1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][a-zA-Z$._][a-zA-Z$._0-9]*``'. Identifiers which require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves.
83 #. Unnamed values are represented as an unsigned numeric value with
84 their prefix. For example, ``%12``, ``@2``, ``%44``.
85 #. Constants, which are described in the section Constants_ below.
87 LLVM requires that values start with a prefix for two reasons: Compilers
88 don't need to worry about name clashes with reserved words, and the set
89 of reserved words may be expanded in the future without penalty.
90 Additionally, unnamed identifiers allow a compiler to quickly come up
91 with a temporary variable without having to avoid symbol table
94 Reserved words in LLVM are very similar to reserved words in other
95 languages. There are keywords for different opcodes ('``add``',
96 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
97 '``i32``', etc...), and others. These reserved words cannot conflict
98 with variable names, because none of them start with a prefix character
101 Here is an example of LLVM code to multiply the integer variable
108 %result = mul i32 %X, 8
110 After strength reduction:
114 %result = shl i32 %X, 3
120 %0 = add i32 %X, %X ; yields {i32}:%0
121 %1 = add i32 %0, %0 ; yields {i32}:%1
122 %result = add i32 %1, %1
124 This last way of multiplying ``%X`` by 8 illustrates several important
125 lexical features of LLVM:
127 #. Comments are delimited with a '``;``' and go until the end of line.
128 #. Unnamed temporaries are created when the result of a computation is
129 not assigned to a named value.
130 #. Unnamed temporaries are numbered sequentially (using a per-function
131 incrementing counter, starting with 0). Note that basic blocks are
132 included in this numbering. For example, if the entry basic block is not
133 given a label name, then it will get number 0.
135 It also shows a convention that we follow in this document. When
136 demonstrating instructions, we will follow an instruction with a comment
137 that defines the type and name of value produced.
145 LLVM programs are composed of ``Module``'s, each of which is a
146 translation unit of the input programs. Each module consists of
147 functions, global variables, and symbol table entries. Modules may be
148 combined together with the LLVM linker, which merges function (and
149 global variable) definitions, resolves forward declarations, and merges
150 symbol table entries. Here is an example of the "hello world" module:
154 ; Declare the string constant as a global constant.
155 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
157 ; External declaration of the puts function
158 declare i32 @puts(i8* nocapture) nounwind
160 ; Definition of main function
161 define i32 @main() { ; i32()*
162 ; Convert [13 x i8]* to i8 *...
163 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
165 ; Call puts function to write out the string to stdout.
166 call i32 @puts(i8* %cast210)
171 !1 = metadata !{i32 42}
174 This example is made up of a :ref:`global variable <globalvars>` named
175 "``.str``", an external declaration of the "``puts``" function, a
176 :ref:`function definition <functionstructure>` for "``main``" and
177 :ref:`named metadata <namedmetadatastructure>` "``foo``".
179 In general, a module is made up of a list of global values (where both
180 functions and global variables are global values). Global values are
181 represented by a pointer to a memory location (in this case, a pointer
182 to an array of char, and a pointer to a function), and have one of the
183 following :ref:`linkage types <linkage>`.
190 All Global Variables and Functions have one of the following types of
194 Global values with "``private``" linkage are only directly
195 accessible by objects in the current module. In particular, linking
196 code into a module with an private global value may cause the
197 private to be renamed as necessary to avoid collisions. Because the
198 symbol is private to the module, all references can be updated. This
199 doesn't show up in any symbol table in the object file.
201 Similar to ``private``, but the symbol is passed through the
202 assembler and evaluated by the linker. Unlike normal strong symbols,
203 they are removed by the linker from the final linked image
204 (executable or dynamic library).
205 ``linker_private_weak``
206 Similar to "``linker_private``", but the symbol is weak. Note that
207 ``linker_private_weak`` symbols are subject to coalescing by the
208 linker. The symbols are removed by the linker from the final linked
209 image (executable or dynamic library).
211 Similar to private, but the value shows as a local symbol
212 (``STB_LOCAL`` in the case of ELF) in the object file. This
213 corresponds to the notion of the '``static``' keyword in C.
214 ``available_externally``
215 Globals with "``available_externally``" linkage are never emitted
216 into the object file corresponding to the LLVM module. They exist to
217 allow inlining and other optimizations to take place given knowledge
218 of the definition of the global, which is known to be somewhere
219 outside the module. Globals with ``available_externally`` linkage
220 are allowed to be discarded at will, and are otherwise the same as
221 ``linkonce_odr``. This linkage type is only allowed on definitions,
224 Globals with "``linkonce``" linkage are merged with other globals of
225 the same name when linkage occurs. This can be used to implement
226 some forms of inline functions, templates, or other code which must
227 be generated in each translation unit that uses it, but where the
228 body may be overridden with a more definitive definition later.
229 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
230 that ``linkonce`` linkage does not actually allow the optimizer to
231 inline the body of this function into callers because it doesn't
232 know if this definition of the function is the definitive definition
233 within the program or whether it will be overridden by a stronger
234 definition. To enable inlining and other optimizations, use
235 "``linkonce_odr``" linkage.
237 "``weak``" linkage has the same merging semantics as ``linkonce``
238 linkage, except that unreferenced globals with ``weak`` linkage may
239 not be discarded. This is used for globals that are declared "weak"
242 "``common``" linkage is most similar to "``weak``" linkage, but they
243 are used for tentative definitions in C, such as "``int X;``" at
244 global scope. Symbols with "``common``" linkage are merged in the
245 same way as ``weak symbols``, and they may not be deleted if
246 unreferenced. ``common`` symbols may not have an explicit section,
247 must have a zero initializer, and may not be marked
248 ':ref:`constant <globalvars>`'. Functions and aliases may not have
251 .. _linkage_appending:
254 "``appending``" linkage may only be applied to global variables of
255 pointer to array type. When two global variables with appending
256 linkage are linked together, the two global arrays are appended
257 together. This is the LLVM, typesafe, equivalent of having the
258 system linker append together "sections" with identical names when
261 The semantics of this linkage follow the ELF object file model: the
262 symbol is weak until linked, if not linked, the symbol becomes null
263 instead of being an undefined reference.
264 ``linkonce_odr``, ``weak_odr``
265 Some languages allow differing globals to be merged, such as two
266 functions with different semantics. Other languages, such as
267 ``C++``, ensure that only equivalent globals are ever merged (the
268 "one definition rule" --- "ODR"). Such languages can use the
269 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
270 global will only be merged with equivalent globals. These linkage
271 types are otherwise the same as their non-``odr`` versions.
273 If none of the above identifiers are used, the global is externally
274 visible, meaning that it participates in linkage and can be used to
275 resolve external symbol references.
277 The next two types of linkage are targeted for Microsoft Windows
278 platform only. They are designed to support importing (exporting)
279 symbols from (to) DLLs (Dynamic Link Libraries).
282 "``dllimport``" linkage causes the compiler to reference a function
283 or variable via a global pointer to a pointer that is set up by the
284 DLL exporting the symbol. On Microsoft Windows targets, the pointer
285 name is formed by combining ``__imp_`` and the function or variable
288 "``dllexport``" linkage causes the compiler to provide a global
289 pointer to a pointer in a DLL, so that it can be referenced with the
290 ``dllimport`` attribute. On Microsoft Windows targets, the pointer
291 name is formed by combining ``__imp_`` and the function or variable
292 name. Since this linkage exists for defining a dll interface, the
293 compiler, assembler and linker know it is externally referenced and
294 must refrain from deleting the symbol.
296 It is illegal for a function *declaration* to have any linkage type
297 other than ``external``, ``dllimport`` or ``extern_weak``.
304 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
305 :ref:`invokes <i_invoke>` can all have an optional calling convention
306 specified for the call. The calling convention of any pair of dynamic
307 caller/callee must match, or the behavior of the program is undefined.
308 The following calling conventions are supported by LLVM, and more may be
311 "``ccc``" - The C calling convention
312 This calling convention (the default if no other calling convention
313 is specified) matches the target C calling conventions. This calling
314 convention supports varargs function calls and tolerates some
315 mismatch in the declared prototype and implemented declaration of
316 the function (as does normal C).
317 "``fastcc``" - The fast calling convention
318 This calling convention attempts to make calls as fast as possible
319 (e.g. by passing things in registers). This calling convention
320 allows the target to use whatever tricks it wants to produce fast
321 code for the target, without having to conform to an externally
322 specified ABI (Application Binary Interface). `Tail calls can only
323 be optimized when this, the GHC or the HiPE convention is
324 used. <CodeGenerator.html#id80>`_ This calling convention does not
325 support varargs and requires the prototype of all callees to exactly
326 match the prototype of the function definition.
327 "``coldcc``" - The cold calling convention
328 This calling convention attempts to make code in the caller as
329 efficient as possible under the assumption that the call is not
330 commonly executed. As such, these calls often preserve all registers
331 so that the call does not break any live ranges in the caller side.
332 This calling convention does not support varargs and requires the
333 prototype of all callees to exactly match the prototype of the
335 "``cc 10``" - GHC convention
336 This calling convention has been implemented specifically for use by
337 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
338 It passes everything in registers, going to extremes to achieve this
339 by disabling callee save registers. This calling convention should
340 not be used lightly but only for specific situations such as an
341 alternative to the *register pinning* performance technique often
342 used when implementing functional programming languages. At the
343 moment only X86 supports this convention and it has the following
346 - On *X86-32* only supports up to 4 bit type parameters. No
347 floating point types are supported.
348 - On *X86-64* only supports up to 10 bit type parameters and 6
349 floating point parameters.
351 This calling convention supports `tail call
352 optimization <CodeGenerator.html#id80>`_ but requires both the
353 caller and callee are using it.
354 "``cc 11``" - The HiPE calling convention
355 This calling convention has been implemented specifically for use by
356 the `High-Performance Erlang
357 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
358 native code compiler of the `Ericsson's Open Source Erlang/OTP
359 system <http://www.erlang.org/download.shtml>`_. It uses more
360 registers for argument passing than the ordinary C calling
361 convention and defines no callee-saved registers. The calling
362 convention properly supports `tail call
363 optimization <CodeGenerator.html#id80>`_ but requires that both the
364 caller and the callee use it. It uses a *register pinning*
365 mechanism, similar to GHC's convention, for keeping frequently
366 accessed runtime components pinned to specific hardware registers.
367 At the moment only X86 supports this convention (both 32 and 64
369 "``webkit_jscc``" - WebKit's JavaScript calling convention
370 This calling convention has been implemented for `WebKit FTL JIT
371 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
372 stack right to left (as cdecl does), and returns a value in the
373 platform's customary return register.
374 "``anyregcc``" - Dynamic calling convention for code patching
375 This is a special convention that supports patching an arbitrary code
376 sequence in place of a call site. This convention forces the call
377 arguments into registers but allows them to be dynamcially
378 allocated. This can currently only be used with calls to
379 llvm.experimental.patchpoint because only this intrinsic records
380 the location of its arguments in a side table. See :doc:`StackMaps`.
381 "``cc <n>``" - Numbered convention
382 Any calling convention may be specified by number, allowing
383 target-specific calling conventions to be used. Target specific
384 calling conventions start at 64.
386 More calling conventions can be added/defined on an as-needed basis, to
387 support Pascal conventions or any other well-known target-independent
390 .. _visibilitystyles:
395 All Global Variables and Functions have one of the following visibility
398 "``default``" - Default style
399 On targets that use the ELF object file format, default visibility
400 means that the declaration is visible to other modules and, in
401 shared libraries, means that the declared entity may be overridden.
402 On Darwin, default visibility means that the declaration is visible
403 to other modules. Default visibility corresponds to "external
404 linkage" in the language.
405 "``hidden``" - Hidden style
406 Two declarations of an object with hidden visibility refer to the
407 same object if they are in the same shared object. Usually, hidden
408 visibility indicates that the symbol will not be placed into the
409 dynamic symbol table, so no other module (executable or shared
410 library) can reference it directly.
411 "``protected``" - Protected style
412 On ELF, protected visibility indicates that the symbol will be
413 placed in the dynamic symbol table, but that references within the
414 defining module will bind to the local symbol. That is, the symbol
415 cannot be overridden by another module.
422 LLVM IR allows you to specify name aliases for certain types. This can
423 make it easier to read the IR and make the IR more condensed
424 (particularly when recursive types are involved). An example of a name
429 %mytype = type { %mytype*, i32 }
431 You may give a name to any :ref:`type <typesystem>` except
432 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
433 expected with the syntax "%mytype".
435 Note that type names are aliases for the structural type that they
436 indicate, and that you can therefore specify multiple names for the same
437 type. This often leads to confusing behavior when dumping out a .ll
438 file. Since LLVM IR uses structural typing, the name is not part of the
439 type. When printing out LLVM IR, the printer will pick *one name* to
440 render all types of a particular shape. This means that if you have code
441 where two different source types end up having the same LLVM type, that
442 the dumper will sometimes print the "wrong" or unexpected type. This is
443 an important design point and isn't going to change.
450 Global variables define regions of memory allocated at compilation time
453 Global variables definitions must be initialized, may have an explicit section
454 to be placed in, and may have an optional explicit alignment specified.
456 Global variables in other translation units can also be declared, in which
457 case they don't have an initializer.
459 A variable may be defined as ``thread_local``, which means that it will
460 not be shared by threads (each thread will have a separated copy of the
461 variable). Not all targets support thread-local variables. Optionally, a
462 TLS model may be specified:
465 For variables that are only used within the current shared library.
467 For variables in modules that will not be loaded dynamically.
469 For variables defined in the executable and only used within it.
471 The models correspond to the ELF TLS models; see `ELF Handling For
472 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
473 more information on under which circumstances the different models may
474 be used. The target may choose a different TLS model if the specified
475 model is not supported, or if a better choice of model can be made.
477 A variable may be defined as a global ``constant``, which indicates that
478 the contents of the variable will **never** be modified (enabling better
479 optimization, allowing the global data to be placed in the read-only
480 section of an executable, etc). Note that variables that need runtime
481 initialization cannot be marked ``constant`` as there is a store to the
484 LLVM explicitly allows *declarations* of global variables to be marked
485 constant, even if the final definition of the global is not. This
486 capability can be used to enable slightly better optimization of the
487 program, but requires the language definition to guarantee that
488 optimizations based on the 'constantness' are valid for the translation
489 units that do not include the definition.
491 As SSA values, global variables define pointer values that are in scope
492 (i.e. they dominate) all basic blocks in the program. Global variables
493 always define a pointer to their "content" type because they describe a
494 region of memory, and all memory objects in LLVM are accessed through
497 Global variables can be marked with ``unnamed_addr`` which indicates
498 that the address is not significant, only the content. Constants marked
499 like this can be merged with other constants if they have the same
500 initializer. Note that a constant with significant address *can* be
501 merged with a ``unnamed_addr`` constant, the result being a constant
502 whose address is significant.
504 A global variable may be declared to reside in a target-specific
505 numbered address space. For targets that support them, address spaces
506 may affect how optimizations are performed and/or what target
507 instructions are used to access the variable. The default address space
508 is zero. The address space qualifier must precede any other attributes.
510 LLVM allows an explicit section to be specified for globals. If the
511 target supports it, it will emit globals to the section specified.
513 By default, global initializers are optimized by assuming that global
514 variables defined within the module are not modified from their
515 initial values before the start of the global initializer. This is
516 true even for variables potentially accessible from outside the
517 module, including those with external linkage or appearing in
518 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
519 by marking the variable with ``externally_initialized``.
521 An explicit alignment may be specified for a global, which must be a
522 power of 2. If not present, or if the alignment is set to zero, the
523 alignment of the global is set by the target to whatever it feels
524 convenient. If an explicit alignment is specified, the global is forced
525 to have exactly that alignment. Targets and optimizers are not allowed
526 to over-align the global if the global has an assigned section. In this
527 case, the extra alignment could be observable: for example, code could
528 assume that the globals are densely packed in their section and try to
529 iterate over them as an array, alignment padding would break this
532 For example, the following defines a global in a numbered address space
533 with an initializer, section, and alignment:
537 @G = addrspace(5) constant float 1.0, section "foo", align 4
539 The following example just declares a global variable
543 @G = external global i32
545 The following example defines a thread-local global with the
546 ``initialexec`` TLS model:
550 @G = thread_local(initialexec) global i32 0, align 4
552 .. _functionstructure:
557 LLVM function definitions consist of the "``define``" keyword, an
558 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
559 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
560 an optional ``unnamed_addr`` attribute, a return type, an optional
561 :ref:`parameter attribute <paramattrs>` for the return type, a function
562 name, a (possibly empty) argument list (each with optional :ref:`parameter
563 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
564 an optional section, an optional alignment, an optional :ref:`garbage
565 collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
566 curly brace, a list of basic blocks, and a closing curly brace.
568 LLVM function declarations consist of the "``declare``" keyword, an
569 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
570 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
571 an optional ``unnamed_addr`` attribute, a return type, an optional
572 :ref:`parameter attribute <paramattrs>` for the return type, a function
573 name, a possibly empty list of arguments, an optional alignment, an optional
574 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
576 A function definition contains a list of basic blocks, forming the CFG (Control
577 Flow Graph) for the function. Each basic block may optionally start with a label
578 (giving the basic block a symbol table entry), contains a list of instructions,
579 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
580 function return). If an explicit label is not provided, a block is assigned an
581 implicit numbered label, using the next value from the same counter as used for
582 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
583 entry block does not have an explicit label, it will be assigned label "%0",
584 then the first unnamed temporary in that block will be "%1", etc.
586 The first basic block in a function is special in two ways: it is
587 immediately executed on entrance to the function, and it is not allowed
588 to have predecessor basic blocks (i.e. there can not be any branches to
589 the entry block of a function). Because the block can have no
590 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
592 LLVM allows an explicit section to be specified for functions. If the
593 target supports it, it will emit functions to the section specified.
595 An explicit alignment may be specified for a function. If not present,
596 or if the alignment is set to zero, the alignment of the function is set
597 by the target to whatever it feels convenient. If an explicit alignment
598 is specified, the function is forced to have at least that much
599 alignment. All alignments must be a power of 2.
601 If the ``unnamed_addr`` attribute is given, the address is know to not
602 be significant and two identical functions can be merged.
606 define [linkage] [visibility]
608 <ResultType> @<FunctionName> ([argument list])
609 [fn Attrs] [section "name"] [align N]
610 [gc] [prefix Constant] { ... }
617 Aliases act as "second name" for the aliasee value (which can be either
618 function, global variable, another alias or bitcast of global value).
619 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
620 :ref:`visibility style <visibility>`.
624 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
626 The linkage must be one of ``private``, ``linker_private``,
627 ``linker_private_weak``, ``internal``, ``linkonce``, ``weak``,
628 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
629 might not correctly handle dropping a weak symbol that is aliased by a non-weak
632 .. _namedmetadatastructure:
637 Named metadata is a collection of metadata. :ref:`Metadata
638 nodes <metadata>` (but not metadata strings) are the only valid
639 operands for a named metadata.
643 ; Some unnamed metadata nodes, which are referenced by the named metadata.
644 !0 = metadata !{metadata !"zero"}
645 !1 = metadata !{metadata !"one"}
646 !2 = metadata !{metadata !"two"}
648 !name = !{!0, !1, !2}
655 The return type and each parameter of a function type may have a set of
656 *parameter attributes* associated with them. Parameter attributes are
657 used to communicate additional information about the result or
658 parameters of a function. Parameter attributes are considered to be part
659 of the function, not of the function type, so functions with different
660 parameter attributes can have the same function type.
662 Parameter attributes are simple keywords that follow the type specified.
663 If multiple parameter attributes are needed, they are space separated.
668 declare i32 @printf(i8* noalias nocapture, ...)
669 declare i32 @atoi(i8 zeroext)
670 declare signext i8 @returns_signed_char()
672 Note that any attributes for the function result (``nounwind``,
673 ``readonly``) come immediately after the argument list.
675 Currently, only the following parameter attributes are defined:
678 This indicates to the code generator that the parameter or return
679 value should be zero-extended to the extent required by the target's
680 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
681 the caller (for a parameter) or the callee (for a return value).
683 This indicates to the code generator that the parameter or return
684 value should be sign-extended to the extent required by the target's
685 ABI (which is usually 32-bits) by the caller (for a parameter) or
686 the callee (for a return value).
688 This indicates that this parameter or return value should be treated
689 in a special target-dependent fashion during while emitting code for
690 a function call or return (usually, by putting it in a register as
691 opposed to memory, though some targets use it to distinguish between
692 two different kinds of registers). Use of this attribute is
695 This indicates that the pointer parameter should really be passed by
696 value to the function. The attribute implies that a hidden copy of
697 the pointee is made between the caller and the callee, so the callee
698 is unable to modify the value in the caller. This attribute is only
699 valid on LLVM pointer arguments. It is generally used to pass
700 structs and arrays by value, but is also valid on pointers to
701 scalars. The copy is considered to belong to the caller not the
702 callee (for example, ``readonly`` functions should not write to
703 ``byval`` parameters). This is not a valid attribute for return
706 The byval attribute also supports specifying an alignment with the
707 align attribute. It indicates the alignment of the stack slot to
708 form and the known alignment of the pointer specified to the call
709 site. If the alignment is not specified, then the code generator
710 makes a target-specific assumption.
716 .. Warning:: This feature is unstable and not fully implemented.
718 The ``inalloca`` argument attribute allows the caller to get the
719 address of an outgoing argument to a ``call`` or ``invoke`` before
720 it executes. It is similar to ``byval`` in that it is used to pass
721 arguments by value, but it guarantees that the argument will not be
724 To be :ref:`well formed <wellformed>`, the caller must pass in an
725 alloca value into an ``inalloca`` parameter, and an alloca may be
726 used as an ``inalloca`` argument at most once. The attribute can
727 only be applied to parameters that would be passed in memory and not
728 registers. The ``inalloca`` attribute cannot be used in conjunction
729 with other attributes that affect argument storage, like ``inreg``,
730 ``nest``, ``sret``, or ``byval``. The ``inalloca`` stack space is
731 considered to be clobbered by any call that uses it, so any
732 ``inalloca`` parameters cannot be marked ``readonly``.
734 Allocas passed with ``inalloca`` to a call must be in the opposite
735 order of the parameter list, meaning that the rightmost argument
736 must be allocated first. If a call has inalloca arguments, no other
737 allocas can occur between the first alloca used by the call and the
738 call site, unless they are are cleared by calls to
739 :ref:`llvm.stackrestore <int_stackrestore>`. Violating these rules
740 results in undefined behavior at runtime.
742 See :doc:`InAlloca` for more information on how to use this
746 This indicates that the pointer parameter specifies the address of a
747 structure that is the return value of the function in the source
748 program. This pointer must be guaranteed by the caller to be valid:
749 loads and stores to the structure may be assumed by the callee
750 not to trap and to be properly aligned. This may only be applied to
751 the first parameter. This is not a valid attribute for return
754 This indicates that pointer values :ref:`based <pointeraliasing>` on
755 the argument or return value do not alias pointer values which are
756 not *based* on it, ignoring certain "irrelevant" dependencies. For a
757 call to the parent function, dependencies between memory references
758 from before or after the call and from those during the call are
759 "irrelevant" to the ``noalias`` keyword for the arguments and return
760 value used in that call. The caller shares the responsibility with
761 the callee for ensuring that these requirements are met. For further
762 details, please see the discussion of the NoAlias response in `alias
763 analysis <AliasAnalysis.html#MustMayNo>`_.
765 Note that this definition of ``noalias`` is intentionally similar
766 to the definition of ``restrict`` in C99 for function arguments,
767 though it is slightly weaker.
769 For function return values, C99's ``restrict`` is not meaningful,
770 while LLVM's ``noalias`` is.
772 This indicates that the callee does not make any copies of the
773 pointer that outlive the callee itself. This is not a valid
774 attribute for return values.
779 This indicates that the pointer parameter can be excised using the
780 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
781 attribute for return values and can only be applied to one parameter.
784 This indicates that the function always returns the argument as its return
785 value. This is an optimization hint to the code generator when generating
786 the caller, allowing tail call optimization and omission of register saves
787 and restores in some cases; it is not checked or enforced when generating
788 the callee. The parameter and the function return type must be valid
789 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
790 valid attribute for return values and can only be applied to one parameter.
794 Garbage Collector Names
795 -----------------------
797 Each function may specify a garbage collector name, which is simply a
802 define void @f() gc "name" { ... }
804 The compiler declares the supported values of *name*. Specifying a
805 collector which will cause the compiler to alter its output in order to
806 support the named garbage collection algorithm.
813 Prefix data is data associated with a function which the code generator
814 will emit immediately before the function body. The purpose of this feature
815 is to allow frontends to associate language-specific runtime metadata with
816 specific functions and make it available through the function pointer while
817 still allowing the function pointer to be called. To access the data for a
818 given function, a program may bitcast the function pointer to a pointer to
819 the constant's type. This implies that the IR symbol points to the start
822 To maintain the semantics of ordinary function calls, the prefix data must
823 have a particular format. Specifically, it must begin with a sequence of
824 bytes which decode to a sequence of machine instructions, valid for the
825 module's target, which transfer control to the point immediately succeeding
826 the prefix data, without performing any other visible action. This allows
827 the inliner and other passes to reason about the semantics of the function
828 definition without needing to reason about the prefix data. Obviously this
829 makes the format of the prefix data highly target dependent.
