takes a single '``label``' value as a target.
3189 Upon execution of a conditional '``br``' instruction, the '``i1``'
3190 argument is evaluated. If the value is ``true``, control flows to the
3191 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3192 to the '``iffalse``' ``label`` argument.
3197 .. code-block:: llvm
3200 %cond = icmp eq i32 %a, %b
3201 br i1 %cond, label %IfEqual, label %IfUnequal
3209 '``switch``' Instruction
3210 ^^^^^^^^^^^^^^^^^^^^^^^^
3217 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3222 The '``switch``' instruction is used to transfer control flow to one of
3223 several different places. It is a generalization of the '``br``'
3224 instruction, allowing a branch to occur to one of many possible
3230 The '``switch``' instruction uses three parameters: an integer
3231 comparison value '``value``', a default '``label``' destination, and an
3232 array of pairs of comparison value constants and '``label``'s. The table
3233 is not allowed to contain duplicate constant entries.
3238 The ``switch`` instruction specifies a table of values and destinations.
3239 When the '``switch``' instruction is executed, this table is searched
3240 for the given value. If the value is found, control flow is transferred
3241 to the corresponding destination; otherwise, control flow is transferred
3242 to the default destination.
3247 Depending on properties of the target machine and the particular
3248 ``switch`` instruction, this instruction may be code generated in
3249 different ways. For example, it could be generated as a series of
3250 chained conditional branches or with a lookup table.
3255 .. code-block:: llvm
3257 ; Emulate a conditional br instruction
3258 %Val = zext i1 %value to i32
3259 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3261 ; Emulate an unconditional br instruction
3262 switch i32 0, label %dest [ ]
3264 ; Implement a jump table:
3265 switch i32 %val, label %otherwise [ i32 0, label %onzero
3267 i32 2, label %ontwo ]
3271 '``indirectbr``' Instruction
3272 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3279 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3284 The '``indirectbr``' instruction implements an indirect branch to a
3285 label within the current function, whose address is specified by
3286 "``address``". Address must be derived from a
3287 :ref:`blockaddress <blockaddress>` constant.
3292 The '``address``' argument is the address of the label to jump to. The
3293 rest of the arguments indicate the full set of possible destinations
3294 that the address may point to. Blocks are allowed to occur multiple
3295 times in the destination list, though this isn't particularly useful.
3297 This destination list is required so that dataflow analysis has an
3298 accurate understanding of the CFG.
3303 Control transfers to the block specified in the address argument. All
3304 possible destination blocks must be listed in the label list, otherwise
3305 this instruction has undefined behavior. This implies that jumps to
3306 labels defined in other functions have undefined behavior as well.
3311 This is typically implemented with a jump through a register.
3316 .. code-block:: llvm
3318 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3322 '``invoke``' Instruction
3323 ^^^^^^^^^^^^^^^^^^^^^^^^
3330 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3331 to label <normal label> unwind label <exception label>
3336 The '``invoke``' instruction causes control to transfer to a specified
3337 function, with the possibility of control flow transfer to either the
3338 '``normal``' label or the '``exception``' label. If the callee function
3339 returns with the "``ret``" instruction, control flow will return to the
3340 "normal" label. If the callee (or any indirect callees) returns via the
3341 ":ref:`resume <i_resume>`" instruction or other exception handling
3342 mechanism, control is interrupted and continued at the dynamically
3343 nearest "exception" label.
3345 The '``exception``' label is a `landing
3346 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3347 '``exception``' label is required to have the
3348 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3349 information about the behavior of the program after unwinding happens,
3350 as its first non-PHI instruction. The restrictions on the
3351 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3352 instruction, so that the important information contained within the
3353 "``landingpad``" instruction can't be lost through normal code motion.
3358 This instruction requires several arguments:
3360 #. The optional "cconv" marker indicates which :ref:`calling
3361 convention <callingconv>` the call should use. If none is
3362 specified, the call defaults to using C calling conventions.
3363 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3364 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3366 #. '``ptr to function ty``': shall be the signature of the pointer to
3367 function value being invoked. In most cases, this is a direct
3368 function invocation, but indirect ``invoke``'s are just as possible,
3369 branching off an arbitrary pointer to function value.
3370 #. '``function ptr val``': An LLVM value containing a pointer to a
3371 function to be invoked.
3372 #. '``function args``': argument list whose types match the function
3373 signature argument types and parameter attributes. All arguments must
3374 be of :ref:`first class <t_firstclass>` type. If the function signature
3375 indicates the function accepts a variable number of arguments, the
3376 extra arguments can be specified.
3377 #. '``normal label``': the label reached when the called function
3378 executes a '``ret``' instruction.
3379 #. '``exception label``': the label reached when a callee returns via
3380 the :ref:`resume <i_resume>` instruction or other exception handling
3382 #. The optional :ref:`function attributes <fnattrs>` list. Only
3383 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3384 attributes are valid here.
3389 This instruction is designed to operate as a standard '``call``'
3390 instruction in most regards. The primary difference is that it
3391 establishes an association with a label, which is used by the runtime
3392 library to unwind the stack.
3394 This instruction is used in languages with destructors to ensure that
3395 proper cleanup is performed in the case of either a ``longjmp`` or a
3396 thrown exception. Additionally, this is important for implementation of
3397 '``catch``' clauses in high-level languages that support them.
3399 For the purposes of the SSA form, the definition of the value returned
3400 by the '``invoke``' instruction is deemed to occur on the edge from the
3401 current block to the "normal" label. If the callee unwinds then no
3402 return value is available.
3407 .. code-block:: llvm
3409 %retval = invoke i32 @Test(i32 15) to label %Continue
3410 unwind label %TestCleanup ; {i32}:retval set
3411 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3412 unwind label %TestCleanup ; {i32}:retval set
3416 '``resume``' Instruction
3417 ^^^^^^^^^^^^^^^^^^^^^^^^
3424 resume <type> <value>
3429 The '``resume``' instruction is a terminator instruction that has no
3435 The '``resume``' instruction requires one argument, which must have the
3436 same type as the result of any '``landingpad``' instruction in the same
3442 The '``resume``' instruction resumes propagation of an existing
3443 (in-flight) exception whose unwinding was interrupted with a
3444 :ref:`landingpad <i_landingpad>` instruction.
3449 .. code-block:: llvm
3451 resume { i8*, i32 } %exn
3455 '``unreachable``' Instruction
3456 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3468 The '``unreachable``' instruction has no defined semantics. This
3469 instruction is used to inform the optimizer that a particular portion of
3470 the code is not reachable. This can be used to indicate that the code
3471 after a no-return function cannot be reached, and other facts.
3476 The '``unreachable``' instruction has no defined semantics.
3483 Binary operators are used to do most of the computation in a program.
3484 They require two operands of the same type, execute an operation on
3485 them, and produce a single value. The operands might represent multiple
3486 data, as is the case with the :ref:`vector <t_vector>` data type. The
3487 result value has the same type as its operands.
3489 There are several different binary operators:
3493 '``add``' Instruction
3494 ^^^^^^^^^^^^^^^^^^^^^
3501 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3502 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3503 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3504 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3509 The '``add``' instruction returns the sum of its two operands.
3514 The two arguments to the '``add``' instruction must be
3515 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3516 arguments must have identical types.
3521 The value produced is the integer sum of the two operands.
3523 If the sum has unsigned overflow, the result returned is the
3524 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3527 Because LLVM integers use a two's complement representation, this
3528 instruction is appropriate for both signed and unsigned integers.
3530 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3531 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3532 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3533 unsigned and/or signed overflow, respectively, occurs.
3538 .. code-block:: llvm
3540 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3544 '``fadd``' Instruction
3545 ^^^^^^^^^^^^^^^^^^^^^^
3552 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3557 The '``fadd``' instruction returns the sum of its two operands.
3562 The two arguments to the '``fadd``' instruction must be :ref:`floating
3563 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3564 Both arguments must have identical types.
3569 The value produced is the floating point sum of the two operands. This
3570 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3571 which are optimization hints to enable otherwise unsafe floating point
3577 .. code-block:: llvm
3579 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3581 '``sub``' Instruction
3582 ^^^^^^^^^^^^^^^^^^^^^
3589 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3590 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3591 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3592 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3597 The '``sub``' instruction returns the difference of its two operands.
3599 Note that the '``sub``' instruction is used to represent the '``neg``'
3600 instruction present in most other intermediate representations.
3605 The two arguments to the '``sub``' instruction must be
3606 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3607 arguments must have identical types.
3612 The value produced is the integer difference of the two operands.
3614 If the difference has unsigned overflow, the result returned is the
3615 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3618 Because LLVM integers use a two's complement representation, this
3619 instruction is appropriate for both signed and unsigned integers.
3621 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3622 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3623 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3624 unsigned and/or signed overflow, respectively, occurs.
3629 .. code-block:: llvm
3631 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3632 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3636 '``fsub``' Instruction
3637 ^^^^^^^^^^^^^^^^^^^^^^
3644 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3649 The '``fsub``' instruction returns the difference of its two operands.
3651 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3652 instruction present in most other intermediate representations.
3657 The two arguments to the '``fsub``' instruction must be :ref:`floating
3658 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3659 Both arguments must have identical types.
3664 The value produced is the floating point difference of the two operands.
3665 This instruction can also take any number of :ref:`fast-math
3666 flags <fastmath>`, which are optimization hints to enable otherwise
3667 unsafe floating point optimizations:
3672 .. code-block:: llvm
3674 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3675 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3677 '``mul``' Instruction
3678 ^^^^^^^^^^^^^^^^^^^^^
3685 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3686 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3687 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3688 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3693 The '``mul``' instruction returns the product of its two operands.
3698 The two arguments to the '``mul``' instruction must be
3699 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3700 arguments must have identical types.
3705 The value produced is the integer product of the two operands.
3707 If the result of the multiplication has unsigned overflow, the result
3708 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3709 bit width of the result.
3711 Because LLVM integers use a two's complement representation, and the
3712 result is the same width as the operands, this instruction returns the
3713 correct result for both signed and unsigned integers. If a full product
3714 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3715 sign-extended or zero-extended as appropriate to the width of the full
3718 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3719 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3720 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3721 unsigned and/or signed overflow, respectively, occurs.
3726 .. code-block:: llvm
3728 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3732 '``fmul``' Instruction
3733 ^^^^^^^^^^^^^^^^^^^^^^
3740 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3745 The '``fmul``' instruction returns the product of its two operands.
3750 The two arguments to the '``fmul``' instruction must be :ref:`floating
3751 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3752 Both arguments must have identical types.
3757 The value produced is the floating point product of the two operands.
3758 This instruction can also take any number of :ref:`fast-math
3759 flags <fastmath>`, which are optimization hints to enable otherwise
3760 unsafe floating point optimizations:
3765 .. code-block:: llvm
3767 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3769 '``udiv``' Instruction
3770 ^^^^^^^^^^^^^^^^^^^^^^
3777 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3778 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3783 The '``udiv``' instruction returns the quotient of its two operands.
3788 The two arguments to the '``udiv``' instruction must be
3789 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3790 arguments must have identical types.
3795 The value produced is the unsigned integer quotient of the two operands.
3797 Note that unsigned integer division and signed integer division are
3798 distinct operations; for signed integer division, use '``sdiv``'.
3800 Division by zero leads to undefined behavior.
3802 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3803 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3804 such, "((a udiv exact b) mul b) == a").
3809 .. code-block:: llvm
3811 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3813 '``sdiv``' Instruction
3814 ^^^^^^^^^^^^^^^^^^^^^^
3821 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3822 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3827 The '``sdiv``' instruction returns the quotient of its two operands.
3832 The two arguments to the '``sdiv``' instruction must be
3833 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3834 arguments must have identical types.
3839 The value produced is the signed integer quotient of the two operands
3840 rounded towards zero.
3842 Note that signed integer division and unsigned integer division are
3843 distinct operations; for unsigned integer division, use '``udiv``'.
3845 Division by zero leads to undefined behavior. Overflow also leads to
3846 undefined behavior; this is a rare case, but can occur, for example, by
3847 doing a 32-bit division of -2147483648 by -1.
3849 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3850 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3855 .. code-block:: llvm
3857 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3861 '``fdiv``' Instruction
3862 ^^^^^^^^^^^^^^^^^^^^^^
3869 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3874 The '``fdiv``' instruction returns the quotient of its two operands.
3879 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3880 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3881 Both arguments must have identical types.
3886 The value produced is the floating point quotient of the two operands.
3887 This instruction can also take any number of :ref:`fast-math
3888 flags <fastmath>`, which are optimization hints to enable otherwise
3889 unsafe floating point optimizations:
3894 .. code-block:: llvm
3896 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3898 '``urem``' Instruction
3899 ^^^^^^^^^^^^^^^^^^^^^^
3906 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3911 The '``urem``' instruction returns the remainder from the unsigned
3912 division of its two arguments.
3917 The two arguments to the '``urem``' instruction must be
3918 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3919 arguments must have identical types.
3924 This instruction returns the unsigned integer *remainder* of a division.
3925 This instruction always performs an unsigned division to get the
3928 Note that unsigned integer remainder and signed integer remainder are
3929 distinct operations; for signed integer remainder, use '``srem``'.
3931 Taking the remainder of a division by zero leads to undefined behavior.
3936 .. code-block:: llvm
3938 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3940 '``srem``' Instruction
3941 ^^^^^^^^^^^^^^^^^^^^^^
3948 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3953 The '``srem``' instruction returns the remainder from the signed
3954 division of its two operands. This instruction can also take
3955 :ref:`vector <t_vector>` versions of the values in which case the elements
3961 The two arguments to the '``srem``' instruction must be
3962 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3963 arguments must have identical types.
