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 that 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. The ``"\01"`` prefix
83 can be used on global variables to suppress mangling.
84 #. Unnamed values are represented as an unsigned numeric value with
85 their prefix. For example, ``%12``, ``@2``, ``%44``.
86 #. Constants, which are described in the section Constants_ below.
88 LLVM requires that values start with a prefix for two reasons: Compilers
89 don't need to worry about name clashes with reserved words, and the set
90 of reserved words may be expanded in the future without penalty.
91 Additionally, unnamed identifiers allow a compiler to quickly come up
92 with a temporary variable without having to avoid symbol table
95 Reserved words in LLVM are very similar to reserved words in other
96 languages. There are keywords for different opcodes ('``add``',
97 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
98 '``i32``', etc...), and others. These reserved words cannot conflict
99 with variable names, because none of them start with a prefix character
100 (``'%'`` or ``'@'``).
102 Here is an example of LLVM code to multiply the integer variable
109 %result = mul i32 %X, 8
111 After strength reduction:
115 %result = shl i32 %X, 3
121 %0 = add i32 %X, %X ; yields i32:%0
122 %1 = add i32 %0, %0 ; yields i32:%1
123 %result = add i32 %1, %1
125 This last way of multiplying ``%X`` by 8 illustrates several important
126 lexical features of LLVM:
128 #. Comments are delimited with a '``;``' and go until the end of line.
129 #. Unnamed temporaries are created when the result of a computation is
130 not assigned to a named value.
131 #. Unnamed temporaries are numbered sequentially (using a per-function
132 incrementing counter, starting with 0). Note that basic blocks and unnamed
133 function parameters are included in this numbering. For example, if the
134 entry basic block is not given a label name and all function parameters are
135 named, then it will get number 0.
137 It also shows a convention that we follow in this document. When
138 demonstrating instructions, we will follow an instruction with a comment
139 that defines the type and name of value produced.
147 LLVM programs are composed of ``Module``'s, each of which is a
148 translation unit of the input programs. Each module consists of
149 functions, global variables, and symbol table entries. Modules may be
150 combined together with the LLVM linker, which merges function (and
151 global variable) definitions, resolves forward declarations, and merges
152 symbol table entries. Here is an example of the "hello world" module:
156 ; Declare the string constant as a global constant.
157 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
159 ; External declaration of the puts function
160 declare i32 @puts(i8* nocapture) nounwind
162 ; Definition of main function
163 define i32 @main() { ; i32()*
164 ; Convert [13 x i8]* to i8 *...
165 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
167 ; Call puts function to write out the string to stdout.
168 call i32 @puts(i8* %cast210)
173 !0 = metadata !{i32 42, null, metadata !"string"}
176 This example is made up of a :ref:`global variable <globalvars>` named
177 "``.str``", an external declaration of the "``puts``" function, a
178 :ref:`function definition <functionstructure>` for "``main``" and
179 :ref:`named metadata <namedmetadatastructure>` "``foo``".
181 In general, a module is made up of a list of global values (where both
182 functions and global variables are global values). Global values are
183 represented by a pointer to a memory location (in this case, a pointer
184 to an array of char, and a pointer to a function), and have one of the
185 following :ref:`linkage types <linkage>`.
192 All Global Variables and Functions have one of the following types of
196 Global values with "``private``" linkage are only directly
197 accessible by objects in the current module. In particular, linking
198 code into a module with an private global value may cause the
199 private to be renamed as necessary to avoid collisions. Because the
200 symbol is private to the module, all references can be updated. This
201 doesn't show up in any symbol table in the object file.
203 Similar to private, but the value shows as a local symbol
204 (``STB_LOCAL`` in the case of ELF) in the object file. This
205 corresponds to the notion of the '``static``' keyword in C.
206 ``available_externally``
207 Globals with "``available_externally``" linkage are never emitted
208 into the object file corresponding to the LLVM module. They exist to
209 allow inlining and other optimizations to take place given knowledge
210 of the definition of the global, which is known to be somewhere
211 outside the module. Globals with ``available_externally`` linkage
212 are allowed to be discarded at will, and are otherwise the same as
213 ``linkonce_odr``. This linkage type is only allowed on definitions,
216 Globals with "``linkonce``" linkage are merged with other globals of
217 the same name when linkage occurs. This can be used to implement
218 some forms of inline functions, templates, or other code which must
219 be generated in each translation unit that uses it, but where the
220 body may be overridden with a more definitive definition later.
221 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
222 that ``linkonce`` linkage does not actually allow the optimizer to
223 inline the body of this function into callers because it doesn't
224 know if this definition of the function is the definitive definition
225 within the program or whether it will be overridden by a stronger
226 definition. To enable inlining and other optimizations, use
227 "``linkonce_odr``" linkage.
229 "``weak``" linkage has the same merging semantics as ``linkonce``
230 linkage, except that unreferenced globals with ``weak`` linkage may
231 not be discarded. This is used for globals that are declared "weak"
234 "``common``" linkage is most similar to "``weak``" linkage, but they
235 are used for tentative definitions in C, such as "``int X;``" at
236 global scope. Symbols with "``common``" linkage are merged in the
237 same way as ``weak symbols``, and they may not be deleted if
238 unreferenced. ``common`` symbols may not have an explicit section,
239 must have a zero initializer, and may not be marked
240 ':ref:`constant <globalvars>`'. Functions and aliases may not have
243 .. _linkage_appending:
246 "``appending``" linkage may only be applied to global variables of
247 pointer to array type. When two global variables with appending
248 linkage are linked together, the two global arrays are appended
249 together. This is the LLVM, typesafe, equivalent of having the
250 system linker append together "sections" with identical names when
253 The semantics of this linkage follow the ELF object file model: the
254 symbol is weak until linked, if not linked, the symbol becomes null
255 instead of being an undefined reference.
256 ``linkonce_odr``, ``weak_odr``
257 Some languages allow differing globals to be merged, such as two
258 functions with different semantics. Other languages, such as
259 ``C++``, ensure that only equivalent globals are ever merged (the
260 "one definition rule" --- "ODR"). Such languages can use the
261 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
262 global will only be merged with equivalent globals. These linkage
263 types are otherwise the same as their non-``odr`` versions.
265 If none of the above identifiers are used, the global is externally
266 visible, meaning that it participates in linkage and can be used to
267 resolve external symbol references.
269 It is illegal for a function *declaration* to have any linkage type
270 other than ``external`` or ``extern_weak``.
277 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
278 :ref:`invokes <i_invoke>` can all have an optional calling convention
279 specified for the call. The calling convention of any pair of dynamic
280 caller/callee must match, or the behavior of the program is undefined.
281 The following calling conventions are supported by LLVM, and more may be
284 "``ccc``" - The C calling convention
285 This calling convention (the default if no other calling convention
286 is specified) matches the target C calling conventions. This calling
287 convention supports varargs function calls and tolerates some
288 mismatch in the declared prototype and implemented declaration of
289 the function (as does normal C).
290 "``fastcc``" - The fast calling convention
291 This calling convention attempts to make calls as fast as possible
292 (e.g. by passing things in registers). This calling convention
293 allows the target to use whatever tricks it wants to produce fast
294 code for the target, without having to conform to an externally
295 specified ABI (Application Binary Interface). `Tail calls can only
296 be optimized when this, the GHC or the HiPE convention is
297 used. <CodeGenerator.html#id80>`_ This calling convention does not
298 support varargs and requires the prototype of all callees to exactly
299 match the prototype of the function definition.
300 "``coldcc``" - The cold calling convention
301 This calling convention attempts to make code in the caller as
302 efficient as possible under the assumption that the call is not
303 commonly executed. As such, these calls often preserve all registers
304 so that the call does not break any live ranges in the caller side.
305 This calling convention does not support varargs and requires the
306 prototype of all callees to exactly match the prototype of the
307 function definition. Furthermore the inliner doesn't consider such function
309 "``cc 10``" - GHC convention
310 This calling convention has been implemented specifically for use by
311 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
312 It passes everything in registers, going to extremes to achieve this
313 by disabling callee save registers. This calling convention should
314 not be used lightly but only for specific situations such as an
315 alternative to the *register pinning* performance technique often
316 used when implementing functional programming languages. At the
317 moment only X86 supports this convention and it has the following
320 - On *X86-32* only supports up to 4 bit type parameters. No
321 floating point types are supported.
322 - On *X86-64* only supports up to 10 bit type parameters and 6
323 floating point parameters.
325 This calling convention supports `tail call
326 optimization <CodeGenerator.html#id80>`_ but requires both the
327 caller and callee are using it.
328 "``cc 11``" - The HiPE calling convention
329 This calling convention has been implemented specifically for use by
330 the `High-Performance Erlang
331 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
332 native code compiler of the `Ericsson's Open Source Erlang/OTP
333 system <http://www.erlang.org/download.shtml>`_. It uses more
334 registers for argument passing than the ordinary C calling
335 convention and defines no callee-saved registers. The calling
336 convention properly supports `tail call
337 optimization <CodeGenerator.html#id80>`_ but requires that both the
338 caller and the callee use it. It uses a *register pinning*
339 mechanism, similar to GHC's convention, for keeping frequently
340 accessed runtime components pinned to specific hardware registers.
341 At the moment only X86 supports this convention (both 32 and 64
343 "``webkit_jscc``" - WebKit's JavaScript calling convention
344 This calling convention has been implemented for `WebKit FTL JIT
345 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
346 stack right to left (as cdecl does), and returns a value in the
347 platform's customary return register.
348 "``anyregcc``" - Dynamic calling convention for code patching
349 This is a special convention that supports patching an arbitrary code
350 sequence in place of a call site. This convention forces the call
351 arguments into registers but allows them to be dynamcially
352 allocated. This can currently only be used with calls to
353 llvm.experimental.patchpoint because only this intrinsic records
354 the location of its arguments in a side table. See :doc:`StackMaps`.
355 "``preserve_mostcc``" - The `PreserveMost` calling convention
356 This calling convention attempts to make the code in the caller as little
357 intrusive as possible. This calling convention behaves identical to the `C`
358 calling convention on how arguments and return values are passed, but it
359 uses a different set of caller/callee-saved registers. This alleviates the
360 burden of saving and recovering a large register set before and after the
361 call in the caller. If the arguments are passed in callee-saved registers,
362 then they will be preserved by the callee across the call. This doesn't
363 apply for values returned in callee-saved registers.
365 - On X86-64 the callee preserves all general purpose registers, except for
366 R11. R11 can be used as a scratch register. Floating-point registers
367 (XMMs/YMMs) are not preserved and need to be saved by the caller.
369 The idea behind this convention is to support calls to runtime functions
370 that have a hot path and a cold path. The hot path is usually a small piece
371 of code that doesn't many registers. The cold path might need to call out to
372 another function and therefore only needs to preserve the caller-saved
373 registers, which haven't already been saved by the caller. The
374 `PreserveMost` calling convention is very similar to the `cold` calling
375 convention in terms of caller/callee-saved registers, but they are used for
376 different types of function calls. `coldcc` is for function calls that are
377 rarely executed, whereas `preserve_mostcc` function calls are intended to be
378 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
379 doesn't prevent the inliner from inlining the function call.
381 This calling convention will be used by a future version of the ObjectiveC
382 runtime and should therefore still be considered experimental at this time.
383 Although this convention was created to optimize certain runtime calls to
384 the ObjectiveC runtime, it is not limited to this runtime and might be used
385 by other runtimes in the future too. The current implementation only
386 supports X86-64, but the intention is to support more architectures in the
388 "``preserve_allcc``" - The `PreserveAll` calling convention
389 This calling convention attempts to make the code in the caller even less
390 intrusive than the `PreserveMost` calling convention. This calling
391 convention also behaves identical to the `C` calling convention on how
392 arguments and return values are passed, but it uses a different set of
393 caller/callee-saved registers. This removes the burden of saving and
394 recovering a large register set before and after the call in the caller. If
395 the arguments are passed in callee-saved registers, then they will be
396 preserved by the callee across the call. This doesn't apply for values
397 returned in callee-saved registers.
399 - On X86-64 the callee preserves all general purpose registers, except for
400 R11. R11 can be used as a scratch register. Furthermore it also preserves
401 all floating-point registers (XMMs/YMMs).
403 The idea behind this convention is to support calls to runtime functions
404 that don't need to call out to any other functions.
406 This calling convention, like the `PreserveMost` calling convention, will be
407 used by a future version of the ObjectiveC runtime and should be considered
408 experimental at this time.
409 "``cc <n>``" - Numbered convention
410 Any calling convention may be specified by number, allowing
411 target-specific calling conventions to be used. Target specific
412 calling conventions start at 64.
414 More calling conventions can be added/defined on an as-needed basis, to
415 support Pascal conventions or any other well-known target-independent
418 .. _visibilitystyles:
423 All Global Variables and Functions have one of the following visibility
426 "``default``" - Default style
427 On targets that use the ELF object file format, default visibility
428 means that the declaration is visible to other modules and, in
429 shared libraries, means that the declared entity may be overridden.
430 On Darwin, default visibility means that the declaration is visible
431 to other modules. Default visibility corresponds to "external
432 linkage" in the language.
433 "``hidden``" - Hidden style
434 Two declarations of an object with hidden visibility refer to the
435 same object if they are in the same shared object. Usually, hidden
436 visibility indicates that the symbol will not be placed into the
437 dynamic symbol table, so no other module (executable or shared
438 library) can reference it directly.
439 "``protected``" - Protected style
440 On ELF, protected visibility indicates that the symbol will be
441 placed in the dynamic symbol table, but that references within the
442 defining module will bind to the local symbol. That is, the symbol
443 cannot be overridden by another module.
445 A symbol with ``internal`` or ``private`` linkage must have ``default``
453 All Global Variables, Functions and Aliases can have one of the following
457 "``dllimport``" causes the compiler to reference a function or variable via
458 a global pointer to a pointer that is set up by the DLL exporting the
459 symbol. On Microsoft Windows targets, the pointer name is formed by
460 combining ``__imp_`` and the function or variable name.
462 "``dllexport``" causes the compiler to provide a global pointer to a pointer
463 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
464 Microsoft Windows targets, the pointer name is formed by combining
465 ``__imp_`` and the function or variable name. Since this storage class
466 exists for defining a dll interface, the compiler, assembler and linker know
467 it is externally referenced and must refrain from deleting the symbol.
471 Thread Local Storage Models
472 ---------------------------
474 A variable may be defined as ``thread_local``, which means that it will
475 not be shared by threads (each thread will have a separated copy of the
476 variable). Not all targets support thread-local variables. Optionally, a
477 TLS model may be specified:
480 For variables that are only used within the current shared library.
482 For variables in modules that will not be loaded dynamically.
484 For variables defined in the executable and only used within it.
486 If no explicit model is given, the "general dynamic" model is used.
488 The models correspond to the ELF TLS models; see `ELF Handling For
489 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
490 more information on under which circumstances the different models may
491 be used. The target may choose a different TLS model if the specified
492 model is not supported, or if a better choice of model can be made.
494 A model can also be specified in a alias, but then it only governs how
495 the alias is accessed. It will not have any effect in the aliasee.
502 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
503 types <t_struct>`. Literal types are uniqued structurally, but identified types
504 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
505 to forward declare a type that is not yet available.
507 An example of a identified structure specification is:
511 %mytype = type { %mytype*, i32 }
513 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
514 literal types are uniqued in recent versions of LLVM.
521 Global variables define regions of memory allocated at compilation time
524 Global variables definitions must be initialized.
526 Global variables in other translation units can also be declared, in which
527 case they don't have an initializer.
529 Either global variable definitions or declarations may have an explicit section
530 to be placed in and may have an optional explicit alignment specified.
532 A variable may be defined as a global ``constant``, which indicates that
533 the contents of the variable will **never** be modified (enabling better
534 optimization, allowing the global data to be placed in the read-only
535 section of an executable, etc). Note that variables that need runtime
536 initialization cannot be marked ``constant`` as there is a store to the
539 LLVM explicitly allows *declarations* of global variables to be marked
540 constant, even if the final definition of the global is not. This
541 capability can be used to enable slightly better optimization of the
542 program, but requires the language definition to guarantee that
543 optimizations based on the 'constantness' are valid for the translation
544 units that do not include the definition.
546 As SSA values, global variables define pointer values that are in scope
547 (i.e. they dominate) all basic blocks in the program. Global variables
548 always define a pointer to their "content" type because they describe a
549 region of memory, and all memory objects in LLVM are accessed through
552 Global variables can be marked with ``unnamed_addr`` which indicates
553 that the address is not significant, only the content. Constants marked
554 like this can be merged with other constants if they have the same
555 initializer. Note that a constant with significant address *can* be
556 merged with a ``unnamed_addr`` constant, the result being a constant
557 whose address is significant.
559 A global variable may be declared to reside in a target-specific
560 numbered address space. For targets that support them, address spaces
561 may affect how optimizations are performed and/or what target
562 instructions are used to access the variable. The default address space
563 is zero. The address space qualifier must precede any other attributes.
565 LLVM allows an explicit section to be specified for globals. If the
566 target supports it, it will emit globals to the section specified.
567 Additionally, the global can placed in a comdat if the target has the necessary
570 By default, global initializers are optimized by assuming that global
571 variables defined within the module are not modified from their
572 initial values before the start of the global initializer. This is
573 true even for variables potentially accessible from outside the
574 module, including those with external linkage or appearing in
575 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
576 by marking the variable with ``externally_initialized``.
578 An explicit alignment may be specified for a global, which must be a
579 power of 2. If not present, or if the alignment is set to zero, the
580 alignment of the global is set by the target to whatever it feels
581 convenient. If an explicit alignment is specified, the global is forced
582 to have exactly that alignment. Targets and optimizers are not allowed
583 to over-align the global if the global has an assigned section. In this
584 case, the extra alignment could be observable: for example, code could
585 assume that the globals are densely packed in their section and try to
586 iterate over them as an array, alignment padding would break this
587 iteration. The maximum alignment is ``1 << 29``.
589 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
591 Variables and aliasaes can have a
592 :ref:`Thread Local Storage Model <tls_model>`.
596 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
597 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
598 <global | constant> <Type> [<InitializerConstant>]
599 [, section "name"] [, align <Alignment>]
601 For example, the following defines a global in a numbered address space
602 with an initializer, section, and alignment:
606 @G = addrspace(5) constant float 1.0, section "foo", align 4
608 The following example just declares a global variable
612 @G = external global i32
614 The following example defines a thread-local global with the
615 ``initialexec`` TLS model:
619 @G = thread_local(initialexec) global i32 0, align 4
621 .. _functionstructure:
626 LLVM function definitions consist of the "``define``" keyword, an
627 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
628 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
629 an optional :ref:`calling convention <callingconv>`,
630 an optional ``unnamed_addr`` attribute, a return type, an optional
631 :ref:`parameter attribute <paramattrs>` for the return type, a function
632 name, a (possibly empty) argument list (each with optional :ref:`parameter
633 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
634 an optional section, an optional alignment,
635 an optional :ref:`comdat <langref_comdats>`,
636 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
637 curly brace, a list of basic blocks, and a closing curly brace.
639 LLVM function declarations consist of the "``declare``" keyword, an
640 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
641 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
642 an optional :ref:`calling convention <callingconv>`,
643 an optional ``unnamed_addr`` attribute, a return type, an optional
644 :ref:`parameter attribute <paramattrs>` for the return type, a function
645 name, a possibly empty list of arguments, an optional alignment, an optional
646 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
648 A function definition contains a list of basic blocks, forming the CFG (Control
649 Flow Graph) for the function. Each basic block may optionally start with a label
650 (giving the basic block a symbol table entry), contains a list of instructions,
651 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
652 function return). If an explicit label is not provided, a block is assigned an
653 implicit numbered label, using the next value from the same counter as used for
654 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
655 entry block does not have an explicit label, it will be assigned label "%0",
656 then the first unnamed temporary in that block will be "%1", etc.
658 The first basic block in a function is special in two ways: it is
659 immediately executed on entrance to the function, and it is not allowed
660 to have predecessor basic blocks (i.e. there can not be any branches to
661 the entry block of a function). Because the block can have no
662 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
664 LLVM allows an explicit section to be specified for functions. If the
665 target supports it, it will emit functions to the section specified.
666 Additionally, the function can placed in a COMDAT.
668 An explicit alignment may be specified for a function. If not present,
669 or if the alignment is set to zero, the alignment of the function is set
670 by the target to whatever it feels convenient. If an explicit alignment
671 is specified, the function is forced to have at least that much
672 alignment. All alignments must be a power of 2.
674 If the ``unnamed_addr`` attribute is given, the address is know to not
675 be significant and two identical functions can be merged.
679 define [linkage] [visibility] [DLLStorageClass]
681 <ResultType> @<FunctionName> ([argument list])
682 [unnamed_addr] [fn Attrs] [section "name"] [comdat $<ComdatName>]
683 [align N] [gc] [prefix Constant] { ... }
685 The argument list is a comma seperated sequence of arguments where each
686 argument is of the following form
690 <type> [parameter Attrs] [name]
698 Aliases, unlike function or variables, don't create any new data. They
699 are just a new symbol and metadata for an existing position.
701 Aliases have a name and an aliasee that is either a global value or a
704 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
705 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
706 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
710 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy> @<Aliasee>
712 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
713 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
714 might not correctly handle dropping a weak symbol that is aliased.
716 Alias that are not ``unnamed_addr`` are guaranteed to have the same address as
717 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
720 Since aliases are only a second name, some restrictions apply, of which
721 some can only be checked when producing an object file:
723 * The expression defining the aliasee must be computable at assembly
724 time. Since it is just a name, no relocations can be used.
726 * No alias in the expression can be weak as the possibility of the
727 intermediate alias being overridden cannot be represented in an
730 * No global value in the expression can be a declaration, since that
731 would require a relocation, which is not possible.
738 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
740 Comdats have a name which represents the COMDAT key. All global objects that
741 specify this key will only end up in the final object file if the linker chooses
742 that key over some other key. Aliases are placed in the same COMDAT that their
743 aliasee computes to, if any.
745 Comdats have a selection kind to provide input on how the linker should
746 choose between keys in two different object files.
750 $<Name> = comdat SelectionKind
752 The selection kind must be one of the following:
755 The linker may choose any COMDAT key, the choice is arbitrary.
757 The linker may choose any COMDAT key but the sections must contain the
760 The linker will choose the section containing the largest COMDAT key.
762 The linker requires that only section with this COMDAT key exist.
764 The linker may choose any COMDAT key but the sections must contain the
767 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
768 ``any`` as a selection kind.
770 Here is an example of a COMDAT group where a function will only be selected if
771 the COMDAT key's section is the largest:
775 $foo = comdat largest
776 @foo = global i32 2, comdat $foo
778 define void @bar() comdat $foo {
782 In a COFF object file, this will create a COMDAT section with selection kind
783 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
784 and another COMDAT section with selection kind
785 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
786 section and contains the contents of the ``@bar`` symbol.
788 There are some restrictions on the properties of the global object.
789 It, or an alias to it, must have the same name as the COMDAT group when
791 The contents and size of this object may be used during link-time to determine
792 which COMDAT groups get selected depending on the selection kind.
793 Because the name of the object must match the name of the COMDAT group, the
794 linkage of the global object must not be local; local symbols can get renamed
795 if a collision occurs in the symbol table.
797 The combined use of COMDATS and section attributes may yield surprising results.
804 @g1 = global i32 42, section "sec", comdat $foo
805 @g2 = global i32 42, section "sec", comdat $bar
807 From the object file perspective, this requires the creation of two sections
808 with the same name. This is necessary because both globals belong to different
809 COMDAT groups and COMDATs, at the object file level, are represented by
812 Note that certain IR constructs like global variables and functions may create
813 COMDATs in the object file in addition to any which are specified using COMDAT
814 IR. This arises, for example, when a global variable has linkonce_odr linkage.
816 .. _namedmetadatastructure:
821 Named metadata is a collection of metadata. :ref:`Metadata
822 nodes <metadata>` (but not metadata strings) are the only valid
823 operands for a named metadata.
827 ; Some unnamed metadata nodes, which are referenced by the named metadata.
828 !0 = metadata !{metadata !"zero"}
829 !1 = metadata !{metadata !"one"}
830 !2 = metadata !{metadata !"two"}
832 !name = !{!0, !1, !2}
839 The return type and each parameter of a function type may have a set of
840 *parameter attributes* associated with them. Parameter attributes are
841 used to communicate additional information about the result or
842 parameters of a function. Parameter attributes are considered to be part
843 of the function, not of the function type, so functions with different
844 parameter attributes can have the same function type.
