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
294 It is illegal for a function *declaration* to have any linkage type
295 other than ``external``, ``dllimport`` or ``extern_weak``.
302 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
303 :ref:`invokes <i_invoke>` can all have an optional calling convention
304 specified for the call. The calling convention of any pair of dynamic
305 caller/callee must match, or the behavior of the program is undefined.
306 The following calling conventions are supported by LLVM, and more may be
309 "``ccc``" - The C calling convention
310 This calling convention (the default if no other calling convention
311 is specified) matches the target C calling conventions. This calling
312 convention supports varargs function calls and tolerates some
313 mismatch in the declared prototype and implemented declaration of
314 the function (as does normal C).
315 "``fastcc``" - The fast calling convention
316 This calling convention attempts to make calls as fast as possible
317 (e.g. by passing things in registers). This calling convention
318 allows the target to use whatever tricks it wants to produce fast
319 code for the target, without having to conform to an externally
320 specified ABI (Application Binary Interface). `Tail calls can only
321 be optimized when this, the GHC or the HiPE convention is
322 used. <CodeGenerator.html#id80>`_ This calling convention does not
323 support varargs and requires the prototype of all callees to exactly
324 match the prototype of the function definition.
325 "``coldcc``" - The cold calling convention
326 This calling convention attempts to make code in the caller as
327 efficient as possible under the assumption that the call is not
328 commonly executed. As such, these calls often preserve all registers
329 so that the call does not break any live ranges in the caller side.
330 This calling convention does not support varargs and requires the
331 prototype of all callees to exactly match the prototype of the
333 "``cc 10``" - GHC convention
334 This calling convention has been implemented specifically for use by
335 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
336 It passes everything in registers, going to extremes to achieve this
337 by disabling callee save registers. This calling convention should
338 not be used lightly but only for specific situations such as an
339 alternative to the *register pinning* performance technique often
340 used when implementing functional programming languages. At the
341 moment only X86 supports this convention and it has the following
344 - On *X86-32* only supports up to 4 bit type parameters. No
345 floating point types are supported.
346 - On *X86-64* only supports up to 10 bit type parameters and 6
347 floating point parameters.
349 This calling convention supports `tail call
350 optimization <CodeGenerator.html#id80>`_ but requires both the
351 caller and callee are using it.
352 "``cc 11``" - The HiPE calling convention
353 This calling convention has been implemented specifically for use by
354 the `High-Performance Erlang
355 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
356 native code compiler of the `Ericsson's Open Source Erlang/OTP
357 system <http://www.erlang.org/download.shtml>`_. It uses more
358 registers for argument passing than the ordinary C calling
359 convention and defines no callee-saved registers. The calling
360 convention properly supports `tail call
361 optimization <CodeGenerator.html#id80>`_ but requires that both the
362 caller and the callee use it. It uses a *register pinning*
363 mechanism, similar to GHC's convention, for keeping frequently
364 accessed runtime components pinned to specific hardware registers.
365 At the moment only X86 supports this convention (both 32 and 64
367 "``cc <n>``" - Numbered convention
368 Any calling convention may be specified by number, allowing
369 target-specific calling conventions to be used. Target specific
370 calling conventions start at 64.
372 More calling conventions can be added/defined on an as-needed basis, to
373 support Pascal conventions or any other well-known target-independent
376 .. _visibilitystyles:
381 All Global Variables and Functions have one of the following visibility
384 "``default``" - Default style
385 On targets that use the ELF object file format, default visibility
386 means that the declaration is visible to other modules and, in
387 shared libraries, means that the declared entity may be overridden.
388 On Darwin, default visibility means that the declaration is visible
389 to other modules. Default visibility corresponds to "external
390 linkage" in the language.
391 "``hidden``" - Hidden style
392 Two declarations of an object with hidden visibility refer to the
393 same object if they are in the same shared object. Usually, hidden
394 visibility indicates that the symbol will not be placed into the
395 dynamic symbol table, so no other module (executable or shared
396 library) can reference it directly.
397 "``protected``" - Protected style
398 On ELF, protected visibility indicates that the symbol will be
399 placed in the dynamic symbol table, but that references within the
400 defining module will bind to the local symbol. That is, the symbol
401 cannot be overridden by another module.
408 LLVM IR allows you to specify name aliases for certain types. This can
409 make it easier to read the IR and make the IR more condensed
410 (particularly when recursive types are involved). An example of a name
415 %mytype = type { %mytype*, i32 }
417 You may give a name to any :ref:`type <typesystem>` except
418 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
419 expected with the syntax "%mytype".
421 Note that type names are aliases for the structural type that they
422 indicate, and that you can therefore specify multiple names for the same
423 type. This often leads to confusing behavior when dumping out a .ll
424 file. Since LLVM IR uses structural typing, the name is not part of the
425 type. When printing out LLVM IR, the printer will pick *one name* to
426 render all types of a particular shape. This means that if you have code
427 where two different source types end up having the same LLVM type, that
428 the dumper will sometimes print the "wrong" or unexpected type. This is
429 an important design point and isn't going to change.
436 Global variables define regions of memory allocated at compilation time
439 Global variables definitions must be initialized, may have an explicit section
440 to be placed in, and may have an optional explicit alignment specified.
442 Global variables in other translation units can also be declared, in which
443 case they don't have an initializer.
445 A variable may be defined as ``thread_local``, which means that it will
446 not be shared by threads (each thread will have a separated copy of the
447 variable). Not all targets support thread-local variables. Optionally, a
448 TLS model may be specified:
451 For variables that are only used within the current shared library.
453 For variables in modules that will not be loaded dynamically.
455 For variables defined in the executable and only used within it.
457 The models correspond to the ELF TLS models; see `ELF Handling For
458 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
459 more information on under which circumstances the different models may
460 be used. The target may choose a different TLS model if the specified
461 model is not supported, or if a better choice of model can be made.
463 A variable may be defined as a global ``constant``, which indicates that
464 the contents of the variable will **never** be modified (enabling better
465 optimization, allowing the global data to be placed in the read-only
466 section of an executable, etc). Note that variables that need runtime
467 initialization cannot be marked ``constant`` as there is a store to the
470 LLVM explicitly allows *declarations* of global variables to be marked
471 constant, even if the final definition of the global is not. This
472 capability can be used to enable slightly better optimization of the
473 program, but requires the language definition to guarantee that
474 optimizations based on the 'constantness' are valid for the translation
475 units that do not include the definition.
477 As SSA values, global variables define pointer values that are in scope
478 (i.e. they dominate) all basic blocks in the program. Global variables
479 always define a pointer to their "content" type because they describe a
480 region of memory, and all memory objects in LLVM are accessed through
483 Global variables can be marked with ``unnamed_addr`` which indicates
484 that the address is not significant, only the content. Constants marked
485 like this can be merged with other constants if they have the same
486 initializer. Note that a constant with significant address *can* be
487 merged with a ``unnamed_addr`` constant, the result being a constant
488 whose address is significant.
490 A global variable may be declared to reside in a target-specific
491 numbered address space. For targets that support them, address spaces
492 may affect how optimizations are performed and/or what target
493 instructions are used to access the variable. The default address space
494 is zero. The address space qualifier must precede any other attributes.
496 LLVM allows an explicit section to be specified for globals. If the
497 target supports it, it will emit globals to the section specified.
499 By default, global initializers are optimized by assuming that global
500 variables defined within the module are not modified from their
501 initial values before the start of the global initializer. This is
502 true even for variables potentially accessible from outside the
503 module, including those with external linkage or appearing in
504 ``@llvm.used``. This assumption may be suppressed by marking the
505 variable with ``externally_initialized``.
507 An explicit alignment may be specified for a global, which must be a
508 power of 2. If not present, or if the alignment is set to zero, the
509 alignment of the global is set by the target to whatever it feels
510 convenient. If an explicit alignment is specified, the global is forced
511 to have exactly that alignment. Targets and optimizers are not allowed
512 to over-align the global if the global has an assigned section. In this
513 case, the extra alignment could be observable: for example, code could
514 assume that the globals are densely packed in their section and try to
515 iterate over them as an array, alignment padding would break this
518 For example, the following defines a global in a numbered address space
519 with an initializer, section, and alignment:
523 @G = addrspace(5) constant float 1.0, section "foo", align 4
525 The following example just declares a global variable
529 @G = external global i32
531 The following example defines a thread-local global with the
532 ``initialexec`` TLS model:
536 @G = thread_local(initialexec) global i32 0, align 4
538 .. _functionstructure:
543 LLVM function definitions consist of the "``define``" keyword, an
544 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
545 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
546 an optional ``unnamed_addr`` attribute, a return type, an optional
547 :ref:`parameter attribute <paramattrs>` for the return type, a function
548 name, a (possibly empty) argument list (each with optional :ref:`parameter
549 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
550 an optional section, an optional alignment, an optional :ref:`garbage
551 collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
552 curly brace, a list of basic blocks, and a closing curly brace.
554 LLVM function declarations consist of the "``declare``" keyword, an
555 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
556 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
557 an optional ``unnamed_addr`` attribute, a return type, an optional
558 :ref:`parameter attribute <paramattrs>` for the return type, a function
559 name, a possibly empty list of arguments, an optional alignment, an optional
560 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
562 A function definition contains a list of basic blocks, forming the CFG (Control
563 Flow Graph) for the function. Each basic block may optionally start with a label
564 (giving the basic block a symbol table entry), contains a list of instructions,
565 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
566 function return). If an explicit label is not provided, a block is assigned an
567 implicit numbered label, using the next value from the same counter as used for
568 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
569 entry block does not have an explicit label, it will be assigned label "%0",
570 then the first unnamed temporary in that block will be "%1", etc.
572 The first basic block in a function is special in two ways: it is
573 immediately executed on entrance to the function, and it is not allowed
574 to have predecessor basic blocks (i.e. there can not be any branches to
575 the entry block of a function). Because the block can have no
576 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
578 LLVM allows an explicit section to be specified for functions. If the
579 target supports it, it will emit functions to the section specified.
581 An explicit alignment may be specified for a function. If not present,
582 or if the alignment is set to zero, the alignment of the function is set
583 by the target to whatever it feels convenient. If an explicit alignment
584 is specified, the function is forced to have at least that much
585 alignment. All alignments must be a power of 2.
587 If the ``unnamed_addr`` attribute is given, the address is know to not
588 be significant and two identical functions can be merged.
592 define [linkage] [visibility]
594 <ResultType> @<FunctionName> ([argument list])
595 [fn Attrs] [section "name"] [align N]
596 [gc] [prefix Constant] { ... }
603 Aliases act as "second name" for the aliasee value (which can be either
604 function, global variable, another alias or bitcast of global value).
605 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
606 :ref:`visibility style <visibility>`.
610 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
612 The linkage must be one of ``private``, ``linker_private``,
613 ``linker_private_weak``, ``internal``, ``linkonce``, ``weak``,
614 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
615 might not correctly handle dropping a weak symbol that is aliased by a non weak
618 .. _namedmetadatastructure:
623 Named metadata is a collection of metadata. :ref:`Metadata
624 nodes <metadata>` (but not metadata strings) are the only valid
625 operands for a named metadata.
629 ; Some unnamed metadata nodes, which are referenced by the named metadata.
630 !0 = metadata !{metadata !"zero"}
631 !1 = metadata !{metadata !"one"}
632 !2 = metadata !{metadata !"two"}
634 !name = !{!0, !1, !2}
641 The return type and each parameter of a function type may have a set of
642 *parameter attributes* associated with them. Parameter attributes are
643 used to communicate additional information about the result or
644 parameters of a function. Parameter attributes are considered to be part
645 of the function, not of the function type, so functions with different
646 parameter attributes can have the same function type.
648 Parameter attributes are simple keywords that follow the type specified.
649 If multiple parameter attributes are needed, they are space separated.
654 declare i32 @printf(i8* noalias nocapture, ...)
655 declare i32 @atoi(i8 zeroext)
656 declare signext i8 @returns_signed_char()
658 Note that any attributes for the function result (``nounwind``,
659 ``readonly``) come immediately after the argument list.
661 Currently, only the following parameter attributes are defined:
664 This indicates to the code generator that the parameter or return
665 value should be zero-extended to the extent required by the target's
666 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
667 the caller (for a parameter) or the callee (for a return value).
669 This indicates to the code generator that the parameter or return
670 value should be sign-extended to the extent required by the target's
671 ABI (which is usually 32-bits) by the caller (for a parameter) or
672 the callee (for a return value).
674 This indicates that this parameter or return value should be treated
675 in a special target-dependent fashion during while emitting code for
676 a function call or return (usually, by putting it in a register as
677 opposed to memory, though some targets use it to distinguish between
678 two different kinds of registers). Use of this attribute is
681 This indicates that the pointer parameter should really be passed by
682 value to the function. The attribute implies that a hidden copy of
683 the pointee is made between the caller and the callee, so the callee
684 is unable to modify the value in the caller. This attribute is only
685 valid on LLVM pointer arguments. It is generally used to pass
686 structs and arrays by value, but is also valid on pointers to
687 scalars. The copy is considered to belong to the caller not the
688 callee (for example, ``readonly`` functions should not write to
689 ``byval`` parameters). This is not a valid attribute for return
692 The byval attribute also supports specifying an alignment with the
693 align attribute. It indicates the alignment of the stack slot to
694 form and the known alignment of the pointer specified to the call
695 site. If the alignment is not specified, then the code generator
696 makes a target-specific assumption.
699 This indicates that the pointer parameter specifies the address of a
700 structure that is the return value of the function in the source
701 program. This pointer must be guaranteed by the caller to be valid:
702 loads and stores to the structure may be assumed by the callee
703 not to trap and to be properly aligned. This may only be applied to
704 the first parameter. This is not a valid attribute for return
707 This indicates that pointer values :ref:`based <pointeraliasing>` on
708 the argument or return value do not alias pointer values which are
709 not *based* on it, ignoring certain "irrelevant" dependencies. For a
710 call to the parent function, dependencies between memory references
711 from before or after the call and from those during the call are
712 "irrelevant" to the ``noalias`` keyword for the arguments and return
713 value used in that call. The caller shares the responsibility with
714 the callee for ensuring that these requirements are met. For further
715 details, please see the discussion of the NoAlias response in `alias
716 analysis <AliasAnalysis.html#MustMayNo>`_.
718 Note that this definition of ``noalias`` is intentionally similar
719 to the definition of ``restrict`` in C99 for function arguments,
720 though it is slightly weaker.
722 For function return values, C99's ``restrict`` is not meaningful,
723 while LLVM's ``noalias`` is.
725 This indicates that the callee does not make any copies of the
726 pointer that outlive the callee itself. This is not a valid
727 attribute for return values.
732 This indicates that the pointer parameter can be excised using the
733 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
734 attribute for return values and can only be applied to one parameter.
737 This indicates that the function always returns the argument as its return
738 value. This is an optimization hint to the code generator when generating
739 the caller, allowing tail call optimization and omission of register saves
740 and restores in some cases; it is not checked or enforced when generating
741 the callee. The parameter and the function return type must be valid
742 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
743 valid attribute for return values and can only be applied to one parameter.
747 Garbage Collector Names
748 -----------------------
750 Each function may specify a garbage collector name, which is simply a
755 define void @f() gc "name" { ... }
757 The compiler declares the supported values of *name*. Specifying a
758 collector which will cause the compiler to alter its output in order to
759 support the named garbage collection algorithm.
766 Prefix data is data associated with a function which the code generator
767 will emit immediately before the function body. The purpose of this feature
768 is to allow frontends to associate language-specific runtime metadata with
769 specific functions and make it available through the function pointer while
770 still allowing the function pointer to be called. To access the data for a
771 given function, a program may bitcast the function pointer to a pointer to
772 the constant's type. This implies that the IR symbol points to the start
775 To maintain the semantics of ordinary function calls, the prefix data must
776 have a particular format. Specifically, it must begin with a sequence of
777 bytes which decode to a sequence of machine instructions, valid for the
778 module's target, which transfer control to the point immediately succeeding
779 the prefix data, without performing any other visible action. This allows
780 the inliner and other passes to reason about the semantics of the function
781 definition without needing to reason about the prefix data. Obviously this
782 makes the format of the prefix data highly target dependent.
784 Prefix data is laid out as if it were an initializer for a global variable
785 of the prefix data's type. No padding is automatically placed between the
786 prefix data and the function body. If padding is required, it must be part
789 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
790 which encodes the ``nop`` instruction:
794 define void @f() prefix i8 144 { ... }
796 Generally prefix data can be formed by encoding a relative branch instruction
797 which skips the metadata, as in this example of valid prefix data for the
798 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
802 %0 = type <{ i8, i8, i8* }>
804 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
806 A function may have prefix data but no body. This has similar semantics
807 to the ``available_externally`` linkage in that the data may be used by the
808 optimizers but will not be emitted in the object file.
815 Attribute groups are groups of attributes that are referenced by objects within
816 the IR. They are important for keeping ``.ll`` files readable, because a lot of
817 functions will use the same set of attributes. In the degenerative case of a
818 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
819 group will capture the important command line flags used to build that file.
821 An attribute group is a module-level object. To use an attribute group, an
822 object references the attribute group's ID (e.g. ``#37``). An object may refer
823 to more than one attribute group. In that situation, the attributes from the
824 different groups are merged.
826 Here is an example of attribute groups for a function that should always be
827 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
831 ; Target-independent attributes:
832 attributes #0 = { alwaysinline alignstack=4 }
834 ; Target-dependent attributes:
835 attributes #1 = { "no-sse" }
837 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
838 define void @f() #0 #1 { ... }
845 Function attributes are set to communicate additional information about
846 a function. Function attributes are considered to be part of the
847 function, not of the function type, so functions with different function
848 attributes can have the same function type.
850 Function attributes are simple keywords that follow the type specified.
851 If multiple attributes are needed, they are space separated. For
856 define void @f() noinline { ... }
857 define void @f() alwaysinline { ... }
858 define void @f() alwaysinline optsize { ... }
859 define void @f() optsize { ... }
862 This attribute indicates that, when emitting the prologue and
863 epilogue, the backend should forcibly align the stack pointer.
864 Specify the desired alignment, which must be a power of two, in
867 This attribute indicates that the inliner should attempt to inline
868 this function into callers whenever possible, ignoring any active
869 inlining size threshold for this caller.
871 This indicates that the callee function at a call site should be
872 recognized as a built-in function, even though the function's declaration
873 uses the ``nobuiltin`` attribute. This is only valid at call sites for
874 direct calls to functions which are declared with the ``nobuiltin``
877 This attribute indicates that this function is rarely called. When
878 computing edge weights, basic blocks post-dominated by a cold
879 function call are also considered to be cold; and, thus, given low
882 This attribute indicates that the source code contained a hint that
883 inlining this function is desirable (such as the "inline" keyword in
884 C/C++). It is just a hint; it imposes no requirements on the
887 This attribute suggests that optimization passes and code generator
888 passes make choices that keep the code size of this function as small
889 as possible and perform optimizations that may sacrifice runtime
890 performance in order to minimize the size of the generated code.
892 This attribute disables prologue / epilogue emission for the
893 function. This can have very system-specific consequences.
895 This indicates that the callee function at a call site is not recognized as
896 a built-in function. LLVM will retain the original call and not replace it
897 with equivalent code based on the semantics of the built-in function, unless
898 the call site uses the ``builtin`` attribute. This is valid at call sites
899 and on function declarations and definitions.
901 This attribute indicates that calls to the function cannot be
902 duplicated. A call to a ``noduplicate`` function may be moved
903 within its parent function, but may not be duplicated within
906 A function containing a ``noduplicate`` call may still
907 be an inlining candidate, provided that the call is not
908 duplicated by inlining. That implies that the function has
909 internal linkage and only has one call site, so the original
910 call is dead after inlining.
912 This attributes disables implicit floating point instructions.
914 This attribute indicates that the inliner should never inline this
915 function in any situation. This attribute may not be used together
916 with the ``alwaysinline`` attribute.
918 This attribute suppresses lazy symbol binding for the function. This
919 may make calls to the function faster, at the cost of extra program
920 startup time if the function is not called during program startup.
922 This attribute indicates that the code generator should not use a
923 red zone, even if the target-specific ABI normally permits it.
925 This function attribute indicates that the function never returns
926 normally. This produces undefined behavior at runtime if the
927 function ever does dynamically return.
929 This function attribute indicates that the function never returns
930 with an unwind or exceptional control flow. If the function does
931 unwind, its runtime behavior is undefined.
933 This function attribute indicates that the function is not optimized
934 by any optimization or code generator passes with the
935 exception of interprocedural optimization passes.
936 This attribute cannot be used together with the ``alwaysinline``
937 attribute; this attribute is also incompatible
938 with the ``minsize`` attribute and the ``optsize`` attribute.
940 This attribute requires the ``noinline`` attribute to be specified on
941 the function as well, so the function is never inlined into any caller.
942 Only functions with the ``alwaysinline`` attribute are valid
943 candidates for inlining into the body of this function.
945 This attribute suggests that optimization passes and code generator
946 passes make choices that keep the code size of this function low,
947 and otherwise do optimizations specifically to reduce code size as
948 long as they do not significantly impact runtime performance.
950 On a function, this attribute indicates that the function computes its
951 result (or decides to unwind an exception) based strictly on its arguments,
952 without dereferencing any pointer arguments or otherwise accessing
953 any mutable state (e.g. memory, control registers, etc) visible to
954 caller functions. It does not write through any pointer arguments
955 (including ``byval`` arguments) and never changes any state visible
956 to callers. This means that it cannot unwind exceptions by calling
957 the ``C++`` exception throwing methods.
959 On an argument, this attribute indicates that the function does not
960 dereference that pointer argument, even though it may read or write the
961 memory that the pointer points to if accessed through other pointers.
963 On a function, this attribute indicates that the function does not write
964 through any pointer arguments (including ``byval`` arguments) or otherwise
965 modify any state (e.g. memory, control registers, etc) visible to
966 caller functions. It may dereference pointer arguments and read
967 state that may be set in the caller. A readonly function always
968 returns the same value (or unwinds an exception identically) when
969 called with the same set of arguments and global state. It cannot
970 unwind an exception by calling the ``C++`` exception throwing
973 On an argument, this attribute indicates that the function does not write
974 through this pointer argument, even though it may write to the memory that
975 the pointer points to.
977 This attribute indicates that this function can return twice. The C
978 ``setjmp`` is an example of such a function. The compiler disables
979 some optimizations (like tail calls) in the caller of these
982 This attribute indicates that AddressSanitizer checks
983 (dynamic address safety analysis) are enabled for this function.
985 This attribute indicates that MemorySanitizer checks (dynamic detection
986 of accesses to uninitialized memory) are enabled for this function.
988 This attribute indicates that ThreadSanitizer checks
989 (dynamic thread safety analysis) are enabled for this function.
991 This attribute indicates that the function should emit a stack
992 smashing protector. It is in the form of a "canary" --- a random value
993 placed on the stack before the local variables that's checked upon
994 return from the function to see if it has been overwritten. A
995 heuristic is used to determine if a function needs stack protectors
996 or not. The heuristic used will enable protectors for functions with:
998 - Character arrays larger than ``ssp-buffer-size`` (default 8).
999 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1000 - Calls to alloca() with variable sizes or constant sizes greater than
1001 ``ssp-buffer-size``.
1003 If a function that has an ``ssp`` attribute is inlined into a
1004 function that doesn't have an ``ssp`` attribute, then the resulting
1005 function will have an ``ssp`` attribute.
1007 This attribute indicates that the function should *always* emit a
1008 stack smashing protector. This overrides the ``ssp`` function
1011 If a function that has an ``sspreq`` attribute is inlined into a
1012 function that doesn't have an ``sspreq`` attribute or which has an
1013 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1014 an ``sspreq`` attribute.
1016 This attribute indicates that the function should emit a stack smashing
1017 protector. This attribute causes a strong heuristic to be used when
1018 determining if a function needs stack protectors. The strong heuristic
1019 will enable protectors for functions with:
1021 - Arrays of any size and type
1022 - Aggregates containing an array of any size and type.