3968 This instruction returns the *remainder* of a division (where the result
3969 is either zero or has the same sign as the dividend, ``op1``), not the
3970 *modulo* operator (where the result is either zero or has the same sign
3971 as the divisor, ``op2``) of a value. For more information about the
3972 difference, see `The Math
3973 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3974 table of how this is implemented in various languages, please see
3976 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3978 Note that signed integer remainder and unsigned integer remainder are
3979 distinct operations; for unsigned integer remainder, use '``urem``'.
3981 Taking the remainder of a division by zero leads to undefined behavior.
3982 Overflow also leads to undefined behavior; this is a rare case, but can
3983 occur, for example, by taking the remainder of a 32-bit division of
3984 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3985 rule lets srem be implemented using instructions that return both the
3986 result of the division and the remainder.)
3991 .. code-block:: llvm
3993 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3997 '``frem``' Instruction
3998 ^^^^^^^^^^^^^^^^^^^^^^
4005 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4010 The '``frem``' instruction returns the remainder from the division of
4016 The two arguments to the '``frem``' instruction must be :ref:`floating
4017 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4018 Both arguments must have identical types.
4023 This instruction returns the *remainder* of a division. The remainder
4024 has the same sign as the dividend. This instruction can also take any
4025 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4026 to enable otherwise unsafe floating point optimizations:
4031 .. code-block:: llvm
4033 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4037 Bitwise Binary Operations
4038 -------------------------
4040 Bitwise binary operators are used to do various forms of bit-twiddling
4041 in a program. They are generally very efficient instructions and can
4042 commonly be strength reduced from other instructions. They require two
4043 operands of the same type, execute an operation on them, and produce a
4044 single value. The resulting value is the same type as its operands.
4046 '``shl``' Instruction
4047 ^^^^^^^^^^^^^^^^^^^^^
4054 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4055 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4056 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4057 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4062 The '``shl``' instruction returns the first operand shifted to the left
4063 a specified number of bits.
4068 Both arguments to the '``shl``' instruction must be the same
4069 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4070 '``op2``' is treated as an unsigned value.
4075 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4076 where ``n`` is the width of the result. If ``op2`` is (statically or
4077 dynamically) negative or equal to or larger than the number of bits in
4078 ``op1``, the result is undefined. If the arguments are vectors, each
4079 vector element of ``op1`` is shifted by the corresponding shift amount
4082 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4083 value <poisonvalues>` if it shifts out any non-zero bits. If the
4084 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4085 value <poisonvalues>` if it shifts out any bits that disagree with the
4086 resultant sign bit. As such, NUW/NSW have the same semantics as they
4087 would if the shift were expressed as a mul instruction with the same
4088 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4093 .. code-block:: llvm
4095 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4096 <result> = shl i32 4, 2 ; yields {i32}: 16
4097 <result> = shl i32 1, 10 ; yields {i32}: 1024
4098 <result> = shl i32 1, 32 ; undefined
4099 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4101 '``lshr``' Instruction
4102 ^^^^^^^^^^^^^^^^^^^^^^
4109 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4110 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4115 The '``lshr``' instruction (logical shift right) returns the first
4116 operand shifted to the right a specified number of bits with zero fill.
4121 Both arguments to the '``lshr``' instruction must be the same
4122 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4123 '``op2``' is treated as an unsigned value.
4128 This instruction always performs a logical shift right operation. The
4129 most significant bits of the result will be filled with zero bits after
4130 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4131 than the number of bits in ``op1``, the result is undefined. If the
4132 arguments are vectors, each vector element of ``op1`` is shifted by the
4133 corresponding shift amount in ``op2``.
4135 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4136 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4142 .. code-block:: llvm
4144 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4145 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4146 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4147 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4148 <result> = lshr i32 1, 32 ; undefined
4149 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4151 '``ashr``' Instruction
4152 ^^^^^^^^^^^^^^^^^^^^^^
4159 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4160 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4165 The '``ashr``' instruction (arithmetic shift right) returns the first
4166 operand shifted to the right a specified number of bits with sign
4172 Both arguments to the '``ashr``' instruction must be the same
4173 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4174 '``op2``' is treated as an unsigned value.
4179 This instruction always performs an arithmetic shift right operation,
4180 The most significant bits of the result will be filled with the sign bit
4181 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4182 than the number of bits in ``op1``, the result is undefined. If the
4183 arguments are vectors, each vector element of ``op1`` is shifted by the
4184 corresponding shift amount in ``op2``.
4186 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4187 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4193 .. code-block:: llvm
4195 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4196 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4197 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4198 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4199 <result> = ashr i32 1, 32 ; undefined
4200 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4202 '``and``' Instruction
4203 ^^^^^^^^^^^^^^^^^^^^^
4210 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4215 The '``and``' instruction returns the bitwise logical and of its two
4221 The two arguments to the '``and``' instruction must be
4222 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4223 arguments must have identical types.
4228 The truth table used for the '``and``' instruction is:
4245 .. code-block:: llvm
4247 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4248 <result> = and i32 15, 40 ; yields {i32}:result = 8
4249 <result> = and i32 4, 8 ; yields {i32}:result = 0
4251 '``or``' Instruction
4252 ^^^^^^^^^^^^^^^^^^^^
4259 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4264 The '``or``' instruction returns the bitwise logical inclusive or of its
4270 The two arguments to the '``or``' instruction must be
4271 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4272 arguments must have identical types.
4277 The truth table used for the '``or``' instruction is:
4296 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4297 <result> = or i32 15, 40 ; yields {i32}:result = 47
4298 <result> = or i32 4, 8 ; yields {i32}:result = 12
4300 '``xor``' Instruction
4301 ^^^^^^^^^^^^^^^^^^^^^
4308 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4313 The '``xor``' instruction returns the bitwise logical exclusive or of
4314 its two operands. The ``xor`` is used to implement the "one's
4315 complement" operation, which is the "~" operator in C.
4320 The two arguments to the '``xor``' instruction must be
4321 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4322 arguments must have identical types.
4327 The truth table used for the '``xor``' instruction is:
4344 .. code-block:: llvm
4346 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4347 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4348 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4349 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4354 LLVM supports several instructions to represent vector operations in a
4355 target-independent manner. These instructions cover the element-access
4356 and vector-specific operations needed to process vectors effectively.
4357 While LLVM does directly support these vector operations, many
4358 sophisticated algorithms will want to use target-specific intrinsics to
4359 take full advantage of a specific target.
4361 .. _i_extractelement:
4363 '``extractelement``' Instruction
4364 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4371 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4376 The '``extractelement``' instruction extracts a single scalar element
4377 from a vector at a specified index.
4382 The first operand of an '``extractelement``' instruction is a value of
4383 :ref:`vector <t_vector>` type. The second operand is an index indicating
4384 the position from which to extract the element. The index may be a
4390 The result is a scalar of the same type as the element type of ``val``.
4391 Its value is the value at position ``idx`` of ``val``. If ``idx``
4392 exceeds the length of ``val``, the results are undefined.
4397 .. code-block:: llvm
4399 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4401 .. _i_insertelement:
4403 '``insertelement``' Instruction
4404 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4411 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4416 The '``insertelement``' instruction inserts a scalar element into a
4417 vector at a specified index.
4422 The first operand of an '``insertelement``' instruction is a value of
4423 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4424 type must equal the element type of the first operand. The third operand
4425 is an index indicating the position at which to insert the value. The
4426 index may be a variable.
4431 The result is a vector of the same type as ``val``. Its element values
4432 are those of ``val`` except at position ``idx``, where it gets the value
4433 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4439 .. code-block:: llvm
4441 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4443 .. _i_shufflevector:
4445 '``shufflevector``' Instruction
4446 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4453 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4458 The '``shufflevector``' instruction constructs a permutation of elements
4459 from two input vectors, returning a vector with the same element type as
4460 the input and length that is the same as the shuffle mask.
4465 The first two operands of a '``shufflevector``' instruction are vectors
4466 with the same type. The third argument is a shuffle mask whose element
4467 type is always 'i32'. The result of the instruction is a vector whose
4468 length is the same as the shuffle mask and whose element type is the
4469 same as the element type of the first two operands.
4471 The shuffle mask operand is required to be a constant vector with either
4472 constant integer or undef values.
4477 The elements of the two input vectors are numbered from left to right
4478 across both of the vectors. The shuffle mask operand specifies, for each
4479 element of the result vector, which element of the two input vectors the
4480 result element gets. The element selector may be undef (meaning "don't
4481 care") and the second operand may be undef if performing a shuffle from
4487 .. code-block:: llvm
4489 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4490 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4491 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4492 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4493 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4494 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4495 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4496 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4498 Aggregate Operations
4499 --------------------
4501 LLVM supports several instructions for working with
4502 :ref:`aggregate <t_aggregate>` values.
4506 '``extractvalue``' Instruction
4507 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4514 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4519 The '``extractvalue``' instruction extracts the value of a member field
4520 from an :ref:`aggregate <t_aggregate>` value.
4525 The first operand of an '``extractvalue``' instruction is a value of
4526 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4527 constant indices to specify which value to extract in a similar manner
4528 as indices in a '``getelementptr``' instruction.
4530 The major differences to ``getelementptr`` indexing are:
4532 - Since the value being indexed is not a pointer, the first index is
4533 omitted and assumed to be zero.
4534 - At least one index must be specified.
4535 - Not only struct indices but also array indices must be in bounds.
4540 The result is the value at the position in the aggregate specified by
4546 .. code-block:: llvm
4548 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4552 '``insertvalue``' Instruction
4553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4560 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4565 The '``insertvalue``' instruction inserts a value into a member field in
4566 an :ref:`aggregate <t_aggregate>` value.
4571 The first operand of an '``insertvalue``' instruction is a value of
4572 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4573 a first-class value to insert. The following operands are constant
4574 indices indicating the position at which to insert the value in a
4575 similar manner as indices in a '``extractvalue``' instruction. The value
4576 to insert must have the same type as the value identified by the
4582 The result is an aggregate of the same type as ``val``. Its value is
4583 that of ``val`` except that the value at the position specified by the
4584 indices is that of ``elt``.
4589 .. code-block:: llvm
4591 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4592 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4593 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4597 Memory Access and Addressing Operations
4598 ---------------------------------------
4600 A key design point of an SSA-based representation is how it represents
4601 memory. In LLVM, no memory locations are in SSA form, which makes things
4602 very simple. This section describes how to read, write, and allocate
4607 '``alloca``' Instruction
4608 ^^^^^^^^^^^^^^^^^^^^^^^^
4615 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4620 The '``alloca``' instruction allocates memory on the stack frame of the
4621 currently executing function, to be automatically released when this
4622 function returns to its caller. The object is always allocated in the
4623 generic address space (address space zero).
4628 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4629 bytes of memory on the runtime stack, returning a pointer of the
4630 appropriate type to the program. If "NumElements" is specified, it is
4631 the number of elements allocated, otherwise "NumElements" is defaulted
4632 to be one. If a constant alignment is specified, the value result of the
4633 allocation is guaranteed to be aligned to at least that boundary. If not
4634 specified, or if zero, the target can choose to align the allocation on
4635 any convenient boundary compatible with the type.
4637 '``type``' may be any sized type.
4642 Memory is allocated; a pointer is returned. The operation is undefined
4643 if there is insufficient stack space for the allocation. '``alloca``'d
4644 memory is automatically released when the function returns. The
4645 '``alloca``' instruction is commonly used to represent automatic
4646 variables that must have an address available. When the function returns
4647 (either with the ``ret`` or ``resume`` instructions), the memory is
4648 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4649 The order in which memory is allocated (ie., which way the stack grows)
4655 .. code-block:: llvm
4657 %ptr = alloca i32 ; yields {i32*}:ptr
4658 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4659 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4660 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4664 '``load``' Instruction
4665 ^^^^^^^^^^^^^^^^^^^^^^
4672 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4673 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4674 !<index> = !{ i32 1 }
4679 The '``load``' instruction is used to read from memory.
4684 The argument to the ``load`` instruction specifies the memory address
4685 from which to load. The pointer must point to a :ref:`first
4686 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4687 then the optimizer is not allowed to modify the number or order of
4688 execution of this ``load`` with other :ref:`volatile
4689 operations <volatile>`.
4691 If the ``load`` is marked as ``atomic``, it takes an extra
4692 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4693 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4694 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4695 when they may see multiple atomic stores. The type of the pointee must
4696 be an integer type whose bit width is a power of two greater than or
4697 equal to eight and less than or equal to a target-specific size limit.
4698 ``align`` must be explicitly specified on atomic loads, and the load has
4699 undefined behavior if the alignment is not set to a value which is at
4700 least the size in bytes of the pointee. ``!nontemporal`` does not have
4701 any defined semantics for atomic loads.
4703 The optional constant ``align`` argument specifies the alignment of the
4704 operation (that is, the alignment of the memory address). A value of 0
4705 or an omitted ``align`` argument means that the operation has the ABI
4706 alignment for the target. It is the responsibility of the code emitter
4707 to ensure that the alignment information is correct. Overestimating the
4708 alignment results in undefined behavior. Underestimating the alignment
4709 may produce less efficient code. An alignment of 1 is always safe.
4711 The optional ``!nontemporal`` metadata must reference a single
4712 metadata name ``<index>`` corresponding to a metadata node with one
4713 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4714 metadata on the instruction tells the optimizer and code generator
4715 that this load is not expected to be reused in the cache. The code
4716 generator may select special instructions to save cache bandwidth, such
4717 as the ``MOVNT`` instruction on x86.
