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. The address space, ``n`` is optional, and if not specified,
1145 denotes the default address space 0. The value of ``n`` must be
1146 in the range [1,2^23).
1147 ``i<size>:<abi>:<pref>``
1148 This specifies the alignment for an integer type of a given bit
1149 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1150 ``v<size>:<abi>:<pref>``
1151 This specifies the alignment for a vector type of a given bit
1153 ``f<size>:<abi>:<pref>``
1154 This specifies the alignment for a floating point type of a given bit
1155 ``<size>``. Only values of ``<size>`` that are supported by the target
1156 will work. 32 (float) and 64 (double) are supported on all targets; 80
1157 or 128 (different flavors of long double) are also supported on some
1160 This specifies the alignment for an object of aggregate type.
1162 If prerest, specifies that llvm names are mangled in the output. The
1164 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1165 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1166 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1167 symbols get a ``_`` prefix.
1168 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1169 functions also get a suffix based on the frame size.
1170 ``n<size1>:<size2>:<size3>...``
1171 This specifies a set of native integer widths for the target CPU in
1172 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1173 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1174 this set are considered to support most general arithmetic operations
1177 On every specification that takes a ``<abi>:<pref>``, specifying the
1178 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1179 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1181 When constructing the data layout for a given target, LLVM starts with a
1182 default set of specifications which are then (possibly) overridden by
1183 the specifications in the ``datalayout`` keyword. The default
1184 specifications are given in this list:
1186 - ``E`` - big endian
1187 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1188 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1189 same as the default address space.
1190 - ``S0`` - natural stack alignment is unspecified
1191 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1192 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1193 - ``i16:16:16`` - i16 is 16-bit aligned
1194 - ``i32:32:32`` - i32 is 32-bit aligned
1195 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1196 alignment of 64-bits
1197 - ``f16:16:16`` - half is 16-bit aligned
1198 - ``f32:32:32`` - float is 32-bit aligned
1199 - ``f64:64:64`` - double is 64-bit aligned
1200 - ``f128:128:128`` - quad is 128-bit aligned
1201 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1202 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1203 - ``a:0:64`` - aggregates are 64-bit aligned
1205 When LLVM is determining the alignment for a given type, it uses the
1208 #. If the type sought is an exact match for one of the specifications,
1209 that specification is used.
1210 #. If no match is found, and the type sought is an integer type, then
1211 the smallest integer type that is larger than the bitwidth of the
1212 sought type is used. If none of the specifications are larger than
1213 the bitwidth then the largest integer type is used. For example,
1214 given the default specifications above, the i7 type will use the
1215 alignment of i8 (next largest) while both i65 and i256 will use the
1216 alignment of i64 (largest specified).
1217 #. If no match is found, and the type sought is a vector type, then the
1218 largest vector type that is smaller than the sought vector type will
1219 be used as a fall back. This happens because <128 x double> can be
1220 implemented in terms of 64 <2 x double>, for example.
1222 The function of the data layout string may not be what you expect.
1223 Notably, this is not a specification from the frontend of what alignment
1224 the code generator should use.
1226 Instead, if specified, the target data layout is required to match what
1227 the ultimate *code generator* expects. This string is used by the
1228 mid-level optimizers to improve code, and this only works if it matches
1229 what the ultimate code generator uses. If you would like to generate IR
1230 that does not embed this target-specific detail into the IR, then you
1231 don't have to specify the string. This will disable some optimizations
1232 that require precise layout information, but this also prevents those
1233 optimizations from introducing target specificity into the IR.
1240 A module may specify a target triple string that describes the target
1241 host. The syntax for the target triple is simply:
1243 .. code-block:: llvm
1245 target triple = "x86_64-apple-macosx10.7.0"
1247 The *target triple* string consists of a series of identifiers delimited
1248 by the minus sign character ('-'). The canonical forms are:
1252 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1253 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1255 This information is passed along to the backend so that it generates
1256 code for the proper architecture. It's possible to override this on the
1257 command line with the ``-mtriple`` command line option.
1259 .. _pointeraliasing:
1261 Pointer Aliasing Rules
1262 ----------------------
1264 Any memory access must be done through a pointer value associated with
1265 an address range of the memory access, otherwise the behavior is
1266 undefined. Pointer values are associated with address ranges according
1267 to the following rules:
1269 - A pointer value is associated with the addresses associated with any
1270 value it is *based* on.
1271 - An address of a global variable is associated with the address range
1272 of the variable's storage.
1273 - The result value of an allocation instruction is associated with the
1274 address range of the allocated storage.
1275 - A null pointer in the default address-space is associated with no
1277 - An integer constant other than zero or a pointer value returned from
1278 a function not defined within LLVM may be associated with address
1279 ranges allocated through mechanisms other than those provided by
1280 LLVM. Such ranges shall not overlap with any ranges of addresses
1281 allocated by mechanisms provided by LLVM.
1283 A pointer value is *based* on another pointer value according to the
1286 - A pointer value formed from a ``getelementptr`` operation is *based*
1287 on the first operand of the ``getelementptr``.
1288 - The result value of a ``bitcast`` is *based* on the operand of the
1290 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1291 values that contribute (directly or indirectly) to the computation of
1292 the pointer's value.
1293 - The "*based* on" relationship is transitive.
1295 Note that this definition of *"based"* is intentionally similar to the
1296 definition of *"based"* in C99, though it is slightly weaker.
1298 LLVM IR does not associate types with memory. The result type of a
1299 ``load`` merely indicates the size and alignment of the memory from
1300 which to load, as well as the interpretation of the value. The first
1301 operand type of a ``store`` similarly only indicates the size and
1302 alignment of the store.
1304 Consequently, type-based alias analysis, aka TBAA, aka
1305 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1306 :ref:`Metadata <metadata>` may be used to encode additional information
1307 which specialized optimization passes may use to implement type-based
1312 Volatile Memory Accesses
1313 ------------------------
1315 Certain memory accesses, such as :ref:`load <i_load>`'s,
1316 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1317 marked ``volatile``. The optimizers must not change the number of
1318 volatile operations or change their order of execution relative to other
1319 volatile operations. The optimizers *may* change the order of volatile
1320 operations relative to non-volatile operations. This is not Java's
1321 "volatile" and has no cross-thread synchronization behavior.
1323 IR-level volatile loads and stores cannot safely be optimized into
1324 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1325 flagged volatile. Likewise, the backend should never split or merge
1326 target-legal volatile load/store instructions.
1328 .. admonition:: Rationale
1330 Platforms may rely on volatile loads and stores of natively supported
1331 data width to be executed as single instruction. For example, in C
1332 this holds for an l-value of volatile primitive type with native
1333 hardware support, but not necessarily for aggregate types. The
1334 frontend upholds these expectations, which are intentionally
1335 unspecified in the IR. The rules above ensure that IR transformation
1336 do not violate the frontend's contract with the language.
1340 Memory Model for Concurrent Operations
1341 --------------------------------------
1343 The LLVM IR does not define any way to start parallel threads of
1344 execution or to register signal handlers. Nonetheless, there are
1345 platform-specific ways to create them, and we define LLVM IR's behavior
1346 in their presence. This model is inspired by the C++0x memory model.
1348 For a more informal introduction to this model, see the :doc:`Atomics`.
1350 We define a *happens-before* partial order as the least partial order
1353 - Is a superset of single-thread program order, and
1354 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1355 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1356 techniques, like pthread locks, thread creation, thread joining,
1357 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1358 Constraints <ordering>`).
1360 Note that program order does not introduce *happens-before* edges
1361 between a thread and signals executing inside that thread.
1363 Every (defined) read operation (load instructions, memcpy, atomic
1364 loads/read-modify-writes, etc.) R reads a series of bytes written by
1365 (defined) write operations (store instructions, atomic
1366 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1367 section, initialized globals are considered to have a write of the
1368 initializer which is atomic and happens before any other read or write
1369 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1370 may see any write to the same byte, except:
1372 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1373 write\ :sub:`2` happens before R\ :sub:`byte`, then
1374 R\ :sub:`byte` does not see write\ :sub:`1`.
1375 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1376 R\ :sub:`byte` does not see write\ :sub:`3`.
1378 Given that definition, R\ :sub:`byte` is defined as follows:
1380 - If R is volatile, the result is target-dependent. (Volatile is
1381 supposed to give guarantees which can support ``sig_atomic_t`` in
1382 C/C++, and may be used for accesses to addresses which do not behave
1383 like normal memory. It does not generally provide cross-thread
1385 - Otherwise, if there is no write to the same byte that happens before
1386 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1387 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1388 R\ :sub:`byte` returns the value written by that write.
1389 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1390 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1391 Memory Ordering Constraints <ordering>` section for additional
1392 constraints on how the choice is made.
1393 - Otherwise R\ :sub:`byte` returns ``undef``.
1395 R returns the value composed of the series of bytes it read. This
1396 implies that some bytes within the value may be ``undef`` **without**
1397 the entire value being ``undef``. Note that this only defines the
1398 semantics of the operation; it doesn't mean that targets will emit more
1399 than one instruction to read the series of bytes.
1401 Note that in cases where none of the atomic intrinsics are used, this
1402 model places only one restriction on IR transformations on top of what
1403 is required for single-threaded execution: introducing a store to a byte
1404 which might not otherwise be stored is not allowed in general.
1405 (Specifically, in the case where another thread might write to and read
1406 from an address, introducing a store can change a load that may see
1407 exactly one write into a load that may see multiple writes.)
1411 Atomic Memory Ordering Constraints
1412 ----------------------------------
1414 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1415 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1416 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1417 an ordering parameter that determines which other atomic instructions on
1418 the same address they *synchronize with*. These semantics are borrowed
1419 from Java and C++0x, but are somewhat more colloquial. If these
1420 descriptions aren't precise enough, check those specs (see spec
1421 references in the :doc:`atomics guide <Atomics>`).
1422 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1423 differently since they don't take an address. See that instruction's
1424 documentation for details.
1426 For a simpler introduction to the ordering constraints, see the
1430 The set of values that can be read is governed by the happens-before
1431 partial order. A value cannot be read unless some operation wrote
1432 it. This is intended to provide a guarantee strong enough to model
1433 Java's non-volatile shared variables. This ordering cannot be
1434 specified for read-modify-write operations; it is not strong enough
1435 to make them atomic in any interesting way.
1437 In addition to the guarantees of ``unordered``, there is a single
1438 total order for modifications by ``monotonic`` operations on each
1439 address. All modification orders must be compatible with the
1440 happens-before order. There is no guarantee that the modification
1441 orders can be combined to a global total order for the whole program
1442 (and this often will not be possible). The read in an atomic
1443 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1444 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1445 order immediately before the value it writes. If one atomic read
1446 happens before another atomic read of the same address, the later
1447 read must see the same value or a later value in the address's
1448 modification order. This disallows reordering of ``monotonic`` (or
1449 stronger) operations on the same address. If an address is written
1450 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1451 read that address repeatedly, the other threads must eventually see
1452 the write. This corresponds to the C++0x/C1x
1453 ``memory_order_relaxed``.
1455 In addition to the guarantees of ``monotonic``, a
1456 *synchronizes-with* edge may be formed with a ``release`` operation.
1457 This is intended to model C++'s ``memory_order_acquire``.
1459 In addition to the guarantees of ``monotonic``, if this operation
1460 writes a value which is subsequently read by an ``acquire``
1461 operation, it *synchronizes-with* that operation. (This isn't a
1462 complete description; see the C++0x definition of a release
1463 sequence.) This corresponds to the C++0x/C1x
1464 ``memory_order_release``.
1465 ``acq_rel`` (acquire+release)
1466 Acts as both an ``acquire`` and ``release`` operation on its
1467 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1468 ``seq_cst`` (sequentially consistent)
1469 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1470 operation which only reads, ``release`` for an operation which only
1471 writes), there is a global total order on all
1472 sequentially-consistent operations on all addresses, which is
1473 consistent with the *happens-before* partial order and with the
1474 modification orders of all the affected addresses. Each
1475 sequentially-consistent read sees the last preceding write to the
1476 same address in this global order. This corresponds to the C++0x/C1x
1477 ``memory_order_seq_cst`` and Java volatile.
1481 If an atomic operation is marked ``singlethread``, it only *synchronizes
1482 with* or participates in modification and seq\_cst total orderings with
1483 other operations running in the same thread (for example, in signal
1491 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1492 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1493 :ref:`frem <i_frem>`) have the following flags that can set to enable
1494 otherwise unsafe floating point operations
1497 No NaNs - Allow optimizations to assume the arguments and result are not
1498 NaN. Such optimizations are required to retain defined behavior over
1499 NaNs, but the value of the result is undefined.
1502 No Infs - Allow optimizations to assume the arguments and result are not
1503 +/-Inf. Such optimizations are required to retain defined behavior over
1504 +/-Inf, but the value of the result is undefined.
1507 No Signed Zeros - Allow optimizations to treat the sign of a zero
1508 argument or result as insignificant.
1511 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1512 argument rather than perform division.
1515 Fast - Allow algebraically equivalent transformations that may
1516 dramatically change results in floating point (e.g. reassociate). This
1517 flag implies all the others.
1524 The LLVM type system is one of the most important features of the
1525 intermediate representation. Being typed enables a number of
1526 optimizations to be performed on the intermediate representation
1527 directly, without having to do extra analyses on the side before the
1528 transformation. A strong type system makes it easier to read the
1529 generated code and enables novel analyses and transformations that are
1530 not feasible to perform on normal three address code representations.
1540 The void type does not represent any value and has no size.
1558 The function type can be thought of as a function signature. It consists of a
1559 return type and a list of formal parameter types. The return type of a function
1560 type is a void type or first class type --- except for :ref:`label <t_label>`
1561 and :ref:`metadata <t_metadata>` types.
1567 <returntype> (<parameter list>)
1569 ...where '``<parameter list>``' is a comma-separated list of type
1570 specifiers. Optionally, the parameter list may include a type ``...``, which
1571 indicates that the function takes a variable number of arguments. Variable
1572 argument functions can access their arguments with the :ref:`variable argument
1573 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1574 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1578 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1579 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1580 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1581 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1582 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1583 | ``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. |
1584 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1585 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1586 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1593 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1594 Values of these types are the only ones which can be produced by
1602 These are the types that are valid in registers from CodeGen's perspective.
1611 The integer type is a very simple type that simply specifies an
1612 arbitrary bit width for the integer type desired. Any bit width from 1
1613 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1621 The number of bits the integer will occupy is specified by the ``N``
1627 +----------------+------------------------------------------------+
1628 | ``i1`` | a single-bit integer. |
1629 +----------------+------------------------------------------------+
1630 | ``i32`` | a 32-bit integer. |
1631 +----------------+------------------------------------------------+
1632 | ``i1942652`` | a really big integer of over 1 million bits. |
1633 +----------------+------------------------------------------------+
1637 Floating Point Types
1638 """"""""""""""""""""
1647 - 16-bit floating point value
1650 - 32-bit floating point value
1653 - 64-bit floating point value
1656 - 128-bit floating point value (112-bit mantissa)
1659 - 80-bit floating point value (X87)
1662 - 128-bit floating point value (two 64-bits)
1671 The x86mmx type represents a value held in an MMX register on an x86
1672 machine. The operations allowed on it are quite limited: parameters and
1673 return values, load and store, and bitcast. User-specified MMX
1674 instructions are represented as intrinsic or asm calls with arguments
1675 and/or results of this type. There are no arrays, vectors or constants
1692 The pointer type is used to specify memory locations. Pointers are
1693 commonly used to reference objects in memory.
1695 Pointer types may have an optional address space attribute defining the
1696 numbered address space where the pointed-to object resides. The default
1697 address space is number zero. The semantics of non-zero address spaces
1698 are target-specific.
1700 Note that LLVM does not permit pointers to void (``void*``) nor does it
1701 permit pointers to labels (``label*``). Use ``i8*`` instead.
1711 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1712 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1713 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1714 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1715 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1716 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1717 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1726 A vector type is a simple derived type that represents a vector of
1727 elements. Vector types are used when multiple primitive data are
1728 operated in parallel using a single instruction (SIMD). A vector type
1729 requires a size (number of elements) and an underlying primitive data
1730 type. Vector types are considered :ref:`first class <t_firstclass>`.
1736 < <# elements> x <elementtype> >
1738 The number of elements is a constant integer value larger than 0;
1739 elementtype may be any integer or floating point type, or a pointer to
1740 these types. Vectors of size zero are not allowed.
1744 +-------------------+--------------------------------------------------+
1745 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1746 +-------------------+--------------------------------------------------+
1747 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1748 +-------------------+--------------------------------------------------+
1749 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1750 +-------------------+--------------------------------------------------+
1751 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1752 +-------------------+--------------------------------------------------+
1761 The label type represents code labels.
1776 The metadata type represents embedded metadata. No derived types may be
1777 created from metadata except for :ref:`function <t_function>` arguments.
1790 Aggregate Types are a subset of derived types that can contain multiple
1791 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1792 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1802 The array type is a very simple derived type that arranges elements
1803 sequentially in memory. The array type requires a size (number of
1804 elements) and an underlying data type.
1810 [<# elements> x <elementtype>]
1812 The number of elements is a constant integer value; ``elementtype`` may
1813 be any type with a size.
1817 +------------------+--------------------------------------+
1818 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1819 +------------------+--------------------------------------+
1820 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1821 +------------------+--------------------------------------+
1822 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1823 +------------------+--------------------------------------+
1825 Here are some examples of multidimensional arrays:
1827 +-----------------------------+----------------------------------------------------------+
1828 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1829 +-----------------------------+----------------------------------------------------------+
1830 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1831 +-----------------------------+----------------------------------------------------------+
1832 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1833 +-----------------------------+----------------------------------------------------------+
1835 There is no restriction on indexing beyond the end of the array implied
1836 by a static type (though there are restrictions on indexing beyond the
1837 bounds of an allocated object in some cases). This means that
1838 single-dimension 'variable sized array' addressing can be implemented in
1839 LLVM with a zero length array type. An implementation of 'pascal style
1840 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1850 The structure type is used to represent a collection of data members
1851 together in memory. The elements of a structure may be any type that has
1854 Structures in memory are accessed using '``load``' and '``store``' by
1855 getting a pointer to a field with the '``getelementptr``' instruction.
1856 Structures in registers are accessed using the '``extractvalue``' and
1857 '``insertvalue``' instructions.
1859 Structures may optionally be "packed" structures, which indicate that
1860 the alignment of the struct is one byte, and that there is no padding
1861 between the elements. In non-packed structs, padding between field types
1862 is inserted as defined by the DataLayout string in the module, which is
1863 required to match what the underlying code generator expects.
1865 Structures can either be "literal" or "identified". A literal structure
1866 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1867 identified types are always defined at the top level with a name.
1868 Literal types are uniqued by their contents and can never be recursive
1869 or opaque since there is no way to write one. Identified types can be
1870 recursive, can be opaqued, and are never uniqued.
1876 %T1 = type { <type list> } ; Identified normal struct type
1877 %T2 = type <{ <type list> }> ; Identified packed struct type
1881 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1882 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1883 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1884 | ``{ 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``. |
1885 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1886 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1887 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1891 Opaque Structure Types
1892 """"""""""""""""""""""
1896 Opaque structure types are used to represent named structure types that
1897 do not have a body specified. This corresponds (for example) to the C
1898 notion of a forward declared structure.
1909 +--------------+-------------------+
1910 | ``opaque`` | An opaque type. |
1911 +--------------+-------------------+
1916 LLVM has several different basic types of constants. This section
1917 describes them all and their syntax.
1922 **Boolean constants**
1923 The two strings '``true``' and '``false``' are both valid constants
1925 **Integer constants**
1926 Standard integers (such as '4') are constants of the
1927 :ref:`integer <t_integer>` type. Negative numbers may be used with
1929 **Floating point constants**
1930 Floating point constants use standard decimal notation (e.g.
1931 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1932 hexadecimal notation (see below). The assembler requires the exact
1933 decimal value of a floating-point constant. For example, the
1934 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1935 decimal in binary. Floating point constants must have a :ref:`floating
1936 point <t_floating>` type.
1937 **Null pointer constants**
1938 The identifier '``null``' is recognized as a null pointer constant
1939 and must be of :ref:`pointer type <t_pointer>`.
1941 The one non-intuitive notation for constants is the hexadecimal form of
1942 floating point constants. For example, the form
1943 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1944 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1945 constants are required (and the only time that they are generated by the
1946 disassembler) is when a floating point constant must be emitted but it
1947 cannot be represented as a decimal floating point number in a reasonable
1948 number of digits. For example, NaN's, infinities, and other special
1949 values are represented in their IEEE hexadecimal format so that assembly
1950 and disassembly do not cause any bits to change in the constants.