1023 - Calls to alloca().
1024 - Local variables that have had their address taken.
1026 This overrides the ``ssp`` function attribute.
1028 If a function that has an ``sspstrong`` attribute is inlined into a
1029 function that doesn't have an ``sspstrong`` attribute, then the
1030 resulting function will have an ``sspstrong`` attribute.
1032 This attribute indicates that the ABI being targeted requires that
1033 an unwind table entry be produce for this function even if we can
1034 show that no exceptions passes by it. This is normally the case for
1035 the ELF x86-64 abi, but it can be disabled for some compilation
1040 Module-Level Inline Assembly
1041 ----------------------------
1043 Modules may contain "module-level inline asm" blocks, which corresponds
1044 to the GCC "file scope inline asm" blocks. These blocks are internally
1045 concatenated by LLVM and treated as a single unit, but may be separated
1046 in the ``.ll`` file if desired. The syntax is very simple:
1048 .. code-block:: llvm
1050 module asm "inline asm code goes here"
1051 module asm "more can go here"
1053 The strings can contain any character by escaping non-printable
1054 characters. The escape sequence used is simply "\\xx" where "xx" is the
1055 two digit hex code for the number.
1057 The inline asm code is simply printed to the machine code .s file when
1058 assembly code is generated.
1060 .. _langref_datalayout:
1065 A module may specify a target specific data layout string that specifies
1066 how data is to be laid out in memory. The syntax for the data layout is
1069 .. code-block:: llvm
1071 target datalayout = "layout specification"
1073 The *layout specification* consists of a list of specifications
1074 separated by the minus sign character ('-'). Each specification starts
1075 with a letter and may include other information after the letter to
1076 define some aspect of the data layout. The specifications accepted are
1080 Specifies that the target lays out data in big-endian form. That is,
1081 the bits with the most significance have the lowest address
1084 Specifies that the target lays out data in little-endian form. That
1085 is, the bits with the least significance have the lowest address
1088 Specifies the natural alignment of the stack in bits. Alignment
1089 promotion of stack variables is limited to the natural stack
1090 alignment to avoid dynamic stack realignment. The stack alignment
1091 must be a multiple of 8-bits. If omitted, the natural stack
1092 alignment defaults to "unspecified", which does not prevent any
1093 alignment promotions.
1094 ``p[n]:<size>:<abi>:<pref>``
1095 This specifies the *size* of a pointer and its ``<abi>`` and
1096 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1097 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1098 preceding ``:`` should be omitted too. The address space, ``n`` is
1099 optional, and if not specified, denotes the default address space 0.
1100 The value of ``n`` must be in the range [1,2^23).
1101 ``i<size>:<abi>:<pref>``
1102 This specifies the alignment for an integer type of a given bit
1103 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1104 ``v<size>:<abi>:<pref>``
1105 This specifies the alignment for a vector type of a given bit
1107 ``f<size>:<abi>:<pref>``
1108 This specifies the alignment for a floating point type of a given bit
1109 ``<size>``. Only values of ``<size>`` that are supported by the target
1110 will work. 32 (float) and 64 (double) are supported on all targets; 80
1111 or 128 (different flavors of long double) are also supported on some
1113 ``a<size>:<abi>:<pref>``
1114 This specifies the alignment for an aggregate type of a given bit
1116 ``s<size>:<abi>:<pref>``
1117 This specifies the alignment for a stack object of a given bit
1119 ``n<size1>:<size2>:<size3>...``
1120 This specifies a set of native integer widths for the target CPU in
1121 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1122 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1123 this set are considered to support most general arithmetic operations
1126 When constructing the data layout for a given target, LLVM starts with a
1127 default set of specifications which are then (possibly) overridden by
1128 the specifications in the ``datalayout`` keyword. The default
1129 specifications are given in this list:
1131 - ``E`` - big endian
1132 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1133 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1134 same as the default address space.
1135 - ``S0`` - natural stack alignment is unspecified
1136 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1137 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1138 - ``i16:16:16`` - i16 is 16-bit aligned
1139 - ``i32:32:32`` - i32 is 32-bit aligned
1140 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1141 alignment of 64-bits
1142 - ``f16:16:16`` - half is 16-bit aligned
1143 - ``f32:32:32`` - float is 32-bit aligned
1144 - ``f64:64:64`` - double is 64-bit aligned
1145 - ``f128:128:128`` - quad is 128-bit aligned
1146 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1147 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1148 - ``a0:0:64`` - aggregates are 64-bit aligned
1150 When LLVM is determining the alignment for a given type, it uses the
1153 #. If the type sought is an exact match for one of the specifications,
1154 that specification is used.
1155 #. If no match is found, and the type sought is an integer type, then
1156 the smallest integer type that is larger than the bitwidth of the
1157 sought type is used. If none of the specifications are larger than
1158 the bitwidth then the largest integer type is used. For example,
1159 given the default specifications above, the i7 type will use the
1160 alignment of i8 (next largest) while both i65 and i256 will use the
1161 alignment of i64 (largest specified).
1162 #. If no match is found, and the type sought is a vector type, then the
1163 largest vector type that is smaller than the sought vector type will
1164 be used as a fall back. This happens because <128 x double> can be
1165 implemented in terms of 64 <2 x double>, for example.
1167 The function of the data layout string may not be what you expect.
1168 Notably, this is not a specification from the frontend of what alignment
1169 the code generator should use.
1171 Instead, if specified, the target data layout is required to match what
1172 the ultimate *code generator* expects. This string is used by the
1173 mid-level optimizers to improve code, and this only works if it matches
1174 what the ultimate code generator uses. If you would like to generate IR
1175 that does not embed this target-specific detail into the IR, then you
1176 don't have to specify the string. This will disable some optimizations
1177 that require precise layout information, but this also prevents those
1178 optimizations from introducing target specificity into the IR.
1185 A module may specify a target triple string that describes the target
1186 host. The syntax for the target triple is simply:
1188 .. code-block:: llvm
1190 target triple = "x86_64-apple-macosx10.7.0"
1192 The *target triple* string consists of a series of identifiers delimited
1193 by the minus sign character ('-'). The canonical forms are:
1197 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1198 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1200 This information is passed along to the backend so that it generates
1201 code for the proper architecture. It's possible to override this on the
1202 command line with the ``-mtriple`` command line option.
1204 .. _pointeraliasing:
1206 Pointer Aliasing Rules
1207 ----------------------
1209 Any memory access must be done through a pointer value associated with
1210 an address range of the memory access, otherwise the behavior is
1211 undefined. Pointer values are associated with address ranges according
1212 to the following rules:
1214 - A pointer value is associated with the addresses associated with any
1215 value it is *based* on.
1216 - An address of a global variable is associated with the address range
1217 of the variable's storage.
1218 - The result value of an allocation instruction is associated with the
1219 address range of the allocated storage.
1220 - A null pointer in the default address-space is associated with no
1222 - An integer constant other than zero or a pointer value returned from
1223 a function not defined within LLVM may be associated with address
1224 ranges allocated through mechanisms other than those provided by
1225 LLVM. Such ranges shall not overlap with any ranges of addresses
1226 allocated by mechanisms provided by LLVM.
1228 A pointer value is *based* on another pointer value according to the
1231 - A pointer value formed from a ``getelementptr`` operation is *based*
1232 on the first operand of the ``getelementptr``.
1233 - The result value of a ``bitcast`` is *based* on the operand of the
1235 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1236 values that contribute (directly or indirectly) to the computation of
1237 the pointer's value.
1238 - The "*based* on" relationship is transitive.
1240 Note that this definition of *"based"* is intentionally similar to the
1241 definition of *"based"* in C99, though it is slightly weaker.
1243 LLVM IR does not associate types with memory. The result type of a
1244 ``load`` merely indicates the size and alignment of the memory from
1245 which to load, as well as the interpretation of the value. The first
1246 operand type of a ``store`` similarly only indicates the size and
1247 alignment of the store.
1249 Consequently, type-based alias analysis, aka TBAA, aka
1250 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1251 :ref:`Metadata <metadata>` may be used to encode additional information
1252 which specialized optimization passes may use to implement type-based
1257 Volatile Memory Accesses
1258 ------------------------
1260 Certain memory accesses, such as :ref:`load <i_load>`'s,
1261 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1262 marked ``volatile``. The optimizers must not change the number of
1263 volatile operations or change their order of execution relative to other
1264 volatile operations. The optimizers *may* change the order of volatile
1265 operations relative to non-volatile operations. This is not Java's
1266 "volatile" and has no cross-thread synchronization behavior.
1268 IR-level volatile loads and stores cannot safely be optimized into
1269 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1270 flagged volatile. Likewise, the backend should never split or merge
1271 target-legal volatile load/store instructions.
1273 .. admonition:: Rationale
1275 Platforms may rely on volatile loads and stores of natively supported
1276 data width to be executed as single instruction. For example, in C
1277 this holds for an l-value of volatile primitive type with native
1278 hardware support, but not necessarily for aggregate types. The
1279 frontend upholds these expectations, which are intentionally
1280 unspecified in the IR. The rules above ensure that IR transformation
1281 do not violate the frontend's contract with the language.
1285 Memory Model for Concurrent Operations
1286 --------------------------------------
1288 The LLVM IR does not define any way to start parallel threads of
1289 execution or to register signal handlers. Nonetheless, there are
1290 platform-specific ways to create them, and we define LLVM IR's behavior
1291 in their presence. This model is inspired by the C++0x memory model.
1293 For a more informal introduction to this model, see the :doc:`Atomics`.
1295 We define a *happens-before* partial order as the least partial order
1298 - Is a superset of single-thread program order, and
1299 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1300 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1301 techniques, like pthread locks, thread creation, thread joining,
1302 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1303 Constraints <ordering>`).
1305 Note that program order does not introduce *happens-before* edges
1306 between a thread and signals executing inside that thread.
1308 Every (defined) read operation (load instructions, memcpy, atomic
1309 loads/read-modify-writes, etc.) R reads a series of bytes written by
1310 (defined) write operations (store instructions, atomic
1311 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1312 section, initialized globals are considered to have a write of the
1313 initializer which is atomic and happens before any other read or write
1314 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1315 may see any write to the same byte, except:
1317 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1318 write\ :sub:`2` happens before R\ :sub:`byte`, then
1319 R\ :sub:`byte` does not see write\ :sub:`1`.
1320 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1321 R\ :sub:`byte` does not see write\ :sub:`3`.
1323 Given that definition, R\ :sub:`byte` is defined as follows:
1325 - If R is volatile, the result is target-dependent. (Volatile is
1326 supposed to give guarantees which can support ``sig_atomic_t`` in
1327 C/C++, and may be used for accesses to addresses which do not behave
1328 like normal memory. It does not generally provide cross-thread
1330 - Otherwise, if there is no write to the same byte that happens before
1331 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1332 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1333 R\ :sub:`byte` returns the value written by that write.
1334 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1335 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1336 Memory Ordering Constraints <ordering>` section for additional
1337 constraints on how the choice is made.
1338 - Otherwise R\ :sub:`byte` returns ``undef``.
1340 R returns the value composed of the series of bytes it read. This
1341 implies that some bytes within the value may be ``undef`` **without**
1342 the entire value being ``undef``. Note that this only defines the
1343 semantics of the operation; it doesn't mean that targets will emit more
1344 than one instruction to read the series of bytes.
1346 Note that in cases where none of the atomic intrinsics are used, this
1347 model places only one restriction on IR transformations on top of what
1348 is required for single-threaded execution: introducing a store to a byte
1349 which might not otherwise be stored is not allowed in general.
1350 (Specifically, in the case where another thread might write to and read
1351 from an address, introducing a store can change a load that may see
1352 exactly one write into a load that may see multiple writes.)
1356 Atomic Memory Ordering Constraints
1357 ----------------------------------
1359 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1360 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1361 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1362 an ordering parameter that determines which other atomic instructions on
1363 the same address they *synchronize with*. These semantics are borrowed
1364 from Java and C++0x, but are somewhat more colloquial. If these
1365 descriptions aren't precise enough, check those specs (see spec
1366 references in the :doc:`atomics guide <Atomics>`).
1367 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1368 differently since they don't take an address. See that instruction's
1369 documentation for details.
1371 For a simpler introduction to the ordering constraints, see the
1375 The set of values that can be read is governed by the happens-before
1376 partial order. A value cannot be read unless some operation wrote
1377 it. This is intended to provide a guarantee strong enough to model
1378 Java's non-volatile shared variables. This ordering cannot be
1379 specified for read-modify-write operations; it is not strong enough
1380 to make them atomic in any interesting way.
1382 In addition to the guarantees of ``unordered``, there is a single
1383 total order for modifications by ``monotonic`` operations on each
1384 address. All modification orders must be compatible with the
1385 happens-before order. There is no guarantee that the modification
1386 orders can be combined to a global total order for the whole program
1387 (and this often will not be possible). The read in an atomic
1388 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1389 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1390 order immediately before the value it writes. If one atomic read
1391 happens before another atomic read of the same address, the later
1392 read must see the same value or a later value in the address's
1393 modification order. This disallows reordering of ``monotonic`` (or
1394 stronger) operations on the same address. If an address is written
1395 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1396 read that address repeatedly, the other threads must eventually see
1397 the write. This corresponds to the C++0x/C1x
1398 ``memory_order_relaxed``.
1400 In addition to the guarantees of ``monotonic``, a
1401 *synchronizes-with* edge may be formed with a ``release`` operation.
1402 This is intended to model C++'s ``memory_order_acquire``.
1404 In addition to the guarantees of ``monotonic``, if this operation
1405 writes a value which is subsequently read by an ``acquire``
1406 operation, it *synchronizes-with* that operation. (This isn't a
1407 complete description; see the C++0x definition of a release
1408 sequence.) This corresponds to the C++0x/C1x
1409 ``memory_order_release``.
1410 ``acq_rel`` (acquire+release)
1411 Acts as both an ``acquire`` and ``release`` operation on its
1412 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1413 ``seq_cst`` (sequentially consistent)
1414 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1415 operation which only reads, ``release`` for an operation which only
1416 writes), there is a global total order on all
1417 sequentially-consistent operations on all addresses, which is
1418 consistent with the *happens-before* partial order and with the
1419 modification orders of all the affected addresses. Each
1420 sequentially-consistent read sees the last preceding write to the
1421 same address in this global order. This corresponds to the C++0x/C1x
1422 ``memory_order_seq_cst`` and Java volatile.
1426 If an atomic operation is marked ``singlethread``, it only *synchronizes
1427 with* or participates in modification and seq\_cst total orderings with
1428 other operations running in the same thread (for example, in signal
1436 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1437 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1438 :ref:`frem <i_frem>`) have the following flags that can set to enable
1439 otherwise unsafe floating point operations
1442 No NaNs - Allow optimizations to assume the arguments and result are not
1443 NaN. Such optimizations are required to retain defined behavior over
1444 NaNs, but the value of the result is undefined.
1447 No Infs - Allow optimizations to assume the arguments and result are not
1448 +/-Inf. Such optimizations are required to retain defined behavior over
1449 +/-Inf, but the value of the result is undefined.
1452 No Signed Zeros - Allow optimizations to treat the sign of a zero
1453 argument or result as insignificant.
1456 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1457 argument rather than perform division.
1460 Fast - Allow algebraically equivalent transformations that may
1461 dramatically change results in floating point (e.g. reassociate). This
1462 flag implies all the others.
1469 The LLVM type system is one of the most important features of the
1470 intermediate representation. Being typed enables a number of
1471 optimizations to be performed on the intermediate representation
1472 directly, without having to do extra analyses on the side before the
1473 transformation. A strong type system makes it easier to read the
1474 generated code and enables novel analyses and transformations that are
1475 not feasible to perform on normal three address code representations.
1477 .. _typeclassifications:
1479 Type Classifications
1480 --------------------
1482 The types fall into a few useful classifications:
1491 * - :ref:`integer <t_integer>`
1492 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1495 * - :ref:`floating point <t_floating>`
1496 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1504 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1505 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1506 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1507 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1509 * - :ref:`primitive <t_primitive>`
1510 - :ref:`label <t_label>`,
1511 :ref:`void <t_void>`,
1512 :ref:`integer <t_integer>`,
1513 :ref:`floating point <t_floating>`,
1514 :ref:`x86mmx <t_x86mmx>`,
1515 :ref:`metadata <t_metadata>`.
1517 * - :ref:`derived <t_derived>`
1518 - :ref:`array <t_array>`,
1519 :ref:`function <t_function>`,
1520 :ref:`pointer <t_pointer>`,
1521 :ref:`structure <t_struct>`,
1522 :ref:`vector <t_vector>`,
1523 :ref:`opaque <t_opaque>`.
1525 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1526 Values of these types are the only ones which can be produced by
1534 The primitive types are the fundamental building blocks of the LLVM
1545 The integer type is a very simple type that simply specifies an
1546 arbitrary bit width for the integer type desired. Any bit width from 1
1547 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1556 The number of bits the integer will occupy is specified by the ``N``
1562 +----------------+------------------------------------------------+
1563 | ``i1`` | a single-bit integer. |
1564 +----------------+------------------------------------------------+
1565 | ``i32`` | a 32-bit integer. |
1566 +----------------+------------------------------------------------+
1567 | ``i1942652`` | a really big integer of over 1 million bits. |
1568 +----------------+------------------------------------------------+
1572 Floating Point Types
1573 ^^^^^^^^^^^^^^^^^^^^
1582 - 16-bit floating point value
1585 - 32-bit floating point value
1588 - 64-bit floating point value
1591 - 128-bit floating point value (112-bit mantissa)
1594 - 80-bit floating point value (X87)
1597 - 128-bit floating point value (two 64-bits)
1607 The x86mmx type represents a value held in an MMX register on an x86
1608 machine. The operations allowed on it are quite limited: parameters and
1609 return values, load and store, and bitcast. User-specified MMX
1610 instructions are represented as intrinsic or asm calls with arguments
1611 and/or results of this type. There are no arrays, vectors or constants
1629 The void type does not represent any value and has no size.
1646 The label type represents code labels.
1663 The metadata type represents embedded metadata. No derived types may be
1664 created from metadata except for :ref:`function <t_function>` arguments.
1678 The real power in LLVM comes from the derived types in the system. This
1679 is what allows a programmer to represent arrays, functions, pointers,
1680 and other useful types. Each of these types contain one or more element
1681 types which may be a primitive type, or another derived type. For
1682 example, it is possible to have a two dimensional array, using an array
1683 as the element type of another array.
1690 Aggregate Types are a subset of derived types that can contain multiple
1691 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1692 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1703 The array type is a very simple derived type that arranges elements
1704 sequentially in memory. The array type requires a size (number of
1705 elements) and an underlying data type.
1712 [<# elements> x <elementtype>]
1714 The number of elements is a constant integer value; ``elementtype`` may
1715 be any type with a size.
1720 +------------------+--------------------------------------+
1721 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1722 +------------------+--------------------------------------+
1723 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1724 +------------------+--------------------------------------+
1725 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1726 +------------------+--------------------------------------+
1728 Here are some examples of multidimensional arrays:
1730 +-----------------------------+----------------------------------------------------------+
1731 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1732 +-----------------------------+----------------------------------------------------------+
1733 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1734 +-----------------------------+----------------------------------------------------------+
1735 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1736 +-----------------------------+----------------------------------------------------------+
1738 There is no restriction on indexing beyond the end of the array implied
1739 by a static type (though there are restrictions on indexing beyond the
1740 bounds of an allocated object in some cases). This means that
1741 single-dimension 'variable sized array' addressing can be implemented in
1742 LLVM with a zero length array type. An implementation of 'pascal style
1743 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1754 The function type can be thought of as a function signature. It consists of a
1755 return type and a list of formal parameter types. The return type of a function
1756 type is a void type or first class type --- except for :ref:`label <t_label>`
1757 and :ref:`metadata <t_metadata>` types.
1764 <returntype> (<parameter list>)
1766 ...where '``<parameter list>``' is a comma-separated list of type
1767 specifiers. Optionally, the parameter list may include a type ``...``, which
1768 indicates that the function takes a variable number of arguments. Variable
1769 argument functions can access their arguments with the :ref:`variable argument
1770 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1771 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1776 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1777 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1778 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1779 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1780 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1781 | ``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. |
1782 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1783 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1784 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1794 The structure type is used to represent a collection of data members
1795 together in memory. The elements of a structure may be any type that has
1798 Structures in memory are accessed using '``load``' and '``store``' by
1799 getting a pointer to a field with the '``getelementptr``' instruction.
1800 Structures in registers are accessed using the '``extractvalue``' and
1801 '``insertvalue``' instructions.
1803 Structures may optionally be "packed" structures, which indicate that
1804 the alignment of the struct is one byte, and that there is no padding
1805 between the elements. In non-packed structs, padding between field types
1806 is inserted as defined by the DataLayout string in the module, which is
1807 required to match what the underlying code generator expects.
1809 Structures can either be "literal" or "identified". A literal structure
1810 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1811 identified types are always defined at the top level with a name.
1812 Literal types are uniqued by their contents and can never be recursive
1813 or opaque since there is no way to write one. Identified types can be
1814 recursive, can be opaqued, and are never uniqued.
1821 %T1 = type { <type list> } ; Identified normal struct type
1822 %T2 = type <{ <type list> }> ; Identified packed struct type
1827 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1828 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1829 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1830 | ``{ 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``. |
1831 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1832 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1833 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1837 Opaque Structure Types
1838 ^^^^^^^^^^^^^^^^^^^^^^
1843 Opaque structure types are used to represent named structure types that
1844 do not have a body specified. This corresponds (for example) to the C
1845 notion of a forward declared structure.
1858 +--------------+-------------------+
1859 | ``opaque`` | An opaque type. |
1860 +--------------+-------------------+
1870 The pointer type is used to specify memory locations. Pointers are
1871 commonly used to reference objects in memory.
1873 Pointer types may have an optional address space attribute defining the
1874 numbered address space where the pointed-to object resides. The default
1875 address space is number zero. The semantics of non-zero address spaces
1876 are target-specific.
1878 Note that LLVM does not permit pointers to void (``void*``) nor does it
1879 permit pointers to labels (``label*``). Use ``i8*`` instead.
1891 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1892 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1893 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1894 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1895 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1896 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1897 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1907 A vector type is a simple derived type that represents a vector of
1908 elements. Vector types are used when multiple primitive data are
1909 operated in parallel using a single instruction (SIMD). A vector type
1910 requires a size (number of elements) and an underlying primitive data
1911 type. Vector types are considered :ref:`first class <t_firstclass>`.
1918 < <# elements> x <elementtype> >
1920 The number of elements is a constant integer value larger than 0;
1921 elementtype may be any integer or floating point type, or a pointer to
1922 these types. Vectors of size zero are not allowed.
1927 +-------------------+--------------------------------------------------+
1928 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1929 +-------------------+--------------------------------------------------+
1930 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1931 +-------------------+--------------------------------------------------+
1932 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1933 +-------------------+--------------------------------------------------+
1934 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1935 +-------------------+--------------------------------------------------+
1940 LLVM has several different basic types of constants. This section
1941 describes them all and their syntax.
1946 **Boolean constants**
1947 The two strings '``true``' and '``false``' are both valid constants
1949 **Integer constants**
1950 Standard integers (such as '4') are constants of the
1951 :ref:`integer <t_integer>` type. Negative numbers may be used with
1953 **Floating point constants**
1954 Floating point constants use standard decimal notation (e.g.
1955 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1956 hexadecimal notation (see below). The assembler requires the exact
1957 decimal value of a floating-point constant. For example, the
1958 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1959 decimal in binary. Floating point constants must have a :ref:`floating
1960 point <t_floating>` type.
1961 **Null pointer constants**
1962 The identifier '``null``' is recognized as a null pointer constant
1963 and must be of :ref:`pointer type <t_pointer>`.