4719 The optional ``!invariant.load`` metadata must reference a single
4720 metadata name ``<index>`` corresponding to a metadata node with no
4721 entries. The existence of the ``!invariant.load`` metadata on the
4722 instruction tells the optimizer and code generator that this load
4723 address points to memory which does not change value during program
4724 execution. The optimizer may then move this load around, for example, by
4725 hoisting it out of loops using loop invariant code motion.
4730 The location of memory pointed to is loaded. If the value being loaded
4731 is of scalar type then the number of bytes read does not exceed the
4732 minimum number of bytes needed to hold all bits of the type. For
4733 example, loading an ``i24`` reads at most three bytes. When loading a
4734 value of a type like ``i20`` with a size that is not an integral number
4735 of bytes, the result is undefined if the value was not originally
4736 written using a store of the same type.
4741 .. code-block:: llvm
4743 %ptr = alloca i32 ; yields {i32*}:ptr
4744 store i32 3, i32* %ptr ; yields {void}
4745 %val = load i32* %ptr ; yields {i32}:val = i32 3
4749 '``store``' Instruction
4750 ^^^^^^^^^^^^^^^^^^^^^^^
4757 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4758 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4763 The '``store``' instruction is used to write to memory.
4768 There are two arguments to the ``store`` instruction: a value to store
4769 and an address at which to store it. The type of the ``<pointer>``
4770 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4771 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4772 then the optimizer is not allowed to modify the number or order of
4773 execution of this ``store`` with other :ref:`volatile
4774 operations <volatile>`.
4776 If the ``store`` is marked as ``atomic``, it takes an extra
4777 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4778 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4779 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4780 when they may see multiple atomic stores. The type of the pointee must
4781 be an integer type whose bit width is a power of two greater than or
4782 equal to eight and less than or equal to a target-specific size limit.
4783 ``align`` must be explicitly specified on atomic stores, and the store
4784 has undefined behavior if the alignment is not set to a value which is
4785 at least the size in bytes of the pointee. ``!nontemporal`` does not
4786 have any defined semantics for atomic stores.
4788 The optional constant ``align`` argument specifies the alignment of the
4789 operation (that is, the alignment of the memory address). A value of 0
4790 or an omitted ``align`` argument means that the operation has the ABI
4791 alignment for the target. It is the responsibility of the code emitter
4792 to ensure that the alignment information is correct. Overestimating the
4793 alignment results in undefined behavior. Underestimating the
4794 alignment may produce less efficient code. An alignment of 1 is always
4797 The optional ``!nontemporal`` metadata must reference a single metadata
4798 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4799 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4800 tells the optimizer and code generator that this load is not expected to
4801 be reused in the cache. The code generator may select special
4802 instructions to save cache bandwidth, such as the MOVNT instruction on
4808 The contents of memory are updated to contain ``<value>`` at the
4809 location specified by the ``<pointer>`` operand. If ``<value>`` is
4810 of scalar type then the number of bytes written does not exceed the
4811 minimum number of bytes needed to hold all bits of the type. For
4812 example, storing an ``i24`` writes at most three bytes. When writing a
4813 value of a type like ``i20`` with a size that is not an integral number
4814 of bytes, it is unspecified what happens to the extra bits that do not
4815 belong to the type, but they will typically be overwritten.
4820 .. code-block:: llvm
4822 %ptr = alloca i32 ; yields {i32*}:ptr
4823 store i32 3, i32* %ptr ; yields {void}
4824 %val = load i32* %ptr ; yields {i32}:val = i32 3
4828 '``fence``' Instruction
4829 ^^^^^^^^^^^^^^^^^^^^^^^
4836 fence [singlethread] <ordering> ; yields {void}
4841 The '``fence``' instruction is used to introduce happens-before edges
4847 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4848 defines what *synchronizes-with* edges they add. They can only be given
4849 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4854 A fence A which has (at least) ``release`` ordering semantics
4855 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4856 semantics if and only if there exist atomic operations X and Y, both
4857 operating on some atomic object M, such that A is sequenced before X, X
4858 modifies M (either directly or through some side effect of a sequence
4859 headed by X), Y is sequenced before B, and Y observes M. This provides a
4860 *happens-before* dependency between A and B. Rather than an explicit
4861 ``fence``, one (but not both) of the atomic operations X or Y might
4862 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4863 still *synchronize-with* the explicit ``fence`` and establish the
4864 *happens-before* edge.
4866 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4867 ``acquire`` and ``release`` semantics specified above, participates in
4868 the global program order of other ``seq_cst`` operations and/or fences.
4870 The optional ":ref:`singlethread <singlethread>`" argument specifies
4871 that the fence only synchronizes with other fences in the same thread.
4872 (This is useful for interacting with signal handlers.)
4877 .. code-block:: llvm
4879 fence acquire ; yields {void}
4880 fence singlethread seq_cst ; yields {void}
4884 '``cmpxchg``' Instruction
4885 ^^^^^^^^^^^^^^^^^^^^^^^^^
4892 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4897 The '``cmpxchg``' instruction is used to atomically modify memory. It
4898 loads a value in memory and compares it to a given value. If they are
4899 equal, it stores a new value into the memory.
4904 There are three arguments to the '``cmpxchg``' instruction: an address
4905 to operate on, a value to compare to the value currently be at that
4906 address, and a new value to place at that address if the compared values
4907 are equal. The type of '<cmp>' must be an integer type whose bit width
4908 is a power of two greater than or equal to eight and less than or equal
4909 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4910 type, and the type of '<pointer>' must be a pointer to that type. If the
4911 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4912 to modify the number or order of execution of this ``cmpxchg`` with
4913 other :ref:`volatile operations <volatile>`.
4915 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4916 synchronizes with other atomic operations.
4918 The optional "``singlethread``" argument declares that the ``cmpxchg``
4919 is only atomic with respect to code (usually signal handlers) running in
4920 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4921 respect to all other code in the system.
4923 The pointer passed into cmpxchg must have alignment greater than or
4924 equal to the size in memory of the operand.
4929 The contents of memory at the location specified by the '``<pointer>``'
4930 operand is read and compared to '``<cmp>``'; if the read value is the
4931 equal, '``<new>``' is written. The original value at the location is
4934 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4935 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4936 atomic load with an ordering parameter determined by dropping any
4937 ``release`` part of the ``cmpxchg``'s ordering.
4942 .. code-block:: llvm
4945 %orig = atomic load i32* %ptr unordered ; yields {i32}
4949 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4950 %squared = mul i32 %cmp, %cmp
4951 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4952 %success = icmp eq i32 %cmp, %old
4953 br i1 %success, label %done, label %loop
4960 '``atomicrmw``' Instruction
4961 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4968 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4973 The '``atomicrmw``' instruction is used to atomically modify memory.
4978 There are three arguments to the '``atomicrmw``' instruction: an
4979 operation to apply, an address whose value to modify, an argument to the
4980 operation. The operation must be one of the following keywords:
4994 The type of '<value>' must be an integer type whose bit width is a power
4995 of two greater than or equal to eight and less than or equal to a
4996 target-specific size limit. The type of the '``<pointer>``' operand must
4997 be a pointer to that type. If the ``atomicrmw`` is marked as
4998 ``volatile``, then the optimizer is not allowed to modify the number or
4999 order of execution of this ``atomicrmw`` with other :ref:`volatile
5000 operations <volatile>`.
5005 The contents of memory at the location specified by the '``<pointer>``'
5006 operand are atomically read, modified, and written back. The original
5007 value at the location is returned. The modification is specified by the
5010 - xchg: ``*ptr = val``
5011 - add: ``*ptr = *ptr + val``
5012 - sub: ``*ptr = *ptr - val``
5013 - and: ``*ptr = *ptr & val``
5014 - nand: ``*ptr = ~(*ptr & val)``
5015 - or: ``*ptr = *ptr | val``
5016 - xor: ``*ptr = *ptr ^ val``
5017 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5018 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5019 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5021 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5027 .. code-block:: llvm
5029 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5031 .. _i_getelementptr:
5033 '``getelementptr``' Instruction
5034 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5041 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5042 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5043 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5048 The '``getelementptr``' instruction is used to get the address of a
5049 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5050 address calculation only and does not access memory.
5055 The first argument is always a pointer or a vector of pointers, and
5056 forms the basis of the calculation. The remaining arguments are indices
5057 that indicate which of the elements of the aggregate object are indexed.
5058 The interpretation of each index is dependent on the type being indexed
5059 into. The first index always indexes the pointer value given as the
5060 first argument, the second index indexes a value of the type pointed to
5061 (not necessarily the value directly pointed to, since the first index
5062 can be non-zero), etc. The first type indexed into must be a pointer
5063 value, subsequent types can be arrays, vectors, and structs. Note that
5064 subsequent types being indexed into can never be pointers, since that
5065 would require loading the pointer before continuing calculation.
5067 The type of each index argument depends on the type it is indexing into.
5068 When indexing into a (optionally packed) structure, only ``i32`` integer
5069 **constants** are allowed (when using a vector of indices they must all
5070 be the **same** ``i32`` integer constant). When indexing into an array,
5071 pointer or vector, integers of any width are allowed, and they are not
5072 required to be constant. These integers are treated as signed values
5075 For example, let's consider a C code fragment and how it gets compiled
5091 int *foo(struct ST *s) {
5092 return &s[1].Z.B[5][13];
5095 The LLVM code generated by Clang is:
5097 .. code-block:: llvm
5099 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5100 %struct.ST = type { i32, double, %struct.RT }
5102 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5104 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5111 In the example above, the first index is indexing into the
5112 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5113 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5114 indexes into the third element of the structure, yielding a
5115 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5116 structure. The third index indexes into the second element of the
5117 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5118 dimensions of the array are subscripted into, yielding an '``i32``'
5119 type. The '``getelementptr``' instruction returns a pointer to this
5120 element, thus computing a value of '``i32*``' type.
5122 Note that it is perfectly legal to index partially through a structure,
5123 returning a pointer to an inner element. Because of this, the LLVM code
5124 for the given testcase is equivalent to:
5126 .. code-block:: llvm
5128 define i32* @foo(%struct.ST* %s) {
5129 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5130 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5131 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5132 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5133 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5137 If the ``inbounds`` keyword is present, the result value of the
5138 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5139 pointer is not an *in bounds* address of an allocated object, or if any
5140 of the addresses that would be formed by successive addition of the
5141 offsets implied by the indices to the base address with infinitely
5142 precise signed arithmetic are not an *in bounds* address of that
5143 allocated object. The *in bounds* addresses for an allocated object are
5144 all the addresses that point into the object, plus the address one byte
5145 past the end. In cases where the base is a vector of pointers the
5146 ``inbounds`` keyword applies to each of the computations element-wise.
5148 If the ``inbounds`` keyword is not present, the offsets are added to the
5149 base address with silently-wrapping two's complement arithmetic. If the
5150 offsets have a different width from the pointer, they are sign-extended
5151 or truncated to the width of the pointer. The result value of the
5152 ``getelementptr`` may be outside the object pointed to by the base
5153 pointer. The result value may not necessarily be used to access memory
5154 though, even if it happens to point into allocated storage. See the
5155 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5158 The getelementptr instruction is often confusing. For some more insight
5159 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5164 .. code-block:: llvm
5166 ; yields [12 x i8]*:aptr
5167 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5169 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5171 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5173 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5175 In cases where the pointer argument is a vector of pointers, each index
5176 must be a vector with the same number of elements. For example:
5178 .. code-block:: llvm
5180 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5182 Conversion Operations
5183 ---------------------
5185 The instructions in this category are the conversion instructions
5186 (casting) which all take a single operand and a type. They perform
5187 various bit conversions on the operand.
5189 '``trunc .. to``' Instruction
5190 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5197 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5202 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5207 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5208 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5209 of the same number of integers. The bit size of the ``value`` must be
5210 larger than the bit size of the destination type, ``ty2``. Equal sized
5211 types are not allowed.
5216 The '``trunc``' instruction truncates the high order bits in ``value``
5217 and converts the remaining bits to ``ty2``. Since the source size must
5218 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5219 It will always truncate bits.
5224 .. code-block:: llvm
5226 %X = trunc i32 257 to i8 ; yields i8:1
5227 %Y = trunc i32 123 to i1 ; yields i1:true
5228 %Z = trunc i32 122 to i1 ; yields i1:false
5229 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5231 '``zext .. to``' Instruction
5232 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5239 <result> = zext <ty> <value> to <ty2> ; yields ty2
5244 The '``zext``' instruction zero extends its operand to type ``ty2``.
5249 The '``zext``' instruction takes a value to cast, and a type to cast it
5250 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5251 the same number of integers. The bit size of the ``value`` must be
5252 smaller than the bit size of the destination type, ``ty2``.
5257 The ``zext`` fills the high order bits of the ``value`` with zero bits
5258 until it reaches the size of the destination type, ``ty2``.
5260 When zero extending from i1, the result will always be either 0 or 1.
5265 .. code-block:: llvm
5267 %X = zext i32 257 to i64 ; yields i64:257
5268 %Y = zext i1 true to i32 ; yields i32:1
5269 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5271 '``sext .. to``' Instruction
5272 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5279 <result> = sext <ty> <value> to <ty2> ; yields ty2
5284 The '``sext``' sign extends ``value`` to the type ``ty2``.
5289 The '``sext``' instruction takes a value to cast, and a type to cast it
5290 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5291 the same number of integers. The bit size of the ``value`` must be
5292 smaller than the bit size of the destination type, ``ty2``.