1952 When using the hexadecimal form, constants of types half, float, and
1953 double are represented using the 16-digit form shown above (which
1954 matches the IEEE754 representation for double); half and float values
1955 must, however, be exactly representable as IEEE 754 half and single
1956 precision, respectively. Hexadecimal format is always used for long
1957 double, and there are three forms of long double. The 80-bit format used
1958 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1959 128-bit format used by PowerPC (two adjacent doubles) is represented by
1960 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1961 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1962 will only work if they match the long double format on your target.
1963 The IEEE 16-bit format (half precision) is represented by ``0xH``
1964 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1965 (sign bit at the left).
1967 There are no constants of type x86mmx.
1969 .. _complexconstants:
1974 Complex constants are a (potentially recursive) combination of simple
1975 constants and smaller complex constants.
1977 **Structure constants**
1978 Structure constants are represented with notation similar to
1979 structure type definitions (a comma separated list of elements,
1980 surrounded by braces (``{}``)). For example:
1981 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1982 "``@G = external global i32``". Structure constants must have
1983 :ref:`structure type <t_struct>`, and the number and types of elements
1984 must match those specified by the type.
1986 Array constants are represented with notation similar to array type
1987 definitions (a comma separated list of elements, surrounded by
1988 square brackets (``[]``)). For example:
1989 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1990 :ref:`array type <t_array>`, and the number and types of elements must
1991 match those specified by the type.
1992 **Vector constants**
1993 Vector constants are represented with notation similar to vector
1994 type definitions (a comma separated list of elements, surrounded by
1995 less-than/greater-than's (``<>``)). For example:
1996 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1997 must have :ref:`vector type <t_vector>`, and the number and types of
1998 elements must match those specified by the type.
1999 **Zero initialization**
2000 The string '``zeroinitializer``' can be used to zero initialize a
2001 value to zero of *any* type, including scalar and
2002 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2003 having to print large zero initializers (e.g. for large arrays) and
2004 is always exactly equivalent to using explicit zero initializers.
2006 A metadata node is a structure-like constant with :ref:`metadata
2007 type <t_metadata>`. For example:
2008 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2009 constants that are meant to be interpreted as part of the
2010 instruction stream, metadata is a place to attach additional
2011 information such as debug info.
2013 Global Variable and Function Addresses
2014 --------------------------------------
2016 The addresses of :ref:`global variables <globalvars>` and
2017 :ref:`functions <functionstructure>` are always implicitly valid
2018 (link-time) constants. These constants are explicitly referenced when
2019 the :ref:`identifier for the global <identifiers>` is used and always have
2020 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2023 .. code-block:: llvm
2027 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2034 The string '``undef``' can be used anywhere a constant is expected, and
2035 indicates that the user of the value may receive an unspecified
2036 bit-pattern. Undefined values may be of any type (other than '``label``'
2037 or '``void``') and be used anywhere a constant is permitted.
2039 Undefined values are useful because they indicate to the compiler that
2040 the program is well defined no matter what value is used. This gives the
2041 compiler more freedom to optimize. Here are some examples of
2042 (potentially surprising) transformations that are valid (in pseudo IR):
2044 .. code-block:: llvm
2054 This is safe because all of the output bits are affected by the undef
2055 bits. Any output bit can have a zero or one depending on the input bits.
2057 .. code-block:: llvm
2068 These logical operations have bits that are not always affected by the
2069 input. For example, if ``%X`` has a zero bit, then the output of the
2070 '``and``' operation will always be a zero for that bit, no matter what
2071 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2072 optimize or assume that the result of the '``and``' is '``undef``'.
2073 However, it is safe to assume that all bits of the '``undef``' could be
2074 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2075 all the bits of the '``undef``' operand to the '``or``' could be set,
2076 allowing the '``or``' to be folded to -1.
2078 .. code-block:: llvm
2080 %A = select undef, %X, %Y
2081 %B = select undef, 42, %Y
2082 %C = select %X, %Y, undef
2092 This set of examples shows that undefined '``select``' (and conditional
2093 branch) conditions can go *either way*, but they have to come from one
2094 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2095 both known to have a clear low bit, then ``%A`` would have to have a
2096 cleared low bit. However, in the ``%C`` example, the optimizer is
2097 allowed to assume that the '``undef``' operand could be the same as
2098 ``%Y``, allowing the whole '``select``' to be eliminated.
2100 .. code-block:: llvm
2102 %A = xor undef, undef
2119 This example points out that two '``undef``' operands are not
2120 necessarily the same. This can be surprising to people (and also matches
2121 C semantics) where they assume that "``X^X``" is always zero, even if
2122 ``X`` is undefined. This isn't true for a number of reasons, but the
2123 short answer is that an '``undef``' "variable" can arbitrarily change
2124 its value over its "live range". This is true because the variable
2125 doesn't actually *have a live range*. Instead, the value is logically
2126 read from arbitrary registers that happen to be around when needed, so
2127 the value is not necessarily consistent over time. In fact, ``%A`` and
2128 ``%C`` need to have the same semantics or the core LLVM "replace all
2129 uses with" concept would not hold.
2131 .. code-block:: llvm
2139 These examples show the crucial difference between an *undefined value*
2140 and *undefined behavior*. An undefined value (like '``undef``') is
2141 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2142 operation can be constant folded to '``undef``', because the '``undef``'
2143 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2144 However, in the second example, we can make a more aggressive
2145 assumption: because the ``undef`` is allowed to be an arbitrary value,
2146 we are allowed to assume that it could be zero. Since a divide by zero
2147 has *undefined behavior*, we are allowed to assume that the operation
2148 does not execute at all. This allows us to delete the divide and all
2149 code after it. Because the undefined operation "can't happen", the
2150 optimizer can assume that it occurs in dead code.
2152 .. code-block:: llvm
2154 a: store undef -> %X
2155 b: store %X -> undef
2160 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2161 value can be assumed to not have any effect; we can assume that the
2162 value is overwritten with bits that happen to match what was already
2163 there. However, a store *to* an undefined location could clobber
2164 arbitrary memory, therefore, it has undefined behavior.
2171 Poison values are similar to :ref:`undef values <undefvalues>`, however
2172 they also represent the fact that an instruction or constant expression
2173 which cannot evoke side effects has nevertheless detected a condition
2174 which results in undefined behavior.
2176 There is currently no way of representing a poison value in the IR; they
2177 only exist when produced by operations such as :ref:`add <i_add>` with
2180 Poison value behavior is defined in terms of value *dependence*:
2182 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2183 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2184 their dynamic predecessor basic block.
2185 - Function arguments depend on the corresponding actual argument values
2186 in the dynamic callers of their functions.
2187 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2188 instructions that dynamically transfer control back to them.
2189 - :ref:`Invoke <i_invoke>` instructions depend on the
2190 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2191 call instructions that dynamically transfer control back to them.
2192 - Non-volatile loads and stores depend on the most recent stores to all
2193 of the referenced memory addresses, following the order in the IR
2194 (including loads and stores implied by intrinsics such as
2195 :ref:`@llvm.memcpy <int_memcpy>`.)
2196 - An instruction with externally visible side effects depends on the
2197 most recent preceding instruction with externally visible side
2198 effects, following the order in the IR. (This includes :ref:`volatile
2199 operations <volatile>`.)
2200 - An instruction *control-depends* on a :ref:`terminator
2201 instruction <terminators>` if the terminator instruction has
2202 multiple successors and the instruction is always executed when
2203 control transfers to one of the successors, and may not be executed
2204 when control is transferred to another.
2205 - Additionally, an instruction also *control-depends* on a terminator
2206 instruction if the set of instructions it otherwise depends on would
2207 be different if the terminator had transferred control to a different
2209 - Dependence is transitive.
2211 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2212 with the additional affect that any instruction which has a *dependence*
2213 on a poison value has undefined behavior.
2215 Here are some examples:
2217 .. code-block:: llvm
2220 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2221 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2222 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2223 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2225 store i32 %poison, i32* @g ; Poison value stored to memory.
2226 %poison2 = load i32* @g ; Poison value loaded back from memory.
2228 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2230 %narrowaddr = bitcast i32* @g to i16*
2231 %wideaddr = bitcast i32* @g to i64*
2232 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2233 %poison4 = load i64* %wideaddr ; Returns a poison value.
2235 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2236 br i1 %cmp, label %true, label %end ; Branch to either destination.
2239 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2240 ; it has undefined behavior.
2244 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2245 ; Both edges into this PHI are
2246 ; control-dependent on %cmp, so this
2247 ; always results in a poison value.
2249 store volatile i32 0, i32* @g ; This would depend on the store in %true
2250 ; if %cmp is true, or the store in %entry
2251 ; otherwise, so this is undefined behavior.
2253 br i1 %cmp, label %second_true, label %second_end
2254 ; The same branch again, but this time the
2255 ; true block doesn't have side effects.
2262 store volatile i32 0, i32* @g ; This time, the instruction always depends
2263 ; on the store in %end. Also, it is
2264 ; control-equivalent to %end, so this is
2265 ; well-defined (ignoring earlier undefined
2266 ; behavior in this example).
2270 Addresses of Basic Blocks
2271 -------------------------
2273 ``blockaddress(@function, %block)``
2275 The '``blockaddress``' constant computes the address of the specified
2276 basic block in the specified function, and always has an ``i8*`` type.
2277 Taking the address of the entry block is illegal.
2279 This value only has defined behavior when used as an operand to the
2280 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2281 against null. Pointer equality tests between labels addresses results in
2282 undefined behavior --- though, again, comparison against null is ok, and
2283 no label is equal to the null pointer. This may be passed around as an
2284 opaque pointer sized value as long as the bits are not inspected. This
2285 allows ``ptrtoint`` and arithmetic to be performed on these values so
2286 long as the original value is reconstituted before the ``indirectbr``
2289 Finally, some targets may provide defined semantics when using the value
2290 as the operand to an inline assembly, but that is target specific.
2294 Constant Expressions
2295 --------------------
2297 Constant expressions are used to allow expressions involving other
2298 constants to be used as constants. Constant expressions may be of any
2299 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2300 that does not have side effects (e.g. load and call are not supported).
2301 The following is the syntax for constant expressions:
2303 ``trunc (CST to TYPE)``
2304 Truncate a constant to another type. The bit size of CST must be
2305 larger than the bit size of TYPE. Both types must be integers.
2306 ``zext (CST to TYPE)``
2307 Zero extend a constant to another type. The bit size of CST must be
2308 smaller than the bit size of TYPE. Both types must be integers.
2309 ``sext (CST to TYPE)``
2310 Sign extend a constant to another type. The bit size of CST must be
2311 smaller than the bit size of TYPE. Both types must be integers.
2312 ``fptrunc (CST to TYPE)``
2313 Truncate a floating point constant to another floating point type.
2314 The size of CST must be larger than the size of TYPE. Both types
2315 must be floating point.
2316 ``fpext (CST to TYPE)``
2317 Floating point extend a constant to another type. The size of CST
2318 must be smaller or equal to the size of TYPE. Both types must be
2320 ``fptoui (CST to TYPE)``
2321 Convert a floating point constant to the corresponding unsigned
2322 integer constant. TYPE must be a scalar or vector integer type. CST
2323 must be of scalar or vector floating point type. Both CST and TYPE
2324 must be scalars, or vectors of the same number of elements. If the
2325 value won't fit in the integer type, the results are undefined.
2326 ``fptosi (CST to TYPE)``
2327 Convert a floating point constant to the corresponding signed
2328 integer constant. TYPE must be a scalar or vector integer type. CST
2329 must be of scalar or vector floating point type. Both CST and TYPE
2330 must be scalars, or vectors of the same number of elements. If the
2331 value won't fit in the integer type, the results are undefined.
2332 ``uitofp (CST to TYPE)``
2333 Convert an unsigned integer constant to the corresponding floating
2334 point constant. TYPE must be a scalar or vector floating point type.
2335 CST must be of scalar or vector integer type. Both CST and TYPE must
2336 be scalars, or vectors of the same number of elements. If the value
2337 won't fit in the floating point type, the results are undefined.
2338 ``sitofp (CST to TYPE)``
2339 Convert a signed integer constant to the corresponding floating
2340 point constant. TYPE must be a scalar or vector floating point type.
2341 CST must be of scalar or vector integer type. Both CST and TYPE must
2342 be scalars, or vectors of the same number of elements. If the value
2343 won't fit in the floating point type, the results are undefined.
2344 ``ptrtoint (CST to TYPE)``
2345 Convert a pointer typed constant to the corresponding integer
2346 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2347 pointer type. The ``CST`` value is zero extended, truncated, or
2348 unchanged to make it fit in ``TYPE``.
2349 ``inttoptr (CST to TYPE)``
2350 Convert an integer constant to a pointer constant. TYPE must be a
2351 pointer type. CST must be of integer type. The CST value is zero
2352 extended, truncated, or unchanged to make it fit in a pointer size.
2353 This one is *really* dangerous!
2354 ``bitcast (CST to TYPE)``
2355 Convert a constant, CST, to another TYPE. The constraints of the
2356 operands are the same as those for the :ref:`bitcast
2357 instruction <i_bitcast>`.
2358 ``addrspacecast (CST to TYPE)``
2359 Convert a constant pointer or constant vector of pointer, CST, to another
2360 TYPE in a different address space. The constraints of the operands are the
2361 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2362 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2363 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2364 constants. As with the :ref:`getelementptr <i_getelementptr>`
2365 instruction, the index list may have zero or more indexes, which are
2366 required to make sense for the type of "CSTPTR".
2367 ``select (COND, VAL1, VAL2)``
2368 Perform the :ref:`select operation <i_select>` on constants.
2369 ``icmp COND (VAL1, VAL2)``
2370 Performs the :ref:`icmp operation <i_icmp>` on constants.
2371 ``fcmp COND (VAL1, VAL2)``
2372 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2373 ``extractelement (VAL, IDX)``
2374 Perform the :ref:`extractelement operation <i_extractelement>` on
2376 ``insertelement (VAL, ELT, IDX)``
2377 Perform the :ref:`insertelement operation <i_insertelement>` on
2379 ``shufflevector (VEC1, VEC2, IDXMASK)``
2380 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2382 ``extractvalue (VAL, IDX0, IDX1, ...)``
2383 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2384 constants. The index list is interpreted in a similar manner as
2385 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2386 least one index value must be specified.
2387 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2388 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2389 The index list is interpreted in a similar manner as indices in a
2390 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2391 value must be specified.
2392 ``OPCODE (LHS, RHS)``
2393 Perform the specified operation of the LHS and RHS constants. OPCODE
2394 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2395 binary <bitwiseops>` operations. The constraints on operands are
2396 the same as those for the corresponding instruction (e.g. no bitwise
2397 operations on floating point values are allowed).
2404 Inline Assembler Expressions
2405 ----------------------------
2407 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2408 Inline Assembly <moduleasm>`) through the use of a special value. This
2409 value represents the inline assembler as a string (containing the
2410 instructions to emit), a list of operand constraints (stored as a
2411 string), a flag that indicates whether or not the inline asm expression
2412 has side effects, and a flag indicating whether the function containing
2413 the asm needs to align its stack conservatively. An example inline
2414 assembler expression is:
2416 .. code-block:: llvm
2418 i32 (i32) asm "bswap $0", "=r,r"
2420 Inline assembler expressions may **only** be used as the callee operand
2421 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2422 Thus, typically we have:
2424 .. code-block:: llvm
2426 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2428 Inline asms with side effects not visible in the constraint list must be
2429 marked as having side effects. This is done through the use of the
2430 '``sideeffect``' keyword, like so:
2432 .. code-block:: llvm
2434 call void asm sideeffect "eieio", ""()
2436 In some cases inline asms will contain code that will not work unless
2437 the stack is aligned in some way, such as calls or SSE instructions on
2438 x86, yet will not contain code that does that alignment within the asm.
2439 The compiler should make conservative assumptions about what the asm
2440 might contain and should generate its usual stack alignment code in the
2441 prologue if the '``alignstack``' keyword is present:
2443 .. code-block:: llvm
2445 call void asm alignstack "eieio", ""()
2447 Inline asms also support using non-standard assembly dialects. The
2448 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2449 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2450 the only supported dialects. An example is:
2452 .. code-block:: llvm
2454 call void asm inteldialect "eieio", ""()
2456 If multiple keywords appear the '``sideeffect``' keyword must come
2457 first, the '``alignstack``' keyword second and the '``inteldialect``'
2463 The call instructions that wrap inline asm nodes may have a
2464 "``!srcloc``" MDNode attached to it that contains a list of constant
2465 integers. If present, the code generator will use the integer as the
2466 location cookie value when report errors through the ``LLVMContext``
2467 error reporting mechanisms. This allows a front-end to correlate backend
2468 errors that occur with inline asm back to the source code that produced
2471 .. code-block:: llvm
2473 call void asm sideeffect "something bad", ""(), !srcloc !42
2475 !42 = !{ i32 1234567 }
2477 It is up to the front-end to make sense of the magic numbers it places
2478 in the IR. If the MDNode contains multiple constants, the code generator
2479 will use the one that corresponds to the line of the asm that the error
2484 Metadata Nodes and Metadata Strings
2485 -----------------------------------
2487 LLVM IR allows metadata to be attached to instructions in the program
2488 that can convey extra information about the code to the optimizers and
2489 code generator. One example application of metadata is source-level
2490 debug information. There are two metadata primitives: strings and nodes.
2491 All metadata has the ``metadata`` type and is identified in syntax by a
2492 preceding exclamation point ('``!``').
2494 A metadata string is a string surrounded by double quotes. It can
2495 contain any character by escaping non-printable characters with
2496 "``\xx``" where "``xx``" is the two digit hex code. For example:
2499 Metadata nodes are represented with notation similar to structure
2500 constants (a comma separated list of elements, surrounded by braces and
2501 preceded by an exclamation point). Metadata nodes can have any values as
2502 their operand. For example:
2504 .. code-block:: llvm
2506 !{ metadata !"test\00", i32 10}
2508 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2509 metadata nodes, which can be looked up in the module symbol table. For
2512 .. code-block:: llvm
2514 !foo = metadata !{!4, !3}
2516 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2517 function is using two metadata arguments:
2519 .. code-block:: llvm
2521 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2523 Metadata can be attached with an instruction. Here metadata ``!21`` is
2524 attached to the ``add`` instruction using the ``!dbg`` identifier:
2526 .. code-block:: llvm
2528 %indvar.next = add i64 %indvar, 1, !dbg !21
2530 More information about specific metadata nodes recognized by the
2531 optimizers and code generator is found below.
2536 In LLVM IR, memory does not have types, so LLVM's own type system is not
2537 suitable for doing TBAA. Instead, metadata is added to the IR to
2538 describe a type system of a higher level language. This can be used to
2539 implement typical C/C++ TBAA, but it can also be used to implement
2540 custom alias analysis behavior for other languages.
2542 The current metadata format is very simple. TBAA metadata nodes have up
2543 to three fields, e.g.:
2545 .. code-block:: llvm
2547 !0 = metadata !{ metadata !"an example type tree" }
2548 !1 = metadata !{ metadata !"int", metadata !0 }
2549 !2 = metadata !{ metadata !"float", metadata !0 }
2550 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2552 The first field is an identity field. It can be any value, usually a
2553 metadata string, which uniquely identifies the type. The most important
2554 name in the tree is the name of the root node. Two trees with different
2555 root node names are entirely disjoint, even if they have leaves with
2558 The second field identifies the type's parent node in the tree, or is
2559 null or omitted for a root node. A type is considered to alias all of
2560 its descendants and all of its ancestors in the tree. Also, a type is
2561 considered to alias all types in other trees, so that bitcode produced
2562 from multiple front-ends is handled conservatively.
2564 If the third field is present, it's an integer which if equal to 1
2565 indicates that the type is "constant" (meaning
2566 ``pointsToConstantMemory`` should return true; see `other useful
2567 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2569 '``tbaa.struct``' Metadata
2570 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2572 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2573 aggregate assignment operations in C and similar languages, however it
2574 is defined to copy a contiguous region of memory, which is more than
2575 strictly necessary for aggregate types which contain holes due to
2576 padding. Also, it doesn't contain any TBAA information about the fields
2579 ``!tbaa.struct`` metadata can describe which memory subregions in a
2580 memcpy are padding and what the TBAA tags of the struct are.
2582 The current metadata format is very simple. ``!tbaa.struct`` metadata
2583 nodes are a list of operands which are in conceptual groups of three.