846 Parameter attributes are simple keywords that follow the type specified.
847 If multiple parameter attributes are needed, they are space separated.
852 declare i32 @printf(i8* noalias nocapture, ...)
853 declare i32 @atoi(i8 zeroext)
854 declare signext i8 @returns_signed_char()
856 Note that any attributes for the function result (``nounwind``,
857 ``readonly``) come immediately after the argument list.
859 Currently, only the following parameter attributes are defined:
862 This indicates to the code generator that the parameter or return
863 value should be zero-extended to the extent required by the target's
864 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
865 the caller (for a parameter) or the callee (for a return value).
867 This indicates to the code generator that the parameter or return
868 value should be sign-extended to the extent required by the target's
869 ABI (which is usually 32-bits) by the caller (for a parameter) or
870 the callee (for a return value).
872 This indicates that this parameter or return value should be treated
873 in a special target-dependent fashion during while emitting code for
874 a function call or return (usually, by putting it in a register as
875 opposed to memory, though some targets use it to distinguish between
876 two different kinds of registers). Use of this attribute is
879 This indicates that the pointer parameter should really be passed by
880 value to the function. The attribute implies that a hidden copy of
881 the pointee is made between the caller and the callee, so the callee
882 is unable to modify the value in the caller. This attribute is only
883 valid on LLVM pointer arguments. It is generally used to pass
884 structs and arrays by value, but is also valid on pointers to
885 scalars. The copy is considered to belong to the caller not the
886 callee (for example, ``readonly`` functions should not write to
887 ``byval`` parameters). This is not a valid attribute for return
890 The byval attribute also supports specifying an alignment with the
891 align attribute. It indicates the alignment of the stack slot to
892 form and the known alignment of the pointer specified to the call
893 site. If the alignment is not specified, then the code generator
894 makes a target-specific assumption.
900 The ``inalloca`` argument attribute allows the caller to take the
901 address of outgoing stack arguments. An ``inalloca`` argument must
902 be a pointer to stack memory produced by an ``alloca`` instruction.
903 The alloca, or argument allocation, must also be tagged with the
904 inalloca keyword. Only the last argument may have the ``inalloca``
905 attribute, and that argument is guaranteed to be passed in memory.
907 An argument allocation may be used by a call at most once because
908 the call may deallocate it. The ``inalloca`` attribute cannot be
909 used in conjunction with other attributes that affect argument
910 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
911 ``inalloca`` attribute also disables LLVM's implicit lowering of
912 large aggregate return values, which means that frontend authors
913 must lower them with ``sret`` pointers.
915 When the call site is reached, the argument allocation must have
916 been the most recent stack allocation that is still live, or the
917 results are undefined. It is possible to allocate additional stack
918 space after an argument allocation and before its call site, but it
919 must be cleared off with :ref:`llvm.stackrestore
922 See :doc:`InAlloca` for more information on how to use this
926 This indicates that the pointer parameter specifies the address of a
927 structure that is the return value of the function in the source
928 program. This pointer must be guaranteed by the caller to be valid:
929 loads and stores to the structure may be assumed by the callee
930 not to trap and to be properly aligned. This may only be applied to
931 the first parameter. This is not a valid attribute for return
935 This indicates that the pointer value may be assumed by the optimizer to
936 have the specified alignment.
938 Note that this attribute has additional semantics when combined with the
944 This indicates that objects accessed via pointer values
945 :ref:`based <pointeraliasing>` on the argument or return value are not also
946 accessed, during the execution of the function, via pointer values not
947 *based* on the argument or return value. The attribute on a return value
948 also has additional semantics described below. The caller shares the
949 responsibility with the callee for ensuring that these requirements are met.
950 For further details, please see the discussion of the NoAlias response in
951 :ref:`alias analysis <Must, May, or No>`.
953 Note that this definition of ``noalias`` is intentionally similar
954 to the definition of ``restrict`` in C99 for function arguments.
956 For function return values, C99's ``restrict`` is not meaningful,
957 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
958 attribute on return values are stronger than the semantics of the attribute
959 when used on function arguments. On function return values, the ``noalias``
960 attribute indicates that the function acts like a system memory allocation
961 function, returning a pointer to allocated storage disjoint from the
962 storage for any other object accessible to the caller.
965 This indicates that the callee does not make any copies of the
966 pointer that outlive the callee itself. This is not a valid
967 attribute for return values.
972 This indicates that the pointer parameter can be excised using the
973 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
974 attribute for return values and can only be applied to one parameter.
977 This indicates that the function always returns the argument as its return
978 value. This is an optimization hint to the code generator when generating
979 the caller, allowing tail call optimization and omission of register saves
980 and restores in some cases; it is not checked or enforced when generating
981 the callee. The parameter and the function return type must be valid
982 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
983 valid attribute for return values and can only be applied to one parameter.
986 This indicates that the parameter or return pointer is not null. This
987 attribute may only be applied to pointer typed parameters. This is not
988 checked or enforced by LLVM, the caller must ensure that the pointer
989 passed in is non-null, or the callee must ensure that the returned pointer
992 ``dereferenceable(<n>)``
993 This indicates that the parameter or return pointer is dereferenceable. This
994 attribute may only be applied to pointer typed parameters. A pointer that
995 is dereferenceable can be loaded from speculatively without a risk of
996 trapping. The number of bytes known to be dereferenceable must be provided
997 in parentheses. It is legal for the number of bytes to be less than the
998 size of the pointee type. The ``nonnull`` attribute does not imply
999 dereferenceability (consider a pointer to one element past the end of an
1000 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1001 ``addrspace(0)`` (which is the default address space).
1005 Garbage Collector Names
1006 -----------------------
1008 Each function may specify a garbage collector name, which is simply a
1011 .. code-block:: llvm
1013 define void @f() gc "name" { ... }
1015 The compiler declares the supported values of *name*. Specifying a
1016 collector will cause the compiler to alter its output in order to
1017 support the named garbage collection algorithm.
1024 Prefix data is data associated with a function which the code generator
1025 will emit immediately before the function body. The purpose of this feature
1026 is to allow frontends to associate language-specific runtime metadata with
1027 specific functions and make it available through the function pointer while
1028 still allowing the function pointer to be called. To access the data for a
1029 given function, a program may bitcast the function pointer to a pointer to
1030 the constant's type. This implies that the IR symbol points to the start
1033 To maintain the semantics of ordinary function calls, the prefix data must
1034 have a particular format. Specifically, it must begin with a sequence of
1035 bytes which decode to a sequence of machine instructions, valid for the
1036 module's target, which transfer control to the point immediately succeeding
1037 the prefix data, without performing any other visible action. This allows
1038 the inliner and other passes to reason about the semantics of the function
1039 definition without needing to reason about the prefix data. Obviously this
1040 makes the format of the prefix data highly target dependent.
1042 Prefix data is laid out as if it were an initializer for a global variable
1043 of the prefix data's type. No padding is automatically placed between the
1044 prefix data and the function body. If padding is required, it must be part
1047 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
1048 which encodes the ``nop`` instruction:
1050 .. code-block:: llvm
1052 define void @f() prefix i8 144 { ... }
1054 Generally prefix data can be formed by encoding a relative branch instruction
1055 which skips the metadata, as in this example of valid prefix data for the
1056 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1058 .. code-block:: llvm
1060 %0 = type <{ i8, i8, i8* }>
1062 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
1064 A function may have prefix data but no body. This has similar semantics
1065 to the ``available_externally`` linkage in that the data may be used by the
1066 optimizers but will not be emitted in the object file.
1073 Attribute groups are groups of attributes that are referenced by objects within
1074 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1075 functions will use the same set of attributes. In the degenerative case of a
1076 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1077 group will capture the important command line flags used to build that file.
1079 An attribute group is a module-level object. To use an attribute group, an
1080 object references the attribute group's ID (e.g. ``#37``). An object may refer
1081 to more than one attribute group. In that situation, the attributes from the
1082 different groups are merged.
1084 Here is an example of attribute groups for a function that should always be
1085 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1087 .. code-block:: llvm
1089 ; Target-independent attributes:
1090 attributes #0 = { alwaysinline alignstack=4 }
1092 ; Target-dependent attributes:
1093 attributes #1 = { "no-sse" }
1095 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1096 define void @f() #0 #1 { ... }
1103 Function attributes are set to communicate additional information about
1104 a function. Function attributes are considered to be part of the
1105 function, not of the function type, so functions with different function
1106 attributes can have the same function type.
1108 Function attributes are simple keywords that follow the type specified.
1109 If multiple attributes are needed, they are space separated. For
1112 .. code-block:: llvm
1114 define void @f() noinline { ... }
1115 define void @f() alwaysinline { ... }
1116 define void @f() alwaysinline optsize { ... }
1117 define void @f() optsize { ... }
1120 This attribute indicates that, when emitting the prologue and
1121 epilogue, the backend should forcibly align the stack pointer.
1122 Specify the desired alignment, which must be a power of two, in
1125 This attribute indicates that the inliner should attempt to inline
1126 this function into callers whenever possible, ignoring any active
1127 inlining size threshold for this caller.
1129 This indicates that the callee function at a call site should be
1130 recognized as a built-in function, even though the function's declaration
1131 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1132 direct calls to functions that are declared with the ``nobuiltin``
1135 This attribute indicates that this function is rarely called. When
1136 computing edge weights, basic blocks post-dominated by a cold
1137 function call are also considered to be cold; and, thus, given low
1140 This attribute indicates that the source code contained a hint that
1141 inlining this function is desirable (such as the "inline" keyword in
1142 C/C++). It is just a hint; it imposes no requirements on the
1145 This attribute indicates that the function should be added to a
1146 jump-instruction table at code-generation time, and that all address-taken
1147 references to this function should be replaced with a reference to the
1148 appropriate jump-instruction-table function pointer. Note that this creates
1149 a new pointer for the original function, which means that code that depends
1150 on function-pointer identity can break. So, any function annotated with
1151 ``jumptable`` must also be ``unnamed_addr``.
1153 This attribute suggests that optimization passes and code generator
1154 passes make choices that keep the code size of this function as small
1155 as possible and perform optimizations that may sacrifice runtime
1156 performance in order to minimize the size of the generated code.
1158 This attribute disables prologue / epilogue emission for the
1159 function. This can have very system-specific consequences.
1161 This indicates that the callee function at a call site is not recognized as
1162 a built-in function. LLVM will retain the original call and not replace it
1163 with equivalent code based on the semantics of the built-in function, unless
1164 the call site uses the ``builtin`` attribute. This is valid at call sites
1165 and on function declarations and definitions.
1167 This attribute indicates that calls to the function cannot be
1168 duplicated. A call to a ``noduplicate`` function may be moved
1169 within its parent function, but may not be duplicated within
1170 its parent function.
1172 A function containing a ``noduplicate`` call may still
1173 be an inlining candidate, provided that the call is not
1174 duplicated by inlining. That implies that the function has
1175 internal linkage and only has one call site, so the original
1176 call is dead after inlining.
1178 This attributes disables implicit floating point instructions.
1180 This attribute indicates that the inliner should never inline this
1181 function in any situation. This attribute may not be used together
1182 with the ``alwaysinline`` attribute.
1184 This attribute suppresses lazy symbol binding for the function. This
1185 may make calls to the function faster, at the cost of extra program
1186 startup time if the function is not called during program startup.
1188 This attribute indicates that the code generator should not use a
1189 red zone, even if the target-specific ABI normally permits it.
1191 This function attribute indicates that the function never returns
1192 normally. This produces undefined behavior at runtime if the
1193 function ever does dynamically return.
1195 This function attribute indicates that the function never returns
1196 with an unwind or exceptional control flow. If the function does
1197 unwind, its runtime behavior is undefined.
1199 This function attribute indicates that the function is not optimized
1200 by any optimization or code generator passes with the
1201 exception of interprocedural optimization passes.
1202 This attribute cannot be used together with the ``alwaysinline``
1203 attribute; this attribute is also incompatible
1204 with the ``minsize`` attribute and the ``optsize`` attribute.
1206 This attribute requires the ``noinline`` attribute to be specified on
1207 the function as well, so the function is never inlined into any caller.
1208 Only functions with the ``alwaysinline`` attribute are valid
1209 candidates for inlining into the body of this function.
1211 This attribute suggests that optimization passes and code generator
1212 passes make choices that keep the code size of this function low,
1213 and otherwise do optimizations specifically to reduce code size as
1214 long as they do not significantly impact runtime performance.
1216 On a function, this attribute indicates that the function computes its
1217 result (or decides to unwind an exception) based strictly on its arguments,
1218 without dereferencing any pointer arguments or otherwise accessing
1219 any mutable state (e.g. memory, control registers, etc) visible to
1220 caller functions. It does not write through any pointer arguments
1221 (including ``byval`` arguments) and never changes any state visible
1222 to callers. This means that it cannot unwind exceptions by calling
1223 the ``C++`` exception throwing methods.
1225 On an argument, this attribute indicates that the function does not
1226 dereference that pointer argument, even though it may read or write the
1227 memory that the pointer points to if accessed through other pointers.
1229 On a function, this attribute indicates that the function does not write
1230 through any pointer arguments (including ``byval`` arguments) or otherwise
1231 modify any state (e.g. memory, control registers, etc) visible to
1232 caller functions. It may dereference pointer arguments and read
1233 state that may be set in the caller. A readonly function always
1234 returns the same value (or unwinds an exception identically) when
1235 called with the same set of arguments and global state. It cannot
1236 unwind an exception by calling the ``C++`` exception throwing
1239 On an argument, this attribute indicates that the function does not write
1240 through this pointer argument, even though it may write to the memory that
1241 the pointer points to.
1243 This attribute indicates that this function can return twice. The C
1244 ``setjmp`` is an example of such a function. The compiler disables
1245 some optimizations (like tail calls) in the caller of these
1247 ``sanitize_address``
1248 This attribute indicates that AddressSanitizer checks
1249 (dynamic address safety analysis) are enabled for this function.
1251 This attribute indicates that MemorySanitizer checks (dynamic detection
1252 of accesses to uninitialized memory) are enabled for this function.
1254 This attribute indicates that ThreadSanitizer checks
1255 (dynamic thread safety analysis) are enabled for this function.
1257 This attribute indicates that the function should emit a stack
1258 smashing protector. It is in the form of a "canary" --- a random value
1259 placed on the stack before the local variables that's checked upon
1260 return from the function to see if it has been overwritten. A
1261 heuristic is used to determine if a function needs stack protectors
1262 or not. The heuristic used will enable protectors for functions with:
1264 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1265 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1266 - Calls to alloca() with variable sizes or constant sizes greater than
1267 ``ssp-buffer-size``.
1269 Variables that are identified as requiring a protector will be arranged
1270 on the stack such that they are adjacent to the stack protector guard.
1272 If a function that has an ``ssp`` attribute is inlined into a
1273 function that doesn't have an ``ssp`` attribute, then the resulting
1274 function will have an ``ssp`` attribute.
1276 This attribute indicates that the function should *always* emit a
1277 stack smashing protector. This overrides the ``ssp`` function
1280 Variables that are identified as requiring a protector will be arranged
1281 on the stack such that they are adjacent to the stack protector guard.
1282 The specific layout rules are:
1284 #. Large arrays and structures containing large arrays
1285 (``>= ssp-buffer-size``) are closest to the stack protector.
1286 #. Small arrays and structures containing small arrays
1287 (``< ssp-buffer-size``) are 2nd closest to the protector.
1288 #. Variables that have had their address taken are 3rd closest to the
1291 If a function that has an ``sspreq`` attribute is inlined into a
1292 function that doesn't have an ``sspreq`` attribute or which has an
1293 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1294 an ``sspreq`` attribute.
1296 This attribute indicates that the function should emit a stack smashing
1297 protector. This attribute causes a strong heuristic to be used when
1298 determining if a function needs stack protectors. The strong heuristic
1299 will enable protectors for functions with:
1301 - Arrays of any size and type
1302 - Aggregates containing an array of any size and type.
1303 - Calls to alloca().
1304 - Local variables that have had their address taken.
1306 Variables that are identified as requiring a protector will be arranged
1307 on the stack such that they are adjacent to the stack protector guard.
1308 The specific layout rules are:
1310 #. Large arrays and structures containing large arrays
1311 (``>= ssp-buffer-size``) are closest to the stack protector.
1312 #. Small arrays and structures containing small arrays
1313 (``< ssp-buffer-size``) are 2nd closest to the protector.
1314 #. Variables that have had their address taken are 3rd closest to the
1317 This overrides the ``ssp`` function attribute.
1319 If a function that has an ``sspstrong`` attribute is inlined into a
1320 function that doesn't have an ``sspstrong`` attribute, then the
1321 resulting function will have an ``sspstrong`` attribute.
1323 This attribute indicates that the ABI being targeted requires that
1324 an unwind table entry be produce for this function even if we can
1325 show that no exceptions passes by it. This is normally the case for
1326 the ELF x86-64 abi, but it can be disabled for some compilation
1331 Module-Level Inline Assembly
1332 ----------------------------
1334 Modules may contain "module-level inline asm" blocks, which corresponds
1335 to the GCC "file scope inline asm" blocks. These blocks are internally
1336 concatenated by LLVM and treated as a single unit, but may be separated
1337 in the ``.ll`` file if desired. The syntax is very simple:
1339 .. code-block:: llvm
1341 module asm "inline asm code goes here"
1342 module asm "more can go here"
1344 The strings can contain any character by escaping non-printable
1345 characters. The escape sequence used is simply "\\xx" where "xx" is the
1346 two digit hex code for the number.
1348 The inline asm code is simply printed to the machine code .s file when
1349 assembly code is generated.
1351 .. _langref_datalayout:
1356 A module may specify a target specific data layout string that specifies
1357 how data is to be laid out in memory. The syntax for the data layout is
1360 .. code-block:: llvm
1362 target datalayout = "layout specification"
1364 The *layout specification* consists of a list of specifications
1365 separated by the minus sign character ('-'). Each specification starts
1366 with a letter and may include other information after the letter to
1367 define some aspect of the data layout. The specifications accepted are
1371 Specifies that the target lays out data in big-endian form. That is,
1372 the bits with the most significance have the lowest address
1375 Specifies that the target lays out data in little-endian form. That
1376 is, the bits with the least significance have the lowest address
1379 Specifies the natural alignment of the stack in bits. Alignment
1380 promotion of stack variables is limited to the natural stack
1381 alignment to avoid dynamic stack realignment. The stack alignment
1382 must be a multiple of 8-bits. If omitted, the natural stack
1383 alignment defaults to "unspecified", which does not prevent any
1384 alignment promotions.
1385 ``p[n]:<size>:<abi>:<pref>``
1386 This specifies the *size* of a pointer and its ``<abi>`` and
1387 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1388 bits. The address space, ``n`` is optional, and if not specified,
1389 denotes the default address space 0. The value of ``n`` must be
1390 in the range [1,2^23).
1391 ``i<size>:<abi>:<pref>``
1392 This specifies the alignment for an integer type of a given bit
1393 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1394 ``v<size>:<abi>:<pref>``
1395 This specifies the alignment for a vector type of a given bit
1397 ``f<size>:<abi>:<pref>``
1398 This specifies the alignment for a floating point type of a given bit
1399 ``<size>``. Only values of ``<size>`` that are supported by the target
1400 will work. 32 (float) and 64 (double) are supported on all targets; 80
1401 or 128 (different flavors of long double) are also supported on some
1404 This specifies the alignment for an object of aggregate type.
1406 If present, specifies that llvm names are mangled in the output. The
1409 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1410 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1411 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1412 symbols get a ``_`` prefix.
1413 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1414 functions also get a suffix based on the frame size.
1415 ``n<size1>:<size2>:<size3>...``
1416 This specifies a set of native integer widths for the target CPU in
1417 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1418 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1419 this set are considered to support most general arithmetic operations
1422 On every specification that takes a ``<abi>:<pref>``, specifying the
1423 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1424 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1426 When constructing the data layout for a given target, LLVM starts with a
1427 default set of specifications which are then (possibly) overridden by
1428 the specifications in the ``datalayout`` keyword. The default
1429 specifications are given in this list:
1431 - ``E`` - big endian
1432 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1433 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1434 same as the default address space.
1435 - ``S0`` - natural stack alignment is unspecified
1436 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1437 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1438 - ``i16:16:16`` - i16 is 16-bit aligned
1439 - ``i32:32:32`` - i32 is 32-bit aligned
1440 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1441 alignment of 64-bits
1442 - ``f16:16:16`` - half is 16-bit aligned
1443 - ``f32:32:32`` - float is 32-bit aligned
1444 - ``f64:64:64`` - double is 64-bit aligned
1445 - ``f128:128:128`` - quad is 128-bit aligned
1446 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1447 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1448 - ``a:0:64`` - aggregates are 64-bit aligned
1450 When LLVM is determining the alignment for a given type, it uses the
1453 #. If the type sought is an exact match for one of the specifications,
1454 that specification is used.
1455 #. If no match is found, and the type sought is an integer type, then
1456 the smallest integer type that is larger than the bitwidth of the
1457 sought type is used. If none of the specifications are larger than
1458 the bitwidth then the largest integer type is used. For example,
1459 given the default specifications above, the i7 type will use the
1460 alignment of i8 (next largest) while both i65 and i256 will use the
1461 alignment of i64 (largest specified).
1462 #. If no match is found, and the type sought is a vector type, then the
1463 largest vector type that is smaller than the sought vector type will
1464 be used as a fall back. This happens because <128 x double> can be
1465 implemented in terms of 64 <2 x double>, for example.
1467 The function of the data layout string may not be what you expect.
1468 Notably, this is not a specification from the frontend of what alignment
1469 the code generator should use.
1471 Instead, if specified, the target data layout is required to match what
1472 the ultimate *code generator* expects. This string is used by the
1473 mid-level optimizers to improve code, and this only works if it matches
1474 what the ultimate code generator uses. If you would like to generate IR
1475 that does not embed this target-specific detail into the IR, then you
1476 don't have to specify the string. This will disable some optimizations
1477 that require precise layout information, but this also prevents those
1478 optimizations from introducing target specificity into the IR.
1485 A module may specify a target triple string that describes the target
1486 host. The syntax for the target triple is simply:
1488 .. code-block:: llvm
1490 target triple = "x86_64-apple-macosx10.7.0"
1492 The *target triple* string consists of a series of identifiers delimited
1493 by the minus sign character ('-'). The canonical forms are:
1497 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1498 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1500 This information is passed along to the backend so that it generates
1501 code for the proper architecture. It's possible to override this on the
1502 command line with the ``-mtriple`` command line option.
1504 .. _pointeraliasing:
1506 Pointer Aliasing Rules
1507 ----------------------
1509 Any memory access must be done through a pointer value associated with
1510 an address range of the memory access, otherwise the behavior is
1511 undefined. Pointer values are associated with address ranges according
1512 to the following rules:
1514 - A pointer value is associated with the addresses associated with any
1515 value it is *based* on.
1516 - An address of a global variable is associated with the address range
1517 of the variable's storage.
1518 - The result value of an allocation instruction is associated with the
1519 address range of the allocated storage.
1520 - A null pointer in the default address-space is associated with no
1522 - An integer constant other than zero or a pointer value returned from
1523 a function not defined within LLVM may be associated with address
1524 ranges allocated through mechanisms other than those provided by
1525 LLVM. Such ranges shall not overlap with any ranges of addresses
1526 allocated by mechanisms provided by LLVM.
1528 A pointer value is *based* on another pointer value according to the
1531 - A pointer value formed from a ``getelementptr`` operation is *based*
1532 on the first operand of the ``getelementptr``.
1533 - The result value of a ``bitcast`` is *based* on the operand of the
1535 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1536 values that contribute (directly or indirectly) to the computation of
1537 the pointer's value.
1538 - The "*based* on" relationship is transitive.
1540 Note that this definition of *"based"* is intentionally similar to the
1541 definition of *"based"* in C99, though it is slightly weaker.
1543 LLVM IR does not associate types with memory. The result type of a
1544 ``load`` merely indicates the size and alignment of the memory from
1545 which to load, as well as the interpretation of the value. The first
1546 operand type of a ``store`` similarly only indicates the size and
1547 alignment of the store.