831 Prefix data is laid out as if it were an initializer for a global variable
832 of the prefix data's type. No padding is automatically placed between the
833 prefix data and the function body. If padding is required, it must be part
836 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
837 which encodes the ``nop`` instruction:
841 define void @f() prefix i8 144 { ... }
843 Generally prefix data can be formed by encoding a relative branch instruction
844 which skips the metadata, as in this example of valid prefix data for the
845 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
849 %0 = type <{ i8, i8, i8* }>
851 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
853 A function may have prefix data but no body. This has similar semantics
854 to the ``available_externally`` linkage in that the data may be used by the
855 optimizers but will not be emitted in the object file.
862 Attribute groups are groups of attributes that are referenced by objects within
863 the IR. They are important for keeping ``.ll`` files readable, because a lot of
864 functions will use the same set of attributes. In the degenerative case of a
865 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
866 group will capture the important command line flags used to build that file.
868 An attribute group is a module-level object. To use an attribute group, an
869 object references the attribute group's ID (e.g. ``#37``). An object may refer
870 to more than one attribute group. In that situation, the attributes from the
871 different groups are merged.
873 Here is an example of attribute groups for a function that should always be
874 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
878 ; Target-independent attributes:
879 attributes #0 = { alwaysinline alignstack=4 }
881 ; Target-dependent attributes:
882 attributes #1 = { "no-sse" }
884 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
885 define void @f() #0 #1 { ... }
892 Function attributes are set to communicate additional information about
893 a function. Function attributes are considered to be part of the
894 function, not of the function type, so functions with different function
895 attributes can have the same function type.
897 Function attributes are simple keywords that follow the type specified.
898 If multiple attributes are needed, they are space separated. For
903 define void @f() noinline { ... }
904 define void @f() alwaysinline { ... }
905 define void @f() alwaysinline optsize { ... }
906 define void @f() optsize { ... }
909 This attribute indicates that, when emitting the prologue and
910 epilogue, the backend should forcibly align the stack pointer.
911 Specify the desired alignment, which must be a power of two, in
914 This attribute indicates that the inliner should attempt to inline
915 this function into callers whenever possible, ignoring any active
916 inlining size threshold for this caller.
918 This indicates that the callee function at a call site should be
919 recognized as a built-in function, even though the function's declaration
920 uses the ``nobuiltin`` attribute. This is only valid at call sites for
921 direct calls to functions which are declared with the ``nobuiltin``
924 This attribute indicates that this function is rarely called. When
925 computing edge weights, basic blocks post-dominated by a cold
926 function call are also considered to be cold; and, thus, given low
929 This attribute indicates that the source code contained a hint that
930 inlining this function is desirable (such as the "inline" keyword in
931 C/C++). It is just a hint; it imposes no requirements on the
934 This attribute suggests that optimization passes and code generator
935 passes make choices that keep the code size of this function as small
936 as possible and perform optimizations that may sacrifice runtime
937 performance in order to minimize the size of the generated code.
939 This attribute disables prologue / epilogue emission for the
940 function. This can have very system-specific consequences.
942 This indicates that the callee function at a call site is not recognized as
943 a built-in function. LLVM will retain the original call and not replace it
944 with equivalent code based on the semantics of the built-in function, unless
945 the call site uses the ``builtin`` attribute. This is valid at call sites
946 and on function declarations and definitions.
948 This attribute indicates that calls to the function cannot be
949 duplicated. A call to a ``noduplicate`` function may be moved
950 within its parent function, but may not be duplicated within
953 A function containing a ``noduplicate`` call may still
954 be an inlining candidate, provided that the call is not
955 duplicated by inlining. That implies that the function has
956 internal linkage and only has one call site, so the original
957 call is dead after inlining.
959 This attributes disables implicit floating point instructions.
961 This attribute indicates that the inliner should never inline this
962 function in any situation. This attribute may not be used together
963 with the ``alwaysinline`` attribute.
965 This attribute suppresses lazy symbol binding for the function. This
966 may make calls to the function faster, at the cost of extra program
967 startup time if the function is not called during program startup.
969 This attribute indicates that the code generator should not use a
970 red zone, even if the target-specific ABI normally permits it.
972 This function attribute indicates that the function never returns
973 normally. This produces undefined behavior at runtime if the
974 function ever does dynamically return.
976 This function attribute indicates that the function never returns
977 with an unwind or exceptional control flow. If the function does
978 unwind, its runtime behavior is undefined.
980 This function attribute indicates that the function is not optimized
981 by any optimization or code generator passes with the
982 exception of interprocedural optimization passes.
983 This attribute cannot be used together with the ``alwaysinline``
984 attribute; this attribute is also incompatible
985 with the ``minsize`` attribute and the ``optsize`` attribute.
987 This attribute requires the ``noinline`` attribute to be specified on
988 the function as well, so the function is never inlined into any caller.
989 Only functions with the ``alwaysinline`` attribute are valid
990 candidates for inlining into the body of this function.
992 This attribute suggests that optimization passes and code generator
993 passes make choices that keep the code size of this function low,
994 and otherwise do optimizations specifically to reduce code size as
995 long as they do not significantly impact runtime performance.
997 On a function, this attribute indicates that the function computes its
998 result (or decides to unwind an exception) based strictly on its arguments,
999 without dereferencing any pointer arguments or otherwise accessing
1000 any mutable state (e.g. memory, control registers, etc) visible to
1001 caller functions. It does not write through any pointer arguments
1002 (including ``byval`` arguments) and never changes any state visible
1003 to callers. This means that it cannot unwind exceptions by calling
1004 the ``C++`` exception throwing methods.
1006 On an argument, this attribute indicates that the function does not
1007 dereference that pointer argument, even though it may read or write the
1008 memory that the pointer points to if accessed through other pointers.
1010 On a function, this attribute indicates that the function does not write
1011 through any pointer arguments (including ``byval`` arguments) or otherwise
1012 modify any state (e.g. memory, control registers, etc) visible to
1013 caller functions. It may dereference pointer arguments and read
1014 state that may be set in the caller. A readonly function always
1015 returns the same value (or unwinds an exception identically) when
1016 called with the same set of arguments and global state. It cannot
1017 unwind an exception by calling the ``C++`` exception throwing
1020 On an argument, this attribute indicates that the function does not write
1021 through this pointer argument, even though it may write to the memory that
1022 the pointer points to.
1024 This attribute indicates that this function can return twice. The C
1025 ``setjmp`` is an example of such a function. The compiler disables
1026 some optimizations (like tail calls) in the caller of these
1028 ``sanitize_address``
1029 This attribute indicates that AddressSanitizer checks
1030 (dynamic address safety analysis) are enabled for this function.
1032 This attribute indicates that MemorySanitizer checks (dynamic detection
1033 of accesses to uninitialized memory) are enabled for this function.
1035 This attribute indicates that ThreadSanitizer checks
1036 (dynamic thread safety analysis) are enabled for this function.
1038 This attribute indicates that the function should emit a stack
1039 smashing protector. It is in the form of a "canary" --- a random value
1040 placed on the stack before the local variables that's checked upon
1041 return from the function to see if it has been overwritten. A
1042 heuristic is used to determine if a function needs stack protectors
1043 or not. The heuristic used will enable protectors for functions with:
1045 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1046 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1047 - Calls to alloca() with variable sizes or constant sizes greater than
1048 ``ssp-buffer-size``.
1050 If a function that has an ``ssp`` attribute is inlined into a
1051 function that doesn't have an ``ssp`` attribute, then the resulting
1052 function will have an ``ssp`` attribute.
1054 This attribute indicates that the function should *always* emit a
1055 stack smashing protector. This overrides the ``ssp`` function
1058 If a function that has an ``sspreq`` attribute is inlined into a
1059 function that doesn't have an ``sspreq`` attribute or which has an
1060 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1061 an ``sspreq`` attribute.
1063 This attribute indicates that the function should emit a stack smashing
1064 protector. This attribute causes a strong heuristic to be used when
1065 determining if a function needs stack protectors. The strong heuristic
1066 will enable protectors for functions with:
1068 - Arrays of any size and type
1069 - Aggregates containing an array of any size and type.
1070 - Calls to alloca().
1071 - Local variables that have had their address taken.
1073 This overrides the ``ssp`` function attribute.
1075 If a function that has an ``sspstrong`` attribute is inlined into a
1076 function that doesn't have an ``sspstrong`` attribute, then the
1077 resulting function will have an ``sspstrong`` attribute.
1079 This attribute indicates that the ABI being targeted requires that
1080 an unwind table entry be produce for this function even if we can
1081 show that no exceptions passes by it. This is normally the case for
1082 the ELF x86-64 abi, but it can be disabled for some compilation
1087 Module-Level Inline Assembly
1088 ----------------------------
1090 Modules may contain "module-level inline asm" blocks, which corresponds
1091 to the GCC "file scope inline asm" blocks. These blocks are internally
1092 concatenated by LLVM and treated as a single unit, but may be separated
1093 in the ``.ll`` file if desired. The syntax is very simple:
1095 .. code-block:: llvm
1097 module asm "inline asm code goes here"
1098 module asm "more can go here"
1100 The strings can contain any character by escaping non-printable
1101 characters. The escape sequence used is simply "\\xx" where "xx" is the
1102 two digit hex code for the number.
1104 The inline asm code is simply printed to the machine code .s file when
1105 assembly code is generated.
1107 .. _langref_datalayout:
1112 A module may specify a target specific data layout string that specifies
1113 how data is to be laid out in memory. The syntax for the data layout is
1116 .. code-block:: llvm
1118 target datalayout = "layout specification"
1120 The *layout specification* consists of a list of specifications
1121 separated by the minus sign character ('-'). Each specification starts
1122 with a letter and may include other information after the letter to
1123 define some aspect of the data layout. The specifications accepted are
1127 Specifies that the target lays out data in big-endian form. That is,
1128 the bits with the most significance have the lowest address
1131 Specifies that the target lays out data in little-endian form. That
1132 is, the bits with the least significance have the lowest address
1135 Specifies the natural alignment of the stack in bits. Alignment
1136 promotion of stack variables is limited to the natural stack
1137 alignment to avoid dynamic stack realignment. The stack alignment
1138 must be a multiple of 8-bits. If omitted, the natural stack
1139 alignment defaults to "unspecified", which does not prevent any
1140 alignment promotions.
1141 ``p[n]:<size>:<abi>:<pref>``
1142 This specifies the *size* of a pointer and its ``<abi>`` and
1143 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1144 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1145 preceding ``:`` should be omitted too. The address space, ``n`` is
1146 optional, and if not specified, denotes the default address space 0.
1147 The value of ``n`` must be in the range [1,2^23).
1148 ``i<size>:<abi>:<pref>``
1149 This specifies the alignment for an integer type of a given bit
1150 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1151 ``v<size>:<abi>:<pref>``
1152 This specifies the alignment for a vector type of a given bit
1154 ``f<size>:<abi>:<pref>``
1155 This specifies the alignment for a floating point type of a given bit
1156 ``<size>``. Only values of ``<size>`` that are supported by the target
1157 will work. 32 (float) and 64 (double) are supported on all targets; 80
1158 or 128 (different flavors of long double) are also supported on some
1160 ``a<size>:<abi>:<pref>``
1161 This specifies the alignment for an aggregate type of a given bit
1163 ``n<size1>:<size2>:<size3>...``
1164 This specifies a set of native integer widths for the target CPU in
1165 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1166 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1167 this set are considered to support most general arithmetic operations
1170 When constructing the data layout for a given target, LLVM starts with a
1171 default set of specifications which are then (possibly) overridden by
1172 the specifications in the ``datalayout`` keyword. The default
1173 specifications are given in this list:
1175 - ``E`` - big endian
1176 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1177 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1178 same as the default address space.
1179 - ``S0`` - natural stack alignment is unspecified
1180 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1181 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1182 - ``i16:16:16`` - i16 is 16-bit aligned
1183 - ``i32:32:32`` - i32 is 32-bit aligned
1184 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1185 alignment of 64-bits
1186 - ``f16:16:16`` - half is 16-bit aligned
1187 - ``f32:32:32`` - float is 32-bit aligned
1188 - ``f64:64:64`` - double is 64-bit aligned
1189 - ``f128:128:128`` - quad is 128-bit aligned
1190 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1191 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1192 - ``a:0:64`` - aggregates are 64-bit aligned
1194 When LLVM is determining the alignment for a given type, it uses the
1197 #. If the type sought is an exact match for one of the specifications,
1198 that specification is used.
1199 #. If no match is found, and the type sought is an integer type, then
1200 the smallest integer type that is larger than the bitwidth of the
1201 sought type is used. If none of the specifications are larger than
1202 the bitwidth then the largest integer type is used. For example,
1203 given the default specifications above, the i7 type will use the
1204 alignment of i8 (next largest) while both i65 and i256 will use the
1205 alignment of i64 (largest specified).
1206 #. If no match is found, and the type sought is a vector type, then the
1207 largest vector type that is smaller than the sought vector type will
1208 be used as a fall back. This happens because <128 x double> can be
1209 implemented in terms of 64 <2 x double>, for example.
1211 The function of the data layout string may not be what you expect.
1212 Notably, this is not a specification from the frontend of what alignment
1213 the code generator should use.
1215 Instead, if specified, the target data layout is required to match what
1216 the ultimate *code generator* expects. This string is used by the
1217 mid-level optimizers to improve code, and this only works if it matches
1218 what the ultimate code generator uses. If you would like to generate IR
1219 that does not embed this target-specific detail into the IR, then you
1220 don't have to specify the string. This will disable some optimizations
1221 that require precise layout information, but this also prevents those
1222 optimizations from introducing target specificity into the IR.
1229 A module may specify a target triple string that describes the target
1230 host. The syntax for the target triple is simply:
1232 .. code-block:: llvm
1234 target triple = "x86_64-apple-macosx10.7.0"
1236 The *target triple* string consists of a series of identifiers delimited
1237 by the minus sign character ('-'). The canonical forms are:
1241 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1242 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1244 This information is passed along to the backend so that it generates
1245 code for the proper architecture. It's possible to override this on the
1246 command line with the ``-mtriple`` command line option.
1248 .. _pointeraliasing:
1250 Pointer Aliasing Rules
1251 ----------------------
1253 Any memory access must be done through a pointer value associated with
1254 an address range of the memory access, otherwise the behavior is
1255 undefined. Pointer values are associated with address ranges according
1256 to the following rules:
1258 - A pointer value is associated with the addresses associated with any
1259 value it is *based* on.
1260 - An address of a global variable is associated with the address range
1261 of the variable's storage.
1262 - The result value of an allocation instruction is associated with the
1263 address range of the allocated storage.
1264 - A null pointer in the default address-space is associated with no
1266 - An integer constant other than zero or a pointer value returned from
1267 a function not defined within LLVM may be associated with address
1268 ranges allocated through mechanisms other than those provided by
1269 LLVM. Such ranges shall not overlap with any ranges of addresses
1270 allocated by mechanisms provided by LLVM.
1272 A pointer value is *based* on another pointer value according to the
1275 - A pointer value formed from a ``getelementptr`` operation is *based*
1276 on the first operand of the ``getelementptr``.
1277 - The result value of a ``bitcast`` is *based* on the operand of the
1279 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1280 values that contribute (directly or indirectly) to the computation of
1281 the pointer's value.
1282 - The "*based* on" relationship is transitive.
1284 Note that this definition of *"based"* is intentionally similar to the
1285 definition of *"based"* in C99, though it is slightly weaker.
1287 LLVM IR does not associate types with memory. The result type of a
1288 ``load`` merely indicates the size and alignment of the memory from
1289 which to load, as well as the interpretation of the value. The first
1290 operand type of a ``store`` similarly only indicates the size and
1291 alignment of the store.
1293 Consequently, type-based alias analysis, aka TBAA, aka
1294 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1295 :ref:`Metadata <metadata>` may be used to encode additional information
1296 which specialized optimization passes may use to implement type-based
1301 Volatile Memory Accesses
1302 ------------------------
1304 Certain memory accesses, such as :ref:`load <i_load>`'s,
1305 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1306 marked ``volatile``. The optimizers must not change the number of
1307 volatile operations or change their order of execution relative to other
1308 volatile operations. The optimizers *may* change the order of volatile
1309 operations relative to non-volatile operations. This is not Java's
1310 "volatile" and has no cross-thread synchronization behavior.
1312 IR-level volatile loads and stores cannot safely be optimized into
1313 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1314 flagged volatile. Likewise, the backend should never split or merge
1315 target-legal volatile load/store instructions.
1317 .. admonition:: Rationale
1319 Platforms may rely on volatile loads and stores of natively supported
1320 data width to be executed as single instruction. For example, in C
1321 this holds for an l-value of volatile primitive type with native
1322 hardware support, but not necessarily for aggregate types. The
1323 frontend upholds these expectations, which are intentionally
1324 unspecified in the IR. The rules above ensure that IR transformation
1325 do not violate the frontend's contract with the language.
1329 Memory Model for Concurrent Operations
1330 --------------------------------------
1332 The LLVM IR does not define any way to start parallel threads of
1333 execution or to register signal handlers. Nonetheless, there are
1334 platform-specific ways to create them, and we define LLVM IR's behavior
1335 in their presence. This model is inspired by the C++0x memory model.
1337 For a more informal introduction to this model, see the :doc:`Atomics`.
1339 We define a *happens-before* partial order as the least partial order
1342 - Is a superset of single-thread program order, and
1343 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1344 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1345 techniques, like pthread locks, thread creation, thread joining,
1346 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1347 Constraints <ordering>`).
1349 Note that program order does not introduce *happens-before* edges
1350 between a thread and signals executing inside that thread.
1352 Every (defined) read operation (load instructions, memcpy, atomic
1353 loads/read-modify-writes, etc.) R reads a series of bytes written by
1354 (defined) write operations (store instructions, atomic
1355 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1356 section, initialized globals are considered to have a write of the
1357 initializer which is atomic and happens before any other read or write
1358 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1359 may see any write to the same byte, except:
1361 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1362 write\ :sub:`2` happens before R\ :sub:`byte`, then
1363 R\ :sub:`byte` does not see write\ :sub:`1`.
1364 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1365 R\ :sub:`byte` does not see write\ :sub:`3`.
1367 Given that definition, R\ :sub:`byte` is defined as follows:
1369 - If R is volatile, the result is target-dependent. (Volatile is
1370 supposed to give guarantees which can support ``sig_atomic_t`` in
1371 C/C++, and may be used for accesses to addresses which do not behave
1372 like normal memory. It does not generally provide cross-thread
1374 - Otherwise, if there is no write to the same byte that happens before
1375 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1376 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1377 R\ :sub:`byte` returns the value written by that write.
1378 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1379 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1380 Memory Ordering Constraints <ordering>` section for additional
1381 constraints on how the choice is made.
1382 - Otherwise R\ :sub:`byte` returns ``undef``.
1384 R returns the value composed of the series of bytes it read. This
1385 implies that some bytes within the value may be ``undef`` **without**
1386 the entire value being ``undef``. Note that this only defines the
1387 semantics of the operation; it doesn't mean that targets will emit more
1388 than one instruction to read the series of bytes.
1390 Note that in cases where none of the atomic intrinsics are used, this
1391 model places only one restriction on IR transformations on top of what
1392 is required for single-threaded execution: introducing a store to a byte
1393 which might not otherwise be stored is not allowed in general.
1394 (Specifically, in the case where another thread might write to and read
1395 from an address, introducing a store can change a load that may see
1396 exactly one write into a load that may see multiple writes.)
1400 Atomic Memory Ordering Constraints
1401 ----------------------------------
1403 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1404 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1405 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1406 an ordering parameter that determines which other atomic instructions on
1407 the same address they *synchronize with*. These semantics are borrowed
1408 from Java and C++0x, but are somewhat more colloquial. If these
1409 descriptions aren't precise enough, check those specs (see spec
1410 references in the :doc:`atomics guide <Atomics>`).
1411 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1412 differently since they don't take an address. See that instruction's
1413 documentation for details.
1415 For a simpler introduction to the ordering constraints, see the
1419 The set of values that can be read is governed by the happens-before
1420 partial order. A value cannot be read unless some operation wrote
1421 it. This is intended to provide a guarantee strong enough to model
1422 Java's non-volatile shared variables. This ordering cannot be
1423 specified for read-modify-write operations; it is not strong enough
1424 to make them atomic in any interesting way.
1426 In addition to the guarantees of ``unordered``, there is a single
1427 total order for modifications by ``monotonic`` operations on each
1428 address. All modification orders must be compatible with the
1429 happens-before order. There is no guarantee that the modification
1430 orders can be combined to a global total order for the whole program
1431 (and this often will not be possible). The read in an atomic
1432 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1433 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1434 order immediately before the value it writes. If one atomic read
1435 happens before another atomic read of the same address, the later
1436 read must see the same value or a later value in the address's
1437 modification order. This disallows reordering of ``monotonic`` (or
1438 stronger) operations on the same address. If an address is written
1439 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1440 read that address repeatedly, the other threads must eventually see
1441 the write. This corresponds to the C++0x/C1x
1442 ``memory_order_relaxed``.
1444 In addition to the guarantees of ``monotonic``, a
1445 *synchronizes-with* edge may be formed with a ``release`` operation.
1446 This is intended to model C++'s ``memory_order_acquire``.
1448 In addition to the guarantees of ``monotonic``, if this operation
1449 writes a value which is subsequently read by an ``acquire``
1450 operation, it *synchronizes-with* that operation. (This isn't a
1451 complete description; see the C++0x definition of a release
1452 sequence.) This corresponds to the C++0x/C1x
1453 ``memory_order_release``.
1454 ``acq_rel`` (acquire+release)
1455 Acts as both an ``acquire`` and ``release`` operation on its
1456 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1457 ``seq_cst`` (sequentially consistent)
1458 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1459 operation which only reads, ``release`` for an operation which only
1460 writes), there is a global total order on all
1461 sequentially-consistent operations on all addresses, which is
1462 consistent with the *happens-before* partial order and with the
1463 modification orders of all the affected addresses. Each
1464 sequentially-consistent read sees the last preceding write to the
1465 same address in this global order. This corresponds to the C++0x/C1x
1466 ``memory_order_seq_cst`` and Java volatile.
1470 If an atomic operation is marked ``singlethread``, it only *synchronizes
1471 with* or participates in modification and seq\_cst total orderings with
1472 other operations running in the same thread (for example, in signal
1480 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1481 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1482 :ref:`frem <i_frem>`) have the following flags that can set to enable
1483 otherwise unsafe floating point operations
1486 No NaNs - Allow optimizations to assume the arguments and result are not
1487 NaN. Such optimizations are required to retain defined behavior over
1488 NaNs, but the value of the result is undefined.
1491 No Infs - Allow optimizations to assume the arguments and result are not
1492 +/-Inf. Such optimizations are required to retain defined behavior over
1493 +/-Inf, but the value of the result is undefined.
1496 No Signed Zeros - Allow optimizations to treat the sign of a zero
1497 argument or result as insignificant.
1500 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1501 argument rather than perform division.
1504 Fast - Allow algebraically equivalent transformations that may
1505 dramatically change results in floating point (e.g. reassociate). This
1506 flag implies all the others.
1513 The LLVM type system is one of the most important features of the
1514 intermediate representation. Being typed enables a number of
1515 optimizations to be performed on the intermediate representation
1516 directly, without having to do extra analyses on the side before the
1517 transformation. A strong type system makes it easier to read the
1518 generated code and enables novel analyses and transformations that are
1519 not feasible to perform on normal three address code representations.
1529 The void type does not represent any value and has no size.
1547 The function type can be thought of as a function signature. It consists of a
1548 return type and a list of formal parameter types. The return type of a function
1549 type is a void type or first class type --- except for :ref:`label <t_label>`
1550 and :ref:`metadata <t_metadata>` types.