2584 For each group of three, the first operand gives the byte offset of a
2585 field in bytes, the second gives its size in bytes, and the third gives
2588 .. code-block:: llvm
2590 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2592 This describes a struct with two fields. The first is at offset 0 bytes
2593 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2594 and has size 4 bytes and has tbaa tag !2.
2596 Note that the fields need not be contiguous. In this example, there is a
2597 4 byte gap between the two fields. This gap represents padding which
2598 does not carry useful data and need not be preserved.
2600 '``fpmath``' Metadata
2601 ^^^^^^^^^^^^^^^^^^^^^
2603 ``fpmath`` metadata may be attached to any instruction of floating point
2604 type. It can be used to express the maximum acceptable error in the
2605 result of that instruction, in ULPs, thus potentially allowing the
2606 compiler to use a more efficient but less accurate method of computing
2607 it. ULP is defined as follows:
2609 If ``x`` is a real number that lies between two finite consecutive
2610 floating-point numbers ``a`` and ``b``, without being equal to one
2611 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2612 distance between the two non-equal finite floating-point numbers
2613 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2615 The metadata node shall consist of a single positive floating point
2616 number representing the maximum relative error, for example:
2618 .. code-block:: llvm
2620 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2622 '``range``' Metadata
2623 ^^^^^^^^^^^^^^^^^^^^
2625 ``range`` metadata may be attached only to loads of integer types. It
2626 expresses the possible ranges the loaded value is in. The ranges are
2627 represented with a flattened list of integers. The loaded value is known
2628 to be in the union of the ranges defined by each consecutive pair. Each
2629 pair has the following properties:
2631 - The type must match the type loaded by the instruction.
2632 - The pair ``a,b`` represents the range ``[a,b)``.
2633 - Both ``a`` and ``b`` are constants.
2634 - The range is allowed to wrap.
2635 - The range should not represent the full or empty set. That is,
2638 In addition, the pairs must be in signed order of the lower bound and
2639 they must be non-contiguous.
2643 .. code-block:: llvm
2645 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2646 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2647 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2648 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2650 !0 = metadata !{ i8 0, i8 2 }
2651 !1 = metadata !{ i8 255, i8 2 }
2652 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2653 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2658 It is sometimes useful to attach information to loop constructs. Currently,
2659 loop metadata is implemented as metadata attached to the branch instruction
2660 in the loop latch block. This type of metadata refer to a metadata node that is
2661 guaranteed to be separate for each loop. The loop identifier metadata is
2662 specified with the name ``llvm.loop``.
2664 The loop identifier metadata is implemented using a metadata that refers to
2665 itself to avoid merging it with any other identifier metadata, e.g.,
2666 during module linkage or function inlining. That is, each loop should refer
2667 to their own identification metadata even if they reside in separate functions.
2668 The following example contains loop identifier metadata for two separate loop
2671 .. code-block:: llvm
2673 !0 = metadata !{ metadata !0 }
2674 !1 = metadata !{ metadata !1 }
2676 The loop identifier metadata can be used to specify additional per-loop
2677 metadata. Any operands after the first operand can be treated as user-defined
2678 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2679 by the loop vectorizer to indicate how many times to unroll the loop:
2681 .. code-block:: llvm
2683 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2685 !0 = metadata !{ metadata !0, metadata !1 }
2686 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2691 Metadata types used to annotate memory accesses with information helpful
2692 for optimizations are prefixed with ``llvm.mem``.
2694 '``llvm.mem.parallel_loop_access``' Metadata
2695 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2697 For a loop to be parallel, in addition to using
2698 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2699 also all of the memory accessing instructions in the loop body need to be
2700 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2701 is at least one memory accessing instruction not marked with the metadata,
2702 the loop must be considered a sequential loop. This causes parallel loops to be
2703 converted to sequential loops due to optimization passes that are unaware of
2704 the parallel semantics and that insert new memory instructions to the loop
2707 Example of a loop that is considered parallel due to its correct use of
2708 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2709 metadata types that refer to the same loop identifier metadata.
2711 .. code-block:: llvm
2715 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2717 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2719 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2723 !0 = metadata !{ metadata !0 }
2725 It is also possible to have nested parallel loops. In that case the
2726 memory accesses refer to a list of loop identifier metadata nodes instead of
2727 the loop identifier metadata node directly:
2729 .. code-block:: llvm
2736 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2738 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2740 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2744 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2746 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2748 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2750 outer.for.end: ; preds = %for.body
2752 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2753 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2754 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2756 '``llvm.vectorizer``'
2757 ^^^^^^^^^^^^^^^^^^^^^
2759 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2760 vectorization parameters such as vectorization factor and unroll factor.
2762 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2763 loop identification metadata.
2765 '``llvm.vectorizer.unroll``' Metadata
2766 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2768 This metadata instructs the loop vectorizer to unroll the specified
2769 loop exactly ``N`` times.
2771 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2772 operand is an integer specifying the unroll factor. For example:
2774 .. code-block:: llvm
2776 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2778 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2781 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2782 determined automatically.
2784 '``llvm.vectorizer.width``' Metadata
2785 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2787 This metadata sets the target width of the vectorizer to ``N``. Without
2788 this metadata, the vectorizer will choose a width automatically.
2789 Regardless of this metadata, the vectorizer will only vectorize loops if
2790 it believes it is valid to do so.
2792 The first operand is the string ``llvm.vectorizer.width`` and the second
2793 operand is an integer specifying the width. For example:
2795 .. code-block:: llvm
2797 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2799 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2802 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2805 Module Flags Metadata
2806 =====================
2808 Information about the module as a whole is difficult to convey to LLVM's
2809 subsystems. The LLVM IR isn't sufficient to transmit this information.
2810 The ``llvm.module.flags`` named metadata exists in order to facilitate
2811 this. These flags are in the form of key / value pairs --- much like a
2812 dictionary --- making it easy for any subsystem who cares about a flag to
2815 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2816 Each triplet has the following form:
2818 - The first element is a *behavior* flag, which specifies the behavior
2819 when two (or more) modules are merged together, and it encounters two
2820 (or more) metadata with the same ID. The supported behaviors are
2822 - The second element is a metadata string that is a unique ID for the
2823 metadata. Each module may only have one flag entry for each unique ID (not
2824 including entries with the **Require** behavior).
2825 - The third element is the value of the flag.
2827 When two (or more) modules are merged together, the resulting
2828 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2829 each unique metadata ID string, there will be exactly one entry in the merged
2830 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2831 be determined by the merge behavior flag, as described below. The only exception
2832 is that entries with the *Require* behavior are always preserved.
2834 The following behaviors are supported:
2845 Emits an error if two values disagree, otherwise the resulting value
2846 is that of the operands.
2850 Emits a warning if two values disagree. The result value will be the
2851 operand for the flag from the first module being linked.
2855 Adds a requirement that another module flag be present and have a
2856 specified value after linking is performed. The value must be a
2857 metadata pair, where the first element of the pair is the ID of the
2858 module flag to be restricted, and the second element of the pair is
2859 the value the module flag should be restricted to. This behavior can
2860 be used to restrict the allowable results (via triggering of an
2861 error) of linking IDs with the **Override** behavior.
2865 Uses the specified value, regardless of the behavior or value of the
2866 other module. If both modules specify **Override**, but the values
2867 differ, an error will be emitted.
2871 Appends the two values, which are required to be metadata nodes.
2875 Appends the two values, which are required to be metadata
2876 nodes. However, duplicate entries in the second list are dropped
2877 during the append operation.
2879 It is an error for a particular unique flag ID to have multiple behaviors,
2880 except in the case of **Require** (which adds restrictions on another metadata
2881 value) or **Override**.
2883 An example of module flags:
2885 .. code-block:: llvm
2887 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2888 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2889 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2890 !3 = metadata !{ i32 3, metadata !"qux",
2892 metadata !"foo", i32 1
2895 !llvm.module.flags = !{ !0, !1, !2, !3 }
2897 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2898 if two or more ``!"foo"`` flags are seen is to emit an error if their
2899 values are not equal.
2901 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2902 behavior if two or more ``!"bar"`` flags are seen is to use the value
2905 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2906 behavior if two or more ``!"qux"`` flags are seen is to emit a
2907 warning if their values are not equal.
2909 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2913 metadata !{ metadata !"foo", i32 1 }
2915 The behavior is to emit an error if the ``llvm.module.flags`` does not
2916 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2919 Objective-C Garbage Collection Module Flags Metadata
2920 ----------------------------------------------------
2922 On the Mach-O platform, Objective-C stores metadata about garbage
2923 collection in a special section called "image info". The metadata
2924 consists of a version number and a bitmask specifying what types of
2925 garbage collection are supported (if any) by the file. If two or more
2926 modules are linked together their garbage collection metadata needs to
2927 be merged rather than appended together.
2929 The Objective-C garbage collection module flags metadata consists of the
2930 following key-value pairs:
2939 * - ``Objective-C Version``
2940 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2942 * - ``Objective-C Image Info Version``
2943 - **[Required]** --- The version of the image info section. Currently
2946 * - ``Objective-C Image Info Section``
2947 - **[Required]** --- The section to place the metadata. Valid values are
2948 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2949 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2950 Objective-C ABI version 2.
2952 * - ``Objective-C Garbage Collection``
2953 - **[Required]** --- Specifies whether garbage collection is supported or
2954 not. Valid values are 0, for no garbage collection, and 2, for garbage
2955 collection supported.
2957 * - ``Objective-C GC Only``
2958 - **[Optional]** --- Specifies that only garbage collection is supported.
2959 If present, its value must be 6. This flag requires that the
2960 ``Objective-C Garbage Collection`` flag have the value 2.
2962 Some important flag interactions:
2964 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2965 merged with a module with ``Objective-C Garbage Collection`` set to
2966 2, then the resulting module has the
2967 ``Objective-C Garbage Collection`` flag set to 0.
2968 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2969 merged with a module with ``Objective-C GC Only`` set to 6.
2971 Automatic Linker Flags Module Flags Metadata
2972 --------------------------------------------
2974 Some targets support embedding flags to the linker inside individual object
2975 files. Typically this is used in conjunction with language extensions which
2976 allow source files to explicitly declare the libraries they depend on, and have
2977 these automatically be transmitted to the linker via object files.
2979 These flags are encoded in the IR using metadata in the module flags section,
2980 using the ``Linker Options`` key. The merge behavior for this flag is required
2981 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2982 node which should be a list of other metadata nodes, each of which should be a
2983 list of metadata strings defining linker options.
2985 For example, the following metadata section specifies two separate sets of
2986 linker options, presumably to link against ``libz`` and the ``Cocoa``
2989 !0 = metadata !{ i32 6, metadata !"Linker Options",
2991 metadata !{ metadata !"-lz" },
2992 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2993 !llvm.module.flags = !{ !0 }
2995 The metadata encoding as lists of lists of options, as opposed to a collapsed
2996 list of options, is chosen so that the IR encoding can use multiple option
2997 strings to specify e.g., a single library, while still having that specifier be
2998 preserved as an atomic element that can be recognized by a target specific
2999 assembly writer or object file emitter.
3001 Each individual option is required to be either a valid option for the target's
3002 linker, or an option that is reserved by the target specific assembly writer or
3003 object file emitter. No other aspect of these options is defined by the IR.
3005 .. _intrinsicglobalvariables:
3007 Intrinsic Global Variables
3008 ==========================
3010 LLVM has a number of "magic" global variables that contain data that
3011 affect code generation or other IR semantics. These are documented here.
3012 All globals of this sort should have a section specified as
3013 "``llvm.metadata``". This section and all globals that start with
3014 "``llvm.``" are reserved for use by LLVM.
3018 The '``llvm.used``' Global Variable
3019 -----------------------------------
3021 The ``@llvm.used`` global is an array which has
3022 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3023 pointers to named global variables, functions and aliases which may optionally
3024 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3027 .. code-block:: llvm
3032 @llvm.used = appending global [2 x i8*] [
3034 i8* bitcast (i32* @Y to i8*)
3035 ], section "llvm.metadata"
3037 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3038 and linker are required to treat the symbol as if there is a reference to the
3039 symbol that it cannot see (which is why they have to be named). For example, if
3040 a variable has internal linkage and no references other than that from the
3041 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3042 references from inline asms and other things the compiler cannot "see", and
3043 corresponds to "``attribute((used))``" in GNU C.
3045 On some targets, the code generator must emit a directive to the
3046 assembler or object file to prevent the assembler and linker from
3047 molesting the symbol.
3049 .. _gv_llvmcompilerused:
3051 The '``llvm.compiler.used``' Global Variable
3052 --------------------------------------------
3054 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3055 directive, except that it only prevents the compiler from touching the
3056 symbol. On targets that support it, this allows an intelligent linker to
3057 optimize references to the symbol without being impeded as it would be
3060 This is a rare construct that should only be used in rare circumstances,
3061 and should not be exposed to source languages.
3063 .. _gv_llvmglobalctors:
3065 The '``llvm.global_ctors``' Global Variable
3066 -------------------------------------------
3068 .. code-block:: llvm
3070 %0 = type { i32, void ()* }
3071 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3073 The ``@llvm.global_ctors`` array contains a list of constructor
3074 functions and associated priorities. The functions referenced by this
3075 array will be called in ascending order of priority (i.e. lowest first)
3076 when the module is loaded. The order of functions with the same priority
3079 .. _llvmglobaldtors:
3081 The '``llvm.global_dtors``' Global Variable
3082 -------------------------------------------
3084 .. code-block:: llvm
3086 %0 = type { i32, void ()* }
3087 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3089 The ``@llvm.global_dtors`` array contains a list of destructor functions
3090 and associated priorities. The functions referenced by this array will
3091 be called in descending order of priority (i.e. highest first) when the
3092 module is loaded. The order of functions with the same priority is not
3095 Instruction Reference
3096 =====================
3098 The LLVM instruction set consists of several different classifications
3099 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3100 instructions <binaryops>`, :ref:`bitwise binary
3101 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3102 :ref:`other instructions <otherops>`.
3106 Terminator Instructions
3107 -----------------------
3109 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3110 program ends with a "Terminator" instruction, which indicates which
3111 block should be executed after the current block is finished. These
3112 terminator instructions typically yield a '``void``' value: they produce
3113 control flow, not values (the one exception being the
3114 ':ref:`invoke <i_invoke>`' instruction).
3116 The terminator instructions are: ':ref:`ret <i_ret>`',
3117 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3118 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3119 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3123 '``ret``' Instruction
3124 ^^^^^^^^^^^^^^^^^^^^^
3131 ret <type> <value> ; Return a value from a non-void function
3132 ret void ; Return from void function
3137 The '``ret``' instruction is used to return control flow (and optionally
3138 a value) from a function back to the caller.
3140 There are two forms of the '``ret``' instruction: one that returns a
3141 value and then causes control flow, and one that just causes control
3147 The '``ret``' instruction optionally accepts a single argument, the
3148 return value. The type of the return value must be a ':ref:`first
3149 class <t_firstclass>`' type.
3151 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3152 return type and contains a '``ret``' instruction with no return value or
3153 a return value with a type that does not match its type, or if it has a
3154 void return type and contains a '``ret``' instruction with a return
3160 When the '``ret``' instruction is executed, control flow returns back to
3161 the calling function's context. If the caller is a
3162 ":ref:`call <i_call>`" instruction, execution continues at the
3163 instruction after the call. If the caller was an
3164 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3165 beginning of the "normal" destination block. If the instruction returns
3166 a value, that value shall set the call or invoke instruction's return
3172 .. code-block:: llvm
3174 ret i32 5 ; Return an integer value of 5
3175 ret void ; Return from a void function
3176 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3180 '``br``' Instruction
3181 ^^^^^^^^^^^^^^^^^^^^
3188 br i1 <cond>, label <iftrue>, label <iffalse>
3189 br label <dest> ; Unconditional branch
3194 The '``br``' instruction is used to cause control flow to transfer to a
3195 different basic block in the current function. There are two forms of
3196 this instruction, corresponding to a conditional branch and an
3197 unconditional branch.
3202 The conditional branch form of the '``br``' instruction takes a single
3203 '``i1``' value and two '``label``' values. The unconditional form of the
3204 '``br``' instruction takes a single '``label``' value as a target.
3209 Upon execution of a conditional '``br``' instruction, the '``i1``'
3210 argument is evaluated. If the value is ``true``, control flows to the
3211 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3212 to the '``iffalse``' ``label`` argument.
3217 .. code-block:: llvm
3220 %cond = icmp eq i32 %a, %b
3221 br i1 %cond, label %IfEqual, label %IfUnequal
3229 '``switch``' Instruction
3230 ^^^^^^^^^^^^^^^^^^^^^^^^
3237 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3242 The '``switch``' instruction is used to transfer control flow to one of
3243 several different places. It is a generalization of the '``br``'
3244 instruction, allowing a branch to occur to one of many possible
3250 The '``switch``' instruction uses three parameters: an integer
3251 comparison value '``value``', a default '``label``' destination, and an
3252 array of pairs of comparison value constants and '``label``'s. The table
3253 is not allowed to contain duplicate constant entries.
3258 The ``switch`` instruction specifies a table of values and destinations.
3259 When the '``switch``' instruction is executed, this table is searched
3260 for the given value. If the value is found, control flow is transferred
3261 to the corresponding destination; otherwise, control flow is transferred
3262 to the default destination.
3267 Depending on properties of the target machine and the particular
3268 ``switch`` instruction, this instruction may be code generated in
3269 different ways. For example, it could be generated as a series of
3270 chained conditional branches or with a lookup table.
3275 .. code-block:: llvm
3277 ; Emulate a conditional br instruction
3278 %Val = zext i1 %value to i32
3279 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3281 ; Emulate an unconditional br instruction
3282 switch i32 0, label %dest [ ]
3284 ; Implement a jump table:
3285 switch i32 %val, label %otherwise [ i32 0, label %onzero
3287 i32 2, label %ontwo ]
3291 '``indirectbr``' Instruction
3292 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3299 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3304 The '``indirectbr``' instruction implements an indirect branch to a
3305 label within the current function, whose address is specified by
3306 "``address``". Address must be derived from a
3307 :ref:`blockaddress <blockaddress>` constant.
3312 The '``address``' argument is the address of the label to jump to. The
3313 rest of the arguments indicate the full set of possible destinations
3314 that the address may point to. Blocks are allowed to occur multiple
3315 times in the destination list, though this isn't particularly useful.
3317 This destination list is required so that dataflow analysis has an
3318 accurate understanding of the CFG.
3323 Control transfers to the block specified in the address argument. All
3324 possible destination blocks must be listed in the label list, otherwise
3325 this instruction has undefined behavior. This implies that jumps to
3326 labels defined in other functions have undefined behavior as well.
3331 This is typically implemented with a jump through a register.
3336 .. code-block:: llvm
3338 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3342 '``invoke``' Instruction
3343 ^^^^^^^^^^^^^^^^^^^^^^^^
3350 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3351 to label <normal label> unwind label <exception label>
3356 The '``invoke``' instruction causes control to transfer to a specified
3357 function, with the possibility of control flow transfer to either the
3358 '``normal``' label or the '``exception``' label. If the callee function
3359 returns with the "``ret``" instruction, control flow will return to the
3360 "normal" label. If the callee (or any indirect callees) returns via the
3361 ":ref:`resume <i_resume>`" instruction or other exception handling
3362 mechanism, control is interrupted and continued at the dynamically
3363 nearest "exception" label.
3365 The '``exception``' label is a `landing
3366 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3367 '``exception``' label is required to have the
3368 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3369 information about the behavior of the program after unwinding happens,
3370 as its first non-PHI instruction. The restrictions on the
3371 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3372 instruction, so that the important information contained within the
3373 "``landingpad``" instruction can't be lost through normal code motion.
3378 This instruction requires several arguments:
3380 #. The optional "cconv" marker indicates which :ref:`calling
3381 convention <callingconv>` the call should use. If none is
3382 specified, the call defaults to using C calling conventions.
3383 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3384 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3386 #. '``ptr to function ty``': shall be the signature of the pointer to
3387 function value being invoked. In most cases, this is a direct
3388 function invocation, but indirect ``invoke``'s are just as possible,
3389 branching off an arbitrary pointer to function value.
3390 #. '``function ptr val``': An LLVM value containing a pointer to a
3391 function to be invoked.
3392 #. '``function args``': argument list whose types match the function
3393 signature argument types and parameter attributes. All arguments must
3394 be of :ref:`first class <t_firstclass>` type. If the function signature
3395 indicates the function accepts a variable number of arguments, the
3396 extra arguments can be specified.
3397 #. '``normal label``': the label reached when the called function
3398 executes a '``ret``' instruction.
3399 #. '``exception label``': the label reached when a callee returns via
3400 the :ref:`resume <i_resume>` instruction or other exception handling
3402 #. The optional :ref:`function attributes <fnattrs>` list. Only
3403 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3404 attributes are valid here.