1549 Consequently, type-based alias analysis, aka TBAA, aka
1550 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1551 :ref:`Metadata <metadata>` may be used to encode additional information
1552 which specialized optimization passes may use to implement type-based
1557 Volatile Memory Accesses
1558 ------------------------
1560 Certain memory accesses, such as :ref:`load <i_load>`'s,
1561 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1562 marked ``volatile``. The optimizers must not change the number of
1563 volatile operations or change their order of execution relative to other
1564 volatile operations. The optimizers *may* change the order of volatile
1565 operations relative to non-volatile operations. This is not Java's
1566 "volatile" and has no cross-thread synchronization behavior.
1568 IR-level volatile loads and stores cannot safely be optimized into
1569 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1570 flagged volatile. Likewise, the backend should never split or merge
1571 target-legal volatile load/store instructions.
1573 .. admonition:: Rationale
1575 Platforms may rely on volatile loads and stores of natively supported
1576 data width to be executed as single instruction. For example, in C
1577 this holds for an l-value of volatile primitive type with native
1578 hardware support, but not necessarily for aggregate types. The
1579 frontend upholds these expectations, which are intentionally
1580 unspecified in the IR. The rules above ensure that IR transformation
1581 do not violate the frontend's contract with the language.
1585 Memory Model for Concurrent Operations
1586 --------------------------------------
1588 The LLVM IR does not define any way to start parallel threads of
1589 execution or to register signal handlers. Nonetheless, there are
1590 platform-specific ways to create them, and we define LLVM IR's behavior
1591 in their presence. This model is inspired by the C++0x memory model.
1593 For a more informal introduction to this model, see the :doc:`Atomics`.
1595 We define a *happens-before* partial order as the least partial order
1598 - Is a superset of single-thread program order, and
1599 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1600 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1601 techniques, like pthread locks, thread creation, thread joining,
1602 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1603 Constraints <ordering>`).
1605 Note that program order does not introduce *happens-before* edges
1606 between a thread and signals executing inside that thread.
1608 Every (defined) read operation (load instructions, memcpy, atomic
1609 loads/read-modify-writes, etc.) R reads a series of bytes written by
1610 (defined) write operations (store instructions, atomic
1611 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1612 section, initialized globals are considered to have a write of the
1613 initializer which is atomic and happens before any other read or write
1614 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1615 may see any write to the same byte, except:
1617 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1618 write\ :sub:`2` happens before R\ :sub:`byte`, then
1619 R\ :sub:`byte` does not see write\ :sub:`1`.
1620 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1621 R\ :sub:`byte` does not see write\ :sub:`3`.
1623 Given that definition, R\ :sub:`byte` is defined as follows:
1625 - If R is volatile, the result is target-dependent. (Volatile is
1626 supposed to give guarantees which can support ``sig_atomic_t`` in
1627 C/C++, and may be used for accesses to addresses that do not behave
1628 like normal memory. It does not generally provide cross-thread
1630 - Otherwise, if there is no write to the same byte that happens before
1631 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1632 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1633 R\ :sub:`byte` returns the value written by that write.
1634 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1635 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1636 Memory Ordering Constraints <ordering>` section for additional
1637 constraints on how the choice is made.
1638 - Otherwise R\ :sub:`byte` returns ``undef``.
1640 R returns the value composed of the series of bytes it read. This
1641 implies that some bytes within the value may be ``undef`` **without**
1642 the entire value being ``undef``. Note that this only defines the
1643 semantics of the operation; it doesn't mean that targets will emit more
1644 than one instruction to read the series of bytes.
1646 Note that in cases where none of the atomic intrinsics are used, this
1647 model places only one restriction on IR transformations on top of what
1648 is required for single-threaded execution: introducing a store to a byte
1649 which might not otherwise be stored is not allowed in general.
1650 (Specifically, in the case where another thread might write to and read
1651 from an address, introducing a store can change a load that may see
1652 exactly one write into a load that may see multiple writes.)
1656 Atomic Memory Ordering Constraints
1657 ----------------------------------
1659 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1660 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1661 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1662 ordering parameters that determine which other atomic instructions on
1663 the same address they *synchronize with*. These semantics are borrowed
1664 from Java and C++0x, but are somewhat more colloquial. If these
1665 descriptions aren't precise enough, check those specs (see spec
1666 references in the :doc:`atomics guide <Atomics>`).
1667 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1668 differently since they don't take an address. See that instruction's
1669 documentation for details.
1671 For a simpler introduction to the ordering constraints, see the
1675 The set of values that can be read is governed by the happens-before
1676 partial order. A value cannot be read unless some operation wrote
1677 it. This is intended to provide a guarantee strong enough to model
1678 Java's non-volatile shared variables. This ordering cannot be
1679 specified for read-modify-write operations; it is not strong enough
1680 to make them atomic in any interesting way.
1682 In addition to the guarantees of ``unordered``, there is a single
1683 total order for modifications by ``monotonic`` operations on each
1684 address. All modification orders must be compatible with the
1685 happens-before order. There is no guarantee that the modification
1686 orders can be combined to a global total order for the whole program
1687 (and this often will not be possible). The read in an atomic
1688 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1689 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1690 order immediately before the value it writes. If one atomic read
1691 happens before another atomic read of the same address, the later
1692 read must see the same value or a later value in the address's
1693 modification order. This disallows reordering of ``monotonic`` (or
1694 stronger) operations on the same address. If an address is written
1695 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1696 read that address repeatedly, the other threads must eventually see
1697 the write. This corresponds to the C++0x/C1x
1698 ``memory_order_relaxed``.
1700 In addition to the guarantees of ``monotonic``, a
1701 *synchronizes-with* edge may be formed with a ``release`` operation.
1702 This is intended to model C++'s ``memory_order_acquire``.
1704 In addition to the guarantees of ``monotonic``, if this operation
1705 writes a value which is subsequently read by an ``acquire``
1706 operation, it *synchronizes-with* that operation. (This isn't a
1707 complete description; see the C++0x definition of a release
1708 sequence.) This corresponds to the C++0x/C1x
1709 ``memory_order_release``.
1710 ``acq_rel`` (acquire+release)
1711 Acts as both an ``acquire`` and ``release`` operation on its
1712 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1713 ``seq_cst`` (sequentially consistent)
1714 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1715 operation that only reads, ``release`` for an operation that only
1716 writes), there is a global total order on all
1717 sequentially-consistent operations on all addresses, which is
1718 consistent with the *happens-before* partial order and with the
1719 modification orders of all the affected addresses. Each
1720 sequentially-consistent read sees the last preceding write to the
1721 same address in this global order. This corresponds to the C++0x/C1x
1722 ``memory_order_seq_cst`` and Java volatile.
1726 If an atomic operation is marked ``singlethread``, it only *synchronizes
1727 with* or participates in modification and seq\_cst total orderings with
1728 other operations running in the same thread (for example, in signal
1736 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1737 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1738 :ref:`frem <i_frem>`) have the following flags that can set to enable
1739 otherwise unsafe floating point operations
1742 No NaNs - Allow optimizations to assume the arguments and result are not
1743 NaN. Such optimizations are required to retain defined behavior over
1744 NaNs, but the value of the result is undefined.
1747 No Infs - Allow optimizations to assume the arguments and result are not
1748 +/-Inf. Such optimizations are required to retain defined behavior over
1749 +/-Inf, but the value of the result is undefined.
1752 No Signed Zeros - Allow optimizations to treat the sign of a zero
1753 argument or result as insignificant.
1756 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1757 argument rather than perform division.
1760 Fast - Allow algebraically equivalent transformations that may
1761 dramatically change results in floating point (e.g. reassociate). This
1762 flag implies all the others.
1766 Use-list Order Directives
1767 -------------------------
1769 Use-list directives encode the in-memory order of each use-list, allowing the
1770 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1771 indexes that are assigned to the referenced value's uses. The referenced
1772 value's use-list is immediately sorted by these indexes.
1774 Use-list directives may appear at function scope or global scope. They are not
1775 instructions, and have no effect on the semantics of the IR. When they're at
1776 function scope, they must appear after the terminator of the final basic block.
1778 If basic blocks have their address taken via ``blockaddress()`` expressions,
1779 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1786 uselistorder <ty> <value>, { <order-indexes> }
1787 uselistorder_bb @function, %block { <order-indexes> }
1793 define void @foo(i32 %arg1, i32 %arg2) {
1795 ; ... instructions ...
1797 ; ... instructions ...
1799 ; At function scope.
1800 uselistorder i32 %arg1, { 1, 0, 2 }
1801 uselistorder label %bb, { 1, 0 }
1805 uselistorder i32* @global, { 1, 2, 0 }
1806 uselistorder i32 7, { 1, 0 }
1807 uselistorder i32 (i32) @bar, { 1, 0 }
1808 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1815 The LLVM type system is one of the most important features of the
1816 intermediate representation. Being typed enables a number of
1817 optimizations to be performed on the intermediate representation
1818 directly, without having to do extra analyses on the side before the
1819 transformation. A strong type system makes it easier to read the
1820 generated code and enables novel analyses and transformations that are
1821 not feasible to perform on normal three address code representations.
1831 The void type does not represent any value and has no size.
1849 The function type can be thought of as a function signature. It consists of a
1850 return type and a list of formal parameter types. The return type of a function
1851 type is a void type or first class type --- except for :ref:`label <t_label>`
1852 and :ref:`metadata <t_metadata>` types.
1858 <returntype> (<parameter list>)
1860 ...where '``<parameter list>``' is a comma-separated list of type
1861 specifiers. Optionally, the parameter list may include a type ``...``, which
1862 indicates that the function takes a variable number of arguments. Variable
1863 argument functions can access their arguments with the :ref:`variable argument
1864 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1865 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1869 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1870 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1871 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1872 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1873 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1874 | ``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. |
1875 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1876 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1877 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1884 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1885 Values of these types are the only ones which can be produced by
1893 These are the types that are valid in registers from CodeGen's perspective.
1902 The integer type is a very simple type that simply specifies an
1903 arbitrary bit width for the integer type desired. Any bit width from 1
1904 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1912 The number of bits the integer will occupy is specified by the ``N``
1918 +----------------+------------------------------------------------+
1919 | ``i1`` | a single-bit integer. |
1920 +----------------+------------------------------------------------+
1921 | ``i32`` | a 32-bit integer. |
1922 +----------------+------------------------------------------------+
1923 | ``i1942652`` | a really big integer of over 1 million bits. |
1924 +----------------+------------------------------------------------+
1928 Floating Point Types
1929 """"""""""""""""""""
1938 - 16-bit floating point value
1941 - 32-bit floating point value
1944 - 64-bit floating point value
1947 - 128-bit floating point value (112-bit mantissa)
1950 - 80-bit floating point value (X87)
1953 - 128-bit floating point value (two 64-bits)
1960 The x86_mmx type represents a value held in an MMX register on an x86
1961 machine. The operations allowed on it are quite limited: parameters and
1962 return values, load and store, and bitcast. User-specified MMX
1963 instructions are represented as intrinsic or asm calls with arguments
1964 and/or results of this type. There are no arrays, vectors or constants
1981 The pointer type is used to specify memory locations. Pointers are
1982 commonly used to reference objects in memory.
1984 Pointer types may have an optional address space attribute defining the
1985 numbered address space where the pointed-to object resides. The default
1986 address space is number zero. The semantics of non-zero address spaces
1987 are target-specific.
1989 Note that LLVM does not permit pointers to void (``void*``) nor does it
1990 permit pointers to labels (``label*``). Use ``i8*`` instead.
2000 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2001 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2002 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2003 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2004 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2005 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2006 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2015 A vector type is a simple derived type that represents a vector of
2016 elements. Vector types are used when multiple primitive data are
2017 operated in parallel using a single instruction (SIMD). A vector type
2018 requires a size (number of elements) and an underlying primitive data
2019 type. Vector types are considered :ref:`first class <t_firstclass>`.
2025 < <# elements> x <elementtype> >
2027 The number of elements is a constant integer value larger than 0;
2028 elementtype may be any integer, floating point or pointer type. Vectors
2029 of size zero are not allowed.
2033 +-------------------+--------------------------------------------------+
2034 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2035 +-------------------+--------------------------------------------------+
2036 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2037 +-------------------+--------------------------------------------------+
2038 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2039 +-------------------+--------------------------------------------------+
2040 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2041 +-------------------+--------------------------------------------------+
2050 The label type represents code labels.
2065 The metadata type represents embedded metadata. No derived types may be
2066 created from metadata except for :ref:`function <t_function>` arguments.
2079 Aggregate Types are a subset of derived types that can contain multiple
2080 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2081 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2091 The array type is a very simple derived type that arranges elements
2092 sequentially in memory. The array type requires a size (number of
2093 elements) and an underlying data type.
2099 [<# elements> x <elementtype>]
2101 The number of elements is a constant integer value; ``elementtype`` may
2102 be any type with a size.
2106 +------------------+--------------------------------------+
2107 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2108 +------------------+--------------------------------------+
2109 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2110 +------------------+--------------------------------------+
2111 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2112 +------------------+--------------------------------------+
2114 Here are some examples of multidimensional arrays:
2116 +-----------------------------+----------------------------------------------------------+
2117 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2118 +-----------------------------+----------------------------------------------------------+
2119 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2120 +-----------------------------+----------------------------------------------------------+
2121 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2122 +-----------------------------+----------------------------------------------------------+
2124 There is no restriction on indexing beyond the end of the array implied
2125 by a static type (though there are restrictions on indexing beyond the
2126 bounds of an allocated object in some cases). This means that
2127 single-dimension 'variable sized array' addressing can be implemented in
2128 LLVM with a zero length array type. An implementation of 'pascal style
2129 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2139 The structure type is used to represent a collection of data members
2140 together in memory. The elements of a structure may be any type that has
2143 Structures in memory are accessed using '``load``' and '``store``' by
2144 getting a pointer to a field with the '``getelementptr``' instruction.
2145 Structures in registers are accessed using the '``extractvalue``' and
2146 '``insertvalue``' instructions.
2148 Structures may optionally be "packed" structures, which indicate that
2149 the alignment of the struct is one byte, and that there is no padding
2150 between the elements. In non-packed structs, padding between field types
2151 is inserted as defined by the DataLayout string in the module, which is
2152 required to match what the underlying code generator expects.
2154 Structures can either be "literal" or "identified". A literal structure
2155 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2156 identified types are always defined at the top level with a name.
2157 Literal types are uniqued by their contents and can never be recursive
2158 or opaque since there is no way to write one. Identified types can be
2159 recursive, can be opaqued, and are never uniqued.
2165 %T1 = type { <type list> } ; Identified normal struct type
2166 %T2 = type <{ <type list> }> ; Identified packed struct type
2170 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2171 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2172 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2173 | ``{ 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``. |
2174 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2175 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2176 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2180 Opaque Structure Types
2181 """"""""""""""""""""""
2185 Opaque structure types are used to represent named structure types that
2186 do not have a body specified. This corresponds (for example) to the C
2187 notion of a forward declared structure.
2198 +--------------+-------------------+
2199 | ``opaque`` | An opaque type. |
2200 +--------------+-------------------+
2207 LLVM has several different basic types of constants. This section
2208 describes them all and their syntax.
2213 **Boolean constants**
2214 The two strings '``true``' and '``false``' are both valid constants
2216 **Integer constants**
2217 Standard integers (such as '4') are constants of the
2218 :ref:`integer <t_integer>` type. Negative numbers may be used with
2220 **Floating point constants**
2221 Floating point constants use standard decimal notation (e.g.
2222 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2223 hexadecimal notation (see below). The assembler requires the exact
2224 decimal value of a floating-point constant. For example, the
2225 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2226 decimal in binary. Floating point constants must have a :ref:`floating
2227 point <t_floating>` type.
2228 **Null pointer constants**
2229 The identifier '``null``' is recognized as a null pointer constant
2230 and must be of :ref:`pointer type <t_pointer>`.
2232 The one non-intuitive notation for constants is the hexadecimal form of
2233 floating point constants. For example, the form
2234 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2235 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2236 constants are required (and the only time that they are generated by the
2237 disassembler) is when a floating point constant must be emitted but it
2238 cannot be represented as a decimal floating point number in a reasonable
2239 number of digits. For example, NaN's, infinities, and other special
2240 values are represented in their IEEE hexadecimal format so that assembly
2241 and disassembly do not cause any bits to change in the constants.
2243 When using the hexadecimal form, constants of types half, float, and
2244 double are represented using the 16-digit form shown above (which
2245 matches the IEEE754 representation for double); half and float values
2246 must, however, be exactly representable as IEEE 754 half and single
2247 precision, respectively. Hexadecimal format is always used for long
2248 double, and there are three forms of long double. The 80-bit format used
2249 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2250 128-bit format used by PowerPC (two adjacent doubles) is represented by
2251 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2252 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2253 will only work if they match the long double format on your target.
2254 The IEEE 16-bit format (half precision) is represented by ``0xH``
2255 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2256 (sign bit at the left).
2258 There are no constants of type x86_mmx.
2260 .. _complexconstants:
2265 Complex constants are a (potentially recursive) combination of simple
2266 constants and smaller complex constants.
2268 **Structure constants**
2269 Structure constants are represented with notation similar to
2270 structure type definitions (a comma separated list of elements,
2271 surrounded by braces (``{}``)). For example:
2272 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2273 "``@G = external global i32``". Structure constants must have
2274 :ref:`structure type <t_struct>`, and the number and types of elements
2275 must match those specified by the type.
2277 Array constants are represented with notation similar to array type
2278 definitions (a comma separated list of elements, surrounded by
2279 square brackets (``[]``)). For example:
2280 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2281 :ref:`array type <t_array>`, and the number and types of elements must
2282 match those specified by the type. As a special case, character array
2283 constants may also be represented as a double-quoted string using the ``c``
2284 prefix. For example: "``c"Hello World\0A\00"``".
2285 **Vector constants**
2286 Vector constants are represented with notation similar to vector
2287 type definitions (a comma separated list of elements, surrounded by
2288 less-than/greater-than's (``<>``)). For example:
2289 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2290 must have :ref:`vector type <t_vector>`, and the number and types of
2291 elements must match those specified by the type.
2292 **Zero initialization**
2293 The string '``zeroinitializer``' can be used to zero initialize a
2294 value to zero of *any* type, including scalar and
2295 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2296 having to print large zero initializers (e.g. for large arrays) and
2297 is always exactly equivalent to using explicit zero initializers.
2299 A metadata node is a structure-like constant with :ref:`metadata
2300 type <t_metadata>`. For example:
2301 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2302 constants that are meant to be interpreted as part of the
2303 instruction stream, metadata is a place to attach additional
2304 information such as debug info.
2306 Global Variable and Function Addresses
2307 --------------------------------------
2309 The addresses of :ref:`global variables <globalvars>` and
2310 :ref:`functions <functionstructure>` are always implicitly valid
2311 (link-time) constants. These constants are explicitly referenced when
2312 the :ref:`identifier for the global <identifiers>` is used and always have
2313 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2316 .. code-block:: llvm
2320 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2327 The string '``undef``' can be used anywhere a constant is expected, and
2328 indicates that the user of the value may receive an unspecified
2329 bit-pattern. Undefined values may be of any type (other than '``label``'
2330 or '``void``') and be used anywhere a constant is permitted.
2332 Undefined values are useful because they indicate to the compiler that
2333 the program is well defined no matter what value is used. This gives the
2334 compiler more freedom to optimize. Here are some examples of
2335 (potentially surprising) transformations that are valid (in pseudo IR):
2337 .. code-block:: llvm
2347 This is safe because all of the output bits are affected by the undef
2348 bits. Any output bit can have a zero or one depending on the input bits.
2350 .. code-block:: llvm
2361 These logical operations have bits that are not always affected by the
2362 input. For example, if ``%X`` has a zero bit, then the output of the
2363 '``and``' operation will always be a zero for that bit, no matter what
2364 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2365 optimize or assume that the result of the '``and``' is '``undef``'.
2366 However, it is safe to assume that all bits of the '``undef``' could be
2367 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2368 all the bits of the '``undef``' operand to the '``or``' could be set,
2369 allowing the '``or``' to be folded to -1.
2371 .. code-block:: llvm
2373 %A = select undef, %X, %Y
2374 %B = select undef, 42, %Y
2375 %C = select %X, %Y, undef
2385 This set of examples shows that undefined '``select``' (and conditional
2386 branch) conditions can go *either way*, but they have to come from one
2387 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2388 both known to have a clear low bit, then ``%A`` would have to have a
2389 cleared low bit. However, in the ``%C`` example, the optimizer is
2390 allowed to assume that the '``undef``' operand could be the same as
2391 ``%Y``, allowing the whole '``select``' to be eliminated.
2393 .. code-block:: llvm
2395 %A = xor undef, undef
2412 This example points out that two '``undef``' operands are not
2413 necessarily the same. This can be surprising to people (and also matches
2414 C semantics) where they assume that "``X^X``" is always zero, even if
2415 ``X`` is undefined. This isn't true for a number of reasons, but the
2416 short answer is that an '``undef``' "variable" can arbitrarily change
2417 its value over its "live range". This is true because the variable
2418 doesn't actually *have a live range*. Instead, the value is logically
2419 read from arbitrary registers that happen to be around when needed, so
2420 the value is not necessarily consistent over time. In fact, ``%A`` and
2421 ``%C`` need to have the same semantics or the core LLVM "replace all
2422 uses with" concept would not hold.
2424 .. code-block:: llvm
2432 These examples show the crucial difference between an *undefined value*
2433 and *undefined behavior*. An undefined value (like '``undef``') is
2434 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2435 operation can be constant folded to '``undef``', because the '``undef``'
2436 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2437 However, in the second example, we can make a more aggressive
2438 assumption: because the ``undef`` is allowed to be an arbitrary value,
2439 we are allowed to assume that it could be zero. Since a divide by zero
2440 has *undefined behavior*, we are allowed to assume that the operation
2441 does not execute at all. This allows us to delete the divide and all
2442 code after it. Because the undefined operation "can't happen", the
2443 optimizer can assume that it occurs in dead code.
2445 .. code-block:: llvm
2447 a: store undef -> %X
2448 b: store %X -> undef
2453 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2454 value can be assumed to not have any effect; we can assume that the
2455 value is overwritten with bits that happen to match what was already
2456 there. However, a store *to* an undefined location could clobber
2457 arbitrary memory, therefore, it has undefined behavior.
2464 Poison values are similar to :ref:`undef values <undefvalues>`, however
2465 they also represent the fact that an instruction or constant expression
2466 that cannot evoke side effects has nevertheless detected a condition
2467 that results in undefined behavior.
2469 There is currently no way of representing a poison value in the IR; they
2470 only exist when produced by operations such as :ref:`add <i_add>` with
2473 Poison value behavior is defined in terms of value *dependence*:
2475 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2476 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2477 their dynamic predecessor basic block.
2478 - Function arguments depend on the corresponding actual argument values
2479 in the dynamic callers of their functions.
2480 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2481 instructions that dynamically transfer control back to them.
2482 - :ref:`Invoke <i_invoke>` instructions depend on the
2483 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2484 call instructions that dynamically transfer control back to them.
2485 - Non-volatile loads and stores depend on the most recent stores to all
2486 of the referenced memory addresses, following the order in the IR
2487 (including loads and stores implied by intrinsics such as
2488 :ref:`@llvm.memcpy <int_memcpy>`.)
2489 - An instruction with externally visible side effects depends on the
2490 most recent preceding instruction with externally visible side
2491 effects, following the order in the IR. (This includes :ref:`volatile
2492 operations <volatile>`.)
2493 - An instruction *control-depends* on a :ref:`terminator
2494 instruction <terminators>` if the terminator instruction has
2495 multiple successors and the instruction is always executed when
2496 control transfers to one of the successors, and may not be executed
2497 when control is transferred to another.
2498 - Additionally, an instruction also *control-depends* on a terminator
2499 instruction if the set of instructions it otherwise depends on would
2500 be different if the terminator had transferred control to a different
2502 - Dependence is transitive.
2504 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2505 with the additional effect that any instruction that has a *dependence*
2506 on a poison value has undefined behavior.
2508 Here are some examples:
2510 .. code-block:: llvm
2513 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2514 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2515 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2516 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2518 store i32 %poison, i32* @g ; Poison value stored to memory.
2519 %poison2 = load i32* @g ; Poison value loaded back from memory.