1556 <returntype> (<parameter list>)
1558 ...where '``<parameter list>``' is a comma-separated list of type
1559 specifiers. Optionally, the parameter list may include a type ``...``, which
1560 indicates that the function takes a variable number of arguments. Variable
1561 argument functions can access their arguments with the :ref:`variable argument
1562 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1563 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1567 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1568 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1569 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1570 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1571 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1572 | ``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. |
1573 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1574 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1575 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1582 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1583 Values of these types are the only ones which can be produced by
1591 These are the types that are valid in registers from CodeGen's perspective.
1600 The integer type is a very simple type that simply specifies an
1601 arbitrary bit width for the integer type desired. Any bit width from 1
1602 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1610 The number of bits the integer will occupy is specified by the ``N``
1616 +----------------+------------------------------------------------+
1617 | ``i1`` | a single-bit integer. |
1618 +----------------+------------------------------------------------+
1619 | ``i32`` | a 32-bit integer. |
1620 +----------------+------------------------------------------------+
1621 | ``i1942652`` | a really big integer of over 1 million bits. |
1622 +----------------+------------------------------------------------+
1626 Floating Point Types
1627 """"""""""""""""""""
1636 - 16-bit floating point value
1639 - 32-bit floating point value
1642 - 64-bit floating point value
1645 - 128-bit floating point value (112-bit mantissa)
1648 - 80-bit floating point value (X87)
1651 - 128-bit floating point value (two 64-bits)
1660 The x86mmx type represents a value held in an MMX register on an x86
1661 machine. The operations allowed on it are quite limited: parameters and
1662 return values, load and store, and bitcast. User-specified MMX
1663 instructions are represented as intrinsic or asm calls with arguments
1664 and/or results of this type. There are no arrays, vectors or constants
1681 The pointer type is used to specify memory locations. Pointers are
1682 commonly used to reference objects in memory.
1684 Pointer types may have an optional address space attribute defining the
1685 numbered address space where the pointed-to object resides. The default
1686 address space is number zero. The semantics of non-zero address spaces
1687 are target-specific.
1689 Note that LLVM does not permit pointers to void (``void*``) nor does it
1690 permit pointers to labels (``label*``). Use ``i8*`` instead.
1700 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1701 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1702 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1703 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1704 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1705 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1706 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1715 A vector type is a simple derived type that represents a vector of
1716 elements. Vector types are used when multiple primitive data are
1717 operated in parallel using a single instruction (SIMD). A vector type
1718 requires a size (number of elements) and an underlying primitive data
1719 type. Vector types are considered :ref:`first class <t_firstclass>`.
1725 < <# elements> x <elementtype> >
1727 The number of elements is a constant integer value larger than 0;
1728 elementtype may be any integer or floating point type, or a pointer to
1729 these types. Vectors of size zero are not allowed.
1733 +-------------------+--------------------------------------------------+
1734 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1735 +-------------------+--------------------------------------------------+
1736 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1737 +-------------------+--------------------------------------------------+
1738 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1739 +-------------------+--------------------------------------------------+
1740 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1741 +-------------------+--------------------------------------------------+
1750 The label type represents code labels.
1765 The metadata type represents embedded metadata. No derived types may be
1766 created from metadata except for :ref:`function <t_function>` arguments.
1779 Aggregate Types are a subset of derived types that can contain multiple
1780 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1781 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1791 The array type is a very simple derived type that arranges elements
1792 sequentially in memory. The array type requires a size (number of
1793 elements) and an underlying data type.
1799 [<# elements> x <elementtype>]
1801 The number of elements is a constant integer value; ``elementtype`` may
1802 be any type with a size.
1806 +------------------+--------------------------------------+
1807 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1808 +------------------+--------------------------------------+
1809 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1810 +------------------+--------------------------------------+
1811 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1812 +------------------+--------------------------------------+
1814 Here are some examples of multidimensional arrays:
1816 +-----------------------------+----------------------------------------------------------+
1817 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1818 +-----------------------------+----------------------------------------------------------+
1819 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1820 +-----------------------------+----------------------------------------------------------+
1821 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1822 +-----------------------------+----------------------------------------------------------+
1824 There is no restriction on indexing beyond the end of the array implied
1825 by a static type (though there are restrictions on indexing beyond the
1826 bounds of an allocated object in some cases). This means that
1827 single-dimension 'variable sized array' addressing can be implemented in
1828 LLVM with a zero length array type. An implementation of 'pascal style
1829 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1839 The structure type is used to represent a collection of data members
1840 together in memory. The elements of a structure may be any type that has
1843 Structures in memory are accessed using '``load``' and '``store``' by
1844 getting a pointer to a field with the '``getelementptr``' instruction.
1845 Structures in registers are accessed using the '``extractvalue``' and
1846 '``insertvalue``' instructions.
1848 Structures may optionally be "packed" structures, which indicate that
1849 the alignment of the struct is one byte, and that there is no padding
1850 between the elements. In non-packed structs, padding between field types
1851 is inserted as defined by the DataLayout string in the module, which is
1852 required to match what the underlying code generator expects.
1854 Structures can either be "literal" or "identified". A literal structure
1855 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1856 identified types are always defined at the top level with a name.
1857 Literal types are uniqued by their contents and can never be recursive
1858 or opaque since there is no way to write one. Identified types can be
1859 recursive, can be opaqued, and are never uniqued.
1865 %T1 = type { <type list> } ; Identified normal struct type
1866 %T2 = type <{ <type list> }> ; Identified packed struct type
1870 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1871 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1872 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1873 | ``{ 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``. |
1874 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1875 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1876 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1880 Opaque Structure Types
1881 """"""""""""""""""""""
1885 Opaque structure types are used to represent named structure types that
1886 do not have a body specified. This corresponds (for example) to the C
1887 notion of a forward declared structure.
1898 +--------------+-------------------+
1899 | ``opaque`` | An opaque type. |
1900 +--------------+-------------------+
1905 LLVM has several different basic types of constants. This section
1906 describes them all and their syntax.
1911 **Boolean constants**
1912 The two strings '``true``' and '``false``' are both valid constants
1914 **Integer constants**
1915 Standard integers (such as '4') are constants of the
1916 :ref:`integer <t_integer>` type. Negative numbers may be used with
1918 **Floating point constants**
1919 Floating point constants use standard decimal notation (e.g.
1920 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1921 hexadecimal notation (see below). The assembler requires the exact
1922 decimal value of a floating-point constant. For example, the
1923 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1924 decimal in binary. Floating point constants must have a :ref:`floating
1925 point <t_floating>` type.
1926 **Null pointer constants**
1927 The identifier '``null``' is recognized as a null pointer constant
1928 and must be of :ref:`pointer type <t_pointer>`.
1930 The one non-intuitive notation for constants is the hexadecimal form of
1931 floating point constants. For example, the form
1932 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1933 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1934 constants are required (and the only time that they are generated by the
1935 disassembler) is when a floating point constant must be emitted but it
1936 cannot be represented as a decimal floating point number in a reasonable
1937 number of digits. For example, NaN's, infinities, and other special
1938 values are represented in their IEEE hexadecimal format so that assembly
1939 and disassembly do not cause any bits to change in the constants.
1941 When using the hexadecimal form, constants of types half, float, and
1942 double are represented using the 16-digit form shown above (which
1943 matches the IEEE754 representation for double); half and float values
1944 must, however, be exactly representable as IEEE 754 half and single
1945 precision, respectively. Hexadecimal format is always used for long
1946 double, and there are three forms of long double. The 80-bit format used
1947 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1948 128-bit format used by PowerPC (two adjacent doubles) is represented by
1949 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1950 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1951 will only work if they match the long double format on your target.
1952 The IEEE 16-bit format (half precision) is represented by ``0xH``
1953 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1954 (sign bit at the left).
1956 There are no constants of type x86mmx.
1958 .. _complexconstants:
1963 Complex constants are a (potentially recursive) combination of simple
1964 constants and smaller complex constants.
1966 **Structure constants**
1967 Structure constants are represented with notation similar to
1968 structure type definitions (a comma separated list of elements,
1969 surrounded by braces (``{}``)). For example:
1970 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1971 "``@G = external global i32``". Structure constants must have
1972 :ref:`structure type <t_struct>`, and the number and types of elements
1973 must match those specified by the type.
1975 Array constants are represented with notation similar to array type
1976 definitions (a comma separated list of elements, surrounded by
1977 square brackets (``[]``)). For example:
1978 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1979 :ref:`array type <t_array>`, and the number and types of elements must
1980 match those specified by the type.
1981 **Vector constants**
1982 Vector constants are represented with notation similar to vector
1983 type definitions (a comma separated list of elements, surrounded by
1984 less-than/greater-than's (``<>``)). For example:
1985 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1986 must have :ref:`vector type <t_vector>`, and the number and types of
1987 elements must match those specified by the type.
1988 **Zero initialization**
1989 The string '``zeroinitializer``' can be used to zero initialize a
1990 value to zero of *any* type, including scalar and
1991 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1992 having to print large zero initializers (e.g. for large arrays) and
1993 is always exactly equivalent to using explicit zero initializers.
1995 A metadata node is a structure-like constant with :ref:`metadata
1996 type <t_metadata>`. For example:
1997 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1998 constants that are meant to be interpreted as part of the
1999 instruction stream, metadata is a place to attach additional
2000 information such as debug info.
2002 Global Variable and Function Addresses
2003 --------------------------------------
2005 The addresses of :ref:`global variables <globalvars>` and
2006 :ref:`functions <functionstructure>` are always implicitly valid
2007 (link-time) constants. These constants are explicitly referenced when
2008 the :ref:`identifier for the global <identifiers>` is used and always have
2009 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2012 .. code-block:: llvm
2016 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2023 The string '``undef``' can be used anywhere a constant is expected, and
2024 indicates that the user of the value may receive an unspecified
2025 bit-pattern. Undefined values may be of any type (other than '``label``'
2026 or '``void``') and be used anywhere a constant is permitted.
2028 Undefined values are useful because they indicate to the compiler that
2029 the program is well defined no matter what value is used. This gives the
2030 compiler more freedom to optimize. Here are some examples of
2031 (potentially surprising) transformations that are valid (in pseudo IR):
2033 .. code-block:: llvm
2043 This is safe because all of the output bits are affected by the undef
2044 bits. Any output bit can have a zero or one depending on the input bits.
2046 .. code-block:: llvm
2057 These logical operations have bits that are not always affected by the
2058 input. For example, if ``%X`` has a zero bit, then the output of the
2059 '``and``' operation will always be a zero for that bit, no matter what
2060 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2061 optimize or assume that the result of the '``and``' is '``undef``'.
2062 However, it is safe to assume that all bits of the '``undef``' could be
2063 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2064 all the bits of the '``undef``' operand to the '``or``' could be set,
2065 allowing the '``or``' to be folded to -1.
2067 .. code-block:: llvm
2069 %A = select undef, %X, %Y
2070 %B = select undef, 42, %Y
2071 %C = select %X, %Y, undef
2081 This set of examples shows that undefined '``select``' (and conditional
2082 branch) conditions can go *either way*, but they have to come from one
2083 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2084 both known to have a clear low bit, then ``%A`` would have to have a
2085 cleared low bit. However, in the ``%C`` example, the optimizer is
2086 allowed to assume that the '``undef``' operand could be the same as
2087 ``%Y``, allowing the whole '``select``' to be eliminated.
2089 .. code-block:: llvm
2091 %A = xor undef, undef
2108 This example points out that two '``undef``' operands are not
2109 necessarily the same. This can be surprising to people (and also matches
2110 C semantics) where they assume that "``X^X``" is always zero, even if
2111 ``X`` is undefined. This isn't true for a number of reasons, but the
2112 short answer is that an '``undef``' "variable" can arbitrarily change
2113 its value over its "live range". This is true because the variable
2114 doesn't actually *have a live range*. Instead, the value is logically
2115 read from arbitrary registers that happen to be around when needed, so
2116 the value is not necessarily consistent over time. In fact, ``%A`` and
2117 ``%C`` need to have the same semantics or the core LLVM "replace all
2118 uses with" concept would not hold.
2120 .. code-block:: llvm
2128 These examples show the crucial difference between an *undefined value*
2129 and *undefined behavior*. An undefined value (like '``undef``') is
2130 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2131 operation can be constant folded to '``undef``', because the '``undef``'
2132 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2133 However, in the second example, we can make a more aggressive
2134 assumption: because the ``undef`` is allowed to be an arbitrary value,
2135 we are allowed to assume that it could be zero. Since a divide by zero
2136 has *undefined behavior*, we are allowed to assume that the operation
2137 does not execute at all. This allows us to delete the divide and all
2138 code after it. Because the undefined operation "can't happen", the
2139 optimizer can assume that it occurs in dead code.
2141 .. code-block:: llvm
2143 a: store undef -> %X
2144 b: store %X -> undef
2149 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2150 value can be assumed to not have any effect; we can assume that the
2151 value is overwritten with bits that happen to match what was already
2152 there. However, a store *to* an undefined location could clobber
2153 arbitrary memory, therefore, it has undefined behavior.
2160 Poison values are similar to :ref:`undef values <undefvalues>`, however
2161 they also represent the fact that an instruction or constant expression
2162 which cannot evoke side effects has nevertheless detected a condition
2163 which results in undefined behavior.
2165 There is currently no way of representing a poison value in the IR; they
2166 only exist when produced by operations such as :ref:`add <i_add>` with
2169 Poison value behavior is defined in terms of value *dependence*:
2171 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2172 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2173 their dynamic predecessor basic block.
2174 - Function arguments depend on the corresponding actual argument values
2175 in the dynamic callers of their functions.
2176 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2177 instructions that dynamically transfer control back to them.
2178 - :ref:`Invoke <i_invoke>` instructions depend on the
2179 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2180 call instructions that dynamically transfer control back to them.
2181 - Non-volatile loads and stores depend on the most recent stores to all
2182 of the referenced memory addresses, following the order in the IR
2183 (including loads and stores implied by intrinsics such as
2184 :ref:`@llvm.memcpy <int_memcpy>`.)
2185 - An instruction with externally visible side effects depends on the
2186 most recent preceding instruction with externally visible side
2187 effects, following the order in the IR. (This includes :ref:`volatile
2188 operations <volatile>`.)
2189 - An instruction *control-depends* on a :ref:`terminator
2190 instruction <terminators>` if the terminator instruction has
2191 multiple successors and the instruction is always executed when
2192 control transfers to one of the successors, and may not be executed
2193 when control is transferred to another.
2194 - Additionally, an instruction also *control-depends* on a terminator
2195 instruction if the set of instructions it otherwise depends on would
2196 be different if the terminator had transferred control to a different
2198 - Dependence is transitive.
2200 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2201 with the additional affect that any instruction which has a *dependence*
2202 on a poison value has undefined behavior.
2204 Here are some examples:
2206 .. code-block:: llvm
2209 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2210 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2211 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2212 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2214 store i32 %poison, i32* @g ; Poison value stored to memory.
2215 %poison2 = load i32* @g ; Poison value loaded back from memory.
2217 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2219 %narrowaddr = bitcast i32* @g to i16*
2220 %wideaddr = bitcast i32* @g to i64*
2221 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2222 %poison4 = load i64* %wideaddr ; Returns a poison value.
2224 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2225 br i1 %cmp, label %true, label %end ; Branch to either destination.
2228 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2229 ; it has undefined behavior.
2233 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2234 ; Both edges into this PHI are
2235 ; control-dependent on %cmp, so this
2236 ; always results in a poison value.
2238 store volatile i32 0, i32* @g ; This would depend on the store in %true
2239 ; if %cmp is true, or the store in %entry
2240 ; otherwise, so this is undefined behavior.
2242 br i1 %cmp, label %second_true, label %second_end
2243 ; The same branch again, but this time the
2244 ; true block doesn't have side effects.
2251 store volatile i32 0, i32* @g ; This time, the instruction always depends
2252 ; on the store in %end. Also, it is
2253 ; control-equivalent to %end, so this is
2254 ; well-defined (ignoring earlier undefined
2255 ; behavior in this example).
2259 Addresses of Basic Blocks
2260 -------------------------
2262 ``blockaddress(@function, %block)``
2264 The '``blockaddress``' constant computes the address of the specified
2265 basic block in the specified function, and always has an ``i8*`` type.
2266 Taking the address of the entry block is illegal.
2268 This value only has defined behavior when used as an operand to the
2269 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2270 against null. Pointer equality tests between labels addresses results in
2271 undefined behavior --- though, again, comparison against null is ok, and
2272 no label is equal to the null pointer. This may be passed around as an
2273 opaque pointer sized value as long as the bits are not inspected. This
2274 allows ``ptrtoint`` and arithmetic to be performed on these values so
2275 long as the original value is reconstituted before the ``indirectbr``
2278 Finally, some targets may provide defined semantics when using the value
2279 as the operand to an inline assembly, but that is target specific.
2283 Constant Expressions
2284 --------------------
2286 Constant expressions are used to allow expressions involving other
2287 constants to be used as constants. Constant expressions may be of any
2288 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2289 that does not have side effects (e.g. load and call are not supported).
2290 The following is the syntax for constant expressions:
2292 ``trunc (CST to TYPE)``
2293 Truncate a constant to another type. The bit size of CST must be
2294 larger than the bit size of TYPE. Both types must be integers.
2295 ``zext (CST to TYPE)``
2296 Zero extend a constant to another type. The bit size of CST must be
2297 smaller than the bit size of TYPE. Both types must be integers.
2298 ``sext (CST to TYPE)``
2299 Sign extend a constant to another type. The bit size of CST must be
2300 smaller than the bit size of TYPE. Both types must be integers.
2301 ``fptrunc (CST to TYPE)``
2302 Truncate a floating point constant to another floating point type.
2303 The size of CST must be larger than the size of TYPE. Both types
2304 must be floating point.
2305 ``fpext (CST to TYPE)``
2306 Floating point extend a constant to another type. The size of CST
2307 must be smaller or equal to the size of TYPE. Both types must be
2309 ``fptoui (CST to TYPE)``
2310 Convert a floating point constant to the corresponding unsigned
2311 integer constant. TYPE must be a scalar or vector integer type. CST
2312 must be of scalar or vector floating point type. Both CST and TYPE
2313 must be scalars, or vectors of the same number of elements. If the
2314 value won't fit in the integer type, the results are undefined.
2315 ``fptosi (CST to TYPE)``
2316 Convert a floating point constant to the corresponding signed
2317 integer constant. TYPE must be a scalar or vector integer type. CST
2318 must be of scalar or vector floating point type. Both CST and TYPE
2319 must be scalars, or vectors of the same number of elements. If the
2320 value won't fit in the integer type, the results are undefined.
2321 ``uitofp (CST to TYPE)``
2322 Convert an unsigned integer constant to the corresponding floating
2323 point constant. TYPE must be a scalar or vector floating point type.
2324 CST must be of scalar or vector integer type. Both CST and TYPE must
2325 be scalars, or vectors of the same number of elements. If the value
2326 won't fit in the floating point type, the results are undefined.
2327 ``sitofp (CST to TYPE)``
2328 Convert a signed integer constant to the corresponding floating
2329 point constant. TYPE must be a scalar or vector floating point type.
2330 CST must be of scalar or vector integer type. Both CST and TYPE must
2331 be scalars, or vectors of the same number of elements. If the value
2332 won't fit in the floating point type, the results are undefined.
2333 ``ptrtoint (CST to TYPE)``
2334 Convert a pointer typed constant to the corresponding integer
2335 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2336 pointer type. The ``CST`` value is zero extended, truncated, or
2337 unchanged to make it fit in ``TYPE``.
2338 ``inttoptr (CST to TYPE)``
2339 Convert an integer constant to a pointer constant. TYPE must be a
2340 pointer type. CST must be of integer type. The CST value is zero
2341 extended, truncated, or unchanged to make it fit in a pointer size.
2342 This one is *really* dangerous!
2343 ``bitcast (CST to TYPE)``
2344 Convert a constant, CST, to another TYPE. The constraints of the
2345 operands are the same as those for the :ref:`bitcast
2346 instruction <i_bitcast>`.
2347 ``addrspacecast (CST to TYPE)``
2348 Convert a constant pointer or constant vector of pointer, CST, to another
2349 TYPE in a different address space. The constraints of the operands are the
2350 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2351 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2352 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2353 constants. As with the :ref:`getelementptr <i_getelementptr>`
2354 instruction, the index list may have zero or more indexes, which are
2355 required to make sense for the type of "CSTPTR".
2356 ``select (COND, VAL1, VAL2)``
2357 Perform the :ref:`select operation <i_select>` on constants.
2358 ``icmp COND (VAL1, VAL2)``
2359 Performs the :ref:`icmp operation <i_icmp>` on constants.
2360 ``fcmp COND (VAL1, VAL2)``
2361 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2362 ``extractelement (VAL, IDX)``
2363 Perform the :ref:`extractelement operation <i_extractelement>` on
2365 ``insertelement (VAL, ELT, IDX)``
2366 Perform the :ref:`insertelement operation <i_insertelement>` on
2368 ``shufflevector (VEC1, VEC2, IDXMASK)``
2369 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2371 ``extractvalue (VAL, IDX0, IDX1, ...)``
2372 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2373 constants. The index list is interpreted in a similar manner as
2374 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2375 least one index value must be specified.
2376 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2377 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2378 The index list is interpreted in a similar manner as indices in a
2379 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2380 value must be specified.
2381 ``OPCODE (LHS, RHS)``
2382 Perform the specified operation of the LHS and RHS constants. OPCODE
2383 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2384 binary <bitwiseops>` operations. The constraints on operands are
2385 the same as those for the corresponding instruction (e.g. no bitwise
2386 operations on floating point values are allowed).
2393 Inline Assembler Expressions
2394 ----------------------------
2396 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2397 Inline Assembly <moduleasm>`) through the use of a special value. This
2398 value represents the inline assembler as a string (containing the
2399 instructions to emit), a list of operand constraints (stored as a
2400 string), a flag that indicates whether or not the inline asm expression
2401 has side effects, and a flag indicating whether the function containing
2402 the asm needs to align its stack conservatively. An example inline
2403 assembler expression is:
2405 .. code-block:: llvm
2407 i32 (i32) asm "bswap $0", "=r,r"
2409 Inline assembler expressions may **only** be used as the callee operand
2410 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2411 Thus, typically we have:
2413 .. code-block:: llvm
2415 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2417 Inline asms with side effects not visible in the constraint list must be
2418 marked as having side effects. This is done through the use of the
2419 '``sideeffect``' keyword, like so:
2421 .. code-block:: llvm
2423 call void asm sideeffect "eieio", ""()
2425 In some cases inline asms will contain code that will not work unless
2426 the stack is aligned in some way, such as calls or SSE instructions on
2427 x86, yet will not contain code that does that alignment within the asm.
2428 The compiler should make conservative assumptions about what the asm
2429 might contain and should generate its usual stack alignment code in the
2430 prologue if the '``alignstack``' keyword is present:
2432 .. code-block:: llvm
2434 call void asm alignstack "eieio", ""()
2436 Inline asms also support using non-standard assembly dialects. The
2437 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2438 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2439 the only supported dialects. An example is:
2441 .. code-block:: llvm
2443 call void asm inteldialect "eieio", ""()
2445 If multiple keywords appear the '``sideeffect``' keyword must come
2446 first, the '``alignstack``' keyword second and the '``inteldialect``'
2452 The call instructions that wrap inline asm nodes may have a
2453 "``!srcloc``" MDNode attached to it that contains a list of constant
2454 integers. If present, the code generator will use the integer as the
2455 location cookie value when report errors through the ``LLVMContext``
2456 error reporting mechanisms. This allows a front-end to correlate backend
2457 errors that occur with inline asm back to the source code that produced
2460 .. code-block:: llvm
2462 call void asm sideeffect "something bad", ""(), !srcloc !42
2464 !42 = !{ i32 1234567 }
2466 It is up to the front-end to make sense of the magic numbers it places
2467 in the IR. If the MDNode contains multiple constants, the code generator
2468 will use the one that corresponds to the line of the asm that the error
2473 Metadata Nodes and Metadata Strings
2474 -----------------------------------
2476 LLVM IR allows metadata to be attached to instructions in the program
2477 that can convey extra information about the code to the optimizers and
2478 code generator. One example application of metadata is source-level
2479 debug information. There are two metadata primitives: strings and nodes.
2480 All metadata has the ``metadata`` type and is identified in syntax by a
2481 preceding exclamation point ('``!``').
2483 A metadata string is a string surrounded by double quotes. It can
2484 contain any character by escaping non-printable characters with
2485 "``\xx``" where "``xx``" is the two digit hex code. For example:
2488 Metadata nodes are represented with notation similar to structure
2489 constants (a comma separated list of elements, surrounded by braces and
2490 preceded by an exclamation point). Metadata nodes can have any values as
2491 their operand. For example:
2493 .. code-block:: llvm
2495 !{ metadata !"test\00", i32 10}
2497 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2498 metadata nodes, which can be looked up in the module symbol table. For
2501 .. code-block:: llvm
2503 !foo = metadata !{!4, !3}
2505 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2506 function is using two metadata arguments:
2508 .. code-block:: llvm
2510 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2512 Metadata can be attached with an instruction. Here metadata ``!21`` is
2513 attached to the ``add`` instruction using the ``!dbg`` identifier:
2515 .. code-block:: llvm
2517 %indvar.next = add i64 %indvar, 1, !dbg !21
2519 More information about specific metadata nodes recognized by the
2520 optimizers and code generator is found below.
2525 In LLVM IR, memory does not have types, so LLVM's own type system is not
2526 suitable for doing TBAA. Instead, metadata is added to the IR to
2527 describe a type system of a higher level language. This can be used to
2528 implement typical C/C++ TBAA, but it can also be used to implement
2529 custom alias analysis behavior for other languages.