1965 The one non-intuitive notation for constants is the hexadecimal form of
1966 floating point constants. For example, the form
1967 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1968 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1969 constants are required (and the only time that they are generated by the
1970 disassembler) is when a floating point constant must be emitted but it
1971 cannot be represented as a decimal floating point number in a reasonable
1972 number of digits. For example, NaN's, infinities, and other special
1973 values are represented in their IEEE hexadecimal format so that assembly
1974 and disassembly do not cause any bits to change in the constants.
1976 When using the hexadecimal form, constants of types half, float, and
1977 double are represented using the 16-digit form shown above (which
1978 matches the IEEE754 representation for double); half and float values
1979 must, however, be exactly representable as IEEE 754 half and single
1980 precision, respectively. Hexadecimal format is always used for long
1981 double, and there are three forms of long double. The 80-bit format used
1982 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1983 128-bit format used by PowerPC (two adjacent doubles) is represented by
1984 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1985 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1986 will only work if they match the long double format on your target.
1987 The IEEE 16-bit format (half precision) is represented by ``0xH``
1988 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1989 (sign bit at the left).
1991 There are no constants of type x86mmx.
1993 .. _complexconstants:
1998 Complex constants are a (potentially recursive) combination of simple
1999 constants and smaller complex constants.
2001 **Structure constants**
2002 Structure constants are represented with notation similar to
2003 structure type definitions (a comma separated list of elements,
2004 surrounded by braces (``{}``)). For example:
2005 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2006 "``@G = external global i32``". Structure constants must have
2007 :ref:`structure type <t_struct>`, and the number and types of elements
2008 must match those specified by the type.
2010 Array constants are represented with notation similar to array type
2011 definitions (a comma separated list of elements, surrounded by
2012 square brackets (``[]``)). For example:
2013 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2014 :ref:`array type <t_array>`, and the number and types of elements must
2015 match those specified by the type.
2016 **Vector constants**
2017 Vector constants are represented with notation similar to vector
2018 type definitions (a comma separated list of elements, surrounded by
2019 less-than/greater-than's (``<>``)). For example:
2020 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2021 must have :ref:`vector type <t_vector>`, and the number and types of
2022 elements must match those specified by the type.
2023 **Zero initialization**
2024 The string '``zeroinitializer``' can be used to zero initialize a
2025 value to zero of *any* type, including scalar and
2026 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2027 having to print large zero initializers (e.g. for large arrays) and
2028 is always exactly equivalent to using explicit zero initializers.
2030 A metadata node is a structure-like constant with :ref:`metadata
2031 type <t_metadata>`. For example:
2032 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2033 constants that are meant to be interpreted as part of the
2034 instruction stream, metadata is a place to attach additional
2035 information such as debug info.
2037 Global Variable and Function Addresses
2038 --------------------------------------
2040 The addresses of :ref:`global variables <globalvars>` and
2041 :ref:`functions <functionstructure>` are always implicitly valid
2042 (link-time) constants. These constants are explicitly referenced when
2043 the :ref:`identifier for the global <identifiers>` is used and always have
2044 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2047 .. code-block:: llvm
2051 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2058 The string '``undef``' can be used anywhere a constant is expected, and
2059 indicates that the user of the value may receive an unspecified
2060 bit-pattern. Undefined values may be of any type (other than '``label``'
2061 or '``void``') and be used anywhere a constant is permitted.
2063 Undefined values are useful because they indicate to the compiler that
2064 the program is well defined no matter what value is used. This gives the
2065 compiler more freedom to optimize. Here are some examples of
2066 (potentially surprising) transformations that are valid (in pseudo IR):
2068 .. code-block:: llvm
2078 This is safe because all of the output bits are affected by the undef
2079 bits. Any output bit can have a zero or one depending on the input bits.
2081 .. code-block:: llvm
2092 These logical operations have bits that are not always affected by the
2093 input. For example, if ``%X`` has a zero bit, then the output of the
2094 '``and``' operation will always be a zero for that bit, no matter what
2095 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2096 optimize or assume that the result of the '``and``' is '``undef``'.
2097 However, it is safe to assume that all bits of the '``undef``' could be
2098 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2099 all the bits of the '``undef``' operand to the '``or``' could be set,
2100 allowing the '``or``' to be folded to -1.
2102 .. code-block:: llvm
2104 %A = select undef, %X, %Y
2105 %B = select undef, 42, %Y
2106 %C = select %X, %Y, undef
2116 This set of examples shows that undefined '``select``' (and conditional
2117 branch) conditions can go *either way*, but they have to come from one
2118 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2119 both known to have a clear low bit, then ``%A`` would have to have a
2120 cleared low bit. However, in the ``%C`` example, the optimizer is
2121 allowed to assume that the '``undef``' operand could be the same as
2122 ``%Y``, allowing the whole '``select``' to be eliminated.
2124 .. code-block:: llvm
2126 %A = xor undef, undef
2143 This example points out that two '``undef``' operands are not
2144 necessarily the same. This can be surprising to people (and also matches
2145 C semantics) where they assume that "``X^X``" is always zero, even if
2146 ``X`` is undefined. This isn't true for a number of reasons, but the
2147 short answer is that an '``undef``' "variable" can arbitrarily change
2148 its value over its "live range". This is true because the variable
2149 doesn't actually *have a live range*. Instead, the value is logically
2150 read from arbitrary registers that happen to be around when needed, so
2151 the value is not necessarily consistent over time. In fact, ``%A`` and
2152 ``%C`` need to have the same semantics or the core LLVM "replace all
2153 uses with" concept would not hold.
2155 .. code-block:: llvm
2163 These examples show the crucial difference between an *undefined value*
2164 and *undefined behavior*. An undefined value (like '``undef``') is
2165 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2166 operation can be constant folded to '``undef``', because the '``undef``'
2167 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2168 However, in the second example, we can make a more aggressive
2169 assumption: because the ``undef`` is allowed to be an arbitrary value,
2170 we are allowed to assume that it could be zero. Since a divide by zero
2171 has *undefined behavior*, we are allowed to assume that the operation
2172 does not execute at all. This allows us to delete the divide and all
2173 code after it. Because the undefined operation "can't happen", the
2174 optimizer can assume that it occurs in dead code.
2176 .. code-block:: llvm
2178 a: store undef -> %X
2179 b: store %X -> undef
2184 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2185 value can be assumed to not have any effect; we can assume that the
2186 value is overwritten with bits that happen to match what was already
2187 there. However, a store *to* an undefined location could clobber
2188 arbitrary memory, therefore, it has undefined behavior.
2195 Poison values are similar to :ref:`undef values <undefvalues>`, however
2196 they also represent the fact that an instruction or constant expression
2197 which cannot evoke side effects has nevertheless detected a condition
2198 which results in undefined behavior.
2200 There is currently no way of representing a poison value in the IR; they
2201 only exist when produced by operations such as :ref:`add <i_add>` with
2204 Poison value behavior is defined in terms of value *dependence*:
2206 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2207 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2208 their dynamic predecessor basic block.
2209 - Function arguments depend on the corresponding actual argument values
2210 in the dynamic callers of their functions.
2211 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2212 instructions that dynamically transfer control back to them.
2213 - :ref:`Invoke <i_invoke>` instructions depend on the
2214 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2215 call instructions that dynamically transfer control back to them.
2216 - Non-volatile loads and stores depend on the most recent stores to all
2217 of the referenced memory addresses, following the order in the IR
2218 (including loads and stores implied by intrinsics such as
2219 :ref:`@llvm.memcpy <int_memcpy>`.)
2220 - An instruction with externally visible side effects depends on the
2221 most recent preceding instruction with externally visible side
2222 effects, following the order in the IR. (This includes :ref:`volatile
2223 operations <volatile>`.)
2224 - An instruction *control-depends* on a :ref:`terminator
2225 instruction <terminators>` if the terminator instruction has
2226 multiple successors and the instruction is always executed when
2227 control transfers to one of the successors, and may not be executed
2228 when control is transferred to another.
2229 - Additionally, an instruction also *control-depends* on a terminator
2230 instruction if the set of instructions it otherwise depends on would
2231 be different if the terminator had transferred control to a different
2233 - Dependence is transitive.
2235 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2236 with the additional affect that any instruction which has a *dependence*
2237 on a poison value has undefined behavior.
2239 Here are some examples:
2241 .. code-block:: llvm
2244 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2245 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2246 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2247 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2249 store i32 %poison, i32* @g ; Poison value stored to memory.
2250 %poison2 = load i32* @g ; Poison value loaded back from memory.
2252 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2254 %narrowaddr = bitcast i32* @g to i16*
2255 %wideaddr = bitcast i32* @g to i64*
2256 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2257 %poison4 = load i64* %wideaddr ; Returns a poison value.
2259 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2260 br i1 %cmp, label %true, label %end ; Branch to either destination.
2263 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2264 ; it has undefined behavior.
2268 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2269 ; Both edges into this PHI are
2270 ; control-dependent on %cmp, so this
2271 ; always results in a poison value.
2273 store volatile i32 0, i32* @g ; This would depend on the store in %true
2274 ; if %cmp is true, or the store in %entry
2275 ; otherwise, so this is undefined behavior.
2277 br i1 %cmp, label %second_true, label %second_end
2278 ; The same branch again, but this time the
2279 ; true block doesn't have side effects.
2286 store volatile i32 0, i32* @g ; This time, the instruction always depends
2287 ; on the store in %end. Also, it is
2288 ; control-equivalent to %end, so this is
2289 ; well-defined (ignoring earlier undefined
2290 ; behavior in this example).
2294 Addresses of Basic Blocks
2295 -------------------------
2297 ``blockaddress(@function, %block)``
2299 The '``blockaddress``' constant computes the address of the specified
2300 basic block in the specified function, and always has an ``i8*`` type.
2301 Taking the address of the entry block is illegal.
2303 This value only has defined behavior when used as an operand to the
2304 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2305 against null. Pointer equality tests between labels addresses results in
2306 undefined behavior --- though, again, comparison against null is ok, and
2307 no label is equal to the null pointer. This may be passed around as an
2308 opaque pointer sized value as long as the bits are not inspected. This
2309 allows ``ptrtoint`` and arithmetic to be performed on these values so
2310 long as the original value is reconstituted before the ``indirectbr``
2313 Finally, some targets may provide defined semantics when using the value
2314 as the operand to an inline assembly, but that is target specific.
2318 Constant Expressions
2319 --------------------
2321 Constant expressions are used to allow expressions involving other
2322 constants to be used as constants. Constant expressions may be of any
2323 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2324 that does not have side effects (e.g. load and call are not supported).
2325 The following is the syntax for constant expressions:
2327 ``trunc (CST to TYPE)``
2328 Truncate a constant to another type. The bit size of CST must be
2329 larger than the bit size of TYPE. Both types must be integers.
2330 ``zext (CST to TYPE)``
2331 Zero extend a constant to another type. The bit size of CST must be
2332 smaller than the bit size of TYPE. Both types must be integers.
2333 ``sext (CST to TYPE)``
2334 Sign extend a constant to another type. The bit size of CST must be
2335 smaller than the bit size of TYPE. Both types must be integers.
2336 ``fptrunc (CST to TYPE)``
2337 Truncate a floating point constant to another floating point type.
2338 The size of CST must be larger than the size of TYPE. Both types
2339 must be floating point.
2340 ``fpext (CST to TYPE)``
2341 Floating point extend a constant to another type. The size of CST
2342 must be smaller or equal to the size of TYPE. Both types must be
2344 ``fptoui (CST to TYPE)``
2345 Convert a floating point constant to the corresponding unsigned
2346 integer constant. TYPE must be a scalar or vector integer type. CST
2347 must be of scalar or vector floating point type. Both CST and TYPE
2348 must be scalars, or vectors of the same number of elements. If the
2349 value won't fit in the integer type, the results are undefined.
2350 ``fptosi (CST to TYPE)``
2351 Convert a floating point constant to the corresponding signed
2352 integer constant. TYPE must be a scalar or vector integer type. CST
2353 must be of scalar or vector floating point type. Both CST and TYPE
2354 must be scalars, or vectors of the same number of elements. If the
2355 value won't fit in the integer type, the results are undefined.
2356 ``uitofp (CST to TYPE)``
2357 Convert an unsigned integer constant to the corresponding floating
2358 point constant. TYPE must be a scalar or vector floating point type.
2359 CST must be of scalar or vector integer type. Both CST and TYPE must
2360 be scalars, or vectors of the same number of elements. If the value
2361 won't fit in the floating point type, the results are undefined.
2362 ``sitofp (CST to TYPE)``
2363 Convert a signed integer constant to the corresponding floating
2364 point constant. TYPE must be a scalar or vector floating point type.
2365 CST must be of scalar or vector integer type. Both CST and TYPE must
2366 be scalars, or vectors of the same number of elements. If the value
2367 won't fit in the floating point type, the results are undefined.
2368 ``ptrtoint (CST to TYPE)``
2369 Convert a pointer typed constant to the corresponding integer
2370 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2371 pointer type. The ``CST`` value is zero extended, truncated, or
2372 unchanged to make it fit in ``TYPE``.
2373 ``inttoptr (CST to TYPE)``
2374 Convert an integer constant to a pointer constant. TYPE must be a
2375 pointer type. CST must be of integer type. The CST value is zero
2376 extended, truncated, or unchanged to make it fit in a pointer size.
2377 This one is *really* dangerous!
2378 ``bitcast (CST to TYPE)``
2379 Convert a constant, CST, to another TYPE. The constraints of the
2380 operands are the same as those for the :ref:`bitcast
2381 instruction <i_bitcast>`.
2382 ``addrspacecast (CST to TYPE)``
2383 Convert a constant pointer or constant vector of pointer, CST, to another
2384 TYPE in a different address space. The constraints of the operands are the
2385 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2386 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2387 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2388 constants. As with the :ref:`getelementptr <i_getelementptr>`
2389 instruction, the index list may have zero or more indexes, which are
2390 required to make sense for the type of "CSTPTR".
2391 ``select (COND, VAL1, VAL2)``
2392 Perform the :ref:`select operation <i_select>` on constants.
2393 ``icmp COND (VAL1, VAL2)``
2394 Performs the :ref:`icmp operation <i_icmp>` on constants.
2395 ``fcmp COND (VAL1, VAL2)``
2396 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2397 ``extractelement (VAL, IDX)``
2398 Perform the :ref:`extractelement operation <i_extractelement>` on
2400 ``insertelement (VAL, ELT, IDX)``
2401 Perform the :ref:`insertelement operation <i_insertelement>` on
2403 ``shufflevector (VEC1, VEC2, IDXMASK)``
2404 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2406 ``extractvalue (VAL, IDX0, IDX1, ...)``
2407 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2408 constants. The index list is interpreted in a similar manner as
2409 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2410 least one index value must be specified.
2411 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2412 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2413 The index list is interpreted in a similar manner as indices in a
2414 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2415 value must be specified.
2416 ``OPCODE (LHS, RHS)``
2417 Perform the specified operation of the LHS and RHS constants. OPCODE
2418 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2419 binary <bitwiseops>` operations. The constraints on operands are
2420 the same as those for the corresponding instruction (e.g. no bitwise
2421 operations on floating point values are allowed).
2428 Inline Assembler Expressions
2429 ----------------------------
2431 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2432 Inline Assembly <moduleasm>`) through the use of a special value. This
2433 value represents the inline assembler as a string (containing the
2434 instructions to emit), a list of operand constraints (stored as a
2435 string), a flag that indicates whether or not the inline asm expression
2436 has side effects, and a flag indicating whether the function containing
2437 the asm needs to align its stack conservatively. An example inline
2438 assembler expression is:
2440 .. code-block:: llvm
2442 i32 (i32) asm "bswap $0", "=r,r"
2444 Inline assembler expressions may **only** be used as the callee operand
2445 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2446 Thus, typically we have:
2448 .. code-block:: llvm
2450 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2452 Inline asms with side effects not visible in the constraint list must be
2453 marked as having side effects. This is done through the use of the
2454 '``sideeffect``' keyword, like so:
2456 .. code-block:: llvm
2458 call void asm sideeffect "eieio", ""()
2460 In some cases inline asms will contain code that will not work unless
2461 the stack is aligned in some way, such as calls or SSE instructions on
2462 x86, yet will not contain code that does that alignment within the asm.
2463 The compiler should make conservative assumptions about what the asm
2464 might contain and should generate its usual stack alignment code in the
2465 prologue if the '``alignstack``' keyword is present:
2467 .. code-block:: llvm
2469 call void asm alignstack "eieio", ""()
2471 Inline asms also support using non-standard assembly dialects. The
2472 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2473 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2474 the only supported dialects. An example is:
2476 .. code-block:: llvm
2478 call void asm inteldialect "eieio", ""()
2480 If multiple keywords appear the '``sideeffect``' keyword must come
2481 first, the '``alignstack``' keyword second and the '``inteldialect``'
2487 The call instructions that wrap inline asm nodes may have a
2488 "``!srcloc``" MDNode attached to it that contains a list of constant
2489 integers. If present, the code generator will use the integer as the
2490 location cookie value when report errors through the ``LLVMContext``
2491 error reporting mechanisms. This allows a front-end to correlate backend
2492 errors that occur with inline asm back to the source code that produced
2495 .. code-block:: llvm
2497 call void asm sideeffect "something bad", ""(), !srcloc !42
2499 !42 = !{ i32 1234567 }
2501 It is up to the front-end to make sense of the magic numbers it places
2502 in the IR. If the MDNode contains multiple constants, the code generator
2503 will use the one that corresponds to the line of the asm that the error
2508 Metadata Nodes and Metadata Strings
2509 -----------------------------------
2511 LLVM IR allows metadata to be attached to instructions in the program
2512 that can convey extra information about the code to the optimizers and
2513 code generator. One example application of metadata is source-level
2514 debug information. There are two metadata primitives: strings and nodes.
2515 All metadata has the ``metadata`` type and is identified in syntax by a
2516 preceding exclamation point ('``!``').
2518 A metadata string is a string surrounded by double quotes. It can
2519 contain any character by escaping non-printable characters with
2520 "``\xx``" where "``xx``" is the two digit hex code. For example:
2523 Metadata nodes are represented with notation similar to structure
2524 constants (a comma separated list of elements, surrounded by braces and
2525 preceded by an exclamation point). Metadata nodes can have any values as
2526 their operand. For example:
2528 .. code-block:: llvm
2530 !{ metadata !"test\00", i32 10}
2532 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2533 metadata nodes, which can be looked up in the module symbol table. For
2536 .. code-block:: llvm
2538 !foo = metadata !{!4, !3}
2540 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2541 function is using two metadata arguments:
2543 .. code-block:: llvm
2545 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2547 Metadata can be attached with an instruction. Here metadata ``!21`` is
2548 attached to the ``add`` instruction using the ``!dbg`` identifier:
2550 .. code-block:: llvm
2552 %indvar.next = add i64 %indvar, 1, !dbg !21
2554 More information about specific metadata nodes recognized by the
2555 optimizers and code generator is found below.
2560 In LLVM IR, memory does not have types, so LLVM's own type system is not
2561 suitable for doing TBAA. Instead, metadata is added to the IR to
2562 describe a type system of a higher level language. This can be used to
2563 implement typical C/C++ TBAA, but it can also be used to implement
2564 custom alias analysis behavior for other languages.
2566 The current metadata format is very simple. TBAA metadata nodes have up
2567 to three fields, e.g.:
2569 .. code-block:: llvm
2571 !0 = metadata !{ metadata !"an example type tree" }
2572 !1 = metadata !{ metadata !"int", metadata !0 }
2573 !2 = metadata !{ metadata !"float", metadata !0 }
2574 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2576 The first field is an identity field. It can be any value, usually a
2577 metadata string, which uniquely identifies the type. The most important
2578 name in the tree is the name of the root node. Two trees with different
2579 root node names are entirely disjoint, even if they have leaves with
2582 The second field identifies the type's parent node in the tree, or is
2583 null or omitted for a root node. A type is considered to alias all of
2584 its descendants and all of its ancestors in the tree. Also, a type is
2585 considered to alias all types in other trees, so that bitcode produced
2586 from multiple front-ends is handled conservatively.
2588 If the third field is present, it's an integer which if equal to 1
2589 indicates that the type is "constant" (meaning
2590 ``pointsToConstantMemory`` should return true; see `other useful
2591 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2593 '``tbaa.struct``' Metadata
2594 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2596 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2597 aggregate assignment operations in C and similar languages, however it
2598 is defined to copy a contiguous region of memory, which is more than
2599 strictly necessary for aggregate types which contain holes due to
2600 padding. Also, it doesn't contain any TBAA information about the fields
2603 ``!tbaa.struct`` metadata can describe which memory subregions in a
2604 memcpy are padding and what the TBAA tags of the struct are.
2606 The current metadata format is very simple. ``!tbaa.struct`` metadata
2607 nodes are a list of operands which are in conceptual groups of three.
2608 For each group of three, the first operand gives the byte offset of a
2609 field in bytes, the second gives its size in bytes, and the third gives
2612 .. code-block:: llvm
2614 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2616 This describes a struct with two fields. The first is at offset 0 bytes
2617 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2618 and has size 4 bytes and has tbaa tag !2.
2620 Note that the fields need not be contiguous. In this example, there is a
2621 4 byte gap between the two fields. This gap represents padding which
2622 does not carry useful data and need not be preserved.
2624 '``fpmath``' Metadata
2625 ^^^^^^^^^^^^^^^^^^^^^
2627 ``fpmath`` metadata may be attached to any instruction of floating point
2628 type. It can be used to express the maximum acceptable error in the
2629 result of that instruction, in ULPs, thus potentially allowing the
2630 compiler to use a more efficient but less accurate method of computing
2631 it. ULP is defined as follows:
2633 If ``x`` is a real number that lies between two finite consecutive
2634 floating-point numbers ``a`` and ``b``, without being equal to one
2635 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2636 distance between the two non-equal finite floating-point numbers
2637 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2639 The metadata node shall consist of a single positive floating point
2640 number representing the maximum relative error, for example:
2642 .. code-block:: llvm
2644 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2646 '``range``' Metadata
2647 ^^^^^^^^^^^^^^^^^^^^
2649 ``range`` metadata may be attached only to loads of integer types. It
2650 expresses the possible ranges the loaded value is in. The ranges are
2651 represented with a flattened list of integers. The loaded value is known
2652 to be in the union of the ranges defined by each consecutive pair. Each
2653 pair has the following properties:
2655 - The type must match the type loaded by the instruction.
2656 - The pair ``a,b`` represents the range ``[a,b)``.
2657 - Both ``a`` and ``b`` are constants.
2658 - The range is allowed to wrap.
2659 - The range should not represent the full or empty set. That is,
2662 In addition, the pairs must be in signed order of the lower bound and
2663 they must be non-contiguous.
2667 .. code-block:: llvm
2669 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2670 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2671 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2672 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2674 !0 = metadata !{ i8 0, i8 2 }
2675 !1 = metadata !{ i8 255, i8 2 }
2676 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2677 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2682 It is sometimes useful to attach information to loop constructs. Currently,
2683 loop metadata is implemented as metadata attached to the branch instruction
2684 in the loop latch block. This type of metadata refer to a metadata node that is
2685 guaranteed to be separate for each loop. The loop identifier metadata is
2686 specified with the name ``llvm.loop``.
2688 The loop identifier metadata is implemented using a metadata that refers to
2689 itself to avoid merging it with any other identifier metadata, e.g.,
2690 during module linkage or function inlining. That is, each loop should refer
2691 to their own identification metadata even if they reside in separate functions.
2692 The following example contains loop identifier metadata for two separate loop
2695 .. code-block:: llvm
2697 !0 = metadata !{ metadata !0 }
2698 !1 = metadata !{ metadata !1 }
2700 The loop identifier metadata can be used to specify additional per-loop
2701 metadata. Any operands after the first operand can be treated as user-defined
2702 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2703 by the loop vectorizer to indicate how many times to unroll the loop:
2705 .. code-block:: llvm
2707 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2709 !0 = metadata !{ metadata !0, metadata !1 }
2710 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2715 Metadata types used to annotate memory accesses with information helpful
2716 for optimizations are prefixed with ``llvm.mem``.