2521 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2523 %narrowaddr = bitcast i32* @g to i16*
2524 %wideaddr = bitcast i32* @g to i64*
2525 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2526 %poison4 = load i64* %wideaddr ; Returns a poison value.
2528 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2529 br i1 %cmp, label %true, label %end ; Branch to either destination.
2532 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2533 ; it has undefined behavior.
2537 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2538 ; Both edges into this PHI are
2539 ; control-dependent on %cmp, so this
2540 ; always results in a poison value.
2542 store volatile i32 0, i32* @g ; This would depend on the store in %true
2543 ; if %cmp is true, or the store in %entry
2544 ; otherwise, so this is undefined behavior.
2546 br i1 %cmp, label %second_true, label %second_end
2547 ; The same branch again, but this time the
2548 ; true block doesn't have side effects.
2555 store volatile i32 0, i32* @g ; This time, the instruction always depends
2556 ; on the store in %end. Also, it is
2557 ; control-equivalent to %end, so this is
2558 ; well-defined (ignoring earlier undefined
2559 ; behavior in this example).
2563 Addresses of Basic Blocks
2564 -------------------------
2566 ``blockaddress(@function, %block)``
2568 The '``blockaddress``' constant computes the address of the specified
2569 basic block in the specified function, and always has an ``i8*`` type.
2570 Taking the address of the entry block is illegal.
2572 This value only has defined behavior when used as an operand to the
2573 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2574 against null. Pointer equality tests between labels addresses results in
2575 undefined behavior --- though, again, comparison against null is ok, and
2576 no label is equal to the null pointer. This may be passed around as an
2577 opaque pointer sized value as long as the bits are not inspected. This
2578 allows ``ptrtoint`` and arithmetic to be performed on these values so
2579 long as the original value is reconstituted before the ``indirectbr``
2582 Finally, some targets may provide defined semantics when using the value
2583 as the operand to an inline assembly, but that is target specific.
2587 Constant Expressions
2588 --------------------
2590 Constant expressions are used to allow expressions involving other
2591 constants to be used as constants. Constant expressions may be of any
2592 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2593 that does not have side effects (e.g. load and call are not supported).
2594 The following is the syntax for constant expressions:
2596 ``trunc (CST to TYPE)``
2597 Truncate a constant to another type. The bit size of CST must be
2598 larger than the bit size of TYPE. Both types must be integers.
2599 ``zext (CST to TYPE)``
2600 Zero extend a constant to another type. The bit size of CST must be
2601 smaller than the bit size of TYPE. Both types must be integers.
2602 ``sext (CST to TYPE)``
2603 Sign extend a constant to another type. The bit size of CST must be
2604 smaller than the bit size of TYPE. Both types must be integers.
2605 ``fptrunc (CST to TYPE)``
2606 Truncate a floating point constant to another floating point type.
2607 The size of CST must be larger than the size of TYPE. Both types
2608 must be floating point.
2609 ``fpext (CST to TYPE)``
2610 Floating point extend a constant to another type. The size of CST
2611 must be smaller or equal to the size of TYPE. Both types must be
2613 ``fptoui (CST to TYPE)``
2614 Convert a floating point constant to the corresponding unsigned
2615 integer constant. TYPE must be a scalar or vector integer type. CST
2616 must be of scalar or vector floating point type. Both CST and TYPE
2617 must be scalars, or vectors of the same number of elements. If the
2618 value won't fit in the integer type, the results are undefined.
2619 ``fptosi (CST to TYPE)``
2620 Convert a floating point constant to the corresponding signed
2621 integer constant. TYPE must be a scalar or vector integer type. CST
2622 must be of scalar or vector floating point type. Both CST and TYPE
2623 must be scalars, or vectors of the same number of elements. If the
2624 value won't fit in the integer type, the results are undefined.
2625 ``uitofp (CST to TYPE)``
2626 Convert an unsigned integer constant to the corresponding floating
2627 point constant. TYPE must be a scalar or vector floating point type.
2628 CST must be of scalar or vector integer type. Both CST and TYPE must
2629 be scalars, or vectors of the same number of elements. If the value
2630 won't fit in the floating point type, the results are undefined.
2631 ``sitofp (CST to TYPE)``
2632 Convert a signed integer constant to the corresponding floating
2633 point constant. TYPE must be a scalar or vector floating point type.
2634 CST must be of scalar or vector integer type. Both CST and TYPE must
2635 be scalars, or vectors of the same number of elements. If the value
2636 won't fit in the floating point type, the results are undefined.
2637 ``ptrtoint (CST to TYPE)``
2638 Convert a pointer typed constant to the corresponding integer
2639 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2640 pointer type. The ``CST`` value is zero extended, truncated, or
2641 unchanged to make it fit in ``TYPE``.
2642 ``inttoptr (CST to TYPE)``
2643 Convert an integer constant to a pointer constant. TYPE must be a
2644 pointer type. CST must be of integer type. The CST value is zero
2645 extended, truncated, or unchanged to make it fit in a pointer size.
2646 This one is *really* dangerous!
2647 ``bitcast (CST to TYPE)``
2648 Convert a constant, CST, to another TYPE. The constraints of the
2649 operands are the same as those for the :ref:`bitcast
2650 instruction <i_bitcast>`.
2651 ``addrspacecast (CST to TYPE)``
2652 Convert a constant pointer or constant vector of pointer, CST, to another
2653 TYPE in a different address space. The constraints of the operands are the
2654 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2655 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2656 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2657 constants. As with the :ref:`getelementptr <i_getelementptr>`
2658 instruction, the index list may have zero or more indexes, which are
2659 required to make sense for the type of "CSTPTR".
2660 ``select (COND, VAL1, VAL2)``
2661 Perform the :ref:`select operation <i_select>` on constants.
2662 ``icmp COND (VAL1, VAL2)``
2663 Performs the :ref:`icmp operation <i_icmp>` on constants.
2664 ``fcmp COND (VAL1, VAL2)``
2665 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2666 ``extractelement (VAL, IDX)``
2667 Perform the :ref:`extractelement operation <i_extractelement>` on
2669 ``insertelement (VAL, ELT, IDX)``
2670 Perform the :ref:`insertelement operation <i_insertelement>` on
2672 ``shufflevector (VEC1, VEC2, IDXMASK)``
2673 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2675 ``extractvalue (VAL, IDX0, IDX1, ...)``
2676 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2677 constants. The index list is interpreted in a similar manner as
2678 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2679 least one index value must be specified.
2680 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2681 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2682 The index list is interpreted in a similar manner as indices in a
2683 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2684 value must be specified.
2685 ``OPCODE (LHS, RHS)``
2686 Perform the specified operation of the LHS and RHS constants. OPCODE
2687 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2688 binary <bitwiseops>` operations. The constraints on operands are
2689 the same as those for the corresponding instruction (e.g. no bitwise
2690 operations on floating point values are allowed).
2697 Inline Assembler Expressions
2698 ----------------------------
2700 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2701 Inline Assembly <moduleasm>`) through the use of a special value. This
2702 value represents the inline assembler as a string (containing the
2703 instructions to emit), a list of operand constraints (stored as a
2704 string), a flag that indicates whether or not the inline asm expression
2705 has side effects, and a flag indicating whether the function containing
2706 the asm needs to align its stack conservatively. An example inline
2707 assembler expression is:
2709 .. code-block:: llvm
2711 i32 (i32) asm "bswap $0", "=r,r"
2713 Inline assembler expressions may **only** be used as the callee operand
2714 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2715 Thus, typically we have:
2717 .. code-block:: llvm
2719 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2721 Inline asms with side effects not visible in the constraint list must be
2722 marked as having side effects. This is done through the use of the
2723 '``sideeffect``' keyword, like so:
2725 .. code-block:: llvm
2727 call void asm sideeffect "eieio", ""()
2729 In some cases inline asms will contain code that will not work unless
2730 the stack is aligned in some way, such as calls or SSE instructions on
2731 x86, yet will not contain code that does that alignment within the asm.
2732 The compiler should make conservative assumptions about what the asm
2733 might contain and should generate its usual stack alignment code in the
2734 prologue if the '``alignstack``' keyword is present:
2736 .. code-block:: llvm
2738 call void asm alignstack "eieio", ""()
2740 Inline asms also support using non-standard assembly dialects. The
2741 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2742 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2743 the only supported dialects. An example is:
2745 .. code-block:: llvm
2747 call void asm inteldialect "eieio", ""()
2749 If multiple keywords appear the '``sideeffect``' keyword must come
2750 first, the '``alignstack``' keyword second and the '``inteldialect``'
2756 The call instructions that wrap inline asm nodes may have a
2757 "``!srcloc``" MDNode attached to it that contains a list of constant
2758 integers. If present, the code generator will use the integer as the
2759 location cookie value when report errors through the ``LLVMContext``
2760 error reporting mechanisms. This allows a front-end to correlate backend
2761 errors that occur with inline asm back to the source code that produced
2764 .. code-block:: llvm
2766 call void asm sideeffect "something bad", ""(), !srcloc !42
2768 !42 = !{ i32 1234567 }
2770 It is up to the front-end to make sense of the magic numbers it places
2771 in the IR. If the MDNode contains multiple constants, the code generator
2772 will use the one that corresponds to the line of the asm that the error
2777 Metadata Nodes and Metadata Strings
2778 -----------------------------------
2780 LLVM IR allows metadata to be attached to instructions in the program
2781 that can convey extra information about the code to the optimizers and
2782 code generator. One example application of metadata is source-level
2783 debug information. There are two metadata primitives: strings and nodes.
2784 All metadata has the ``metadata`` type and is identified in syntax by a
2785 preceding exclamation point ('``!``').
2787 A metadata string is a string surrounded by double quotes. It can
2788 contain any character by escaping non-printable characters with
2789 "``\xx``" where "``xx``" is the two digit hex code. For example:
2792 Metadata nodes are represented with notation similar to structure
2793 constants (a comma separated list of elements, surrounded by braces and
2794 preceded by an exclamation point). Metadata nodes can have any values as
2795 their operand. For example:
2797 .. code-block:: llvm
2799 !{ metadata !"test\00", i32 10}
2801 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2802 metadata nodes, which can be looked up in the module symbol table. For
2805 .. code-block:: llvm
2807 !foo = metadata !{!4, !3}
2809 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2810 function is using two metadata arguments:
2812 .. code-block:: llvm
2814 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2816 Metadata can be attached with an instruction. Here metadata ``!21`` is
2817 attached to the ``add`` instruction using the ``!dbg`` identifier:
2819 .. code-block:: llvm
2821 %indvar.next = add i64 %indvar, 1, !dbg !21
2823 More information about specific metadata nodes recognized by the
2824 optimizers and code generator is found below.
2829 In LLVM IR, memory does not have types, so LLVM's own type system is not
2830 suitable for doing TBAA. Instead, metadata is added to the IR to
2831 describe a type system of a higher level language. This can be used to
2832 implement typical C/C++ TBAA, but it can also be used to implement
2833 custom alias analysis behavior for other languages.
2835 The current metadata format is very simple. TBAA metadata nodes have up
2836 to three fields, e.g.:
2838 .. code-block:: llvm
2840 !0 = metadata !{ metadata !"an example type tree" }
2841 !1 = metadata !{ metadata !"int", metadata !0 }
2842 !2 = metadata !{ metadata !"float", metadata !0 }
2843 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2845 The first field is an identity field. It can be any value, usually a
2846 metadata string, which uniquely identifies the type. The most important
2847 name in the tree is the name of the root node. Two trees with different
2848 root node names are entirely disjoint, even if they have leaves with
2851 The second field identifies the type's parent node in the tree, or is
2852 null or omitted for a root node. A type is considered to alias all of
2853 its descendants and all of its ancestors in the tree. Also, a type is
2854 considered to alias all types in other trees, so that bitcode produced
2855 from multiple front-ends is handled conservatively.
2857 If the third field is present, it's an integer which if equal to 1
2858 indicates that the type is "constant" (meaning
2859 ``pointsToConstantMemory`` should return true; see `other useful
2860 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2862 '``tbaa.struct``' Metadata
2863 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2865 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2866 aggregate assignment operations in C and similar languages, however it
2867 is defined to copy a contiguous region of memory, which is more than
2868 strictly necessary for aggregate types which contain holes due to
2869 padding. Also, it doesn't contain any TBAA information about the fields
2872 ``!tbaa.struct`` metadata can describe which memory subregions in a
2873 memcpy are padding and what the TBAA tags of the struct are.
2875 The current metadata format is very simple. ``!tbaa.struct`` metadata
2876 nodes are a list of operands which are in conceptual groups of three.
2877 For each group of three, the first operand gives the byte offset of a
2878 field in bytes, the second gives its size in bytes, and the third gives
2881 .. code-block:: llvm
2883 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2885 This describes a struct with two fields. The first is at offset 0 bytes
2886 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2887 and has size 4 bytes and has tbaa tag !2.
2889 Note that the fields need not be contiguous. In this example, there is a
2890 4 byte gap between the two fields. This gap represents padding which
2891 does not carry useful data and need not be preserved.
2893 '``noalias``' and '``alias.scope``' Metadata
2894 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2896 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
2897 noalias memory-access sets. This means that some collection of memory access
2898 instructions (loads, stores, memory-accessing calls, etc.) that carry
2899 ``noalias`` metadata can specifically be specified not to alias with some other
2900 collection of memory access instructions that carry ``alias.scope`` metadata.
2901 Each type of metadata specifies a list of scopes where each scope has an id and
2902 a domain. When evaluating an aliasing query, if for some some domain, the set
2903 of scopes with that domain in one instruction's ``alias.scope`` list is a
2904 subset of (or qual to) the set of scopes for that domain in another
2905 instruction's ``noalias`` list, then the two memory accesses are assumed not to
2908 The metadata identifying each domain is itself a list containing one or two
2909 entries. The first entry is the name of the domain. Note that if the name is a
2910 string then it can be combined accross functions and translation units. A
2911 self-reference can be used to create globally unique domain names. A
2912 descriptive string may optionally be provided as a second list entry.
2914 The metadata identifying each scope is also itself a list containing two or
2915 three entries. The first entry is the name of the scope. Note that if the name
2916 is a string then it can be combined accross functions and translation units. A
2917 self-reference can be used to create globally unique scope names. A metadata
2918 reference to the scope's domain is the second entry. A descriptive string may
2919 optionally be provided as a third list entry.
2923 .. code-block:: llvm
2925 ; Two scope domains:
2926 !0 = metadata !{metadata !0}
2927 !1 = metadata !{metadata !1}
2929 ; Some scopes in these domains:
2930 !2 = metadata !{metadata !2, metadata !0}
2931 !3 = metadata !{metadata !3, metadata !0}
2932 !4 = metadata !{metadata !4, metadata !1}
2935 !5 = metadata !{metadata !4} ; A list containing only scope !4
2936 !6 = metadata !{metadata !4, metadata !3, metadata !2}
2937 !7 = metadata !{metadata !3}
2939 ; These two instructions don't alias:
2940 %0 = load float* %c, align 4, !alias.scope !5
2941 store float %0, float* %arrayidx.i, align 4, !noalias !5
2943 ; These two instructions also don't alias (for domain !1, the set of scopes
2944 ; in the !alias.scope equals that in the !noalias list):
2945 %2 = load float* %c, align 4, !alias.scope !5
2946 store float %2, float* %arrayidx.i2, align 4, !noalias !6
2948 ; These two instructions don't alias (for domain !0, the set of scopes in
2949 ; the !noalias list is not a superset of, or equal to, the scopes in the
2950 ; !alias.scope list):
2951 %2 = load float* %c, align 4, !alias.scope !6
2952 store float %0, float* %arrayidx.i, align 4, !noalias !7
2954 '``fpmath``' Metadata
2955 ^^^^^^^^^^^^^^^^^^^^^
2957 ``fpmath`` metadata may be attached to any instruction of floating point
2958 type. It can be used to express the maximum acceptable error in the
2959 result of that instruction, in ULPs, thus potentially allowing the
2960 compiler to use a more efficient but less accurate method of computing
2961 it. ULP is defined as follows:
2963 If ``x`` is a real number that lies between two finite consecutive
2964 floating-point numbers ``a`` and ``b``, without being equal to one
2965 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2966 distance between the two non-equal finite floating-point numbers
2967 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2969 The metadata node shall consist of a single positive floating point
2970 number representing the maximum relative error, for example:
2972 .. code-block:: llvm
2974 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2976 '``range``' Metadata
2977 ^^^^^^^^^^^^^^^^^^^^
2979 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
2980 integer types. It expresses the possible ranges the loaded value or the value
2981 returned by the called function at this call site is in. The ranges are
2982 represented with a flattened list of integers. The loaded value or the value
2983 returned is known to be in the union of the ranges defined by each consecutive
2984 pair. Each pair has the following properties:
2986 - The type must match the type loaded by the instruction.
2987 - The pair ``a,b`` represents the range ``[a,b)``.
2988 - Both ``a`` and ``b`` are constants.
2989 - The range is allowed to wrap.
2990 - The range should not represent the full or empty set. That is,
2993 In addition, the pairs must be in signed order of the lower bound and
2994 they must be non-contiguous.
2998 .. code-block:: llvm
3000 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3001 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3002 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
3003 %d = invoke i8 @bar() to label %cont
3004 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
3006 !0 = metadata !{ i8 0, i8 2 }
3007 !1 = metadata !{ i8 255, i8 2 }
3008 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
3009 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
3014 It is sometimes useful to attach information to loop constructs. Currently,
3015 loop metadata is implemented as metadata attached to the branch instruction
3016 in the loop latch block. This type of metadata refer to a metadata node that is
3017 guaranteed to be separate for each loop. The loop identifier metadata is
3018 specified with the name ``llvm.loop``.
3020 The loop identifier metadata is implemented using a metadata that refers to
3021 itself to avoid merging it with any other identifier metadata, e.g.,
3022 during module linkage or function inlining. That is, each loop should refer
3023 to their own identification metadata even if they reside in separate functions.
3024 The following example contains loop identifier metadata for two separate loop
3027 .. code-block:: llvm
3029 !0 = metadata !{ metadata !0 }
3030 !1 = metadata !{ metadata !1 }
3032 The loop identifier metadata can be used to specify additional
3033 per-loop metadata. Any operands after the first operand can be treated
3034 as user-defined metadata. For example the ``llvm.loop.unroll.count``
3035 suggests an unroll factor to the loop unroller:
3037 .. code-block:: llvm
3039 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
3041 !0 = metadata !{ metadata !0, metadata !1 }
3042 !1 = metadata !{ metadata !"llvm.loop.unroll.count", i32 4 }
3044 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
3045 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3047 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
3048 used to control per-loop vectorization and interleaving parameters such as
3049 vectorization width and interleave count. These metadata should be used in
3050 conjunction with ``llvm.loop`` loop identification metadata. The
3051 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
3052 optimization hints and the optimizer will only interleave and vectorize loops if
3053 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
3054 which contains information about loop-carried memory dependencies can be helpful
3055 in determining the safety of these transformations.
3057 '``llvm.loop.interleave.count``' Metadata
3058 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3060 This metadata suggests an interleave count to the loop interleaver.
3061 The first operand is the string ``llvm.loop.interleave.count`` and the
3062 second operand is an integer specifying the interleave count. For
3065 .. code-block:: llvm
3067 !0 = metadata !{ metadata !"llvm.loop.interleave.count", i32 4 }
3069 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
3070 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
3071 then the interleave count will be determined automatically.
3073 '``llvm.loop.vectorize.enable``' Metadata
3074 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3076 This metadata selectively enables or disables vectorization for the loop. The
3077 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
3078 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
3079 0 disables vectorization:
3081 .. code-block:: llvm
3083 !0 = metadata !{ metadata !"llvm.loop.vectorize.enable", i1 0 }
3084 !1 = metadata !{ metadata !"llvm.loop.vectorize.enable", i1 1 }
3086 '``llvm.loop.vectorize.width``' Metadata
3087 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3089 This metadata sets the target width of the vectorizer. The first
3090 operand is the string ``llvm.loop.vectorize.width`` and the second
3091 operand is an integer specifying the width. For example:
3093 .. code-block:: llvm
3095 !0 = metadata !{ metadata !"llvm.loop.vectorize.width", i32 4 }
3097 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
3098 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
3099 0 or if the loop does not have this metadata the width will be
3100 determined automatically.
3102 '``llvm.loop.unroll``'
3103 ^^^^^^^^^^^^^^^^^^^^^^
3105 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
3106 optimization hints such as the unroll factor. ``llvm.loop.unroll``
3107 metadata should be used in conjunction with ``llvm.loop`` loop
3108 identification metadata. The ``llvm.loop.unroll`` metadata are only
3109 optimization hints and the unrolling will only be performed if the
3110 optimizer believes it is safe to do so.
3112 '``llvm.loop.unroll.count``' Metadata
3113 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3115 This metadata suggests an unroll factor to the loop unroller. The
3116 first operand is the string ``llvm.loop.unroll.count`` and the second
3117 operand is a positive integer specifying the unroll factor. For
3120 .. code-block:: llvm
3122 !0 = metadata !{ metadata !"llvm.loop.unroll.count", i32 4 }
3124 If the trip count of the loop is less than the unroll count the loop
3125 will be partially unrolled.
3127 '``llvm.loop.unroll.disable``' Metadata
3128 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3130 This metadata either disables loop unrolling. The metadata has a single operand
3131 which is the string ``llvm.loop.unroll.disable``. For example:
3133 .. code-block:: llvm
3135 !0 = metadata !{ metadata !"llvm.loop.unroll.disable" }
3137 '``llvm.loop.unroll.full``' Metadata
3138 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3140 This metadata either suggests that the loop should be unrolled fully. The
3141 metadata has a single operand which is the string ``llvm.loop.unroll.disable``.
3144 .. code-block:: llvm
3146 !0 = metadata !{ metadata !"llvm.loop.unroll.full" }
3151 Metadata types used to annotate memory accesses with information helpful
3152 for optimizations are prefixed with ``llvm.mem``.
3154 '``llvm.mem.parallel_loop_access``' Metadata
3155 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3157 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
3158 or metadata containing a list of loop identifiers for nested loops.
3159 The metadata is attached to memory accessing instructions and denotes that
3160 no loop carried memory dependence exist between it and other instructions denoted
3161 with the same loop identifier.
3163 Precisely, given two instructions ``m1`` and ``m2`` that both have the
3164 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
3165 set of loops associated with that metadata, respectively, then there is no loop
3166 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
3169 As a special case, if all memory accessing instructions in a loop have
3170 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
3171 loop has no loop carried memory dependences and is considered to be a parallel
3174 Note that if not all memory access instructions have such metadata referring to
3175 the loop, then the loop is considered not being trivially parallel. Additional
3176 memory dependence analysis is required to make that determination. As a fail
3177 safe mechanism, this causes loops that were originally parallel to be considered
3178 sequential (if optimization passes that are unaware of the parallel semantics
3179 insert new memory instructions into the loop body).
3181 Example of a loop that is considered parallel due to its correct use of
3182 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
3183 metadata types that refer to the same loop identifier metadata.
3185 .. code-block:: llvm
3189 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
3191 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3193 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
3197 !0 = metadata !{ metadata !0 }
3199 It is also possible to have nested parallel loops. In that case the
3200 memory accesses refer to a list of loop identifier metadata nodes instead of
3201 the loop identifier metadata node directly:
3203 .. code-block:: llvm
3207 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
3209 br label %inner.for.body
3213 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3215 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
3217 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
3221 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
3223 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
3225 outer.for.end: ; preds = %for.body
3227 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
3228 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
3229 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
3231 Module Flags Metadata
3232 =====================
3234 Information about the module as a whole is difficult to convey to LLVM's
3235 subsystems. The LLVM IR isn't sufficient to transmit this information.
3236 The ``llvm.module.flags`` named metadata exists in order to facilitate
3237 this. These flags are in the form of key / value pairs --- much like a
3238 dictionary --- making it easy for any subsystem who cares about a flag to
3241 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
3242 Each triplet has the following form:
3244 - The first element is a *behavior* flag, which specifies the behavior
3245 when two (or more) modules are merged together, and it encounters two
3246 (or more) metadata with the same ID. The supported behaviors are
3248 - The second element is a metadata string that is a unique ID for the
3249 metadata. Each module may only have one flag entry for each unique ID (not
3250 including entries with the **Require** behavior).