5297 The '``sext``' instruction performs a sign extension by copying the sign
5298 bit (highest order bit) of the ``value`` until it reaches the bit size
5299 of the type ``ty2``.
5301 When sign extending from i1, the extension always results in -1 or 0.
5306 .. code-block:: llvm
5308 %X = sext i8 -1 to i16 ; yields i16 :65535
5309 %Y = sext i1 true to i32 ; yields i32:-1
5310 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5312 '``fptrunc .. to``' Instruction
5313 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5320 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5325 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5330 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5331 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5332 The size of ``value`` must be larger than the size of ``ty2``. This
5333 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5338 The '``fptrunc``' instruction truncates a ``value`` from a larger
5339 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5340 point <t_floating>` type. If the value cannot fit within the
5341 destination type, ``ty2``, then the results are undefined.
5346 .. code-block:: llvm
5348 %X = fptrunc double 123.0 to float ; yields float:123.0
5349 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5351 '``fpext .. to``' Instruction
5352 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5359 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5364 The '``fpext``' extends a floating point ``value`` to a larger floating
5370 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5371 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5372 to. The source type must be smaller than the destination type.
5377 The '``fpext``' instruction extends the ``value`` from a smaller
5378 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5379 point <t_floating>` type. The ``fpext`` cannot be used to make a
5380 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5381 *no-op cast* for a floating point cast.
5386 .. code-block:: llvm
5388 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5389 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5391 '``fptoui .. to``' Instruction
5392 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5399 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5404 The '``fptoui``' converts a floating point ``value`` to its unsigned
5405 integer equivalent of type ``ty2``.
5410 The '``fptoui``' instruction takes a value to cast, which must be a
5411 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5412 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5413 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5414 type with the same number of elements as ``ty``
5419 The '``fptoui``' instruction converts its :ref:`floating
5420 point <t_floating>` operand into the nearest (rounding towards zero)
5421 unsigned integer value. If the value cannot fit in ``ty2``, the results
5427 .. code-block:: llvm
5429 %X = fptoui double 123.0 to i32 ; yields i32:123
5430 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5431 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5433 '``fptosi .. to``' Instruction
5434 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5441 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5446 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5447 ``value`` to type ``ty2``.
5452 The '``fptosi``' instruction takes a value to cast, which must be a
5453 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5454 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5455 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5456 type with the same number of elements as ``ty``
5461 The '``fptosi``' instruction converts its :ref:`floating
5462 point <t_floating>` operand into the nearest (rounding towards zero)
5463 signed integer value. If the value cannot fit in ``ty2``, the results
5469 .. code-block:: llvm
5471 %X = fptosi double -123.0 to i32 ; yields i32:-123
5472 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5473 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5475 '``uitofp .. to``' Instruction
5476 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5483 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5488 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5489 and converts that value to the ``ty2`` type.
5494 The '``uitofp``' instruction takes a value to cast, which must be a
5495 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5496 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5497 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5498 type with the same number of elements as ``ty``
5503 The '``uitofp``' instruction interprets its operand as an unsigned
5504 integer quantity and converts it to the corresponding floating point
5505 value. If the value cannot fit in the floating point value, the results
5511 .. code-block:: llvm
5513 %X = uitofp i32 257 to float ; yields float:257.0
5514 %Y = uitofp i8 -1 to double ; yields double:255.0
5516 '``sitofp .. to``' Instruction
5517 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5524 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5529 The '``sitofp``' instruction regards ``value`` as a signed integer and
5530 converts that value to the ``ty2`` type.
5535 The '``sitofp``' instruction takes a value to cast, which must be a
5536 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5537 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5538 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5539 type with the same number of elements as ``ty``
5544 The '``sitofp``' instruction interprets its operand as a signed integer
5545 quantity and converts it to the corresponding floating point value. If
5546 the value cannot fit in the floating point value, the results are
5552 .. code-block:: llvm
5554 %X = sitofp i32 257 to float ; yields float:257.0
5555 %Y = sitofp i8 -1 to double ; yields double:-1.0
5559 '``ptrtoint .. to``' Instruction
5560 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5567 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5572 The '``ptrtoint``' instruction converts the pointer or a vector of
5573 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5578 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5579 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5580 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5581 a vector of integers type.
5586 The '``ptrtoint``' instruction converts ``value`` to integer type
5587 ``ty2`` by interpreting the pointer value as an integer and either
5588 truncating or zero extending that value to the size of the integer type.
5589 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5590 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5591 the same size, then nothing is done (*no-op cast*) other than a type
5597 .. code-block:: llvm
5599 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5600 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5601 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5605 '``inttoptr .. to``' Instruction
5606 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5613 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5618 The '``inttoptr``' instruction converts an integer ``value`` to a
5619 pointer type, ``ty2``.
5624 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5625 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5631 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5632 applying either a zero extension or a truncation depending on the size
5633 of the integer ``value``. If ``value`` is larger than the size of a
5634 pointer then a truncation is done. If ``value`` is smaller than the size
5635 of a pointer then a zero extension is done. If they are the same size,
5636 nothing is done (*no-op cast*).
5641 .. code-block:: llvm
5643 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5644 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5645 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5646 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5650 '``bitcast .. to``' Instruction
5651 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5658 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5663 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5669 The '``bitcast``' instruction takes a value to cast, which must be a
5670 non-aggregate first class value, and a type to cast it to, which must
5671 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5672 bit sizes of ``value`` and the destination type, ``ty2``, must be
5673 identical. If the source type is a pointer, the destination type must
5674 also be a pointer of the same size. This instruction supports bitwise
5675 conversion of vectors to integers and to vectors of other types (as
5676 long as they have the same size).
5681 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5682 is always a *no-op cast* because no bits change with this
5683 conversion. The conversion is done as if the ``value`` had been stored
5684 to memory and read back as type ``ty2``. Pointer (or vector of
5685 pointers) types may only be converted to other pointer (or vector of
5686 pointers) types with the same address space through this instruction.
5687 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5688 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5693 .. code-block:: llvm
5695 %X = bitcast i8 255 to i8 ; yields i8 :-1
5696 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5697 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5698 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5700 .. _i_addrspacecast:
5702 '``addrspacecast .. to``' Instruction
5703 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5710 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5715 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5716 address space ``n`` to type ``pty2`` in address space ``m``.
5721 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5722 to cast and a pointer type to cast it to, which must have a different
5728 The '``addrspacecast``' instruction converts the pointer value
5729 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5730 value modification, depending on the target and the address space
5731 pair. Pointer conversions within the same address space must be
5732 performed with the ``bitcast`` instruction. Note that if the address space
5733 conversion is legal then both result and operand refer to the same memory
5739 .. code-block:: llvm
5741 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5742 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5743 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5750 The instructions in this category are the "miscellaneous" instructions,
5751 which defy better classification.
5755 '``icmp``' Instruction
5756 ^^^^^^^^^^^^^^^^^^^^^^
5763 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5768 The '``icmp``' instruction returns a boolean value or a vector of
5769 boolean values based on comparison of its two integer, integer vector,
5770 pointer, or pointer vector operands.
5775 The '``icmp``' instruction takes three operands. The first operand is
5776 the condition code indicating the kind of comparison to perform. It is
5777 not a value, just a keyword. The possible condition code are:
5780 #. ``ne``: not equal
5781 #. ``ugt``: unsigned greater than
5782 #. ``uge``: unsigned greater or equal
5783 #. ``ult``: unsigned less than
5784 #. ``ule``: unsigned less or equal
5785 #. ``sgt``: signed greater than
5786 #. ``sge``: signed greater or equal
5787 #. ``slt``: signed less than
5788 #. ``sle``: signed less or equal
5790 The remaining two arguments must be :ref:`integer <t_integer>` or
5791 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5792 must also be identical types.
5797 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5798 code given as ``cond``. The comparison performed always yields either an
5799 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5801 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5802 otherwise. No sign interpretation is necessary or performed.
5803 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5804 otherwise. No sign interpretation is necessary or performed.
5805 #. ``ugt``: interprets the operands as unsigned values and yields
5806 ``true`` if ``op1`` is greater than ``op2``.
5807 #. ``uge``: interprets the operands as unsigned values and yields
5808 ``true`` if ``op1`` is greater than or equal to ``op2``.
5809 #. ``ult``: interprets the operands as unsigned values and yields
5810 ``true`` if ``op1`` is less than ``op2``.
5811 #. ``ule``: interprets the operands as unsigned values and yields
5812 ``true`` if ``op1`` is less than or equal to ``op2``.
5813 #. ``sgt``: interprets the operands as signed values and yields ``true``
5814 if ``op1`` is greater than ``op2``.
5815 #. ``sge``: interprets the operands as signed values and yields ``true``
5816 if ``op1`` is greater than or equal to ``op2``.
5817 #. ``slt``: interprets the operands as signed values and yields ``true``
5818 if ``op1`` is less than ``op2``.
5819 #. ``sle``: interprets the operands as signed values and yields ``true``
5820 if ``op1`` is less than or equal to ``op2``.
5822 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5823 are compared as if they were integers.
5825 If the operands are integer vectors, then they are compared element by
5826 element. The result is an ``i1`` vector with the same number of elements
5827 as the values being compared. Otherwise, the result is an ``i1``.
5832 .. code-block:: llvm
5834 <result> = icmp eq i32 4, 5 ; yields: result=false
5835 <result> = icmp ne float* %X, %X ; yields: result=false
5836 <result> = icmp ult i16 4, 5 ; yields: result=true
5837 <result> = icmp sgt i16 4, 5 ; yields: result=false
5838 <result> = icmp ule i16 -4, 5 ; yields: result=false
5839 <result> = icmp sge i16 4, 5 ; yields: result=false
5841 Note that the code generator does not yet support vector types with the
5842 ``icmp`` instruction.
5846 '``fcmp``' Instruction
5847 ^^^^^^^^^^^^^^^^^^^^^^
5854 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5859 The '``fcmp``' instruction returns a boolean value or vector of boolean
5860 values based on comparison of its operands.
5862 If the operands are floating point scalars, then the result type is a
5863 boolean (:ref:`i1 <t_integer>`).
5865 If the operands are floating point vectors, then the result type is a
5866 vector of boolean with the same number of elements as the operands being
5872 The '``fcmp``' instruction takes three operands. The first operand is
5873 the condition code indicating the kind of comparison to perform. It is
5874 not a value, just a keyword. The possible condition code are:
5876 #. ``false``: no comparison, always returns false
5877 #. ``oeq``: ordered and equal
5878 #. ``ogt``: ordered and greater than
5879 #. ``oge``: ordered and greater than or equal
5880 #. ``olt``: ordered and less than
5881 #. ``ole``: ordered and less than or equal
5882 #. ``one``: ordered and not equal
5883 #. ``ord``: ordered (no nans)
5884 #. ``ueq``: unordered or equal
5885 #. ``ugt``: unordered or greater than
5886 #. ``uge``: unordered or greater than or equal
5887 #. ``ult``: unordered or less than
5888 #. ``ule``: unordered or less than or equal
5889 #. ``une``: unordered or not equal
5890 #. ``uno``: unordered (either nans)
5891 #. ``true``: no comparison, always returns true
5893 *Ordered* means that neither operand is a QNAN while *unordered* means
5894 that either operand may be a QNAN.
5896 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5897 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5898 type. They must have identical types.
5903 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5904 condition code given as ``cond``. If the operands are vectors, then the
5905 vectors are compared element by element. Each comparison performed
5906 always yields an :ref:`i1 <t_integer>` result, as follows:
5908 #. ``false``: always yields ``false``, regardless of operands.
5909 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5910 is equal to ``op2``.
5911 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5912 is greater than ``op2``.
5913 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5914 is greater than or equal to ``op2``.
5915 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5916 is less than ``op2``.
5917 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5918 is less than or equal to ``op2``.
5919 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5920 is not equal to ``op2``.
5921 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5922 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5924 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5925 greater than ``op2``.
5926 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5927 greater than or equal to ``op2``.
5928 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5930 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5931 less than or equal to ``op2``.
5932 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5933 not equal to ``op2``.
5934 #. ``uno``: yields ``true`` if either operand is a QNAN.
5935 #. ``true``: always yields ``true``, regardless of operands.
5940 .. code-block:: llvm
5942 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5943 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5944 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5945 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5947 Note that the code generator does not yet support vector types with the
5948 ``fcmp`` instruction.
5952 '``phi``' Instruction
5953 ^^^^^^^^^^^^^^^^^^^^^
5960 <result> = phi <ty> [ <val0>, <label0>], ...
5965 The '``phi``' instruction is used to implement the φ node in the SSA
5966 graph representing the function.
5971 The type of the incoming values is specified with the first type field.
5972 After this, the '``phi``' instruction takes a list of pairs as
5973 arguments, with one pair for each predecessor basic block of the current
5974 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5975 the value arguments to the PHI node. Only labels may be used as the
5978 There must be no non-phi instructions between the start of a basic block
5979 and the PHI instructions: i.e. PHI instructions must be first in a basic
5982 For the purposes of the SSA form, the use of each incoming value is
5983 deemed to occur on the edge from the corresponding predecessor block to
5984 the current block (but after any definition of an '``invoke``'
5985 instruction's return value on the same edge).
5990 At runtime, the '``phi``' instruction logically takes on the value
5991 specified by the pair corresponding to the predecessor basic block that
5992 executed just prior to the current block.