2718 '``llvm.mem.parallel_loop_access``' Metadata
2719 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2721 For a loop to be parallel, in addition to using
2722 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2723 also all of the memory accessing instructions in the loop body need to be
2724 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2725 is at least one memory accessing instruction not marked with the metadata,
2726 the loop must be considered a sequential loop. This causes parallel loops to be
2727 converted to sequential loops due to optimization passes that are unaware of
2728 the parallel semantics and that insert new memory instructions to the loop
2731 Example of a loop that is considered parallel due to its correct use of
2732 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2733 metadata types that refer to the same loop identifier metadata.
2735 .. code-block:: llvm
2739 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2741 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2743 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2747 !0 = metadata !{ metadata !0 }
2749 It is also possible to have nested parallel loops. In that case the
2750 memory accesses refer to a list of loop identifier metadata nodes instead of
2751 the loop identifier metadata node directly:
2753 .. code-block:: llvm
2760 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2762 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2764 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2768 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2770 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2772 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2774 outer.for.end: ; preds = %for.body
2776 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2777 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2778 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2780 '``llvm.vectorizer``'
2781 ^^^^^^^^^^^^^^^^^^^^^
2783 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2784 vectorization parameters such as vectorization factor and unroll factor.
2786 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2787 loop identification metadata.
2789 '``llvm.vectorizer.unroll``' Metadata
2790 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2792 This metadata instructs the loop vectorizer to unroll the specified
2793 loop exactly ``N`` times.
2795 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2796 operand is an integer specifying the unroll factor. For example:
2798 .. code-block:: llvm
2800 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2802 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2805 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2806 determined automatically.
2808 '``llvm.vectorizer.width``' Metadata
2809 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2811 This metadata sets the target width of the vectorizer to ``N``. Without
2812 this metadata, the vectorizer will choose a width automatically.
2813 Regardless of this metadata, the vectorizer will only vectorize loops if
2814 it believes it is valid to do so.
2816 The first operand is the string ``llvm.vectorizer.width`` and the second
2817 operand is an integer specifying the width. For example:
2819 .. code-block:: llvm
2821 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2823 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2826 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2829 Module Flags Metadata
2830 =====================
2832 Information about the module as a whole is difficult to convey to LLVM's
2833 subsystems. The LLVM IR isn't sufficient to transmit this information.
2834 The ``llvm.module.flags`` named metadata exists in order to facilitate
2835 this. These flags are in the form of key / value pairs --- much like a
2836 dictionary --- making it easy for any subsystem who cares about a flag to
2839 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2840 Each triplet has the following form:
2842 - The first element is a *behavior* flag, which specifies the behavior
2843 when two (or more) modules are merged together, and it encounters two
2844 (or more) metadata with the same ID. The supported behaviors are
2846 - The second element is a metadata string that is a unique ID for the
2847 metadata. Each module may only have one flag entry for each unique ID (not
2848 including entries with the **Require** behavior).
2849 - The third element is the value of the flag.
2851 When two (or more) modules are merged together, the resulting
2852 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2853 each unique metadata ID string, there will be exactly one entry in the merged
2854 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2855 be determined by the merge behavior flag, as described below. The only exception
2856 is that entries with the *Require* behavior are always preserved.
2858 The following behaviors are supported:
2869 Emits an error if two values disagree, otherwise the resulting value
2870 is that of the operands.
2874 Emits a warning if two values disagree. The result value will be the
2875 operand for the flag from the first module being linked.
2879 Adds a requirement that another module flag be present and have a
2880 specified value after linking is performed. The value must be a
2881 metadata pair, where the first element of the pair is the ID of the
2882 module flag to be restricted, and the second element of the pair is
2883 the value the module flag should be restricted to. This behavior can
2884 be used to restrict the allowable results (via triggering of an
2885 error) of linking IDs with the **Override** behavior.
2889 Uses the specified value, regardless of the behavior or value of the
2890 other module. If both modules specify **Override**, but the values
2891 differ, an error will be emitted.
2895 Appends the two values, which are required to be metadata nodes.
2899 Appends the two values, which are required to be metadata
2900 nodes. However, duplicate entries in the second list are dropped
2901 during the append operation.
2903 It is an error for a particular unique flag ID to have multiple behaviors,
2904 except in the case of **Require** (which adds restrictions on another metadata
2905 value) or **Override**.
2907 An example of module flags:
2909 .. code-block:: llvm
2911 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2912 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2913 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2914 !3 = metadata !{ i32 3, metadata !"qux",
2916 metadata !"foo", i32 1
2919 !llvm.module.flags = !{ !0, !1, !2, !3 }
2921 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2922 if two or more ``!"foo"`` flags are seen is to emit an error if their
2923 values are not equal.
2925 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2926 behavior if two or more ``!"bar"`` flags are seen is to use the value
2929 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2930 behavior if two or more ``!"qux"`` flags are seen is to emit a
2931 warning if their values are not equal.
2933 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2937 metadata !{ metadata !"foo", i32 1 }
2939 The behavior is to emit an error if the ``llvm.module.flags`` does not
2940 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2943 Objective-C Garbage Collection Module Flags Metadata
2944 ----------------------------------------------------
2946 On the Mach-O platform, Objective-C stores metadata about garbage
2947 collection in a special section called "image info". The metadata
2948 consists of a version number and a bitmask specifying what types of
2949 garbage collection are supported (if any) by the file. If two or more
2950 modules are linked together their garbage collection metadata needs to
2951 be merged rather than appended together.
2953 The Objective-C garbage collection module flags metadata consists of the
2954 following key-value pairs:
2963 * - ``Objective-C Version``
2964 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2966 * - ``Objective-C Image Info Version``
2967 - **[Required]** --- The version of the image info section. Currently
2970 * - ``Objective-C Image Info Section``
2971 - **[Required]** --- The section to place the metadata. Valid values are
2972 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2973 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2974 Objective-C ABI version 2.
2976 * - ``Objective-C Garbage Collection``
2977 - **[Required]** --- Specifies whether garbage collection is supported or
2978 not. Valid values are 0, for no garbage collection, and 2, for garbage
2979 collection supported.
2981 * - ``Objective-C GC Only``
2982 - **[Optional]** --- Specifies that only garbage collection is supported.
2983 If present, its value must be 6. This flag requires that the
2984 ``Objective-C Garbage Collection`` flag have the value 2.
2986 Some important flag interactions:
2988 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2989 merged with a module with ``Objective-C Garbage Collection`` set to
2990 2, then the resulting module has the
2991 ``Objective-C Garbage Collection`` flag set to 0.
2992 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2993 merged with a module with ``Objective-C GC Only`` set to 6.
2995 Automatic Linker Flags Module Flags Metadata
2996 --------------------------------------------
2998 Some targets support embedding flags to the linker inside individual object
2999 files. Typically this is used in conjunction with language extensions which
3000 allow source files to explicitly declare the libraries they depend on, and have
3001 these automatically be transmitted to the linker via object files.
3003 These flags are encoded in the IR using metadata in the module flags section,
3004 using the ``Linker Options`` key. The merge behavior for this flag is required
3005 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3006 node which should be a list of other metadata nodes, each of which should be a
3007 list of metadata strings defining linker options.
3009 For example, the following metadata section specifies two separate sets of
3010 linker options, presumably to link against ``libz`` and the ``Cocoa``
3013 !0 = metadata !{ i32 6, metadata !"Linker Options",
3015 metadata !{ metadata !"-lz" },
3016 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3017 !llvm.module.flags = !{ !0 }
3019 The metadata encoding as lists of lists of options, as opposed to a collapsed
3020 list of options, is chosen so that the IR encoding can use multiple option
3021 strings to specify e.g., a single library, while still having that specifier be
3022 preserved as an atomic element that can be recognized by a target specific
3023 assembly writer or object file emitter.
3025 Each individual option is required to be either a valid option for the target's
3026 linker, or an option that is reserved by the target specific assembly writer or
3027 object file emitter. No other aspect of these options is defined by the IR.
3029 .. _intrinsicglobalvariables:
3031 Intrinsic Global Variables
3032 ==========================
3034 LLVM has a number of "magic" global variables that contain data that
3035 affect code generation or other IR semantics. These are documented here.
3036 All globals of this sort should have a section specified as
3037 "``llvm.metadata``". This section and all globals that start with
3038 "``llvm.``" are reserved for use by LLVM.
3042 The '``llvm.used``' Global Variable
3043 -----------------------------------
3045 The ``@llvm.used`` global is an array which has
3046 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3047 pointers to named global variables, functions and aliases which may optionally
3048 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3051 .. code-block:: llvm
3056 @llvm.used = appending global [2 x i8*] [
3058 i8* bitcast (i32* @Y to i8*)
3059 ], section "llvm.metadata"
3061 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3062 and linker are required to treat the symbol as if there is a reference to the
3063 symbol that it cannot see (which is why they have to be named). For example, if
3064 a variable has internal linkage and no references other than that from the
3065 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3066 references from inline asms and other things the compiler cannot "see", and
3067 corresponds to "``attribute((used))``" in GNU C.
3069 On some targets, the code generator must emit a directive to the
3070 assembler or object file to prevent the assembler and linker from
3071 molesting the symbol.
3073 .. _gv_llvmcompilerused:
3075 The '``llvm.compiler.used``' Global Variable
3076 --------------------------------------------
3078 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3079 directive, except that it only prevents the compiler from touching the
3080 symbol. On targets that support it, this allows an intelligent linker to
3081 optimize references to the symbol without being impeded as it would be
3084 This is a rare construct that should only be used in rare circumstances,
3085 and should not be exposed to source languages.
3087 .. _gv_llvmglobalctors:
3089 The '``llvm.global_ctors``' Global Variable
3090 -------------------------------------------
3092 .. code-block:: llvm
3094 %0 = type { i32, void ()* }
3095 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3097 The ``@llvm.global_ctors`` array contains a list of constructor
3098 functions and associated priorities. The functions referenced by this
3099 array will be called in ascending order of priority (i.e. lowest first)
3100 when the module is loaded. The order of functions with the same priority
3103 .. _llvmglobaldtors:
3105 The '``llvm.global_dtors``' Global Variable
3106 -------------------------------------------
3108 .. code-block:: llvm
3110 %0 = type { i32, void ()* }
3111 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3113 The ``@llvm.global_dtors`` array contains a list of destructor functions
3114 and associated priorities. The functions referenced by this array will
3115 be called in descending order of priority (i.e. highest first) when the
3116 module is loaded. The order of functions with the same priority is not
3119 Instruction Reference
3120 =====================
3122 The LLVM instruction set consists of several different classifications
3123 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3124 instructions <binaryops>`, :ref:`bitwise binary
3125 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3126 :ref:`other instructions <otherops>`.
3130 Terminator Instructions
3131 -----------------------
3133 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3134 program ends with a "Terminator" instruction, which indicates which
3135 block should be executed after the current block is finished. These
3136 terminator instructions typically yield a '``void``' value: they produce
3137 control flow, not values (the one exception being the
3138 ':ref:`invoke <i_invoke>`' instruction).
3140 The terminator instructions are: ':ref:`ret <i_ret>`',
3141 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3142 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3143 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3147 '``ret``' Instruction
3148 ^^^^^^^^^^^^^^^^^^^^^
3155 ret <type> <value> ; Return a value from a non-void function
3156 ret void ; Return from void function
3161 The '``ret``' instruction is used to return control flow (and optionally
3162 a value) from a function back to the caller.
3164 There are two forms of the '``ret``' instruction: one that returns a
3165 value and then causes control flow, and one that just causes control
3171 The '``ret``' instruction optionally accepts a single argument, the
3172 return value. The type of the return value must be a ':ref:`first
3173 class <t_firstclass>`' type.
3175 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3176 return type and contains a '``ret``' instruction with no return value or
3177 a return value with a type that does not match its type, or if it has a
3178 void return type and contains a '``ret``' instruction with a return
3184 When the '``ret``' instruction is executed, control flow returns back to
3185 the calling function's context. If the caller is a
3186 ":ref:`call <i_call>`" instruction, execution continues at the
3187 instruction after the call. If the caller was an
3188 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3189 beginning of the "normal" destination block. If the instruction returns
3190 a value, that value shall set the call or invoke instruction's return
3196 .. code-block:: llvm
3198 ret i32 5 ; Return an integer value of 5
3199 ret void ; Return from a void function
3200 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3204 '``br``' Instruction
3205 ^^^^^^^^^^^^^^^^^^^^
3212 br i1 <cond>, label <iftrue>, label <iffalse>
3213 br label <dest> ; Unconditional branch
3218 The '``br``' instruction is used to cause control flow to transfer to a
3219 different basic block in the current function. There are two forms of
3220 this instruction, corresponding to a conditional branch and an
3221 unconditional branch.
3226 The conditional branch form of the '``br``' instruction takes a single
3227 '``i1``' value and two '``label``' values. The unconditional form of the
3228 '``br``' instruction takes a single '``label``' value as a target.
3233 Upon execution of a conditional '``br``' instruction, the '``i1``'
3234 argument is evaluated. If the value is ``true``, control flows to the
3235 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3236 to the '``iffalse``' ``label`` argument.
3241 .. code-block:: llvm
3244 %cond = icmp eq i32 %a, %b
3245 br i1 %cond, label %IfEqual, label %IfUnequal
3253 '``switch``' Instruction
3254 ^^^^^^^^^^^^^^^^^^^^^^^^
3261 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3266 The '``switch``' instruction is used to transfer control flow to one of
3267 several different places. It is a generalization of the '``br``'
3268 instruction, allowing a branch to occur to one of many possible
3274 The '``switch``' instruction uses three parameters: an integer
3275 comparison value '``value``', a default '``label``' destination, and an
3276 array of pairs of comparison value constants and '``label``'s. The table
3277 is not allowed to contain duplicate constant entries.
3282 The ``switch`` instruction specifies a table of values and destinations.
3283 When the '``switch``' instruction is executed, this table is searched
3284 for the given value. If the value is found, control flow is transferred
3285 to the corresponding destination; otherwise, control flow is transferred
3286 to the default destination.
3291 Depending on properties of the target machine and the particular
3292 ``switch`` instruction, this instruction may be code generated in
3293 different ways. For example, it could be generated as a series of
3294 chained conditional branches or with a lookup table.
3299 .. code-block:: llvm
3301 ; Emulate a conditional br instruction
3302 %Val = zext i1 %value to i32
3303 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3305 ; Emulate an unconditional br instruction
3306 switch i32 0, label %dest [ ]
3308 ; Implement a jump table:
3309 switch i32 %val, label %otherwise [ i32 0, label %onzero
3311 i32 2, label %ontwo ]
3315 '``indirectbr``' Instruction
3316 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3323 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3328 The '``indirectbr``' instruction implements an indirect branch to a
3329 label within the current function, whose address is specified by
3330 "``address``". Address must be derived from a
3331 :ref:`blockaddress <blockaddress>` constant.
3336 The '``address``' argument is the address of the label to jump to. The
3337 rest of the arguments indicate the full set of possible destinations
3338 that the address may point to. Blocks are allowed to occur multiple
3339 times in the destination list, though this isn't particularly useful.
3341 This destination list is required so that dataflow analysis has an
3342 accurate understanding of the CFG.
3347 Control transfers to the block specified in the address argument. All
3348 possible destination blocks must be listed in the label list, otherwise
3349 this instruction has undefined behavior. This implies that jumps to
3350 labels defined in other functions have undefined behavior as well.
3355 This is typically implemented with a jump through a register.
3360 .. code-block:: llvm
3362 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3366 '``invoke``' Instruction
3367 ^^^^^^^^^^^^^^^^^^^^^^^^
3374 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3375 to label <normal label> unwind label <exception label>
3380 The '``invoke``' instruction causes control to transfer to a specified
3381 function, with the possibility of control flow transfer to either the
3382 '``normal``' label or the '``exception``' label. If the callee function
3383 returns with the "``ret``" instruction, control flow will return to the
3384 "normal" label. If the callee (or any indirect callees) returns via the
3385 ":ref:`resume <i_resume>`" instruction or other exception handling
3386 mechanism, control is interrupted and continued at the dynamically
3387 nearest "exception" label.
3389 The '``exception``' label is a `landing
3390 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3391 '``exception``' label is required to have the
3392 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3393 information about the behavior of the program after unwinding happens,
3394 as its first non-PHI instruction. The restrictions on the
3395 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3396 instruction, so that the important information contained within the
3397 "``landingpad``" instruction can't be lost through normal code motion.
3402 This instruction requires several arguments:
3404 #. The optional "cconv" marker indicates which :ref:`calling
3405 convention <callingconv>` the call should use. If none is
3406 specified, the call defaults to using C calling conventions.
3407 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3408 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3410 #. '``ptr to function ty``': shall be the signature of the pointer to
3411 function value being invoked. In most cases, this is a direct
3412 function invocation, but indirect ``invoke``'s are just as possible,
3413 branching off an arbitrary pointer to function value.
3414 #. '``function ptr val``': An LLVM value containing a pointer to a
3415 function to be invoked.
3416 #. '``function args``': argument list whose types match the function
3417 signature argument types and parameter attributes. All arguments must
3418 be of :ref:`first class <t_firstclass>` type. If the function signature
3419 indicates the function accepts a variable number of arguments, the
3420 extra arguments can be specified.
3421 #. '``normal label``': the label reached when the called function
3422 executes a '``ret``' instruction.
3423 #. '``exception label``': the label reached when a callee returns via
3424 the :ref:`resume <i_resume>` instruction or other exception handling
3426 #. The optional :ref:`function attributes <fnattrs>` list. Only
3427 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3428 attributes are valid here.
3433 This instruction is designed to operate as a standard '``call``'
3434 instruction in most regards. The primary difference is that it
3435 establishes an association with a label, which is used by the runtime
3436 library to unwind the stack.
3438 This instruction is used in languages with destructors to ensure that
3439 proper cleanup is performed in the case of either a ``longjmp`` or a
3440 thrown exception. Additionally, this is important for implementation of
3441 '``catch``' clauses in high-level languages that support them.
3443 For the purposes of the SSA form, the definition of the value returned
3444 by the '``invoke``' instruction is deemed to occur on the edge from the
3445 current block to the "normal" label. If the callee unwinds then no
3446 return value is available.
3451 .. code-block:: llvm
3453 %retval = invoke i32 @Test(i32 15) to label %Continue
3454 unwind label %TestCleanup ; {i32}:retval set
3455 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3456 unwind label %TestCleanup ; {i32}:retval set
3460 '``resume``' Instruction
3461 ^^^^^^^^^^^^^^^^^^^^^^^^
3468 resume <type> <value>
3473 The '``resume``' instruction is a terminator instruction that has no
3479 The '``resume``' instruction requires one argument, which must have the
3480 same type as the result of any '``landingpad``' instruction in the same
3486 The '``resume``' instruction resumes propagation of an existing
3487 (in-flight) exception whose unwinding was interrupted with a
3488 :ref:`landingpad <i_landingpad>` instruction.
3493 .. code-block:: llvm
3495 resume { i8*, i32 } %exn
3499 '``unreachable``' Instruction
3500 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3512 The '``unreachable``' instruction has no defined semantics. This
3513 instruction is used to inform the optimizer that a particular portion of
3514 the code is not reachable. This can be used to indicate that the code
3515 after a no-return function cannot be reached, and other facts.
3520 The '``unreachable``' instruction has no defined semantics.
3527 Binary operators are used to do most of the computation in a program.
3528 They require two operands of the same type, execute an operation on
3529 them, and produce a single value. The operands might represent multiple
3530 data, as is the case with the :ref:`vector <t_vector>` data type. The
3531 result value has the same type as its operands.
3533 There are several different binary operators:
3537 '``add``' Instruction
3538 ^^^^^^^^^^^^^^^^^^^^^
3545 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3546 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3547 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3548 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3553 The '``add``' instruction returns the sum of its two operands.
3558 The two arguments to the '``add``' instruction must be
3559 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3560 arguments must have identical types.
3565 The value produced is the integer sum of the two operands.
3567 If the sum has unsigned overflow, the result returned is the
3568 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3571 Because LLVM integers use a two's complement representation, this
3572 instruction is appropriate for both signed and unsigned integers.
3574 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3575 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3576 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3577 unsigned and/or signed overflow, respectively, occurs.
3582 .. code-block:: llvm
3584 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3588 '``fadd``' Instruction
3589 ^^^^^^^^^^^^^^^^^^^^^^
3596 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3601 The '``fadd``' instruction returns the sum of its two operands.
3606 The two arguments to the '``fadd``' instruction must be :ref:`floating
3607 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3608 Both arguments must have identical types.
3613 The value produced is the floating point sum of the two operands. This
3614 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3615 which are optimization hints to enable otherwise unsafe floating point
3621 .. code-block:: llvm
3623 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3625 '``sub``' Instruction
3626 ^^^^^^^^^^^^^^^^^^^^^
3633 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3634 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3635 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3636 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3641 The '``sub``' instruction returns the difference of its two operands.
3643 Note that the '``sub``' instruction is used to represent the '``neg``'
3644 instruction present in most other intermediate representations.
3649 The two arguments to the '``sub``' instruction must be
3650 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3651 arguments must have identical types.
3656 The value produced is the integer difference of the two operands.
3658 If the difference has unsigned overflow, the result returned is the
3659 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3662 Because LLVM integers use a two's complement representation, this
3663 instruction is appropriate for both signed and unsigned integers.
3665 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3666 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3667 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3668 unsigned and/or signed overflow, respectively, occurs.
3673 .. code-block:: llvm
3675 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3676 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3680 '``fsub``' Instruction
3681 ^^^^^^^^^^^^^^^^^^^^^^
3688 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3693 The '``fsub``' instruction returns the difference of its two operands.
3695 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3696 instruction present in most other intermediate representations.
3701 The two arguments to the '``fsub``' instruction must be :ref:`floating
3702 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3703 Both arguments must have identical types.
3708 The value produced is the floating point difference of the two operands.
3709 This instruction can also take any number of :ref:`fast-math
3710 flags <fastmath>`, which are optimization hints to enable otherwise
3711 unsafe floating point optimizations:
3716 .. code-block:: llvm
3718 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3719 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3721 '``mul``' Instruction
3722 ^^^^^^^^^^^^^^^^^^^^^
3729 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3730 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3731 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3732 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3737 The '``mul``' instruction returns the product of its two operands.
3742 The two arguments to the '``mul``' instruction must be
3743 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3744 arguments must have identical types.
3749 The value produced is the integer product of the two operands.
3751 If the result of the multiplication has unsigned overflow, the result
3752 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3753 bit width of the result.
3755 Because LLVM integers use a two's complement representation, and the
3756 result is the same width as the operands, this instruction returns the
3757 correct result for both signed and unsigned integers. If a full product
3758 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3759 sign-extended or zero-extended as appropriate to the width of the full
3762 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3763 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3764 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3765 unsigned and/or signed overflow, respectively, occurs.
3770 .. code-block:: llvm
3772 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3776 '``fmul``' Instruction
3777 ^^^^^^^^^^^^^^^^^^^^^^
3784 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3789 The '``fmul``' instruction returns the product of its two operands.
3794 The two arguments to the '``fmul``' instruction must be :ref:`floating
3795 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3796 Both arguments must have identical types.
3801 The value produced is the floating point product of the two operands.
3802 This instruction can also take any number of :ref:`fast-math
3803 flags <fastmath>`, which are optimization hints to enable otherwise
3804 unsafe floating point optimizations:
3809 .. code-block:: llvm
3811 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3813 '``udiv``' Instruction
3814 ^^^^^^^^^^^^^^^^^^^^^^
3821 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3822 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3827 The '``udiv``' instruction returns the quotient of its two operands.
3832 The two arguments to the '``udiv``' 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 unsigned integer quotient of the two operands.
3841 Note that unsigned integer division and signed integer division are
3842 distinct operations; for signed integer division, use '``sdiv``'.
3844 Division by zero leads to undefined behavior.
3846 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3847 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3848 such, "((a udiv exact b) mul b) == a").