3251 - The third element is the value of the flag.
3253 When two (or more) modules are merged together, the resulting
3254 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
3255 each unique metadata ID string, there will be exactly one entry in the merged
3256 modules ``llvm.module.flags`` metadata table, and the value for that entry will
3257 be determined by the merge behavior flag, as described below. The only exception
3258 is that entries with the *Require* behavior are always preserved.
3260 The following behaviors are supported:
3271 Emits an error if two values disagree, otherwise the resulting value
3272 is that of the operands.
3276 Emits a warning if two values disagree. The result value will be the
3277 operand for the flag from the first module being linked.
3281 Adds a requirement that another module flag be present and have a
3282 specified value after linking is performed. The value must be a
3283 metadata pair, where the first element of the pair is the ID of the
3284 module flag to be restricted, and the second element of the pair is
3285 the value the module flag should be restricted to. This behavior can
3286 be used to restrict the allowable results (via triggering of an
3287 error) of linking IDs with the **Override** behavior.
3291 Uses the specified value, regardless of the behavior or value of the
3292 other module. If both modules specify **Override**, but the values
3293 differ, an error will be emitted.
3297 Appends the two values, which are required to be metadata nodes.
3301 Appends the two values, which are required to be metadata
3302 nodes. However, duplicate entries in the second list are dropped
3303 during the append operation.
3305 It is an error for a particular unique flag ID to have multiple behaviors,
3306 except in the case of **Require** (which adds restrictions on another metadata
3307 value) or **Override**.
3309 An example of module flags:
3311 .. code-block:: llvm
3313 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3314 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3315 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3316 !3 = metadata !{ i32 3, metadata !"qux",
3318 metadata !"foo", i32 1
3321 !llvm.module.flags = !{ !0, !1, !2, !3 }
3323 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3324 if two or more ``!"foo"`` flags are seen is to emit an error if their
3325 values are not equal.
3327 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3328 behavior if two or more ``!"bar"`` flags are seen is to use the value
3331 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3332 behavior if two or more ``!"qux"`` flags are seen is to emit a
3333 warning if their values are not equal.
3335 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3339 metadata !{ metadata !"foo", i32 1 }
3341 The behavior is to emit an error if the ``llvm.module.flags`` does not
3342 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3345 Objective-C Garbage Collection Module Flags Metadata
3346 ----------------------------------------------------
3348 On the Mach-O platform, Objective-C stores metadata about garbage
3349 collection in a special section called "image info". The metadata
3350 consists of a version number and a bitmask specifying what types of
3351 garbage collection are supported (if any) by the file. If two or more
3352 modules are linked together their garbage collection metadata needs to
3353 be merged rather than appended together.
3355 The Objective-C garbage collection module flags metadata consists of the
3356 following key-value pairs:
3365 * - ``Objective-C Version``
3366 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3368 * - ``Objective-C Image Info Version``
3369 - **[Required]** --- The version of the image info section. Currently
3372 * - ``Objective-C Image Info Section``
3373 - **[Required]** --- The section to place the metadata. Valid values are
3374 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3375 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3376 Objective-C ABI version 2.
3378 * - ``Objective-C Garbage Collection``
3379 - **[Required]** --- Specifies whether garbage collection is supported or
3380 not. Valid values are 0, for no garbage collection, and 2, for garbage
3381 collection supported.
3383 * - ``Objective-C GC Only``
3384 - **[Optional]** --- Specifies that only garbage collection is supported.
3385 If present, its value must be 6. This flag requires that the
3386 ``Objective-C Garbage Collection`` flag have the value 2.
3388 Some important flag interactions:
3390 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3391 merged with a module with ``Objective-C Garbage Collection`` set to
3392 2, then the resulting module has the
3393 ``Objective-C Garbage Collection`` flag set to 0.
3394 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3395 merged with a module with ``Objective-C GC Only`` set to 6.
3397 Automatic Linker Flags Module Flags Metadata
3398 --------------------------------------------
3400 Some targets support embedding flags to the linker inside individual object
3401 files. Typically this is used in conjunction with language extensions which
3402 allow source files to explicitly declare the libraries they depend on, and have
3403 these automatically be transmitted to the linker via object files.
3405 These flags are encoded in the IR using metadata in the module flags section,
3406 using the ``Linker Options`` key. The merge behavior for this flag is required
3407 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3408 node which should be a list of other metadata nodes, each of which should be a
3409 list of metadata strings defining linker options.
3411 For example, the following metadata section specifies two separate sets of
3412 linker options, presumably to link against ``libz`` and the ``Cocoa``
3415 !0 = metadata !{ i32 6, metadata !"Linker Options",
3417 metadata !{ metadata !"-lz" },
3418 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3419 !llvm.module.flags = !{ !0 }
3421 The metadata encoding as lists of lists of options, as opposed to a collapsed
3422 list of options, is chosen so that the IR encoding can use multiple option
3423 strings to specify e.g., a single library, while still having that specifier be
3424 preserved as an atomic element that can be recognized by a target specific
3425 assembly writer or object file emitter.
3427 Each individual option is required to be either a valid option for the target's
3428 linker, or an option that is reserved by the target specific assembly writer or
3429 object file emitter. No other aspect of these options is defined by the IR.
3431 C type width Module Flags Metadata
3432 ----------------------------------
3434 The ARM backend emits a section into each generated object file describing the
3435 options that it was compiled with (in a compiler-independent way) to prevent
3436 linking incompatible objects, and to allow automatic library selection. Some
3437 of these options are not visible at the IR level, namely wchar_t width and enum
3440 To pass this information to the backend, these options are encoded in module
3441 flags metadata, using the following key-value pairs:
3451 - * 0 --- sizeof(wchar_t) == 4
3452 * 1 --- sizeof(wchar_t) == 2
3455 - * 0 --- Enums are at least as large as an ``int``.
3456 * 1 --- Enums are stored in the smallest integer type which can
3457 represent all of its values.
3459 For example, the following metadata section specifies that the module was
3460 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
3461 enum is the smallest type which can represent all of its values::
3463 !llvm.module.flags = !{!0, !1}
3464 !0 = metadata !{i32 1, metadata !"short_wchar", i32 1}
3465 !1 = metadata !{i32 1, metadata !"short_enum", i32 0}
3467 .. _intrinsicglobalvariables:
3469 Intrinsic Global Variables
3470 ==========================
3472 LLVM has a number of "magic" global variables that contain data that
3473 affect code generation or other IR semantics. These are documented here.
3474 All globals of this sort should have a section specified as
3475 "``llvm.metadata``". This section and all globals that start with
3476 "``llvm.``" are reserved for use by LLVM.
3480 The '``llvm.used``' Global Variable
3481 -----------------------------------
3483 The ``@llvm.used`` global is an array which has
3484 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3485 pointers to named global variables, functions and aliases which may optionally
3486 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3489 .. code-block:: llvm
3494 @llvm.used = appending global [2 x i8*] [
3496 i8* bitcast (i32* @Y to i8*)
3497 ], section "llvm.metadata"
3499 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3500 and linker are required to treat the symbol as if there is a reference to the
3501 symbol that it cannot see (which is why they have to be named). For example, if
3502 a variable has internal linkage and no references other than that from the
3503 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3504 references from inline asms and other things the compiler cannot "see", and
3505 corresponds to "``attribute((used))``" in GNU C.
3507 On some targets, the code generator must emit a directive to the
3508 assembler or object file to prevent the assembler and linker from
3509 molesting the symbol.
3511 .. _gv_llvmcompilerused:
3513 The '``llvm.compiler.used``' Global Variable
3514 --------------------------------------------
3516 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3517 directive, except that it only prevents the compiler from touching the
3518 symbol. On targets that support it, this allows an intelligent linker to
3519 optimize references to the symbol without being impeded as it would be
3522 This is a rare construct that should only be used in rare circumstances,
3523 and should not be exposed to source languages.
3525 .. _gv_llvmglobalctors:
3527 The '``llvm.global_ctors``' Global Variable
3528 -------------------------------------------
3530 .. code-block:: llvm
3532 %0 = type { i32, void ()*, i8* }
3533 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
3535 The ``@llvm.global_ctors`` array contains a list of constructor
3536 functions, priorities, and an optional associated global or function.
3537 The functions referenced by this array will be called in ascending order
3538 of priority (i.e. lowest first) when the module is loaded. The order of
3539 functions with the same priority is not defined.
3541 If the third field is present, non-null, and points to a global variable
3542 or function, the initializer function will only run if the associated
3543 data from the current module is not discarded.
3545 .. _llvmglobaldtors:
3547 The '``llvm.global_dtors``' Global Variable
3548 -------------------------------------------
3550 .. code-block:: llvm
3552 %0 = type { i32, void ()*, i8* }
3553 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
3555 The ``@llvm.global_dtors`` array contains a list of destructor
3556 functions, priorities, and an optional associated global or function.
3557 The functions referenced by this array will be called in descending
3558 order of priority (i.e. highest first) when the module is unloaded. The
3559 order of functions with the same priority is not defined.
3561 If the third field is present, non-null, and points to a global variable
3562 or function, the destructor function will only run if the associated
3563 data from the current module is not discarded.
3565 Instruction Reference
3566 =====================
3568 The LLVM instruction set consists of several different classifications
3569 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3570 instructions <binaryops>`, :ref:`bitwise binary
3571 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3572 :ref:`other instructions <otherops>`.
3576 Terminator Instructions
3577 -----------------------
3579 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3580 program ends with a "Terminator" instruction, which indicates which
3581 block should be executed after the current block is finished. These
3582 terminator instructions typically yield a '``void``' value: they produce
3583 control flow, not values (the one exception being the
3584 ':ref:`invoke <i_invoke>`' instruction).
3586 The terminator instructions are: ':ref:`ret <i_ret>`',
3587 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3588 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3589 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3593 '``ret``' Instruction
3594 ^^^^^^^^^^^^^^^^^^^^^
3601 ret <type> <value> ; Return a value from a non-void function
3602 ret void ; Return from void function
3607 The '``ret``' instruction is used to return control flow (and optionally
3608 a value) from a function back to the caller.
3610 There are two forms of the '``ret``' instruction: one that returns a
3611 value and then causes control flow, and one that just causes control
3617 The '``ret``' instruction optionally accepts a single argument, the
3618 return value. The type of the return value must be a ':ref:`first
3619 class <t_firstclass>`' type.
3621 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3622 return type and contains a '``ret``' instruction with no return value or
3623 a return value with a type that does not match its type, or if it has a
3624 void return type and contains a '``ret``' instruction with a return
3630 When the '``ret``' instruction is executed, control flow returns back to
3631 the calling function's context. If the caller is a
3632 ":ref:`call <i_call>`" instruction, execution continues at the
3633 instruction after the call. If the caller was an
3634 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3635 beginning of the "normal" destination block. If the instruction returns
3636 a value, that value shall set the call or invoke instruction's return
3642 .. code-block:: llvm
3644 ret i32 5 ; Return an integer value of 5
3645 ret void ; Return from a void function
3646 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3650 '``br``' Instruction
3651 ^^^^^^^^^^^^^^^^^^^^
3658 br i1 <cond>, label <iftrue>, label <iffalse>
3659 br label <dest> ; Unconditional branch
3664 The '``br``' instruction is used to cause control flow to transfer to a
3665 different basic block in the current function. There are two forms of
3666 this instruction, corresponding to a conditional branch and an
3667 unconditional branch.
3672 The conditional branch form of the '``br``' instruction takes a single
3673 '``i1``' value and two '``label``' values. The unconditional form of the
3674 '``br``' instruction takes a single '``label``' value as a target.
3679 Upon execution of a conditional '``br``' instruction, the '``i1``'
3680 argument is evaluated. If the value is ``true``, control flows to the
3681 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3682 to the '``iffalse``' ``label`` argument.
3687 .. code-block:: llvm
3690 %cond = icmp eq i32 %a, %b
3691 br i1 %cond, label %IfEqual, label %IfUnequal
3699 '``switch``' Instruction
3700 ^^^^^^^^^^^^^^^^^^^^^^^^
3707 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3712 The '``switch``' instruction is used to transfer control flow to one of
3713 several different places. It is a generalization of the '``br``'
3714 instruction, allowing a branch to occur to one of many possible
3720 The '``switch``' instruction uses three parameters: an integer
3721 comparison value '``value``', a default '``label``' destination, and an
3722 array of pairs of comparison value constants and '``label``'s. The table
3723 is not allowed to contain duplicate constant entries.
3728 The ``switch`` instruction specifies a table of values and destinations.
3729 When the '``switch``' instruction is executed, this table is searched
3730 for the given value. If the value is found, control flow is transferred
3731 to the corresponding destination; otherwise, control flow is transferred
3732 to the default destination.
3737 Depending on properties of the target machine and the particular
3738 ``switch`` instruction, this instruction may be code generated in
3739 different ways. For example, it could be generated as a series of
3740 chained conditional branches or with a lookup table.
3745 .. code-block:: llvm
3747 ; Emulate a conditional br instruction
3748 %Val = zext i1 %value to i32
3749 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3751 ; Emulate an unconditional br instruction
3752 switch i32 0, label %dest [ ]
3754 ; Implement a jump table:
3755 switch i32 %val, label %otherwise [ i32 0, label %onzero
3757 i32 2, label %ontwo ]
3761 '``indirectbr``' Instruction
3762 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3769 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3774 The '``indirectbr``' instruction implements an indirect branch to a
3775 label within the current function, whose address is specified by
3776 "``address``". Address must be derived from a
3777 :ref:`blockaddress <blockaddress>` constant.
3782 The '``address``' argument is the address of the label to jump to. The
3783 rest of the arguments indicate the full set of possible destinations
3784 that the address may point to. Blocks are allowed to occur multiple
3785 times in the destination list, though this isn't particularly useful.
3787 This destination list is required so that dataflow analysis has an
3788 accurate understanding of the CFG.
3793 Control transfers to the block specified in the address argument. All
3794 possible destination blocks must be listed in the label list, otherwise
3795 this instruction has undefined behavior. This implies that jumps to
3796 labels defined in other functions have undefined behavior as well.
3801 This is typically implemented with a jump through a register.
3806 .. code-block:: llvm
3808 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3812 '``invoke``' Instruction
3813 ^^^^^^^^^^^^^^^^^^^^^^^^
3820 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3821 to label <normal label> unwind label <exception label>
3826 The '``invoke``' instruction causes control to transfer to a specified
3827 function, with the possibility of control flow transfer to either the
3828 '``normal``' label or the '``exception``' label. If the callee function
3829 returns with the "``ret``" instruction, control flow will return to the
3830 "normal" label. If the callee (or any indirect callees) returns via the
3831 ":ref:`resume <i_resume>`" instruction or other exception handling
3832 mechanism, control is interrupted and continued at the dynamically
3833 nearest "exception" label.
3835 The '``exception``' label is a `landing
3836 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3837 '``exception``' label is required to have the
3838 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3839 information about the behavior of the program after unwinding happens,
3840 as its first non-PHI instruction. The restrictions on the
3841 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3842 instruction, so that the important information contained within the
3843 "``landingpad``" instruction can't be lost through normal code motion.
3848 This instruction requires several arguments:
3850 #. The optional "cconv" marker indicates which :ref:`calling
3851 convention <callingconv>` the call should use. If none is
3852 specified, the call defaults to using C calling conventions.
3853 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3854 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3856 #. '``ptr to function ty``': shall be the signature of the pointer to
3857 function value being invoked. In most cases, this is a direct
3858 function invocation, but indirect ``invoke``'s are just as possible,
3859 branching off an arbitrary pointer to function value.
3860 #. '``function ptr val``': An LLVM value containing a pointer to a
3861 function to be invoked.
3862 #. '``function args``': argument list whose types match the function
3863 signature argument types and parameter attributes. All arguments must
3864 be of :ref:`first class <t_firstclass>` type. If the function signature
3865 indicates the function accepts a variable number of arguments, the
3866 extra arguments can be specified.
3867 #. '``normal label``': the label reached when the called function
3868 executes a '``ret``' instruction.
3869 #. '``exception label``': the label reached when a callee returns via
3870 the :ref:`resume <i_resume>` instruction or other exception handling
3872 #. The optional :ref:`function attributes <fnattrs>` list. Only
3873 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3874 attributes are valid here.
3879 This instruction is designed to operate as a standard '``call``'
3880 instruction in most regards. The primary difference is that it
3881 establishes an association with a label, which is used by the runtime
3882 library to unwind the stack.
3884 This instruction is used in languages with destructors to ensure that
3885 proper cleanup is performed in the case of either a ``longjmp`` or a
3886 thrown exception. Additionally, this is important for implementation of
3887 '``catch``' clauses in high-level languages that support them.
3889 For the purposes of the SSA form, the definition of the value returned
3890 by the '``invoke``' instruction is deemed to occur on the edge from the
3891 current block to the "normal" label. If the callee unwinds then no
3892 return value is available.
3897 .. code-block:: llvm
3899 %retval = invoke i32 @Test(i32 15) to label %Continue
3900 unwind label %TestCleanup ; i32:retval set
3901 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3902 unwind label %TestCleanup ; i32:retval set
3906 '``resume``' Instruction
3907 ^^^^^^^^^^^^^^^^^^^^^^^^
3914 resume <type> <value>
3919 The '``resume``' instruction is a terminator instruction that has no
3925 The '``resume``' instruction requires one argument, which must have the
3926 same type as the result of any '``landingpad``' instruction in the same
3932 The '``resume``' instruction resumes propagation of an existing
3933 (in-flight) exception whose unwinding was interrupted with a
3934 :ref:`landingpad <i_landingpad>` instruction.
3939 .. code-block:: llvm
3941 resume { i8*, i32 } %exn
3945 '``unreachable``' Instruction
3946 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3958 The '``unreachable``' instruction has no defined semantics. This
3959 instruction is used to inform the optimizer that a particular portion of
3960 the code is not reachable. This can be used to indicate that the code
3961 after a no-return function cannot be reached, and other facts.
3966 The '``unreachable``' instruction has no defined semantics.
3973 Binary operators are used to do most of the computation in a program.
3974 They require two operands of the same type, execute an operation on
3975 them, and produce a single value. The operands might represent multiple
3976 data, as is the case with the :ref:`vector <t_vector>` data type. The
3977 result value has the same type as its operands.
3979 There are several different binary operators:
3983 '``add``' Instruction
3984 ^^^^^^^^^^^^^^^^^^^^^
3991 <result> = add <ty> <op1>, <op2> ; yields ty:result
3992 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
3993 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
3994 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
3999 The '``add``' instruction returns the sum of its two operands.
4004 The two arguments to the '``add``' instruction must be
4005 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4006 arguments must have identical types.
4011 The value produced is the integer sum of the two operands.
4013 If the sum has unsigned overflow, the result returned is the
4014 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4017 Because LLVM integers use a two's complement representation, this
4018 instruction is appropriate for both signed and unsigned integers.
4020 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4021 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4022 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
4023 unsigned and/or signed overflow, respectively, occurs.
4028 .. code-block:: llvm
4030 <result> = add i32 4, %var ; yields i32:result = 4 + %var
4034 '``fadd``' Instruction
4035 ^^^^^^^^^^^^^^^^^^^^^^
4042 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4047 The '``fadd``' instruction returns the sum of its two operands.
4052 The two arguments to the '``fadd``' instruction must be :ref:`floating
4053 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4054 Both arguments must have identical types.
4059 The value produced is the floating point sum of the two operands. This
4060 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
4061 which are optimization hints to enable otherwise unsafe floating point
4067 .. code-block:: llvm
4069 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
4071 '``sub``' Instruction
4072 ^^^^^^^^^^^^^^^^^^^^^
4079 <result> = sub <ty> <op1>, <op2> ; yields ty:result
4080 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
4081 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
4082 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
4087 The '``sub``' instruction returns the difference of its two operands.
4089 Note that the '``sub``' instruction is used to represent the '``neg``'
4090 instruction present in most other intermediate representations.
4095 The two arguments to the '``sub``' instruction must be
4096 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4097 arguments must have identical types.
4102 The value produced is the integer difference of the two operands.
4104 If the difference has unsigned overflow, the result returned is the
4105 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4108 Because LLVM integers use a two's complement representation, this
4109 instruction is appropriate for both signed and unsigned integers.
4111 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4112 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4113 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
4114 unsigned and/or signed overflow, respectively, occurs.
4119 .. code-block:: llvm
4121 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
4122 <result> = sub i32 0, %val ; yields i32:result = -%var
4126 '``fsub``' Instruction
4127 ^^^^^^^^^^^^^^^^^^^^^^
4134 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4139 The '``fsub``' instruction returns the difference of its two operands.
4141 Note that the '``fsub``' instruction is used to represent the '``fneg``'
4142 instruction present in most other intermediate representations.
4147 The two arguments to the '``fsub``' instruction must be :ref:`floating
4148 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4149 Both arguments must have identical types.
4154 The value produced is the floating point difference of the two operands.
4155 This instruction can also take any number of :ref:`fast-math
4156 flags <fastmath>`, which are optimization hints to enable otherwise
4157 unsafe floating point optimizations:
4162 .. code-block:: llvm
4164 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
4165 <result> = fsub float -0.0, %val ; yields float:result = -%var
4167 '``mul``' Instruction
4168 ^^^^^^^^^^^^^^^^^^^^^
4175 <result> = mul <ty> <op1>, <op2> ; yields ty:result
4176 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
4177 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
4178 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
4183 The '``mul``' instruction returns the product of its two operands.
4188 The two arguments to the '``mul``' instruction must be
4189 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4190 arguments must have identical types.
4195 The value produced is the integer product of the two operands.
4197 If the result of the multiplication has unsigned overflow, the result
4198 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
4199 bit width of the result.
4201 Because LLVM integers use a two's complement representation, and the
4202 result is the same width as the operands, this instruction returns the
4203 correct result for both signed and unsigned integers. If a full product
4204 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
4205 sign-extended or zero-extended as appropriate to the width of the full
4208 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4209 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4210 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
4211 unsigned and/or signed overflow, respectively, occurs.
4216 .. code-block:: llvm
4218 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
4222 '``fmul``' Instruction
4223 ^^^^^^^^^^^^^^^^^^^^^^
4230 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4235 The '``fmul``' instruction returns the product of its two operands.
4240 The two arguments to the '``fmul``' instruction must be :ref:`floating
4241 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4242 Both arguments must have identical types.
4247 The value produced is the floating point product of the two operands.
4248 This instruction can also take any number of :ref:`fast-math
4249 flags <fastmath>`, which are optimization hints to enable otherwise
4250 unsafe floating point optimizations:
4255 .. code-block:: llvm
4257 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
4259 '``udiv``' Instruction
4260 ^^^^^^^^^^^^^^^^^^^^^^
4267 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
4268 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
4273 The '``udiv``' instruction returns the quotient of its two operands.
4278 The two arguments to the '``udiv``' instruction must be
4279 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4280 arguments must have identical types.
4285 The value produced is the unsigned integer quotient of the two operands.
4287 Note that unsigned integer division and signed integer division are
4288 distinct operations; for signed integer division, use '``sdiv``'.
4290 Division by zero leads to undefined behavior.
4292 If the ``exact`` keyword is present, the result value of the ``udiv`` is
4293 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
4294 such, "((a udiv exact b) mul b) == a").
4299 .. code-block:: llvm
4301 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
4303 '``sdiv``' Instruction
4304 ^^^^^^^^^^^^^^^^^^^^^^
4311 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
4312 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
4317 The '``sdiv``' instruction returns the quotient of its two operands.
4322 The two arguments to the '``sdiv``' instruction must be
4323 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4324 arguments must have identical types.
4329 The value produced is the signed integer quotient of the two operands
4330 rounded towards zero.
4332 Note that signed integer division and unsigned integer division are
4333 distinct operations; for unsigned integer division, use '``udiv``'.
4335 Division by zero leads to undefined behavior. Overflow also leads to
4336 undefined behavior; this is a rare case, but can occur, for example, by
4337 doing a 32-bit division of -2147483648 by -1.
4339 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4340 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4345 .. code-block:: llvm
4347 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
4351 '``fdiv``' Instruction
4352 ^^^^^^^^^^^^^^^^^^^^^^
4359 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4364 The '``fdiv``' instruction returns the quotient of its two operands.
4369 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4370 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4371 Both arguments must have identical types.