2531 The current metadata format is very simple. TBAA metadata nodes have up
2532 to three fields, e.g.:
2534 .. code-block:: llvm
2536 !0 = metadata !{ metadata !"an example type tree" }
2537 !1 = metadata !{ metadata !"int", metadata !0 }
2538 !2 = metadata !{ metadata !"float", metadata !0 }
2539 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2541 The first field is an identity field. It can be any value, usually a
2542 metadata string, which uniquely identifies the type. The most important
2543 name in the tree is the name of the root node. Two trees with different
2544 root node names are entirely disjoint, even if they have leaves with
2547 The second field identifies the type's parent node in the tree, or is
2548 null or omitted for a root node. A type is considered to alias all of
2549 its descendants and all of its ancestors in the tree. Also, a type is
2550 considered to alias all types in other trees, so that bitcode produced
2551 from multiple front-ends is handled conservatively.
2553 If the third field is present, it's an integer which if equal to 1
2554 indicates that the type is "constant" (meaning
2555 ``pointsToConstantMemory`` should return true; see `other useful
2556 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2558 '``tbaa.struct``' Metadata
2559 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2561 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2562 aggregate assignment operations in C and similar languages, however it
2563 is defined to copy a contiguous region of memory, which is more than
2564 strictly necessary for aggregate types which contain holes due to
2565 padding. Also, it doesn't contain any TBAA information about the fields
2568 ``!tbaa.struct`` metadata can describe which memory subregions in a
2569 memcpy are padding and what the TBAA tags of the struct are.
2571 The current metadata format is very simple. ``!tbaa.struct`` metadata
2572 nodes are a list of operands which are in conceptual groups of three.
2573 For each group of three, the first operand gives the byte offset of a
2574 field in bytes, the second gives its size in bytes, and the third gives
2577 .. code-block:: llvm
2579 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2581 This describes a struct with two fields. The first is at offset 0 bytes
2582 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2583 and has size 4 bytes and has tbaa tag !2.
2585 Note that the fields need not be contiguous. In this example, there is a
2586 4 byte gap between the two fields. This gap represents padding which
2587 does not carry useful data and need not be preserved.
2589 '``fpmath``' Metadata
2590 ^^^^^^^^^^^^^^^^^^^^^
2592 ``fpmath`` metadata may be attached to any instruction of floating point
2593 type. It can be used to express the maximum acceptable error in the
2594 result of that instruction, in ULPs, thus potentially allowing the
2595 compiler to use a more efficient but less accurate method of computing
2596 it. ULP is defined as follows:
2598 If ``x`` is a real number that lies between two finite consecutive
2599 floating-point numbers ``a`` and ``b``, without being equal to one
2600 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2601 distance between the two non-equal finite floating-point numbers
2602 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2604 The metadata node shall consist of a single positive floating point
2605 number representing the maximum relative error, for example:
2607 .. code-block:: llvm
2609 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2611 '``range``' Metadata
2612 ^^^^^^^^^^^^^^^^^^^^
2614 ``range`` metadata may be attached only to loads of integer types. It
2615 expresses the possible ranges the loaded value is in. The ranges are
2616 represented with a flattened list of integers. The loaded value is known
2617 to be in the union of the ranges defined by each consecutive pair. Each
2618 pair has the following properties:
2620 - The type must match the type loaded by the instruction.
2621 - The pair ``a,b`` represents the range ``[a,b)``.
2622 - Both ``a`` and ``b`` are constants.
2623 - The range is allowed to wrap.
2624 - The range should not represent the full or empty set. That is,
2627 In addition, the pairs must be in signed order of the lower bound and
2628 they must be non-contiguous.
2632 .. code-block:: llvm
2634 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2635 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2636 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2637 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2639 !0 = metadata !{ i8 0, i8 2 }
2640 !1 = metadata !{ i8 255, i8 2 }
2641 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2642 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2647 It is sometimes useful to attach information to loop constructs. Currently,
2648 loop metadata is implemented as metadata attached to the branch instruction
2649 in the loop latch block. This type of metadata refer to a metadata node that is
2650 guaranteed to be separate for each loop. The loop identifier metadata is
2651 specified with the name ``llvm.loop``.
2653 The loop identifier metadata is implemented using a metadata that refers to
2654 itself to avoid merging it with any other identifier metadata, e.g.,
2655 during module linkage or function inlining. That is, each loop should refer
2656 to their own identification metadata even if they reside in separate functions.
2657 The following example contains loop identifier metadata for two separate loop
2660 .. code-block:: llvm
2662 !0 = metadata !{ metadata !0 }
2663 !1 = metadata !{ metadata !1 }
2665 The loop identifier metadata can be used to specify additional per-loop
2666 metadata. Any operands after the first operand can be treated as user-defined
2667 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2668 by the loop vectorizer to indicate how many times to unroll the loop:
2670 .. code-block:: llvm
2672 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2674 !0 = metadata !{ metadata !0, metadata !1 }
2675 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2680 Metadata types used to annotate memory accesses with information helpful
2681 for optimizations are prefixed with ``llvm.mem``.
2683 '``llvm.mem.parallel_loop_access``' Metadata
2684 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2686 For a loop to be parallel, in addition to using
2687 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2688 also all of the memory accessing instructions in the loop body need to be
2689 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2690 is at least one memory accessing instruction not marked with the metadata,
2691 the loop must be considered a sequential loop. This causes parallel loops to be
2692 converted to sequential loops due to optimization passes that are unaware of
2693 the parallel semantics and that insert new memory instructions to the loop
2696 Example of a loop that is considered parallel due to its correct use of
2697 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2698 metadata types that refer to the same loop identifier metadata.
2700 .. code-block:: llvm
2704 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2706 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2708 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2712 !0 = metadata !{ metadata !0 }
2714 It is also possible to have nested parallel loops. In that case the
2715 memory accesses refer to a list of loop identifier metadata nodes instead of
2716 the loop identifier metadata node directly:
2718 .. code-block:: llvm
2725 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2727 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2729 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2733 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2735 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2737 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2739 outer.for.end: ; preds = %for.body
2741 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2742 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2743 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2745 '``llvm.vectorizer``'
2746 ^^^^^^^^^^^^^^^^^^^^^
2748 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2749 vectorization parameters such as vectorization factor and unroll factor.
2751 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2752 loop identification metadata.
2754 '``llvm.vectorizer.unroll``' Metadata
2755 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2757 This metadata instructs the loop vectorizer to unroll the specified
2758 loop exactly ``N`` times.
2760 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2761 operand is an integer specifying the unroll factor. For example:
2763 .. code-block:: llvm
2765 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2767 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2770 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2771 determined automatically.
2773 '``llvm.vectorizer.width``' Metadata
2774 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2776 This metadata sets the target width of the vectorizer to ``N``. Without
2777 this metadata, the vectorizer will choose a width automatically.
2778 Regardless of this metadata, the vectorizer will only vectorize loops if
2779 it believes it is valid to do so.
2781 The first operand is the string ``llvm.vectorizer.width`` and the second
2782 operand is an integer specifying the width. For example:
2784 .. code-block:: llvm
2786 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2788 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2791 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2794 Module Flags Metadata
2795 =====================
2797 Information about the module as a whole is difficult to convey to LLVM's
2798 subsystems. The LLVM IR isn't sufficient to transmit this information.
2799 The ``llvm.module.flags`` named metadata exists in order to facilitate
2800 this. These flags are in the form of key / value pairs --- much like a
2801 dictionary --- making it easy for any subsystem who cares about a flag to
2804 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2805 Each triplet has the following form:
2807 - The first element is a *behavior* flag, which specifies the behavior
2808 when two (or more) modules are merged together, and it encounters two
2809 (or more) metadata with the same ID. The supported behaviors are
2811 - The second element is a metadata string that is a unique ID for the
2812 metadata. Each module may only have one flag entry for each unique ID (not
2813 including entries with the **Require** behavior).
2814 - The third element is the value of the flag.
2816 When two (or more) modules are merged together, the resulting
2817 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2818 each unique metadata ID string, there will be exactly one entry in the merged
2819 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2820 be determined by the merge behavior flag, as described below. The only exception
2821 is that entries with the *Require* behavior are always preserved.
2823 The following behaviors are supported:
2834 Emits an error if two values disagree, otherwise the resulting value
2835 is that of the operands.
2839 Emits a warning if two values disagree. The result value will be the
2840 operand for the flag from the first module being linked.
2844 Adds a requirement that another module flag be present and have a
2845 specified value after linking is performed. The value must be a
2846 metadata pair, where the first element of the pair is the ID of the
2847 module flag to be restricted, and the second element of the pair is
2848 the value the module flag should be restricted to. This behavior can
2849 be used to restrict the allowable results (via triggering of an
2850 error) of linking IDs with the **Override** behavior.
2854 Uses the specified value, regardless of the behavior or value of the
2855 other module. If both modules specify **Override**, but the values
2856 differ, an error will be emitted.
2860 Appends the two values, which are required to be metadata nodes.
2864 Appends the two values, which are required to be metadata
2865 nodes. However, duplicate entries in the second list are dropped
2866 during the append operation.
2868 It is an error for a particular unique flag ID to have multiple behaviors,
2869 except in the case of **Require** (which adds restrictions on another metadata
2870 value) or **Override**.
2872 An example of module flags:
2874 .. code-block:: llvm
2876 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2877 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2878 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2879 !3 = metadata !{ i32 3, metadata !"qux",
2881 metadata !"foo", i32 1
2884 !llvm.module.flags = !{ !0, !1, !2, !3 }
2886 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2887 if two or more ``!"foo"`` flags are seen is to emit an error if their
2888 values are not equal.
2890 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2891 behavior if two or more ``!"bar"`` flags are seen is to use the value
2894 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2895 behavior if two or more ``!"qux"`` flags are seen is to emit a
2896 warning if their values are not equal.
2898 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2902 metadata !{ metadata !"foo", i32 1 }
2904 The behavior is to emit an error if the ``llvm.module.flags`` does not
2905 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2908 Objective-C Garbage Collection Module Flags Metadata
2909 ----------------------------------------------------
2911 On the Mach-O platform, Objective-C stores metadata about garbage
2912 collection in a special section called "image info". The metadata
2913 consists of a version number and a bitmask specifying what types of
2914 garbage collection are supported (if any) by the file. If two or more
2915 modules are linked together their garbage collection metadata needs to
2916 be merged rather than appended together.
2918 The Objective-C garbage collection module flags metadata consists of the
2919 following key-value pairs:
2928 * - ``Objective-C Version``
2929 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2931 * - ``Objective-C Image Info Version``
2932 - **[Required]** --- The version of the image info section. Currently
2935 * - ``Objective-C Image Info Section``
2936 - **[Required]** --- The section to place the metadata. Valid values are
2937 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2938 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2939 Objective-C ABI version 2.
2941 * - ``Objective-C Garbage Collection``
2942 - **[Required]** --- Specifies whether garbage collection is supported or
2943 not. Valid values are 0, for no garbage collection, and 2, for garbage
2944 collection supported.
2946 * - ``Objective-C GC Only``
2947 - **[Optional]** --- Specifies that only garbage collection is supported.
2948 If present, its value must be 6. This flag requires that the
2949 ``Objective-C Garbage Collection`` flag have the value 2.
2951 Some important flag interactions:
2953 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2954 merged with a module with ``Objective-C Garbage Collection`` set to
2955 2, then the resulting module has the
2956 ``Objective-C Garbage Collection`` flag set to 0.
2957 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2958 merged with a module with ``Objective-C GC Only`` set to 6.
2960 Automatic Linker Flags Module Flags Metadata
2961 --------------------------------------------
2963 Some targets support embedding flags to the linker inside individual object
2964 files. Typically this is used in conjunction with language extensions which
2965 allow source files to explicitly declare the libraries they depend on, and have
2966 these automatically be transmitted to the linker via object files.
2968 These flags are encoded in the IR using metadata in the module flags section,
2969 using the ``Linker Options`` key. The merge behavior for this flag is required
2970 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2971 node which should be a list of other metadata nodes, each of which should be a
2972 list of metadata strings defining linker options.
2974 For example, the following metadata section specifies two separate sets of
2975 linker options, presumably to link against ``libz`` and the ``Cocoa``
2978 !0 = metadata !{ i32 6, metadata !"Linker Options",
2980 metadata !{ metadata !"-lz" },
2981 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2982 !llvm.module.flags = !{ !0 }
2984 The metadata encoding as lists of lists of options, as opposed to a collapsed
2985 list of options, is chosen so that the IR encoding can use multiple option
2986 strings to specify e.g., a single library, while still having that specifier be
2987 preserved as an atomic element that can be recognized by a target specific
2988 assembly writer or object file emitter.
2990 Each individual option is required to be either a valid option for the target's
2991 linker, or an option that is reserved by the target specific assembly writer or
2992 object file emitter. No other aspect of these options is defined by the IR.
2994 .. _intrinsicglobalvariables:
2996 Intrinsic Global Variables
2997 ==========================
2999 LLVM has a number of "magic" global variables that contain data that
3000 affect code generation or other IR semantics. These are documented here.
3001 All globals of this sort should have a section specified as
3002 "``llvm.metadata``". This section and all globals that start with
3003 "``llvm.``" are reserved for use by LLVM.
3007 The '``llvm.used``' Global Variable
3008 -----------------------------------
3010 The ``@llvm.used`` global is an array which has
3011 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3012 pointers to named global variables, functions and aliases which may optionally
3013 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3016 .. code-block:: llvm
3021 @llvm.used = appending global [2 x i8*] [
3023 i8* bitcast (i32* @Y to i8*)
3024 ], section "llvm.metadata"
3026 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3027 and linker are required to treat the symbol as if there is a reference to the
3028 symbol that it cannot see (which is why they have to be named). For example, if
3029 a variable has internal linkage and no references other than that from the
3030 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3031 references from inline asms and other things the compiler cannot "see", and
3032 corresponds to "``attribute((used))``" in GNU C.
3034 On some targets, the code generator must emit a directive to the
3035 assembler or object file to prevent the assembler and linker from
3036 molesting the symbol.
3038 .. _gv_llvmcompilerused:
3040 The '``llvm.compiler.used``' Global Variable
3041 --------------------------------------------
3043 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3044 directive, except that it only prevents the compiler from touching the
3045 symbol. On targets that support it, this allows an intelligent linker to
3046 optimize references to the symbol without being impeded as it would be
3049 This is a rare construct that should only be used in rare circumstances,
3050 and should not be exposed to source languages.
3052 .. _gv_llvmglobalctors:
3054 The '``llvm.global_ctors``' Global Variable
3055 -------------------------------------------
3057 .. code-block:: llvm
3059 %0 = type { i32, void ()* }
3060 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3062 The ``@llvm.global_ctors`` array contains a list of constructor
3063 functions and associated priorities. The functions referenced by this
3064 array will be called in ascending order of priority (i.e. lowest first)
3065 when the module is loaded. The order of functions with the same priority
3068 .. _llvmglobaldtors:
3070 The '``llvm.global_dtors``' Global Variable
3071 -------------------------------------------
3073 .. code-block:: llvm
3075 %0 = type { i32, void ()* }
3076 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3078 The ``@llvm.global_dtors`` array contains a list of destructor functions
3079 and associated priorities. The functions referenced by this array will
3080 be called in descending order of priority (i.e. highest first) when the
3081 module is loaded. The order of functions with the same priority is not
3084 Instruction Reference
3085 =====================
3087 The LLVM instruction set consists of several different classifications
3088 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3089 instructions <binaryops>`, :ref:`bitwise binary
3090 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3091 :ref:`other instructions <otherops>`.
3095 Terminator Instructions
3096 -----------------------
3098 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3099 program ends with a "Terminator" instruction, which indicates which
3100 block should be executed after the current block is finished. These
3101 terminator instructions typically yield a '``void``' value: they produce
3102 control flow, not values (the one exception being the
3103 ':ref:`invoke <i_invoke>`' instruction).
3105 The terminator instructions are: ':ref:`ret <i_ret>`',
3106 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3107 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3108 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3112 '``ret``' Instruction
3113 ^^^^^^^^^^^^^^^^^^^^^
3120 ret <type> <value> ; Return a value from a non-void function
3121 ret void ; Return from void function
3126 The '``ret``' instruction is used to return control flow (and optionally
3127 a value) from a function back to the caller.
3129 There are two forms of the '``ret``' instruction: one that returns a
3130 value and then causes control flow, and one that just causes control
3136 The '``ret``' instruction optionally accepts a single argument, the
3137 return value. The type of the return value must be a ':ref:`first
3138 class <t_firstclass>`' type.
3140 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3141 return type and contains a '``ret``' instruction with no return value or
3142 a return value with a type that does not match its type, or if it has a
3143 void return type and contains a '``ret``' instruction with a return
3149 When the '``ret``' instruction is executed, control flow returns back to
3150 the calling function's context. If the caller is a
3151 ":ref:`call <i_call>`" instruction, execution continues at the
3152 instruction after the call. If the caller was an
3153 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3154 beginning of the "normal" destination block. If the instruction returns
3155 a value, that value shall set the call or invoke instruction's return
3161 .. code-block:: llvm
3163 ret i32 5 ; Return an integer value of 5
3164 ret void ; Return from a void function
3165 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3169 '``br``' Instruction
3170 ^^^^^^^^^^^^^^^^^^^^
3177 br i1 <cond>, label <iftrue>, label <iffalse>
3178 br label <dest> ; Unconditional branch
3183 The '``br``' instruction is used to cause control flow to transfer to a
3184 different basic block in the current function. There are two forms of
3185 this instruction, corresponding to a conditional branch and an
3186 unconditional branch.
3191 The conditional branch form of the '``br``' instruction takes a single
3192 '``i1``' value and two '``label``' values. The unconditional form of the
3193 '``br``' instruction takes a single '``label``' value as a target.
3198 Upon execution of a conditional '``br``' instruction, the '``i1``'
3199 argument is evaluated. If the value is ``true``, control flows to the
3200 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3201 to the '``iffalse``' ``label`` argument.
3206 .. code-block:: llvm
3209 %cond = icmp eq i32 %a, %b
3210 br i1 %cond, label %IfEqual, label %IfUnequal
3218 '``switch``' Instruction
3219 ^^^^^^^^^^^^^^^^^^^^^^^^
3226 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3231 The '``switch``' instruction is used to transfer control flow to one of
3232 several different places. It is a generalization of the '``br``'
3233 instruction, allowing a branch to occur to one of many possible
3239 The '``switch``' instruction uses three parameters: an integer
3240 comparison value '``value``', a default '``label``' destination, and an
3241 array of pairs of comparison value constants and '``label``'s. The table
3242 is not allowed to contain duplicate constant entries.
3247 The ``switch`` instruction specifies a table of values and destinations.
3248 When the '``switch``' instruction is executed, this table is searched
3249 for the given value. If the value is found, control flow is transferred
3250 to the corresponding destination; otherwise, control flow is transferred
3251 to the default destination.
3256 Depending on properties of the target machine and the particular
3257 ``switch`` instruction, this instruction may be code generated in
3258 different ways. For example, it could be generated as a series of
3259 chained conditional branches or with a lookup table.
3264 .. code-block:: llvm
3266 ; Emulate a conditional br instruction
3267 %Val = zext i1 %value to i32
3268 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3270 ; Emulate an unconditional br instruction
3271 switch i32 0, label %dest [ ]
3273 ; Implement a jump table:
3274 switch i32 %val, label %otherwise [ i32 0, label %onzero
3276 i32 2, label %ontwo ]
3280 '``indirectbr``' Instruction
3281 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3288 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3293 The '``indirectbr``' instruction implements an indirect branch to a
3294 label within the current function, whose address is specified by
3295 "``address``". Address must be derived from a
3296 :ref:`blockaddress <blockaddress>` constant.
3301 The '``address``' argument is the address of the label to jump to. The
3302 rest of the arguments indicate the full set of possible destinations
3303 that the address may point to. Blocks are allowed to occur multiple
3304 times in the destination list, though this isn't particularly useful.
3306 This destination list is required so that dataflow analysis has an
3307 accurate understanding of the CFG.
3312 Control transfers to the block specified in the address argument. All
3313 possible destination blocks must be listed in the label list, otherwise
3314 this instruction has undefined behavior. This implies that jumps to
3315 labels defined in other functions have undefined behavior as well.
3320 This is typically implemented with a jump through a register.
3325 .. code-block:: llvm
3327 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3331 '``invoke``' Instruction
3332 ^^^^^^^^^^^^^^^^^^^^^^^^
3339 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3340 to label <normal label> unwind label <exception label>
3345 The '``invoke``' instruction causes control to transfer to a specified
3346 function, with the possibility of control flow transfer to either the
3347 '``normal``' label or the '``exception``' label. If the callee function
3348 returns with the "``ret``" instruction, control flow will return to the
3349 "normal" label. If the callee (or any indirect callees) returns via the
3350 ":ref:`resume <i_resume>`" instruction or other exception handling
3351 mechanism, control is interrupted and continued at the dynamically
3352 nearest "exception" label.
3354 The '``exception``' label is a `landing
3355 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3356 '``exception``' label is required to have the
3357 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3358 information about the behavior of the program after unwinding happens,
3359 as its first non-PHI instruction. The restrictions on the
3360 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3361 instruction, so that the important information contained within the
3362 "``landingpad``" instruction can't be lost through normal code motion.
3367 This instruction requires several arguments:
3369 #. The optional "cconv" marker indicates which :ref:`calling
3370 convention <callingconv>` the call should use. If none is
3371 specified, the call defaults to using C calling conventions.
3372 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3373 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3375 #. '``ptr to function ty``': shall be the signature of the pointer to
3376 function value being invoked. In most cases, this is a direct
3377 function invocation, but indirect ``invoke``'s are just as possible,
3378 branching off an arbitrary pointer to function value.
3379 #. '``function ptr val``': An LLVM value containing a pointer to a
3380 function to be invoked.
3381 #. '``function args``': argument list whose types match the function
3382 signature argument types and parameter attributes. All arguments must
3383 be of :ref:`first class <t_firstclass>` type. If the function signature
3384 indicates the function accepts a variable number of arguments, the
3385 extra arguments can be specified.
3386 #. '``normal label``': the label reached when the called function
3387 executes a '``ret``' instruction.
3388 #. '``exception label``': the label reached when a callee returns via
3389 the :ref:`resume <i_resume>` instruction or other exception handling
3391 #. The optional :ref:`function attributes <fnattrs>` list. Only
3392 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3393 attributes are valid here.
3398 This instruction is designed to operate as a standard '``call``'
3399 instruction in most regards. The primary difference is that it
3400 establishes an association with a label, which is used by the runtime
3401 library to unwind the stack.
3403 This instruction is used in languages with destructors to ensure that
3404 proper cleanup is performed in the case of either a ``longjmp`` or a
3405 thrown exception. Additionally, this is important for implementation of
3406 '``catch``' clauses in high-level languages that support them.
3408 For the purposes of the SSA form, the definition of the value returned
3409 by the '``invoke``' instruction is deemed to occur on the edge from the
3410 current block to the "normal" label. If the callee unwinds then no
3411 return value is available.
3416 .. code-block:: llvm
3418 %retval = invoke i32 @Test(i32 15) to label %Continue
3419 unwind label %TestCleanup ; {i32}:retval set
3420 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3421 unwind label %TestCleanup ; {i32}:retval set
3425 '``resume``' Instruction
3426 ^^^^^^^^^^^^^^^^^^^^^^^^
3433 resume <type> <value>
3438 The '``resume``' instruction is a terminator instruction that has no
3444 The '``resume``' instruction requires one argument, which must have the
3445 same type as the result of any '``landingpad``' instruction in the same
3451 The '``resume``' instruction resumes propagation of an existing
3452 (in-flight) exception whose unwinding was interrupted with a
3453 :ref:`landingpad <i_landingpad>` instruction.
3458 .. code-block:: llvm
3460 resume { i8*, i32 } %exn
3464 '``unreachable``' Instruction
3465 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3477 The '``unreachable``' instruction has no defined semantics. This
3478 instruction is used to inform the optimizer that a particular portion of
3479 the code is not reachable. This can be used to indicate that the code
3480 after a no-return function cannot be reached, and other facts.
3485 The '``unreachable``' instruction has no defined semantics.
3492 Binary operators are used to do most of the computation in a program.
3493 They require two operands of the same type, execute an operation on
3494 them, and produce a single value. The operands might represent multiple
3495 data, as is the case with the :ref:`vector <t_vector>` data type. The
3496 result value has the same type as its operands.
3498 There are several different binary operators:
3502 '``add``' Instruction
3503 ^^^^^^^^^^^^^^^^^^^^^
3510 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3511 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3512 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3513 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3518 The '``add``' instruction returns the sum of its two operands.
3523 The two arguments to the '``add``' instruction must be
3524 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3525 arguments must have identical types.
3530 The value produced is the integer sum of the two operands.
3532 If the sum has unsigned overflow, the result returned is the
3533 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3536 Because LLVM integers use a two's complement representation, this
3537 instruction is appropriate for both signed and unsigned integers.