3853 .. code-block:: llvm
3855 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3857 '``sdiv``' Instruction
3858 ^^^^^^^^^^^^^^^^^^^^^^
3865 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3866 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3871 The '``sdiv``' instruction returns the quotient of its two operands.
3876 The two arguments to the '``sdiv``' instruction must be
3877 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3878 arguments must have identical types.
3883 The value produced is the signed integer quotient of the two operands
3884 rounded towards zero.
3886 Note that signed integer division and unsigned integer division are
3887 distinct operations; for unsigned integer division, use '``udiv``'.
3889 Division by zero leads to undefined behavior. Overflow also leads to
3890 undefined behavior; this is a rare case, but can occur, for example, by
3891 doing a 32-bit division of -2147483648 by -1.
3893 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3894 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3899 .. code-block:: llvm
3901 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3905 '``fdiv``' Instruction
3906 ^^^^^^^^^^^^^^^^^^^^^^
3913 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3918 The '``fdiv``' instruction returns the quotient of its two operands.
3923 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3924 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3925 Both arguments must have identical types.
3930 The value produced is the floating point quotient of the two operands.
3931 This instruction can also take any number of :ref:`fast-math
3932 flags <fastmath>`, which are optimization hints to enable otherwise
3933 unsafe floating point optimizations:
3938 .. code-block:: llvm
3940 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3942 '``urem``' Instruction
3943 ^^^^^^^^^^^^^^^^^^^^^^
3950 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3955 The '``urem``' instruction returns the remainder from the unsigned
3956 division of its two arguments.
3961 The two arguments to the '``urem``' 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 unsigned integer *remainder* of a division.
3969 This instruction always performs an unsigned division to get the
3972 Note that unsigned integer remainder and signed integer remainder are
3973 distinct operations; for signed integer remainder, use '``srem``'.
3975 Taking the remainder of a division by zero leads to undefined behavior.
3980 .. code-block:: llvm
3982 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3984 '``srem``' Instruction
3985 ^^^^^^^^^^^^^^^^^^^^^^
3992 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3997 The '``srem``' instruction returns the remainder from the signed
3998 division of its two operands. This instruction can also take
3999 :ref:`vector <t_vector>` versions of the values in which case the elements
4005 The two arguments to the '``srem``' instruction must be
4006 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4007 arguments must have identical types.
4012 This instruction returns the *remainder* of a division (where the result
4013 is either zero or has the same sign as the dividend, ``op1``), not the
4014 *modulo* operator (where the result is either zero or has the same sign
4015 as the divisor, ``op2``) of a value. For more information about the
4016 difference, see `The Math
4017 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4018 table of how this is implemented in various languages, please see
4020 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4022 Note that signed integer remainder and unsigned integer remainder are
4023 distinct operations; for unsigned integer remainder, use '``urem``'.
4025 Taking the remainder of a division by zero leads to undefined behavior.
4026 Overflow also leads to undefined behavior; this is a rare case, but can
4027 occur, for example, by taking the remainder of a 32-bit division of
4028 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4029 rule lets srem be implemented using instructions that return both the
4030 result of the division and the remainder.)
4035 .. code-block:: llvm
4037 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4041 '``frem``' Instruction
4042 ^^^^^^^^^^^^^^^^^^^^^^
4049 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4054 The '``frem``' instruction returns the remainder from the division of
4060 The two arguments to the '``frem``' instruction must be :ref:`floating
4061 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4062 Both arguments must have identical types.
4067 This instruction returns the *remainder* of a division. The remainder
4068 has the same sign as the dividend. This instruction can also take any
4069 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4070 to enable otherwise unsafe floating point optimizations:
4075 .. code-block:: llvm
4077 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4081 Bitwise Binary Operations
4082 -------------------------
4084 Bitwise binary operators are used to do various forms of bit-twiddling
4085 in a program. They are generally very efficient instructions and can
4086 commonly be strength reduced from other instructions. They require two
4087 operands of the same type, execute an operation on them, and produce a
4088 single value. The resulting value is the same type as its operands.
4090 '``shl``' Instruction
4091 ^^^^^^^^^^^^^^^^^^^^^
4098 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4099 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4100 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4101 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4106 The '``shl``' instruction returns the first operand shifted to the left
4107 a specified number of bits.
4112 Both arguments to the '``shl``' instruction must be the same
4113 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4114 '``op2``' is treated as an unsigned value.
4119 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4120 where ``n`` is the width of the result. If ``op2`` is (statically or
4121 dynamically) negative or equal to or larger than the number of bits in
4122 ``op1``, the result is undefined. If the arguments are vectors, each
4123 vector element of ``op1`` is shifted by the corresponding shift amount
4126 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4127 value <poisonvalues>` if it shifts out any non-zero bits. If the
4128 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4129 value <poisonvalues>` if it shifts out any bits that disagree with the
4130 resultant sign bit. As such, NUW/NSW have the same semantics as they
4131 would if the shift were expressed as a mul instruction with the same
4132 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4137 .. code-block:: llvm
4139 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4140 <result> = shl i32 4, 2 ; yields {i32}: 16
4141 <result> = shl i32 1, 10 ; yields {i32}: 1024
4142 <result> = shl i32 1, 32 ; undefined
4143 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4145 '``lshr``' Instruction
4146 ^^^^^^^^^^^^^^^^^^^^^^
4153 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4154 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4159 The '``lshr``' instruction (logical shift right) returns the first
4160 operand shifted to the right a specified number of bits with zero fill.
4165 Both arguments to the '``lshr``' instruction must be the same
4166 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4167 '``op2``' is treated as an unsigned value.
4172 This instruction always performs a logical shift right operation. The
4173 most significant bits of the result will be filled with zero bits after
4174 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4175 than the number of bits in ``op1``, the result is undefined. If the
4176 arguments are vectors, each vector element of ``op1`` is shifted by the
4177 corresponding shift amount in ``op2``.
4179 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4180 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4186 .. code-block:: llvm
4188 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4189 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4190 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4191 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4192 <result> = lshr i32 1, 32 ; undefined
4193 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4195 '``ashr``' Instruction
4196 ^^^^^^^^^^^^^^^^^^^^^^
4203 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4204 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4209 The '``ashr``' instruction (arithmetic shift right) returns the first
4210 operand shifted to the right a specified number of bits with sign
4216 Both arguments to the '``ashr``' instruction must be the same
4217 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4218 '``op2``' is treated as an unsigned value.
4223 This instruction always performs an arithmetic shift right operation,
4224 The most significant bits of the result will be filled with the sign bit
4225 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4226 than the number of bits in ``op1``, the result is undefined. If the
4227 arguments are vectors, each vector element of ``op1`` is shifted by the
4228 corresponding shift amount in ``op2``.
4230 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4231 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4237 .. code-block:: llvm
4239 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4240 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4241 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4242 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4243 <result> = ashr i32 1, 32 ; undefined
4244 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4246 '``and``' Instruction
4247 ^^^^^^^^^^^^^^^^^^^^^
4254 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4259 The '``and``' instruction returns the bitwise logical and of its two
4265 The two arguments to the '``and``' instruction must be
4266 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4267 arguments must have identical types.
4272 The truth table used for the '``and``' instruction is:
4289 .. code-block:: llvm
4291 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4292 <result> = and i32 15, 40 ; yields {i32}:result = 8
4293 <result> = and i32 4, 8 ; yields {i32}:result = 0
4295 '``or``' Instruction
4296 ^^^^^^^^^^^^^^^^^^^^
4303 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4308 The '``or``' instruction returns the bitwise logical inclusive or of its
4314 The two arguments to the '``or``' instruction must be
4315 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4316 arguments must have identical types.
4321 The truth table used for the '``or``' instruction is:
4340 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4341 <result> = or i32 15, 40 ; yields {i32}:result = 47
4342 <result> = or i32 4, 8 ; yields {i32}:result = 12
4344 '``xor``' Instruction
4345 ^^^^^^^^^^^^^^^^^^^^^
4352 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4357 The '``xor``' instruction returns the bitwise logical exclusive or of
4358 its two operands. The ``xor`` is used to implement the "one's
4359 complement" operation, which is the "~" operator in C.
4364 The two arguments to the '``xor``' instruction must be
4365 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4366 arguments must have identical types.
4371 The truth table used for the '``xor``' instruction is:
4388 .. code-block:: llvm
4390 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4391 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4392 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4393 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4398 LLVM supports several instructions to represent vector operations in a
4399 target-independent manner. These instructions cover the element-access
4400 and vector-specific operations needed to process vectors effectively.
4401 While LLVM does directly support these vector operations, many
4402 sophisticated algorithms will want to use target-specific intrinsics to
4403 take full advantage of a specific target.
4405 .. _i_extractelement:
4407 '``extractelement``' Instruction
4408 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4415 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4420 The '``extractelement``' instruction extracts a single scalar element
4421 from a vector at a specified index.
4426 The first operand of an '``extractelement``' instruction is a value of
4427 :ref:`vector <t_vector>` type. The second operand is an index indicating
4428 the position from which to extract the element. The index may be a
4434 The result is a scalar of the same type as the element type of ``val``.
4435 Its value is the value at position ``idx`` of ``val``. If ``idx``
4436 exceeds the length of ``val``, the results are undefined.
4441 .. code-block:: llvm
4443 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4445 .. _i_insertelement:
4447 '``insertelement``' Instruction
4448 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4455 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4460 The '``insertelement``' instruction inserts a scalar element into a
4461 vector at a specified index.
4466 The first operand of an '``insertelement``' instruction is a value of
4467 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4468 type must equal the element type of the first operand. The third operand
4469 is an index indicating the position at which to insert the value. The
4470 index may be a variable.
4475 The result is a vector of the same type as ``val``. Its element values
4476 are those of ``val`` except at position ``idx``, where it gets the value
4477 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4483 .. code-block:: llvm
4485 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4487 .. _i_shufflevector:
4489 '``shufflevector``' Instruction
4490 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4497 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4502 The '``shufflevector``' instruction constructs a permutation of elements
4503 from two input vectors, returning a vector with the same element type as
4504 the input and length that is the same as the shuffle mask.
4509 The first two operands of a '``shufflevector``' instruction are vectors
4510 with the same type. The third argument is a shuffle mask whose element
4511 type is always 'i32'. The result of the instruction is a vector whose
4512 length is the same as the shuffle mask and whose element type is the
4513 same as the element type of the first two operands.
4515 The shuffle mask operand is required to be a constant vector with either
4516 constant integer or undef values.
4521 The elements of the two input vectors are numbered from left to right
4522 across both of the vectors. The shuffle mask operand specifies, for each
4523 element of the result vector, which element of the two input vectors the
4524 result element gets. The element selector may be undef (meaning "don't
4525 care") and the second operand may be undef if performing a shuffle from
4531 .. code-block:: llvm
4533 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4534 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4535 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4536 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4537 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4538 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4539 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4540 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4542 Aggregate Operations
4543 --------------------
4545 LLVM supports several instructions for working with
4546 :ref:`aggregate <t_aggregate>` values.
4550 '``extractvalue``' Instruction
4551 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4558 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4563 The '``extractvalue``' instruction extracts the value of a member field
4564 from an :ref:`aggregate <t_aggregate>` value.
4569 The first operand of an '``extractvalue``' instruction is a value of
4570 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4571 constant indices to specify which value to extract in a similar manner
4572 as indices in a '``getelementptr``' instruction.
4574 The major differences to ``getelementptr`` indexing are:
4576 - Since the value being indexed is not a pointer, the first index is
4577 omitted and assumed to be zero.
4578 - At least one index must be specified.
4579 - Not only struct indices but also array indices must be in bounds.
4584 The result is the value at the position in the aggregate specified by
4590 .. code-block:: llvm
4592 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4596 '``insertvalue``' Instruction
4597 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4604 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4609 The '``insertvalue``' instruction inserts a value into a member field in
4610 an :ref:`aggregate <t_aggregate>` value.
4615 The first operand of an '``insertvalue``' instruction is a value of
4616 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4617 a first-class value to insert. The following operands are constant
4618 indices indicating the position at which to insert the value in a
4619 similar manner as indices in a '``extractvalue``' instruction. The value
4620 to insert must have the same type as the value identified by the
4626 The result is an aggregate of the same type as ``val``. Its value is
4627 that of ``val`` except that the value at the position specified by the
4628 indices is that of ``elt``.
4633 .. code-block:: llvm
4635 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4636 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4637 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4641 Memory Access and Addressing Operations
4642 ---------------------------------------
4644 A key design point of an SSA-based representation is how it represents
4645 memory. In LLVM, no memory locations are in SSA form, which makes things
4646 very simple. This section describes how to read, write, and allocate
4651 '``alloca``' Instruction
4652 ^^^^^^^^^^^^^^^^^^^^^^^^
4659 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4664 The '``alloca``' instruction allocates memory on the stack frame of the
4665 currently executing function, to be automatically released when this
4666 function returns to its caller. The object is always allocated in the
4667 generic address space (address space zero).
4672 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4673 bytes of memory on the runtime stack, returning a pointer of the
4674 appropriate type to the program. If "NumElements" is specified, it is
4675 the number of elements allocated, otherwise "NumElements" is defaulted
4676 to be one. If a constant alignment is specified, the value result of the
4677 allocation is guaranteed to be aligned to at least that boundary. If not
4678 specified, or if zero, the target can choose to align the allocation on
4679 any convenient boundary compatible with the type.
4681 '``type``' may be any sized type.
4686 Memory is allocated; a pointer is returned. The operation is undefined
4687 if there is insufficient stack space for the allocation. '``alloca``'d
4688 memory is automatically released when the function returns. The
4689 '``alloca``' instruction is commonly used to represent automatic
4690 variables that must have an address available. When the function returns
4691 (either with the ``ret`` or ``resume`` instructions), the memory is
4692 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4693 The order in which memory is allocated (ie., which way the stack grows)
4699 .. code-block:: llvm
4701 %ptr = alloca i32 ; yields {i32*}:ptr
4702 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4703 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4704 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4708 '``load``' Instruction
4709 ^^^^^^^^^^^^^^^^^^^^^^
4716 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4717 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4718 !<index> = !{ i32 1 }
4723 The '``load``' instruction is used to read from memory.
4728 The argument to the ``load`` instruction specifies the memory address
4729 from which to load. The pointer must point to a :ref:`first
4730 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4731 then the optimizer is not allowed to modify the number or order of
4732 execution of this ``load`` with other :ref:`volatile
4733 operations <volatile>`.
4735 If the ``load`` is marked as ``atomic``, it takes an extra
4736 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4737 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4738 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4739 when they may see multiple atomic stores. The type of the pointee must
4740 be an integer type whose bit width is a power of two greater than or
4741 equal to eight and less than or equal to a target-specific size limit.
4742 ``align`` must be explicitly specified on atomic loads, and the load has
4743 undefined behavior if the alignment is not set to a value which is at
4744 least the size in bytes of the pointee. ``!nontemporal`` does not have
4745 any defined semantics for atomic loads.
4747 The optional constant ``align`` argument specifies the alignment of the
4748 operation (that is, the alignment of the memory address). A value of 0
4749 or an omitted ``align`` argument means that the operation has the ABI
4750 alignment for the target. It is the responsibility of the code emitter
4751 to ensure that the alignment information is correct. Overestimating the
4752 alignment results in undefined behavior. Underestimating the alignment
4753 may produce less efficient code. An alignment of 1 is always safe.
4755 The optional ``!nontemporal`` metadata must reference a single
4756 metadata name ``<index>`` corresponding to a metadata node with one
4757 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4758 metadata on the instruction tells the optimizer and code generator
4759 that this load is not expected to be reused in the cache. The code
4760 generator may select special instructions to save cache bandwidth, such
4761 as the ``MOVNT`` instruction on x86.
4763 The optional ``!invariant.load`` metadata must reference a single
4764 metadata name ``<index>`` corresponding to a metadata node with no
4765 entries. The existence of the ``!invariant.load`` metadata on the
4766 instruction tells the optimizer and code generator that this load
4767 address points to memory which does not change value during program
4768 execution. The optimizer may then move this load around, for example, by
4769 hoisting it out of loops using loop invariant code motion.
4774 The location of memory pointed to is loaded. If the value being loaded
4775 is of scalar type then the number of bytes read does not exceed the
4776 minimum number of bytes needed to hold all bits of the type. For
4777 example, loading an ``i24`` reads at most three bytes. When loading a
4778 value of a type like ``i20`` with a size that is not an integral number
4779 of bytes, the result is undefined if the value was not originally
4780 written using a store of the same type.
4785 .. code-block:: llvm
4787 %ptr = alloca i32 ; yields {i32*}:ptr
4788 store i32 3, i32* %ptr ; yields {void}
4789 %val = load i32* %ptr ; yields {i32}:val = i32 3
4793 '``store``' Instruction
4794 ^^^^^^^^^^^^^^^^^^^^^^^
4801 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4802 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4807 The '``store``' instruction is used to write to memory.
4812 There are two arguments to the ``store`` instruction: a value to store
4813 and an address at which to store it. The type of the ``<pointer>``
4814 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4815 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4816 then the optimizer is not allowed to modify the number or order of
4817 execution of this ``store`` with other :ref:`volatile
4818 operations <volatile>`.
4820 If the ``store`` is marked as ``atomic``, it takes an extra
4821 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4822 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4823 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4824 when they may see multiple atomic stores. The type of the pointee must
4825 be an integer type whose bit width is a power of two greater than or
4826 equal to eight and less than or equal to a target-specific size limit.
4827 ``align`` must be explicitly specified on atomic stores, and the store
4828 has undefined behavior if the alignment is not set to a value which is
4829 at least the size in bytes of the pointee. ``!nontemporal`` does not
4830 have any defined semantics for atomic stores.
4832 The optional constant ``align`` argument specifies the alignment of the
4833 operation (that is, the alignment of the memory address). A value of 0
4834 or an omitted ``align`` argument means that the operation has the ABI
4835 alignment for the target. It is the responsibility of the code emitter
4836 to ensure that the alignment information is correct. Overestimating the
4837 alignment results in undefined behavior. Underestimating the
4838 alignment may produce less efficient code. An alignment of 1 is always
4841 The optional ``!nontemporal`` metadata must reference a single metadata
4842 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4843 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4844 tells the optimizer and code generator that this load is not expected to
4845 be reused in the cache. The code generator may select special
4846 instructions to save cache bandwidth, such as the MOVNT instruction on
4852 The contents of memory are updated to contain ``<value>`` at the
4853 location specified by the ``<pointer>`` operand. If ``<value>`` is
4854 of scalar type then the number of bytes written does not exceed the
4855 minimum number of bytes needed to hold all bits of the type. For
4856 example, storing an ``i24`` writes at most three bytes. When writing a
4857 value of a type like ``i20`` with a size that is not an integral number
4858 of bytes, it is unspecified what happens to the extra bits that do not
4859 belong to the type, but they will typically be overwritten.
4864 .. code-block:: llvm
4866 %ptr = alloca i32 ; yields {i32*}:ptr
4867 store i32 3, i32* %ptr ; yields {void}
4868 %val = load i32* %ptr ; yields {i32}:val = i32 3
4872 '``fence``' Instruction
4873 ^^^^^^^^^^^^^^^^^^^^^^^
4880 fence [singlethread] <ordering> ; yields {void}
4885 The '``fence``' instruction is used to introduce happens-before edges
4891 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4892 defines what *synchronizes-with* edges they add. They can only be given
4893 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4898 A fence A which has (at least) ``release`` ordering semantics
4899 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4900 semantics if and only if there exist atomic operations X and Y, both
4901 operating on some atomic object M, such that A is sequenced before X, X
4902 modifies M (either directly or through some side effect of a sequence
4903 headed by X), Y is sequenced before B, and Y observes M. This provides a
4904 *happens-before* dependency between A and B. Rather than an explicit
4905 ``fence``, one (but not both) of the atomic operations X or Y might
4906 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4907 still *synchronize-with* the explicit ``fence`` and establish the
4908 *happens-before* edge.
4910 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4911 ``acquire`` and ``release`` semantics specified above, participates in
4912 the global program order of other ``seq_cst`` operations and/or fences.
4914 The optional ":ref:`singlethread <singlethread>`" argument specifies
4915 that the fence only synchronizes with other fences in the same thread.
4916 (This is useful for interacting with signal handlers.)
4921 .. code-block:: llvm
4923 fence acquire ; yields {void}
4924 fence singlethread seq_cst ; yields {void}
4928 '``cmpxchg``' Instruction
4929 ^^^^^^^^^^^^^^^^^^^^^^^^^
4936 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4941 The '``cmpxchg``' instruction is used to atomically modify memory. It
4942 loads a value in memory and compares it to a given value. If they are
4943 equal, it stores a new value into the memory.
4948 There are three arguments to the '``cmpxchg``' instruction: an address
4949 to operate on, a value to compare to the value currently be at that
4950 address, and a new value to place at that address if the compared values
4951 are equal. The type of '<cmp>' must be an integer type whose bit width
4952 is a power of two greater than or equal to eight and less than or equal
4953 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4954 type, and the type of '<pointer>' must be a pointer to that type. If the
4955 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4956 to modify the number or order of execution of this ``cmpxchg`` with
4957 other :ref:`volatile operations <volatile>`.
4959 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4960 synchronizes with other atomic operations.
4962 The optional "``singlethread``" argument declares that the ``cmpxchg``
4963 is only atomic with respect to code (usually signal handlers) running in
4964 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4965 respect to all other code in the system.
4967 The pointer passed into cmpxchg must have alignment greater than or
4968 equal to the size in memory of the operand.
4973 The contents of memory at the location specified by the '``<pointer>``'
4974 operand is read and compared to '``<cmp>``'; if the read value is the
4975 equal, '``<new>``' is written. The original value at the location is
4978 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4979 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4980 atomic load with an ordering parameter determined by dropping any
4981 ``release`` part of the ``cmpxchg``'s ordering.
4986 .. code-block:: llvm
4989 %orig = atomic load i32* %ptr unordered ; yields {i32}
4993 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4994 %squared = mul i32 %cmp, %cmp
4995 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4996 %success = icmp eq i32 %cmp, %old
4997 br i1 %success, label %done, label %loop
5004 '``atomicrmw``' Instruction
5005 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5012 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
5017 The '``atomicrmw``' instruction is used to atomically modify memory.
5022 There are three arguments to the '``atomicrmw``' instruction: an
5023 operation to apply, an address whose value to modify, an argument to the
5024 operation. The operation must be one of the following keywords:
5038 The type of '<value>' must be an integer type whose bit width is a power
5039 of two greater than or equal to eight and less than or equal to a
5040 target-specific size limit. The type of the '``<pointer>``' operand must
5041 be a pointer to that type. If the ``atomicrmw`` is marked as
5042 ``volatile``, then the optimizer is not allowed to modify the number or
5043 order of execution of this ``atomicrmw`` with other :ref:`volatile
5044 operations <volatile>`.
5049 The contents of memory at the location specified by the '``<pointer>``'
5050 operand are atomically read, modified, and written back. The original
5051 value at the location is returned. The modification is specified by the
5054 - xchg: ``*ptr = val``
5055 - add: ``*ptr = *ptr + val``
5056 - sub: ``*ptr = *ptr - val``
5057 - and: ``*ptr = *ptr & val``
5058 - nand: ``*ptr = ~(*ptr & val)``
5059 - or: ``*ptr = *ptr | val``
5060 - xor: ``*ptr = *ptr ^ val``
5061 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5062 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5063 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5065 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5071 .. code-block:: llvm
5073 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5075 .. _i_getelementptr:
5077 '``getelementptr``' Instruction
5078 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5085 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5086 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5087 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5092 The '``getelementptr``' instruction is used to get the address of a
5093 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5094 address calculation only and does not access memory.
5099 The first argument is always a pointer or a vector of pointers, and
5100 forms the basis of the calculation. The remaining arguments are indices
5101 that indicate which of the elements of the aggregate object are indexed.