4376 The value produced is the floating point quotient of the two operands.
4377 This instruction can also take any number of :ref:`fast-math
4378 flags <fastmath>`, which are optimization hints to enable otherwise
4379 unsafe floating point optimizations:
4384 .. code-block:: llvm
4386 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
4388 '``urem``' Instruction
4389 ^^^^^^^^^^^^^^^^^^^^^^
4396 <result> = urem <ty> <op1>, <op2> ; yields ty:result
4401 The '``urem``' instruction returns the remainder from the unsigned
4402 division of its two arguments.
4407 The two arguments to the '``urem``' instruction must be
4408 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4409 arguments must have identical types.
4414 This instruction returns the unsigned integer *remainder* of a division.
4415 This instruction always performs an unsigned division to get the
4418 Note that unsigned integer remainder and signed integer remainder are
4419 distinct operations; for signed integer remainder, use '``srem``'.
4421 Taking the remainder of a division by zero leads to undefined behavior.
4426 .. code-block:: llvm
4428 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
4430 '``srem``' Instruction
4431 ^^^^^^^^^^^^^^^^^^^^^^
4438 <result> = srem <ty> <op1>, <op2> ; yields ty:result
4443 The '``srem``' instruction returns the remainder from the signed
4444 division of its two operands. This instruction can also take
4445 :ref:`vector <t_vector>` versions of the values in which case the elements
4451 The two arguments to the '``srem``' instruction must be
4452 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4453 arguments must have identical types.
4458 This instruction returns the *remainder* of a division (where the result
4459 is either zero or has the same sign as the dividend, ``op1``), not the
4460 *modulo* operator (where the result is either zero or has the same sign
4461 as the divisor, ``op2``) of a value. For more information about the
4462 difference, see `The Math
4463 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4464 table of how this is implemented in various languages, please see
4466 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4468 Note that signed integer remainder and unsigned integer remainder are
4469 distinct operations; for unsigned integer remainder, use '``urem``'.
4471 Taking the remainder of a division by zero leads to undefined behavior.
4472 Overflow also leads to undefined behavior; this is a rare case, but can
4473 occur, for example, by taking the remainder of a 32-bit division of
4474 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4475 rule lets srem be implemented using instructions that return both the
4476 result of the division and the remainder.)
4481 .. code-block:: llvm
4483 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
4487 '``frem``' Instruction
4488 ^^^^^^^^^^^^^^^^^^^^^^
4495 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4500 The '``frem``' instruction returns the remainder from the division of
4506 The two arguments to the '``frem``' instruction must be :ref:`floating
4507 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4508 Both arguments must have identical types.
4513 This instruction returns the *remainder* of a division. The remainder
4514 has the same sign as the dividend. This instruction can also take any
4515 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4516 to enable otherwise unsafe floating point optimizations:
4521 .. code-block:: llvm
4523 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
4527 Bitwise Binary Operations
4528 -------------------------
4530 Bitwise binary operators are used to do various forms of bit-twiddling
4531 in a program. They are generally very efficient instructions and can
4532 commonly be strength reduced from other instructions. They require two
4533 operands of the same type, execute an operation on them, and produce a
4534 single value. The resulting value is the same type as its operands.
4536 '``shl``' Instruction
4537 ^^^^^^^^^^^^^^^^^^^^^
4544 <result> = shl <ty> <op1>, <op2> ; yields ty:result
4545 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
4546 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
4547 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
4552 The '``shl``' instruction returns the first operand shifted to the left
4553 a specified number of bits.
4558 Both arguments to the '``shl``' instruction must be the same
4559 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4560 '``op2``' is treated as an unsigned value.
4565 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4566 where ``n`` is the width of the result. If ``op2`` is (statically or
4567 dynamically) negative or equal to or larger than the number of bits in
4568 ``op1``, the result is undefined. If the arguments are vectors, each
4569 vector element of ``op1`` is shifted by the corresponding shift amount
4572 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4573 value <poisonvalues>` if it shifts out any non-zero bits. If the
4574 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4575 value <poisonvalues>` if it shifts out any bits that disagree with the
4576 resultant sign bit. As such, NUW/NSW have the same semantics as they
4577 would if the shift were expressed as a mul instruction with the same
4578 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4583 .. code-block:: llvm
4585 <result> = shl i32 4, %var ; yields i32: 4 << %var
4586 <result> = shl i32 4, 2 ; yields i32: 16
4587 <result> = shl i32 1, 10 ; yields i32: 1024
4588 <result> = shl i32 1, 32 ; undefined
4589 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4591 '``lshr``' Instruction
4592 ^^^^^^^^^^^^^^^^^^^^^^
4599 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
4600 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
4605 The '``lshr``' instruction (logical shift right) returns the first
4606 operand shifted to the right a specified number of bits with zero fill.
4611 Both arguments to the '``lshr``' instruction must be the same
4612 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4613 '``op2``' is treated as an unsigned value.
4618 This instruction always performs a logical shift right operation. The
4619 most significant bits of the result will be filled with zero bits after
4620 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4621 than the number of bits in ``op1``, the result is undefined. If the
4622 arguments are vectors, each vector element of ``op1`` is shifted by the
4623 corresponding shift amount in ``op2``.
4625 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4626 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4632 .. code-block:: llvm
4634 <result> = lshr i32 4, 1 ; yields i32:result = 2
4635 <result> = lshr i32 4, 2 ; yields i32:result = 1
4636 <result> = lshr i8 4, 3 ; yields i8:result = 0
4637 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
4638 <result> = lshr i32 1, 32 ; undefined
4639 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4641 '``ashr``' Instruction
4642 ^^^^^^^^^^^^^^^^^^^^^^
4649 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
4650 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
4655 The '``ashr``' instruction (arithmetic shift right) returns the first
4656 operand shifted to the right a specified number of bits with sign
4662 Both arguments to the '``ashr``' instruction must be the same
4663 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4664 '``op2``' is treated as an unsigned value.
4669 This instruction always performs an arithmetic shift right operation,
4670 The most significant bits of the result will be filled with the sign bit
4671 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4672 than the number of bits in ``op1``, the result is undefined. If the
4673 arguments are vectors, each vector element of ``op1`` is shifted by the
4674 corresponding shift amount in ``op2``.
4676 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4677 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4683 .. code-block:: llvm
4685 <result> = ashr i32 4, 1 ; yields i32:result = 2
4686 <result> = ashr i32 4, 2 ; yields i32:result = 1
4687 <result> = ashr i8 4, 3 ; yields i8:result = 0
4688 <result> = ashr i8 -2, 1 ; yields i8:result = -1
4689 <result> = ashr i32 1, 32 ; undefined
4690 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4692 '``and``' Instruction
4693 ^^^^^^^^^^^^^^^^^^^^^
4700 <result> = and <ty> <op1>, <op2> ; yields ty:result
4705 The '``and``' instruction returns the bitwise logical and of its two
4711 The two arguments to the '``and``' instruction must be
4712 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4713 arguments must have identical types.
4718 The truth table used for the '``and``' instruction is:
4735 .. code-block:: llvm
4737 <result> = and i32 4, %var ; yields i32:result = 4 & %var
4738 <result> = and i32 15, 40 ; yields i32:result = 8
4739 <result> = and i32 4, 8 ; yields i32:result = 0
4741 '``or``' Instruction
4742 ^^^^^^^^^^^^^^^^^^^^
4749 <result> = or <ty> <op1>, <op2> ; yields ty:result
4754 The '``or``' instruction returns the bitwise logical inclusive or of its
4760 The two arguments to the '``or``' instruction must be
4761 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4762 arguments must have identical types.
4767 The truth table used for the '``or``' instruction is:
4786 <result> = or i32 4, %var ; yields i32:result = 4 | %var
4787 <result> = or i32 15, 40 ; yields i32:result = 47
4788 <result> = or i32 4, 8 ; yields i32:result = 12
4790 '``xor``' Instruction
4791 ^^^^^^^^^^^^^^^^^^^^^
4798 <result> = xor <ty> <op1>, <op2> ; yields ty:result
4803 The '``xor``' instruction returns the bitwise logical exclusive or of
4804 its two operands. The ``xor`` is used to implement the "one's
4805 complement" operation, which is the "~" operator in C.
4810 The two arguments to the '``xor``' instruction must be
4811 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4812 arguments must have identical types.
4817 The truth table used for the '``xor``' instruction is:
4834 .. code-block:: llvm
4836 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
4837 <result> = xor i32 15, 40 ; yields i32:result = 39
4838 <result> = xor i32 4, 8 ; yields i32:result = 12
4839 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
4844 LLVM supports several instructions to represent vector operations in a
4845 target-independent manner. These instructions cover the element-access
4846 and vector-specific operations needed to process vectors effectively.
4847 While LLVM does directly support these vector operations, many
4848 sophisticated algorithms will want to use target-specific intrinsics to
4849 take full advantage of a specific target.
4851 .. _i_extractelement:
4853 '``extractelement``' Instruction
4854 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4861 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
4866 The '``extractelement``' instruction extracts a single scalar element
4867 from a vector at a specified index.
4872 The first operand of an '``extractelement``' instruction is a value of
4873 :ref:`vector <t_vector>` type. The second operand is an index indicating
4874 the position from which to extract the element. The index may be a
4875 variable of any integer type.
4880 The result is a scalar of the same type as the element type of ``val``.
4881 Its value is the value at position ``idx`` of ``val``. If ``idx``
4882 exceeds the length of ``val``, the results are undefined.
4887 .. code-block:: llvm
4889 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4891 .. _i_insertelement:
4893 '``insertelement``' Instruction
4894 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4901 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
4906 The '``insertelement``' instruction inserts a scalar element into a
4907 vector at a specified index.
4912 The first operand of an '``insertelement``' instruction is a value of
4913 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4914 type must equal the element type of the first operand. The third operand
4915 is an index indicating the position at which to insert the value. The
4916 index may be a variable of any integer type.
4921 The result is a vector of the same type as ``val``. Its element values
4922 are those of ``val`` except at position ``idx``, where it gets the value
4923 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4929 .. code-block:: llvm
4931 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4933 .. _i_shufflevector:
4935 '``shufflevector``' Instruction
4936 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4943 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4948 The '``shufflevector``' instruction constructs a permutation of elements
4949 from two input vectors, returning a vector with the same element type as
4950 the input and length that is the same as the shuffle mask.
4955 The first two operands of a '``shufflevector``' instruction are vectors
4956 with the same type. The third argument is a shuffle mask whose element
4957 type is always 'i32'. The result of the instruction is a vector whose
4958 length is the same as the shuffle mask and whose element type is the
4959 same as the element type of the first two operands.
4961 The shuffle mask operand is required to be a constant vector with either
4962 constant integer or undef values.
4967 The elements of the two input vectors are numbered from left to right
4968 across both of the vectors. The shuffle mask operand specifies, for each
4969 element of the result vector, which element of the two input vectors the
4970 result element gets. The element selector may be undef (meaning "don't
4971 care") and the second operand may be undef if performing a shuffle from
4977 .. code-block:: llvm
4979 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4980 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4981 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4982 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4983 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4984 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4985 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4986 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4988 Aggregate Operations
4989 --------------------
4991 LLVM supports several instructions for working with
4992 :ref:`aggregate <t_aggregate>` values.
4996 '``extractvalue``' Instruction
4997 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5004 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
5009 The '``extractvalue``' instruction extracts the value of a member field
5010 from an :ref:`aggregate <t_aggregate>` value.
5015 The first operand of an '``extractvalue``' instruction is a value of
5016 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
5017 constant indices to specify which value to extract in a similar manner
5018 as indices in a '``getelementptr``' instruction.
5020 The major differences to ``getelementptr`` indexing are:
5022 - Since the value being indexed is not a pointer, the first index is
5023 omitted and assumed to be zero.
5024 - At least one index must be specified.
5025 - Not only struct indices but also array indices must be in bounds.
5030 The result is the value at the position in the aggregate specified by
5036 .. code-block:: llvm
5038 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
5042 '``insertvalue``' Instruction
5043 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5050 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
5055 The '``insertvalue``' instruction inserts a value into a member field in
5056 an :ref:`aggregate <t_aggregate>` value.
5061 The first operand of an '``insertvalue``' instruction is a value of
5062 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
5063 a first-class value to insert. The following operands are constant
5064 indices indicating the position at which to insert the value in a
5065 similar manner as indices in a '``extractvalue``' instruction. The value
5066 to insert must have the same type as the value identified by the
5072 The result is an aggregate of the same type as ``val``. Its value is
5073 that of ``val`` except that the value at the position specified by the
5074 indices is that of ``elt``.
5079 .. code-block:: llvm
5081 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
5082 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
5083 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
5087 Memory Access and Addressing Operations
5088 ---------------------------------------
5090 A key design point of an SSA-based representation is how it represents
5091 memory. In LLVM, no memory locations are in SSA form, which makes things
5092 very simple. This section describes how to read, write, and allocate
5097 '``alloca``' Instruction
5098 ^^^^^^^^^^^^^^^^^^^^^^^^
5105 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
5110 The '``alloca``' instruction allocates memory on the stack frame of the
5111 currently executing function, to be automatically released when this
5112 function returns to its caller. The object is always allocated in the
5113 generic address space (address space zero).
5118 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
5119 bytes of memory on the runtime stack, returning a pointer of the
5120 appropriate type to the program. If "NumElements" is specified, it is
5121 the number of elements allocated, otherwise "NumElements" is defaulted
5122 to be one. If a constant alignment is specified, the value result of the
5123 allocation is guaranteed to be aligned to at least that boundary. The
5124 alignment may not be greater than ``1 << 29``. If not specified, or if
5125 zero, the target can choose to align the allocation on any convenient
5126 boundary compatible with the type.
5128 '``type``' may be any sized type.
5133 Memory is allocated; a pointer is returned. The operation is undefined
5134 if there is insufficient stack space for the allocation. '``alloca``'d
5135 memory is automatically released when the function returns. The
5136 '``alloca``' instruction is commonly used to represent automatic
5137 variables that must have an address available. When the function returns
5138 (either with the ``ret`` or ``resume`` instructions), the memory is
5139 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5140 The order in which memory is allocated (ie., which way the stack grows)
5146 .. code-block:: llvm
5148 %ptr = alloca i32 ; yields i32*:ptr
5149 %ptr = alloca i32, i32 4 ; yields i32*:ptr
5150 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
5151 %ptr = alloca i32, align 1024 ; yields i32*:ptr
5155 '``load``' Instruction
5156 ^^^^^^^^^^^^^^^^^^^^^^
5163 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !nonnull !<index>]
5164 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5165 !<index> = !{ i32 1 }
5170 The '``load``' instruction is used to read from memory.
5175 The argument to the ``load`` instruction specifies the memory address
5176 from which to load. The pointer must point to a :ref:`first
5177 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
5178 then the optimizer is not allowed to modify the number or order of
5179 execution of this ``load`` with other :ref:`volatile
5180 operations <volatile>`.
5182 If the ``load`` is marked as ``atomic``, it takes an extra
5183 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5184 ``release`` and ``acq_rel`` orderings are not valid on ``load``
5185 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5186 when they may see multiple atomic stores. The type of the pointee must
5187 be an integer type whose bit width is a power of two greater than or
5188 equal to eight and less than or equal to a target-specific size limit.
5189 ``align`` must be explicitly specified on atomic loads, and the load has
5190 undefined behavior if the alignment is not set to a value which is at
5191 least the size in bytes of the pointee. ``!nontemporal`` does not have
5192 any defined semantics for atomic loads.
5194 The optional constant ``align`` argument specifies the alignment of the
5195 operation (that is, the alignment of the memory address). A value of 0
5196 or an omitted ``align`` argument means that the operation has the ABI
5197 alignment for the target. It is the responsibility of the code emitter
5198 to ensure that the alignment information is correct. Overestimating the
5199 alignment results in undefined behavior. Underestimating the alignment
5200 may produce less efficient code. An alignment of 1 is always safe. The
5201 maximum possible alignment is ``1 << 29``.
5203 The optional ``!nontemporal`` metadata must reference a single
5204 metadata name ``<index>`` corresponding to a metadata node with one
5205 ``i32`` entry of value 1. The existence of the ``!nontemporal``
5206 metadata on the instruction tells the optimizer and code generator
5207 that this load is not expected to be reused in the cache. The code
5208 generator may select special instructions to save cache bandwidth, such
5209 as the ``MOVNT`` instruction on x86.
5211 The optional ``!invariant.load`` metadata must reference a single
5212 metadata name ``<index>`` corresponding to a metadata node with no
5213 entries. The existence of the ``!invariant.load`` metadata on the
5214 instruction tells the optimizer and code generator that the address
5215 operand to this load points to memory which can be assumed unchanged.
5216 Being invariant does not imply that a location is dereferenceable,
5217 but it does imply that once the location is known dereferenceable
5218 its value is henceforth unchanging.
5220 The optional ``!nonnull`` metadata must reference a single
5221 metadata name ``<index>`` corresponding to a metadata node with no
5222 entries. The existence of the ``!nonnull`` metadata on the
5223 instruction tells the optimizer that the value loaded is known to
5224 never be null. This is analogous to the ''nonnull'' attribute
5225 on parameters and return values. This metadata can only be applied
5226 to loads of a pointer type.
5231 The location of memory pointed to is loaded. If the value being loaded
5232 is of scalar type then the number of bytes read does not exceed the
5233 minimum number of bytes needed to hold all bits of the type. For
5234 example, loading an ``i24`` reads at most three bytes. When loading a
5235 value of a type like ``i20`` with a size that is not an integral number
5236 of bytes, the result is undefined if the value was not originally
5237 written using a store of the same type.
5242 .. code-block:: llvm
5244 %ptr = alloca i32 ; yields i32*:ptr
5245 store i32 3, i32* %ptr ; yields void
5246 %val = load i32* %ptr ; yields i32:val = i32 3
5250 '``store``' Instruction
5251 ^^^^^^^^^^^^^^^^^^^^^^^
5258 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
5259 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
5264 The '``store``' instruction is used to write to memory.
5269 There are two arguments to the ``store`` instruction: a value to store
5270 and an address at which to store it. The type of the ``<pointer>``
5271 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
5272 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
5273 then the optimizer is not allowed to modify the number or order of
5274 execution of this ``store`` with other :ref:`volatile
5275 operations <volatile>`.
5277 If the ``store`` is marked as ``atomic``, it takes an extra
5278 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5279 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
5280 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5281 when they may see multiple atomic stores. The type of the pointee must
5282 be an integer type whose bit width is a power of two greater than or
5283 equal to eight and less than or equal to a target-specific size limit.
5284 ``align`` must be explicitly specified on atomic stores, and the store
5285 has undefined behavior if the alignment is not set to a value which is
5286 at least the size in bytes of the pointee. ``!nontemporal`` does not
5287 have any defined semantics for atomic stores.
5289 The optional constant ``align`` argument specifies the alignment of the
5290 operation (that is, the alignment of the memory address). A value of 0
5291 or an omitted ``align`` argument means that the operation has the ABI
5292 alignment for the target. It is the responsibility of the code emitter
5293 to ensure that the alignment information is correct. Overestimating the
5294 alignment results in undefined behavior. Underestimating the
5295 alignment may produce less efficient code. An alignment of 1 is always
5296 safe. The maximum possible alignment is ``1 << 29``.
5298 The optional ``!nontemporal`` metadata must reference a single metadata
5299 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
5300 value 1. The existence of the ``!nontemporal`` metadata on the instruction
5301 tells the optimizer and code generator that this load is not expected to
5302 be reused in the cache. The code generator may select special
5303 instructions to save cache bandwidth, such as the MOVNT instruction on
5309 The contents of memory are updated to contain ``<value>`` at the
5310 location specified by the ``<pointer>`` operand. If ``<value>`` is
5311 of scalar type then the number of bytes written does not exceed the
5312 minimum number of bytes needed to hold all bits of the type. For
5313 example, storing an ``i24`` writes at most three bytes. When writing a
5314 value of a type like ``i20`` with a size that is not an integral number
5315 of bytes, it is unspecified what happens to the extra bits that do not
5316 belong to the type, but they will typically be overwritten.
5321 .. code-block:: llvm
5323 %ptr = alloca i32 ; yields i32*:ptr
5324 store i32 3, i32* %ptr ; yields void
5325 %val = load i32* %ptr ; yields i32:val = i32 3
5329 '``fence``' Instruction
5330 ^^^^^^^^^^^^^^^^^^^^^^^
5337 fence [singlethread] <ordering> ; yields void
5342 The '``fence``' instruction is used to introduce happens-before edges
5348 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5349 defines what *synchronizes-with* edges they add. They can only be given
5350 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5355 A fence A which has (at least) ``release`` ordering semantics
5356 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5357 semantics if and only if there exist atomic operations X and Y, both
5358 operating on some atomic object M, such that A is sequenced before X, X
5359 modifies M (either directly or through some side effect of a sequence
5360 headed by X), Y is sequenced before B, and Y observes M. This provides a
5361 *happens-before* dependency between A and B. Rather than an explicit
5362 ``fence``, one (but not both) of the atomic operations X or Y might
5363 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5364 still *synchronize-with* the explicit ``fence`` and establish the
5365 *happens-before* edge.
5367 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5368 ``acquire`` and ``release`` semantics specified above, participates in
5369 the global program order of other ``seq_cst`` operations and/or fences.
5371 The optional ":ref:`singlethread <singlethread>`" argument specifies
5372 that the fence only synchronizes with other fences in the same thread.
5373 (This is useful for interacting with signal handlers.)
5378 .. code-block:: llvm
5380 fence acquire ; yields void
5381 fence singlethread seq_cst ; yields void
5385 '``cmpxchg``' Instruction
5386 ^^^^^^^^^^^^^^^^^^^^^^^^^
5393 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
5398 The '``cmpxchg``' instruction is used to atomically modify memory. It
5399 loads a value in memory and compares it to a given value. If they are
5400 equal, it tries to store a new value into the memory.
5405 There are three arguments to the '``cmpxchg``' instruction: an address
5406 to operate on, a value to compare to the value currently be at that
5407 address, and a new value to place at that address if the compared values
5408 are equal. The type of '<cmp>' must be an integer type whose bit width
5409 is a power of two greater than or equal to eight and less than or equal
5410 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5411 type, and the type of '<pointer>' must be a pointer to that type. If the
5412 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5413 to modify the number or order of execution of this ``cmpxchg`` with
5414 other :ref:`volatile operations <volatile>`.
5416 The success and failure :ref:`ordering <ordering>` arguments specify how this
5417 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
5418 must be at least ``monotonic``, the ordering constraint on failure must be no
5419 stronger than that on success, and the failure ordering cannot be either
5420 ``release`` or ``acq_rel``.
5422 The optional "``singlethread``" argument declares that the ``cmpxchg``
5423 is only atomic with respect to code (usually signal handlers) running in
5424 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5425 respect to all other code in the system.
5427 The pointer passed into cmpxchg must have alignment greater than or
5428 equal to the size in memory of the operand.
5433 The contents of memory at the location specified by the '``<pointer>``' operand
5434 is read and compared to '``<cmp>``'; if the read value is the equal, the
5435 '``<new>``' is written. The original value at the location is returned, together
5436 with a flag indicating success (true) or failure (false).
5438 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
5439 permitted: the operation may not write ``<new>`` even if the comparison
5442 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
5443 if the value loaded equals ``cmp``.
5445 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5446 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5447 load with an ordering parameter determined the second ordering parameter.
5452 .. code-block:: llvm
5455 %orig = atomic load i32* %ptr unordered ; yields i32
5459 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5460 %squared = mul i32 %cmp, %cmp
5461 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
5462 %value_loaded = extractvalue { i32, i1 } %val_success, 0
5463 %success = extractvalue { i32, i1 } %val_success, 1
5464 br i1 %success, label %done, label %loop
5471 '``atomicrmw``' Instruction
5472 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5479 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
5484 The '``atomicrmw``' instruction is used to atomically modify memory.
5489 There are three arguments to the '``atomicrmw``' instruction: an
5490 operation to apply, an address whose value to modify, an argument to the
5491 operation. The operation must be one of the following keywords:
5505 The type of '<value>' must be an integer type whose bit width is a power
5506 of two greater than or equal to eight and less than or equal to a
5507 target-specific size limit. The type of the '``<pointer>``' operand must
5508 be a pointer to that type. If the ``atomicrmw`` is marked as
5509 ``volatile``, then the optimizer is not allowed to modify the number or
5510 order of execution of this ``atomicrmw`` with other :ref:`volatile
5511 operations <volatile>`.