3409 This instruction is designed to operate as a standard '``call``'
3410 instruction in most regards. The primary difference is that it
3411 establishes an association with a label, which is used by the runtime
3412 library to unwind the stack.
3414 This instruction is used in languages with destructors to ensure that
3415 proper cleanup is performed in the case of either a ``longjmp`` or a
3416 thrown exception. Additionally, this is important for implementation of
3417 '``catch``' clauses in high-level languages that support them.
3419 For the purposes of the SSA form, the definition of the value returned
3420 by the '``invoke``' instruction is deemed to occur on the edge from the
3421 current block to the "normal" label. If the callee unwinds then no
3422 return value is available.
3427 .. code-block:: llvm
3429 %retval = invoke i32 @Test(i32 15) to label %Continue
3430 unwind label %TestCleanup ; {i32}:retval set
3431 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3432 unwind label %TestCleanup ; {i32}:retval set
3436 '``resume``' Instruction
3437 ^^^^^^^^^^^^^^^^^^^^^^^^
3444 resume <type> <value>
3449 The '``resume``' instruction is a terminator instruction that has no
3455 The '``resume``' instruction requires one argument, which must have the
3456 same type as the result of any '``landingpad``' instruction in the same
3462 The '``resume``' instruction resumes propagation of an existing
3463 (in-flight) exception whose unwinding was interrupted with a
3464 :ref:`landingpad <i_landingpad>` instruction.
3469 .. code-block:: llvm
3471 resume { i8*, i32 } %exn
3475 '``unreachable``' Instruction
3476 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3488 The '``unreachable``' instruction has no defined semantics. This
3489 instruction is used to inform the optimizer that a particular portion of
3490 the code is not reachable. This can be used to indicate that the code
3491 after a no-return function cannot be reached, and other facts.
3496 The '``unreachable``' instruction has no defined semantics.
3503 Binary operators are used to do most of the computation in a program.
3504 They require two operands of the same type, execute an operation on
3505 them, and produce a single value. The operands might represent multiple
3506 data, as is the case with the :ref:`vector <t_vector>` data type. The
3507 result value has the same type as its operands.
3509 There are several different binary operators:
3513 '``add``' Instruction
3514 ^^^^^^^^^^^^^^^^^^^^^
3521 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3522 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3523 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3524 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3529 The '``add``' instruction returns the sum of its two operands.
3534 The two arguments to the '``add``' instruction must be
3535 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3536 arguments must have identical types.
3541 The value produced is the integer sum of the two operands.
3543 If the sum has unsigned overflow, the result returned is the
3544 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3547 Because LLVM integers use a two's complement representation, this
3548 instruction is appropriate for both signed and unsigned integers.
3550 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3551 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3552 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3553 unsigned and/or signed overflow, respectively, occurs.
3558 .. code-block:: llvm
3560 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3564 '``fadd``' Instruction
3565 ^^^^^^^^^^^^^^^^^^^^^^
3572 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3577 The '``fadd``' instruction returns the sum of its two operands.
3582 The two arguments to the '``fadd``' instruction must be :ref:`floating
3583 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3584 Both arguments must have identical types.
3589 The value produced is the floating point sum of the two operands. This
3590 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3591 which are optimization hints to enable otherwise unsafe floating point
3597 .. code-block:: llvm
3599 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3601 '``sub``' Instruction
3602 ^^^^^^^^^^^^^^^^^^^^^
3609 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3610 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3611 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3612 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3617 The '``sub``' instruction returns the difference of its two operands.
3619 Note that the '``sub``' instruction is used to represent the '``neg``'
3620 instruction present in most other intermediate representations.
3625 The two arguments to the '``sub``' instruction must be
3626 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3627 arguments must have identical types.
3632 The value produced is the integer difference of the two operands.
3634 If the difference has unsigned overflow, the result returned is the
3635 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3638 Because LLVM integers use a two's complement representation, this
3639 instruction is appropriate for both signed and unsigned integers.
3641 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3642 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3643 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3644 unsigned and/or signed overflow, respectively, occurs.
3649 .. code-block:: llvm
3651 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3652 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3656 '``fsub``' Instruction
3657 ^^^^^^^^^^^^^^^^^^^^^^
3664 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3669 The '``fsub``' instruction returns the difference of its two operands.
3671 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3672 instruction present in most other intermediate representations.
3677 The two arguments to the '``fsub``' instruction must be :ref:`floating
3678 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3679 Both arguments must have identical types.
3684 The value produced is the floating point difference of the two operands.
3685 This instruction can also take any number of :ref:`fast-math
3686 flags <fastmath>`, which are optimization hints to enable otherwise
3687 unsafe floating point optimizations:
3692 .. code-block:: llvm
3694 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3695 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3697 '``mul``' Instruction
3698 ^^^^^^^^^^^^^^^^^^^^^
3705 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3706 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3707 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3708 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3713 The '``mul``' instruction returns the product of its two operands.
3718 The two arguments to the '``mul``' instruction must be
3719 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3720 arguments must have identical types.
3725 The value produced is the integer product of the two operands.
3727 If the result of the multiplication has unsigned overflow, the result
3728 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3729 bit width of the result.
3731 Because LLVM integers use a two's complement representation, and the
3732 result is the same width as the operands, this instruction returns the
3733 correct result for both signed and unsigned integers. If a full product
3734 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3735 sign-extended or zero-extended as appropriate to the width of the full
3738 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3739 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3740 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3741 unsigned and/or signed overflow, respectively, occurs.
3746 .. code-block:: llvm
3748 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3752 '``fmul``' Instruction
3753 ^^^^^^^^^^^^^^^^^^^^^^
3760 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3765 The '``fmul``' instruction returns the product of its two operands.
3770 The two arguments to the '``fmul``' instruction must be :ref:`floating
3771 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3772 Both arguments must have identical types.
3777 The value produced is the floating point product of the two operands.
3778 This instruction can also take any number of :ref:`fast-math
3779 flags <fastmath>`, which are optimization hints to enable otherwise
3780 unsafe floating point optimizations:
3785 .. code-block:: llvm
3787 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3789 '``udiv``' Instruction
3790 ^^^^^^^^^^^^^^^^^^^^^^
3797 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3798 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3803 The '``udiv``' instruction returns the quotient of its two operands.
3808 The two arguments to the '``udiv``' instruction must be
3809 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3810 arguments must have identical types.
3815 The value produced is the unsigned integer quotient of the two operands.
3817 Note that unsigned integer division and signed integer division are
3818 distinct operations; for signed integer division, use '``sdiv``'.
3820 Division by zero leads to undefined behavior.
3822 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3823 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3824 such, "((a udiv exact b) mul b) == a").
3829 .. code-block:: llvm
3831 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3833 '``sdiv``' Instruction
3834 ^^^^^^^^^^^^^^^^^^^^^^
3841 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3842 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3847 The '``sdiv``' instruction returns the quotient of its two operands.
3852 The two arguments to the '``sdiv``' instruction must be
3853 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3854 arguments must have identical types.
3859 The value produced is the signed integer quotient of the two operands
3860 rounded towards zero.
3862 Note that signed integer division and unsigned integer division are
3863 distinct operations; for unsigned integer division, use '``udiv``'.
3865 Division by zero leads to undefined behavior. Overflow also leads to
3866 undefined behavior; this is a rare case, but can occur, for example, by
3867 doing a 32-bit division of -2147483648 by -1.
3869 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3870 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3875 .. code-block:: llvm
3877 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3881 '``fdiv``' Instruction
3882 ^^^^^^^^^^^^^^^^^^^^^^
3889 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3894 The '``fdiv``' instruction returns the quotient of its two operands.
3899 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3900 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3901 Both arguments must have identical types.
3906 The value produced is the floating point quotient of the two operands.
3907 This instruction can also take any number of :ref:`fast-math
3908 flags <fastmath>`, which are optimization hints to enable otherwise
3909 unsafe floating point optimizations:
3914 .. code-block:: llvm
3916 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3918 '``urem``' Instruction
3919 ^^^^^^^^^^^^^^^^^^^^^^
3926 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3931 The '``urem``' instruction returns the remainder from the unsigned
3932 division of its two arguments.
3937 The two arguments to the '``urem``' instruction must be
3938 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3939 arguments must have identical types.
3944 This instruction returns the unsigned integer *remainder* of a division.
3945 This instruction always performs an unsigned division to get the
3948 Note that unsigned integer remainder and signed integer remainder are
3949 distinct operations; for signed integer remainder, use '``srem``'.
3951 Taking the remainder of a division by zero leads to undefined behavior.
3956 .. code-block:: llvm
3958 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3960 '``srem``' Instruction
3961 ^^^^^^^^^^^^^^^^^^^^^^
3968 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3973 The '``srem``' instruction returns the remainder from the signed
3974 division of its two operands. This instruction can also take
3975 :ref:`vector <t_vector>` versions of the values in which case the elements
3981 The two arguments to the '``srem``' instruction must be
3982 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3983 arguments must have identical types.
3988 This instruction returns the *remainder* of a division (where the result
3989 is either zero or has the same sign as the dividend, ``op1``), not the
3990 *modulo* operator (where the result is either zero or has the same sign
3991 as the divisor, ``op2``) of a value. For more information about the
3992 difference, see `The Math
3993 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3994 table of how this is implemented in various languages, please see
3996 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3998 Note that signed integer remainder and unsigned integer remainder are
3999 distinct operations; for unsigned integer remainder, use '``urem``'.
4001 Taking the remainder of a division by zero leads to undefined behavior.
4002 Overflow also leads to undefined behavior; this is a rare case, but can
4003 occur, for example, by taking the remainder of a 32-bit division of
4004 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4005 rule lets srem be implemented using instructions that return both the
4006 result of the division and the remainder.)
4011 .. code-block:: llvm
4013 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4017 '``frem``' Instruction
4018 ^^^^^^^^^^^^^^^^^^^^^^
4025 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4030 The '``frem``' instruction returns the remainder from the division of
4036 The two arguments to the '``frem``' instruction must be :ref:`floating
4037 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4038 Both arguments must have identical types.
4043 This instruction returns the *remainder* of a division. The remainder
4044 has the same sign as the dividend. This instruction can also take any
4045 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4046 to enable otherwise unsafe floating point optimizations:
4051 .. code-block:: llvm
4053 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4057 Bitwise Binary Operations
4058 -------------------------
4060 Bitwise binary operators are used to do various forms of bit-twiddling
4061 in a program. They are generally very efficient instructions and can
4062 commonly be strength reduced from other instructions. They require two
4063 operands of the same type, execute an operation on them, and produce a
4064 single value. The resulting value is the same type as its operands.
4066 '``shl``' Instruction
4067 ^^^^^^^^^^^^^^^^^^^^^
4074 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4075 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4076 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4077 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4082 The '``shl``' instruction returns the first operand shifted to the left
4083 a specified number of bits.
4088 Both arguments to the '``shl``' instruction must be the same
4089 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4090 '``op2``' is treated as an unsigned value.
4095 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4096 where ``n`` is the width of the result. If ``op2`` is (statically or
4097 dynamically) negative or equal to or larger than the number of bits in
4098 ``op1``, the result is undefined. If the arguments are vectors, each
4099 vector element of ``op1`` is shifted by the corresponding shift amount
4102 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4103 value <poisonvalues>` if it shifts out any non-zero bits. If the
4104 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4105 value <poisonvalues>` if it shifts out any bits that disagree with the
4106 resultant sign bit. As such, NUW/NSW have the same semantics as they
4107 would if the shift were expressed as a mul instruction with the same
4108 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4113 .. code-block:: llvm
4115 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4116 <result> = shl i32 4, 2 ; yields {i32}: 16
4117 <result> = shl i32 1, 10 ; yields {i32}: 1024
4118 <result> = shl i32 1, 32 ; undefined
4119 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4121 '``lshr``' Instruction
4122 ^^^^^^^^^^^^^^^^^^^^^^
4129 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4130 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4135 The '``lshr``' instruction (logical shift right) returns the first
4136 operand shifted to the right a specified number of bits with zero fill.
4141 Both arguments to the '``lshr``' instruction must be the same
4142 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4143 '``op2``' is treated as an unsigned value.
4148 This instruction always performs a logical shift right operation. The
4149 most significant bits of the result will be filled with zero bits after
4150 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4151 than the number of bits in ``op1``, the result is undefined. If the
4152 arguments are vectors, each vector element of ``op1`` is shifted by the
4153 corresponding shift amount in ``op2``.
4155 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4156 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4162 .. code-block:: llvm
4164 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4165 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4166 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4167 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4168 <result> = lshr i32 1, 32 ; undefined
4169 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4171 '``ashr``' Instruction
4172 ^^^^^^^^^^^^^^^^^^^^^^
4179 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4180 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4185 The '``ashr``' instruction (arithmetic shift right) returns the first
4186 operand shifted to the right a specified number of bits with sign
4192 Both arguments to the '``ashr``' instruction must be the same
4193 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4194 '``op2``' is treated as an unsigned value.
4199 This instruction always performs an arithmetic shift right operation,
4200 The most significant bits of the result will be filled with the sign bit
4201 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4202 than the number of bits in ``op1``, the result is undefined. If the
4203 arguments are vectors, each vector element of ``op1`` is shifted by the
4204 corresponding shift amount in ``op2``.
4206 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4207 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4213 .. code-block:: llvm
4215 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4216 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4217 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4218 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4219 <result> = ashr i32 1, 32 ; undefined
4220 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4222 '``and``' Instruction
4223 ^^^^^^^^^^^^^^^^^^^^^
4230 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4235 The '``and``' instruction returns the bitwise logical and of its two
4241 The two arguments to the '``and``' instruction must be
4242 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4243 arguments must have identical types.
4248 The truth table used for the '``and``' instruction is:
4265 .. code-block:: llvm
4267 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4268 <result> = and i32 15, 40 ; yields {i32}:result = 8
4269 <result> = and i32 4, 8 ; yields {i32}:result = 0
4271 '``or``' Instruction
4272 ^^^^^^^^^^^^^^^^^^^^
4279 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4284 The '``or``' instruction returns the bitwise logical inclusive or of its
4290 The two arguments to the '``or``' instruction must be
4291 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4292 arguments must have identical types.
4297 The truth table used for the '``or``' instruction is:
4316 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4317 <result> = or i32 15, 40 ; yields {i32}:result = 47
4318 <result> = or i32 4, 8 ; yields {i32}:result = 12
4320 '``xor``' Instruction
4321 ^^^^^^^^^^^^^^^^^^^^^
4328 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4333 The '``xor``' instruction returns the bitwise logical exclusive or of
4334 its two operands. The ``xor`` is used to implement the "one's
4335 complement" operation, which is the "~" operator in C.
4340 The two arguments to the '``xor``' instruction must be
4341 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4342 arguments must have identical types.
4347 The truth table used for the '``xor``' instruction is:
4364 .. code-block:: llvm
4366 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4367 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4368 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4369 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4374 LLVM supports several instructions to represent vector operations in a
4375 target-independent manner. These instructions cover the element-access
4376 and vector-specific operations needed to process vectors effectively.
4377 While LLVM does directly support these vector operations, many
4378 sophisticated algorithms will want to use target-specific intrinsics to
4379 take full advantage of a specific target.
4381 .. _i_extractelement:
4383 '``extractelement``' Instruction
4384 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4391 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4396 The '``extractelement``' instruction extracts a single scalar element
4397 from a vector at a specified index.
4402 The first operand of an '``extractelement``' instruction is a value of
4403 :ref:`vector <t_vector>` type. The second operand is an index indicating
4404 the position from which to extract the element. The index may be a
4410 The result is a scalar of the same type as the element type of ``val``.
4411 Its value is the value at position ``idx`` of ``val``. If ``idx``
4412 exceeds the length of ``val``, the results are undefined.
4417 .. code-block:: llvm
4419 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4421 .. _i_insertelement:
4423 '``insertelement``' Instruction
4424 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4431 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4436 The '``insertelement``' instruction inserts a scalar element into a
4437 vector at a specified index.
4442 The first operand of an '``insertelement``' instruction is a value of
4443 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4444 type must equal the element type of the first operand. The third operand
4445 is an index indicating the position at which to insert the value. The
4446 index may be a variable.
4451 The result is a vector of the same type as ``val``. Its element values
4452 are those of ``val`` except at position ``idx``, where it gets the value
4453 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4459 .. code-block:: llvm
4461 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4463 .. _i_shufflevector:
4465 '``shufflevector``' Instruction
4466 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4473 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4478 The '``shufflevector``' instruction constructs a permutation of elements
4479 from two input vectors, returning a vector with the same element type as
4480 the input and length that is the same as the shuffle mask.
4485 The first two operands of a '``shufflevector``' instruction are vectors
4486 with the same type. The third argument is a shuffle mask whose element
4487 type is always 'i32'. The result of the instruction is a vector whose
4488 length is the same as the shuffle mask and whose element type is the
4489 same as the element type of the first two operands.
4491 The shuffle mask operand is required to be a constant vector with either
4492 constant integer or undef values.
4497 The elements of the two input vectors are numbered from left to right
4498 across both of the vectors. The shuffle mask operand specifies, for each
4499 element of the result vector, which element of the two input vectors the
4500 result element gets. The element selector may be undef (meaning "don't
4501 care") and the second operand may be undef if performing a shuffle from
4507 .. code-block:: llvm
4509 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4510 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4511 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4512 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4513 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4514 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4515 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4516 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4518 Aggregate Operations
4519 --------------------
4521 LLVM supports several instructions for working with
4522 :ref:`aggregate <t_aggregate>` values.
4526 '``extractvalue``' Instruction
4527 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4534 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4539 The '``extractvalue``' instruction extracts the value of a member field
4540 from an :ref:`aggregate <t_aggregate>` value.
4545 The first operand of an '``extractvalue``' instruction is a value of
4546 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4547 constant indices to specify which value to extract in a similar manner
4548 as indices in a '``getelementptr``' instruction.
4550 The major differences to ``getelementptr`` indexing are:
4552 - Since the value being indexed is not a pointer, the first index is
4553 omitted and assumed to be zero.
4554 - At least one index must be specified.
4555 - Not only struct indices but also array indices must be in bounds.
4560 The result is the value at the position in the aggregate specified by
4566 .. code-block:: llvm
4568 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4572 '``insertvalue``' Instruction
4573 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4580 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4585 The '``insertvalue``' instruction inserts a value into a member field in
4586 an :ref:`aggregate <t_aggregate>` value.
4591 The first operand of an '``insertvalue``' instruction is a value of
4592 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4593 a first-class value to insert. The following operands are constant
4594 indices indicating the position at which to insert the value in a
4595 similar manner as indices in a '``extractvalue``' instruction. The value
4596 to insert must have the same type as the value identified by the
4602 The result is an aggregate of the same type as ``val``. Its value is
4603 that of ``val`` except that the value at the position specified by the
4604 indices is that of ``elt``.
4609 .. code-block:: llvm
4611 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4612 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4613 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4617 Memory Access and Addressing Operations
4618 ---------------------------------------
4620 A key design point of an SSA-based representation is how it represents
4621 memory. In LLVM, no memory locations are in SSA form, which makes things
4622 very simple. This section describes how to read, write, and allocate
4627 '``alloca``' Instruction
4628 ^^^^^^^^^^^^^^^^^^^^^^^^
4635 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4640 The '``alloca``' instruction allocates memory on the stack frame of the
4641 currently executing function, to be automatically released when this
4642 function returns to its caller. The object is always allocated in the
4643 generic address space (address space zero).
4648 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4649 bytes of memory on the runtime stack, returning a pointer of the
4650 appropriate type to the program. If "NumElements" is specified, it is
4651 the number of elements allocated, otherwise "NumElements" is defaulted
4652 to be one. If a constant alignment is specified, the value result of the
4653 allocation is guaranteed to be aligned to at least that boundary. If not
4654 specified, or if zero, the target can choose to align the allocation on
4655 any convenient boundary compatible with the type.
4657 '``type``' may be any sized type.
4662 Memory is allocated; a pointer is returned. The operation is undefined
4663 if there is insufficient stack space for the allocation. '``alloca``'d
4664 memory is automatically released when the function returns. The
4665 '``alloca``' instruction is commonly used to represent automatic
4666 variables that must have an address available. When the function returns
4667 (either with the ``ret`` or ``resume`` instructions), the memory is
4668 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4669 The order in which memory is allocated (ie., which way the stack grows)
4675 .. code-block:: llvm
4677 %ptr = alloca i32 ; yields {i32*}:ptr
4678 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4679 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4680 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4684 '``load``' Instruction
4685 ^^^^^^^^^^^^^^^^^^^^^^
4692 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4693 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4694 !<index> = !{ i32 1 }
4699 The '``load``' instruction is used to read from memory.
4704 The argument to the ``load`` instruction specifies the memory address
4705 from which to load. The pointer must point to a :ref:`first
4706 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4707 then the optimizer is not allowed to modify the number or order of
4708 execution of this ``load`` with other :ref:`volatile
4709 operations <volatile>`.