5102 The interpretation of each index is dependent on the type being indexed
5103 into. The first index always indexes the pointer value given as the
5104 first argument, the second index indexes a value of the type pointed to
5105 (not necessarily the value directly pointed to, since the first index
5106 can be non-zero), etc. The first type indexed into must be a pointer
5107 value, subsequent types can be arrays, vectors, and structs. Note that
5108 subsequent types being indexed into can never be pointers, since that
5109 would require loading the pointer before continuing calculation.
5111 The type of each index argument depends on the type it is indexing into.
5112 When indexing into a (optionally packed) structure, only ``i32`` integer
5113 **constants** are allowed (when using a vector of indices they must all
5114 be the **same** ``i32`` integer constant). When indexing into an array,
5115 pointer or vector, integers of any width are allowed, and they are not
5116 required to be constant. These integers are treated as signed values
5119 For example, let's consider a C code fragment and how it gets compiled
5135 int *foo(struct ST *s) {
5136 return &s[1].Z.B[5][13];
5139 The LLVM code generated by Clang is:
5141 .. code-block:: llvm
5143 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5144 %struct.ST = type { i32, double, %struct.RT }
5146 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5148 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5155 In the example above, the first index is indexing into the
5156 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5157 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5158 indexes into the third element of the structure, yielding a
5159 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5160 structure. The third index indexes into the second element of the
5161 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5162 dimensions of the array are subscripted into, yielding an '``i32``'
5163 type. The '``getelementptr``' instruction returns a pointer to this
5164 element, thus computing a value of '``i32*``' type.
5166 Note that it is perfectly legal to index partially through a structure,
5167 returning a pointer to an inner element. Because of this, the LLVM code
5168 for the given testcase is equivalent to:
5170 .. code-block:: llvm
5172 define i32* @foo(%struct.ST* %s) {
5173 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5174 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5175 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5176 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5177 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5181 If the ``inbounds`` keyword is present, the result value of the
5182 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5183 pointer is not an *in bounds* address of an allocated object, or if any
5184 of the addresses that would be formed by successive addition of the
5185 offsets implied by the indices to the base address with infinitely
5186 precise signed arithmetic are not an *in bounds* address of that
5187 allocated object. The *in bounds* addresses for an allocated object are
5188 all the addresses that point into the object, plus the address one byte
5189 past the end. In cases where the base is a vector of pointers the
5190 ``inbounds`` keyword applies to each of the computations element-wise.
5192 If the ``inbounds`` keyword is not present, the offsets are added to the
5193 base address with silently-wrapping two's complement arithmetic. If the
5194 offsets have a different width from the pointer, they are sign-extended
5195 or truncated to the width of the pointer. The result value of the
5196 ``getelementptr`` may be outside the object pointed to by the base
5197 pointer. The result value may not necessarily be used to access memory
5198 though, even if it happens to point into allocated storage. See the
5199 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5202 The getelementptr instruction is often confusing. For some more insight
5203 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5208 .. code-block:: llvm
5210 ; yields [12 x i8]*:aptr
5211 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5213 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5215 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5217 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5219 In cases where the pointer argument is a vector of pointers, each index
5220 must be a vector with the same number of elements. For example:
5222 .. code-block:: llvm
5224 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5226 Conversion Operations
5227 ---------------------
5229 The instructions in this category are the conversion instructions
5230 (casting) which all take a single operand and a type. They perform
5231 various bit conversions on the operand.
5233 '``trunc .. to``' Instruction
5234 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5241 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5246 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5251 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5252 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5253 of the same number of integers. The bit size of the ``value`` must be
5254 larger than the bit size of the destination type, ``ty2``. Equal sized
5255 types are not allowed.
5260 The '``trunc``' instruction truncates the high order bits in ``value``
5261 and converts the remaining bits to ``ty2``. Since the source size must
5262 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5263 It will always truncate bits.
5268 .. code-block:: llvm
5270 %X = trunc i32 257 to i8 ; yields i8:1
5271 %Y = trunc i32 123 to i1 ; yields i1:true
5272 %Z = trunc i32 122 to i1 ; yields i1:false
5273 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5275 '``zext .. to``' Instruction
5276 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5283 <result> = zext <ty> <value> to <ty2> ; yields ty2
5288 The '``zext``' instruction zero extends its operand to type ``ty2``.
5293 The '``zext``' instruction takes a value to cast, and a type to cast it
5294 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5295 the same number of integers. The bit size of the ``value`` must be
5296 smaller than the bit size of the destination type, ``ty2``.
5301 The ``zext`` fills the high order bits of the ``value`` with zero bits
5302 until it reaches the size of the destination type, ``ty2``.
5304 When zero extending from i1, the result will always be either 0 or 1.
5309 .. code-block:: llvm
5311 %X = zext i32 257 to i64 ; yields i64:257
5312 %Y = zext i1 true to i32 ; yields i32:1
5313 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5315 '``sext .. to``' Instruction
5316 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5323 <result> = sext <ty> <value> to <ty2> ; yields ty2
5328 The '``sext``' sign extends ``value`` to the type ``ty2``.
5333 The '``sext``' instruction takes a value to cast, and a type to cast it
5334 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5335 the same number of integers. The bit size of the ``value`` must be
5336 smaller than the bit size of the destination type, ``ty2``.
5341 The '``sext``' instruction performs a sign extension by copying the sign
5342 bit (highest order bit) of the ``value`` until it reaches the bit size
5343 of the type ``ty2``.
5345 When sign extending from i1, the extension always results in -1 or 0.
5350 .. code-block:: llvm
5352 %X = sext i8 -1 to i16 ; yields i16 :65535
5353 %Y = sext i1 true to i32 ; yields i32:-1
5354 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5356 '``fptrunc .. to``' Instruction
5357 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5364 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5369 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5374 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5375 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5376 The size of ``value`` must be larger than the size of ``ty2``. This
5377 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5382 The '``fptrunc``' instruction truncates a ``value`` from a larger
5383 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5384 point <t_floating>` type. If the value cannot fit within the
5385 destination type, ``ty2``, then the results are undefined.
5390 .. code-block:: llvm
5392 %X = fptrunc double 123.0 to float ; yields float:123.0
5393 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5395 '``fpext .. to``' Instruction
5396 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5403 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5408 The '``fpext``' extends a floating point ``value`` to a larger floating
5414 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5415 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5416 to. The source type must be smaller than the destination type.
5421 The '``fpext``' instruction extends the ``value`` from a smaller
5422 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5423 point <t_floating>` type. The ``fpext`` cannot be used to make a
5424 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5425 *no-op cast* for a floating point cast.
5430 .. code-block:: llvm
5432 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5433 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5435 '``fptoui .. to``' Instruction
5436 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5443 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5448 The '``fptoui``' converts a floating point ``value`` to its unsigned
5449 integer equivalent of type ``ty2``.
5454 The '``fptoui``' instruction takes a value to cast, which must be a
5455 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5456 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5457 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5458 type with the same number of elements as ``ty``
5463 The '``fptoui``' instruction converts its :ref:`floating
5464 point <t_floating>` operand into the nearest (rounding towards zero)
5465 unsigned integer value. If the value cannot fit in ``ty2``, the results
5471 .. code-block:: llvm
5473 %X = fptoui double 123.0 to i32 ; yields i32:123
5474 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5475 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5477 '``fptosi .. to``' Instruction
5478 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5485 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5490 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5491 ``value`` to type ``ty2``.
5496 The '``fptosi``' instruction takes a value to cast, which must be a
5497 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5498 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5499 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5500 type with the same number of elements as ``ty``
5505 The '``fptosi``' instruction converts its :ref:`floating
5506 point <t_floating>` operand into the nearest (rounding towards zero)
5507 signed integer value. If the value cannot fit in ``ty2``, the results
5513 .. code-block:: llvm
5515 %X = fptosi double -123.0 to i32 ; yields i32:-123
5516 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5517 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5519 '``uitofp .. to``' Instruction
5520 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5527 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5532 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5533 and converts that value to the ``ty2`` type.
5538 The '``uitofp``' instruction takes a value to cast, which must be a
5539 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5540 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5541 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5542 type with the same number of elements as ``ty``
5547 The '``uitofp``' instruction interprets its operand as an unsigned
5548 integer quantity and converts it to the corresponding floating point
5549 value. If the value cannot fit in the floating point value, the results
5555 .. code-block:: llvm
5557 %X = uitofp i32 257 to float ; yields float:257.0
5558 %Y = uitofp i8 -1 to double ; yields double:255.0
5560 '``sitofp .. to``' Instruction
5561 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5568 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5573 The '``sitofp``' instruction regards ``value`` as a signed integer and
5574 converts that value to the ``ty2`` type.
5579 The '``sitofp``' instruction takes a value to cast, which must be a
5580 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5581 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5582 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5583 type with the same number of elements as ``ty``
5588 The '``sitofp``' instruction interprets its operand as a signed integer
5589 quantity and converts it to the corresponding floating point value. If
5590 the value cannot fit in the floating point value, the results are
5596 .. code-block:: llvm
5598 %X = sitofp i32 257 to float ; yields float:257.0
5599 %Y = sitofp i8 -1 to double ; yields double:-1.0
5603 '``ptrtoint .. to``' Instruction
5604 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5611 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5616 The '``ptrtoint``' instruction converts the pointer or a vector of
5617 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5622 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5623 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5624 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5625 a vector of integers type.
5630 The '``ptrtoint``' instruction converts ``value`` to integer type
5631 ``ty2`` by interpreting the pointer value as an integer and either
5632 truncating or zero extending that value to the size of the integer type.
5633 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5634 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5635 the same size, then nothing is done (*no-op cast*) other than a type
5641 .. code-block:: llvm
5643 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5644 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5645 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5649 '``inttoptr .. to``' Instruction
5650 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5657 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5662 The '``inttoptr``' instruction converts an integer ``value`` to a
5663 pointer type, ``ty2``.
5668 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5669 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5675 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5676 applying either a zero extension or a truncation depending on the size
5677 of the integer ``value``. If ``value`` is larger than the size of a
5678 pointer then a truncation is done. If ``value`` is smaller than the size
5679 of a pointer then a zero extension is done. If they are the same size,
5680 nothing is done (*no-op cast*).
5685 .. code-block:: llvm
5687 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5688 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5689 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5690 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5694 '``bitcast .. to``' Instruction
5695 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5702 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5707 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5713 The '``bitcast``' instruction takes a value to cast, which must be a
5714 non-aggregate first class value, and a type to cast it to, which must
5715 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5716 bit sizes of ``value`` and the destination type, ``ty2``, must be
5717 identical. If the source type is a pointer, the destination type must
5718 also be a pointer of the same size. This instruction supports bitwise
5719 conversion of vectors to integers and to vectors of other types (as
5720 long as they have the same size).
5725 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5726 is always a *no-op cast* because no bits change with this
5727 conversion. The conversion is done as if the ``value`` had been stored
5728 to memory and read back as type ``ty2``. Pointer (or vector of
5729 pointers) types may only be converted to other pointer (or vector of
5730 pointers) types with the same address space through this instruction.
5731 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5732 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5737 .. code-block:: llvm
5739 %X = bitcast i8 255 to i8 ; yields i8 :-1
5740 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5741 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5742 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5744 .. _i_addrspacecast:
5746 '``addrspacecast .. to``' Instruction
5747 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5754 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5759 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5760 address space ``n`` to type ``pty2`` in address space ``m``.
5765 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5766 to cast and a pointer type to cast it to, which must have a different
5772 The '``addrspacecast``' instruction converts the pointer value
5773 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5774 value modification, depending on the target and the address space
5775 pair. Pointer conversions within the same address space must be
5776 performed with the ``bitcast`` instruction. Note that if the address space
5777 conversion is legal then both result and operand refer to the same memory
5783 .. code-block:: llvm
5785 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5786 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5787 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5794 The instructions in this category are the "miscellaneous" instructions,
5795 which defy better classification.
5799 '``icmp``' Instruction
5800 ^^^^^^^^^^^^^^^^^^^^^^
5807 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5812 The '``icmp``' instruction returns a boolean value or a vector of
5813 boolean values based on comparison of its two integer, integer vector,
5814 pointer, or pointer vector operands.
5819 The '``icmp``' instruction takes three operands. The first operand is
5820 the condition code indicating the kind of comparison to perform. It is
5821 not a value, just a keyword. The possible condition code are:
5824 #. ``ne``: not equal
5825 #. ``ugt``: unsigned greater than
5826 #. ``uge``: unsigned greater or equal
5827 #. ``ult``: unsigned less than
5828 #. ``ule``: unsigned less or equal
5829 #. ``sgt``: signed greater than
5830 #. ``sge``: signed greater or equal
5831 #. ``slt``: signed less than
5832 #. ``sle``: signed less or equal
5834 The remaining two arguments must be :ref:`integer <t_integer>` or
5835 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5836 must also be identical types.
5841 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5842 code given as ``cond``. The comparison performed always yields either an
5843 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5845 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5846 otherwise. No sign interpretation is necessary or performed.
5847 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5848 otherwise. No sign interpretation is necessary or performed.
5849 #. ``ugt``: interprets the operands as unsigned values and yields
5850 ``true`` if ``op1`` is greater than ``op2``.
5851 #. ``uge``: interprets the operands as unsigned values and yields
5852 ``true`` if ``op1`` is greater than or equal to ``op2``.
5853 #. ``ult``: interprets the operands as unsigned values and yields
5854 ``true`` if ``op1`` is less than ``op2``.
5855 #. ``ule``: interprets the operands as unsigned values and yields
5856 ``true`` if ``op1`` is less than or equal to ``op2``.
5857 #. ``sgt``: interprets the operands as signed values and yields ``true``
5858 if ``op1`` is greater than ``op2``.
5859 #. ``sge``: interprets the operands as signed values and yields ``true``
5860 if ``op1`` is greater than or equal to ``op2``.
5861 #. ``slt``: interprets the operands as signed values and yields ``true``
5862 if ``op1`` is less than ``op2``.
5863 #. ``sle``: interprets the operands as signed values and yields ``true``
5864 if ``op1`` is less than or equal to ``op2``.
5866 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5867 are compared as if they were integers.
5869 If the operands are integer vectors, then they are compared element by
5870 element. The result is an ``i1`` vector with the same number of elements
5871 as the values being compared. Otherwise, the result is an ``i1``.
5876 .. code-block:: llvm
5878 <result> = icmp eq i32 4, 5 ; yields: result=false
5879 <result> = icmp ne float* %X, %X ; yields: result=false
5880 <result> = icmp ult i16 4, 5 ; yields: result=true
5881 <result> = icmp sgt i16 4, 5 ; yields: result=false
5882 <result> = icmp ule i16 -4, 5 ; yields: result=false
5883 <result> = icmp sge i16 4, 5 ; yields: result=false
5885 Note that the code generator does not yet support vector types with the
5886 ``icmp`` instruction.
5890 '``fcmp``' Instruction
5891 ^^^^^^^^^^^^^^^^^^^^^^
5898 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5903 The '``fcmp``' instruction returns a boolean value or vector of boolean
5904 values based on comparison of its operands.
5906 If the operands are floating point scalars, then the result type is a
5907 boolean (:ref:`i1 <t_integer>`).
5909 If the operands are floating point vectors, then the result type is a
5910 vector of boolean with the same number of elements as the operands being
5916 The '``fcmp``' instruction takes three operands. The first operand is
5917 the condition code indicating the kind of comparison to perform. It is
5918 not a value, just a keyword. The possible condition code are:
5920 #. ``false``: no comparison, always returns false
5921 #. ``oeq``: ordered and equal
5922 #. ``ogt``: ordered and greater than
5923 #. ``oge``: ordered and greater than or equal
5924 #. ``olt``: ordered and less than
5925 #. ``ole``: ordered and less than or equal
5926 #. ``one``: ordered and not equal
5927 #. ``ord``: ordered (no nans)
5928 #. ``ueq``: unordered or equal
5929 #. ``ugt``: unordered or greater than
5930 #. ``uge``: unordered or greater than or equal
5931 #. ``ult``: unordered or less than
5932 #. ``ule``: unordered or less than or equal
5933 #. ``une``: unordered or not equal
5934 #. ``uno``: unordered (either nans)
5935 #. ``true``: no comparison, always returns true
5937 *Ordered* means that neither operand is a QNAN while *unordered* means
5938 that either operand may be a QNAN.
5940 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5941 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5942 type. They must have identical types.
5947 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5948 condition code given as ``cond``. If the operands are vectors, then the
5949 vectors are compared element by element. Each comparison performed
5950 always yields an :ref:`i1 <t_integer>` result, as follows:
5952 #. ``false``: always yields ``false``, regardless of operands.
5953 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5954 is equal to ``op2``.
5955 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5956 is greater than ``op2``.
5957 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5958 is greater than or equal to ``op2``.
5959 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5960 is less than ``op2``.
5961 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5962 is less than or equal to ``op2``.
5963 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5964 is not equal to ``op2``.
5965 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5966 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5968 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5969 greater than ``op2``.
5970 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5971 greater than or equal to ``op2``.
5972 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5974 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5975 less than or equal to ``op2``.
5976 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5977 not equal to ``op2``.
5978 #. ``uno``: yields ``true`` if either operand is a QNAN.
5979 #. ``true``: always yields ``true``, regardless of operands.
5984 .. code-block:: llvm
5986 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5987 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5988 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5989 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5991 Note that the code generator does not yet support vector types with the
5992 ``fcmp`` instruction.
5996 '``phi``' Instruction
5997 ^^^^^^^^^^^^^^^^^^^^^
6004 <result> = phi <ty> [ <val0>, <label0>], ...
6009 The '``phi``' instruction is used to implement the φ node in the SSA
6010 graph representing the function.
6015 The type of the incoming values is specified with the first type field.
6016 After this, the '``phi``' instruction takes a list of pairs as
6017 arguments, with one pair for each predecessor basic block of the current
6018 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6019 the value arguments to the PHI node. Only labels may be used as the
6022 There must be no non-phi instructions between the start of a basic block
6023 and the PHI instructions: i.e. PHI instructions must be first in a basic
6026 For the purposes of the SSA form, the use of each incoming value is
6027 deemed to occur on the edge from the corresponding predecessor block to
6028 the current block (but after any definition of an '``invoke``'
6029 instruction's return value on the same edge).
6034 At runtime, the '``phi``' instruction logically takes on the value
6035 specified by the pair corresponding to the predecessor basic block that
6036 executed just prior to the current block.
6041 .. code-block:: llvm
6043 Loop: ; Infinite loop that counts from 0 on up...
6044 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6045 %nextindvar = add i32 %indvar, 1
6050 '``select``' Instruction
6051 ^^^^^^^^^^^^^^^^^^^^^^^^
6058 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6060 selty is either i1 or {<N x i1>}
6065 The '``select``' instruction is used to choose one value based on a
6066 condition, without branching.
6071 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6072 values indicating the condition, and two values of the same :ref:`first
6073 class <t_firstclass>` type. If the val1/val2 are vectors and the
6074 condition is a scalar, then entire vectors are selected, not individual
6080 If the condition is an i1 and it evaluates to 1, the instruction returns
6081 the first value argument; otherwise, it returns the second value
6084 If the condition is a vector of i1, then the value arguments must be
6085 vectors of the same size, and the selection is done element by element.
6090 .. code-block:: llvm
6092 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6096 '``call``' Instruction
6097 ^^^^^^^^^^^^^^^^^^^^^^
6104 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6109 The '``call``' instruction represents a simple function call.
6114 This instruction requires several arguments:
6116 #. The optional "tail" marker indicates that the callee function does
6117 not access any allocas or varargs in the caller. Note that calls may
6118 be marked "tail" even if they do not occur before a
6119 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6120 function call is eligible for tail call optimization, but `might not
6121 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6122 The code generator may optimize calls marked "tail" with either 1)
6123 automatic `sibling call
6124 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6125 callee have matching signatures, or 2) forced tail call optimization
6126 when the following extra requirements are met:
6128 - Caller and callee both have the calling convention ``fastcc``.
6129 - The call is in tail position (ret immediately follows call and ret
6130 uses value of call or is void).
6131 - Option ``-tailcallopt`` is enabled, or
6132 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6133 - `Platform specific constraints are
6134 met. <CodeGenerator.html#tailcallopt>`_
6136 #. The optional "cconv" marker indicates which :ref:`calling
6137 convention <callingconv>` the call should use. If none is
6138 specified, the call defaults to using C calling conventions. The
6139 calling convention of the call must match the calling convention of
6140 the target function, or else the behavior is undefined.
6141 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6142 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6144 #. '``ty``': the type of the call instruction itself which is also the
6145 type of the return value. Functions that return no value are marked
6147 #. '``fnty``': shall be the signature of the pointer to function value
6148 being invoked. The argument types must match the types implied by
6149 this signature. This type can be omitted if the function is not
6150 varargs and if the function type does not return a pointer to a
6152 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6153 be invoked. In most cases, this is a direct function invocation, but
6154 indirect ``call``'s are just as possible, calling an arbitrary pointer
6156 #. '``function args``': argument list whose types match the function
6157 signature argument types and parameter attributes. All arguments must
6158 be of :ref:`first class <t_firstclass>` type. If the function signature
6159 indicates the function accepts a variable number of arguments, the
6160 extra arguments can be specified.
6161 #. The optional :ref:`function attributes <fnattrs>` list. Only
6162 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6163 attributes are valid here.
6168 The '``call``' instruction is used to cause control flow to transfer to
6169 a specified function, with its incoming arguments bound to the specified
6170 values. Upon a '``ret``' instruction in the called function, control
6171 flow continues with the instruction after the function call, and the
6172 return value of the function is bound to the result argument.
6177 .. code-block:: llvm
6179 %retval = call i32 @test(i32 %argc)
6180 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6181 %X = tail call i32 @foo() ; yields i32
6182 %Y = tail call fastcc i32 @foo() ; yields i32
6183 call void %foo(i8 97 signext)
6185 %struct.A = type { i32, i8 }
6186 %r = call %struct.A @foo() ; yields { 32, i8 }
6187 %gr = extractvalue %struct.A %r, 0 ; yields i32
6188 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6189 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6190 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6192 llvm treats calls to some functions with names and arguments that match
6193 the standard C99 library as being the C99 library functions, and may
6194 perform optimizations or generate code for them under that assumption.
6195 This is something we'd like to change in the future to provide better
6196 support for freestanding environments and non-C-based languages.
6200 '``va_arg``' Instruction
6201 ^^^^^^^^^^^^^^^^^^^^^^^^
6208 <resultval> = va_arg <va_list*> <arglist>, <argty>
6213 The '``va_arg``' instruction is used to access arguments passed through
6214 the "variable argument" area of a function call. It is used to implement
6215 the ``va_arg`` macro in C.
6220 This instruction takes a ``va_list*`` value and the type of the
6221 argument. It returns a value of the specified argument type and
6222 increments the ``va_list`` to point to the next argument. The actual
6223 type of ``va_list`` is target specific.
6228 The '``va_arg``' instruction loads an argument of the specified type
6229 from the specified ``va_list`` and causes the ``va_list`` to point to
6230 the next argument. For more information, see the variable argument
6231 handling :ref:`Intrinsic Functions <int_varargs>`.
6233 It is legal for this instruction to be called in a function which does
6234 not take a variable number of arguments, for example, the ``vfprintf``
6237 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6238 function <intrinsics>` because it takes a type as an argument.
6243 See the :ref:`variable argument processing <int_varargs>` section.
6245 Note that the code generator does not yet fully support va\_arg on many
6246 targets. Also, it does not currently support va\_arg with aggregate
6247 types on any target.
6251 '``landingpad``' Instruction
6252 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6259 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6260 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6262 <clause> := catch <type> <value>
6263 <clause> := filter <array constant type> <array constant>
6268 The '``landingpad``' instruction is used by `LLVM's exception handling
6269 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6270 is a landing pad --- one where the exception lands, and corresponds to the
6271 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6272 defines values supplied by the personality function (``pers_fn``) upon
6273 re-entry to the function. The ``resultval`` has the type ``resultty``.