5516 The contents of memory at the location specified by the '``<pointer>``'
5517 operand are atomically read, modified, and written back. The original
5518 value at the location is returned. The modification is specified by the
5521 - xchg: ``*ptr = val``
5522 - add: ``*ptr = *ptr + val``
5523 - sub: ``*ptr = *ptr - val``
5524 - and: ``*ptr = *ptr & val``
5525 - nand: ``*ptr = ~(*ptr & val)``
5526 - or: ``*ptr = *ptr | val``
5527 - xor: ``*ptr = *ptr ^ val``
5528 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5529 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5530 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5532 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5538 .. code-block:: llvm
5540 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
5542 .. _i_getelementptr:
5544 '``getelementptr``' Instruction
5545 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5552 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5553 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5554 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5559 The '``getelementptr``' instruction is used to get the address of a
5560 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5561 address calculation only and does not access memory.
5566 The first argument is always a pointer or a vector of pointers, and
5567 forms the basis of the calculation. The remaining arguments are indices
5568 that indicate which of the elements of the aggregate object are indexed.
5569 The interpretation of each index is dependent on the type being indexed
5570 into. The first index always indexes the pointer value given as the
5571 first argument, the second index indexes a value of the type pointed to
5572 (not necessarily the value directly pointed to, since the first index
5573 can be non-zero), etc. The first type indexed into must be a pointer
5574 value, subsequent types can be arrays, vectors, and structs. Note that
5575 subsequent types being indexed into can never be pointers, since that
5576 would require loading the pointer before continuing calculation.
5578 The type of each index argument depends on the type it is indexing into.
5579 When indexing into a (optionally packed) structure, only ``i32`` integer
5580 **constants** are allowed (when using a vector of indices they must all
5581 be the **same** ``i32`` integer constant). When indexing into an array,
5582 pointer or vector, integers of any width are allowed, and they are not
5583 required to be constant. These integers are treated as signed values
5586 For example, let's consider a C code fragment and how it gets compiled
5602 int *foo(struct ST *s) {
5603 return &s[1].Z.B[5][13];
5606 The LLVM code generated by Clang is:
5608 .. code-block:: llvm
5610 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5611 %struct.ST = type { i32, double, %struct.RT }
5613 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5615 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5622 In the example above, the first index is indexing into the
5623 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5624 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5625 indexes into the third element of the structure, yielding a
5626 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5627 structure. The third index indexes into the second element of the
5628 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5629 dimensions of the array are subscripted into, yielding an '``i32``'
5630 type. The '``getelementptr``' instruction returns a pointer to this
5631 element, thus computing a value of '``i32*``' type.
5633 Note that it is perfectly legal to index partially through a structure,
5634 returning a pointer to an inner element. Because of this, the LLVM code
5635 for the given testcase is equivalent to:
5637 .. code-block:: llvm
5639 define i32* @foo(%struct.ST* %s) {
5640 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5641 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5642 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5643 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5644 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5648 If the ``inbounds`` keyword is present, the result value of the
5649 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5650 pointer is not an *in bounds* address of an allocated object, or if any
5651 of the addresses that would be formed by successive addition of the
5652 offsets implied by the indices to the base address with infinitely
5653 precise signed arithmetic are not an *in bounds* address of that
5654 allocated object. The *in bounds* addresses for an allocated object are
5655 all the addresses that point into the object, plus the address one byte
5656 past the end. In cases where the base is a vector of pointers the
5657 ``inbounds`` keyword applies to each of the computations element-wise.
5659 If the ``inbounds`` keyword is not present, the offsets are added to the
5660 base address with silently-wrapping two's complement arithmetic. If the
5661 offsets have a different width from the pointer, they are sign-extended
5662 or truncated to the width of the pointer. The result value of the
5663 ``getelementptr`` may be outside the object pointed to by the base
5664 pointer. The result value may not necessarily be used to access memory
5665 though, even if it happens to point into allocated storage. See the
5666 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5669 The getelementptr instruction is often confusing. For some more insight
5670 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5675 .. code-block:: llvm
5677 ; yields [12 x i8]*:aptr
5678 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5680 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5682 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5684 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5686 In cases where the pointer argument is a vector of pointers, each index
5687 must be a vector with the same number of elements. For example:
5689 .. code-block:: llvm
5691 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5693 Conversion Operations
5694 ---------------------
5696 The instructions in this category are the conversion instructions
5697 (casting) which all take a single operand and a type. They perform
5698 various bit conversions on the operand.
5700 '``trunc .. to``' Instruction
5701 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5708 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5713 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5718 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5719 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5720 of the same number of integers. The bit size of the ``value`` must be
5721 larger than the bit size of the destination type, ``ty2``. Equal sized
5722 types are not allowed.
5727 The '``trunc``' instruction truncates the high order bits in ``value``
5728 and converts the remaining bits to ``ty2``. Since the source size must
5729 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5730 It will always truncate bits.
5735 .. code-block:: llvm
5737 %X = trunc i32 257 to i8 ; yields i8:1
5738 %Y = trunc i32 123 to i1 ; yields i1:true
5739 %Z = trunc i32 122 to i1 ; yields i1:false
5740 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5742 '``zext .. to``' Instruction
5743 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5750 <result> = zext <ty> <value> to <ty2> ; yields ty2
5755 The '``zext``' instruction zero extends its operand to type ``ty2``.
5760 The '``zext``' instruction takes a value to cast, and a type to cast it
5761 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5762 the same number of integers. The bit size of the ``value`` must be
5763 smaller than the bit size of the destination type, ``ty2``.
5768 The ``zext`` fills the high order bits of the ``value`` with zero bits
5769 until it reaches the size of the destination type, ``ty2``.
5771 When zero extending from i1, the result will always be either 0 or 1.
5776 .. code-block:: llvm
5778 %X = zext i32 257 to i64 ; yields i64:257
5779 %Y = zext i1 true to i32 ; yields i32:1
5780 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5782 '``sext .. to``' Instruction
5783 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5790 <result> = sext <ty> <value> to <ty2> ; yields ty2
5795 The '``sext``' sign extends ``value`` to the type ``ty2``.
5800 The '``sext``' instruction takes a value to cast, and a type to cast it
5801 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5802 the same number of integers. The bit size of the ``value`` must be
5803 smaller than the bit size of the destination type, ``ty2``.
5808 The '``sext``' instruction performs a sign extension by copying the sign
5809 bit (highest order bit) of the ``value`` until it reaches the bit size
5810 of the type ``ty2``.
5812 When sign extending from i1, the extension always results in -1 or 0.
5817 .. code-block:: llvm
5819 %X = sext i8 -1 to i16 ; yields i16 :65535
5820 %Y = sext i1 true to i32 ; yields i32:-1
5821 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5823 '``fptrunc .. to``' Instruction
5824 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5831 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5836 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5841 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5842 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5843 The size of ``value`` must be larger than the size of ``ty2``. This
5844 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5849 The '``fptrunc``' instruction truncates a ``value`` from a larger
5850 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5851 point <t_floating>` type. If the value cannot fit within the
5852 destination type, ``ty2``, then the results are undefined.
5857 .. code-block:: llvm
5859 %X = fptrunc double 123.0 to float ; yields float:123.0
5860 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5862 '``fpext .. to``' Instruction
5863 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5870 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5875 The '``fpext``' extends a floating point ``value`` to a larger floating
5881 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5882 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5883 to. The source type must be smaller than the destination type.
5888 The '``fpext``' instruction extends the ``value`` from a smaller
5889 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5890 point <t_floating>` type. The ``fpext`` cannot be used to make a
5891 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5892 *no-op cast* for a floating point cast.
5897 .. code-block:: llvm
5899 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5900 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5902 '``fptoui .. to``' Instruction
5903 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5910 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5915 The '``fptoui``' converts a floating point ``value`` to its unsigned
5916 integer equivalent of type ``ty2``.
5921 The '``fptoui``' instruction takes a value to cast, which must be a
5922 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5923 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5924 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5925 type with the same number of elements as ``ty``
5930 The '``fptoui``' instruction converts its :ref:`floating
5931 point <t_floating>` operand into the nearest (rounding towards zero)
5932 unsigned integer value. If the value cannot fit in ``ty2``, the results
5938 .. code-block:: llvm
5940 %X = fptoui double 123.0 to i32 ; yields i32:123
5941 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5942 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5944 '``fptosi .. to``' Instruction
5945 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5952 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5957 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5958 ``value`` to type ``ty2``.
5963 The '``fptosi``' instruction takes a value to cast, which must be a
5964 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5965 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5966 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5967 type with the same number of elements as ``ty``
5972 The '``fptosi``' instruction converts its :ref:`floating
5973 point <t_floating>` operand into the nearest (rounding towards zero)
5974 signed integer value. If the value cannot fit in ``ty2``, the results
5980 .. code-block:: llvm
5982 %X = fptosi double -123.0 to i32 ; yields i32:-123
5983 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5984 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5986 '``uitofp .. to``' Instruction
5987 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5994 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5999 The '``uitofp``' instruction regards ``value`` as an unsigned integer
6000 and converts that value to the ``ty2`` type.
6005 The '``uitofp``' instruction takes a value to cast, which must be a
6006 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6007 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6008 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6009 type with the same number of elements as ``ty``
6014 The '``uitofp``' instruction interprets its operand as an unsigned
6015 integer quantity and converts it to the corresponding floating point
6016 value. If the value cannot fit in the floating point value, the results
6022 .. code-block:: llvm
6024 %X = uitofp i32 257 to float ; yields float:257.0
6025 %Y = uitofp i8 -1 to double ; yields double:255.0
6027 '``sitofp .. to``' Instruction
6028 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6035 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
6040 The '``sitofp``' instruction regards ``value`` as a signed integer and
6041 converts that value to the ``ty2`` type.
6046 The '``sitofp``' instruction takes a value to cast, which must be a
6047 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6048 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6049 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6050 type with the same number of elements as ``ty``
6055 The '``sitofp``' instruction interprets its operand as a signed integer
6056 quantity and converts it to the corresponding floating point value. If
6057 the value cannot fit in the floating point value, the results are
6063 .. code-block:: llvm
6065 %X = sitofp i32 257 to float ; yields float:257.0
6066 %Y = sitofp i8 -1 to double ; yields double:-1.0
6070 '``ptrtoint .. to``' Instruction
6071 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6078 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
6083 The '``ptrtoint``' instruction converts the pointer or a vector of
6084 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
6089 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
6090 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
6091 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
6092 a vector of integers type.
6097 The '``ptrtoint``' instruction converts ``value`` to integer type
6098 ``ty2`` by interpreting the pointer value as an integer and either
6099 truncating or zero extending that value to the size of the integer type.
6100 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
6101 ``value`` is larger than ``ty2`` then a truncation is done. If they are
6102 the same size, then nothing is done (*no-op cast*) other than a type
6108 .. code-block:: llvm
6110 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
6111 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
6112 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
6116 '``inttoptr .. to``' Instruction
6117 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6124 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
6129 The '``inttoptr``' instruction converts an integer ``value`` to a
6130 pointer type, ``ty2``.
6135 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
6136 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
6142 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
6143 applying either a zero extension or a truncation depending on the size
6144 of the integer ``value``. If ``value`` is larger than the size of a
6145 pointer then a truncation is done. If ``value`` is smaller than the size
6146 of a pointer then a zero extension is done. If they are the same size,
6147 nothing is done (*no-op cast*).
6152 .. code-block:: llvm
6154 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
6155 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
6156 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
6157 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
6161 '``bitcast .. to``' Instruction
6162 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6169 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
6174 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
6180 The '``bitcast``' instruction takes a value to cast, which must be a
6181 non-aggregate first class value, and a type to cast it to, which must
6182 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
6183 bit sizes of ``value`` and the destination type, ``ty2``, must be
6184 identical. If the source type is a pointer, the destination type must
6185 also be a pointer of the same size. This instruction supports bitwise
6186 conversion of vectors to integers and to vectors of other types (as
6187 long as they have the same size).
6192 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
6193 is always a *no-op cast* because no bits change with this
6194 conversion. The conversion is done as if the ``value`` had been stored
6195 to memory and read back as type ``ty2``. Pointer (or vector of
6196 pointers) types may only be converted to other pointer (or vector of
6197 pointers) types with the same address space through this instruction.
6198 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
6199 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
6204 .. code-block:: llvm
6206 %X = bitcast i8 255 to i8 ; yields i8 :-1
6207 %Y = bitcast i32* %x to sint* ; yields sint*:%x
6208 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
6209 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
6211 .. _i_addrspacecast:
6213 '``addrspacecast .. to``' Instruction
6214 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6221 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
6226 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
6227 address space ``n`` to type ``pty2`` in address space ``m``.
6232 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
6233 to cast and a pointer type to cast it to, which must have a different
6239 The '``addrspacecast``' instruction converts the pointer value
6240 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
6241 value modification, depending on the target and the address space
6242 pair. Pointer conversions within the same address space must be
6243 performed with the ``bitcast`` instruction. Note that if the address space
6244 conversion is legal then both result and operand refer to the same memory
6250 .. code-block:: llvm
6252 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
6253 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
6254 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
6261 The instructions in this category are the "miscellaneous" instructions,
6262 which defy better classification.
6266 '``icmp``' Instruction
6267 ^^^^^^^^^^^^^^^^^^^^^^
6274 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6279 The '``icmp``' instruction returns a boolean value or a vector of
6280 boolean values based on comparison of its two integer, integer vector,
6281 pointer, or pointer vector operands.
6286 The '``icmp``' instruction takes three operands. The first operand is
6287 the condition code indicating the kind of comparison to perform. It is
6288 not a value, just a keyword. The possible condition code are:
6291 #. ``ne``: not equal
6292 #. ``ugt``: unsigned greater than
6293 #. ``uge``: unsigned greater or equal
6294 #. ``ult``: unsigned less than
6295 #. ``ule``: unsigned less or equal
6296 #. ``sgt``: signed greater than
6297 #. ``sge``: signed greater or equal
6298 #. ``slt``: signed less than
6299 #. ``sle``: signed less or equal
6301 The remaining two arguments must be :ref:`integer <t_integer>` or
6302 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
6303 must also be identical types.
6308 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
6309 code given as ``cond``. The comparison performed always yields either an
6310 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
6312 #. ``eq``: yields ``true`` if the operands are equal, ``false``
6313 otherwise. No sign interpretation is necessary or performed.
6314 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
6315 otherwise. No sign interpretation is necessary or performed.
6316 #. ``ugt``: interprets the operands as unsigned values and yields
6317 ``true`` if ``op1`` is greater than ``op2``.
6318 #. ``uge``: interprets the operands as unsigned values and yields
6319 ``true`` if ``op1`` is greater than or equal to ``op2``.
6320 #. ``ult``: interprets the operands as unsigned values and yields
6321 ``true`` if ``op1`` is less than ``op2``.
6322 #. ``ule``: interprets the operands as unsigned values and yields
6323 ``true`` if ``op1`` is less than or equal to ``op2``.
6324 #. ``sgt``: interprets the operands as signed values and yields ``true``
6325 if ``op1`` is greater than ``op2``.
6326 #. ``sge``: interprets the operands as signed values and yields ``true``
6327 if ``op1`` is greater than or equal to ``op2``.
6328 #. ``slt``: interprets the operands as signed values and yields ``true``
6329 if ``op1`` is less than ``op2``.
6330 #. ``sle``: interprets the operands as signed values and yields ``true``
6331 if ``op1`` is less than or equal to ``op2``.
6333 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
6334 are compared as if they were integers.
6336 If the operands are integer vectors, then they are compared element by
6337 element. The result is an ``i1`` vector with the same number of elements
6338 as the values being compared. Otherwise, the result is an ``i1``.
6343 .. code-block:: llvm
6345 <result> = icmp eq i32 4, 5 ; yields: result=false
6346 <result> = icmp ne float* %X, %X ; yields: result=false
6347 <result> = icmp ult i16 4, 5 ; yields: result=true
6348 <result> = icmp sgt i16 4, 5 ; yields: result=false
6349 <result> = icmp ule i16 -4, 5 ; yields: result=false
6350 <result> = icmp sge i16 4, 5 ; yields: result=false
6352 Note that the code generator does not yet support vector types with the
6353 ``icmp`` instruction.
6357 '``fcmp``' Instruction
6358 ^^^^^^^^^^^^^^^^^^^^^^
6365 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6370 The '``fcmp``' instruction returns a boolean value or vector of boolean
6371 values based on comparison of its operands.
6373 If the operands are floating point scalars, then the result type is a
6374 boolean (:ref:`i1 <t_integer>`).
6376 If the operands are floating point vectors, then the result type is a
6377 vector of boolean with the same number of elements as the operands being
6383 The '``fcmp``' instruction takes three operands. The first operand is
6384 the condition code indicating the kind of comparison to perform. It is
6385 not a value, just a keyword. The possible condition code are:
6387 #. ``false``: no comparison, always returns false
6388 #. ``oeq``: ordered and equal
6389 #. ``ogt``: ordered and greater than
6390 #. ``oge``: ordered and greater than or equal
6391 #. ``olt``: ordered and less than
6392 #. ``ole``: ordered and less than or equal
6393 #. ``one``: ordered and not equal
6394 #. ``ord``: ordered (no nans)
6395 #. ``ueq``: unordered or equal
6396 #. ``ugt``: unordered or greater than
6397 #. ``uge``: unordered or greater than or equal
6398 #. ``ult``: unordered or less than
6399 #. ``ule``: unordered or less than or equal
6400 #. ``une``: unordered or not equal
6401 #. ``uno``: unordered (either nans)
6402 #. ``true``: no comparison, always returns true
6404 *Ordered* means that neither operand is a QNAN while *unordered* means
6405 that either operand may be a QNAN.
6407 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6408 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6409 type. They must have identical types.
6414 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6415 condition code given as ``cond``. If the operands are vectors, then the
6416 vectors are compared element by element. Each comparison performed
6417 always yields an :ref:`i1 <t_integer>` result, as follows:
6419 #. ``false``: always yields ``false``, regardless of operands.
6420 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6421 is equal to ``op2``.
6422 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6423 is greater than ``op2``.
6424 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6425 is greater than or equal to ``op2``.
6426 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6427 is less than ``op2``.
6428 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6429 is less than or equal to ``op2``.
6430 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6431 is not equal to ``op2``.
6432 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6433 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6435 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6436 greater than ``op2``.
6437 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6438 greater than or equal to ``op2``.
6439 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6441 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6442 less than or equal to ``op2``.
6443 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6444 not equal to ``op2``.
6445 #. ``uno``: yields ``true`` if either operand is a QNAN.
6446 #. ``true``: always yields ``true``, regardless of operands.
6451 .. code-block:: llvm
6453 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6454 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6455 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6456 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6458 Note that the code generator does not yet support vector types with the
6459 ``fcmp`` instruction.
6463 '``phi``' Instruction
6464 ^^^^^^^^^^^^^^^^^^^^^
6471 <result> = phi <ty> [ <val0>, <label0>], ...
6476 The '``phi``' instruction is used to implement the φ node in the SSA
6477 graph representing the function.
6482 The type of the incoming values is specified with the first type field.
6483 After this, the '``phi``' instruction takes a list of pairs as
6484 arguments, with one pair for each predecessor basic block of the current
6485 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6486 the value arguments to the PHI node. Only labels may be used as the
6489 There must be no non-phi instructions between the start of a basic block
6490 and the PHI instructions: i.e. PHI instructions must be first in a basic
6493 For the purposes of the SSA form, the use of each incoming value is
6494 deemed to occur on the edge from the corresponding predecessor block to
6495 the current block (but after any definition of an '``invoke``'
6496 instruction's return value on the same edge).
6501 At runtime, the '``phi``' instruction logically takes on the value
6502 specified by the pair corresponding to the predecessor basic block that
6503 executed just prior to the current block.
6508 .. code-block:: llvm
6510 Loop: ; Infinite loop that counts from 0 on up...
6511 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6512 %nextindvar = add i32 %indvar, 1
6517 '``select``' Instruction
6518 ^^^^^^^^^^^^^^^^^^^^^^^^
6525 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6527 selty is either i1 or {<N x i1>}
6532 The '``select``' instruction is used to choose one value based on a
6533 condition, without IR-level branching.
6538 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6539 values indicating the condition, and two values of the same :ref:`first
6540 class <t_firstclass>` type. If the val1/val2 are vectors and the
6541 condition is a scalar, then entire vectors are selected, not individual
6547 If the condition is an i1 and it evaluates to 1, the instruction returns
6548 the first value argument; otherwise, it returns the second value
6551 If the condition is a vector of i1, then the value arguments must be
6552 vectors of the same size, and the selection is done element by element.
6557 .. code-block:: llvm
6559 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6563 '``call``' Instruction
6564 ^^^^^^^^^^^^^^^^^^^^^^
6571 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6576 The '``call``' instruction represents a simple function call.
6581 This instruction requires several arguments:
6583 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
6584 should perform tail call optimization. The ``tail`` marker is a hint that
6585 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
6586 means that the call must be tail call optimized in order for the program to
6587 be correct. The ``musttail`` marker provides these guarantees:
6589 #. The call will not cause unbounded stack growth if it is part of a
6590 recursive cycle in the call graph.
6591 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
6594 Both markers imply that the callee does not access allocas or varargs from
6595 the caller. Calls marked ``musttail`` must obey the following additional
6598 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
6599 or a pointer bitcast followed by a ret instruction.
6600 - The ret instruction must return the (possibly bitcasted) value
6601 produced by the call or void.
6602 - The caller and callee prototypes must match. Pointer types of
6603 parameters or return types may differ in pointee type, but not
6605 - The calling conventions of the caller and callee must match.
6606 - All ABI-impacting function attributes, such as sret, byval, inreg,
6607 returned, and inalloca, must match.
6608 - The callee must be varargs iff the caller is varargs. Bitcasting a
6609 non-varargs function to the appropriate varargs type is legal so
6610 long as the non-varargs prefixes obey the other rules.
6612 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
6613 the following conditions are met:
6615 - Caller and callee both have the calling convention ``fastcc``.
6616 - The call is in tail position (ret immediately follows call and ret
6617 uses value of call or is void).
6618 - Option ``-tailcallopt`` is enabled, or
6619 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6620 - `Platform-specific constraints are
6621 met. <CodeGenerator.html#tailcallopt>`_
6623 #. The optional "cconv" marker indicates which :ref:`calling
6624 convention <callingconv>` the call should use. If none is
6625 specified, the call defaults to using C calling conventions. The
6626 calling convention of the call must match the calling convention of
6627 the target function, or else the behavior is undefined.
6628 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6629 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6631 #. '``ty``': the type of the call instruction itself which is also the
6632 type of the return value. Functions that return no value are marked
6634 #. '``fnty``': shall be the signature of the pointer to function value
6635 being invoked. The argument types must match the types implied by
6636 this signature. This type can be omitted if the function is not
6637 varargs and if the function type does not return a pointer to a
6639 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6640 be invoked. In most cases, this is a direct function invocation, but
6641 indirect ``call``'s are just as possible, calling an arbitrary pointer
6643 #. '``function args``': argument list whose types match the function
6644 signature argument types and parameter attributes. All arguments must
6645 be of :ref:`first class <t_firstclass>` type. If the function signature
6646 indicates the function accepts a variable number of arguments, the
6647 extra arguments can be specified.
6648 #. The optional :ref:`function attributes <fnattrs>` list. Only
6649 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6650 attributes are valid here.
6655 The '``call``' instruction is used to cause control flow to transfer to
6656 a specified function, with its incoming arguments bound to the specified
6657 values. Upon a '``ret``' instruction in the called function, control
6658 flow continues with the instruction after the function call, and the
6659 return value of the function is bound to the result argument.
6664 .. code-block:: llvm
6666 %retval = call i32 @test(i32 %argc)
6667 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6668 %X = tail call i32 @foo() ; yields i32
6669 %Y = tail call fastcc i32 @foo() ; yields i32
6670 call void %foo(i8 97 signext)
6672 %struct.A = type { i32, i8 }
6673 %r = call %struct.A @foo() ; yields { i32, i8 }
6674 %gr = extractvalue %struct.A %r, 0 ; yields i32
6675 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6676 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6677 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6679 llvm treats calls to some functions with names and arguments that match
6680 the standard C99 library as being the C99 library functions, and may
6681 perform optimizations or generate code for them under that assumption.