5997 .. code-block:: llvm
5999 Loop: ; Infinite loop that counts from 0 on up...
6000 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6001 %nextindvar = add i32 %indvar, 1
6006 '``select``' Instruction
6007 ^^^^^^^^^^^^^^^^^^^^^^^^
6014 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6016 selty is either i1 or {<N x i1>}
6021 The '``select``' instruction is used to choose one value based on a
6022 condition, without branching.
6027 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6028 values indicating the condition, and two values of the same :ref:`first
6029 class <t_firstclass>` type. If the val1/val2 are vectors and the
6030 condition is a scalar, then entire vectors are selected, not individual
6036 If the condition is an i1 and it evaluates to 1, the instruction returns
6037 the first value argument; otherwise, it returns the second value
6040 If the condition is a vector of i1, then the value arguments must be
6041 vectors of the same size, and the selection is done element by element.
6046 .. code-block:: llvm
6048 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6052 '``call``' Instruction
6053 ^^^^^^^^^^^^^^^^^^^^^^
6060 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6065 The '``call``' instruction represents a simple function call.
6070 This instruction requires several arguments:
6072 #. The optional "tail" marker indicates that the callee function does
6073 not access any allocas or varargs in the caller. Note that calls may
6074 be marked "tail" even if they do not occur before a
6075 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6076 function call is eligible for tail call optimization, but `might not
6077 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6078 The code generator may optimize calls marked "tail" with either 1)
6079 automatic `sibling call
6080 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6081 callee have matching signatures, or 2) forced tail call optimization
6082 when the following extra requirements are met:
6084 - Caller and callee both have the calling convention ``fastcc``.
6085 - The call is in tail position (ret immediately follows call and ret
6086 uses value of call or is void).
6087 - Option ``-tailcallopt`` is enabled, or
6088 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6089 - `Platform specific constraints are
6090 met. <CodeGenerator.html#tailcallopt>`_
6092 #. The optional "cconv" marker indicates which :ref:`calling
6093 convention <callingconv>` the call should use. If none is
6094 specified, the call defaults to using C calling conventions. The
6095 calling convention of the call must match the calling convention of
6096 the target function, or else the behavior is undefined.
6097 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6098 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6100 #. '``ty``': the type of the call instruction itself which is also the
6101 type of the return value. Functions that return no value are marked
6103 #. '``fnty``': shall be the signature of the pointer to function value
6104 being invoked. The argument types must match the types implied by
6105 this signature. This type can be omitted if the function is not
6106 varargs and if the function type does not return a pointer to a
6108 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6109 be invoked. In most cases, this is a direct function invocation, but
6110 indirect ``call``'s are just as possible, calling an arbitrary pointer
6112 #. '``function args``': argument list whose types match the function
6113 signature argument types and parameter attributes. All arguments must
6114 be of :ref:`first class <t_firstclass>` type. If the function signature
6115 indicates the function accepts a variable number of arguments, the
6116 extra arguments can be specified.
6117 #. The optional :ref:`function attributes <fnattrs>` list. Only
6118 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6119 attributes are valid here.
6124 The '``call``' instruction is used to cause control flow to transfer to
6125 a specified function, with its incoming arguments bound to the specified
6126 values. Upon a '``ret``' instruction in the called function, control
6127 flow continues with the instruction after the function call, and the
6128 return value of the function is bound to the result argument.
6133 .. code-block:: llvm
6135 %retval = call i32 @test(i32 %argc)
6136 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6137 %X = tail call i32 @foo() ; yields i32
6138 %Y = tail call fastcc i32 @foo() ; yields i32
6139 call void %foo(i8 97 signext)
6141 %struct.A = type { i32, i8 }
6142 %r = call %struct.A @foo() ; yields { 32, i8 }
6143 %gr = extractvalue %struct.A %r, 0 ; yields i32
6144 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6145 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6146 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6148 llvm treats calls to some functions with names and arguments that match
6149 the standard C99 library as being the C99 library functions, and may
6150 perform optimizations or generate code for them under that assumption.
6151 This is something we'd like to change in the future to provide better
6152 support for freestanding environments and non-C-based languages.
6156 '``va_arg``' Instruction
6157 ^^^^^^^^^^^^^^^^^^^^^^^^
6164 <resultval> = va_arg <va_list*> <arglist>, <argty>
6169 The '``va_arg``' instruction is used to access arguments passed through
6170 the "variable argument" area of a function call. It is used to implement
6171 the ``va_arg`` macro in C.
6176 This instruction takes a ``va_list*`` value and the type of the
6177 argument. It returns a value of the specified argument type and
6178 increments the ``va_list`` to point to the next argument. The actual
6179 type of ``va_list`` is target specific.
6184 The '``va_arg``' instruction loads an argument of the specified type
6185 from the specified ``va_list`` and causes the ``va_list`` to point to
6186 the next argument. For more information, see the variable argument
6187 handling :ref:`Intrinsic Functions <int_varargs>`.
6189 It is legal for this instruction to be called in a function which does
6190 not take a variable number of arguments, for example, the ``vfprintf``
6193 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6194 function <intrinsics>` because it takes a type as an argument.
6199 See the :ref:`variable argument processing <int_varargs>` section.
6201 Note that the code generator does not yet fully support va\_arg on many
6202 targets. Also, it does not currently support va\_arg with aggregate
6203 types on any target.
6207 '``landingpad``' Instruction
6208 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6215 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6216 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6218 <clause> := catch <type> <value>
6219 <clause> := filter <array constant type> <array constant>
6224 The '``landingpad``' instruction is used by `LLVM's exception handling
6225 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6226 is a landing pad --- one where the exception lands, and corresponds to the
6227 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6228 defines values supplied by the personality function (``pers_fn``) upon
6229 re-entry to the function. The ``resultval`` has the type ``resultty``.
6234 This instruction takes a ``pers_fn`` value. This is the personality
6235 function associated with the unwinding mechanism. The optional
6236 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6238 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6239 contains the global variable representing the "type" that may be caught
6240 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6241 clause takes an array constant as its argument. Use
6242 "``[0 x i8**] undef``" for a filter which cannot throw. The
6243 '``landingpad``' instruction must contain *at least* one ``clause`` or
6244 the ``cleanup`` flag.
6249 The '``landingpad``' instruction defines the values which are set by the
6250 personality function (``pers_fn``) upon re-entry to the function, and
6251 therefore the "result type" of the ``landingpad`` instruction. As with
6252 calling conventions, how the personality function results are
6253 represented in LLVM IR is target specific.
6255 The clauses are applied in order from top to bottom. If two
6256 ``landingpad`` instructions are merged together through inlining, the
6257 clauses from the calling function are appended to the list of clauses.
6258 When the call stack is being unwound due to an exception being thrown,
6259 the exception is compared against each ``clause`` in turn. If it doesn't
6260 match any of the clauses, and the ``cleanup`` flag is not set, then
6261 unwinding continues further up the call stack.
6263 The ``landingpad`` instruction has several restrictions:
6265 - A landing pad block is a basic block which is the unwind destination
6266 of an '``invoke``' instruction.
6267 - A landing pad block must have a '``landingpad``' instruction as its
6268 first non-PHI instruction.
6269 - There can be only one '``landingpad``' instruction within the landing
6271 - A basic block that is not a landing pad block may not include a
6272 '``landingpad``' instruction.
6273 - All '``landingpad``' instructions in a function must have the same
6274 personality function.
6279 .. code-block:: llvm
6281 ;; A landing pad which can catch an integer.
6282 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6284 ;; A landing pad that is a cleanup.
6285 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6287 ;; A landing pad which can catch an integer and can only throw a double.
6288 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6290 filter [1 x i8**] [@_ZTId]
6297 LLVM supports the notion of an "intrinsic function". These functions
6298 have well known names and semantics and are required to follow certain
6299 restrictions. Overall, these intrinsics represent an extension mechanism
6300 for the LLVM language that does not require changing all of the
6301 transformations in LLVM when adding to the language (or the bitcode
6302 reader/writer, the parser, etc...).
6304 Intrinsic function names must all start with an "``llvm.``" prefix. This
6305 prefix is reserved in LLVM for intrinsic names; thus, function names may
6306 not begin with this prefix. Intrinsic functions must always be external
6307 functions: you cannot define the body of intrinsic functions. Intrinsic
6308 functions may only be used in call or invoke instructions: it is illegal
6309 to take the address of an intrinsic function. Additionally, because
6310 intrinsic functions are part of the LLVM language, it is required if any
6311 are added that they be documented here.
6313 Some intrinsic functions can be overloaded, i.e., the intrinsic
6314 represents a family of functions that perform the same operation but on
6315 different data types. Because LLVM can represent over 8 million
6316 different integer types, overloading is used commonly to allow an
6317 intrinsic function to operate on any integer type. One or more of the
6318 argument types or the result type can be overloaded to accept any
6319 integer type. Argument types may also be defined as exactly matching a
6320 previous argument's type or the result type. This allows an intrinsic
6321 function which accepts multiple arguments, but needs all of them to be
6322 of the same type, to only be overloaded with respect to a single
6323 argument or the result.
6325 Overloaded intrinsics will have the names of its overloaded argument
6326 types encoded into its function name, each preceded by a period. Only
6327 those types which are overloaded result in a name suffix. Arguments
6328 whose type is matched against another type do not. For example, the
6329 ``llvm.ctpop`` function can take an integer of any width and returns an
6330 integer of exactly the same integer width. This leads to a family of
6331 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6332 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6333 overloaded, and only one type suffix is required. Because the argument's
6334 type is matched against the return type, it does not require its own
6337 To learn how to add an intrinsic function, please see the `Extending
6338 LLVM Guide <ExtendingLLVM.html>`_.
6342 Variable Argument Handling Intrinsics
6343 -------------------------------------
6345 Variable argument support is defined in LLVM with the
6346 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6347 functions. These functions are related to the similarly named macros
6348 defined in the ``<stdarg.h>`` header file.
6350 All of these functions operate on arguments that use a target-specific
6351 value type "``va_list``". The LLVM assembly language reference manual
6352 does not define what this type is, so all transformations should be
6353 prepared to handle these functions regardless of the type used.
6355 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6356 variable argument handling intrinsic functions are used.
6358 .. code-block:: llvm
6360 define i32 @test(i32 %X, ...) {
6361 ; Initialize variable argument processing
6363 %ap2 = bitcast i8** %ap to i8*
6364 call void @llvm.va_start(i8* %ap2)
6366 ; Read a single integer argument
6367 %tmp = va_arg i8** %ap, i32
6369 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6371 %aq2 = bitcast i8** %aq to i8*
6372 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6373 call void @llvm.va_end(i8* %aq2)
6375 ; Stop processing of arguments.
6376 call void @llvm.va_end(i8* %ap2)
6380 declare void @llvm.va_start(i8*)
6381 declare void @llvm.va_copy(i8*, i8*)
6382 declare void @llvm.va_end(i8*)
6386 '``llvm.va_start``' Intrinsic
6387 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6394 declare void @llvm.va_start(i8* <arglist>)
6399 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6400 subsequent use by ``va_arg``.
6405 The argument is a pointer to a ``va_list`` element to initialize.
6410 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6411 available in C. In a target-dependent way, it initializes the
6412 ``va_list`` element to which the argument points, so that the next call
6413 to ``va_arg`` will produce the first variable argument passed to the
6414 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6415 to know the last argument of the function as the compiler can figure
6418 '``llvm.va_end``' Intrinsic
6419 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6426 declare void @llvm.va_end(i8* <arglist>)
6431 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6432 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6437 The argument is a pointer to a ``va_list`` to destroy.
6442 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6443 available in C. In a target-dependent way, it destroys the ``va_list``
6444 element to which the argument points. Calls to
6445 :ref:`llvm.va_start <int_va_start>` and
6446 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6451 '``llvm.va_copy``' Intrinsic
6452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6459 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6464 The '``llvm.va_copy``' intrinsic copies the current argument position
6465 from the source argument list to the destination argument list.
6470 The first argument is a pointer to a ``va_list`` element to initialize.
6471 The second argument is a pointer to a ``va_list`` element to copy from.
6476 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6477 available in C. In a target-dependent way, it copies the source
6478 ``va_list`` element into the destination ``va_list`` element. This
6479 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6480 arbitrarily complex and require, for example, memory allocation.
6482 Accurate Garbage Collection Intrinsics
6483 --------------------------------------
6485 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6486 (GC) requires the implementation and generation of these intrinsics.
6487 These intrinsics allow identification of :ref:`GC roots on the
6488 stack <int_gcroot>`, as well as garbage collector implementations that
6489 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6490 Front-ends for type-safe garbage collected languages should generate
6491 these intrinsics to make use of the LLVM garbage collectors. For more
6492 details, see `Accurate Garbage Collection with
6493 LLVM <GarbageCollection.html>`_.
6495 The garbage collection intrinsics only operate on objects in the generic
6496 address space (address space zero).
6500 '``llvm.gcroot``' Intrinsic
6501 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6508 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6513 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6514 the code generator, and allows some metadata to be associated with it.
6519 The first argument specifies the address of a stack object that contains
6520 the root pointer. The second pointer (which must be either a constant or
6521 a global value address) contains the meta-data to be associated with the
6527 At runtime, a call to this intrinsic stores a null pointer into the
6528 "ptrloc" location. At compile-time, the code generator generates
6529 information to allow the runtime to find the pointer at GC safe points.
6530 The '``llvm.gcroot``' intrinsic may only be used in a function which
6531 :ref:`specifies a GC algorithm <gc>`.