3539 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3540 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3541 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3542 unsigned and/or signed overflow, respectively, occurs.
3547 .. code-block:: llvm
3549 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3553 '``fadd``' Instruction
3554 ^^^^^^^^^^^^^^^^^^^^^^
3561 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3566 The '``fadd``' instruction returns the sum of its two operands.
3571 The two arguments to the '``fadd``' instruction must be :ref:`floating
3572 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3573 Both arguments must have identical types.
3578 The value produced is the floating point sum of the two operands. This
3579 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3580 which are optimization hints to enable otherwise unsafe floating point
3586 .. code-block:: llvm
3588 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3590 '``sub``' Instruction
3591 ^^^^^^^^^^^^^^^^^^^^^
3598 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3599 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3600 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3601 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3606 The '``sub``' instruction returns the difference of its two operands.
3608 Note that the '``sub``' instruction is used to represent the '``neg``'
3609 instruction present in most other intermediate representations.
3614 The two arguments to the '``sub``' instruction must be
3615 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3616 arguments must have identical types.
3621 The value produced is the integer difference of the two operands.
3623 If the difference has unsigned overflow, the result returned is the
3624 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3627 Because LLVM integers use a two's complement representation, this
3628 instruction is appropriate for both signed and unsigned integers.
3630 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3631 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3632 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3633 unsigned and/or signed overflow, respectively, occurs.
3638 .. code-block:: llvm
3640 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3641 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3645 '``fsub``' Instruction
3646 ^^^^^^^^^^^^^^^^^^^^^^
3653 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3658 The '``fsub``' instruction returns the difference of its two operands.
3660 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3661 instruction present in most other intermediate representations.
3666 The two arguments to the '``fsub``' instruction must be :ref:`floating
3667 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3668 Both arguments must have identical types.
3673 The value produced is the floating point difference of the two operands.
3674 This instruction can also take any number of :ref:`fast-math
3675 flags <fastmath>`, which are optimization hints to enable otherwise
3676 unsafe floating point optimizations:
3681 .. code-block:: llvm
3683 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3684 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3686 '``mul``' Instruction
3687 ^^^^^^^^^^^^^^^^^^^^^
3694 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3695 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3696 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3697 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3702 The '``mul``' instruction returns the product of its two operands.
3707 The two arguments to the '``mul``' instruction must be
3708 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3709 arguments must have identical types.
3714 The value produced is the integer product of the two operands.
3716 If the result of the multiplication has unsigned overflow, the result
3717 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3718 bit width of the result.
3720 Because LLVM integers use a two's complement representation, and the
3721 result is the same width as the operands, this instruction returns the
3722 correct result for both signed and unsigned integers. If a full product
3723 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3724 sign-extended or zero-extended as appropriate to the width of the full
3727 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3728 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3729 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3730 unsigned and/or signed overflow, respectively, occurs.
3735 .. code-block:: llvm
3737 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3741 '``fmul``' Instruction
3742 ^^^^^^^^^^^^^^^^^^^^^^
3749 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3754 The '``fmul``' instruction returns the product of its two operands.
3759 The two arguments to the '``fmul``' instruction must be :ref:`floating
3760 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3761 Both arguments must have identical types.
3766 The value produced is the floating point product of the two operands.
3767 This instruction can also take any number of :ref:`fast-math
3768 flags <fastmath>`, which are optimization hints to enable otherwise
3769 unsafe floating point optimizations:
3774 .. code-block:: llvm
3776 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3778 '``udiv``' Instruction
3779 ^^^^^^^^^^^^^^^^^^^^^^
3786 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3787 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3792 The '``udiv``' instruction returns the quotient of its two operands.
3797 The two arguments to the '``udiv``' instruction must be
3798 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3799 arguments must have identical types.
3804 The value produced is the unsigned integer quotient of the two operands.
3806 Note that unsigned integer division and signed integer division are
3807 distinct operations; for signed integer division, use '``sdiv``'.
3809 Division by zero leads to undefined behavior.
3811 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3812 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3813 such, "((a udiv exact b) mul b) == a").
3818 .. code-block:: llvm
3820 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3822 '``sdiv``' Instruction
3823 ^^^^^^^^^^^^^^^^^^^^^^
3830 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3831 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3836 The '``sdiv``' instruction returns the quotient of its two operands.
3841 The two arguments to the '``sdiv``' instruction must be
3842 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3843 arguments must have identical types.
3848 The value produced is the signed integer quotient of the two operands
3849 rounded towards zero.
3851 Note that signed integer division and unsigned integer division are
3852 distinct operations; for unsigned integer division, use '``udiv``'.
3854 Division by zero leads to undefined behavior. Overflow also leads to
3855 undefined behavior; this is a rare case, but can occur, for example, by
3856 doing a 32-bit division of -2147483648 by -1.
3858 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3859 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3864 .. code-block:: llvm
3866 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3870 '``fdiv``' Instruction
3871 ^^^^^^^^^^^^^^^^^^^^^^
3878 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3883 The '``fdiv``' instruction returns the quotient of its two operands.
3888 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3889 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3890 Both arguments must have identical types.
3895 The value produced is the floating point quotient of the two operands.
3896 This instruction can also take any number of :ref:`fast-math
3897 flags <fastmath>`, which are optimization hints to enable otherwise
3898 unsafe floating point optimizations:
3903 .. code-block:: llvm
3905 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3907 '``urem``' Instruction
3908 ^^^^^^^^^^^^^^^^^^^^^^
3915 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3920 The '``urem``' instruction returns the remainder from the unsigned
3921 division of its two arguments.
3926 The two arguments to the '``urem``' instruction must be
3927 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3928 arguments must have identical types.
3933 This instruction returns the unsigned integer *remainder* of a division.
3934 This instruction always performs an unsigned division to get the
3937 Note that unsigned integer remainder and signed integer remainder are
3938 distinct operations; for signed integer remainder, use '``srem``'.
3940 Taking the remainder of a division by zero leads to undefined behavior.
3945 .. code-block:: llvm
3947 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3949 '``srem``' Instruction
3950 ^^^^^^^^^^^^^^^^^^^^^^
3957 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3962 The '``srem``' instruction returns the remainder from the signed
3963 division of its two operands. This instruction can also take
3964 :ref:`vector <t_vector>` versions of the values in which case the elements
3970 The two arguments to the '``srem``' instruction must be
3971 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3972 arguments must have identical types.
3977 This instruction returns the *remainder* of a division (where the result
3978 is either zero or has the same sign as the dividend, ``op1``), not the
3979 *modulo* operator (where the result is either zero or has the same sign
3980 as the divisor, ``op2``) of a value. For more information about the
3981 difference, see `The Math
3982 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3983 table of how this is implemented in various languages, please see
3985 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3987 Note that signed integer remainder and unsigned integer remainder are
3988 distinct operations; for unsigned integer remainder, use '``urem``'.
3990 Taking the remainder of a division by zero leads to undefined behavior.
3991 Overflow also leads to undefined behavior; this is a rare case, but can
3992 occur, for example, by taking the remainder of a 32-bit division of
3993 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3994 rule lets srem be implemented using instructions that return both the
3995 result of the division and the remainder.)
4000 .. code-block:: llvm
4002 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4006 '``frem``' Instruction
4007 ^^^^^^^^^^^^^^^^^^^^^^
4014 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4019 The '``frem``' instruction returns the remainder from the division of
4025 The two arguments to the '``frem``' instruction must be :ref:`floating
4026 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4027 Both arguments must have identical types.
4032 This instruction returns the *remainder* of a division. The remainder
4033 has the same sign as the dividend. This instruction can also take any
4034 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4035 to enable otherwise unsafe floating point optimizations:
4040 .. code-block:: llvm
4042 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4046 Bitwise Binary Operations
4047 -------------------------
4049 Bitwise binary operators are used to do various forms of bit-twiddling
4050 in a program. They are generally very efficient instructions and can
4051 commonly be strength reduced from other instructions. They require two
4052 operands of the same type, execute an operation on them, and produce a
4053 single value. The resulting value is the same type as its operands.
4055 '``shl``' Instruction
4056 ^^^^^^^^^^^^^^^^^^^^^
4063 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4064 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4065 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4066 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4071 The '``shl``' instruction returns the first operand shifted to the left
4072 a specified number of bits.
4077 Both arguments to the '``shl``' instruction must be the same
4078 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4079 '``op2``' is treated as an unsigned value.
4084 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4085 where ``n`` is the width of the result. If ``op2`` is (statically or
4086 dynamically) negative or equal to or larger than the number of bits in
4087 ``op1``, the result is undefined. If the arguments are vectors, each
4088 vector element of ``op1`` is shifted by the corresponding shift amount
4091 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4092 value <poisonvalues>` if it shifts out any non-zero bits. If the
4093 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4094 value <poisonvalues>` if it shifts out any bits that disagree with the
4095 resultant sign bit. As such, NUW/NSW have the same semantics as they
4096 would if the shift were expressed as a mul instruction with the same
4097 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4102 .. code-block:: llvm
4104 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4105 <result> = shl i32 4, 2 ; yields {i32}: 16
4106 <result> = shl i32 1, 10 ; yields {i32}: 1024
4107 <result> = shl i32 1, 32 ; undefined
4108 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4110 '``lshr``' Instruction
4111 ^^^^^^^^^^^^^^^^^^^^^^
4118 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4119 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4124 The '``lshr``' instruction (logical shift right) returns the first
4125 operand shifted to the right a specified number of bits with zero fill.
4130 Both arguments to the '``lshr``' instruction must be the same
4131 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4132 '``op2``' is treated as an unsigned value.
4137 This instruction always performs a logical shift right operation. The
4138 most significant bits of the result will be filled with zero bits after
4139 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4140 than the number of bits in ``op1``, the result is undefined. If the
4141 arguments are vectors, each vector element of ``op1`` is shifted by the
4142 corresponding shift amount in ``op2``.
4144 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4145 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4151 .. code-block:: llvm
4153 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4154 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4155 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4156 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4157 <result> = lshr i32 1, 32 ; undefined
4158 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4160 '``ashr``' Instruction
4161 ^^^^^^^^^^^^^^^^^^^^^^
4168 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4169 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4174 The '``ashr``' instruction (arithmetic shift right) returns the first
4175 operand shifted to the right a specified number of bits with sign
4181 Both arguments to the '``ashr``' instruction must be the same
4182 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4183 '``op2``' is treated as an unsigned value.
4188 This instruction always performs an arithmetic shift right operation,
4189 The most significant bits of the result will be filled with the sign bit
4190 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4191 than the number of bits in ``op1``, the result is undefined. If the
4192 arguments are vectors, each vector element of ``op1`` is shifted by the
4193 corresponding shift amount in ``op2``.
4195 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4196 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4202 .. code-block:: llvm
4204 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4205 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4206 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4207 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4208 <result> = ashr i32 1, 32 ; undefined
4209 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4211 '``and``' Instruction
4212 ^^^^^^^^^^^^^^^^^^^^^
4219 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4224 The '``and``' instruction returns the bitwise logical and of its two
4230 The two arguments to the '``and``' instruction must be
4231 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4232 arguments must have identical types.
4237 The truth table used for the '``and``' instruction is:
4254 .. code-block:: llvm
4256 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4257 <result> = and i32 15, 40 ; yields {i32}:result = 8
4258 <result> = and i32 4, 8 ; yields {i32}:result = 0
4260 '``or``' Instruction
4261 ^^^^^^^^^^^^^^^^^^^^
4268 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4273 The '``or``' instruction returns the bitwise logical inclusive or of its
4279 The two arguments to the '``or``' instruction must be
4280 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4281 arguments must have identical types.
4286 The truth table used for the '``or``' instruction is:
4305 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4306 <result> = or i32 15, 40 ; yields {i32}:result = 47
4307 <result> = or i32 4, 8 ; yields {i32}:result = 12
4309 '``xor``' Instruction
4310 ^^^^^^^^^^^^^^^^^^^^^
4317 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4322 The '``xor``' instruction returns the bitwise logical exclusive or of
4323 its two operands. The ``xor`` is used to implement the "one's
4324 complement" operation, which is the "~" operator in C.
4329 The two arguments to the '``xor``' instruction must be
4330 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4331 arguments must have identical types.
4336 The truth table used for the '``xor``' instruction is:
4353 .. code-block:: llvm
4355 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4356 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4357 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4358 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4363 LLVM supports several instructions to represent vector operations in a
4364 target-independent manner. These instructions cover the element-access
4365 and vector-specific operations needed to process vectors effectively.
4366 While LLVM does directly support these vector operations, many
4367 sophisticated algorithms will want to use target-specific intrinsics to
4368 take full advantage of a specific target.
4370 .. _i_extractelement:
4372 '``extractelement``' Instruction
4373 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4380 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4385 The '``extractelement``' instruction extracts a single scalar element
4386 from a vector at a specified index.
4391 The first operand of an '``extractelement``' instruction is a value of
4392 :ref:`vector <t_vector>` type. The second operand is an index indicating
4393 the position from which to extract the element. The index may be a
4399 The result is a scalar of the same type as the element type of ``val``.
4400 Its value is the value at position ``idx`` of ``val``. If ``idx``
4401 exceeds the length of ``val``, the results are undefined.
4406 .. code-block:: llvm
4408 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4410 .. _i_insertelement:
4412 '``insertelement``' Instruction
4413 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4420 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4425 The '``insertelement``' instruction inserts a scalar element into a
4426 vector at a specified index.
4431 The first operand of an '``insertelement``' instruction is a value of
4432 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4433 type must equal the element type of the first operand. The third operand
4434 is an index indicating the position at which to insert the value. The
4435 index may be a variable.
4440 The result is a vector of the same type as ``val``. Its element values
4441 are those of ``val`` except at position ``idx``, where it gets the value
4442 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4448 .. code-block:: llvm
4450 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4452 .. _i_shufflevector:
4454 '``shufflevector``' Instruction
4455 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4462 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4467 The '``shufflevector``' instruction constructs a permutation of elements
4468 from two input vectors, returning a vector with the same element type as
4469 the input and length that is the same as the shuffle mask.
4474 The first two operands of a '``shufflevector``' instruction are vectors
4475 with the same type. The third argument is a shuffle mask whose element
4476 type is always 'i32'. The result of the instruction is a vector whose
4477 length is the same as the shuffle mask and whose element type is the
4478 same as the element type of the first two operands.
4480 The shuffle mask operand is required to be a constant vector with either
4481 constant integer or undef values.
4486 The elements of the two input vectors are numbered from left to right
4487 across both of the vectors. The shuffle mask operand specifies, for each
4488 element of the result vector, which element of the two input vectors the
4489 result element gets. The element selector may be undef (meaning "don't
4490 care") and the second operand may be undef if performing a shuffle from
4496 .. code-block:: llvm
4498 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4499 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4500 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4501 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4502 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4503 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4504 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4505 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4507 Aggregate Operations
4508 --------------------
4510 LLVM supports several instructions for working with
4511 :ref:`aggregate <t_aggregate>` values.
4515 '``extractvalue``' Instruction
4516 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4523 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4528 The '``extractvalue``' instruction extracts the value of a member field
4529 from an :ref:`aggregate <t_aggregate>` value.
4534 The first operand of an '``extractvalue``' instruction is a value of
4535 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4536 constant indices to specify which value to extract in a similar manner
4537 as indices in a '``getelementptr``' instruction.
4539 The major differences to ``getelementptr`` indexing are:
4541 - Since the value being indexed is not a pointer, the first index is
4542 omitted and assumed to be zero.
4543 - At least one index must be specified.
4544 - Not only struct indices but also array indices must be in bounds.
4549 The result is the value at the position in the aggregate specified by
4555 .. code-block:: llvm
4557 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4561 '``insertvalue``' Instruction
4562 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4569 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4574 The '``insertvalue``' instruction inserts a value into a member field in
4575 an :ref:`aggregate <t_aggregate>` value.
4580 The first operand of an '``insertvalue``' instruction is a value of
4581 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4582 a first-class value to insert. The following operands are constant
4583 indices indicating the position at which to insert the value in a
4584 similar manner as indices in a '``extractvalue``' instruction. The value
4585 to insert must have the same type as the value identified by the
4591 The result is an aggregate of the same type as ``val``. Its value is
4592 that of ``val`` except that the value at the position specified by the
4593 indices is that of ``elt``.
4598 .. code-block:: llvm
4600 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4601 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4602 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4606 Memory Access and Addressing Operations
4607 ---------------------------------------
4609 A key design point of an SSA-based representation is how it represents
4610 memory. In LLVM, no memory locations are in SSA form, which makes things
4611 very simple. This section describes how to read, write, and allocate
4616 '``alloca``' Instruction
4617 ^^^^^^^^^^^^^^^^^^^^^^^^
4624 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4629 The '``alloca``' instruction allocates memory on the stack frame of the
4630 currently executing function, to be automatically released when this
4631 function returns to its caller. The object is always allocated in the
4632 generic address space (address space zero).
4637 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4638 bytes of memory on the runtime stack, returning a pointer of the
4639 appropriate type to the program. If "NumElements" is specified, it is
4640 the number of elements allocated, otherwise "NumElements" is defaulted
4641 to be one. If a constant alignment is specified, the value result of the
4642 allocation is guaranteed to be aligned to at least that boundary. If not
4643 specified, or if zero, the target can choose to align the allocation on
4644 any convenient boundary compatible with the type.
4646 '``type``' may be any sized type.
4651 Memory is allocated; a pointer is returned. The operation is undefined
4652 if there is insufficient stack space for the allocation. '``alloca``'d
4653 memory is automatically released when the function returns. The
4654 '``alloca``' instruction is commonly used to represent automatic
4655 variables that must have an address available. When the function returns
4656 (either with the ``ret`` or ``resume`` instructions), the memory is
4657 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4658 The order in which memory is allocated (ie., which way the stack grows)
4664 .. code-block:: llvm
4666 %ptr = alloca i32 ; yields {i32*}:ptr
4667 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4668 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4669 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4673 '``load``' Instruction
4674 ^^^^^^^^^^^^^^^^^^^^^^
4681 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4682 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4683 !<index> = !{ i32 1 }
4688 The '``load``' instruction is used to read from memory.
4693 The argument to the ``load`` instruction specifies the memory address
4694 from which to load. The pointer must point to a :ref:`first
4695 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4696 then the optimizer is not allowed to modify the number or order of
4697 execution of this ``load`` with other :ref:`volatile
4698 operations <volatile>`.
4700 If the ``load`` is marked as ``atomic``, it takes an extra
4701 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4702 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4703 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4704 when they may see multiple atomic stores. The type of the pointee must
4705 be an integer type whose bit width is a power of two greater than or
4706 equal to eight and less than or equal to a target-specific size limit.
4707 ``align`` must be explicitly specified on atomic loads, and the load has
4708 undefined behavior if the alignment is not set to a value which is at
4709 least the size in bytes of the pointee. ``!nontemporal`` does not have
4710 any defined semantics for atomic loads.
4712 The optional constant ``align`` argument specifies the alignment of the
4713 operation (that is, the alignment of the memory address). A value of 0
4714 or an omitted ``align`` argument means that the operation has the ABI
4715 alignment for the target. It is the responsibility of the code emitter
4716 to ensure that the alignment information is correct. Overestimating the
4717 alignment results in undefined behavior. Underestimating the alignment
4718 may produce less efficient code. An alignment of 1 is always safe.
4720 The optional ``!nontemporal`` metadata must reference a single
4721 metadata name ``<index>`` corresponding to a metadata node with one
4722 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4723 metadata on the instruction tells the optimizer and code generator
4724 that this load is not expected to be reused in the cache. The code
4725 generator may select special instructions to save cache bandwidth, such
4726 as the ``MOVNT`` instruction on x86.
4728 The optional ``!invariant.load`` metadata must reference a single
4729 metadata name ``<index>`` corresponding to a metadata node with no
4730 entries. The existence of the ``!invariant.load`` metadata on the
4731 instruction tells the optimizer and code generator that this load
4732 address points to memory which does not change value during program
4733 execution. The optimizer may then move this load around, for example, by
4734 hoisting it out of loops using loop invariant code motion.
4739 The location of memory pointed to is loaded. If the value being loaded
4740 is of scalar type then the number of bytes read does not exceed the
4741 minimum number of bytes needed to hold all bits of the type. For
4742 example, loading an ``i24`` reads at most three bytes. When loading a
4743 value of a type like ``i20`` with a size that is not an integral number
4744 of bytes, the result is undefined if the value was not originally
4745 written using a store of the same type.
4750 .. code-block:: llvm
4752 %ptr = alloca i32 ; yields {i32*}:ptr
4753 store i32 3, i32* %ptr ; yields {void}
4754 %val = load i32* %ptr ; yields {i32}:val = i32 3
4758 '``store``' Instruction
4759 ^^^^^^^^^^^^^^^^^^^^^^^
4766 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4767 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4772 The '``store``' instruction is used to write to memory.
4777 There are two arguments to the ``store`` instruction: a value to store
4778 and an address at which to store it. The type of the ``<pointer>``
4779 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4780 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4781 then the optimizer is not allowed to modify the number or order of
4782 execution of this ``store`` with other :ref:`volatile
4783 operations <volatile>`.
4785 If the ``store`` is marked as ``atomic``, it takes an extra
4786 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4787 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4788 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4789 when they may see multiple atomic stores. The type of the pointee must
4790 be an integer type whose bit width is a power of two greater than or
4791 equal to eight and less than or equal to a target-specific size limit.
4792 ``align`` must be explicitly specified on atomic stores, and the store
4793 has undefined behavior if the alignment is not set to a value which is
4794 at least the size in bytes of the pointee. ``!nontemporal`` does not
4795 have any defined semantics for atomic stores.
4797 The optional constant ``align`` argument specifies the alignment of the
4798 operation (that is, the alignment of the memory address). A value of 0
4799 or an omitted ``align`` argument means that the operation has the ABI
4800 alignment for the target. It is the responsibility of the code emitter
4801 to ensure that the alignment information is correct. Overestimating the
4802 alignment results in undefined behavior. Underestimating the
4803 alignment may produce less efficient code. An alignment of 1 is always
4806 The optional ``!nontemporal`` metadata must reference a single metadata
4807 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4808 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4809 tells the optimizer and code generator that this load is not expected to
4810 be reused in the cache. The code generator may select special
4811 instructions to save cache bandwidth, such as the MOVNT instruction on
4817 The contents of memory are updated to contain ``<value>`` at the
4818 location specified by the ``<pointer>`` operand. If ``<value>`` is
4819 of scalar type then the number of bytes written does not exceed the
4820 minimum number of bytes needed to hold all bits of the type. For
4821 example, storing an ``i24`` writes at most three bytes. When writing a
4822 value of a type like ``i20`` with a size that is not an integral number
4823 of bytes, it is unspecified what happens to the extra bits that do not
4824 belong to the type, but they will typically be overwritten.
4829 .. code-block:: llvm
4831 %ptr = alloca i32 ; yields {i32*}:ptr
4832 store i32 3, i32* %ptr ; yields {void}
4833 %val = load i32* %ptr ; yields {i32}:val = i32 3
4837 '``fence``' Instruction
4838 ^^^^^^^^^^^^^^^^^^^^^^^
4845 fence [singlethread] <ordering> ; yields {void}
4850 The '``fence``' instruction is used to introduce happens-before edges
4856 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4857 defines what *synchronizes-with* edges they add. They can only be given
4858 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4863 A fence A which has (at least) ``release`` ordering semantics
4864 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4865 semantics if and only if there exist atomic operations X and Y, both
4866 operating on some atomic object M, such that A is sequenced before X, X
4867 modifies M (either directly or through some side effect of a sequence
4868 headed by X), Y is sequenced before B, and Y observes M. This provides a
4869 *happens-before* dependency between A and B. Rather than an explicit
4870 ``fence``, one (but not both) of the atomic operations X or Y might
4871 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4872 still *synchronize-with* the explicit ``fence`` and establish the
4873 *happens-before* edge.
4875 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4876 ``acquire`` and ``release`` semantics specified above, participates in
4877 the global program order of other ``seq_cst`` operations and/or fences.
4879 The optional ":ref:`singlethread <singlethread>`" argument specifies
4880 that the fence only synchronizes with other fences in the same thread.
4881 (This is useful for interacting with signal handlers.)
4886 .. code-block:: llvm
4888 fence acquire ; yields {void}
4889 fence singlethread seq_cst ; yields {void}
4893 '``cmpxchg``' Instruction
4894 ^^^^^^^^^^^^^^^^^^^^^^^^^
4901 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4906 The '``cmpxchg``' instruction is used to atomically modify memory. It
4907 loads a value in memory and compares it to a given value. If they are
4908 equal, it stores a new value into the memory.
4913 There are three arguments to the '``cmpxchg``' instruction: an address
4914 to operate on, a value to compare to the value currently be at that
4915 address, and a new value to place at that address if the compared values
4916 are equal. The type of '<cmp>' must be an integer type whose bit width
4917 is a power of two greater than or equal to eight and less than or equal
4918 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4919 type, and the type of '<pointer>' must be a pointer to that type. If the
4920 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4921 to modify the number or order of execution of this ``cmpxchg`` with
4922 other :ref:`volatile operations <volatile>`.