4711 If the ``load`` is marked as ``atomic``, it takes an extra
4712 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4713 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4714 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4715 when they may see multiple atomic stores. The type of the pointee must
4716 be an integer type whose bit width is a power of two greater than or
4717 equal to eight and less than or equal to a target-specific size limit.
4718 ``align`` must be explicitly specified on atomic loads, and the load has
4719 undefined behavior if the alignment is not set to a value which is at
4720 least the size in bytes of the pointee. ``!nontemporal`` does not have
4721 any defined semantics for atomic loads.
4723 The optional constant ``align`` argument specifies the alignment of the
4724 operation (that is, the alignment of the memory address). A value of 0
4725 or an omitted ``align`` argument means that the operation has the ABI
4726 alignment for the target. It is the responsibility of the code emitter
4727 to ensure that the alignment information is correct. Overestimating the
4728 alignment results in undefined behavior. Underestimating the alignment
4729 may produce less efficient code. An alignment of 1 is always safe.
4731 The optional ``!nontemporal`` metadata must reference a single
4732 metadata name ``<index>`` corresponding to a metadata node with one
4733 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4734 metadata on the instruction tells the optimizer and code generator
4735 that this load is not expected to be reused in the cache. The code
4736 generator may select special instructions to save cache bandwidth, such
4737 as the ``MOVNT`` instruction on x86.
4739 The optional ``!invariant.load`` metadata must reference a single
4740 metadata name ``<index>`` corresponding to a metadata node with no
4741 entries. The existence of the ``!invariant.load`` metadata on the
4742 instruction tells the optimizer and code generator that this load
4743 address points to memory which does not change value during program
4744 execution. The optimizer may then move this load around, for example, by
4745 hoisting it out of loops using loop invariant code motion.
4750 The location of memory pointed to is loaded. If the value being loaded
4751 is of scalar type then the number of bytes read does not exceed the
4752 minimum number of bytes needed to hold all bits of the type. For
4753 example, loading an ``i24`` reads at most three bytes. When loading a
4754 value of a type like ``i20`` with a size that is not an integral number
4755 of bytes, the result is undefined if the value was not originally
4756 written using a store of the same type.
4761 .. code-block:: llvm
4763 %ptr = alloca i32 ; yields {i32*}:ptr
4764 store i32 3, i32* %ptr ; yields {void}
4765 %val = load i32* %ptr ; yields {i32}:val = i32 3
4769 '``store``' Instruction
4770 ^^^^^^^^^^^^^^^^^^^^^^^
4777 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4778 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4783 The '``store``' instruction is used to write to memory.
4788 There are two arguments to the ``store`` instruction: a value to store
4789 and an address at which to store it. The type of the ``<pointer>``
4790 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4791 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4792 then the optimizer is not allowed to modify the number or order of
4793 execution of this ``store`` with other :ref:`volatile
4794 operations <volatile>`.
4796 If the ``store`` is marked as ``atomic``, it takes an extra
4797 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4798 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4799 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4800 when they may see multiple atomic stores. The type of the pointee must
4801 be an integer type whose bit width is a power of two greater than or
4802 equal to eight and less than or equal to a target-specific size limit.
4803 ``align`` must be explicitly specified on atomic stores, and the store
4804 has undefined behavior if the alignment is not set to a value which is
4805 at least the size in bytes of the pointee. ``!nontemporal`` does not
4806 have any defined semantics for atomic stores.
4808 The optional constant ``align`` argument specifies the alignment of the
4809 operation (that is, the alignment of the memory address). A value of 0
4810 or an omitted ``align`` argument means that the operation has the ABI
4811 alignment for the target. It is the responsibility of the code emitter
4812 to ensure that the alignment information is correct. Overestimating the
4813 alignment results in undefined behavior. Underestimating the
4814 alignment may produce less efficient code. An alignment of 1 is always
4817 The optional ``!nontemporal`` metadata must reference a single metadata
4818 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4819 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4820 tells the optimizer and code generator that this load is not expected to
4821 be reused in the cache. The code generator may select special
4822 instructions to save cache bandwidth, such as the MOVNT instruction on
4828 The contents of memory are updated to contain ``<value>`` at the
4829 location specified by the ``<pointer>`` operand. If ``<value>`` is
4830 of scalar type then the number of bytes written does not exceed the
4831 minimum number of bytes needed to hold all bits of the type. For
4832 example, storing an ``i24`` writes at most three bytes. When writing a
4833 value of a type like ``i20`` with a size that is not an integral number
4834 of bytes, it is unspecified what happens to the extra bits that do not
4835 belong to the type, but they will typically be overwritten.
4840 .. code-block:: llvm
4842 %ptr = alloca i32 ; yields {i32*}:ptr
4843 store i32 3, i32* %ptr ; yields {void}
4844 %val = load i32* %ptr ; yields {i32}:val = i32 3
4848 '``fence``' Instruction
4849 ^^^^^^^^^^^^^^^^^^^^^^^
4856 fence [singlethread] <ordering> ; yields {void}
4861 The '``fence``' instruction is used to introduce happens-before edges
4867 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4868 defines what *synchronizes-with* edges they add. They can only be given
4869 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4874 A fence A which has (at least) ``release`` ordering semantics
4875 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4876 semantics if and only if there exist atomic operations X and Y, both
4877 operating on some atomic object M, such that A is sequenced before X, X
4878 modifies M (either directly or through some side effect of a sequence
4879 headed by X), Y is sequenced before B, and Y observes M. This provides a
4880 *happens-before* dependency between A and B. Rather than an explicit
4881 ``fence``, one (but not both) of the atomic operations X or Y might
4882 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4883 still *synchronize-with* the explicit ``fence`` and establish the
4884 *happens-before* edge.
4886 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4887 ``acquire`` and ``release`` semantics specified above, participates in
4888 the global program order of other ``seq_cst`` operations and/or fences.
4890 The optional ":ref:`singlethread <singlethread>`" argument specifies
4891 that the fence only synchronizes with other fences in the same thread.
4892 (This is useful for interacting with signal handlers.)
4897 .. code-block:: llvm
4899 fence acquire ; yields {void}
4900 fence singlethread seq_cst ; yields {void}
4904 '``cmpxchg``' Instruction
4905 ^^^^^^^^^^^^^^^^^^^^^^^^^
4912 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4917 The '``cmpxchg``' instruction is used to atomically modify memory. It
4918 loads a value in memory and compares it to a given value. If they are
4919 equal, it stores a new value into the memory.
4924 There are three arguments to the '``cmpxchg``' instruction: an address
4925 to operate on, a value to compare to the value currently be at that
4926 address, and a new value to place at that address if the compared values
4927 are equal. The type of '<cmp>' must be an integer type whose bit width
4928 is a power of two greater than or equal to eight and less than or equal
4929 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4930 type, and the type of '<pointer>' must be a pointer to that type. If the
4931 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4932 to modify the number or order of execution of this ``cmpxchg`` with
4933 other :ref:`volatile operations <volatile>`.
4935 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4936 synchronizes with other atomic operations.
4938 The optional "``singlethread``" argument declares that the ``cmpxchg``
4939 is only atomic with respect to code (usually signal handlers) running in
4940 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4941 respect to all other code in the system.
4943 The pointer passed into cmpxchg must have alignment greater than or
4944 equal to the size in memory of the operand.
4949 The contents of memory at the location specified by the '``<pointer>``'
4950 operand is read and compared to '``<cmp>``'; if the read value is the
4951 equal, '``<new>``' is written. The original value at the location is
4954 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4955 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4956 atomic load with an ordering parameter determined by dropping any
4957 ``release`` part of the ``cmpxchg``'s ordering.
4962 .. code-block:: llvm
4965 %orig = atomic load i32* %ptr unordered ; yields {i32}
4969 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4970 %squared = mul i32 %cmp, %cmp
4971 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4972 %success = icmp eq i32 %cmp, %old
4973 br i1 %success, label %done, label %loop
4980 '``atomicrmw``' Instruction
4981 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4988 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4993 The '``atomicrmw``' instruction is used to atomically modify memory.
4998 There are three arguments to the '``atomicrmw``' instruction: an
4999 operation to apply, an address whose value to modify, an argument to the
5000 operation. The operation must be one of the following keywords:
5014 The type of '<value>' must be an integer type whose bit width is a power
5015 of two greater than or equal to eight and less than or equal to a
5016 target-specific size limit. The type of the '``<pointer>``' operand must
5017 be a pointer to that type. If the ``atomicrmw`` is marked as
5018 ``volatile``, then the optimizer is not allowed to modify the number or
5019 order of execution of this ``atomicrmw`` with other :ref:`volatile
5020 operations <volatile>`.
5025 The contents of memory at the location specified by the '``<pointer>``'
5026 operand are atomically read, modified, and written back. The original
5027 value at the location is returned. The modification is specified by the
5030 - xchg: ``*ptr = val``
5031 - add: ``*ptr = *ptr + val``
5032 - sub: ``*ptr = *ptr - val``
5033 - and: ``*ptr = *ptr & val``
5034 - nand: ``*ptr = ~(*ptr & val)``
5035 - or: ``*ptr = *ptr | val``
5036 - xor: ``*ptr = *ptr ^ val``
5037 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5038 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5039 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5041 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5047 .. code-block:: llvm
5049 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5051 .. _i_getelementptr:
5053 '``getelementptr``' Instruction
5054 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5061 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5062 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5063 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5068 The '``getelementptr``' instruction is used to get the address of a
5069 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5070 address calculation only and does not access memory.
5075 The first argument is always a pointer or a vector of pointers, and
5076 forms the basis of the calculation. The remaining arguments are indices
5077 that indicate which of the elements of the aggregate object are indexed.
5078 The interpretation of each index is dependent on the type being indexed
5079 into. The first index always indexes the pointer value given as the
5080 first argument, the second index indexes a value of the type pointed to
5081 (not necessarily the value directly pointed to, since the first index
5082 can be non-zero), etc. The first type indexed into must be a pointer
5083 value, subsequent types can be arrays, vectors, and structs. Note that
5084 subsequent types being indexed into can never be pointers, since that
5085 would require loading the pointer before continuing calculation.
5087 The type of each index argument depends on the type it is indexing into.
5088 When indexing into a (optionally packed) structure, only ``i32`` integer
5089 **constants** are allowed (when using a vector of indices they must all
5090 be the **same** ``i32`` integer constant). When indexing into an array,
5091 pointer or vector, integers of any width are allowed, and they are not
5092 required to be constant. These integers are treated as signed values
5095 For example, let's consider a C code fragment and how it gets compiled
5111 int *foo(struct ST *s) {
5112 return &s[1].Z.B[5][13];
5115 The LLVM code generated by Clang is:
5117 .. code-block:: llvm
5119 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5120 %struct.ST = type { i32, double, %struct.RT }
5122 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5124 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5131 In the example above, the first index is indexing into the
5132 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5133 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5134 indexes into the third element of the structure, yielding a
5135 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5136 structure. The third index indexes into the second element of the
5137 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5138 dimensions of the array are subscripted into, yielding an '``i32``'
5139 type. The '``getelementptr``' instruction returns a pointer to this
5140 element, thus computing a value of '``i32*``' type.
5142 Note that it is perfectly legal to index partially through a structure,
5143 returning a pointer to an inner element. Because of this, the LLVM code
5144 for the given testcase is equivalent to:
5146 .. code-block:: llvm
5148 define i32* @foo(%struct.ST* %s) {
5149 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5150 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5151 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5152 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5153 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5157 If the ``inbounds`` keyword is present, the result value of the
5158 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5159 pointer is not an *in bounds* address of an allocated object, or if any
5160 of the addresses that would be formed by successive addition of the
5161 offsets implied by the indices to the base address with infinitely
5162 precise signed arithmetic are not an *in bounds* address of that
5163 allocated object. The *in bounds* addresses for an allocated object are
5164 all the addresses that point into the object, plus the address one byte
5165 past the end. In cases where the base is a vector of pointers the
5166 ``inbounds`` keyword applies to each of the computations element-wise.
5168 If the ``inbounds`` keyword is not present, the offsets are added to the
5169 base address with silently-wrapping two's complement arithmetic. If the
5170 offsets have a different width from the pointer, they are sign-extended
5171 or truncated to the width of the pointer. The result value of the
5172 ``getelementptr`` may be outside the object pointed to by the base
5173 pointer. The result value may not necessarily be used to access memory
5174 though, even if it happens to point into allocated storage. See the
5175 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5178 The getelementptr instruction is often confusing. For some more insight
5179 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5184 .. code-block:: llvm
5186 ; yields [12 x i8]*:aptr
5187 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5189 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5191 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5193 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5195 In cases where the pointer argument is a vector of pointers, each index
5196 must be a vector with the same number of elements. For example:
5198 .. code-block:: llvm
5200 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5202 Conversion Operations
5203 ---------------------
5205 The instructions in this category are the conversion instructions
5206 (casting) which all take a single operand and a type. They perform
5207 various bit conversions on the operand.
5209 '``trunc .. to``' Instruction
5210 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5217 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5222 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5227 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5228 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5229 of the same number of integers. The bit size of the ``value`` must be
5230 larger than the bit size of the destination type, ``ty2``. Equal sized
5231 types are not allowed.
5236 The '``trunc``' instruction truncates the high order bits in ``value``
5237 and converts the remaining bits to ``ty2``. Since the source size must
5238 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5239 It will always truncate bits.
5244 .. code-block:: llvm
5246 %X = trunc i32 257 to i8 ; yields i8:1
5247 %Y = trunc i32 123 to i1 ; yields i1:true
5248 %Z = trunc i32 122 to i1 ; yields i1:false
5249 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5251 '``zext .. to``' Instruction
5252 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5259 <result> = zext <ty> <value> to <ty2> ; yields ty2
5264 The '``zext``' instruction zero extends its operand to type ``ty2``.
5269 The '``zext``' instruction takes a value to cast, and a type to cast it
5270 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5271 the same number of integers. The bit size of the ``value`` must be
5272 smaller than the bit size of the destination type, ``ty2``.
5277 The ``zext`` fills the high order bits of the ``value`` with zero bits
5278 until it reaches the size of the destination type, ``ty2``.
5280 When zero extending from i1, the result will always be either 0 or 1.
5285 .. code-block:: llvm
5287 %X = zext i32 257 to i64 ; yields i64:257
5288 %Y = zext i1 true to i32 ; yields i32:1
5289 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5291 '``sext .. to``' Instruction
5292 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5299 <result> = sext <ty> <value> to <ty2> ; yields ty2
5304 The '``sext``' sign extends ``value`` to the type ``ty2``.
5309 The '``sext``' instruction takes a value to cast, and a type to cast it
5310 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5311 the same number of integers. The bit size of the ``value`` must be
5312 smaller than the bit size of the destination type, ``ty2``.
5317 The '``sext``' instruction performs a sign extension by copying the sign
5318 bit (highest order bit) of the ``value`` until it reaches the bit size
5319 of the type ``ty2``.
5321 When sign extending from i1, the extension always results in -1 or 0.
5326 .. code-block:: llvm
5328 %X = sext i8 -1 to i16 ; yields i16 :65535
5329 %Y = sext i1 true to i32 ; yields i32:-1
5330 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5332 '``fptrunc .. to``' Instruction
5333 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5340 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5345 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5350 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5351 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5352 The size of ``value`` must be larger than the size of ``ty2``. This
5353 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5358 The '``fptrunc``' instruction truncates a ``value`` from a larger
5359 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5360 point <t_floating>` type. If the value cannot fit within the
5361 destination type, ``ty2``, then the results are undefined.
5366 .. code-block:: llvm
5368 %X = fptrunc double 123.0 to float ; yields float:123.0
5369 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5371 '``fpext .. to``' Instruction
5372 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5379 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5384 The '``fpext``' extends a floating point ``value`` to a larger floating
5390 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5391 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5392 to. The source type must be smaller than the destination type.
5397 The '``fpext``' instruction extends the ``value`` from a smaller
5398 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5399 point <t_floating>` type. The ``fpext`` cannot be used to make a
5400 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5401 *no-op cast* for a floating point cast.
5406 .. code-block:: llvm
5408 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5409 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5411 '``fptoui .. to``' Instruction
5412 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5419 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5424 The '``fptoui``' converts a floating point ``value`` to its unsigned
5425 integer equivalent of type ``ty2``.
5430 The '``fptoui``' instruction takes a value to cast, which must be a
5431 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5432 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5433 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5434 type with the same number of elements as ``ty``
5439 The '``fptoui``' instruction converts its :ref:`floating
5440 point <t_floating>` operand into the nearest (rounding towards zero)
5441 unsigned integer value. If the value cannot fit in ``ty2``, the results
5447 .. code-block:: llvm
5449 %X = fptoui double 123.0 to i32 ; yields i32:123
5450 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5451 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5453 '``fptosi .. to``' Instruction
5454 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5461 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5466 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5467 ``value`` to type ``ty2``.
5472 The '``fptosi``' instruction takes a value to cast, which must be a
5473 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5474 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5475 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5476 type with the same number of elements as ``ty``
5481 The '``fptosi``' instruction converts its :ref:`floating
5482 point <t_floating>` operand into the nearest (rounding towards zero)
5483 signed integer value. If the value cannot fit in ``ty2``, the results
5489 .. code-block:: llvm
5491 %X = fptosi double -123.0 to i32 ; yields i32:-123
5492 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5493 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5495 '``uitofp .. to``' Instruction
5496 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5503 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5508 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5509 and converts that value to the ``ty2`` type.
5514 The '``uitofp``' instruction takes a value to cast, which must be a
5515 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5516 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5517 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5518 type with the same number of elements as ``ty``
5523 The '``uitofp``' instruction interprets its operand as an unsigned
5524 integer quantity and converts it to the corresponding floating point
5525 value. If the value cannot fit in the floating point value, the results
5531 .. code-block:: llvm
5533 %X = uitofp i32 257 to float ; yields float:257.0
5534 %Y = uitofp i8 -1 to double ; yields double:255.0
5536 '``sitofp .. to``' Instruction
5537 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5544 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5549 The '``sitofp``' instruction regards ``value`` as a signed integer and
5550 converts that value to the ``ty2`` type.
5555 The '``sitofp``' instruction takes a value to cast, which must be a
5556 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5557 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5558 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5559 type with the same number of elements as ``ty``
5564 The '``sitofp``' instruction interprets its operand as a signed integer
5565 quantity and converts it to the corresponding floating point value. If
5566 the value cannot fit in the floating point value, the results are
5572 .. code-block:: llvm
5574 %X = sitofp i32 257 to float ; yields float:257.0
5575 %Y = sitofp i8 -1 to double ; yields double:-1.0
5579 '``ptrtoint .. to``' Instruction
5580 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5587 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5592 The '``ptrtoint``' instruction converts the pointer or a vector of
5593 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5598 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5599 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5600 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5601 a vector of integers type.
5606 The '``ptrtoint``' instruction converts ``value`` to integer type
5607 ``ty2`` by interpreting the pointer value as an integer and either
5608 truncating or zero extending that value to the size of the integer type.
5609 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5610 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5611 the same size, then nothing is done (*no-op cast*) other than a type
5617 .. code-block:: llvm
5619 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5620 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5621 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5625 '``inttoptr .. to``' Instruction
5626 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5633 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5638 The '``inttoptr``' instruction converts an integer ``value`` to a
5639 pointer type, ``ty2``.
5644 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5645 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5651 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5652 applying either a zero extension or a truncation depending on the size
5653 of the integer ``value``. If ``value`` is larger than the size of a
5654 pointer then a truncation is done. If ``value`` is smaller than the size
5655 of a pointer then a zero extension is done. If they are the same size,
5656 nothing is done (*no-op cast*).
5661 .. code-block:: llvm
5663 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5664 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5665 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5666 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5670 '``bitcast .. to``' Instruction
5671 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5678 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5683 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5689 The '``bitcast``' instruction takes a value to cast, which must be a
5690 non-aggregate first class value, and a type to cast it to, which must
5691 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5692 bit sizes of ``value`` and the destination type, ``ty2``, must be
5693 identical. If the source type is a pointer, the destination type must
5694 also be a pointer of the same size. This instruction supports bitwise
5695 conversion of vectors to integers and to vectors of other types (as
5696 long as they have the same size).