6535 '``llvm.gcread``' Intrinsic
6536 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6543 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6548 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6549 locations, allowing garbage collector implementations that require read
6555 The second argument is the address to read from, which should be an
6556 address allocated from the garbage collector. The first object is a
6557 pointer to the start of the referenced object, if needed by the language
6558 runtime (otherwise null).
6563 The '``llvm.gcread``' intrinsic has the same semantics as a load
6564 instruction, but may be replaced with substantially more complex code by
6565 the garbage collector runtime, as needed. The '``llvm.gcread``'
6566 intrinsic may only be used in a function which :ref:`specifies a GC
6571 '``llvm.gcwrite``' Intrinsic
6572 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6579 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6584 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6585 locations, allowing garbage collector implementations that require write
6586 barriers (such as generational or reference counting collectors).
6591 The first argument is the reference to store, the second is the start of
6592 the object to store it to, and the third is the address of the field of
6593 Obj to store to. If the runtime does not require a pointer to the
6594 object, Obj may be null.
6599 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6600 instruction, but may be replaced with substantially more complex code by
6601 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6602 intrinsic may only be used in a function which :ref:`specifies a GC
6605 Code Generator Intrinsics
6606 -------------------------
6608 These intrinsics are provided by LLVM to expose special features that
6609 may only be implemented with code generator support.
6611 '``llvm.returnaddress``' Intrinsic
6612 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6619 declare i8 *@llvm.returnaddress(i32 <level>)
6624 The '``llvm.returnaddress``' intrinsic attempts to compute a
6625 target-specific value indicating the return address of the current
6626 function or one of its callers.
6631 The argument to this intrinsic indicates which function to return the
6632 address for. Zero indicates the calling function, one indicates its
6633 caller, etc. The argument is **required** to be a constant integer
6639 The '``llvm.returnaddress``' intrinsic either returns a pointer
6640 indicating the return address of the specified call frame, or zero if it
6641 cannot be identified. The value returned by this intrinsic is likely to
6642 be incorrect or 0 for arguments other than zero, so it should only be
6643 used for debugging purposes.
6645 Note that calling this intrinsic does not prevent function inlining or
6646 other aggressive transformations, so the value returned may not be that
6647 of the obvious source-language caller.
6649 '``llvm.frameaddress``' Intrinsic
6650 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6657 declare i8* @llvm.frameaddress(i32 <level>)
6662 The '``llvm.frameaddress``' intrinsic attempts to return the
6663 target-specific frame pointer value for the specified stack frame.
6668 The argument to this intrinsic indicates which function to return the
6669 frame pointer for. Zero indicates the calling function, one indicates
6670 its caller, etc. The argument is **required** to be a constant integer
6676 The '``llvm.frameaddress``' intrinsic either returns a pointer
6677 indicating the frame address of the specified call frame, or zero if it
6678 cannot be identified. The value returned by this intrinsic is likely to
6679 be incorrect or 0 for arguments other than zero, so it should only be
6680 used for debugging purposes.
6682 Note that calling this intrinsic does not prevent function inlining or
6683 other aggressive transformations, so the value returned may not be that
6684 of the obvious source-language caller.
6688 '``llvm.stacksave``' Intrinsic
6689 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6696 declare i8* @llvm.stacksave()
6701 The '``llvm.stacksave``' intrinsic is used to remember the current state
6702 of the function stack, for use with
6703 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6704 implementing language features like scoped automatic variable sized
6710 This intrinsic returns a opaque pointer value that can be passed to
6711 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6712 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6713 ``llvm.stacksave``, it effectively restores the state of the stack to
6714 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6715 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6716 were allocated after the ``llvm.stacksave`` was executed.
6718 .. _int_stackrestore:
6720 '``llvm.stackrestore``' Intrinsic
6721 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6728 declare void @llvm.stackrestore(i8* %ptr)
6733 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6734 the function stack to the state it was in when the corresponding
6735 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6736 useful for implementing language features like scoped automatic variable
6737 sized arrays in C99.
6742 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6744 '``llvm.prefetch``' Intrinsic
6745 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6752 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6757 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6758 insert a prefetch instruction if supported; otherwise, it is a noop.
6759 Prefetches have no effect on the behavior of the program but can change
6760 its performance characteristics.
6765 ``address`` is the address to be prefetched, ``rw`` is the specifier
6766 determining if the fetch should be for a read (0) or write (1), and
6767 ``locality`` is a temporal locality specifier ranging from (0) - no
6768 locality, to (3) - extremely local keep in cache. The ``cache type``
6769 specifies whether the prefetch is performed on the data (1) or
6770 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6771 arguments must be constant integers.
6776 This intrinsic does not modify the behavior of the program. In
6777 particular, prefetches cannot trap and do not produce a value. On
6778 targets that support this intrinsic, the prefetch can provide hints to
6779 the processor cache for better performance.
6781 '``llvm.pcmarker``' Intrinsic
6782 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6789 declare void @llvm.pcmarker(i32 <id>)
6794 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6795 Counter (PC) in a region of code to simulators and other tools. The
6796 method is target specific, but it is expected that the marker will use
6797 exported symbols to transmit the PC of the marker. The marker makes no
6798 guarantees that it will remain with any specific instruction after
6799 optimizations. It is possible that the presence of a marker will inhibit
6800 optimizations. The intended use is to be inserted after optimizations to
6801 allow correlations of simulation runs.
6806 ``id`` is a numerical id identifying the marker.
6811 This intrinsic does not modify the behavior of the program. Backends
6812 that do not support this intrinsic may ignore it.
6814 '``llvm.readcyclecounter``' Intrinsic
6815 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6822 declare i64 @llvm.readcyclecounter()
6827 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6828 counter register (or similar low latency, high accuracy clocks) on those
6829 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6830 should map to RPCC. As the backing counters overflow quickly (on the
6831 order of 9 seconds on alpha), this should only be used for small
6837 When directly supported, reading the cycle counter should not modify any
6838 memory. Implementations are allowed to either return a application
6839 specific value or a system wide value. On backends without support, this
6840 is lowered to a constant 0.
6842 Note that runtime support may be conditional on the privilege-level code is
6843 running at and the host platform.
6845 Standard C Library Intrinsics
6846 -----------------------------
6848 LLVM provides intrinsics for a few important standard C library
6849 functions. These intrinsics allow source-language front-ends to pass
6850 information about the alignment of the pointer arguments to the code
6851 generator, providing opportunity for more efficient code generation.
6855 '``llvm.memcpy``' Intrinsic
6856 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6861 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6862 integer bit width and for different address spaces. Not all targets
6863 support all bit widths however.
6867 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6868 i32 <len>, i32 <align>, i1 <isvolatile>)
6869 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6870 i64 <len>, i32 <align>, i1 <isvolatile>)
6875 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6876 source location to the destination location.
6878 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6879 intrinsics do not return a value, takes extra alignment/isvolatile
6880 arguments and the pointers can be in specified address spaces.
6885 The first argument is a pointer to the destination, the second is a
6886 pointer to the source. The third argument is an integer argument
6887 specifying the number of bytes to copy, the fourth argument is the
6888 alignment of the source and destination locations, and the fifth is a
6889 boolean indicating a volatile access.
6891 If the call to this intrinsic has an alignment value that is not 0 or 1,
6892 then the caller guarantees that both the source and destination pointers
6893 are aligned to that boundary.
6895 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6896 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6897 very cleanly specified and it is unwise to depend on it.
6902 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6903 source location to the destination location, which are not allowed to
6904 overlap. It copies "len" bytes of memory over. If the argument is known
6905 to be aligned to some boundary, this can be specified as the fourth
6906 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
6908 '``llvm.memmove``' Intrinsic
6909 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6914 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6915 bit width and for different address space. Not all targets support all
6920 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6921 i32 <len>, i32 <align>, i1 <isvolatile>)
6922 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6923 i64 <len>, i32 <align>, i1 <isvolatile>)
6928 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6929 source location to the destination location. It is similar to the
6930 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6933 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6934 intrinsics do not return a value, takes extra alignment/isvolatile
6935 arguments and the pointers can be in specified address spaces.
6940 The first argument is a pointer to the destination, the second is a
6941 pointer to the source. The third argument is an integer argument
6942 specifying the number of bytes to copy, the fourth argument is the
6943 alignment of the source and destination locations, and the fifth is a
6944 boolean indicating a volatile access.
6946 If the call to this intrinsic has an alignment value that is not 0 or 1,
6947 then the caller guarantees that the source and destination pointers are
6948 aligned to that boundary.
6950 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6951 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6952 not very cleanly specified and it is unwise to depend on it.
6957 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6958 source location to the destination location, which may overlap. It
6959 copies "len" bytes of memory over. If the argument is known to be
6960 aligned to some boundary, this can be specified as the fourth argument,
6961 otherwise it should be set to 0 or 1 (both meaning no alignment).
6963 '``llvm.memset.*``' Intrinsics
6964 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6969 This is an overloaded intrinsic. You can use llvm.memset on any integer
6970 bit width and for different address spaces. However, not all targets
6971 support all bit widths.
6975 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6976 i32 <len>, i32 <align>, i1 <isvolatile>)
6977 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6978 i64 <len>, i32 <align>, i1 <isvolatile>)
6983 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6984 particular byte value.
6986 Note that, unlike the standard libc function, the ``llvm.memset``
6987 intrinsic does not return a value and takes extra alignment/volatile
6988 arguments. Also, the destination can be in an arbitrary address space.
6993 The first argument is a pointer to the destination to fill, the second
6994 is the byte value with which to fill it, the third argument is an
6995 integer argument specifying the number of bytes to fill, and the fourth
6996 argument is the known alignment of the destination location.
6998 If the call to this intrinsic has an alignment value that is not 0 or 1,
6999 then the caller guarantees that the destination pointer is aligned to
7002 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7003 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7004 very cleanly specified and it is unwise to depend on it.
7009 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7010 at the destination location. If the argument is known to be aligned to
7011 some boundary, this can be specified as the fourth argument, otherwise
7012 it should be set to 0 or 1 (both meaning no alignment).
7014 '``llvm.sqrt.*``' Intrinsic
7015 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7020 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7021 floating point or vector of floating point type. Not all targets support
7026 declare float @llvm.sqrt.f32(float %Val)
7027 declare double @llvm.sqrt.f64(double %Val)
7028 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7029 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7030 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7035 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7036 returning the same value as the libm '``sqrt``' functions would. Unlike
7037 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7038 negative numbers other than -0.0 (which allows for better optimization,
7039 because there is no need to worry about errno being set).
7040 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7045 The argument and return value are floating point numbers of the same
7051 This function returns the sqrt of the specified operand if it is a
7052 nonnegative floating point number.
7054 '``llvm.powi.*``' Intrinsic
7055 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7060 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7061 floating point or vector of floating point type. Not all targets support
7066 declare float @llvm.powi.f32(float %Val, i32 %power)
7067 declare double @llvm.powi.f64(double %Val, i32 %power)
7068 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7069 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7070 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7075 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7076 specified (positive or negative) power. The order of evaluation of
7077 multiplications is not defined. When a vector of floating point type is
7078 used, the second argument remains a scalar integer value.
7083 The second argument is an integer power, and the first is a value to
7084 raise to that power.
7089 This function returns the first value raised to the second power with an
7090 unspecified sequence of rounding operations.
7092 '``llvm.sin.*``' Intrinsic
7093 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7098 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7099 floating point or vector of floating point type. Not all targets support
7104 declare float @llvm.sin.f32(float %Val)
7105 declare double @llvm.sin.f64(double %Val)
7106 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7107 declare fp128 @llvm.sin.f128(fp128 %Val)
7108 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7113 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7118 The argument and return value are floating point numbers of the same
7124 This function returns the sine of the specified operand, returning the
7125 same values as the libm ``sin`` functions would, and handles error
7126 conditions in the same way.
7128 '``llvm.cos.*``' Intrinsic
7129 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7134 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7135 floating point or vector of floating point type. Not all targets support
7140 declare float @llvm.cos.f32(float %Val)
7141 declare double @llvm.cos.f64(double %Val)
7142 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7143 declare fp128 @llvm.cos.f128(fp128 %Val)
7144 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7149 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7154 The argument and return value are floating point numbers of the same
7160 This function returns the cosine of the specified operand, returning the
7161 same values as the libm ``cos`` functions would, and handles error
7162 conditions in the same way.
7164 '``llvm.pow.*``' Intrinsic
7165 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7170 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7171 floating point or vector of floating point type. Not all targets support
7176 declare float @llvm.pow.f32(float %Val, float %Power)
7177 declare double @llvm.pow.f64(double %Val, double %Power)
7178 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7179 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7180 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7185 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7186 specified (positive or negative) power.
7191 The second argument is a floating point power, and the first is a value
7192 to raise to that power.
7197 This function returns the first value raised to the second power,
7198 returning the same values as the libm ``pow`` functions would, and
7199 handles error conditions in the same way.
7201 '``llvm.exp.*``' Intrinsic
7202 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7207 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7208 floating point or vector of floating point type. Not all targets support
7213 declare float @llvm.exp.f32(float %Val)
7214 declare double @llvm.exp.f64(double %Val)
7215 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7216 declare fp128 @llvm.exp.f128(fp128 %Val)
7217 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7222 The '``llvm.exp.*``' intrinsics perform the exp function.
7227 The argument and return value are floating point numbers of the same
7233 This function returns the same values as the libm ``exp`` functions
7234 would, and handles error conditions in the same way.