4924 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4925 synchronizes with other atomic operations.
4927 The optional "``singlethread``" argument declares that the ``cmpxchg``
4928 is only atomic with respect to code (usually signal handlers) running in
4929 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4930 respect to all other code in the system.
4932 The pointer passed into cmpxchg must have alignment greater than or
4933 equal to the size in memory of the operand.
4938 The contents of memory at the location specified by the '``<pointer>``'
4939 operand is read and compared to '``<cmp>``'; if the read value is the
4940 equal, '``<new>``' is written. The original value at the location is
4943 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4944 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4945 atomic load with an ordering parameter determined by dropping any
4946 ``release`` part of the ``cmpxchg``'s ordering.
4951 .. code-block:: llvm
4954 %orig = atomic load i32* %ptr unordered ; yields {i32}
4958 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4959 %squared = mul i32 %cmp, %cmp
4960 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4961 %success = icmp eq i32 %cmp, %old
4962 br i1 %success, label %done, label %loop
4969 '``atomicrmw``' Instruction
4970 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4977 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4982 The '``atomicrmw``' instruction is used to atomically modify memory.
4987 There are three arguments to the '``atomicrmw``' instruction: an
4988 operation to apply, an address whose value to modify, an argument to the
4989 operation. The operation must be one of the following keywords:
5003 The type of '<value>' must be an integer type whose bit width is a power
5004 of two greater than or equal to eight and less than or equal to a
5005 target-specific size limit. The type of the '``<pointer>``' operand must
5006 be a pointer to that type. If the ``atomicrmw`` is marked as
5007 ``volatile``, then the optimizer is not allowed to modify the number or
5008 order of execution of this ``atomicrmw`` with other :ref:`volatile
5009 operations <volatile>`.
5014 The contents of memory at the location specified by the '``<pointer>``'
5015 operand are atomically read, modified, and written back. The original
5016 value at the location is returned. The modification is specified by the
5019 - xchg: ``*ptr = val``
5020 - add: ``*ptr = *ptr + val``
5021 - sub: ``*ptr = *ptr - val``
5022 - and: ``*ptr = *ptr & val``
5023 - nand: ``*ptr = ~(*ptr & val)``
5024 - or: ``*ptr = *ptr | val``
5025 - xor: ``*ptr = *ptr ^ val``
5026 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5027 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5028 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5030 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5036 .. code-block:: llvm
5038 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5040 .. _i_getelementptr:
5042 '``getelementptr``' Instruction
5043 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5050 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5051 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5052 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5057 The '``getelementptr``' instruction is used to get the address of a
5058 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5059 address calculation only and does not access memory.
5064 The first argument is always a pointer or a vector of pointers, and
5065 forms the basis of the calculation. The remaining arguments are indices
5066 that indicate which of the elements of the aggregate object are indexed.
5067 The interpretation of each index is dependent on the type being indexed
5068 into. The first index always indexes the pointer value given as the
5069 first argument, the second index indexes a value of the type pointed to
5070 (not necessarily the value directly pointed to, since the first index
5071 can be non-zero), etc. The first type indexed into must be a pointer
5072 value, subsequent types can be arrays, vectors, and structs. Note that
5073 subsequent types being indexed into can never be pointers, since that
5074 would require loading the pointer before continuing calculation.
5076 The type of each index argument depends on the type it is indexing into.
5077 When indexing into a (optionally packed) structure, only ``i32`` integer
5078 **constants** are allowed (when using a vector of indices they must all
5079 be the **same** ``i32`` integer constant). When indexing into an array,
5080 pointer or vector, integers of any width are allowed, and they are not
5081 required to be constant. These integers are treated as signed values
5084 For example, let's consider a C code fragment and how it gets compiled
5100 int *foo(struct ST *s) {
5101 return &s[1].Z.B[5][13];
5104 The LLVM code generated by Clang is:
5106 .. code-block:: llvm
5108 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5109 %struct.ST = type { i32, double, %struct.RT }
5111 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5113 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5120 In the example above, the first index is indexing into the
5121 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5122 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5123 indexes into the third element of the structure, yielding a
5124 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5125 structure. The third index indexes into the second element of the
5126 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5127 dimensions of the array are subscripted into, yielding an '``i32``'
5128 type. The '``getelementptr``' instruction returns a pointer to this
5129 element, thus computing a value of '``i32*``' type.
5131 Note that it is perfectly legal to index partially through a structure,
5132 returning a pointer to an inner element. Because of this, the LLVM code
5133 for the given testcase is equivalent to:
5135 .. code-block:: llvm
5137 define i32* @foo(%struct.ST* %s) {
5138 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5139 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5140 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5141 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5142 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5146 If the ``inbounds`` keyword is present, the result value of the
5147 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5148 pointer is not an *in bounds* address of an allocated object, or if any
5149 of the addresses that would be formed by successive addition of the
5150 offsets implied by the indices to the base address with infinitely
5151 precise signed arithmetic are not an *in bounds* address of that
5152 allocated object. The *in bounds* addresses for an allocated object are
5153 all the addresses that point into the object, plus the address one byte
5154 past the end. In cases where the base is a vector of pointers the
5155 ``inbounds`` keyword applies to each of the computations element-wise.
5157 If the ``inbounds`` keyword is not present, the offsets are added to the
5158 base address with silently-wrapping two's complement arithmetic. If the
5159 offsets have a different width from the pointer, they are sign-extended
5160 or truncated to the width of the pointer. The result value of the
5161 ``getelementptr`` may be outside the object pointed to by the base
5162 pointer. The result value may not necessarily be used to access memory
5163 though, even if it happens to point into allocated storage. See the
5164 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5167 The getelementptr instruction is often confusing. For some more insight
5168 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5173 .. code-block:: llvm
5175 ; yields [12 x i8]*:aptr
5176 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5178 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5180 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5182 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5184 In cases where the pointer argument is a vector of pointers, each index
5185 must be a vector with the same number of elements. For example:
5187 .. code-block:: llvm
5189 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5191 Conversion Operations
5192 ---------------------
5194 The instructions in this category are the conversion instructions
5195 (casting) which all take a single operand and a type. They perform
5196 various bit conversions on the operand.
5198 '``trunc .. to``' Instruction
5199 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5206 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5211 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5216 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5217 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5218 of the same number of integers. The bit size of the ``value`` must be
5219 larger than the bit size of the destination type, ``ty2``. Equal sized
5220 types are not allowed.
5225 The '``trunc``' instruction truncates the high order bits in ``value``
5226 and converts the remaining bits to ``ty2``. Since the source size must
5227 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5228 It will always truncate bits.
5233 .. code-block:: llvm
5235 %X = trunc i32 257 to i8 ; yields i8:1
5236 %Y = trunc i32 123 to i1 ; yields i1:true
5237 %Z = trunc i32 122 to i1 ; yields i1:false
5238 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5240 '``zext .. to``' Instruction
5241 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5248 <result> = zext <ty> <value> to <ty2> ; yields ty2
5253 The '``zext``' instruction zero extends its operand to type ``ty2``.
5258 The '``zext``' instruction takes a value to cast, and a type to cast it
5259 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5260 the same number of integers. The bit size of the ``value`` must be
5261 smaller than the bit size of the destination type, ``ty2``.
5266 The ``zext`` fills the high order bits of the ``value`` with zero bits
5267 until it reaches the size of the destination type, ``ty2``.
5269 When zero extending from i1, the result will always be either 0 or 1.
5274 .. code-block:: llvm
5276 %X = zext i32 257 to i64 ; yields i64:257
5277 %Y = zext i1 true to i32 ; yields i32:1
5278 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5280 '``sext .. to``' Instruction
5281 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5288 <result> = sext <ty> <value> to <ty2> ; yields ty2
5293 The '``sext``' sign extends ``value`` to the type ``ty2``.
5298 The '``sext``' instruction takes a value to cast, and a type to cast it
5299 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5300 the same number of integers. The bit size of the ``value`` must be
5301 smaller than the bit size of the destination type, ``ty2``.
5306 The '``sext``' instruction performs a sign extension by copying the sign
5307 bit (highest order bit) of the ``value`` until it reaches the bit size
5308 of the type ``ty2``.
5310 When sign extending from i1, the extension always results in -1 or 0.
5315 .. code-block:: llvm
5317 %X = sext i8 -1 to i16 ; yields i16 :65535
5318 %Y = sext i1 true to i32 ; yields i32:-1
5319 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5321 '``fptrunc .. to``' Instruction
5322 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5329 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5334 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5339 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5340 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5341 The size of ``value`` must be larger than the size of ``ty2``. This
5342 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5347 The '``fptrunc``' instruction truncates a ``value`` from a larger
5348 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5349 point <t_floating>` type. If the value cannot fit within the
5350 destination type, ``ty2``, then the results are undefined.
5355 .. code-block:: llvm
5357 %X = fptrunc double 123.0 to float ; yields float:123.0
5358 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5360 '``fpext .. to``' Instruction
5361 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5368 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5373 The '``fpext``' extends a floating point ``value`` to a larger floating
5379 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5380 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5381 to. The source type must be smaller than the destination type.
5386 The '``fpext``' instruction extends the ``value`` from a smaller
5387 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5388 point <t_floating>` type. The ``fpext`` cannot be used to make a
5389 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5390 *no-op cast* for a floating point cast.
5395 .. code-block:: llvm
5397 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5398 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5400 '``fptoui .. to``' Instruction
5401 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5408 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5413 The '``fptoui``' converts a floating point ``value`` to its unsigned
5414 integer equivalent of type ``ty2``.
5419 The '``fptoui``' instruction takes a value to cast, which must be a
5420 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5421 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5422 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5423 type with the same number of elements as ``ty``
5428 The '``fptoui``' instruction converts its :ref:`floating
5429 point <t_floating>` operand into the nearest (rounding towards zero)
5430 unsigned integer value. If the value cannot fit in ``ty2``, the results
5436 .. code-block:: llvm
5438 %X = fptoui double 123.0 to i32 ; yields i32:123
5439 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5440 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5442 '``fptosi .. to``' Instruction
5443 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5450 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5455 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5456 ``value`` to type ``ty2``.
5461 The '``fptosi``' instruction takes a value to cast, which must be a
5462 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5463 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5464 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5465 type with the same number of elements as ``ty``
5470 The '``fptosi``' instruction converts its :ref:`floating
5471 point <t_floating>` operand into the nearest (rounding towards zero)
5472 signed integer value. If the value cannot fit in ``ty2``, the results
5478 .. code-block:: llvm
5480 %X = fptosi double -123.0 to i32 ; yields i32:-123
5481 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5482 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5484 '``uitofp .. to``' Instruction
5485 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5492 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5497 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5498 and converts that value to the ``ty2`` type.
5503 The '``uitofp``' instruction takes a value to cast, which must be a
5504 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5505 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5506 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5507 type with the same number of elements as ``ty``
5512 The '``uitofp``' instruction interprets its operand as an unsigned
5513 integer quantity and converts it to the corresponding floating point
5514 value. If the value cannot fit in the floating point value, the results
5520 .. code-block:: llvm
5522 %X = uitofp i32 257 to float ; yields float:257.0
5523 %Y = uitofp i8 -1 to double ; yields double:255.0
5525 '``sitofp .. to``' Instruction
5526 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5533 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5538 The '``sitofp``' instruction regards ``value`` as a signed integer and
5539 converts that value to the ``ty2`` type.
5544 The '``sitofp``' instruction takes a value to cast, which must be a
5545 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5546 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5547 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5548 type with the same number of elements as ``ty``
5553 The '``sitofp``' instruction interprets its operand as a signed integer
5554 quantity and converts it to the corresponding floating point value. If
5555 the value cannot fit in the floating point value, the results are
5561 .. code-block:: llvm
5563 %X = sitofp i32 257 to float ; yields float:257.0
5564 %Y = sitofp i8 -1 to double ; yields double:-1.0
5568 '``ptrtoint .. to``' Instruction
5569 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5576 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5581 The '``ptrtoint``' instruction converts the pointer or a vector of
5582 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5587 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5588 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5589 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5590 a vector of integers type.
5595 The '``ptrtoint``' instruction converts ``value`` to integer type
5596 ``ty2`` by interpreting the pointer value as an integer and either
5597 truncating or zero extending that value to the size of the integer type.
5598 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5599 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5600 the same size, then nothing is done (*no-op cast*) other than a type
5606 .. code-block:: llvm
5608 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5609 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5610 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5614 '``inttoptr .. to``' Instruction
5615 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5622 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5627 The '``inttoptr``' instruction converts an integer ``value`` to a
5628 pointer type, ``ty2``.
5633 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5634 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5640 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5641 applying either a zero extension or a truncation depending on the size
5642 of the integer ``value``. If ``value`` is larger than the size of a
5643 pointer then a truncation is done. If ``value`` is smaller than the size
5644 of a pointer then a zero extension is done. If they are the same size,
5645 nothing is done (*no-op cast*).
5650 .. code-block:: llvm
5652 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5653 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5654 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5655 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5659 '``bitcast .. to``' Instruction
5660 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5667 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5672 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5678 The '``bitcast``' instruction takes a value to cast, which must be a
5679 non-aggregate first class value, and a type to cast it to, which must
5680 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5681 bit sizes of ``value`` and the destination type, ``ty2``, must be
5682 identical. If the source type is a pointer, the destination type must
5683 also be a pointer of the same size. This instruction supports bitwise
5684 conversion of vectors to integers and to vectors of other types (as
5685 long as they have the same size).
5690 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5691 is always a *no-op cast* because no bits change with this
5692 conversion. The conversion is done as if the ``value`` had been stored
5693 to memory and read back as type ``ty2``. Pointer (or vector of
5694 pointers) types may only be converted to other pointer (or vector of
5695 pointers) types with the same address space through this instruction.
5696 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5697 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5702 .. code-block:: llvm
5704 %X = bitcast i8 255 to i8 ; yields i8 :-1
5705 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5706 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5707 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5709 .. _i_addrspacecast:
5711 '``addrspacecast .. to``' Instruction
5712 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5719 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5724 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5725 address space ``n`` to type ``pty2`` in address space ``m``.
5730 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5731 to cast and a pointer type to cast it to, which must have a different
5737 The '``addrspacecast``' instruction converts the pointer value
5738 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5739 value modification, depending on the target and the address space
5740 pair. Pointer conversions within the same address space must be
5741 performed with the ``bitcast`` instruction. Note that if the address space
5742 conversion is legal then both result and operand refer to the same memory
5748 .. code-block:: llvm
5750 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5751 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5752 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5759 The instructions in this category are the "miscellaneous" instructions,
5760 which defy better classification.
5764 '``icmp``' Instruction
5765 ^^^^^^^^^^^^^^^^^^^^^^
5772 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5777 The '``icmp``' instruction returns a boolean value or a vector of
5778 boolean values based on comparison of its two integer, integer vector,
5779 pointer, or pointer vector operands.
5784 The '``icmp``' instruction takes three operands. The first operand is
5785 the condition code indicating the kind of comparison to perform. It is
5786 not a value, just a keyword. The possible condition code are:
5789 #. ``ne``: not equal
5790 #. ``ugt``: unsigned greater than
5791 #. ``uge``: unsigned greater or equal
5792 #. ``ult``: unsigned less than
5793 #. ``ule``: unsigned less or equal
5794 #. ``sgt``: signed greater than
5795 #. ``sge``: signed greater or equal
5796 #. ``slt``: signed less than
5797 #. ``sle``: signed less or equal
5799 The remaining two arguments must be :ref:`integer <t_integer>` or
5800 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5801 must also be identical types.
5806 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5807 code given as ``cond``. The comparison performed always yields either an
5808 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5810 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5811 otherwise. No sign interpretation is necessary or performed.
5812 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5813 otherwise. No sign interpretation is necessary or performed.
5814 #. ``ugt``: interprets the operands as unsigned values and yields
5815 ``true`` if ``op1`` is greater than ``op2``.
5816 #. ``uge``: interprets the operands as unsigned values and yields
5817 ``true`` if ``op1`` is greater than or equal to ``op2``.
5818 #. ``ult``: interprets the operands as unsigned values and yields
5819 ``true`` if ``op1`` is less than ``op2``.
5820 #. ``ule``: interprets the operands as unsigned values and yields
5821 ``true`` if ``op1`` is less than or equal to ``op2``.
5822 #. ``sgt``: interprets the operands as signed values and yields ``true``
5823 if ``op1`` is greater than ``op2``.
5824 #. ``sge``: interprets the operands as signed values and yields ``true``
5825 if ``op1`` is greater than or equal to ``op2``.
5826 #. ``slt``: interprets the operands as signed values and yields ``true``
5827 if ``op1`` is less than ``op2``.
5828 #. ``sle``: interprets the operands as signed values and yields ``true``
5829 if ``op1`` is less than or equal to ``op2``.
5831 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5832 are compared as if they were integers.
5834 If the operands are integer vectors, then they are compared element by
5835 element. The result is an ``i1`` vector with the same number of elements
5836 as the values being compared. Otherwise, the result is an ``i1``.
5841 .. code-block:: llvm
5843 <result> = icmp eq i32 4, 5 ; yields: result=false
5844 <result> = icmp ne float* %X, %X ; yields: result=false
5845 <result> = icmp ult i16 4, 5 ; yields: result=true
5846 <result> = icmp sgt i16 4, 5 ; yields: result=false
5847 <result> = icmp ule i16 -4, 5 ; yields: result=false
5848 <result> = icmp sge i16 4, 5 ; yields: result=false
5850 Note that the code generator does not yet support vector types with the
5851 ``icmp`` instruction.
5855 '``fcmp``' Instruction
5856 ^^^^^^^^^^^^^^^^^^^^^^
5863 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5868 The '``fcmp``' instruction returns a boolean value or vector of boolean
5869 values based on comparison of its operands.
5871 If the operands are floating point scalars, then the result type is a
5872 boolean (:ref:`i1 <t_integer>`).
5874 If the operands are floating point vectors, then the result type is a
5875 vector of boolean with the same number of elements as the operands being
5881 The '``fcmp``' instruction takes three operands. The first operand is
5882 the condition code indicating the kind of comparison to perform. It is
5883 not a value, just a keyword. The possible condition code are:
5885 #. ``false``: no comparison, always returns false
5886 #. ``oeq``: ordered and equal
5887 #. ``ogt``: ordered and greater than
5888 #. ``oge``: ordered and greater than or equal
5889 #. ``olt``: ordered and less than
5890 #. ``ole``: ordered and less than or equal
5891 #. ``one``: ordered and not equal
5892 #. ``ord``: ordered (no nans)
5893 #. ``ueq``: unordered or equal
5894 #. ``ugt``: unordered or greater than
5895 #. ``uge``: unordered or greater than or equal
5896 #. ``ult``: unordered or less than
5897 #. ``ule``: unordered or less than or equal
5898 #. ``une``: unordered or not equal
5899 #. ``uno``: unordered (either nans)
5900 #. ``true``: no comparison, always returns true
5902 *Ordered* means that neither operand is a QNAN while *unordered* means
5903 that either operand may be a QNAN.
5905 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5906 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5907 type. They must have identical types.
5912 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5913 condition code given as ``cond``. If the operands are vectors, then the
5914 vectors are compared element by element. Each comparison performed
5915 always yields an :ref:`i1 <t_integer>` result, as follows:
5917 #. ``false``: always yields ``false``, regardless of operands.
5918 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5919 is equal to ``op2``.
5920 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5921 is greater than ``op2``.
5922 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5923 is greater than or equal to ``op2``.
5924 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5925 is less than ``op2``.
5926 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5927 is less than or equal to ``op2``.
5928 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5929 is not equal to ``op2``.
5930 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5931 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5933 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5934 greater than ``op2``.
5935 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5936 greater than or equal to ``op2``.
5937 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5939 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5940 less than or equal to ``op2``.
5941 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5942 not equal to ``op2``.
5943 #. ``uno``: yields ``true`` if either operand is a QNAN.
5944 #. ``true``: always yields ``true``, regardless of operands.
5949 .. code-block:: llvm
5951 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5952 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5953 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5954 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5956 Note that the code generator does not yet support vector types with the
5957 ``fcmp`` instruction.
5961 '``phi``' Instruction
5962 ^^^^^^^^^^^^^^^^^^^^^
5969 <result> = phi <ty> [ <val0>, <label0>], ...
5974 The '``phi``' instruction is used to implement the φ node in the SSA
5975 graph representing the function.
5980 The type of the incoming values is specified with the first type field.
5981 After this, the '``phi``' instruction takes a list of pairs as
5982 arguments, with one pair for each predecessor basic block of the current
5983 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5984 the value arguments to the PHI node. Only labels may be used as the
5987 There must be no non-phi instructions between the start of a basic block
5988 and the PHI instructions: i.e. PHI instructions must be first in a basic
5991 For the purposes of the SSA form, the use of each incoming value is
5992 deemed to occur on the edge from the corresponding predecessor block to
5993 the current block (but after any definition of an '``invoke``'
5994 instruction's return value on the same edge).
5999 At runtime, the '``phi``' instruction logically takes on the value
6000 specified by the pair corresponding to the predecessor basic block that
6001 executed just prior to the current block.
6006 .. code-block:: llvm
6008 Loop: ; Infinite loop that counts from 0 on up...
6009 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6010 %nextindvar = add i32 %indvar, 1
6015 '``select``' Instruction
6016 ^^^^^^^^^^^^^^^^^^^^^^^^
6023 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6025 selty is either i1 or {<N x i1>}
6030 The '``select``' instruction is used to choose one value based on a
6031 condition, without branching.
6036 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6037 values indicating the condition, and two values of the same :ref:`first
6038 class <t_firstclass>` type. If the val1/val2 are vectors and the
6039 condition is a scalar, then entire vectors are selected, not individual
6045 If the condition is an i1 and it evaluates to 1, the instruction returns
6046 the first value argument; otherwise, it returns the second value
6049 If the condition is a vector of i1, then the value arguments must be
6050 vectors of the same size, and the selection is done element by element.
6055 .. code-block:: llvm
6057 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6061 '``call``' Instruction
6062 ^^^^^^^^^^^^^^^^^^^^^^
6069 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6074 The '``call``' instruction represents a simple function call.
6079 This instruction requires several arguments:
6081 #. The optional "tail" marker indicates that the callee function does
6082 not access any allocas or varargs in the caller. Note that calls may
6083 be marked "tail" even if they do not occur before a
6084 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6085 function call is eligible for tail call optimization, but `might not
6086 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6087 The code generator may optimize calls marked "tail" with either 1)
6088 automatic `sibling call
6089 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6090 callee have matching signatures, or 2) forced tail call optimization
6091 when the following extra requirements are met:
6093 - Caller and callee both have the calling convention ``fastcc``.
6094 - The call is in tail position (ret immediately follows call and ret
6095 uses value of call or is void).
6096 - Option ``-tailcallopt`` is enabled, or
6097 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6098 - `Platform specific constraints are
6099 met. <CodeGenerator.html#tailcallopt>`_
6101 #. The optional "cconv" marker indicates which :ref:`calling
6102 convention <callingconv>` the call should use. If none is
6103 specified, the call defaults to using C calling conventions. The
6104 calling convention of the call must match the calling convention of
6105 the target function, or else the behavior is undefined.
6106 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6107 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6109 #. '``ty``': the type of the call instruction itself which is also the
6110 type of the return value. Functions that return no value are marked
6112 #. '``fnty``': shall be the signature of the pointer to function value
6113 being invoked. The argument types must match the types implied by
6114 this signature. This type can be omitted if the function is not
6115 varargs and if the function type does not return a pointer to a
6117 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6118 be invoked. In most cases, this is a direct function invocation, but
6119 indirect ``call``'s are just as possible, calling an arbitrary pointer
6121 #. '``function args``': argument list whose types match the function
6122 signature argument types and parameter attributes. All arguments must
6123 be of :ref:`first class <t_firstclass>` type. If the function signature
6124 indicates the function accepts a variable number of arguments, the
6125 extra arguments can be specified.
6126 #. The optional :ref:`function attributes <fnattrs>` list. Only
6127 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6128 attributes are valid here.
6133 The '``call``' instruction is used to cause control flow to transfer to
6134 a specified function, with its incoming arguments bound to the specified
6135 values. Upon a '``ret``' instruction in the called function, control
6136 flow continues with the instruction after the function call, and the
6137 return value of the function is bound to the result argument.
6142 .. code-block:: llvm
6144 %retval = call i32 @test(i32 %argc)
6145 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6146 %X = tail call i32 @foo() ; yields i32
6147 %Y = tail call fastcc i32 @foo() ; yields i32
6148 call void %foo(i8 97 signext)
6150 %struct.A = type { i32, i8 }
6151 %r = call %struct.A @foo() ; yields { 32, i8 }
6152 %gr = extractvalue %struct.A %r, 0 ; yields i32
6153 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6154 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6155 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6157 llvm treats calls to some functions with names and arguments that match
6158 the standard C99 library as being the C99 library functions, and may
6159 perform optimizations or generate code for them under that assumption.