5701 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5702 is always a *no-op cast* because no bits change with this
5703 conversion. The conversion is done as if the ``value`` had been stored
5704 to memory and read back as type ``ty2``. Pointer (or vector of
5705 pointers) types may only be converted to other pointer (or vector of
5706 pointers) types with the same address space through this instruction.
5707 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5708 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5713 .. code-block:: llvm
5715 %X = bitcast i8 255 to i8 ; yields i8 :-1
5716 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5717 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5718 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5720 .. _i_addrspacecast:
5722 '``addrspacecast .. to``' Instruction
5723 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5730 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5735 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5736 address space ``n`` to type ``pty2`` in address space ``m``.
5741 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5742 to cast and a pointer type to cast it to, which must have a different
5748 The '``addrspacecast``' instruction converts the pointer value
5749 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5750 value modification, depending on the target and the address space
5751 pair. Pointer conversions within the same address space must be
5752 performed with the ``bitcast`` instruction. Note that if the address space
5753 conversion is legal then both result and operand refer to the same memory
5759 .. code-block:: llvm
5761 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5762 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5763 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5770 The instructions in this category are the "miscellaneous" instructions,
5771 which defy better classification.
5775 '``icmp``' Instruction
5776 ^^^^^^^^^^^^^^^^^^^^^^
5783 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5788 The '``icmp``' instruction returns a boolean value or a vector of
5789 boolean values based on comparison of its two integer, integer vector,
5790 pointer, or pointer vector operands.
5795 The '``icmp``' instruction takes three operands. The first operand is
5796 the condition code indicating the kind of comparison to perform. It is
5797 not a value, just a keyword. The possible condition code are:
5800 #. ``ne``: not equal
5801 #. ``ugt``: unsigned greater than
5802 #. ``uge``: unsigned greater or equal
5803 #. ``ult``: unsigned less than
5804 #. ``ule``: unsigned less or equal
5805 #. ``sgt``: signed greater than
5806 #. ``sge``: signed greater or equal
5807 #. ``slt``: signed less than
5808 #. ``sle``: signed less or equal
5810 The remaining two arguments must be :ref:`integer <t_integer>` or
5811 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5812 must also be identical types.
5817 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5818 code given as ``cond``. The comparison performed always yields either an
5819 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5821 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5822 otherwise. No sign interpretation is necessary or performed.
5823 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5824 otherwise. No sign interpretation is necessary or performed.
5825 #. ``ugt``: interprets the operands as unsigned values and yields
5826 ``true`` if ``op1`` is greater than ``op2``.
5827 #. ``uge``: interprets the operands as unsigned values and yields
5828 ``true`` if ``op1`` is greater than or equal to ``op2``.
5829 #. ``ult``: interprets the operands as unsigned values and yields
5830 ``true`` if ``op1`` is less than ``op2``.
5831 #. ``ule``: interprets the operands as unsigned values and yields
5832 ``true`` if ``op1`` is less than or equal to ``op2``.
5833 #. ``sgt``: interprets the operands as signed values and yields ``true``
5834 if ``op1`` is greater than ``op2``.
5835 #. ``sge``: interprets the operands as signed values and yields ``true``
5836 if ``op1`` is greater than or equal to ``op2``.
5837 #. ``slt``: interprets the operands as signed values and yields ``true``
5838 if ``op1`` is less than ``op2``.
5839 #. ``sle``: interprets the operands as signed values and yields ``true``
5840 if ``op1`` is less than or equal to ``op2``.
5842 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5843 are compared as if they were integers.
5845 If the operands are integer vectors, then they are compared element by
5846 element. The result is an ``i1`` vector with the same number of elements
5847 as the values being compared. Otherwise, the result is an ``i1``.
5852 .. code-block:: llvm
5854 <result> = icmp eq i32 4, 5 ; yields: result=false
5855 <result> = icmp ne float* %X, %X ; yields: result=false
5856 <result> = icmp ult i16 4, 5 ; yields: result=true
5857 <result> = icmp sgt i16 4, 5 ; yields: result=false
5858 <result> = icmp ule i16 -4, 5 ; yields: result=false
5859 <result> = icmp sge i16 4, 5 ; yields: result=false
5861 Note that the code generator does not yet support vector types with the
5862 ``icmp`` instruction.
5866 '``fcmp``' Instruction
5867 ^^^^^^^^^^^^^^^^^^^^^^
5874 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5879 The '``fcmp``' instruction returns a boolean value or vector of boolean
5880 values based on comparison of its operands.
5882 If the operands are floating point scalars, then the result type is a
5883 boolean (:ref:`i1 <t_integer>`).
5885 If the operands are floating point vectors, then the result type is a
5886 vector of boolean with the same number of elements as the operands being
5892 The '``fcmp``' instruction takes three operands. The first operand is
5893 the condition code indicating the kind of comparison to perform. It is
5894 not a value, just a keyword. The possible condition code are:
5896 #. ``false``: no comparison, always returns false
5897 #. ``oeq``: ordered and equal
5898 #. ``ogt``: ordered and greater than
5899 #. ``oge``: ordered and greater than or equal
5900 #. ``olt``: ordered and less than
5901 #. ``ole``: ordered and less than or equal
5902 #. ``one``: ordered and not equal
5903 #. ``ord``: ordered (no nans)
5904 #. ``ueq``: unordered or equal
5905 #. ``ugt``: unordered or greater than
5906 #. ``uge``: unordered or greater than or equal
5907 #. ``ult``: unordered or less than
5908 #. ``ule``: unordered or less than or equal
5909 #. ``une``: unordered or not equal
5910 #. ``uno``: unordered (either nans)
5911 #. ``true``: no comparison, always returns true
5913 *Ordered* means that neither operand is a QNAN while *unordered* means
5914 that either operand may be a QNAN.
5916 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5917 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5918 type. They must have identical types.
5923 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5924 condition code given as ``cond``. If the operands are vectors, then the
5925 vectors are compared element by element. Each comparison performed
5926 always yields an :ref:`i1 <t_integer>` result, as follows:
5928 #. ``false``: always yields ``false``, regardless of operands.
5929 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5930 is equal to ``op2``.
5931 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5932 is greater than ``op2``.
5933 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5934 is greater than or equal to ``op2``.
5935 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5936 is less than ``op2``.
5937 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5938 is less than or equal to ``op2``.
5939 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5940 is not equal to ``op2``.
5941 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5942 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5944 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5945 greater than ``op2``.
5946 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5947 greater than or equal to ``op2``.
5948 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5950 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5951 less than or equal to ``op2``.
5952 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5953 not equal to ``op2``.
5954 #. ``uno``: yields ``true`` if either operand is a QNAN.
5955 #. ``true``: always yields ``true``, regardless of operands.
5960 .. code-block:: llvm
5962 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5963 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5964 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5965 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5967 Note that the code generator does not yet support vector types with the
5968 ``fcmp`` instruction.
5972 '``phi``' Instruction
5973 ^^^^^^^^^^^^^^^^^^^^^
5980 <result> = phi <ty> [ <val0>, <label0>], ...
5985 The '``phi``' instruction is used to implement the φ node in the SSA
5986 graph representing the function.
5991 The type of the incoming values is specified with the first type field.
5992 After this, the '``phi``' instruction takes a list of pairs as
5993 arguments, with one pair for each predecessor basic block of the current
5994 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5995 the value arguments to the PHI node. Only labels may be used as the
5998 There must be no non-phi instructions between the start of a basic block
5999 and the PHI instructions: i.e. PHI instructions must be first in a basic
6002 For the purposes of the SSA form, the use of each incoming value is
6003 deemed to occur on the edge from the corresponding predecessor block to
6004 the current block (but after any definition of an '``invoke``'
6005 instruction's return value on the same edge).
6010 At runtime, the '``phi``' instruction logically takes on the value
6011 specified by the pair corresponding to the predecessor basic block that
6012 executed just prior to the current block.
6017 .. code-block:: llvm
6019 Loop: ; Infinite loop that counts from 0 on up...
6020 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6021 %nextindvar = add i32 %indvar, 1
6026 '``select``' Instruction
6027 ^^^^^^^^^^^^^^^^^^^^^^^^
6034 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6036 selty is either i1 or {<N x i1>}
6041 The '``select``' instruction is used to choose one value based on a
6042 condition, without branching.
6047 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6048 values indicating the condition, and two values of the same :ref:`first
6049 class <t_firstclass>` type. If the val1/val2 are vectors and the
6050 condition is a scalar, then entire vectors are selected, not individual
6056 If the condition is an i1 and it evaluates to 1, the instruction returns
6057 the first value argument; otherwise, it returns the second value
6060 If the condition is a vector of i1, then the value arguments must be
6061 vectors of the same size, and the selection is done element by element.
6066 .. code-block:: llvm
6068 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6072 '``call``' Instruction
6073 ^^^^^^^^^^^^^^^^^^^^^^
6080 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6085 The '``call``' instruction represents a simple function call.
6090 This instruction requires several arguments:
6092 #. The optional "tail" marker indicates that the callee function does
6093 not access any allocas or varargs in the caller. Note that calls may
6094 be marked "tail" even if they do not occur before a
6095 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6096 function call is eligible for tail call optimization, but `might not
6097 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6098 The code generator may optimize calls marked "tail" with either 1)
6099 automatic `sibling call
6100 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6101 callee have matching signatures, or 2) forced tail call optimization
6102 when the following extra requirements are met:
6104 - Caller and callee both have the calling convention ``fastcc``.
6105 - The call is in tail position (ret immediately follows call and ret
6106 uses value of call or is void).
6107 - Option ``-tailcallopt`` is enabled, or
6108 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6109 - `Platform specific constraints are
6110 met. <CodeGenerator.html#tailcallopt>`_
6112 #. The optional "cconv" marker indicates which :ref:`calling
6113 convention <callingconv>` the call should use. If none is
6114 specified, the call defaults to using C calling conventions. The
6115 calling convention of the call must match the calling convention of
6116 the target function, or else the behavior is undefined.
6117 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6118 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6120 #. '``ty``': the type of the call instruction itself which is also the
6121 type of the return value. Functions that return no value are marked
6123 #. '``fnty``': shall be the signature of the pointer to function value
6124 being invoked. The argument types must match the types implied by
6125 this signature. This type can be omitted if the function is not
6126 varargs and if the function type does not return a pointer to a
6128 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6129 be invoked. In most cases, this is a direct function invocation, but
6130 indirect ``call``'s are just as possible, calling an arbitrary pointer
6132 #. '``function args``': argument list whose types match the function
6133 signature argument types and parameter attributes. All arguments must
6134 be of :ref:`first class <t_firstclass>` type. If the function signature
6135 indicates the function accepts a variable number of arguments, the
6136 extra arguments can be specified.
6137 #. The optional :ref:`function attributes <fnattrs>` list. Only
6138 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6139 attributes are valid here.
6144 The '``call``' instruction is used to cause control flow to transfer to
6145 a specified function, with its incoming arguments bound to the specified
6146 values. Upon a '``ret``' instruction in the called function, control
6147 flow continues with the instruction after the function call, and the
6148 return value of the function is bound to the result argument.
6153 .. code-block:: llvm
6155 %retval = call i32 @test(i32 %argc)
6156 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6157 %X = tail call i32 @foo() ; yields i32
6158 %Y = tail call fastcc i32 @foo() ; yields i32
6159 call void %foo(i8 97 signext)
6161 %struct.A = type { i32, i8 }
6162 %r = call %struct.A @foo() ; yields { 32, i8 }
6163 %gr = extractvalue %struct.A %r, 0 ; yields i32
6164 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6165 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6166 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6168 llvm treats calls to some functions with names and arguments that match
6169 the standard C99 library as being the C99 library functions, and may
6170 perform optimizations or generate code for them under that assumption.
6171 This is something we'd like to change in the future to provide better
6172 support for freestanding environments and non-C-based languages.
6176 '``va_arg``' Instruction
6177 ^^^^^^^^^^^^^^^^^^^^^^^^
6184 <resultval> = va_arg <va_list*> <arglist>, <argty>
6189 The '``va_arg``' instruction is used to access arguments passed through
6190 the "variable argument" area of a function call. It is used to implement
6191 the ``va_arg`` macro in C.
6196 This instruction takes a ``va_list*`` value and the type of the
6197 argument. It returns a value of the specified argument type and
6198 increments the ``va_list`` to point to the next argument. The actual
6199 type of ``va_list`` is target specific.
6204 The '``va_arg``' instruction loads an argument of the specified type
6205 from the specified ``va_list`` and causes the ``va_list`` to point to
6206 the next argument. For more information, see the variable argument
6207 handling :ref:`Intrinsic Functions <int_varargs>`.
6209 It is legal for this instruction to be called in a function which does
6210 not take a variable number of arguments, for example, the ``vfprintf``
6213 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6214 function <intrinsics>` because it takes a type as an argument.
6219 See the :ref:`variable argument processing <int_varargs>` section.
6221 Note that the code generator does not yet fully support va\_arg on many
6222 targets. Also, it does not currently support va\_arg with aggregate
6223 types on any target.
6227 '``landingpad``' Instruction
6228 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6235 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6236 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6238 <clause> := catch <type> <value>
6239 <clause> := filter <array constant type> <array constant>
6244 The '``landingpad``' instruction is used by `LLVM's exception handling
6245 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6246 is a landing pad --- one where the exception lands, and corresponds to the
6247 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6248 defines values supplied by the personality function (``pers_fn``) upon
6249 re-entry to the function. The ``resultval`` has the type ``resultty``.
6254 This instruction takes a ``pers_fn`` value. This is the personality
6255 function associated with the unwinding mechanism. The optional
6256 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6258 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6259 contains the global variable representing the "type" that may be caught
6260 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6261 clause takes an array constant as its argument. Use
6262 "``[0 x i8**] undef``" for a filter which cannot throw. The
6263 '``landingpad``' instruction must contain *at least* one ``clause`` or
6264 the ``cleanup`` flag.
6269 The '``landingpad``' instruction defines the values which are set by the
6270 personality function (``pers_fn``) upon re-entry to the function, and
6271 therefore the "result type" of the ``landingpad`` instruction. As with
6272 calling conventions, how the personality function results are
6273 represented in LLVM IR is target specific.
6275 The clauses are applied in order from top to bottom. If two
6276 ``landingpad`` instructions are merged together through inlining, the
6277 clauses from the calling function are appended to the list of clauses.
6278 When the call stack is being unwound due to an exception being thrown,
6279 the exception is compared against each ``clause`` in turn. If it doesn't
6280 match any of the clauses, and the ``cleanup`` flag is not set, then
6281 unwinding continues further up the call stack.
6283 The ``landingpad`` instruction has several restrictions:
6285 - A landing pad block is a basic block which is the unwind destination
6286 of an '``invoke``' instruction.
6287 - A landing pad block must have a '``landingpad``' instruction as its
6288 first non-PHI instruction.
6289 - There can be only one '``landingpad``' instruction within the landing
6291 - A basic block that is not a landing pad block may not include a
6292 '``landingpad``' instruction.
6293 - All '``landingpad``' instructions in a function must have the same
6294 personality function.
6299 .. code-block:: llvm
6301 ;; A landing pad which can catch an integer.
6302 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6304 ;; A landing pad that is a cleanup.
6305 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6307 ;; A landing pad which can catch an integer and can only throw a double.
6308 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6310 filter [1 x i8**] [@_ZTId]
6317 LLVM supports the notion of an "intrinsic function". These functions
6318 have well known names and semantics and are required to follow certain
6319 restrictions. Overall, these intrinsics represent an extension mechanism
6320 for the LLVM language that does not require changing all of the
6321 transformations in LLVM when adding to the language (or the bitcode
6322 reader/writer, the parser, etc...).
6324 Intrinsic function names must all start with an "``llvm.``" prefix. This
6325 prefix is reserved in LLVM for intrinsic names; thus, function names may
6326 not begin with this prefix. Intrinsic functions must always be external
6327 functions: you cannot define the body of intrinsic functions. Intrinsic
6328 functions may only be used in call or invoke instructions: it is illegal
6329 to take the address of an intrinsic function. Additionally, because
6330 intrinsic functions are part of the LLVM language, it is required if any
6331 are added that they be documented here.
6333 Some intrinsic functions can be overloaded, i.e., the intrinsic
6334 represents a family of functions that perform the same operation but on
6335 different data types. Because LLVM can represent over 8 million
6336 different integer types, overloading is used commonly to allow an
6337 intrinsic function to operate on any integer type. One or more of the
6338 argument types or the result type can be overloaded to accept any
6339 integer type. Argument types may also be defined as exactly matching a
6340 previous argument's type or the result type. This allows an intrinsic
6341 function which accepts multiple arguments, but needs all of them to be
6342 of the same type, to only be overloaded with respect to a single
6343 argument or the result.
6345 Overloaded intrinsics will have the names of its overloaded argument
6346 types encoded into its function name, each preceded by a period. Only
6347 those types which are overloaded result in a name suffix. Arguments
6348 whose type is matched against another type do not. For example, the
6349 ``llvm.ctpop`` function can take an integer of any width and returns an
6350 integer of exactly the same integer width. This leads to a family of
6351 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6352 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6353 overloaded, and only one type suffix is required. Because the argument's
6354 type is matched against the return type, it does not require its own
6357 To learn how to add an intrinsic function, please see the `Extending
6358 LLVM Guide <ExtendingLLVM.html>`_.
6362 Variable Argument Handling Intrinsics
6363 -------------------------------------
6365 Variable argument support is defined in LLVM with the
6366 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6367 functions. These functions are related to the similarly named macros
6368 defined in the ``<stdarg.h>`` header file.
6370 All of these functions operate on arguments that use a target-specific
6371 value type "``va_list``". The LLVM assembly language reference manual
6372 does not define what this type is, so all transformations should be
6373 prepared to handle these functions regardless of the type used.
6375 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6376 variable argument handling intrinsic functions are used.
6378 .. code-block:: llvm
6380 define i32 @test(i32 %X, ...) {
6381 ; Initialize variable argument processing
6383 %ap2 = bitcast i8** %ap to i8*
6384 call void @llvm.va_start(i8* %ap2)
6386 ; Read a single integer argument
6387 %tmp = va_arg i8** %ap, i32
6389 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6391 %aq2 = bitcast i8** %aq to i8*
6392 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6393 call void @llvm.va_end(i8* %aq2)
6395 ; Stop processing of arguments.
6396 call void @llvm.va_end(i8* %ap2)
6400 declare void @llvm.va_start(i8*)
6401 declare void @llvm.va_copy(i8*, i8*)
6402 declare void @llvm.va_end(i8*)
6406 '``llvm.va_start``' Intrinsic
6407 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6414 declare void @llvm.va_start(i8* <arglist>)
6419 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6420 subsequent use by ``va_arg``.
6425 The argument is a pointer to a ``va_list`` element to initialize.
6430 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6431 available in C. In a target-dependent way, it initializes the
6432 ``va_list`` element to which the argument points, so that the next call
6433 to ``va_arg`` will produce the first variable argument passed to the
6434 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6435 to know the last argument of the function as the compiler can figure
6438 '``llvm.va_end``' Intrinsic
6439 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6446 declare void @llvm.va_end(i8* <arglist>)
6451 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6452 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6457 The argument is a pointer to a ``va_list`` to destroy.
6462 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6463 available in C. In a target-dependent way, it destroys the ``va_list``
6464 element to which the argument points. Calls to
6465 :ref:`llvm.va_start <int_va_start>` and
6466 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6471 '``llvm.va_copy``' Intrinsic
6472 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6479 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6484 The '``llvm.va_copy``' intrinsic copies the current argument position
6485 from the source argument list to the destination argument list.
6490 The first argument is a pointer to a ``va_list`` element to initialize.
6491 The second argument is a pointer to a ``va_list`` element to copy from.
6496 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6497 available in C. In a target-dependent way, it copies the source
6498 ``va_list`` element into the destination ``va_list`` element. This
6499 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6500 arbitrarily complex and require, for example, memory allocation.
6502 Accurate Garbage Collection Intrinsics
6503 --------------------------------------
6505 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6506 (GC) requires the implementation and generation of these intrinsics.
6507 These intrinsics allow identification of :ref:`GC roots on the
6508 stack <int_gcroot>`, as well as garbage collector implementations that
6509 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6510 Front-ends for type-safe garbage collected languages should generate
6511 these intrinsics to make use of the LLVM garbage collectors. For more
6512 details, see `Accurate Garbage Collection with
6513 LLVM <GarbageCollection.html>`_.
6515 The garbage collection intrinsics only operate on objects in the generic
6516 address space (address space zero).