7236 '``llvm.exp2.*``' Intrinsic
7237 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7242 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7243 floating point or vector of floating point type. Not all targets support
7248 declare float @llvm.exp2.f32(float %Val)
7249 declare double @llvm.exp2.f64(double %Val)
7250 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7251 declare fp128 @llvm.exp2.f128(fp128 %Val)
7252 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7257 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7262 The argument and return value are floating point numbers of the same
7268 This function returns the same values as the libm ``exp2`` functions
7269 would, and handles error conditions in the same way.
7271 '``llvm.log.*``' Intrinsic
7272 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7277 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7278 floating point or vector of floating point type. Not all targets support
7283 declare float @llvm.log.f32(float %Val)
7284 declare double @llvm.log.f64(double %Val)
7285 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7286 declare fp128 @llvm.log.f128(fp128 %Val)
7287 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7292 The '``llvm.log.*``' intrinsics perform the log function.
7297 The argument and return value are floating point numbers of the same
7303 This function returns the same values as the libm ``log`` functions
7304 would, and handles error conditions in the same way.
7306 '``llvm.log10.*``' Intrinsic
7307 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7312 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7313 floating point or vector of floating point type. Not all targets support
7318 declare float @llvm.log10.f32(float %Val)
7319 declare double @llvm.log10.f64(double %Val)
7320 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7321 declare fp128 @llvm.log10.f128(fp128 %Val)
7322 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7327 The '``llvm.log10.*``' intrinsics perform the log10 function.
7332 The argument and return value are floating point numbers of the same
7338 This function returns the same values as the libm ``log10`` functions
7339 would, and handles error conditions in the same way.
7341 '``llvm.log2.*``' Intrinsic
7342 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7347 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7348 floating point or vector of floating point type. Not all targets support
7353 declare float @llvm.log2.f32(float %Val)
7354 declare double @llvm.log2.f64(double %Val)
7355 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7356 declare fp128 @llvm.log2.f128(fp128 %Val)
7357 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7362 The '``llvm.log2.*``' intrinsics perform the log2 function.
7367 The argument and return value are floating point numbers of the same
7373 This function returns the same values as the libm ``log2`` functions
7374 would, and handles error conditions in the same way.
7376 '``llvm.fma.*``' Intrinsic
7377 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7382 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7383 floating point or vector of floating point type. Not all targets support
7388 declare float @llvm.fma.f32(float %a, float %b, float %c)
7389 declare double @llvm.fma.f64(double %a, double %b, double %c)
7390 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7391 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7392 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7397 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7403 The argument and return value are floating point numbers of the same
7409 This function returns the same values as the libm ``fma`` functions
7412 '``llvm.fabs.*``' Intrinsic
7413 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7418 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7419 floating point or vector of floating point type. Not all targets support
7424 declare float @llvm.fabs.f32(float %Val)
7425 declare double @llvm.fabs.f64(double %Val)
7426 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7427 declare fp128 @llvm.fabs.f128(fp128 %Val)
7428 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7433 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7439 The argument and return value are floating point numbers of the same
7445 This function returns the same values as the libm ``fabs`` functions
7446 would, and handles error conditions in the same way.
7448 '``llvm.copysign.*``' Intrinsic
7449 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7454 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7455 floating point or vector of floating point type. Not all targets support
7460 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7461 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7462 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7463 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7464 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7469 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7470 first operand and the sign of the second operand.
7475 The arguments and return value are floating point numbers of the same
7481 This function returns the same values as the libm ``copysign``
7482 functions would, and handles error conditions in the same way.
7484 '``llvm.floor.*``' Intrinsic
7485 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7490 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7491 floating point or vector of floating point type. Not all targets support
7496 declare float @llvm.floor.f32(float %Val)
7497 declare double @llvm.floor.f64(double %Val)
7498 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7499 declare fp128 @llvm.floor.f128(fp128 %Val)
7500 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7505 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7510 The argument and return value are floating point numbers of the same
7516 This function returns the same values as the libm ``floor`` functions
7517 would, and handles error conditions in the same way.
7519 '``llvm.ceil.*``' Intrinsic
7520 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7525 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7526 floating point or vector of floating point type. Not all targets support
7531 declare float @llvm.ceil.f32(float %Val)
7532 declare double @llvm.ceil.f64(double %Val)
7533 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7534 declare fp128 @llvm.ceil.f128(fp128 %Val)
7535 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7540 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7545 The argument and return value are floating point numbers of the same
7551 This function returns the same values as the libm ``ceil`` functions
7552 would, and handles error conditions in the same way.
7554 '``llvm.trunc.*``' Intrinsic
7555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7560 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7561 floating point or vector of floating point type. Not all targets support
7566 declare float @llvm.trunc.f32(float %Val)
7567 declare double @llvm.trunc.f64(double %Val)
7568 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7569 declare fp128 @llvm.trunc.f128(fp128 %Val)
7570 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7575 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7576 nearest integer not larger in magnitude than the operand.
7581 The argument and return value are floating point numbers of the same
7587 This function returns the same values as the libm ``trunc`` functions
7588 would, and handles error conditions in the same way.
7590 '``llvm.rint.*``' Intrinsic
7591 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7596 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7597 floating point or vector of floating point type. Not all targets support
7602 declare float @llvm.rint.f32(float %Val)
7603 declare double @llvm.rint.f64(double %Val)
7604 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7605 declare fp128 @llvm.rint.f128(fp128 %Val)
7606 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7611 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7612 nearest integer. It may raise an inexact floating-point exception if the
7613 operand isn't an integer.
7618 The argument and return value are floating point numbers of the same
7624 This function returns the same values as the libm ``rint`` functions
7625 would, and handles error conditions in the same way.
7627 '``llvm.nearbyint.*``' Intrinsic
7628 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7633 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7634 floating point or vector of floating point type. Not all targets support
7639 declare float @llvm.nearbyint.f32(float %Val)
7640 declare double @llvm.nearbyint.f64(double %Val)
7641 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7642 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7643 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7648 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7654 The argument and return value are floating point numbers of the same
7660 This function returns the same values as the libm ``nearbyint``
7661 functions would, and handles error conditions in the same way.
7663 '``llvm.round.*``' Intrinsic
7664 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7669 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7670 floating point or vector of floating point type. Not all targets support
7675 declare float @llvm.round.f32(float %Val)
7676 declare double @llvm.round.f64(double %Val)
7677 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7678 declare fp128 @llvm.round.f128(fp128 %Val)
7679 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7684 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7690 The argument and return value are floating point numbers of the same
7696 This function returns the same values as the libm ``round``
7697 functions would, and handles error conditions in the same way.
7699 Bit Manipulation Intrinsics
7700 ---------------------------
7702 LLVM provides intrinsics for a few important bit manipulation
7703 operations. These allow efficient code generation for some algorithms.
7705 '``llvm.bswap.*``' Intrinsics
7706 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7711 This is an overloaded intrinsic function. You can use bswap on any
7712 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7716 declare i16 @llvm.bswap.i16(i16 <id>)
7717 declare i32 @llvm.bswap.i32(i32 <id>)
7718 declare i64 @llvm.bswap.i64(i64 <id>)
7723 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7724 values with an even number of bytes (positive multiple of 16 bits).
7725 These are useful for performing operations on data that is not in the
7726 target's native byte order.
7731 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7732 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7733 intrinsic returns an i32 value that has the four bytes of the input i32
7734 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7735 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7736 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7737 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7740 '``llvm.ctpop.*``' Intrinsic
7741 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7746 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7747 bit width, or on any vector with integer elements. Not all targets
7748 support all bit widths or vector types, however.
7752 declare i8 @llvm.ctpop.i8(i8 <src>)
7753 declare i16 @llvm.ctpop.i16(i16 <src>)
7754 declare i32 @llvm.ctpop.i32(i32 <src>)
7755 declare i64 @llvm.ctpop.i64(i64 <src>)
7756 declare i256 @llvm.ctpop.i256(i256 <src>)
7757 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7762 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7768 The only argument is the value to be counted. The argument may be of any
7769 integer type, or a vector with integer elements. The return type must
7770 match the argument type.
7775 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7776 each element of a vector.
7778 '``llvm.ctlz.*``' Intrinsic
7779 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7784 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7785 integer bit width, or any vector whose elements are integers. Not all
7786 targets support all bit widths or vector types, however.
7790 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7791 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7792 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7793 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7794 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7795 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7800 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7801 leading zeros in a variable.
7806 The first argument is the value to be counted. This argument may be of
7807 any integer type, or a vectory with integer element type. The return
7808 type must match the first argument type.
7810 The second argument must be a constant and is a flag to indicate whether
7811 the intrinsic should ensure that a zero as the first argument produces a
7812 defined result. Historically some architectures did not provide a
7813 defined result for zero values as efficiently, and many algorithms are
7814 now predicated on avoiding zero-value inputs.
7819 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7820 zeros in a variable, or within each element of the vector. If
7821 ``src == 0`` then the result is the size in bits of the type of ``src``
7822 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7823 ``llvm.ctlz(i32 2) = 30``.
7825 '``llvm.cttz.*``' Intrinsic
7826 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7831 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7832 integer bit width, or any vector of integer elements. Not all targets
7833 support all bit widths or vector types, however.
7837 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7838 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7839 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7840 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7841 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7842 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7847 The '``llvm.cttz``' family of intrinsic functions counts the number of
7853 The first argument is the value to be counted. This argument may be of
7854 any integer type, or a vectory with integer element type. The return
7855 type must match the first argument type.
7857 The second argument must be a constant and is a flag to indicate whether
7858 the intrinsic should ensure that a zero as the first argument produces a
7859 defined result. Historically some architectures did not provide a
7860 defined result for zero values as efficiently, and many algorithms are
7861 now predicated on avoiding zero-value inputs.
7866 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7867 zeros in a variable, or within each element of a vector. If ``src == 0``
7868 then the result is the size in bits of the type of ``src`` if
7869 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7870 ``llvm.cttz(2) = 1``.
7872 Arithmetic with Overflow Intrinsics
7873 -----------------------------------
7875 LLVM provides intrinsics for some arithmetic with overflow operations.
7877 '``llvm.sadd.with.overflow.*``' Intrinsics
7878 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7883 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7884 on any integer bit width.
7888 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7889 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7890 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7895 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7896 a signed addition of the two arguments, and indicate whether an overflow
7897 occurred during the signed summation.
7902 The arguments (%a and %b) and the first element of the result structure
7903 may be of integer types of any bit width, but they must have the same
7904 bit width. The second element of the result structure must be of type
7905 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7911 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7912 a signed addition of the two variables. They return a structure --- the
7913 first element of which is the signed summation, and the second element
7914 of which is a bit specifying if the signed summation resulted in an
7920 .. code-block:: llvm
7922 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7923 %sum = extractvalue {i32, i1} %res, 0
7924 %obit = extractvalue {i32, i1} %res, 1
7925 br i1 %obit, label %overflow, label %normal
7927 '``llvm.uadd.with.overflow.*``' Intrinsics
7928 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7933 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7934 on any integer bit width.
7938 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7939 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7940 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7945 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7946 an unsigned addition of the two arguments, and indicate whether a carry
7947 occurred during the unsigned summation.
7952 The arguments (%a and %b) and the first element of the result structure
7953 may be of integer types of any bit width, but they must have the same
7954 bit width. The second element of the result structure must be of type
7955 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7961 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7962 an unsigned addition of the two arguments. They return a structure --- the
7963 first element of which is the sum, and the second element of which is a
7964 bit specifying if the unsigned summation resulted in a carry.
7969 .. code-block:: llvm
7971 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7972 %sum = extractvalue {i32, i1} %res, 0
7973 %obit = extractvalue {i32, i1} %res, 1
7974 br i1 %obit, label %carry, label %normal
7976 '``llvm.ssub.with.overflow.*``' Intrinsics
7977 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7982 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7983 on any integer bit width.
7987 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7988 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7989 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7994 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7995 a signed subtraction of the two arguments, and indicate whether an
7996 overflow occurred during the signed subtraction.
8001 The arguments (%a and %b) and the first element of the result structure
8002 may be of integer types of any bit width, but they must have the same
8003 bit width. The second element of the result structure must be of type
8004 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8010 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8011 a signed subtraction of the two arguments. They return a structure --- the
8012 first element of which is the subtraction, and the second element of
8013 which is a bit specifying if the signed subtraction resulted in an
8019 .. code-block:: llvm
8021 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8022 %sum = extractvalue {i32, i1} %res, 0
8023 %obit = extractvalue {i32, i1} %res, 1
8024 br i1 %obit, label %overflow, label %normal
8026 '``llvm.usub.with.overflow.*``' Intrinsics
8027 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8032 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8033 on any integer bit width.
8037 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8038 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8039 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8044 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8045 an unsigned subtraction of the two arguments, and indicate whether an
8046 overflow occurred during the unsigned subtraction.
8051 The arguments (%a and %b) and the first element of the result structure
8052 may be of integer types of any bit width, but they must have the same
8053 bit width. The second element of the result structure must be of type
8054 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8060 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8061 an unsigned subtraction of the two arguments. They return a structure ---
8062 the first element of which is the subtraction, and the second element of
8063 which is a bit specifying if the unsigned subtraction resulted in an
8069 .. code-block:: llvm
8071 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8072 %sum = extractvalue {i32, i1} %res, 0
8073 %obit = extractvalue {i32, i1} %res, 1
8074 br i1 %obit, label %overflow, label %normal
8076 '``llvm.smul.with.overflow.*``' Intrinsics
8077 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8082 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8083 on any integer bit width.
8087 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8088 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8089 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8094 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8095 a signed multiplication of the two arguments, and indicate whether an
8096 overflow occurred during the signed multiplication.