6160 This is something we'd like to change in the future to provide better
6161 support for freestanding environments and non-C-based languages.
6165 '``va_arg``' Instruction
6166 ^^^^^^^^^^^^^^^^^^^^^^^^
6173 <resultval> = va_arg <va_list*> <arglist>, <argty>
6178 The '``va_arg``' instruction is used to access arguments passed through
6179 the "variable argument" area of a function call. It is used to implement
6180 the ``va_arg`` macro in C.
6185 This instruction takes a ``va_list*`` value and the type of the
6186 argument. It returns a value of the specified argument type and
6187 increments the ``va_list`` to point to the next argument. The actual
6188 type of ``va_list`` is target specific.
6193 The '``va_arg``' instruction loads an argument of the specified type
6194 from the specified ``va_list`` and causes the ``va_list`` to point to
6195 the next argument. For more information, see the variable argument
6196 handling :ref:`Intrinsic Functions <int_varargs>`.
6198 It is legal for this instruction to be called in a function which does
6199 not take a variable number of arguments, for example, the ``vfprintf``
6202 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6203 function <intrinsics>` because it takes a type as an argument.
6208 See the :ref:`variable argument processing <int_varargs>` section.
6210 Note that the code generator does not yet fully support va\_arg on many
6211 targets. Also, it does not currently support va\_arg with aggregate
6212 types on any target.
6216 '``landingpad``' Instruction
6217 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6224 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6225 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6227 <clause> := catch <type> <value>
6228 <clause> := filter <array constant type> <array constant>
6233 The '``landingpad``' instruction is used by `LLVM's exception handling
6234 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6235 is a landing pad --- one where the exception lands, and corresponds to the
6236 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6237 defines values supplied by the personality function (``pers_fn``) upon
6238 re-entry to the function. The ``resultval`` has the type ``resultty``.
6243 This instruction takes a ``pers_fn`` value. This is the personality
6244 function associated with the unwinding mechanism. The optional
6245 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6247 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6248 contains the global variable representing the "type" that may be caught
6249 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6250 clause takes an array constant as its argument. Use
6251 "``[0 x i8**] undef``" for a filter which cannot throw. The
6252 '``landingpad``' instruction must contain *at least* one ``clause`` or
6253 the ``cleanup`` flag.
6258 The '``landingpad``' instruction defines the values which are set by the
6259 personality function (``pers_fn``) upon re-entry to the function, and
6260 therefore the "result type" of the ``landingpad`` instruction. As with
6261 calling conventions, how the personality function results are
6262 represented in LLVM IR is target specific.
6264 The clauses are applied in order from top to bottom. If two
6265 ``landingpad`` instructions are merged together through inlining, the
6266 clauses from the calling function are appended to the list of clauses.
6267 When the call stack is being unwound due to an exception being thrown,
6268 the exception is compared against each ``clause`` in turn. If it doesn't
6269 match any of the clauses, and the ``cleanup`` flag is not set, then
6270 unwinding continues further up the call stack.
6272 The ``landingpad`` instruction has several restrictions:
6274 - A landing pad block is a basic block which is the unwind destination
6275 of an '``invoke``' instruction.
6276 - A landing pad block must have a '``landingpad``' instruction as its
6277 first non-PHI instruction.
6278 - There can be only one '``landingpad``' instruction within the landing
6280 - A basic block that is not a landing pad block may not include a
6281 '``landingpad``' instruction.
6282 - All '``landingpad``' instructions in a function must have the same
6283 personality function.
6288 .. code-block:: llvm
6290 ;; A landing pad which can catch an integer.
6291 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6293 ;; A landing pad that is a cleanup.
6294 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6296 ;; A landing pad which can catch an integer and can only throw a double.
6297 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6299 filter [1 x i8**] [@_ZTId]
6306 LLVM supports the notion of an "intrinsic function". These functions
6307 have well known names and semantics and are required to follow certain
6308 restrictions. Overall, these intrinsics represent an extension mechanism
6309 for the LLVM language that does not require changing all of the
6310 transformations in LLVM when adding to the language (or the bitcode
6311 reader/writer, the parser, etc...).
6313 Intrinsic function names must all start with an "``llvm.``" prefix. This
6314 prefix is reserved in LLVM for intrinsic names; thus, function names may
6315 not begin with this prefix. Intrinsic functions must always be external
6316 functions: you cannot define the body of intrinsic functions. Intrinsic
6317 functions may only be used in call or invoke instructions: it is illegal
6318 to take the address of an intrinsic function. Additionally, because
6319 intrinsic functions are part of the LLVM language, it is required if any
6320 are added that they be documented here.
6322 Some intrinsic functions can be overloaded, i.e., the intrinsic
6323 represents a family of functions that perform the same operation but on
6324 different data types. Because LLVM can represent over 8 million
6325 different integer types, overloading is used commonly to allow an
6326 intrinsic function to operate on any integer type. One or more of the
6327 argument types or the result type can be overloaded to accept any
6328 integer type. Argument types may also be defined as exactly matching a
6329 previous argument's type or the result type. This allows an intrinsic
6330 function which accepts multiple arguments, but needs all of them to be
6331 of the same type, to only be overloaded with respect to a single
6332 argument or the result.
6334 Overloaded intrinsics will have the names of its overloaded argument
6335 types encoded into its function name, each preceded by a period. Only
6336 those types which are overloaded result in a name suffix. Arguments
6337 whose type is matched against another type do not. For example, the
6338 ``llvm.ctpop`` function can take an integer of any width and returns an
6339 integer of exactly the same integer width. This leads to a family of
6340 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6341 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6342 overloaded, and only one type suffix is required. Because the argument's
6343 type is matched against the return type, it does not require its own
6346 To learn how to add an intrinsic function, please see the `Extending
6347 LLVM Guide <ExtendingLLVM.html>`_.
6351 Variable Argument Handling Intrinsics
6352 -------------------------------------
6354 Variable argument support is defined in LLVM with the
6355 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6356 functions. These functions are related to the similarly named macros
6357 defined in the ``<stdarg.h>`` header file.
6359 All of these functions operate on arguments that use a target-specific
6360 value type "``va_list``". The LLVM assembly language reference manual
6361 does not define what this type is, so all transformations should be
6362 prepared to handle these functions regardless of the type used.
6364 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6365 variable argument handling intrinsic functions are used.
6367 .. code-block:: llvm
6369 define i32 @test(i32 %X, ...) {
6370 ; Initialize variable argument processing
6372 %ap2 = bitcast i8** %ap to i8*
6373 call void @llvm.va_start(i8* %ap2)
6375 ; Read a single integer argument
6376 %tmp = va_arg i8** %ap, i32
6378 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6380 %aq2 = bitcast i8** %aq to i8*
6381 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6382 call void @llvm.va_end(i8* %aq2)
6384 ; Stop processing of arguments.
6385 call void @llvm.va_end(i8* %ap2)
6389 declare void @llvm.va_start(i8*)
6390 declare void @llvm.va_copy(i8*, i8*)
6391 declare void @llvm.va_end(i8*)
6395 '``llvm.va_start``' Intrinsic
6396 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6403 declare void @llvm.va_start(i8* <arglist>)
6408 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6409 subsequent use by ``va_arg``.
6414 The argument is a pointer to a ``va_list`` element to initialize.
6419 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6420 available in C. In a target-dependent way, it initializes the
6421 ``va_list`` element to which the argument points, so that the next call
6422 to ``va_arg`` will produce the first variable argument passed to the
6423 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6424 to know the last argument of the function as the compiler can figure
6427 '``llvm.va_end``' Intrinsic
6428 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6435 declare void @llvm.va_end(i8* <arglist>)
6440 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6441 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6446 The argument is a pointer to a ``va_list`` to destroy.
6451 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6452 available in C. In a target-dependent way, it destroys the ``va_list``
6453 element to which the argument points. Calls to
6454 :ref:`llvm.va_start <int_va_start>` and
6455 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6460 '``llvm.va_copy``' Intrinsic
6461 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6468 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6473 The '``llvm.va_copy``' intrinsic copies the current argument position
6474 from the source argument list to the destination argument list.
6479 The first argument is a pointer to a ``va_list`` element to initialize.
6480 The second argument is a pointer to a ``va_list`` element to copy from.
6485 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6486 available in C. In a target-dependent way, it copies the source
6487 ``va_list`` element into the destination ``va_list`` element. This
6488 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6489 arbitrarily complex and require, for example, memory allocation.
6491 Accurate Garbage Collection Intrinsics
6492 --------------------------------------
6494 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6495 (GC) requires the implementation and generation of these intrinsics.
6496 These intrinsics allow identification of :ref:`GC roots on the
6497 stack <int_gcroot>`, as well as garbage collector implementations that
6498 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6499 Front-ends for type-safe garbage collected languages should generate
6500 these intrinsics to make use of the LLVM garbage collectors. For more
6501 details, see `Accurate Garbage Collection with
6502 LLVM <GarbageCollection.html>`_.
6504 The garbage collection intrinsics only operate on objects in the generic
6505 address space (address space zero).
6509 '``llvm.gcroot``' Intrinsic
6510 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6517 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6522 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6523 the code generator, and allows some metadata to be associated with it.
6528 The first argument specifies the address of a stack object that contains
6529 the root pointer. The second pointer (which must be either a constant or
6530 a global value address) contains the meta-data to be associated with the
6536 At runtime, a call to this intrinsic stores a null pointer into the
6537 "ptrloc" location. At compile-time, the code generator generates
6538 information to allow the runtime to find the pointer at GC safe points.
6539 The '``llvm.gcroot``' intrinsic may only be used in a function which
6540 :ref:`specifies a GC algorithm <gc>`.
6544 '``llvm.gcread``' Intrinsic
6545 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6552 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6557 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6558 locations, allowing garbage collector implementations that require read
6564 The second argument is the address to read from, which should be an
6565 address allocated from the garbage collector. The first object is a
6566 pointer to the start of the referenced object, if needed by the language
6567 runtime (otherwise null).
6572 The '``llvm.gcread``' intrinsic has the same semantics as a load
6573 instruction, but may be replaced with substantially more complex code by
6574 the garbage collector runtime, as needed. The '``llvm.gcread``'
6575 intrinsic may only be used in a function which :ref:`specifies a GC
6580 '``llvm.gcwrite``' Intrinsic
6581 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6588 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6593 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6594 locations, allowing garbage collector implementations that require write
6595 barriers (such as generational or reference counting collectors).
6600 The first argument is the reference to store, the second is the start of
6601 the object to store it to, and the third is the address of the field of
6602 Obj to store to. If the runtime does not require a pointer to the
6603 object, Obj may be null.
6608 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6609 instruction, but may be replaced with substantially more complex code by
6610 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6611 intrinsic may only be used in a function which :ref:`specifies a GC
6614 Code Generator Intrinsics
6615 -------------------------
6617 These intrinsics are provided by LLVM to expose special features that
6618 may only be implemented with code generator support.
6620 '``llvm.returnaddress``' Intrinsic
6621 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6628 declare i8 *@llvm.returnaddress(i32 <level>)
6633 The '``llvm.returnaddress``' intrinsic attempts to compute a
6634 target-specific value indicating the return address of the current
6635 function or one of its callers.
6640 The argument to this intrinsic indicates which function to return the
6641 address for. Zero indicates the calling function, one indicates its
6642 caller, etc. The argument is **required** to be a constant integer
6648 The '``llvm.returnaddress``' intrinsic either returns a pointer
6649 indicating the return address of the specified call frame, or zero if it
6650 cannot be identified. The value returned by this intrinsic is likely to
6651 be incorrect or 0 for arguments other than zero, so it should only be
6652 used for debugging purposes.
6654 Note that calling this intrinsic does not prevent function inlining or
6655 other aggressive transformations, so the value returned may not be that
6656 of the obvious source-language caller.
6658 '``llvm.frameaddress``' Intrinsic
6659 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6666 declare i8* @llvm.frameaddress(i32 <level>)
6671 The '``llvm.frameaddress``' intrinsic attempts to return the
6672 target-specific frame pointer value for the specified stack frame.
6677 The argument to this intrinsic indicates which function to return the
6678 frame pointer for. Zero indicates the calling function, one indicates
6679 its caller, etc. The argument is **required** to be a constant integer
6685 The '``llvm.frameaddress``' intrinsic either returns a pointer
6686 indicating the frame address of the specified call frame, or zero if it
6687 cannot be identified. The value returned by this intrinsic is likely to
6688 be incorrect or 0 for arguments other than zero, so it should only be
6689 used for debugging purposes.
6691 Note that calling this intrinsic does not prevent function inlining or
6692 other aggressive transformations, so the value returned may not be that
6693 of the obvious source-language caller.
6697 '``llvm.stacksave``' Intrinsic
6698 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6705 declare i8* @llvm.stacksave()
6710 The '``llvm.stacksave``' intrinsic is used to remember the current state
6711 of the function stack, for use with
6712 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6713 implementing language features like scoped automatic variable sized
6719 This intrinsic returns a opaque pointer value that can be passed to
6720 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6721 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6722 ``llvm.stacksave``, it effectively restores the state of the stack to
6723 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6724 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6725 were allocated after the ``llvm.stacksave`` was executed.
6727 .. _int_stackrestore:
6729 '``llvm.stackrestore``' Intrinsic
6730 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6737 declare void @llvm.stackrestore(i8* %ptr)
6742 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6743 the function stack to the state it was in when the corresponding
6744 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6745 useful for implementing language features like scoped automatic variable
6746 sized arrays in C99.
6751 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6753 '``llvm.prefetch``' Intrinsic
6754 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6761 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6766 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6767 insert a prefetch instruction if supported; otherwise, it is a noop.
6768 Prefetches have no effect on the behavior of the program but can change
6769 its performance characteristics.
6774 ``address`` is the address to be prefetched, ``rw`` is the specifier
6775 determining if the fetch should be for a read (0) or write (1), and
6776 ``locality`` is a temporal locality specifier ranging from (0) - no
6777 locality, to (3) - extremely local keep in cache. The ``cache type``
6778 specifies whether the prefetch is performed on the data (1) or
6779 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6780 arguments must be constant integers.
6785 This intrinsic does not modify the behavior of the program. In
6786 particular, prefetches cannot trap and do not produce a value. On
6787 targets that support this intrinsic, the prefetch can provide hints to
6788 the processor cache for better performance.
6790 '``llvm.pcmarker``' Intrinsic
6791 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6798 declare void @llvm.pcmarker(i32 <id>)
6803 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6804 Counter (PC) in a region of code to simulators and other tools. The
6805 method is target specific, but it is expected that the marker will use
6806 exported symbols to transmit the PC of the marker. The marker makes no
6807 guarantees that it will remain with any specific instruction after
6808 optimizations. It is possible that the presence of a marker will inhibit
6809 optimizations. The intended use is to be inserted after optimizations to
6810 allow correlations of simulation runs.
6815 ``id`` is a numerical id identifying the marker.
6820 This intrinsic does not modify the behavior of the program. Backends
6821 that do not support this intrinsic may ignore it.
6823 '``llvm.readcyclecounter``' Intrinsic
6824 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6831 declare i64 @llvm.readcyclecounter()
6836 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6837 counter register (or similar low latency, high accuracy clocks) on those
6838 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6839 should map to RPCC. As the backing counters overflow quickly (on the
6840 order of 9 seconds on alpha), this should only be used for small
6846 When directly supported, reading the cycle counter should not modify any
6847 memory. Implementations are allowed to either return a application
6848 specific value or a system wide value. On backends without support, this
6849 is lowered to a constant 0.
6851 Note that runtime support may be conditional on the privilege-level code is
6852 running at and the host platform.
6854 Standard C Library Intrinsics
6855 -----------------------------
6857 LLVM provides intrinsics for a few important standard C library
6858 functions. These intrinsics allow source-language front-ends to pass
6859 information about the alignment of the pointer arguments to the code
6860 generator, providing opportunity for more efficient code generation.
6864 '``llvm.memcpy``' Intrinsic
6865 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6870 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6871 integer bit width and for different address spaces. Not all targets
6872 support all bit widths however.
6876 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6877 i32 <len>, i32 <align>, i1 <isvolatile>)
6878 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6879 i64 <len>, i32 <align>, i1 <isvolatile>)
6884 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6885 source location to the destination location.
6887 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6888 intrinsics do not return a value, takes extra alignment/isvolatile
6889 arguments and the pointers can be in specified address spaces.
6894 The first argument is a pointer to the destination, the second is a
6895 pointer to the source. The third argument is an integer argument
6896 specifying the number of bytes to copy, the fourth argument is the
6897 alignment of the source and destination locations, and the fifth is a
6898 boolean indicating a volatile access.
6900 If the call to this intrinsic has an alignment value that is not 0 or 1,
6901 then the caller guarantees that both the source and destination pointers
6902 are aligned to that boundary.
6904 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6905 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6906 very cleanly specified and it is unwise to depend on it.
6911 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6912 source location to the destination location, which are not allowed to
6913 overlap. It copies "len" bytes of memory over. If the argument is known
6914 to be aligned to some boundary, this can be specified as the fourth
6915 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
6917 '``llvm.memmove``' Intrinsic
6918 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6923 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6924 bit width and for different address space. Not all targets support all
6929 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6930 i32 <len>, i32 <align>, i1 <isvolatile>)
6931 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6932 i64 <len>, i32 <align>, i1 <isvolatile>)
6937 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6938 source location to the destination location. It is similar to the
6939 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6942 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6943 intrinsics do not return a value, takes extra alignment/isvolatile
6944 arguments and the pointers can be in specified address spaces.
6949 The first argument is a pointer to the destination, the second is a
6950 pointer to the source. The third argument is an integer argument
6951 specifying the number of bytes to copy, the fourth argument is the
6952 alignment of the source and destination locations, and the fifth is a
6953 boolean indicating a volatile access.
6955 If the call to this intrinsic has an alignment value that is not 0 or 1,
6956 then the caller guarantees that the source and destination pointers are
6957 aligned to that boundary.
6959 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6960 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6961 not very cleanly specified and it is unwise to depend on it.
6966 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6967 source location to the destination location, which may overlap. It
6968 copies "len" bytes of memory over. If the argument is known to be
6969 aligned to some boundary, this can be specified as the fourth argument,
6970 otherwise it should be set to 0 or 1 (both meaning no alignment).
6972 '``llvm.memset.*``' Intrinsics
6973 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6978 This is an overloaded intrinsic. You can use llvm.memset on any integer
6979 bit width and for different address spaces. However, not all targets
6980 support all bit widths.
6984 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6985 i32 <len>, i32 <align>, i1 <isvolatile>)
6986 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6987 i64 <len>, i32 <align>, i1 <isvolatile>)
6992 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6993 particular byte value.
6995 Note that, unlike the standard libc function, the ``llvm.memset``
6996 intrinsic does not return a value and takes extra alignment/volatile
6997 arguments. Also, the destination can be in an arbitrary address space.
7002 The first argument is a pointer to the destination to fill, the second
7003 is the byte value with which to fill it, the third argument is an
7004 integer argument specifying the number of bytes to fill, and the fourth
7005 argument is the known alignment of the destination location.
7007 If the call to this intrinsic has an alignment value that is not 0 or 1,
7008 then the caller guarantees that the destination pointer is aligned to
7011 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7012 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7013 very cleanly specified and it is unwise to depend on it.
7018 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7019 at the destination location. If the argument is known to be aligned to
7020 some boundary, this can be specified as the fourth argument, otherwise
7021 it should be set to 0 or 1 (both meaning no alignment).
7023 '``llvm.sqrt.*``' Intrinsic
7024 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7029 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7030 floating point or vector of floating point type. Not all targets support
7035 declare float @llvm.sqrt.f32(float %Val)
7036 declare double @llvm.sqrt.f64(double %Val)
7037 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7038 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7039 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7044 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7045 returning the same value as the libm '``sqrt``' functions would. Unlike
7046 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7047 negative numbers other than -0.0 (which allows for better optimization,
7048 because there is no need to worry about errno being set).
7049 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7054 The argument and return value are floating point numbers of the same
7060 This function returns the sqrt of the specified operand if it is a
7061 nonnegative floating point number.
7063 '``llvm.powi.*``' Intrinsic
7064 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7069 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7070 floating point or vector of floating point type. Not all targets support
7075 declare float @llvm.powi.f32(float %Val, i32 %power)
7076 declare double @llvm.powi.f64(double %Val, i32 %power)
7077 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7078 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7079 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7084 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7085 specified (positive or negative) power. The order of evaluation of
7086 multiplications is not defined. When a vector of floating point type is
7087 used, the second argument remains a scalar integer value.
7092 The second argument is an integer power, and the first is a value to
7093 raise to that power.
7098 This function returns the first value raised to the second power with an
7099 unspecified sequence of rounding operations.
7101 '``llvm.sin.*``' Intrinsic
7102 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7107 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7108 floating point or vector of floating point type. Not all targets support
7113 declare float @llvm.sin.f32(float %Val)
7114 declare double @llvm.sin.f64(double %Val)
7115 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7116 declare fp128 @llvm.sin.f128(fp128 %Val)
7117 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7122 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7127 The argument and return value are floating point numbers of the same
7133 This function returns the sine of the specified operand, returning the
7134 same values as the libm ``sin`` functions would, and handles error
7135 conditions in the same way.
7137 '``llvm.cos.*``' Intrinsic
7138 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7143 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7144 floating point or vector of floating point type. Not all targets support
7149 declare float @llvm.cos.f32(float %Val)
7150 declare double @llvm.cos.f64(double %Val)
7151 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7152 declare fp128 @llvm.cos.f128(fp128 %Val)
7153 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7158 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7163 The argument and return value are floating point numbers of the same
7169 This function returns the cosine of the specified operand, returning the
7170 same values as the libm ``cos`` functions would, and handles error
7171 conditions in the same way.
7173 '``llvm.pow.*``' Intrinsic
7174 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7179 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7180 floating point or vector of floating point type. Not all targets support
7185 declare float @llvm.pow.f32(float %Val, float %Power)
7186 declare double @llvm.pow.f64(double %Val, double %Power)
7187 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7188 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7189 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7194 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7195 specified (positive or negative) power.
7200 The second argument is a floating point power, and the first is a value
7201 to raise to that power.
7206 This function returns the first value raised to the second power,
7207 returning the same values as the libm ``pow`` functions would, and
7208 handles error conditions in the same way.
7210 '``llvm.exp.*``' Intrinsic
7211 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7216 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7217 floating point or vector of floating point type. Not all targets support
7222 declare float @llvm.exp.f32(float %Val)
7223 declare double @llvm.exp.f64(double %Val)
7224 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7225 declare fp128 @llvm.exp.f128(fp128 %Val)
7226 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7231 The '``llvm.exp.*``' intrinsics perform the exp function.
7236 The argument and return value are floating point numbers of the same
7242 This function returns the same values as the libm ``exp`` functions
7243 would, and handles error conditions in the same way.
7245 '``llvm.exp2.*``' Intrinsic
7246 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7251 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7252 floating point or vector of floating point type. Not all targets support
7257 declare float @llvm.exp2.f32(float %Val)
7258 declare double @llvm.exp2.f64(double %Val)
7259 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7260 declare fp128 @llvm.exp2.f128(fp128 %Val)
7261 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7266 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7271 The argument and return value are floating point numbers of the same
7277 This function returns the same values as the libm ``exp2`` functions
7278 would, and handles error conditions in the same way.
7280 '``llvm.log.*``' Intrinsic
7281 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7286 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7287 floating point or vector of floating point type. Not all targets support
7292 declare float @llvm.log.f32(float %Val)
7293 declare double @llvm.log.f64(double %Val)
7294 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7295 declare fp128 @llvm.log.f128(fp128 %Val)
7296 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7301 The '``llvm.log.*``' intrinsics perform the log function.
7306 The argument and return value are floating point numbers of the same
7312 This function returns the same values as the libm ``log`` functions
7313 would, and handles error conditions in the same way.
7315 '``llvm.log10.*``' Intrinsic
7316 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7321 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7322 floating point or vector of floating point type. Not all targets support
7327 declare float @llvm.log10.f32(float %Val)
7328 declare double @llvm.log10.f64(double %Val)
7329 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7330 declare fp128 @llvm.log10.f128(fp128 %Val)
7331 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7336 The '``llvm.log10.*``' intrinsics perform the log10 function.
7341 The argument and return value are floating point numbers of the same
7347 This function returns the same values as the libm ``log10`` functions
7348 would, and handles error conditions in the same way.
7350 '``llvm.log2.*``' Intrinsic
7351 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7356 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7357 floating point or vector of floating point type. Not all targets support
7362 declare float @llvm.log2.f32(float %Val)
7363 declare double @llvm.log2.f64(double %Val)
7364 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7365 declare fp128 @llvm.log2.f128(fp128 %Val)
7366 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7371 The '``llvm.log2.*``' intrinsics perform the log2 function.
7376 The argument and return value are floating point numbers of the same
7382 This function returns the same values as the libm ``log2`` functions
7383 would, and handles error conditions in the same way.