6520 '``llvm.gcroot``' Intrinsic
6521 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6528 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6533 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6534 the code generator, and allows some metadata to be associated with it.
6539 The first argument specifies the address of a stack object that contains
6540 the root pointer. The second pointer (which must be either a constant or
6541 a global value address) contains the meta-data to be associated with the
6547 At runtime, a call to this intrinsic stores a null pointer into the
6548 "ptrloc" location. At compile-time, the code generator generates
6549 information to allow the runtime to find the pointer at GC safe points.
6550 The '``llvm.gcroot``' intrinsic may only be used in a function which
6551 :ref:`specifies a GC algorithm <gc>`.
6555 '``llvm.gcread``' Intrinsic
6556 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6563 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6568 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6569 locations, allowing garbage collector implementations that require read
6575 The second argument is the address to read from, which should be an
6576 address allocated from the garbage collector. The first object is a
6577 pointer to the start of the referenced object, if needed by the language
6578 runtime (otherwise null).
6583 The '``llvm.gcread``' intrinsic has the same semantics as a load
6584 instruction, but may be replaced with substantially more complex code by
6585 the garbage collector runtime, as needed. The '``llvm.gcread``'
6586 intrinsic may only be used in a function which :ref:`specifies a GC
6591 '``llvm.gcwrite``' Intrinsic
6592 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6599 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6604 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6605 locations, allowing garbage collector implementations that require write
6606 barriers (such as generational or reference counting collectors).
6611 The first argument is the reference to store, the second is the start of
6612 the object to store it to, and the third is the address of the field of
6613 Obj to store to. If the runtime does not require a pointer to the
6614 object, Obj may be null.
6619 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6620 instruction, but may be replaced with substantially more complex code by
6621 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6622 intrinsic may only be used in a function which :ref:`specifies a GC
6625 Code Generator Intrinsics
6626 -------------------------
6628 These intrinsics are provided by LLVM to expose special features that
6629 may only be implemented with code generator support.
6631 '``llvm.returnaddress``' Intrinsic
6632 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6639 declare i8 *@llvm.returnaddress(i32 <level>)
6644 The '``llvm.returnaddress``' intrinsic attempts to compute a
6645 target-specific value indicating the return address of the current
6646 function or one of its callers.
6651 The argument to this intrinsic indicates which function to return the
6652 address for. Zero indicates the calling function, one indicates its
6653 caller, etc. The argument is **required** to be a constant integer
6659 The '``llvm.returnaddress``' intrinsic either returns a pointer
6660 indicating the return address of the specified call frame, or zero if it
6661 cannot be identified. The value returned by this intrinsic is likely to
6662 be incorrect or 0 for arguments other than zero, so it should only be
6663 used for debugging purposes.
6665 Note that calling this intrinsic does not prevent function inlining or
6666 other aggressive transformations, so the value returned may not be that
6667 of the obvious source-language caller.
6669 '``llvm.frameaddress``' Intrinsic
6670 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6677 declare i8* @llvm.frameaddress(i32 <level>)
6682 The '``llvm.frameaddress``' intrinsic attempts to return the
6683 target-specific frame pointer value for the specified stack frame.
6688 The argument to this intrinsic indicates which function to return the
6689 frame pointer for. Zero indicates the calling function, one indicates
6690 its caller, etc. The argument is **required** to be a constant integer
6696 The '``llvm.frameaddress``' intrinsic either returns a pointer
6697 indicating the frame address of the specified call frame, or zero if it
6698 cannot be identified. The value returned by this intrinsic is likely to
6699 be incorrect or 0 for arguments other than zero, so it should only be
6700 used for debugging purposes.
6702 Note that calling this intrinsic does not prevent function inlining or
6703 other aggressive transformations, so the value returned may not be that
6704 of the obvious source-language caller.
6708 '``llvm.stacksave``' Intrinsic
6709 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6716 declare i8* @llvm.stacksave()
6721 The '``llvm.stacksave``' intrinsic is used to remember the current state
6722 of the function stack, for use with
6723 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6724 implementing language features like scoped automatic variable sized
6730 This intrinsic returns a opaque pointer value that can be passed to
6731 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6732 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6733 ``llvm.stacksave``, it effectively restores the state of the stack to
6734 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6735 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6736 were allocated after the ``llvm.stacksave`` was executed.
6738 .. _int_stackrestore:
6740 '``llvm.stackrestore``' Intrinsic
6741 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6748 declare void @llvm.stackrestore(i8* %ptr)
6753 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6754 the function stack to the state it was in when the corresponding
6755 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6756 useful for implementing language features like scoped automatic variable
6757 sized arrays in C99.
6762 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6764 '``llvm.prefetch``' Intrinsic
6765 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6772 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6777 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6778 insert a prefetch instruction if supported; otherwise, it is a noop.
6779 Prefetches have no effect on the behavior of the program but can change
6780 its performance characteristics.
6785 ``address`` is the address to be prefetched, ``rw`` is the specifier
6786 determining if the fetch should be for a read (0) or write (1), and
6787 ``locality`` is a temporal locality specifier ranging from (0) - no
6788 locality, to (3) - extremely local keep in cache. The ``cache type``
6789 specifies whether the prefetch is performed on the data (1) or
6790 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6791 arguments must be constant integers.
6796 This intrinsic does not modify the behavior of the program. In
6797 particular, prefetches cannot trap and do not produce a value. On
6798 targets that support this intrinsic, the prefetch can provide hints to
6799 the processor cache for better performance.
6801 '``llvm.pcmarker``' Intrinsic
6802 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6809 declare void @llvm.pcmarker(i32 <id>)
6814 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6815 Counter (PC) in a region of code to simulators and other tools. The
6816 method is target specific, but it is expected that the marker will use
6817 exported symbols to transmit the PC of the marker. The marker makes no
6818 guarantees that it will remain with any specific instruction after
6819 optimizations. It is possible that the presence of a marker will inhibit
6820 optimizations. The intended use is to be inserted after optimizations to
6821 allow correlations of simulation runs.
6826 ``id`` is a numerical id identifying the marker.
6831 This intrinsic does not modify the behavior of the program. Backends
6832 that do not support this intrinsic may ignore it.
6834 '``llvm.readcyclecounter``' Intrinsic
6835 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6842 declare i64 @llvm.readcyclecounter()
6847 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6848 counter register (or similar low latency, high accuracy clocks) on those
6849 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6850 should map to RPCC. As the backing counters overflow quickly (on the
6851 order of 9 seconds on alpha), this should only be used for small
6857 When directly supported, reading the cycle counter should not modify any
6858 memory. Implementations are allowed to either return a application
6859 specific value or a system wide value. On backends without support, this
6860 is lowered to a constant 0.
6862 Note that runtime support may be conditional on the privilege-level code is
6863 running at and the host platform.
6865 Standard C Library Intrinsics
6866 -----------------------------
6868 LLVM provides intrinsics for a few important standard C library
6869 functions. These intrinsics allow source-language front-ends to pass
6870 information about the alignment of the pointer arguments to the code
6871 generator, providing opportunity for more efficient code generation.
6875 '``llvm.memcpy``' Intrinsic
6876 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6881 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6882 integer bit width and for different address spaces. Not all targets
6883 support all bit widths however.
6887 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6888 i32 <len>, i32 <align>, i1 <isvolatile>)
6889 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6890 i64 <len>, i32 <align>, i1 <isvolatile>)
6895 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6896 source location to the destination location.
6898 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6899 intrinsics do not return a value, takes extra alignment/isvolatile
6900 arguments and the pointers can be in specified address spaces.
6905 The first argument is a pointer to the destination, the second is a
6906 pointer to the source. The third argument is an integer argument
6907 specifying the number of bytes to copy, the fourth argument is the
6908 alignment of the source and destination locations, and the fifth is a
6909 boolean indicating a volatile access.
6911 If the call to this intrinsic has an alignment value that is not 0 or 1,
6912 then the caller guarantees that both the source and destination pointers
6913 are aligned to that boundary.
6915 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6916 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6917 very cleanly specified and it is unwise to depend on it.
6922 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6923 source location to the destination location, which are not allowed to
6924 overlap. It copies "len" bytes of memory over. If the argument is known
6925 to be aligned to some boundary, this can be specified as the fourth
6926 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
6928 '``llvm.memmove``' Intrinsic
6929 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6934 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6935 bit width and for different address space. Not all targets support all
6940 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6941 i32 <len>, i32 <align>, i1 <isvolatile>)
6942 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6943 i64 <len>, i32 <align>, i1 <isvolatile>)
6948 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6949 source location to the destination location. It is similar to the
6950 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6953 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6954 intrinsics do not return a value, takes extra alignment/isvolatile
6955 arguments and the pointers can be in specified address spaces.
6960 The first argument is a pointer to the destination, the second is a
6961 pointer to the source. The third argument is an integer argument
6962 specifying the number of bytes to copy, the fourth argument is the
6963 alignment of the source and destination locations, and the fifth is a
6964 boolean indicating a volatile access.
6966 If the call to this intrinsic has an alignment value that is not 0 or 1,
6967 then the caller guarantees that the source and destination pointers are
6968 aligned to that boundary.
6970 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6971 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6972 not very cleanly specified and it is unwise to depend on it.
6977 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6978 source location to the destination location, which may overlap. It
6979 copies "len" bytes of memory over. If the argument is known to be
6980 aligned to some boundary, this can be specified as the fourth argument,
6981 otherwise it should be set to 0 or 1 (both meaning no alignment).
6983 '``llvm.memset.*``' Intrinsics
6984 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6989 This is an overloaded intrinsic. You can use llvm.memset on any integer
6990 bit width and for different address spaces. However, not all targets
6991 support all bit widths.
6995 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6996 i32 <len>, i32 <align>, i1 <isvolatile>)
6997 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6998 i64 <len>, i32 <align>, i1 <isvolatile>)
7003 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7004 particular byte value.
7006 Note that, unlike the standard libc function, the ``llvm.memset``
7007 intrinsic does not return a value and takes extra alignment/volatile
7008 arguments. Also, the destination can be in an arbitrary address space.
7013 The first argument is a pointer to the destination to fill, the second
7014 is the byte value with which to fill it, the third argument is an
7015 integer argument specifying the number of bytes to fill, and the fourth
7016 argument is the known alignment of the destination location.
7018 If the call to this intrinsic has an alignment value that is not 0 or 1,
7019 then the caller guarantees that the destination pointer is aligned to
7022 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7023 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7024 very cleanly specified and it is unwise to depend on it.
7029 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7030 at the destination location. If the argument is known to be aligned to
7031 some boundary, this can be specified as the fourth argument, otherwise
7032 it should be set to 0 or 1 (both meaning no alignment).
7034 '``llvm.sqrt.*``' Intrinsic
7035 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7040 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7041 floating point or vector of floating point type. Not all targets support
7046 declare float @llvm.sqrt.f32(float %Val)
7047 declare double @llvm.sqrt.f64(double %Val)
7048 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7049 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7050 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7055 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7056 returning the same value as the libm '``sqrt``' functions would. Unlike
7057 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7058 negative numbers other than -0.0 (which allows for better optimization,
7059 because there is no need to worry about errno being set).
7060 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7065 The argument and return value are floating point numbers of the same
7071 This function returns the sqrt of the specified operand if it is a
7072 nonnegative floating point number.
7074 '``llvm.powi.*``' Intrinsic
7075 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7080 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7081 floating point or vector of floating point type. Not all targets support
7086 declare float @llvm.powi.f32(float %Val, i32 %power)
7087 declare double @llvm.powi.f64(double %Val, i32 %power)
7088 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7089 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7090 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7095 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7096 specified (positive or negative) power. The order of evaluation of
7097 multiplications is not defined. When a vector of floating point type is
7098 used, the second argument remains a scalar integer value.
7103 The second argument is an integer power, and the first is a value to
7104 raise to that power.
7109 This function returns the first value raised to the second power with an
7110 unspecified sequence of rounding operations.
7112 '``llvm.sin.*``' Intrinsic
7113 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7118 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7119 floating point or vector of floating point type. Not all targets support
7124 declare float @llvm.sin.f32(float %Val)
7125 declare double @llvm.sin.f64(double %Val)
7126 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7127 declare fp128 @llvm.sin.f128(fp128 %Val)
7128 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7133 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7138 The argument and return value are floating point numbers of the same
7144 This function returns the sine of the specified operand, returning the
7145 same values as the libm ``sin`` functions would, and handles error
7146 conditions in the same way.
7148 '``llvm.cos.*``' Intrinsic
7149 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7154 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7155 floating point or vector of floating point type. Not all targets support
7160 declare float @llvm.cos.f32(float %Val)
7161 declare double @llvm.cos.f64(double %Val)
7162 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7163 declare fp128 @llvm.cos.f128(fp128 %Val)
7164 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7169 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7174 The argument and return value are floating point numbers of the same
7180 This function returns the cosine of the specified operand, returning the
7181 same values as the libm ``cos`` functions would, and handles error
7182 conditions in the same way.
7184 '``llvm.pow.*``' Intrinsic
7185 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7190 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7191 floating point or vector of floating point type. Not all targets support
7196 declare float @llvm.pow.f32(float %Val, float %Power)
7197 declare double @llvm.pow.f64(double %Val, double %Power)
7198 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7199 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7200 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7205 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7206 specified (positive or negative) power.
7211 The second argument is a floating point power, and the first is a value
7212 to raise to that power.
7217 This function returns the first value raised to the second power,
7218 returning the same values as the libm ``pow`` functions would, and
7219 handles error conditions in the same way.
7221 '``llvm.exp.*``' Intrinsic
7222 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7227 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7228 floating point or vector of floating point type. Not all targets support
7233 declare float @llvm.exp.f32(float %Val)
7234 declare double @llvm.exp.f64(double %Val)
7235 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7236 declare fp128 @llvm.exp.f128(fp128 %Val)
7237 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7242 The '``llvm.exp.*``' intrinsics perform the exp function.
7247 The argument and return value are floating point numbers of the same
7253 This function returns the same values as the libm ``exp`` functions
7254 would, and handles error conditions in the same way.
7256 '``llvm.exp2.*``' Intrinsic
7257 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7262 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7263 floating point or vector of floating point type. Not all targets support
7268 declare float @llvm.exp2.f32(float %Val)
7269 declare double @llvm.exp2.f64(double %Val)
7270 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7271 declare fp128 @llvm.exp2.f128(fp128 %Val)
7272 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7277 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7282 The argument and return value are floating point numbers of the same
7288 This function returns the same values as the libm ``exp2`` functions
7289 would, and handles error conditions in the same way.
7291 '``llvm.log.*``' Intrinsic
7292 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7297 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7298 floating point or vector of floating point type. Not all targets support
7303 declare float @llvm.log.f32(float %Val)
7304 declare double @llvm.log.f64(double %Val)
7305 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7306 declare fp128 @llvm.log.f128(fp128 %Val)
7307 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7312 The '``llvm.log.*``' intrinsics perform the log function.
7317 The argument and return value are floating point numbers of the same
7323 This function returns the same values as the libm ``log`` functions
7324 would, and handles error conditions in the same way.
7326 '``llvm.log10.*``' Intrinsic
7327 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7332 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7333 floating point or vector of floating point type. Not all targets support
7338 declare float @llvm.log10.f32(float %Val)
7339 declare double @llvm.log10.f64(double %Val)
7340 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7341 declare fp128 @llvm.log10.f128(fp128 %Val)
7342 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7347 The '``llvm.log10.*``' intrinsics perform the log10 function.
7352 The argument and return value are floating point numbers of the same
7358 This function returns the same values as the libm ``log10`` functions
7359 would, and handles error conditions in the same way.
7361 '``llvm.log2.*``' Intrinsic
7362 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7367 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7368 floating point or vector of floating point type. Not all targets support
7373 declare float @llvm.log2.f32(float %Val)
7374 declare double @llvm.log2.f64(double %Val)
7375 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7376 declare fp128 @llvm.log2.f128(fp128 %Val)
7377 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7382 The '``llvm.log2.*``' intrinsics perform the log2 function.
7387 The argument and return value are floating point numbers of the same
7393 This function returns the same values as the libm ``log2`` functions
7394 would, and handles error conditions in the same way.
7396 '``llvm.fma.*``' Intrinsic
7397 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7402 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7403 floating point or vector of floating point type. Not all targets support
7408 declare float @llvm.fma.f32(float %a, float %b, float %c)
7409 declare double @llvm.fma.f64(double %a, double %b, double %c)
7410 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7411 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7412 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7417 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7423 The argument and return value are floating point numbers of the same
7429 This function returns the same values as the libm ``fma`` functions
7432 '``llvm.fabs.*``' Intrinsic
7433 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7438 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7439 floating point or vector of floating point type. Not all targets support
7444 declare float @llvm.fabs.f32(float %Val)
7445 declare double @llvm.fabs.f64(double %Val)
7446 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7447 declare fp128 @llvm.fabs.f128(fp128 %Val)
7448 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7453 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7459 The argument and return value are floating point numbers of the same
7465 This function returns the same values as the libm ``fabs`` functions
7466 would, and handles error conditions in the same way.
7468 '``llvm.copysign.*``' Intrinsic
7469 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7474 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7475 floating point or vector of floating point type. Not all targets support
7480 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7481 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7482 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7483 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7484 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7489 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7490 first operand and the sign of the second operand.
7495 The arguments and return value are floating point numbers of the same
7501 This function returns the same values as the libm ``copysign``
7502 functions would, and handles error conditions in the same way.
7504 '``llvm.floor.*``' Intrinsic
7505 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7510 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7511 floating point or vector of floating point type. Not all targets support
7516 declare float @llvm.floor.f32(float %Val)
7517 declare double @llvm.floor.f64(double %Val)
7518 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7519 declare fp128 @llvm.floor.f128(fp128 %Val)
7520 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7525 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7530 The argument and return value are floating point numbers of the same
7536 This function returns the same values as the libm ``floor`` functions
7537 would, and handles error conditions in the same way.
7539 '``llvm.ceil.*``' Intrinsic
7540 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7545 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7546 floating point or vector of floating point type. Not all targets support
7551 declare float @llvm.ceil.f32(float %Val)
7552 declare double @llvm.ceil.f64(double %Val)
7553 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7554 declare fp128 @llvm.ceil.f128(fp128 %Val)
7555 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7560 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7565 The argument and return value are floating point numbers of the same
7571 This function returns the same values as the libm ``ceil`` functions
7572 would, and handles error conditions in the same way.
7574 '``llvm.trunc.*``' Intrinsic
7575 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7580 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7581 floating point or vector of floating point type. Not all targets support
7586 declare float @llvm.trunc.f32(float %Val)
7587 declare double @llvm.trunc.f64(double %Val)
7588 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7589 declare fp128 @llvm.trunc.f128(fp128 %Val)
7590 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7595 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7596 nearest integer not larger in magnitude than the operand.
7601 The argument and return value are floating point numbers of the same
7607 This function returns the same values as the libm ``trunc`` functions
7608 would, and handles error conditions in the same way.
7610 '``llvm.rint.*``' Intrinsic
7611 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7616 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7617 floating point or vector of floating point type. Not all targets support
7622 declare float @llvm.rint.f32(float %Val)
7623 declare double @llvm.rint.f64(double %Val)
7624 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7625 declare fp128 @llvm.rint.f128(fp128 %Val)
7626 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7631 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7632 nearest integer. It may raise an inexact floating-point exception if the
7633 operand isn't an integer.
7638 The argument and return value are floating point numbers of the same
7644 This function returns the same values as the libm ``rint`` functions
7645 would, and handles error conditions in the same way.
7647 '``llvm.nearbyint.*``' Intrinsic
7648 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7653 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7654 floating point or vector of floating point type. Not all targets support
7659 declare float @llvm.nearbyint.f32(float %Val)
7660 declare double @llvm.nearbyint.f64(double %Val)
7661 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7662 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7663 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7668 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7674 The argument and return value are floating point numbers of the same
7680 This function returns the same values as the libm ``nearbyint``
7681 functions would, and handles error conditions in the same way.
7683 '``llvm.round.*``' Intrinsic
7684 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7689 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7690 floating point or vector of floating point type. Not all targets support
7695 declare float @llvm.round.f32(float %Val)
7696 declare double @llvm.round.f64(double %Val)
7697 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7698 declare fp128 @llvm.round.f128(fp128 %Val)
7699 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7704 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7710 The argument and return value are floating point numbers of the same
7716 This function returns the same values as the libm ``round``
7717 functions would, and handles error conditions in the same way.
7719 Bit Manipulation Intrinsics
7720 ---------------------------
7722 LLVM provides intrinsics for a few important bit manipulation
7723 operations. These allow efficient code generation for some algorithms.