8101 The arguments (%a and %b) and the first element of the result structure
8102 may be of integer types of any bit width, but they must have the same
8103 bit width. The second element of the result structure must be of type
8104 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8110 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8111 a signed multiplication of the two arguments. They return a structure ---
8112 the first element of which is the multiplication, and the second element
8113 of which is a bit specifying if the signed multiplication resulted in an
8119 .. code-block:: llvm
8121 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8122 %sum = extractvalue {i32, i1} %res, 0
8123 %obit = extractvalue {i32, i1} %res, 1
8124 br i1 %obit, label %overflow, label %normal
8126 '``llvm.umul.with.overflow.*``' Intrinsics
8127 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8132 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8133 on any integer bit width.
8137 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8138 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8139 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8144 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8145 a unsigned multiplication of the two arguments, and indicate whether an
8146 overflow occurred during the unsigned multiplication.
8151 The arguments (%a and %b) and the first element of the result structure
8152 may be of integer types of any bit width, but they must have the same
8153 bit width. The second element of the result structure must be of type
8154 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8160 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8161 an unsigned multiplication of the two arguments. They return a structure ---
8162 the first element of which is the multiplication, and the second
8163 element of which is a bit specifying if the unsigned multiplication
8164 resulted in an overflow.
8169 .. code-block:: llvm
8171 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8172 %sum = extractvalue {i32, i1} %res, 0
8173 %obit = extractvalue {i32, i1} %res, 1
8174 br i1 %obit, label %overflow, label %normal
8176 Specialised Arithmetic Intrinsics
8177 ---------------------------------
8179 '``llvm.fmuladd.*``' Intrinsic
8180 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8187 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8188 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8193 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8194 expressions that can be fused if the code generator determines that (a) the
8195 target instruction set has support for a fused operation, and (b) that the
8196 fused operation is more efficient than the equivalent, separate pair of mul
8197 and add instructions.
8202 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8203 multiplicands, a and b, and an addend c.
8212 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8214 is equivalent to the expression a \* b + c, except that rounding will
8215 not be performed between the multiplication and addition steps if the
8216 code generator fuses the operations. Fusion is not guaranteed, even if
8217 the target platform supports it. If a fused multiply-add is required the
8218 corresponding llvm.fma.\* intrinsic function should be used instead.
8223 .. code-block:: llvm
8225 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8227 Half Precision Floating Point Intrinsics
8228 ----------------------------------------
8230 For most target platforms, half precision floating point is a
8231 storage-only format. This means that it is a dense encoding (in memory)
8232 but does not support computation in the format.
8234 This means that code must first load the half-precision floating point
8235 value as an i16, then convert it to float with
8236 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8237 then be performed on the float value (including extending to double
8238 etc). To store the value back to memory, it is first converted to float
8239 if needed, then converted to i16 with
8240 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8243 .. _int_convert_to_fp16:
8245 '``llvm.convert.to.fp16``' Intrinsic
8246 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8253 declare i16 @llvm.convert.to.fp16(f32 %a)
8258 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8259 from single precision floating point format to half precision floating
8265 The intrinsic function contains single argument - the value to be
8271 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8272 from single precision floating point format to half precision floating
8273 point format. The return value is an ``i16`` which contains the
8279 .. code-block:: llvm
8281 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8282 store i16 %res, i16* @x, align 2
8284 .. _int_convert_from_fp16:
8286 '``llvm.convert.from.fp16``' Intrinsic
8287 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8294 declare f32 @llvm.convert.from.fp16(i16 %a)
8299 The '``llvm.convert.from.fp16``' intrinsic function performs a
8300 conversion from half precision floating point format to single precision
8301 floating point format.
8306 The intrinsic function contains single argument - the value to be
8312 The '``llvm.convert.from.fp16``' intrinsic function performs a
8313 conversion from half single precision floating point format to single
8314 precision floating point format. The input half-float value is
8315 represented by an ``i16`` value.
8320 .. code-block:: llvm
8322 %a = load i16* @x, align 2
8323 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8328 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8329 prefix), are described in the `LLVM Source Level
8330 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8333 Exception Handling Intrinsics
8334 -----------------------------
8336 The LLVM exception handling intrinsics (which all start with
8337 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8338 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8342 Trampoline Intrinsics
8343 ---------------------
8345 These intrinsics make it possible to excise one parameter, marked with
8346 the :ref:`nest <nest>` attribute, from a function. The result is a
8347 callable function pointer lacking the nest parameter - the caller does
8348 not need to provide a value for it. Instead, the value to use is stored
8349 in advance in a "trampoline", a block of memory usually allocated on the
8350 stack, which also contains code to splice the nest value into the
8351 argument list. This is used to implement the GCC nested function address
8354 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8355 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8356 It can be created as follows:
8358 .. code-block:: llvm
8360 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8361 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8362 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8363 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8364 %fp = bitcast i8* %p to i32 (i32, i32)*
8366 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8367 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8371 '``llvm.init.trampoline``' Intrinsic
8372 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8379 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8384 This fills the memory pointed to by ``tramp`` with executable code,
8385 turning it into a trampoline.
8390 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8391 pointers. The ``tramp`` argument must point to a sufficiently large and
8392 sufficiently aligned block of memory; this memory is written to by the
8393 intrinsic. Note that the size and the alignment are target-specific -
8394 LLVM currently provides no portable way of determining them, so a
8395 front-end that generates this intrinsic needs to have some
8396 target-specific knowledge. The ``func`` argument must hold a function
8397 bitcast to an ``i8*``.
8402 The block of memory pointed to by ``tramp`` is filled with target
8403 dependent code, turning it into a function. Then ``tramp`` needs to be
8404 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8405 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8406 function's signature is the same as that of ``func`` with any arguments
8407 marked with the ``nest`` attribute removed. At most one such ``nest``
8408 argument is allowed, and it must be of pointer type. Calling the new
8409 function is equivalent to calling ``func`` with the same argument list,
8410 but with ``nval`` used for the missing ``nest`` argument. If, after
8411 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8412 modified, then the effect of any later call to the returned function
8413 pointer is undefined.
8417 '``llvm.adjust.trampoline``' Intrinsic
8418 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8425 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8430 This performs any required machine-specific adjustment to the address of
8431 a trampoline (passed as ``tramp``).
8436 ``tramp`` must point to a block of memory which already has trampoline
8437 code filled in by a previous call to
8438 :ref:`llvm.init.trampoline <int_it>`.
8443 On some architectures the address of the code to be executed needs to be
8444 different to the address where the trampoline is actually stored. This
8445 intrinsic returns the executable address corresponding to ``tramp``
8446 after performing the required machine specific adjustments. The pointer
8447 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8452 This class of intrinsics exists to information about the lifetime of
8453 memory objects and ranges where variables are immutable.
8457 '``llvm.lifetime.start``' Intrinsic
8458 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8465 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8470 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8476 The first argument is a constant integer representing the size of the
8477 object, or -1 if it is variable sized. The second argument is a pointer
8483 This intrinsic indicates that before this point in the code, the value
8484 of the memory pointed to by ``ptr`` is dead. This means that it is known
8485 to never be used and has an undefined value. A load from the pointer
8486 that precedes this intrinsic can be replaced with ``'undef'``.
8490 '``llvm.lifetime.end``' Intrinsic
8491 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8498 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8503 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8509 The first argument is a constant integer representing the size of the
8510 object, or -1 if it is variable sized. The second argument is a pointer
8516 This intrinsic indicates that after this point in the code, the value of
8517 the memory pointed to by ``ptr`` is dead. This means that it is known to
8518 never be used and has an undefined value. Any stores into the memory
8519 object following this intrinsic may be removed as dead.
8521 '``llvm.invariant.start``' Intrinsic
8522 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8529 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8534 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8535 a memory object will not change.
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 until an ``llvm.invariant.end`` that uses
8548 the return value, the referenced memory location is constant and
8551 '``llvm.invariant.end``' Intrinsic
8552 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8559 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8564 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8565 memory object are mutable.
8570 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8571 The second argument is a constant integer representing the size of the
8572 object, or -1 if it is variable sized and the third argument is a
8573 pointer to the object.
8578 This intrinsic indicates that the memory is mutable again.
8583 This class of intrinsics is designed to be generic and has no specific
8586 '``llvm.var.annotation``' Intrinsic
8587 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8594 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8599 The '``llvm.var.annotation``' intrinsic.
8604 The first argument is a pointer to a value, the second is a pointer to a
8605 global string, the third is a pointer to a global string which is the
8606 source file name, and the last argument is the line number.
8611 This intrinsic allows annotation of local variables with arbitrary
8612 strings. This can be useful for special purpose optimizations that want
8613 to look for these annotations. These have no other defined use; they are
8614 ignored by code generation and optimization.
8616 '``llvm.ptr.annotation.*``' Intrinsic
8617 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8622 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8623 pointer to an integer of any width. *NOTE* you must specify an address space for
8624 the pointer. The identifier for the default address space is the integer
8629 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8630 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8631 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8632 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8633 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8638 The '``llvm.ptr.annotation``' intrinsic.
8643 The first argument is a pointer to an integer value of arbitrary bitwidth
8644 (result of some expression), the second is a pointer to a global string, the
8645 third is a pointer to a global string which is the source file name, and the
8646 last argument is the line number. It returns the value of the first argument.
8651 This intrinsic allows annotation of a pointer to an integer with arbitrary
8652 strings. This can be useful for special purpose optimizations that want to look
8653 for these annotations. These have no other defined use; they are ignored by code
8654 generation and optimization.
8656 '``llvm.annotation.*``' Intrinsic
8657 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8662 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8663 any integer bit width.
8667 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8668 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8669 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8670 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8671 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8676 The '``llvm.annotation``' intrinsic.
8681 The first argument is an integer value (result of some expression), the
8682 second is a pointer to a global string, the third is a pointer to a
8683 global string which is the source file name, and the last argument is
8684 the line number. It returns the value of the first argument.
8689 This intrinsic allows annotations to be put on arbitrary expressions
8690 with arbitrary strings. This can be useful for special purpose
8691 optimizations that want to look for these annotations. These have no
8692 other defined use; they are ignored by code generation and optimization.
8694 '``llvm.trap``' Intrinsic
8695 ^^^^^^^^^^^^^^^^^^^^^^^^^
8702 declare void @llvm.trap() noreturn nounwind
8707 The '``llvm.trap``' intrinsic.
8717 This intrinsic is lowered to the target dependent trap instruction. If
8718 the target does not have a trap instruction, this intrinsic will be
8719 lowered to a call of the ``abort()`` function.
8721 '``llvm.debugtrap``' Intrinsic
8722 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8729 declare void @llvm.debugtrap() nounwind
8734 The '``llvm.debugtrap``' intrinsic.
8744 This intrinsic is lowered to code which is intended to cause an
8745 execution trap with the intention of requesting the attention of a
8748 '``llvm.stackprotector``' Intrinsic
8749 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8756 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8761 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8762 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8763 is placed on the stack before local variables.
8768 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8769 The first argument is the value loaded from the stack guard
8770 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8771 enough space to hold the value of the guard.
8776 This intrinsic causes the prologue/epilogue inserter to force the position of
8777 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8778 to ensure that if a local variable on the stack is overwritten, it will destroy
8779 the value of the guard. When the function exits, the guard on the stack is
8780 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8781 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8782 calling the ``__stack_chk_fail()`` function.
8784 '``llvm.stackprotectorcheck``' Intrinsic
8785 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8792 declare void @llvm.stackprotectorcheck(i8** <guard>)
8797 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8798 created stack protector and if they are not equal calls the
8799 ``__stack_chk_fail()`` function.
8804 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8805 the variable ``@__stack_chk_guard``.
8810 This intrinsic is provided to perform the stack protector check by comparing
8811 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8812 values do not match call the ``__stack_chk_fail()`` function.
8814 The reason to provide this as an IR level intrinsic instead of implementing it
8815 via other IR operations is that in order to perform this operation at the IR
8816 level without an intrinsic, one would need to create additional basic blocks to
8817 handle the success/failure cases. This makes it difficult to stop the stack
8818 protector check from disrupting sibling tail calls in Codegen. With this
8819 intrinsic, we are able to generate the stack protector basic blocks late in
8820 codegen after the tail call decision has occurred.
8822 '``llvm.objectsize``' Intrinsic
8823 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8830 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8831 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8836 The ``llvm.objectsize`` intrinsic is designed to provide information to
8837 the optimizers to determine at compile time whether a) an operation
8838 (like memcpy) will overflow a buffer that corresponds to an object, or
8839 b) that a runtime check for overflow isn't necessary. An object in this
8840 context means an allocation of a specific class, structure, array, or
8846 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8847 argument is a pointer to or into the ``object``. The second argument is
8848 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8849 or -1 (if false) when the object size is unknown. The second argument
8850 only accepts constants.
8855 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8856 the size of the object concerned. If the size cannot be determined at
8857 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8858 on the ``min`` argument).
8860 '``llvm.expect``' Intrinsic
8861 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8868 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8869 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8874 The ``llvm.expect`` intrinsic provides information about expected (the
8875 most probable) value of ``val``, which can be used by optimizers.
8880 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8881 a value. The second argument is an expected value, this needs to be a
8882 constant value, variables are not allowed.
8887 This intrinsic is lowered to the ``val``.
8889 '``llvm.donothing``' Intrinsic
8890 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8897 declare void @llvm.donothing() nounwind readnone
8902 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8903 only intrinsic that can be called with an invoke instruction.
8913 This intrinsic does nothing, and it's removed by optimizers and ignored