7385 '``llvm.fma.*``' Intrinsic
7386 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7391 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7392 floating point or vector of floating point type. Not all targets support
7397 declare float @llvm.fma.f32(float %a, float %b, float %c)
7398 declare double @llvm.fma.f64(double %a, double %b, double %c)
7399 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7400 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7401 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7406 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7412 The argument and return value are floating point numbers of the same
7418 This function returns the same values as the libm ``fma`` functions
7421 '``llvm.fabs.*``' Intrinsic
7422 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7427 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7428 floating point or vector of floating point type. Not all targets support
7433 declare float @llvm.fabs.f32(float %Val)
7434 declare double @llvm.fabs.f64(double %Val)
7435 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7436 declare fp128 @llvm.fabs.f128(fp128 %Val)
7437 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7442 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7448 The argument and return value are floating point numbers of the same
7454 This function returns the same values as the libm ``fabs`` functions
7455 would, and handles error conditions in the same way.
7457 '``llvm.copysign.*``' Intrinsic
7458 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7463 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7464 floating point or vector of floating point type. Not all targets support
7469 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7470 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7471 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7472 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7473 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7478 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7479 first operand and the sign of the second operand.
7484 The arguments and return value are floating point numbers of the same
7490 This function returns the same values as the libm ``copysign``
7491 functions would, and handles error conditions in the same way.
7493 '``llvm.floor.*``' Intrinsic
7494 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7499 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7500 floating point or vector of floating point type. Not all targets support
7505 declare float @llvm.floor.f32(float %Val)
7506 declare double @llvm.floor.f64(double %Val)
7507 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7508 declare fp128 @llvm.floor.f128(fp128 %Val)
7509 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7514 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7519 The argument and return value are floating point numbers of the same
7525 This function returns the same values as the libm ``floor`` functions
7526 would, and handles error conditions in the same way.
7528 '``llvm.ceil.*``' Intrinsic
7529 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7534 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7535 floating point or vector of floating point type. Not all targets support
7540 declare float @llvm.ceil.f32(float %Val)
7541 declare double @llvm.ceil.f64(double %Val)
7542 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7543 declare fp128 @llvm.ceil.f128(fp128 %Val)
7544 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7549 The '``llvm.ceil.*``' intrinsics return the ceiling 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 ``ceil`` functions
7561 would, and handles error conditions in the same way.
7563 '``llvm.trunc.*``' Intrinsic
7564 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7569 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7570 floating point or vector of floating point type. Not all targets support
7575 declare float @llvm.trunc.f32(float %Val)
7576 declare double @llvm.trunc.f64(double %Val)
7577 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7578 declare fp128 @llvm.trunc.f128(fp128 %Val)
7579 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7584 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7585 nearest integer not larger in magnitude than the operand.
7590 The argument and return value are floating point numbers of the same
7596 This function returns the same values as the libm ``trunc`` functions
7597 would, and handles error conditions in the same way.
7599 '``llvm.rint.*``' Intrinsic
7600 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7605 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7606 floating point or vector of floating point type. Not all targets support
7611 declare float @llvm.rint.f32(float %Val)
7612 declare double @llvm.rint.f64(double %Val)
7613 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7614 declare fp128 @llvm.rint.f128(fp128 %Val)
7615 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7620 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7621 nearest integer. It may raise an inexact floating-point exception if the
7622 operand isn't an integer.
7627 The argument and return value are floating point numbers of the same
7633 This function returns the same values as the libm ``rint`` functions
7634 would, and handles error conditions in the same way.
7636 '``llvm.nearbyint.*``' Intrinsic
7637 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7642 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7643 floating point or vector of floating point type. Not all targets support
7648 declare float @llvm.nearbyint.f32(float %Val)
7649 declare double @llvm.nearbyint.f64(double %Val)
7650 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7651 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7652 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7657 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7663 The argument and return value are floating point numbers of the same
7669 This function returns the same values as the libm ``nearbyint``
7670 functions would, and handles error conditions in the same way.
7672 '``llvm.round.*``' Intrinsic
7673 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7678 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7679 floating point or vector of floating point type. Not all targets support
7684 declare float @llvm.round.f32(float %Val)
7685 declare double @llvm.round.f64(double %Val)
7686 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7687 declare fp128 @llvm.round.f128(fp128 %Val)
7688 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7693 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7699 The argument and return value are floating point numbers of the same
7705 This function returns the same values as the libm ``round``
7706 functions would, and handles error conditions in the same way.
7708 Bit Manipulation Intrinsics
7709 ---------------------------
7711 LLVM provides intrinsics for a few important bit manipulation
7712 operations. These allow efficient code generation for some algorithms.
7714 '``llvm.bswap.*``' Intrinsics
7715 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7720 This is an overloaded intrinsic function. You can use bswap on any
7721 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7725 declare i16 @llvm.bswap.i16(i16 <id>)
7726 declare i32 @llvm.bswap.i32(i32 <id>)
7727 declare i64 @llvm.bswap.i64(i64 <id>)
7732 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7733 values with an even number of bytes (positive multiple of 16 bits).
7734 These are useful for performing operations on data that is not in the
7735 target's native byte order.
7740 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7741 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7742 intrinsic returns an i32 value that has the four bytes of the input i32
7743 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7744 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7745 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7746 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7749 '``llvm.ctpop.*``' Intrinsic
7750 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7755 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7756 bit width, or on any vector with integer elements. Not all targets
7757 support all bit widths or vector types, however.
7761 declare i8 @llvm.ctpop.i8(i8 <src>)
7762 declare i16 @llvm.ctpop.i16(i16 <src>)
7763 declare i32 @llvm.ctpop.i32(i32 <src>)
7764 declare i64 @llvm.ctpop.i64(i64 <src>)
7765 declare i256 @llvm.ctpop.i256(i256 <src>)
7766 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7771 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7777 The only argument is the value to be counted. The argument may be of any
7778 integer type, or a vector with integer elements. The return type must
7779 match the argument type.
7784 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7785 each element of a vector.
7787 '``llvm.ctlz.*``' Intrinsic
7788 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7793 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7794 integer bit width, or any vector whose elements are integers. Not all
7795 targets support all bit widths or vector types, however.
7799 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7800 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7801 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7802 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7803 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7804 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7809 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7810 leading zeros in a variable.
7815 The first argument is the value to be counted. This argument may be of
7816 any integer type, or a vectory with integer element type. The return
7817 type must match the first argument type.
7819 The second argument must be a constant and is a flag to indicate whether
7820 the intrinsic should ensure that a zero as the first argument produces a
7821 defined result. Historically some architectures did not provide a
7822 defined result for zero values as efficiently, and many algorithms are
7823 now predicated on avoiding zero-value inputs.
7828 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7829 zeros in a variable, or within each element of the vector. If
7830 ``src == 0`` then the result is the size in bits of the type of ``src``
7831 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7832 ``llvm.ctlz(i32 2) = 30``.
7834 '``llvm.cttz.*``' Intrinsic
7835 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7840 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7841 integer bit width, or any vector of integer elements. Not all targets
7842 support all bit widths or vector types, however.
7846 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7847 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7848 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7849 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7850 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7851 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7856 The '``llvm.cttz``' family of intrinsic functions counts the number of
7862 The first argument is the value to be counted. This argument may be of
7863 any integer type, or a vectory with integer element type. The return
7864 type must match the first argument type.
7866 The second argument must be a constant and is a flag to indicate whether
7867 the intrinsic should ensure that a zero as the first argument produces a
7868 defined result. Historically some architectures did not provide a
7869 defined result for zero values as efficiently, and many algorithms are
7870 now predicated on avoiding zero-value inputs.
7875 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7876 zeros in a variable, or within each element of a vector. If ``src == 0``
7877 then the result is the size in bits of the type of ``src`` if
7878 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7879 ``llvm.cttz(2) = 1``.
7881 Arithmetic with Overflow Intrinsics
7882 -----------------------------------
7884 LLVM provides intrinsics for some arithmetic with overflow operations.
7886 '``llvm.sadd.with.overflow.*``' Intrinsics
7887 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7892 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7893 on any integer bit width.
7897 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7898 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7899 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7904 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7905 a signed addition of the two arguments, and indicate whether an overflow
7906 occurred during the signed summation.
7911 The arguments (%a and %b) and the first element of the result structure
7912 may be of integer types of any bit width, but they must have the same
7913 bit width. The second element of the result structure must be of type
7914 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7920 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7921 a signed addition of the two variables. They return a structure --- the
7922 first element of which is the signed summation, and the second element
7923 of which is a bit specifying if the signed summation resulted in an
7929 .. code-block:: llvm
7931 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7932 %sum = extractvalue {i32, i1} %res, 0
7933 %obit = extractvalue {i32, i1} %res, 1
7934 br i1 %obit, label %overflow, label %normal
7936 '``llvm.uadd.with.overflow.*``' Intrinsics
7937 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7942 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7943 on any integer bit width.
7947 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7948 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7949 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7954 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7955 an unsigned addition of the two arguments, and indicate whether a carry
7956 occurred during the unsigned summation.
7961 The arguments (%a and %b) and the first element of the result structure
7962 may be of integer types of any bit width, but they must have the same
7963 bit width. The second element of the result structure must be of type
7964 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7970 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7971 an unsigned addition of the two arguments. They return a structure --- the
7972 first element of which is the sum, and the second element of which is a
7973 bit specifying if the unsigned summation resulted in a carry.
7978 .. code-block:: llvm
7980 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7981 %sum = extractvalue {i32, i1} %res, 0
7982 %obit = extractvalue {i32, i1} %res, 1
7983 br i1 %obit, label %carry, label %normal
7985 '``llvm.ssub.with.overflow.*``' Intrinsics
7986 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7991 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7992 on any integer bit width.
7996 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7997 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7998 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8003 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8004 a signed subtraction of the two arguments, and indicate whether an
8005 overflow occurred during the signed subtraction.
8010 The arguments (%a and %b) and the first element of the result structure
8011 may be of integer types of any bit width, but they must have the same
8012 bit width. The second element of the result structure must be of type
8013 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8019 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8020 a signed subtraction of the two arguments. They return a structure --- the
8021 first element of which is the subtraction, and the second element of
8022 which is a bit specifying if the signed subtraction resulted in an
8028 .. code-block:: llvm
8030 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8031 %sum = extractvalue {i32, i1} %res, 0
8032 %obit = extractvalue {i32, i1} %res, 1
8033 br i1 %obit, label %overflow, label %normal
8035 '``llvm.usub.with.overflow.*``' Intrinsics
8036 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8041 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8042 on any integer bit width.
8046 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8047 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8048 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8053 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8054 an unsigned subtraction of the two arguments, and indicate whether an
8055 overflow occurred during the unsigned subtraction.
8060 The arguments (%a and %b) and the first element of the result structure
8061 may be of integer types of any bit width, but they must have the same
8062 bit width. The second element of the result structure must be of type
8063 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8069 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8070 an unsigned subtraction of the two arguments. They return a structure ---
8071 the first element of which is the subtraction, and the second element of
8072 which is a bit specifying if the unsigned subtraction resulted in an
8078 .. code-block:: llvm
8080 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8081 %sum = extractvalue {i32, i1} %res, 0
8082 %obit = extractvalue {i32, i1} %res, 1
8083 br i1 %obit, label %overflow, label %normal
8085 '``llvm.smul.with.overflow.*``' Intrinsics
8086 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8091 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8092 on any integer bit width.
8096 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8097 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8098 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8103 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8104 a signed multiplication of the two arguments, and indicate whether an
8105 overflow occurred during the signed multiplication.
8110 The arguments (%a and %b) and the first element of the result structure
8111 may be of integer types of any bit width, but they must have the same
8112 bit width. The second element of the result structure must be of type
8113 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8119 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8120 a signed multiplication of the two arguments. They return a structure ---
8121 the first element of which is the multiplication, and the second element
8122 of which is a bit specifying if the signed multiplication resulted in an
8128 .. code-block:: llvm
8130 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8131 %sum = extractvalue {i32, i1} %res, 0
8132 %obit = extractvalue {i32, i1} %res, 1
8133 br i1 %obit, label %overflow, label %normal
8135 '``llvm.umul.with.overflow.*``' Intrinsics
8136 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8141 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8142 on any integer bit width.
8146 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8147 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8148 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8153 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8154 a unsigned multiplication of the two arguments, and indicate whether an
8155 overflow occurred during the unsigned multiplication.
8160 The arguments (%a and %b) and the first element of the result structure
8161 may be of integer types of any bit width, but they must have the same
8162 bit width. The second element of the result structure must be of type
8163 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8169 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8170 an unsigned multiplication of the two arguments. They return a structure ---
8171 the first element of which is the multiplication, and the second
8172 element of which is a bit specifying if the unsigned multiplication
8173 resulted in an overflow.
8178 .. code-block:: llvm
8180 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8181 %sum = extractvalue {i32, i1} %res, 0
8182 %obit = extractvalue {i32, i1} %res, 1
8183 br i1 %obit, label %overflow, label %normal
8185 Specialised Arithmetic Intrinsics
8186 ---------------------------------
8188 '``llvm.fmuladd.*``' Intrinsic
8189 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8196 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8197 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8202 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8203 expressions that can be fused if the code generator determines that (a) the
8204 target instruction set has support for a fused operation, and (b) that the
8205 fused operation is more efficient than the equivalent, separate pair of mul
8206 and add instructions.
8211 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8212 multiplicands, a and b, and an addend c.
8221 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8223 is equivalent to the expression a \* b + c, except that rounding will
8224 not be performed between the multiplication and addition steps if the
8225 code generator fuses the operations. Fusion is not guaranteed, even if
8226 the target platform supports it. If a fused multiply-add is required the
8227 corresponding llvm.fma.\* intrinsic function should be used instead.
8232 .. code-block:: llvm
8234 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8236 Half Precision Floating Point Intrinsics
8237 ----------------------------------------
8239 For most target platforms, half precision floating point is a
8240 storage-only format. This means that it is a dense encoding (in memory)
8241 but does not support computation in the format.
8243 This means that code must first load the half-precision floating point
8244 value as an i16, then convert it to float with
8245 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8246 then be performed on the float value (including extending to double
8247 etc). To store the value back to memory, it is first converted to float
8248 if needed, then converted to i16 with
8249 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8252 .. _int_convert_to_fp16:
8254 '``llvm.convert.to.fp16``' Intrinsic
8255 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8262 declare i16 @llvm.convert.to.fp16(f32 %a)
8267 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8268 from single precision floating point format to half precision floating
8274 The intrinsic function contains single argument - the value to be
8280 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8281 from single precision floating point format to half precision floating
8282 point format. The return value is an ``i16`` which contains the
8288 .. code-block:: llvm
8290 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8291 store i16 %res, i16* @x, align 2
8293 .. _int_convert_from_fp16:
8295 '``llvm.convert.from.fp16``' Intrinsic
8296 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8303 declare f32 @llvm.convert.from.fp16(i16 %a)
8308 The '``llvm.convert.from.fp16``' intrinsic function performs a
8309 conversion from half precision floating point format to single precision
8310 floating point format.
8315 The intrinsic function contains single argument - the value to be
8321 The '``llvm.convert.from.fp16``' intrinsic function performs a
8322 conversion from half single precision floating point format to single
8323 precision floating point format. The input half-float value is
8324 represented by an ``i16`` value.
8329 .. code-block:: llvm
8331 %a = load i16* @x, align 2
8332 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8337 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8338 prefix), are described in the `LLVM Source Level
8339 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8342 Exception Handling Intrinsics
8343 -----------------------------
8345 The LLVM exception handling intrinsics (which all start with
8346 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8347 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8351 Trampoline Intrinsics
8352 ---------------------
8354 These intrinsics make it possible to excise one parameter, marked with
8355 the :ref:`nest <nest>` attribute, from a function. The result is a
8356 callable function pointer lacking the nest parameter - the caller does
8357 not need to provide a value for it. Instead, the value to use is stored
8358 in advance in a "trampoline", a block of memory usually allocated on the
8359 stack, which also contains code to splice the nest value into the
8360 argument list. This is used to implement the GCC nested function address
8363 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8364 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8365 It can be created as follows:
8367 .. code-block:: llvm
8369 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8370 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8371 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8372 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8373 %fp = bitcast i8* %p to i32 (i32, i32)*
8375 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8376 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8380 '``llvm.init.trampoline``' Intrinsic
8381 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8388 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8393 This fills the memory pointed to by ``tramp`` with executable code,
8394 turning it into a trampoline.
8399 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8400 pointers. The ``tramp`` argument must point to a sufficiently large and
8401 sufficiently aligned block of memory; this memory is written to by the
8402 intrinsic. Note that the size and the alignment are target-specific -
8403 LLVM currently provides no portable way of determining them, so a
8404 front-end that generates this intrinsic needs to have some
8405 target-specific knowledge. The ``func`` argument must hold a function
8406 bitcast to an ``i8*``.
8411 The block of memory pointed to by ``tramp`` is filled with target
8412 dependent code, turning it into a function. Then ``tramp`` needs to be
8413 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8414 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8415 function's signature is the same as that of ``func`` with any arguments
8416 marked with the ``nest`` attribute removed. At most one such ``nest``
8417 argument is allowed, and it must be of pointer type. Calling the new
8418 function is equivalent to calling ``func`` with the same argument list,
8419 but with ``nval`` used for the missing ``nest`` argument. If, after
8420 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8421 modified, then the effect of any later call to the returned function
8422 pointer is undefined.
8426 '``llvm.adjust.trampoline``' Intrinsic
8427 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8434 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8439 This performs any required machine-specific adjustment to the address of
8440 a trampoline (passed as ``tramp``).
8445 ``tramp`` must point to a block of memory which already has trampoline
8446 code filled in by a previous call to
8447 :ref:`llvm.init.trampoline <int_it>`.
8452 On some architectures the address of the code to be executed needs to be
8453 different to the address where the trampoline is actually stored. This
8454 intrinsic returns the executable address corresponding to ``tramp``
8455 after performing the required machine specific adjustments. The pointer
8456 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8461 This class of intrinsics exists to information about the lifetime of
8462 memory objects and ranges where variables are immutable.
8466 '``llvm.lifetime.start``' Intrinsic
8467 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8474 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8479 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8485 The first argument is a constant integer representing the size of the
8486 object, or -1 if it is variable sized. The second argument is a pointer
8492 This intrinsic indicates that before this point in the code, the value
8493 of the memory pointed to by ``ptr`` is dead. This means that it is known
8494 to never be used and has an undefined value. A load from the pointer
8495 that precedes this intrinsic can be replaced with ``'undef'``.
8499 '``llvm.lifetime.end``' Intrinsic
8500 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8507 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8512 The '``llvm.lifetime.end``' intrinsic specifies the end 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 after this point in the code, the value of
8526 the memory pointed to by ``ptr`` is dead. This means that it is known to
8527 never be used and has an undefined value. Any stores into the memory
8528 object following this intrinsic may be removed as dead.
8530 '``llvm.invariant.start``' Intrinsic
8531 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8538 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8543 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8544 a memory object will not change.
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 until an ``llvm.invariant.end`` that uses
8557 the return value, the referenced memory location is constant and
8560 '``llvm.invariant.end``' Intrinsic
8561 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8568 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8573 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8574 memory object are mutable.
8579 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8580 The second argument is a constant integer representing the size of the
8581 object, or -1 if it is variable sized and the third argument is a
8582 pointer to the object.
8587 This intrinsic indicates that the memory is mutable again.
8592 This class of intrinsics is designed to be generic and has no specific
8595 '``llvm.var.annotation``' Intrinsic
8596 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8603 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8608 The '``llvm.var.annotation``' intrinsic.
8613 The first argument is a pointer to a value, the second is a pointer to a
8614 global string, the third is a pointer to a global string which is the
8615 source file name, and the last argument is the line number.
8620 This intrinsic allows annotation of local variables with arbitrary
8621 strings. This can be useful for special purpose optimizations that want
8622 to look for these annotations. These have no other defined use; they are
8623 ignored by code generation and optimization.
8625 '``llvm.ptr.annotation.*``' Intrinsic
8626 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8631 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8632 pointer to an integer of any width. *NOTE* you must specify an address space for
8633 the pointer. The identifier for the default address space is the integer
8638 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8639 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8640 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8641 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8642 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8647 The '``llvm.ptr.annotation``' intrinsic.
8652 The first argument is a pointer to an integer value of arbitrary bitwidth
8653 (result of some expression), the second is a pointer to a global string, the
8654 third is a pointer to a global string which is the source file name, and the
8655 last argument is the line number. It returns the value of the first argument.
8660 This intrinsic allows annotation of a pointer to an integer with arbitrary
8661 strings. This can be useful for special purpose optimizations that want to look
8662 for these annotations. These have no other defined use; they are ignored by code
8663 generation and optimization.
8665 '``llvm.annotation.*``' Intrinsic
8666 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8671 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8672 any integer bit width.
8676 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8677 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8678 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8679 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8680 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8685 The '``llvm.annotation``' intrinsic.
8690 The first argument is an integer value (result of some expression), the
8691 second is a pointer to a global string, the third is a pointer to a
8692 global string which is the source file name, and the last argument is
8693 the line number. It returns the value of the first argument.
8698 This intrinsic allows annotations to be put on arbitrary expressions
8699 with arbitrary strings. This can be useful for special purpose
8700 optimizations that want to look for these annotations. These have no
8701 other defined use; they are ignored by code generation and optimization.
8703 '``llvm.trap``' Intrinsic
8704 ^^^^^^^^^^^^^^^^^^^^^^^^^
8711 declare void @llvm.trap() noreturn nounwind
8716 The '``llvm.trap``' intrinsic.
8726 This intrinsic is lowered to the target dependent trap instruction. If
8727 the target does not have a trap instruction, this intrinsic will be
8728 lowered to a call of the ``abort()`` function.
8730 '``llvm.debugtrap``' Intrinsic
8731 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8738 declare void @llvm.debugtrap() nounwind
8743 The '``llvm.debugtrap``' intrinsic.
8753 This intrinsic is lowered to code which is intended to cause an
8754 execution trap with the intention of requesting the attention of a
8757 '``llvm.stackprotector``' Intrinsic
8758 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8765 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8770 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8771 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8772 is placed on the stack before local variables.
8777 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8778 The first argument is the value loaded from the stack guard
8779 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8780 enough space to hold the value of the guard.
8785 This intrinsic causes the prologue/epilogue inserter to force the position of
8786 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8787 to ensure that if a local variable on the stack is overwritten, it will destroy
8788 the value of the guard. When the function exits, the guard on the stack is
8789 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8790 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8791 calling the ``__stack_chk_fail()`` function.
8793 '``llvm.stackprotectorcheck``' Intrinsic
8794 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8801 declare void @llvm.stackprotectorcheck(i8** <guard>)
8806 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8807 created stack protector and if they are not equal calls the
8808 ``__stack_chk_fail()`` function.
8813 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8814 the variable ``@__stack_chk_guard``.
8819 This intrinsic is provided to perform the stack protector check by comparing
8820 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8821 values do not match call the ``__stack_chk_fail()`` function.
8823 The reason to provide this as an IR level intrinsic instead of implementing it
8824 via other IR operations is that in order to perform this operation at the IR
8825 level without an intrinsic, one would need to create additional basic blocks to
8826 handle the success/failure cases. This makes it difficult to stop the stack
8827 protector check from disrupting sibling tail calls in Codegen. With this
8828 intrinsic, we are able to generate the stack protector basic blocks late in
8829 codegen after the tail call decision has occurred.
8831 '``llvm.objectsize``' Intrinsic
8832 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8839 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8840 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8845 The ``llvm.objectsize`` intrinsic is designed to provide information to
8846 the optimizers to determine at compile time whether a) an operation
8847 (like memcpy) will overflow a buffer that corresponds to an object, or
8848 b) that a runtime check for overflow isn't necessary. An object in this
8849 context means an allocation of a specific class, structure, array, or
8855 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8856 argument is a pointer to or into the ``object``. The second argument is
8857 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8858 or -1 (if false) when the object size is unknown. The second argument
8859 only accepts constants.
8864 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8865 the size of the object concerned. If the size cannot be determined at
8866 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8867 on the ``min`` argument).
8869 '``llvm.expect``' Intrinsic
8870 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8877 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8878 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8883 The ``llvm.expect`` intrinsic provides information about expected (the
8884 most probable) value of ``val``, which can be used by optimizers.
8889 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8890 a value. The second argument is an expected value, this needs to be a
8891 constant value, variables are not allowed.
8896 This intrinsic is lowered to the ``val``.
8898 '``llvm.donothing``' Intrinsic
8899 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8906 declare void @llvm.donothing() nounwind readnone
8911 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8912 only intrinsic that can be called with an invoke instruction.
8922 This intrinsic does nothing, and it's removed by optimizers and ignored
8925 Stack Map Intrinsics
8926 --------------------
8928 LLVM provides experimental intrinsics to support runtime patching
8929 mechanisms commonly desired in dynamic language JITs. These intrinsics
8930 are described in :doc:`StackMaps`.