6278 This instruction takes a ``pers_fn`` value. This is the personality
6279 function associated with the unwinding mechanism. The optional
6280 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6282 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6283 contains the global variable representing the "type" that may be caught
6284 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6285 clause takes an array constant as its argument. Use
6286 "``[0 x i8**] undef``" for a filter which cannot throw. The
6287 '``landingpad``' instruction must contain *at least* one ``clause`` or
6288 the ``cleanup`` flag.
6293 The '``landingpad``' instruction defines the values which are set by the
6294 personality function (``pers_fn``) upon re-entry to the function, and
6295 therefore the "result type" of the ``landingpad`` instruction. As with
6296 calling conventions, how the personality function results are
6297 represented in LLVM IR is target specific.
6299 The clauses are applied in order from top to bottom. If two
6300 ``landingpad`` instructions are merged together through inlining, the
6301 clauses from the calling function are appended to the list of clauses.
6302 When the call stack is being unwound due to an exception being thrown,
6303 the exception is compared against each ``clause`` in turn. If it doesn't
6304 match any of the clauses, and the ``cleanup`` flag is not set, then
6305 unwinding continues further up the call stack.
6307 The ``landingpad`` instruction has several restrictions:
6309 - A landing pad block is a basic block which is the unwind destination
6310 of an '``invoke``' instruction.
6311 - A landing pad block must have a '``landingpad``' instruction as its
6312 first non-PHI instruction.
6313 - There can be only one '``landingpad``' instruction within the landing
6315 - A basic block that is not a landing pad block may not include a
6316 '``landingpad``' instruction.
6317 - All '``landingpad``' instructions in a function must have the same
6318 personality function.
6323 .. code-block:: llvm
6325 ;; A landing pad which can catch an integer.
6326 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6328 ;; A landing pad that is a cleanup.
6329 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6331 ;; A landing pad which can catch an integer and can only throw a double.
6332 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6334 filter [1 x i8**] [@_ZTId]
6341 LLVM supports the notion of an "intrinsic function". These functions
6342 have well known names and semantics and are required to follow certain
6343 restrictions. Overall, these intrinsics represent an extension mechanism
6344 for the LLVM language that does not require changing all of the
6345 transformations in LLVM when adding to the language (or the bitcode
6346 reader/writer, the parser, etc...).
6348 Intrinsic function names must all start with an "``llvm.``" prefix. This
6349 prefix is reserved in LLVM for intrinsic names; thus, function names may
6350 not begin with this prefix. Intrinsic functions must always be external
6351 functions: you cannot define the body of intrinsic functions. Intrinsic
6352 functions may only be used in call or invoke instructions: it is illegal
6353 to take the address of an intrinsic function. Additionally, because
6354 intrinsic functions are part of the LLVM language, it is required if any
6355 are added that they be documented here.
6357 Some intrinsic functions can be overloaded, i.e., the intrinsic
6358 represents a family of functions that perform the same operation but on
6359 different data types. Because LLVM can represent over 8 million
6360 different integer types, overloading is used commonly to allow an
6361 intrinsic function to operate on any integer type. One or more of the
6362 argument types or the result type can be overloaded to accept any
6363 integer type. Argument types may also be defined as exactly matching a
6364 previous argument's type or the result type. This allows an intrinsic
6365 function which accepts multiple arguments, but needs all of them to be
6366 of the same type, to only be overloaded with respect to a single
6367 argument or the result.
6369 Overloaded intrinsics will have the names of its overloaded argument
6370 types encoded into its function name, each preceded by a period. Only
6371 those types which are overloaded result in a name suffix. Arguments
6372 whose type is matched against another type do not. For example, the
6373 ``llvm.ctpop`` function can take an integer of any width and returns an
6374 integer of exactly the same integer width. This leads to a family of
6375 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6376 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6377 overloaded, and only one type suffix is required. Because the argument's
6378 type is matched against the return type, it does not require its own
6381 To learn how to add an intrinsic function, please see the `Extending
6382 LLVM Guide <ExtendingLLVM.html>`_.
6386 Variable Argument Handling Intrinsics
6387 -------------------------------------
6389 Variable argument support is defined in LLVM with the
6390 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6391 functions. These functions are related to the similarly named macros
6392 defined in the ``<stdarg.h>`` header file.
6394 All of these functions operate on arguments that use a target-specific
6395 value type "``va_list``". The LLVM assembly language reference manual
6396 does not define what this type is, so all transformations should be
6397 prepared to handle these functions regardless of the type used.
6399 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6400 variable argument handling intrinsic functions are used.
6402 .. code-block:: llvm
6404 define i32 @test(i32 %X, ...) {
6405 ; Initialize variable argument processing
6407 %ap2 = bitcast i8** %ap to i8*
6408 call void @llvm.va_start(i8* %ap2)
6410 ; Read a single integer argument
6411 %tmp = va_arg i8** %ap, i32
6413 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6415 %aq2 = bitcast i8** %aq to i8*
6416 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6417 call void @llvm.va_end(i8* %aq2)
6419 ; Stop processing of arguments.
6420 call void @llvm.va_end(i8* %ap2)
6424 declare void @llvm.va_start(i8*)
6425 declare void @llvm.va_copy(i8*, i8*)
6426 declare void @llvm.va_end(i8*)
6430 '``llvm.va_start``' Intrinsic
6431 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6438 declare void @llvm.va_start(i8* <arglist>)
6443 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6444 subsequent use by ``va_arg``.
6449 The argument is a pointer to a ``va_list`` element to initialize.
6454 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6455 available in C. In a target-dependent way, it initializes the
6456 ``va_list`` element to which the argument points, so that the next call
6457 to ``va_arg`` will produce the first variable argument passed to the
6458 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6459 to know the last argument of the function as the compiler can figure
6462 '``llvm.va_end``' Intrinsic
6463 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6470 declare void @llvm.va_end(i8* <arglist>)
6475 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6476 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6481 The argument is a pointer to a ``va_list`` to destroy.
6486 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6487 available in C. In a target-dependent way, it destroys the ``va_list``
6488 element to which the argument points. Calls to
6489 :ref:`llvm.va_start <int_va_start>` and
6490 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6495 '``llvm.va_copy``' Intrinsic
6496 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6503 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6508 The '``llvm.va_copy``' intrinsic copies the current argument position
6509 from the source argument list to the destination argument list.
6514 The first argument is a pointer to a ``va_list`` element to initialize.
6515 The second argument is a pointer to a ``va_list`` element to copy from.
6520 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6521 available in C. In a target-dependent way, it copies the source
6522 ``va_list`` element into the destination ``va_list`` element. This
6523 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6524 arbitrarily complex and require, for example, memory allocation.
6526 Accurate Garbage Collection Intrinsics
6527 --------------------------------------
6529 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6530 (GC) requires the implementation and generation of these intrinsics.
6531 These intrinsics allow identification of :ref:`GC roots on the
6532 stack <int_gcroot>`, as well as garbage collector implementations that
6533 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6534 Front-ends for type-safe garbage collected languages should generate
6535 these intrinsics to make use of the LLVM garbage collectors. For more
6536 details, see `Accurate Garbage Collection with
6537 LLVM <GarbageCollection.html>`_.
6539 The garbage collection intrinsics only operate on objects in the generic
6540 address space (address space zero).
6544 '``llvm.gcroot``' Intrinsic
6545 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6552 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6557 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6558 the code generator, and allows some metadata to be associated with it.
6563 The first argument specifies the address of a stack object that contains
6564 the root pointer. The second pointer (which must be either a constant or
6565 a global value address) contains the meta-data to be associated with the
6571 At runtime, a call to this intrinsic stores a null pointer into the
6572 "ptrloc" location. At compile-time, the code generator generates
6573 information to allow the runtime to find the pointer at GC safe points.
6574 The '``llvm.gcroot``' intrinsic may only be used in a function which
6575 :ref:`specifies a GC algorithm <gc>`.
6579 '``llvm.gcread``' Intrinsic
6580 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6587 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6592 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6593 locations, allowing garbage collector implementations that require read
6599 The second argument is the address to read from, which should be an
6600 address allocated from the garbage collector. The first object is a
6601 pointer to the start of the referenced object, if needed by the language
6602 runtime (otherwise null).
6607 The '``llvm.gcread``' intrinsic has the same semantics as a load
6608 instruction, but may be replaced with substantially more complex code by
6609 the garbage collector runtime, as needed. The '``llvm.gcread``'
6610 intrinsic may only be used in a function which :ref:`specifies a GC
6615 '``llvm.gcwrite``' Intrinsic
6616 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6623 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6628 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6629 locations, allowing garbage collector implementations that require write
6630 barriers (such as generational or reference counting collectors).
6635 The first argument is the reference to store, the second is the start of
6636 the object to store it to, and the third is the address of the field of
6637 Obj to store to. If the runtime does not require a pointer to the
6638 object, Obj may be null.
6643 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6644 instruction, but may be replaced with substantially more complex code by
6645 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6646 intrinsic may only be used in a function which :ref:`specifies a GC
6649 Code Generator Intrinsics
6650 -------------------------
6652 These intrinsics are provided by LLVM to expose special features that
6653 may only be implemented with code generator support.
6655 '``llvm.returnaddress``' Intrinsic
6656 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6663 declare i8 *@llvm.returnaddress(i32 <level>)
6668 The '``llvm.returnaddress``' intrinsic attempts to compute a
6669 target-specific value indicating the return address of the current
6670 function or one of its callers.
6675 The argument to this intrinsic indicates which function to return the
6676 address for. Zero indicates the calling function, one indicates its
6677 caller, etc. The argument is **required** to be a constant integer
6683 The '``llvm.returnaddress``' intrinsic either returns a pointer
6684 indicating the return address of the specified call frame, or zero if it
6685 cannot be identified. The value returned by this intrinsic is likely to
6686 be incorrect or 0 for arguments other than zero, so it should only be
6687 used for debugging purposes.
6689 Note that calling this intrinsic does not prevent function inlining or
6690 other aggressive transformations, so the value returned may not be that
6691 of the obvious source-language caller.
6693 '``llvm.frameaddress``' Intrinsic
6694 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6701 declare i8* @llvm.frameaddress(i32 <level>)
6706 The '``llvm.frameaddress``' intrinsic attempts to return the
6707 target-specific frame pointer value for the specified stack frame.
6712 The argument to this intrinsic indicates which function to return the
6713 frame pointer for. Zero indicates the calling function, one indicates
6714 its caller, etc. The argument is **required** to be a constant integer
6720 The '``llvm.frameaddress``' intrinsic either returns a pointer
6721 indicating the frame address of the specified call frame, or zero if it
6722 cannot be identified. The value returned by this intrinsic is likely to
6723 be incorrect or 0 for arguments other than zero, so it should only be
6724 used for debugging purposes.
6726 Note that calling this intrinsic does not prevent function inlining or
6727 other aggressive transformations, so the value returned may not be that
6728 of the obvious source-language caller.
6732 '``llvm.stacksave``' Intrinsic
6733 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6740 declare i8* @llvm.stacksave()
6745 The '``llvm.stacksave``' intrinsic is used to remember the current state
6746 of the function stack, for use with
6747 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6748 implementing language features like scoped automatic variable sized
6754 This intrinsic returns a opaque pointer value that can be passed to
6755 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6756 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6757 ``llvm.stacksave``, it effectively restores the state of the stack to
6758 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6759 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6760 were allocated after the ``llvm.stacksave`` was executed.
6762 .. _int_stackrestore:
6764 '``llvm.stackrestore``' Intrinsic
6765 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6772 declare void @llvm.stackrestore(i8* %ptr)
6777 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6778 the function stack to the state it was in when the corresponding
6779 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6780 useful for implementing language features like scoped automatic variable
6781 sized arrays in C99.
6786 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6788 '``llvm.prefetch``' Intrinsic
6789 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6796 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6801 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6802 insert a prefetch instruction if supported; otherwise, it is a noop.
6803 Prefetches have no effect on the behavior of the program but can change
6804 its performance characteristics.
6809 ``address`` is the address to be prefetched, ``rw`` is the specifier
6810 determining if the fetch should be for a read (0) or write (1), and
6811 ``locality`` is a temporal locality specifier ranging from (0) - no
6812 locality, to (3) - extremely local keep in cache. The ``cache type``
6813 specifies whether the prefetch is performed on the data (1) or
6814 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6815 arguments must be constant integers.
6820 This intrinsic does not modify the behavior of the program. In
6821 particular, prefetches cannot trap and do not produce a value. On
6822 targets that support this intrinsic, the prefetch can provide hints to
6823 the processor cache for better performance.
6825 '``llvm.pcmarker``' Intrinsic
6826 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6833 declare void @llvm.pcmarker(i32 <id>)
6838 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6839 Counter (PC) in a region of code to simulators and other tools. The
6840 method is target specific, but it is expected that the marker will use
6841 exported symbols to transmit the PC of the marker. The marker makes no
6842 guarantees that it will remain with any specific instruction after
6843 optimizations. It is possible that the presence of a marker will inhibit
6844 optimizations. The intended use is to be inserted after optimizations to
6845 allow correlations of simulation runs.
6850 ``id`` is a numerical id identifying the marker.
6855 This intrinsic does not modify the behavior of the program. Backends
6856 that do not support this intrinsic may ignore it.
6858 '``llvm.readcyclecounter``' Intrinsic
6859 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6866 declare i64 @llvm.readcyclecounter()
6871 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6872 counter register (or similar low latency, high accuracy clocks) on those
6873 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6874 should map to RPCC. As the backing counters overflow quickly (on the
6875 order of 9 seconds on alpha), this should only be used for small
6881 When directly supported, reading the cycle counter should not modify any
6882 memory. Implementations are allowed to either return a application
6883 specific value or a system wide value. On backends without support, this
6884 is lowered to a constant 0.
6886 Note that runtime support may be conditional on the privilege-level code is
6887 running at and the host platform.
6889 Standard C Library Intrinsics
6890 -----------------------------
6892 LLVM provides intrinsics for a few important standard C library
6893 functions. These intrinsics allow source-language front-ends to pass
6894 information about the alignment of the pointer arguments to the code
6895 generator, providing opportunity for more efficient code generation.
6899 '``llvm.memcpy``' Intrinsic
6900 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6905 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6906 integer bit width and for different address spaces. Not all targets
6907 support all bit widths however.
6911 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6912 i32 <len>, i32 <align>, i1 <isvolatile>)
6913 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6914 i64 <len>, i32 <align>, i1 <isvolatile>)
6919 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6920 source location to the destination location.
6922 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6923 intrinsics do not return a value, takes extra alignment/isvolatile
6924 arguments and the pointers can be in specified address spaces.
6929 The first argument is a pointer to the destination, the second is a
6930 pointer to the source. The third argument is an integer argument
6931 specifying the number of bytes to copy, the fourth argument is the
6932 alignment of the source and destination locations, and the fifth is a
6933 boolean indicating a volatile access.
6935 If the call to this intrinsic has an alignment value that is not 0 or 1,
6936 then the caller guarantees that both the source and destination pointers
6937 are aligned to that boundary.
6939 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6940 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6941 very cleanly specified and it is unwise to depend on it.
6946 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6947 source location to the destination location, which are not allowed to
6948 overlap. It copies "len" bytes of memory over. If the argument is known
6949 to be aligned to some boundary, this can be specified as the fourth
6950 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
6952 '``llvm.memmove``' Intrinsic
6953 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6958 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6959 bit width and for different address space. Not all targets support all
6964 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6965 i32 <len>, i32 <align>, i1 <isvolatile>)
6966 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6967 i64 <len>, i32 <align>, i1 <isvolatile>)
6972 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6973 source location to the destination location. It is similar to the
6974 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6977 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6978 intrinsics do not return a value, takes extra alignment/isvolatile
6979 arguments and the pointers can be in specified address spaces.
6984 The first argument is a pointer to the destination, the second is a
6985 pointer to the source. The third argument is an integer argument
6986 specifying the number of bytes to copy, the fourth argument is the
6987 alignment of the source and destination locations, and the fifth is a
6988 boolean indicating a volatile access.
6990 If the call to this intrinsic has an alignment value that is not 0 or 1,
6991 then the caller guarantees that the source and destination pointers are
6992 aligned to that boundary.
6994 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6995 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6996 not very cleanly specified and it is unwise to depend on it.
7001 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7002 source location to the destination location, which may overlap. It
7003 copies "len" bytes of memory over. If the argument is known to be
7004 aligned to some boundary, this can be specified as the fourth argument,
7005 otherwise it should be set to 0 or 1 (both meaning no alignment).
7007 '``llvm.memset.*``' Intrinsics
7008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7013 This is an overloaded intrinsic. You can use llvm.memset on any integer
7014 bit width and for different address spaces. However, not all targets
7015 support all bit widths.
7019 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7020 i32 <len>, i32 <align>, i1 <isvolatile>)
7021 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7022 i64 <len>, i32 <align>, i1 <isvolatile>)
7027 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7028 particular byte value.
7030 Note that, unlike the standard libc function, the ``llvm.memset``
7031 intrinsic does not return a value and takes extra alignment/volatile
7032 arguments. Also, the destination can be in an arbitrary address space.
7037 The first argument is a pointer to the destination to fill, the second
7038 is the byte value with which to fill it, the third argument is an
7039 integer argument specifying the number of bytes to fill, and the fourth
7040 argument is the known alignment of the destination location.
7042 If the call to this intrinsic has an alignment value that is not 0 or 1,
7043 then the caller guarantees that the destination pointer is aligned to
7046 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7047 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7048 very cleanly specified and it is unwise to depend on it.
7053 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7054 at the destination location. If the argument is known to be aligned to
7055 some boundary, this can be specified as the fourth argument, otherwise
7056 it should be set to 0 or 1 (both meaning no alignment).
7058 '``llvm.sqrt.*``' Intrinsic
7059 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7064 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7065 floating point or vector of floating point type. Not all targets support
7070 declare float @llvm.sqrt.f32(float %Val)
7071 declare double @llvm.sqrt.f64(double %Val)
7072 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7073 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7074 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7079 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7080 returning the same value as the libm '``sqrt``' functions would. Unlike
7081 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7082 negative numbers other than -0.0 (which allows for better optimization,
7083 because there is no need to worry about errno being set).
7084 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7089 The argument and return value are floating point numbers of the same
7095 This function returns the sqrt of the specified operand if it is a
7096 nonnegative floating point number.
7098 '``llvm.powi.*``' Intrinsic
7099 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7104 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7105 floating point or vector of floating point type. Not all targets support
7110 declare float @llvm.powi.f32(float %Val, i32 %power)
7111 declare double @llvm.powi.f64(double %Val, i32 %power)
7112 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7113 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7114 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7119 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7120 specified (positive or negative) power. The order of evaluation of
7121 multiplications is not defined. When a vector of floating point type is
7122 used, the second argument remains a scalar integer value.
7127 The second argument is an integer power, and the first is a value to
7128 raise to that power.
7133 This function returns the first value raised to the second power with an
7134 unspecified sequence of rounding operations.
7136 '``llvm.sin.*``' Intrinsic
7137 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7142 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7143 floating point or vector of floating point type. Not all targets support
7148 declare float @llvm.sin.f32(float %Val)
7149 declare double @llvm.sin.f64(double %Val)
7150 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7151 declare fp128 @llvm.sin.f128(fp128 %Val)
7152 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7157 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7162 The argument and return value are floating point numbers of the same
7168 This function returns the sine of the specified operand, returning the
7169 same values as the libm ``sin`` functions would, and handles error
7170 conditions in the same way.
7172 '``llvm.cos.*``' Intrinsic
7173 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7178 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7179 floating point or vector of floating point type. Not all targets support
7184 declare float @llvm.cos.f32(float %Val)
7185 declare double @llvm.cos.f64(double %Val)
7186 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7187 declare fp128 @llvm.cos.f128(fp128 %Val)
7188 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7193 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7198 The argument and return value are floating point numbers of the same
7204 This function returns the cosine of the specified operand, returning the
7205 same values as the libm ``cos`` functions would, and handles error
7206 conditions in the same way.
7208 '``llvm.pow.*``' Intrinsic
7209 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7214 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7215 floating point or vector of floating point type. Not all targets support
7220 declare float @llvm.pow.f32(float %Val, float %Power)
7221 declare double @llvm.pow.f64(double %Val, double %Power)
7222 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7223 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7224 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7229 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7230 specified (positive or negative) power.
7235 The second argument is a floating point power, and the first is a value
7236 to raise to that power.
7241 This function returns the first value raised to the second power,
7242 returning the same values as the libm ``pow`` functions would, and
7243 handles error conditions in the same way.
7245 '``llvm.exp.*``' Intrinsic
7246 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7251 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7252 floating point or vector of floating point type. Not all targets support
7257 declare float @llvm.exp.f32(float %Val)
7258 declare double @llvm.exp.f64(double %Val)
7259 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7260 declare fp128 @llvm.exp.f128(fp128 %Val)
7261 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7266 The '``llvm.exp.*``' intrinsics perform the exp function.
7271 The argument and return value are floating point numbers of the same
7277 This function returns the same values as the libm ``exp`` functions
7278 would, and handles error conditions in the same way.
7280 '``llvm.exp2.*``' Intrinsic
7281 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7286 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7287 floating point or vector of floating point type. Not all targets support
7292 declare float @llvm.exp2.f32(float %Val)
7293 declare double @llvm.exp2.f64(double %Val)
7294 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7295 declare fp128 @llvm.exp2.f128(fp128 %Val)
7296 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7301 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7306 The argument and return value are floating point numbers of the same
7312 This function returns the same values as the libm ``exp2`` functions
7313 would, and handles error conditions in the same way.
7315 '``llvm.log.*``' Intrinsic
7316 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7321 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7322 floating point or vector of floating point type. Not all targets support
7327 declare float @llvm.log.f32(float %Val)
7328 declare double @llvm.log.f64(double %Val)
7329 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7330 declare fp128 @llvm.log.f128(fp128 %Val)
7331 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7336 The '``llvm.log.*``' intrinsics perform the log function.
7341 The argument and return value are floating point numbers of the same
7347 This function returns the same values as the libm ``log`` functions
7348 would, and handles error conditions in the same way.
7350 '``llvm.log10.*``' Intrinsic
7351 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7356 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7357 floating point or vector of floating point type. Not all targets support
7362 declare float @llvm.log10.f32(float %Val)
7363 declare double @llvm.log10.f64(double %Val)
7364 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7365 declare fp128 @llvm.log10.f128(fp128 %Val)
7366 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7371 The '``llvm.log10.*``' intrinsics perform the log10 function.
7376 The argument and return value are floating point numbers of the same
7382 This function returns the same values as the libm ``log10`` functions
7383 would, and handles error conditions in the same way.
7385 '``llvm.log2.*``' Intrinsic
7386 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7391 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7392 floating point or vector of floating point type. Not all targets support
7397 declare float @llvm.log2.f32(float %Val)
7398 declare double @llvm.log2.f64(double %Val)
7399 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7400 declare fp128 @llvm.log2.f128(fp128 %Val)
7401 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7406 The '``llvm.log2.*``' intrinsics perform the log2 function.
7411 The argument and return value are floating point numbers of the same
7417 This function returns the same values as the libm ``log2`` functions
7418 would, and handles error conditions in the same way.
7420 '``llvm.fma.*``' Intrinsic
7421 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7426 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7427 floating point or vector of floating point type. Not all targets support
7432 declare float @llvm.fma.f32(float %a, float %b, float %c)
7433 declare double @llvm.fma.f64(double %a, double %b, double %c)
7434 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7435 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7436 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7441 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7447 The argument and return value are floating point numbers of the same
7453 This function returns the same values as the libm ``fma`` functions
7456 '``llvm.fabs.*``' Intrinsic
7457 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7462 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7463 floating point or vector of floating point type. Not all targets support
7468 declare float @llvm.fabs.f32(float %Val)
7469 declare double @llvm.fabs.f64(double %Val)
7470 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7471 declare fp128 @llvm.fabs.f128(fp128 %Val)
7472 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7477 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7483 The argument and return value are floating point numbers of the same
7489 This function returns the same values as the libm ``fabs`` functions
7490 would, and handles error conditions in the same way.