7725 '``llvm.bswap.*``' Intrinsics
7726 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7731 This is an overloaded intrinsic function. You can use bswap on any
7732 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7736 declare i16 @llvm.bswap.i16(i16 <id>)
7737 declare i32 @llvm.bswap.i32(i32 <id>)
7738 declare i64 @llvm.bswap.i64(i64 <id>)
7743 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7744 values with an even number of bytes (positive multiple of 16 bits).
7745 These are useful for performing operations on data that is not in the
7746 target's native byte order.
7751 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7752 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7753 intrinsic returns an i32 value that has the four bytes of the input i32
7754 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7755 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7756 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7757 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7760 '``llvm.ctpop.*``' Intrinsic
7761 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7766 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7767 bit width, or on any vector with integer elements. Not all targets
7768 support all bit widths or vector types, however.
7772 declare i8 @llvm.ctpop.i8(i8 <src>)
7773 declare i16 @llvm.ctpop.i16(i16 <src>)
7774 declare i32 @llvm.ctpop.i32(i32 <src>)
7775 declare i64 @llvm.ctpop.i64(i64 <src>)
7776 declare i256 @llvm.ctpop.i256(i256 <src>)
7777 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7782 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7788 The only argument is the value to be counted. The argument may be of any
7789 integer type, or a vector with integer elements. The return type must
7790 match the argument type.
7795 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7796 each element of a vector.
7798 '``llvm.ctlz.*``' Intrinsic
7799 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7804 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7805 integer bit width, or any vector whose elements are integers. Not all
7806 targets support all bit widths or vector types, however.
7810 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7811 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7812 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7813 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7814 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7815 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7820 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7821 leading zeros in a variable.
7826 The first argument is the value to be counted. This argument may be of
7827 any integer type, or a vectory with integer element type. The return
7828 type must match the first argument type.
7830 The second argument must be a constant and is a flag to indicate whether
7831 the intrinsic should ensure that a zero as the first argument produces a
7832 defined result. Historically some architectures did not provide a
7833 defined result for zero values as efficiently, and many algorithms are
7834 now predicated on avoiding zero-value inputs.
7839 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7840 zeros in a variable, or within each element of the vector. If
7841 ``src == 0`` then the result is the size in bits of the type of ``src``
7842 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7843 ``llvm.ctlz(i32 2) = 30``.
7845 '``llvm.cttz.*``' Intrinsic
7846 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7851 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7852 integer bit width, or any vector of integer elements. Not all targets
7853 support all bit widths or vector types, however.
7857 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7858 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7859 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7860 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7861 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7862 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7867 The '``llvm.cttz``' family of intrinsic functions counts the number of
7873 The first argument is the value to be counted. This argument may be of
7874 any integer type, or a vectory with integer element type. The return
7875 type must match the first argument type.
7877 The second argument must be a constant and is a flag to indicate whether
7878 the intrinsic should ensure that a zero as the first argument produces a
7879 defined result. Historically some architectures did not provide a
7880 defined result for zero values as efficiently, and many algorithms are
7881 now predicated on avoiding zero-value inputs.
7886 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7887 zeros in a variable, or within each element of a vector. If ``src == 0``
7888 then the result is the size in bits of the type of ``src`` if
7889 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7890 ``llvm.cttz(2) = 1``.
7892 Arithmetic with Overflow Intrinsics
7893 -----------------------------------
7895 LLVM provides intrinsics for some arithmetic with overflow operations.
7897 '``llvm.sadd.with.overflow.*``' Intrinsics
7898 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7903 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7904 on any integer bit width.
7908 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7909 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7910 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7915 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7916 a signed addition of the two arguments, and indicate whether an overflow
7917 occurred during the signed summation.
7922 The arguments (%a and %b) and the first element of the result structure
7923 may be of integer types of any bit width, but they must have the same
7924 bit width. The second element of the result structure must be of type
7925 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7931 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7932 a signed addition of the two variables. They return a structure --- the
7933 first element of which is the signed summation, and the second element
7934 of which is a bit specifying if the signed summation resulted in an
7940 .. code-block:: llvm
7942 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7943 %sum = extractvalue {i32, i1} %res, 0
7944 %obit = extractvalue {i32, i1} %res, 1
7945 br i1 %obit, label %overflow, label %normal
7947 '``llvm.uadd.with.overflow.*``' Intrinsics
7948 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7953 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7954 on any integer bit width.
7958 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7959 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7960 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7965 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7966 an unsigned addition of the two arguments, and indicate whether a carry
7967 occurred during the unsigned summation.
7972 The arguments (%a and %b) and the first element of the result structure
7973 may be of integer types of any bit width, but they must have the same
7974 bit width. The second element of the result structure must be of type
7975 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7981 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7982 an unsigned addition of the two arguments. They return a structure --- the
7983 first element of which is the sum, and the second element of which is a
7984 bit specifying if the unsigned summation resulted in a carry.
7989 .. code-block:: llvm
7991 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7992 %sum = extractvalue {i32, i1} %res, 0
7993 %obit = extractvalue {i32, i1} %res, 1
7994 br i1 %obit, label %carry, label %normal
7996 '``llvm.ssub.with.overflow.*``' Intrinsics
7997 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8002 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8003 on any integer bit width.
8007 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8008 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8009 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8014 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8015 a signed subtraction of the two arguments, and indicate whether an
8016 overflow occurred during the signed subtraction.
8021 The arguments (%a and %b) and the first element of the result structure
8022 may be of integer types of any bit width, but they must have the same
8023 bit width. The second element of the result structure must be of type
8024 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8030 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8031 a signed subtraction of the two arguments. They return a structure --- the
8032 first element of which is the subtraction, and the second element of
8033 which is a bit specifying if the signed subtraction resulted in an
8039 .. code-block:: llvm
8041 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8042 %sum = extractvalue {i32, i1} %res, 0
8043 %obit = extractvalue {i32, i1} %res, 1
8044 br i1 %obit, label %overflow, label %normal
8046 '``llvm.usub.with.overflow.*``' Intrinsics
8047 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8052 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8053 on any integer bit width.
8057 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8058 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8059 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8064 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8065 an unsigned subtraction of the two arguments, and indicate whether an
8066 overflow occurred during the unsigned subtraction.
8071 The arguments (%a and %b) and the first element of the result structure
8072 may be of integer types of any bit width, but they must have the same
8073 bit width. The second element of the result structure must be of type
8074 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8080 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8081 an unsigned subtraction of the two arguments. They return a structure ---
8082 the first element of which is the subtraction, and the second element of
8083 which is a bit specifying if the unsigned subtraction resulted in an
8089 .. code-block:: llvm
8091 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8092 %sum = extractvalue {i32, i1} %res, 0
8093 %obit = extractvalue {i32, i1} %res, 1
8094 br i1 %obit, label %overflow, label %normal
8096 '``llvm.smul.with.overflow.*``' Intrinsics
8097 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8102 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8103 on any integer bit width.
8107 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8108 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8109 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8114 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8115 a signed multiplication of the two arguments, and indicate whether an
8116 overflow occurred during the signed multiplication.
8121 The arguments (%a and %b) and the first element of the result structure
8122 may be of integer types of any bit width, but they must have the same
8123 bit width. The second element of the result structure must be of type
8124 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8130 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8131 a signed multiplication of the two arguments. They return a structure ---
8132 the first element of which is the multiplication, and the second element
8133 of which is a bit specifying if the signed multiplication resulted in an
8139 .. code-block:: llvm
8141 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8142 %sum = extractvalue {i32, i1} %res, 0
8143 %obit = extractvalue {i32, i1} %res, 1
8144 br i1 %obit, label %overflow, label %normal
8146 '``llvm.umul.with.overflow.*``' Intrinsics
8147 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8152 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8153 on any integer bit width.
8157 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8158 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8159 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8164 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8165 a unsigned multiplication of the two arguments, and indicate whether an
8166 overflow occurred during the unsigned multiplication.
8171 The arguments (%a and %b) and the first element of the result structure
8172 may be of integer types of any bit width, but they must have the same
8173 bit width. The second element of the result structure must be of type
8174 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8180 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8181 an unsigned multiplication of the two arguments. They return a structure ---
8182 the first element of which is the multiplication, and the second
8183 element of which is a bit specifying if the unsigned multiplication
8184 resulted in an overflow.
8189 .. code-block:: llvm
8191 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8192 %sum = extractvalue {i32, i1} %res, 0
8193 %obit = extractvalue {i32, i1} %res, 1
8194 br i1 %obit, label %overflow, label %normal
8196 Specialised Arithmetic Intrinsics
8197 ---------------------------------
8199 '``llvm.fmuladd.*``' Intrinsic
8200 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8207 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8208 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8213 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8214 expressions that can be fused if the code generator determines that (a) the
8215 target instruction set has support for a fused operation, and (b) that the
8216 fused operation is more efficient than the equivalent, separate pair of mul
8217 and add instructions.
8222 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8223 multiplicands, a and b, and an addend c.
8232 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8234 is equivalent to the expression a \* b + c, except that rounding will
8235 not be performed between the multiplication and addition steps if the
8236 code generator fuses the operations. Fusion is not guaranteed, even if
8237 the target platform supports it. If a fused multiply-add is required the
8238 corresponding llvm.fma.\* intrinsic function should be used instead.
8243 .. code-block:: llvm
8245 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8247 Half Precision Floating Point Intrinsics
8248 ----------------------------------------
8250 For most target platforms, half precision floating point is a
8251 storage-only format. This means that it is a dense encoding (in memory)
8252 but does not support computation in the format.
8254 This means that code must first load the half-precision floating point
8255 value as an i16, then convert it to float with
8256 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8257 then be performed on the float value (including extending to double
8258 etc). To store the value back to memory, it is first converted to float
8259 if needed, then converted to i16 with
8260 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8263 .. _int_convert_to_fp16:
8265 '``llvm.convert.to.fp16``' Intrinsic
8266 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8273 declare i16 @llvm.convert.to.fp16(f32 %a)
8278 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8279 from single precision floating point format to half precision floating
8285 The intrinsic function contains single argument - the value to be
8291 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8292 from single precision floating point format to half precision floating
8293 point format. The return value is an ``i16`` which contains the
8299 .. code-block:: llvm
8301 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8302 store i16 %res, i16* @x, align 2
8304 .. _int_convert_from_fp16:
8306 '``llvm.convert.from.fp16``' Intrinsic
8307 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8314 declare f32 @llvm.convert.from.fp16(i16 %a)
8319 The '``llvm.convert.from.fp16``' intrinsic function performs a
8320 conversion from half precision floating point format to single precision
8321 floating point format.
8326 The intrinsic function contains single argument - the value to be
8332 The '``llvm.convert.from.fp16``' intrinsic function performs a
8333 conversion from half single precision floating point format to single
8334 precision floating point format. The input half-float value is
8335 represented by an ``i16`` value.
8340 .. code-block:: llvm
8342 %a = load i16* @x, align 2
8343 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8348 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8349 prefix), are described in the `LLVM Source Level
8350 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8353 Exception Handling Intrinsics
8354 -----------------------------
8356 The LLVM exception handling intrinsics (which all start with
8357 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8358 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8362 Trampoline Intrinsics
8363 ---------------------
8365 These intrinsics make it possible to excise one parameter, marked with
8366 the :ref:`nest <nest>` attribute, from a function. The result is a
8367 callable function pointer lacking the nest parameter - the caller does
8368 not need to provide a value for it. Instead, the value to use is stored
8369 in advance in a "trampoline", a block of memory usually allocated on the
8370 stack, which also contains code to splice the nest value into the
8371 argument list. This is used to implement the GCC nested function address
8374 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8375 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8376 It can be created as follows:
8378 .. code-block:: llvm
8380 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8381 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8382 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8383 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8384 %fp = bitcast i8* %p to i32 (i32, i32)*
8386 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8387 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8391 '``llvm.init.trampoline``' Intrinsic
8392 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8399 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8404 This fills the memory pointed to by ``tramp`` with executable code,
8405 turning it into a trampoline.
8410 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8411 pointers. The ``tramp`` argument must point to a sufficiently large and
8412 sufficiently aligned block of memory; this memory is written to by the
8413 intrinsic. Note that the size and the alignment are target-specific -
8414 LLVM currently provides no portable way of determining them, so a
8415 front-end that generates this intrinsic needs to have some
8416 target-specific knowledge. The ``func`` argument must hold a function
8417 bitcast to an ``i8*``.
8422 The block of memory pointed to by ``tramp`` is filled with target
8423 dependent code, turning it into a function. Then ``tramp`` needs to be
8424 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8425 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8426 function's signature is the same as that of ``func`` with any arguments
8427 marked with the ``nest`` attribute removed. At most one such ``nest``
8428 argument is allowed, and it must be of pointer type. Calling the new
8429 function is equivalent to calling ``func`` with the same argument list,
8430 but with ``nval`` used for the missing ``nest`` argument. If, after
8431 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8432 modified, then the effect of any later call to the returned function
8433 pointer is undefined.
8437 '``llvm.adjust.trampoline``' Intrinsic
8438 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8445 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8450 This performs any required machine-specific adjustment to the address of
8451 a trampoline (passed as ``tramp``).
8456 ``tramp`` must point to a block of memory which already has trampoline
8457 code filled in by a previous call to
8458 :ref:`llvm.init.trampoline <int_it>`.
8463 On some architectures the address of the code to be executed needs to be
8464 different to the address where the trampoline is actually stored. This
8465 intrinsic returns the executable address corresponding to ``tramp``
8466 after performing the required machine specific adjustments. The pointer
8467 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8472 This class of intrinsics exists to information about the lifetime of
8473 memory objects and ranges where variables are immutable.
8477 '``llvm.lifetime.start``' Intrinsic
8478 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8485 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8490 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8496 The first argument is a constant integer representing the size of the
8497 object, or -1 if it is variable sized. The second argument is a pointer
8503 This intrinsic indicates that before this point in the code, the value
8504 of the memory pointed to by ``ptr`` is dead. This means that it is known
8505 to never be used and has an undefined value. A load from the pointer
8506 that precedes this intrinsic can be replaced with ``'undef'``.
8510 '``llvm.lifetime.end``' Intrinsic
8511 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8518 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8523 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8529 The first argument is a constant integer representing the size of the
8530 object, or -1 if it is variable sized. The second argument is a pointer
8536 This intrinsic indicates that after this point in the code, the value of
8537 the memory pointed to by ``ptr`` is dead. This means that it is known to
8538 never be used and has an undefined value. Any stores into the memory
8539 object following this intrinsic may be removed as dead.
8541 '``llvm.invariant.start``' Intrinsic
8542 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8549 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8554 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8555 a memory object will not change.
8560 The first argument is a constant integer representing the size of the
8561 object, or -1 if it is variable sized. The second argument is a pointer
8567 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8568 the return value, the referenced memory location is constant and
8571 '``llvm.invariant.end``' Intrinsic
8572 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8579 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8584 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8585 memory object are mutable.
8590 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8591 The second argument is a constant integer representing the size of the
8592 object, or -1 if it is variable sized and the third argument is a
8593 pointer to the object.
8598 This intrinsic indicates that the memory is mutable again.
8603 This class of intrinsics is designed to be generic and has no specific
8606 '``llvm.var.annotation``' Intrinsic
8607 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8614 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8619 The '``llvm.var.annotation``' intrinsic.
8624 The first argument is a pointer to a value, the second is a pointer to a
8625 global string, the third is a pointer to a global string which is the
8626 source file name, and the last argument is the line number.
8631 This intrinsic allows annotation of local variables with arbitrary
8632 strings. This can be useful for special purpose optimizations that want
8633 to look for these annotations. These have no other defined use; they are
8634 ignored by code generation and optimization.
8636 '``llvm.ptr.annotation.*``' Intrinsic
8637 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8642 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8643 pointer to an integer of any width. *NOTE* you must specify an address space for
8644 the pointer. The identifier for the default address space is the integer
8649 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8650 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8651 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8652 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8653 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8658 The '``llvm.ptr.annotation``' intrinsic.
8663 The first argument is a pointer to an integer value of arbitrary bitwidth
8664 (result of some expression), the second is a pointer to a global string, the
8665 third is a pointer to a global string which is the source file name, and the
8666 last argument is the line number. It returns the value of the first argument.
8671 This intrinsic allows annotation of a pointer to an integer with arbitrary
8672 strings. This can be useful for special purpose optimizations that want to look
8673 for these annotations. These have no other defined use; they are ignored by code
8674 generation and optimization.
8676 '``llvm.annotation.*``' Intrinsic
8677 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8682 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8683 any integer bit width.
8687 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8688 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8689 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8690 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8691 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8696 The '``llvm.annotation``' intrinsic.
8701 The first argument is an integer value (result of some expression), the
8702 second is a pointer to a global string, the third is a pointer to a
8703 global string which is the source file name, and the last argument is
8704 the line number. It returns the value of the first argument.
8709 This intrinsic allows annotations to be put on arbitrary expressions
8710 with arbitrary strings. This can be useful for special purpose
8711 optimizations that want to look for these annotations. These have no
8712 other defined use; they are ignored by code generation and optimization.
8714 '``llvm.trap``' Intrinsic
8715 ^^^^^^^^^^^^^^^^^^^^^^^^^
8722 declare void @llvm.trap() noreturn nounwind
8727 The '``llvm.trap``' intrinsic.
8737 This intrinsic is lowered to the target dependent trap instruction. If
8738 the target does not have a trap instruction, this intrinsic will be
8739 lowered to a call of the ``abort()`` function.
8741 '``llvm.debugtrap``' Intrinsic
8742 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8749 declare void @llvm.debugtrap() nounwind
8754 The '``llvm.debugtrap``' intrinsic.
8764 This intrinsic is lowered to code which is intended to cause an
8765 execution trap with the intention of requesting the attention of a
8768 '``llvm.stackprotector``' Intrinsic
8769 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8776 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8781 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8782 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8783 is placed on the stack before local variables.
8788 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8789 The first argument is the value loaded from the stack guard
8790 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8791 enough space to hold the value of the guard.
8796 This intrinsic causes the prologue/epilogue inserter to force the position of
8797 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8798 to ensure that if a local variable on the stack is overwritten, it will destroy
8799 the value of the guard. When the function exits, the guard on the stack is
8800 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8801 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8802 calling the ``__stack_chk_fail()`` function.
8804 '``llvm.stackprotectorcheck``' Intrinsic
8805 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8812 declare void @llvm.stackprotectorcheck(i8** <guard>)
8817 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8818 created stack protector and if they are not equal calls the
8819 ``__stack_chk_fail()`` function.
8824 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8825 the variable ``@__stack_chk_guard``.
8830 This intrinsic is provided to perform the stack protector check by comparing
8831 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8832 values do not match call the ``__stack_chk_fail()`` function.
8834 The reason to provide this as an IR level intrinsic instead of implementing it
8835 via other IR operations is that in order to perform this operation at the IR
8836 level without an intrinsic, one would need to create additional basic blocks to
8837 handle the success/failure cases. This makes it difficult to stop the stack
8838 protector check from disrupting sibling tail calls in Codegen. With this
8839 intrinsic, we are able to generate the stack protector basic blocks late in
8840 codegen after the tail call decision has occurred.
8842 '``llvm.objectsize``' Intrinsic
8843 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8850 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8851 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8856 The ``llvm.objectsize`` intrinsic is designed to provide information to
8857 the optimizers to determine at compile time whether a) an operation
8858 (like memcpy) will overflow a buffer that corresponds to an object, or
8859 b) that a runtime check for overflow isn't necessary. An object in this
8860 context means an allocation of a specific class, structure, array, or
8866 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8867 argument is a pointer to or into the ``object``. The second argument is
8868 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8869 or -1 (if false) when the object size is unknown. The second argument
8870 only accepts constants.
8875 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8876 the size of the object concerned. If the size cannot be determined at
8877 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8878 on the ``min`` argument).
8880 '``llvm.expect``' Intrinsic
8881 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8888 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8889 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8894 The ``llvm.expect`` intrinsic provides information about expected (the
8895 most probable) value of ``val``, which can be used by optimizers.
8900 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8901 a value. The second argument is an expected value, this needs to be a
8902 constant value, variables are not allowed.
8907 This intrinsic is lowered to the ``val``.
8909 '``llvm.donothing``' Intrinsic
8910 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8917 declare void @llvm.donothing() nounwind readnone
8922 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8923 only intrinsic that can be called with an invoke instruction.
8933 This intrinsic does nothing, and it's removed by optimizers and ignored
8936 Stack Map Intrinsics
8937 --------------------
8939 LLVM provides experimental intrinsics to support runtime patching
8940 mechanisms commonly desired in dynamic language JITs. These intrinsics
8941 are described in :doc:`StackMaps`.