7492 '``llvm.copysign.*``' Intrinsic
7493 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7498 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7499 floating point or vector of floating point type. Not all targets support
7504 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7505 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7506 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7507 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7508 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7513 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7514 first operand and the sign of the second operand.
7519 The arguments and return value are floating point numbers of the same
7525 This function returns the same values as the libm ``copysign``
7526 functions would, and handles error conditions in the same way.
7528 '``llvm.floor.*``' Intrinsic
7529 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7534 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7535 floating point or vector of floating point type. Not all targets support
7540 declare float @llvm.floor.f32(float %Val)
7541 declare double @llvm.floor.f64(double %Val)
7542 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7543 declare fp128 @llvm.floor.f128(fp128 %Val)
7544 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7549 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7554 The argument and return value are floating point numbers of the same
7560 This function returns the same values as the libm ``floor`` functions
7561 would, and handles error conditions in the same way.
7563 '``llvm.ceil.*``' Intrinsic
7564 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7569 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7570 floating point or vector of floating point type. Not all targets support
7575 declare float @llvm.ceil.f32(float %Val)
7576 declare double @llvm.ceil.f64(double %Val)
7577 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7578 declare fp128 @llvm.ceil.f128(fp128 %Val)
7579 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7584 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7589 The argument and return value are floating point numbers of the same
7595 This function returns the same values as the libm ``ceil`` functions
7596 would, and handles error conditions in the same way.
7598 '``llvm.trunc.*``' Intrinsic
7599 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7604 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7605 floating point or vector of floating point type. Not all targets support
7610 declare float @llvm.trunc.f32(float %Val)
7611 declare double @llvm.trunc.f64(double %Val)
7612 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7613 declare fp128 @llvm.trunc.f128(fp128 %Val)
7614 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7619 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7620 nearest integer not larger in magnitude than the operand.
7625 The argument and return value are floating point numbers of the same
7631 This function returns the same values as the libm ``trunc`` functions
7632 would, and handles error conditions in the same way.
7634 '``llvm.rint.*``' Intrinsic
7635 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7640 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7641 floating point or vector of floating point type. Not all targets support
7646 declare float @llvm.rint.f32(float %Val)
7647 declare double @llvm.rint.f64(double %Val)
7648 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7649 declare fp128 @llvm.rint.f128(fp128 %Val)
7650 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7655 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7656 nearest integer. It may raise an inexact floating-point exception if the
7657 operand isn't an integer.
7662 The argument and return value are floating point numbers of the same
7668 This function returns the same values as the libm ``rint`` functions
7669 would, and handles error conditions in the same way.
7671 '``llvm.nearbyint.*``' Intrinsic
7672 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7677 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7678 floating point or vector of floating point type. Not all targets support
7683 declare float @llvm.nearbyint.f32(float %Val)
7684 declare double @llvm.nearbyint.f64(double %Val)
7685 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7686 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7687 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7692 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7698 The argument and return value are floating point numbers of the same
7704 This function returns the same values as the libm ``nearbyint``
7705 functions would, and handles error conditions in the same way.
7707 '``llvm.round.*``' Intrinsic
7708 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7713 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7714 floating point or vector of floating point type. Not all targets support
7719 declare float @llvm.round.f32(float %Val)
7720 declare double @llvm.round.f64(double %Val)
7721 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7722 declare fp128 @llvm.round.f128(fp128 %Val)
7723 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7728 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7734 The argument and return value are floating point numbers of the same
7740 This function returns the same values as the libm ``round``
7741 functions would, and handles error conditions in the same way.
7743 Bit Manipulation Intrinsics
7744 ---------------------------
7746 LLVM provides intrinsics for a few important bit manipulation
7747 operations. These allow efficient code generation for some algorithms.
7749 '``llvm.bswap.*``' Intrinsics
7750 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7755 This is an overloaded intrinsic function. You can use bswap on any
7756 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7760 declare i16 @llvm.bswap.i16(i16 <id>)
7761 declare i32 @llvm.bswap.i32(i32 <id>)
7762 declare i64 @llvm.bswap.i64(i64 <id>)
7767 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7768 values with an even number of bytes (positive multiple of 16 bits).
7769 These are useful for performing operations on data that is not in the
7770 target's native byte order.
7775 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7776 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7777 intrinsic returns an i32 value that has the four bytes of the input i32
7778 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7779 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7780 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7781 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7784 '``llvm.ctpop.*``' Intrinsic
7785 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7790 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7791 bit width, or on any vector with integer elements. Not all targets
7792 support all bit widths or vector types, however.
7796 declare i8 @llvm.ctpop.i8(i8 <src>)
7797 declare i16 @llvm.ctpop.i16(i16 <src>)
7798 declare i32 @llvm.ctpop.i32(i32 <src>)
7799 declare i64 @llvm.ctpop.i64(i64 <src>)
7800 declare i256 @llvm.ctpop.i256(i256 <src>)
7801 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7806 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7812 The only argument is the value to be counted. The argument may be of any
7813 integer type, or a vector with integer elements. The return type must
7814 match the argument type.
7819 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7820 each element of a vector.
7822 '``llvm.ctlz.*``' Intrinsic
7823 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7828 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7829 integer bit width, or any vector whose elements are integers. Not all
7830 targets support all bit widths or vector types, however.
7834 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7835 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7836 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7837 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7838 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7839 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7844 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7845 leading zeros in a variable.
7850 The first argument is the value to be counted. This argument may be of
7851 any integer type, or a vectory with integer element type. The return
7852 type must match the first argument type.
7854 The second argument must be a constant and is a flag to indicate whether
7855 the intrinsic should ensure that a zero as the first argument produces a
7856 defined result. Historically some architectures did not provide a
7857 defined result for zero values as efficiently, and many algorithms are
7858 now predicated on avoiding zero-value inputs.
7863 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7864 zeros in a variable, or within each element of the vector. If
7865 ``src == 0`` then the result is the size in bits of the type of ``src``
7866 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7867 ``llvm.ctlz(i32 2) = 30``.
7869 '``llvm.cttz.*``' Intrinsic
7870 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7875 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7876 integer bit width, or any vector of integer elements. Not all targets
7877 support all bit widths or vector types, however.
7881 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7882 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7883 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7884 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7885 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7886 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7891 The '``llvm.cttz``' family of intrinsic functions counts the number of
7897 The first argument is the value to be counted. This argument may be of
7898 any integer type, or a vectory with integer element type. The return
7899 type must match the first argument type.
7901 The second argument must be a constant and is a flag to indicate whether
7902 the intrinsic should ensure that a zero as the first argument produces a
7903 defined result. Historically some architectures did not provide a
7904 defined result for zero values as efficiently, and many algorithms are
7905 now predicated on avoiding zero-value inputs.
7910 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7911 zeros in a variable, or within each element of a vector. If ``src == 0``
7912 then the result is the size in bits of the type of ``src`` if
7913 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7914 ``llvm.cttz(2) = 1``.
7916 Arithmetic with Overflow Intrinsics
7917 -----------------------------------
7919 LLVM provides intrinsics for some arithmetic with overflow operations.
7921 '``llvm.sadd.with.overflow.*``' Intrinsics
7922 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7927 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7928 on any integer bit width.
7932 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7933 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7934 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7939 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7940 a signed addition of the two arguments, and indicate whether an overflow
7941 occurred during the signed summation.
7946 The arguments (%a and %b) and the first element of the result structure
7947 may be of integer types of any bit width, but they must have the same
7948 bit width. The second element of the result structure must be of type
7949 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7955 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7956 a signed addition of the two variables. They return a structure --- the
7957 first element of which is the signed summation, and the second element
7958 of which is a bit specifying if the signed summation resulted in an
7964 .. code-block:: llvm
7966 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7967 %sum = extractvalue {i32, i1} %res, 0
7968 %obit = extractvalue {i32, i1} %res, 1
7969 br i1 %obit, label %overflow, label %normal
7971 '``llvm.uadd.with.overflow.*``' Intrinsics
7972 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7977 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7978 on any integer bit width.
7982 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7983 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7984 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7989 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7990 an unsigned addition of the two arguments, and indicate whether a carry
7991 occurred during the unsigned summation.
7996 The arguments (%a and %b) and the first element of the result structure
7997 may be of integer types of any bit width, but they must have the same
7998 bit width. The second element of the result structure must be of type
7999 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8005 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8006 an unsigned addition of the two arguments. They return a structure --- the
8007 first element of which is the sum, and the second element of which is a
8008 bit specifying if the unsigned summation resulted in a carry.
8013 .. code-block:: llvm
8015 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8016 %sum = extractvalue {i32, i1} %res, 0
8017 %obit = extractvalue {i32, i1} %res, 1
8018 br i1 %obit, label %carry, label %normal
8020 '``llvm.ssub.with.overflow.*``' Intrinsics
8021 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8026 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8027 on any integer bit width.
8031 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8032 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8033 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8038 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8039 a signed subtraction of the two arguments, and indicate whether an
8040 overflow occurred during the signed subtraction.
8045 The arguments (%a and %b) and the first element of the result structure
8046 may be of integer types of any bit width, but they must have the same
8047 bit width. The second element of the result structure must be of type
8048 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8054 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8055 a signed subtraction of the two arguments. They return a structure --- the
8056 first element of which is the subtraction, and the second element of
8057 which is a bit specifying if the signed subtraction resulted in an
8063 .. code-block:: llvm
8065 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8066 %sum = extractvalue {i32, i1} %res, 0
8067 %obit = extractvalue {i32, i1} %res, 1
8068 br i1 %obit, label %overflow, label %normal
8070 '``llvm.usub.with.overflow.*``' Intrinsics
8071 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8076 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8077 on any integer bit width.
8081 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8082 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8083 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8088 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8089 an unsigned subtraction of the two arguments, and indicate whether an
8090 overflow occurred during the unsigned subtraction.
8095 The arguments (%a and %b) and the first element of the result structure
8096 may be of integer types of any bit width, but they must have the same
8097 bit width. The second element of the result structure must be of type
8098 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8104 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8105 an unsigned subtraction of the two arguments. They return a structure ---
8106 the first element of which is the subtraction, and the second element of
8107 which is a bit specifying if the unsigned subtraction resulted in an
8113 .. code-block:: llvm
8115 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8116 %sum = extractvalue {i32, i1} %res, 0
8117 %obit = extractvalue {i32, i1} %res, 1
8118 br i1 %obit, label %overflow, label %normal
8120 '``llvm.smul.with.overflow.*``' Intrinsics
8121 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8126 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8127 on any integer bit width.
8131 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8132 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8133 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8138 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8139 a signed multiplication of the two arguments, and indicate whether an
8140 overflow occurred during the signed multiplication.
8145 The arguments (%a and %b) and the first element of the result structure
8146 may be of integer types of any bit width, but they must have the same
8147 bit width. The second element of the result structure must be of type
8148 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8154 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8155 a signed multiplication of the two arguments. They return a structure ---
8156 the first element of which is the multiplication, and the second element
8157 of which is a bit specifying if the signed multiplication resulted in an
8163 .. code-block:: llvm
8165 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8166 %sum = extractvalue {i32, i1} %res, 0
8167 %obit = extractvalue {i32, i1} %res, 1
8168 br i1 %obit, label %overflow, label %normal
8170 '``llvm.umul.with.overflow.*``' Intrinsics
8171 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8176 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8177 on any integer bit width.
8181 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8182 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8183 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8188 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8189 a unsigned multiplication of the two arguments, and indicate whether an
8190 overflow occurred during the unsigned multiplication.
8195 The arguments (%a and %b) and the first element of the result structure
8196 may be of integer types of any bit width, but they must have the same
8197 bit width. The second element of the result structure must be of type
8198 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8204 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8205 an unsigned multiplication of the two arguments. They return a structure ---
8206 the first element of which is the multiplication, and the second
8207 element of which is a bit specifying if the unsigned multiplication
8208 resulted in an overflow.
8213 .. code-block:: llvm
8215 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8216 %sum = extractvalue {i32, i1} %res, 0
8217 %obit = extractvalue {i32, i1} %res, 1
8218 br i1 %obit, label %overflow, label %normal
8220 Specialised Arithmetic Intrinsics
8221 ---------------------------------
8223 '``llvm.fmuladd.*``' Intrinsic
8224 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8231 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8232 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8237 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8238 expressions that can be fused if the code generator determines that (a) the
8239 target instruction set has support for a fused operation, and (b) that the
8240 fused operation is more efficient than the equivalent, separate pair of mul
8241 and add instructions.
8246 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8247 multiplicands, a and b, and an addend c.
8256 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8258 is equivalent to the expression a \* b + c, except that rounding will
8259 not be performed between the multiplication and addition steps if the
8260 code generator fuses the operations. Fusion is not guaranteed, even if
8261 the target platform supports it. If a fused multiply-add is required the
8262 corresponding llvm.fma.\* intrinsic function should be used instead.
8267 .. code-block:: llvm
8269 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8271 Half Precision Floating Point Intrinsics
8272 ----------------------------------------
8274 For most target platforms, half precision floating point is a
8275 storage-only format. This means that it is a dense encoding (in memory)
8276 but does not support computation in the format.
8278 This means that code must first load the half-precision floating point
8279 value as an i16, then convert it to float with
8280 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8281 then be performed on the float value (including extending to double
8282 etc). To store the value back to memory, it is first converted to float
8283 if needed, then converted to i16 with
8284 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8287 .. _int_convert_to_fp16:
8289 '``llvm.convert.to.fp16``' Intrinsic
8290 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8297 declare i16 @llvm.convert.to.fp16(f32 %a)
8302 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8303 from single precision floating point format to half precision floating
8309 The intrinsic function contains single argument - the value to be
8315 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8316 from single precision floating point format to half precision floating
8317 point format. The return value is an ``i16`` which contains the
8323 .. code-block:: llvm
8325 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8326 store i16 %res, i16* @x, align 2
8328 .. _int_convert_from_fp16:
8330 '``llvm.convert.from.fp16``' Intrinsic
8331 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8338 declare f32 @llvm.convert.from.fp16(i16 %a)
8343 The '``llvm.convert.from.fp16``' intrinsic function performs a
8344 conversion from half precision floating point format to single precision
8345 floating point format.
8350 The intrinsic function contains single argument - the value to be
8356 The '``llvm.convert.from.fp16``' intrinsic function performs a
8357 conversion from half single precision floating point format to single
8358 precision floating point format. The input half-float value is
8359 represented by an ``i16`` value.
8364 .. code-block:: llvm
8366 %a = load i16* @x, align 2
8367 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8372 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8373 prefix), are described in the `LLVM Source Level
8374 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8377 Exception Handling Intrinsics
8378 -----------------------------
8380 The LLVM exception handling intrinsics (which all start with
8381 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8382 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8386 Trampoline Intrinsics
8387 ---------------------
8389 These intrinsics make it possible to excise one parameter, marked with
8390 the :ref:`nest <nest>` attribute, from a function. The result is a
8391 callable function pointer lacking the nest parameter - the caller does
8392 not need to provide a value for it. Instead, the value to use is stored
8393 in advance in a "trampoline", a block of memory usually allocated on the
8394 stack, which also contains code to splice the nest value into the
8395 argument list. This is used to implement the GCC nested function address
8398 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8399 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8400 It can be created as follows:
8402 .. code-block:: llvm
8404 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8405 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8406 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8407 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8408 %fp = bitcast i8* %p to i32 (i32, i32)*
8410 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8411 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8415 '``llvm.init.trampoline``' Intrinsic
8416 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8423 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8428 This fills the memory pointed to by ``tramp`` with executable code,
8429 turning it into a trampoline.
8434 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8435 pointers. The ``tramp`` argument must point to a sufficiently large and
8436 sufficiently aligned block of memory; this memory is written to by the
8437 intrinsic. Note that the size and the alignment are target-specific -
8438 LLVM currently provides no portable way of determining them, so a
8439 front-end that generates this intrinsic needs to have some
8440 target-specific knowledge. The ``func`` argument must hold a function
8441 bitcast to an ``i8*``.
8446 The block of memory pointed to by ``tramp`` is filled with target
8447 dependent code, turning it into a function. Then ``tramp`` needs to be
8448 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8449 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8450 function's signature is the same as that of ``func`` with any arguments
8451 marked with the ``nest`` attribute removed. At most one such ``nest``
8452 argument is allowed, and it must be of pointer type. Calling the new
8453 function is equivalent to calling ``func`` with the same argument list,
8454 but with ``nval`` used for the missing ``nest`` argument. If, after
8455 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8456 modified, then the effect of any later call to the returned function
8457 pointer is undefined.
8461 '``llvm.adjust.trampoline``' Intrinsic
8462 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8469 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8474 This performs any required machine-specific adjustment to the address of
8475 a trampoline (passed as ``tramp``).
8480 ``tramp`` must point to a block of memory which already has trampoline
8481 code filled in by a previous call to
8482 :ref:`llvm.init.trampoline <int_it>`.
8487 On some architectures the address of the code to be executed needs to be
8488 different to the address where the trampoline is actually stored. This
8489 intrinsic returns the executable address corresponding to ``tramp``
8490 after performing the required machine specific adjustments. The pointer
8491 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8496 This class of intrinsics exists to information about the lifetime of
8497 memory objects and ranges where variables are immutable.
8499 '``llvm.lifetime.start``' Intrinsic
8500 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8507 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8512 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8518 The first argument is a constant integer representing the size of the
8519 object, or -1 if it is variable sized. The second argument is a pointer
8525 This intrinsic indicates that before this point in the code, the value
8526 of the memory pointed to by ``ptr`` is dead. This means that it is known
8527 to never be used and has an undefined value. A load from the pointer
8528 that precedes this intrinsic can be replaced with ``'undef'``.
8530 '``llvm.lifetime.end``' Intrinsic
8531 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8538 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8543 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8549 The first argument is a constant integer representing the size of the
8550 object, or -1 if it is variable sized. The second argument is a pointer
8556 This intrinsic indicates that after this point in the code, the value of
8557 the memory pointed to by ``ptr`` is dead. This means that it is known to
8558 never be used and has an undefined value. Any stores into the memory
8559 object following this intrinsic may be removed as dead.
8561 '``llvm.invariant.start``' Intrinsic
8562 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8569 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8574 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8575 a memory object will not change.
8580 The first argument is a constant integer representing the size of the
8581 object, or -1 if it is variable sized. The second argument is a pointer
8587 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8588 the return value, the referenced memory location is constant and
8591 '``llvm.invariant.end``' Intrinsic
8592 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8599 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8604 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8605 memory object are mutable.
8610 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8611 The second argument is a constant integer representing the size of the
8612 object, or -1 if it is variable sized and the third argument is a
8613 pointer to the object.
8618 This intrinsic indicates that the memory is mutable again.
8623 This class of intrinsics is designed to be generic and has no specific
8626 '``llvm.var.annotation``' Intrinsic
8627 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8634 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8639 The '``llvm.var.annotation``' intrinsic.
8644 The first argument is a pointer to a value, the second is a pointer to a
8645 global string, the third is a pointer to a global string which is the
8646 source file name, and the last argument is the line number.
8651 This intrinsic allows annotation of local variables with arbitrary
8652 strings. This can be useful for special purpose optimizations that want
8653 to look for these annotations. These have no other defined use; they are
8654 ignored by code generation and optimization.
8656 '``llvm.ptr.annotation.*``' Intrinsic
8657 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8662 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8663 pointer to an integer of any width. *NOTE* you must specify an address space for
8664 the pointer. The identifier for the default address space is the integer
8669 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8670 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8671 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8672 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8673 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8678 The '``llvm.ptr.annotation``' intrinsic.
8683 The first argument is a pointer to an integer value of arbitrary bitwidth
8684 (result of some expression), the second is a pointer to a global string, the
8685 third is a pointer to a global string which is the source file name, and the
8686 last argument is the line number. It returns the value of the first argument.
8691 This intrinsic allows annotation of a pointer to an integer with arbitrary
8692 strings. This can be useful for special purpose optimizations that want to look
8693 for these annotations. These have no other defined use; they are ignored by code
8694 generation and optimization.
8696 '``llvm.annotation.*``' Intrinsic
8697 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8702 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8703 any integer bit width.
8707 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8708 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8709 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8710 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8711 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8716 The '``llvm.annotation``' intrinsic.
8721 The first argument is an integer value (result of some expression), the
8722 second is a pointer to a global string, the third is a pointer to a
8723 global string which is the source file name, and the last argument is
8724 the line number. It returns the value of the first argument.
8729 This intrinsic allows annotations to be put on arbitrary expressions
8730 with arbitrary strings. This can be useful for special purpose
8731 optimizations that want to look for these annotations. These have no
8732 other defined use; they are ignored by code generation and optimization.
8734 '``llvm.trap``' Intrinsic
8735 ^^^^^^^^^^^^^^^^^^^^^^^^^
8742 declare void @llvm.trap() noreturn nounwind
8747 The '``llvm.trap``' intrinsic.
8757 This intrinsic is lowered to the target dependent trap instruction. If
8758 the target does not have a trap instruction, this intrinsic will be
8759 lowered to a call of the ``abort()`` function.
8761 '``llvm.debugtrap``' Intrinsic
8762 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8769 declare void @llvm.debugtrap() nounwind
8774 The '``llvm.debugtrap``' intrinsic.
8784 This intrinsic is lowered to code which is intended to cause an
8785 execution trap with the intention of requesting the attention of a
8788 '``llvm.stackprotector``' Intrinsic
8789 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8796 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8801 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8802 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8803 is placed on the stack before local variables.
8808 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8809 The first argument is the value loaded from the stack guard
8810 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8811 enough space to hold the value of the guard.
8816 This intrinsic causes the prologue/epilogue inserter to force the position of
8817 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8818 to ensure that if a local variable on the stack is overwritten, it will destroy
8819 the value of the guard. When the function exits, the guard on the stack is
8820 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8821 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8822 calling the ``__stack_chk_fail()`` function.
8824 '``llvm.stackprotectorcheck``' Intrinsic
8825 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8832 declare void @llvm.stackprotectorcheck(i8** <guard>)
8837 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8838 created stack protector and if they are not equal calls the
8839 ``__stack_chk_fail()`` function.
8844 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8845 the variable ``@__stack_chk_guard``.
8850 This intrinsic is provided to perform the stack protector check by comparing
8851 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8852 values do not match call the ``__stack_chk_fail()`` function.
8854 The reason to provide this as an IR level intrinsic instead of implementing it
8855 via other IR operations is that in order to perform this operation at the IR
8856 level without an intrinsic, one would need to create additional basic blocks to
8857 handle the success/failure cases. This makes it difficult to stop the stack
8858 protector check from disrupting sibling tail calls in Codegen. With this
8859 intrinsic, we are able to generate the stack protector basic blocks late in
8860 codegen after the tail call decision has occurred.
8862 '``llvm.objectsize``' Intrinsic
8863 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8870 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8871 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8876 The ``llvm.objectsize`` intrinsic is designed to provide information to
8877 the optimizers to determine at compile time whether a) an operation
8878 (like memcpy) will overflow a buffer that corresponds to an object, or
8879 b) that a runtime check for overflow isn't necessary. An object in this
8880 context means an allocation of a specific class, structure, array, or
8886 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8887 argument is a pointer to or into the ``object``. The second argument is
8888 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8889 or -1 (if false) when the object size is unknown. The second argument
8890 only accepts constants.
8895 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8896 the size of the object concerned. If the size cannot be determined at
8897 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8898 on the ``min`` argument).
8900 '``llvm.expect``' Intrinsic
8901 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8908 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8909 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8914 The ``llvm.expect`` intrinsic provides information about expected (the
8915 most probable) value of ``val``, which can be used by optimizers.
8920 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8921 a value. The second argument is an expected value, this needs to be a
8922 constant value, variables are not allowed.
8927 This intrinsic is lowered to the ``val``.
8929 '``llvm.donothing``' Intrinsic
8930 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8937 declare void @llvm.donothing() nounwind readnone
8942 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8943 only intrinsic that can be called with an invoke instruction.
8953 This intrinsic does nothing, and it's removed by optimizers and ignored