6682 This is something we'd like to change in the future to provide better
6683 support for freestanding environments and non-C-based languages.
6687 '``va_arg``' Instruction
6688 ^^^^^^^^^^^^^^^^^^^^^^^^
6695 <resultval> = va_arg <va_list*> <arglist>, <argty>
6700 The '``va_arg``' instruction is used to access arguments passed through
6701 the "variable argument" area of a function call. It is used to implement
6702 the ``va_arg`` macro in C.
6707 This instruction takes a ``va_list*`` value and the type of the
6708 argument. It returns a value of the specified argument type and
6709 increments the ``va_list`` to point to the next argument. The actual
6710 type of ``va_list`` is target specific.
6715 The '``va_arg``' instruction loads an argument of the specified type
6716 from the specified ``va_list`` and causes the ``va_list`` to point to
6717 the next argument. For more information, see the variable argument
6718 handling :ref:`Intrinsic Functions <int_varargs>`.
6720 It is legal for this instruction to be called in a function which does
6721 not take a variable number of arguments, for example, the ``vfprintf``
6724 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6725 function <intrinsics>` because it takes a type as an argument.
6730 See the :ref:`variable argument processing <int_varargs>` section.
6732 Note that the code generator does not yet fully support va\_arg on many
6733 targets. Also, it does not currently support va\_arg with aggregate
6734 types on any target.
6738 '``landingpad``' Instruction
6739 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6746 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6747 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6749 <clause> := catch <type> <value>
6750 <clause> := filter <array constant type> <array constant>
6755 The '``landingpad``' instruction is used by `LLVM's exception handling
6756 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6757 is a landing pad --- one where the exception lands, and corresponds to the
6758 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6759 defines values supplied by the personality function (``pers_fn``) upon
6760 re-entry to the function. The ``resultval`` has the type ``resultty``.
6765 This instruction takes a ``pers_fn`` value. This is the personality
6766 function associated with the unwinding mechanism. The optional
6767 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6769 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6770 contains the global variable representing the "type" that may be caught
6771 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6772 clause takes an array constant as its argument. Use
6773 "``[0 x i8**] undef``" for a filter which cannot throw. The
6774 '``landingpad``' instruction must contain *at least* one ``clause`` or
6775 the ``cleanup`` flag.
6780 The '``landingpad``' instruction defines the values which are set by the
6781 personality function (``pers_fn``) upon re-entry to the function, and
6782 therefore the "result type" of the ``landingpad`` instruction. As with
6783 calling conventions, how the personality function results are
6784 represented in LLVM IR is target specific.
6786 The clauses are applied in order from top to bottom. If two
6787 ``landingpad`` instructions are merged together through inlining, the
6788 clauses from the calling function are appended to the list of clauses.
6789 When the call stack is being unwound due to an exception being thrown,
6790 the exception is compared against each ``clause`` in turn. If it doesn't
6791 match any of the clauses, and the ``cleanup`` flag is not set, then
6792 unwinding continues further up the call stack.
6794 The ``landingpad`` instruction has several restrictions:
6796 - A landing pad block is a basic block which is the unwind destination
6797 of an '``invoke``' instruction.
6798 - A landing pad block must have a '``landingpad``' instruction as its
6799 first non-PHI instruction.
6800 - There can be only one '``landingpad``' instruction within the landing
6802 - A basic block that is not a landing pad block may not include a
6803 '``landingpad``' instruction.
6804 - All '``landingpad``' instructions in a function must have the same
6805 personality function.
6810 .. code-block:: llvm
6812 ;; A landing pad which can catch an integer.
6813 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6815 ;; A landing pad that is a cleanup.
6816 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6818 ;; A landing pad which can catch an integer and can only throw a double.
6819 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6821 filter [1 x i8**] [@_ZTId]
6828 LLVM supports the notion of an "intrinsic function". These functions
6829 have well known names and semantics and are required to follow certain
6830 restrictions. Overall, these intrinsics represent an extension mechanism
6831 for the LLVM language that does not require changing all of the
6832 transformations in LLVM when adding to the language (or the bitcode
6833 reader/writer, the parser, etc...).
6835 Intrinsic function names must all start with an "``llvm.``" prefix. This
6836 prefix is reserved in LLVM for intrinsic names; thus, function names may
6837 not begin with this prefix. Intrinsic functions must always be external
6838 functions: you cannot define the body of intrinsic functions. Intrinsic
6839 functions may only be used in call or invoke instructions: it is illegal
6840 to take the address of an intrinsic function. Additionally, because
6841 intrinsic functions are part of the LLVM language, it is required if any
6842 are added that they be documented here.
6844 Some intrinsic functions can be overloaded, i.e., the intrinsic
6845 represents a family of functions that perform the same operation but on
6846 different data types. Because LLVM can represent over 8 million
6847 different integer types, overloading is used commonly to allow an
6848 intrinsic function to operate on any integer type. One or more of the
6849 argument types or the result type can be overloaded to accept any
6850 integer type. Argument types may also be defined as exactly matching a
6851 previous argument's type or the result type. This allows an intrinsic
6852 function which accepts multiple arguments, but needs all of them to be
6853 of the same type, to only be overloaded with respect to a single
6854 argument or the result.
6856 Overloaded intrinsics will have the names of its overloaded argument
6857 types encoded into its function name, each preceded by a period. Only
6858 those types which are overloaded result in a name suffix. Arguments
6859 whose type is matched against another type do not. For example, the
6860 ``llvm.ctpop`` function can take an integer of any width and returns an
6861 integer of exactly the same integer width. This leads to a family of
6862 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6863 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6864 overloaded, and only one type suffix is required. Because the argument's
6865 type is matched against the return type, it does not require its own
6868 To learn how to add an intrinsic function, please see the `Extending
6869 LLVM Guide <ExtendingLLVM.html>`_.
6873 Variable Argument Handling Intrinsics
6874 -------------------------------------
6876 Variable argument support is defined in LLVM with the
6877 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6878 functions. These functions are related to the similarly named macros
6879 defined in the ``<stdarg.h>`` header file.
6881 All of these functions operate on arguments that use a target-specific
6882 value type "``va_list``". The LLVM assembly language reference manual
6883 does not define what this type is, so all transformations should be
6884 prepared to handle these functions regardless of the type used.
6886 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6887 variable argument handling intrinsic functions are used.
6889 .. code-block:: llvm
6891 ; This struct is different for every platform. For most platforms,
6892 ; it is merely an i8*.
6893 %struct.va_list = type { i8* }
6895 ; For Unix x86_64 platforms, va_list is the following struct:
6896 ; %struct.va_list = type { i32, i32, i8*, i8* }
6898 define i32 @test(i32 %X, ...) {
6899 ; Initialize variable argument processing
6900 %ap = alloca %struct.va_list
6901 %ap2 = bitcast %struct.va_list* %ap to i8*
6902 call void @llvm.va_start(i8* %ap2)
6904 ; Read a single integer argument
6905 %tmp = va_arg i8* %ap2, i32
6907 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6909 %aq2 = bitcast i8** %aq to i8*
6910 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6911 call void @llvm.va_end(i8* %aq2)
6913 ; Stop processing of arguments.
6914 call void @llvm.va_end(i8* %ap2)
6918 declare void @llvm.va_start(i8*)
6919 declare void @llvm.va_copy(i8*, i8*)
6920 declare void @llvm.va_end(i8*)
6924 '``llvm.va_start``' Intrinsic
6925 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6932 declare void @llvm.va_start(i8* <arglist>)
6937 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6938 subsequent use by ``va_arg``.
6943 The argument is a pointer to a ``va_list`` element to initialize.
6948 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6949 available in C. In a target-dependent way, it initializes the
6950 ``va_list`` element to which the argument points, so that the next call
6951 to ``va_arg`` will produce the first variable argument passed to the
6952 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6953 to know the last argument of the function as the compiler can figure
6956 '``llvm.va_end``' Intrinsic
6957 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6964 declare void @llvm.va_end(i8* <arglist>)
6969 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6970 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6975 The argument is a pointer to a ``va_list`` to destroy.
6980 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6981 available in C. In a target-dependent way, it destroys the ``va_list``
6982 element to which the argument points. Calls to
6983 :ref:`llvm.va_start <int_va_start>` and
6984 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6989 '``llvm.va_copy``' Intrinsic
6990 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6997 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
7002 The '``llvm.va_copy``' intrinsic copies the current argument position
7003 from the source argument list to the destination argument list.
7008 The first argument is a pointer to a ``va_list`` element to initialize.
7009 The second argument is a pointer to a ``va_list`` element to copy from.
7014 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
7015 available in C. In a target-dependent way, it copies the source
7016 ``va_list`` element into the destination ``va_list`` element. This
7017 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
7018 arbitrarily complex and require, for example, memory allocation.
7020 Accurate Garbage Collection Intrinsics
7021 --------------------------------------
7023 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
7024 (GC) requires the implementation and generation of these intrinsics.
7025 These intrinsics allow identification of :ref:`GC roots on the
7026 stack <int_gcroot>`, as well as garbage collector implementations that
7027 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
7028 Front-ends for type-safe garbage collected languages should generate
7029 these intrinsics to make use of the LLVM garbage collectors. For more
7030 details, see `Accurate Garbage Collection with
7031 LLVM <GarbageCollection.html>`_.
7033 The garbage collection intrinsics only operate on objects in the generic
7034 address space (address space zero).
7038 '``llvm.gcroot``' Intrinsic
7039 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7046 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
7051 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
7052 the code generator, and allows some metadata to be associated with it.
7057 The first argument specifies the address of a stack object that contains
7058 the root pointer. The second pointer (which must be either a constant or
7059 a global value address) contains the meta-data to be associated with the
7065 At runtime, a call to this intrinsic stores a null pointer into the
7066 "ptrloc" location. At compile-time, the code generator generates
7067 information to allow the runtime to find the pointer at GC safe points.
7068 The '``llvm.gcroot``' intrinsic may only be used in a function which
7069 :ref:`specifies a GC algorithm <gc>`.
7073 '``llvm.gcread``' Intrinsic
7074 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7081 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
7086 The '``llvm.gcread``' intrinsic identifies reads of references from heap
7087 locations, allowing garbage collector implementations that require read
7093 The second argument is the address to read from, which should be an
7094 address allocated from the garbage collector. The first object is a
7095 pointer to the start of the referenced object, if needed by the language
7096 runtime (otherwise null).
7101 The '``llvm.gcread``' intrinsic has the same semantics as a load
7102 instruction, but may be replaced with substantially more complex code by
7103 the garbage collector runtime, as needed. The '``llvm.gcread``'
7104 intrinsic may only be used in a function which :ref:`specifies a GC
7109 '``llvm.gcwrite``' Intrinsic
7110 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7117 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
7122 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
7123 locations, allowing garbage collector implementations that require write
7124 barriers (such as generational or reference counting collectors).
7129 The first argument is the reference to store, the second is the start of
7130 the object to store it to, and the third is the address of the field of
7131 Obj to store to. If the runtime does not require a pointer to the
7132 object, Obj may be null.
7137 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
7138 instruction, but may be replaced with substantially more complex code by
7139 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
7140 intrinsic may only be used in a function which :ref:`specifies a GC
7143 Code Generator Intrinsics
7144 -------------------------
7146 These intrinsics are provided by LLVM to expose special features that
7147 may only be implemented with code generator support.
7149 '``llvm.returnaddress``' Intrinsic
7150 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7157 declare i8 *@llvm.returnaddress(i32 <level>)
7162 The '``llvm.returnaddress``' intrinsic attempts to compute a
7163 target-specific value indicating the return address of the current
7164 function or one of its callers.
7169 The argument to this intrinsic indicates which function to return the
7170 address for. Zero indicates the calling function, one indicates its
7171 caller, etc. The argument is **required** to be a constant integer
7177 The '``llvm.returnaddress``' intrinsic either returns a pointer
7178 indicating the return address of the specified call frame, or zero if it
7179 cannot be identified. The value returned by this intrinsic is likely to
7180 be incorrect or 0 for arguments other than zero, so it should only be
7181 used for debugging purposes.
7183 Note that calling this intrinsic does not prevent function inlining or
7184 other aggressive transformations, so the value returned may not be that
7185 of the obvious source-language caller.
7187 '``llvm.frameaddress``' Intrinsic
7188 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7195 declare i8* @llvm.frameaddress(i32 <level>)
7200 The '``llvm.frameaddress``' intrinsic attempts to return the
7201 target-specific frame pointer value for the specified stack frame.
7206 The argument to this intrinsic indicates which function to return the
7207 frame pointer for. Zero indicates the calling function, one indicates
7208 its caller, etc. The argument is **required** to be a constant integer
7214 The '``llvm.frameaddress``' intrinsic either returns a pointer
7215 indicating the frame address of the specified call frame, or zero if it
7216 cannot be identified. The value returned by this intrinsic is likely to
7217 be incorrect or 0 for arguments other than zero, so it should only be
7218 used for debugging purposes.
7220 Note that calling this intrinsic does not prevent function inlining or
7221 other aggressive transformations, so the value returned may not be that
7222 of the obvious source-language caller.
7224 .. _int_read_register:
7225 .. _int_write_register:
7227 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
7228 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7235 declare i32 @llvm.read_register.i32(metadata)
7236 declare i64 @llvm.read_register.i64(metadata)
7237 declare void @llvm.write_register.i32(metadata, i32 @value)
7238 declare void @llvm.write_register.i64(metadata, i64 @value)
7239 !0 = metadata !{metadata !"sp\00"}
7244 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
7245 provides access to the named register. The register must be valid on
7246 the architecture being compiled to. The type needs to be compatible
7247 with the register being read.
7252 The '``llvm.read_register``' intrinsic returns the current value of the
7253 register, where possible. The '``llvm.write_register``' intrinsic sets
7254 the current value of the register, where possible.
7256 This is useful to implement named register global variables that need
7257 to always be mapped to a specific register, as is common practice on
7258 bare-metal programs including OS kernels.
7260 The compiler doesn't check for register availability or use of the used
7261 register in surrounding code, including inline assembly. Because of that,
7262 allocatable registers are not supported.
7264 Warning: So far it only works with the stack pointer on selected
7265 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
7266 work is needed to support other registers and even more so, allocatable
7271 '``llvm.stacksave``' Intrinsic
7272 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7279 declare i8* @llvm.stacksave()
7284 The '``llvm.stacksave``' intrinsic is used to remember the current state
7285 of the function stack, for use with
7286 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
7287 implementing language features like scoped automatic variable sized
7293 This intrinsic returns a opaque pointer value that can be passed to
7294 :ref:`llvm.stackrestore <int_stackrestore>`. When an
7295 ``llvm.stackrestore`` intrinsic is executed with a value saved from
7296 ``llvm.stacksave``, it effectively restores the state of the stack to
7297 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
7298 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
7299 were allocated after the ``llvm.stacksave`` was executed.
7301 .. _int_stackrestore:
7303 '``llvm.stackrestore``' Intrinsic
7304 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7311 declare void @llvm.stackrestore(i8* %ptr)
7316 The '``llvm.stackrestore``' intrinsic is used to restore the state of
7317 the function stack to the state it was in when the corresponding
7318 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
7319 useful for implementing language features like scoped automatic variable
7320 sized arrays in C99.
7325 See the description for :ref:`llvm.stacksave <int_stacksave>`.
7327 '``llvm.prefetch``' Intrinsic
7328 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7335 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7340 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
7341 insert a prefetch instruction if supported; otherwise, it is a noop.
7342 Prefetches have no effect on the behavior of the program but can change
7343 its performance characteristics.
7348 ``address`` is the address to be prefetched, ``rw`` is the specifier
7349 determining if the fetch should be for a read (0) or write (1), and
7350 ``locality`` is a temporal locality specifier ranging from (0) - no
7351 locality, to (3) - extremely local keep in cache. The ``cache type``
7352 specifies whether the prefetch is performed on the data (1) or
7353 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
7354 arguments must be constant integers.
7359 This intrinsic does not modify the behavior of the program. In
7360 particular, prefetches cannot trap and do not produce a value. On
7361 targets that support this intrinsic, the prefetch can provide hints to
7362 the processor cache for better performance.
7364 '``llvm.pcmarker``' Intrinsic
7365 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7372 declare void @llvm.pcmarker(i32 <id>)
7377 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7378 Counter (PC) in a region of code to simulators and other tools. The
7379 method is target specific, but it is expected that the marker will use
7380 exported symbols to transmit the PC of the marker. The marker makes no
7381 guarantees that it will remain with any specific instruction after
7382 optimizations. It is possible that the presence of a marker will inhibit
7383 optimizations. The intended use is to be inserted after optimizations to
7384 allow correlations of simulation runs.
7389 ``id`` is a numerical id identifying the marker.
7394 This intrinsic does not modify the behavior of the program. Backends
7395 that do not support this intrinsic may ignore it.
7397 '``llvm.readcyclecounter``' Intrinsic
7398 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7405 declare i64 @llvm.readcyclecounter()
7410 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
7411 counter register (or similar low latency, high accuracy clocks) on those
7412 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7413 should map to RPCC. As the backing counters overflow quickly (on the
7414 order of 9 seconds on alpha), this should only be used for small
7420 When directly supported, reading the cycle counter should not modify any
7421 memory. Implementations are allowed to either return a application
7422 specific value or a system wide value. On backends without support, this
7423 is lowered to a constant 0.
7425 Note that runtime support may be conditional on the privilege-level code is
7426 running at and the host platform.
7428 '``llvm.clear_cache``' Intrinsic
7429 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7436 declare void @llvm.clear_cache(i8*, i8*)
7441 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
7442 in the specified range to the execution unit of the processor. On
7443 targets with non-unified instruction and data cache, the implementation
7444 flushes the instruction cache.
7449 On platforms with coherent instruction and data caches (e.g. x86), this
7450 intrinsic is a nop. On platforms with non-coherent instruction and data
7451 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
7452 instructions or a system call, if cache flushing requires special
7455 The default behavior is to emit a call to ``__clear_cache`` from the run
7458 This instrinsic does *not* empty the instruction pipeline. Modifications
7459 of the current function are outside the scope of the intrinsic.
7461 Standard C Library Intrinsics
7462 -----------------------------
7464 LLVM provides intrinsics for a few important standard C library
7465 functions. These intrinsics allow source-language front-ends to pass
7466 information about the alignment of the pointer arguments to the code
7467 generator, providing opportunity for more efficient code generation.
7471 '``llvm.memcpy``' Intrinsic
7472 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7477 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
7478 integer bit width and for different address spaces. Not all targets
7479 support all bit widths however.
7483 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7484 i32 <len>, i32 <align>, i1 <isvolatile>)
7485 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7486 i64 <len>, i32 <align>, i1 <isvolatile>)
7491 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7492 source location to the destination location.
7494 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
7495 intrinsics do not return a value, takes extra alignment/isvolatile
7496 arguments and the pointers can be in specified address spaces.
7501 The first argument is a pointer to the destination, the second is a
7502 pointer to the source. The third argument is an integer argument
7503 specifying the number of bytes to copy, the fourth argument is the
7504 alignment of the source and destination locations, and the fifth is a
7505 boolean indicating a volatile access.
7507 If the call to this intrinsic has an alignment value that is not 0 or 1,
7508 then the caller guarantees that both the source and destination pointers
7509 are aligned to that boundary.
7511 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
7512 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7513 very cleanly specified and it is unwise to depend on it.
7518 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7519 source location to the destination location, which are not allowed to
7520 overlap. It copies "len" bytes of memory over. If the argument is known
7521 to be aligned to some boundary, this can be specified as the fourth
7522 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
7524 '``llvm.memmove``' Intrinsic
7525 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7530 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7531 bit width and for different address space. Not all targets support all
7536 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7537 i32 <len>, i32 <align>, i1 <isvolatile>)
7538 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7539 i64 <len>, i32 <align>, i1 <isvolatile>)
7544 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7545 source location to the destination location. It is similar to the
7546 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7549 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7550 intrinsics do not return a value, takes extra alignment/isvolatile
7551 arguments and the pointers can be in specified address spaces.
7556 The first argument is a pointer to the destination, the second is a
7557 pointer to the source. The third argument is an integer argument
7558 specifying the number of bytes to copy, the fourth argument is the
7559 alignment of the source and destination locations, and the fifth is a
7560 boolean indicating a volatile access.
7562 If the call to this intrinsic has an alignment value that is not 0 or 1,
7563 then the caller guarantees that the source and destination pointers are
7564 aligned to that boundary.
7566 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7567 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7568 not very cleanly specified and it is unwise to depend on it.
7573 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7574 source location to the destination location, which may overlap. It
7575 copies "len" bytes of memory over. If the argument is known to be
7576 aligned to some boundary, this can be specified as the fourth argument,
7577 otherwise it should be set to 0 or 1 (both meaning no alignment).
7579 '``llvm.memset.*``' Intrinsics
7580 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7585 This is an overloaded intrinsic. You can use llvm.memset on any integer
7586 bit width and for different address spaces. However, not all targets
7587 support all bit widths.
7591 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7592 i32 <len>, i32 <align>, i1 <isvolatile>)
7593 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7594 i64 <len>, i32 <align>, i1 <isvolatile>)
7599 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7600 particular byte value.
7602 Note that, unlike the standard libc function, the ``llvm.memset``
7603 intrinsic does not return a value and takes extra alignment/volatile
7604 arguments. Also, the destination can be in an arbitrary address space.
7609 The first argument is a pointer to the destination to fill, the second
7610 is the byte value with which to fill it, the third argument is an
7611 integer argument specifying the number of bytes to fill, and the fourth
7612 argument is the known alignment of the destination location.
7614 If the call to this intrinsic has an alignment value that is not 0 or 1,
7615 then the caller guarantees that the destination pointer is aligned to
7618 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7619 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7620 very cleanly specified and it is unwise to depend on it.
7625 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7626 at the destination location. If the argument is known to be aligned to
7627 some boundary, this can be specified as the fourth argument, otherwise
7628 it should be set to 0 or 1 (both meaning no alignment).
7630 '``llvm.sqrt.*``' Intrinsic
7631 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7636 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7637 floating point or vector of floating point type. Not all targets support
7642 declare float @llvm.sqrt.f32(float %Val)
7643 declare double @llvm.sqrt.f64(double %Val)
7644 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7645 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7646 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7651 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7652 returning the same value as the libm '``sqrt``' functions would. Unlike
7653 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7654 negative numbers other than -0.0 (which allows for better optimization,
7655 because there is no need to worry about errno being set).
7656 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7661 The argument and return value are floating point numbers of the same
7667 This function returns the sqrt of the specified operand if it is a
7668 nonnegative floating point number.
7670 '``llvm.powi.*``' Intrinsic
7671 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7676 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7677 floating point or vector of floating point type. Not all targets support
7682 declare float @llvm.powi.f32(float %Val, i32 %power)
7683 declare double @llvm.powi.f64(double %Val, i32 %power)
7684 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7685 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7686 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7691 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7692 specified (positive or negative) power. The order of evaluation of
7693 multiplications is not defined. When a vector of floating point type is
7694 used, the second argument remains a scalar integer value.
7699 The second argument is an integer power, and the first is a value to
7700 raise to that power.
7705 This function returns the first value raised to the second power with an
7706 unspecified sequence of rounding operations.
7708 '``llvm.sin.*``' Intrinsic
7709 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7714 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7715 floating point or vector of floating point type. Not all targets support
7720 declare float @llvm.sin.f32(float %Val)
7721 declare double @llvm.sin.f64(double %Val)
7722 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7723 declare fp128 @llvm.sin.f128(fp128 %Val)
7724 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7729 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7734 The argument and return value are floating point numbers of the same
7740 This function returns the sine of the specified operand, returning the
7741 same values as the libm ``sin`` functions would, and handles error
7742 conditions in the same way.
7744 '``llvm.cos.*``' Intrinsic
7745 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7750 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7751 floating point or vector of floating point type. Not all targets support
7756 declare float @llvm.cos.f32(float %Val)
7757 declare double @llvm.cos.f64(double %Val)
7758 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7759 declare fp128 @llvm.cos.f128(fp128 %Val)
7760 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7765 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7770 The argument and return value are floating point numbers of the same
7776 This function returns the cosine of the specified operand, returning the
7777 same values as the libm ``cos`` functions would, and handles error
7778 conditions in the same way.
7780 '``llvm.pow.*``' Intrinsic
7781 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7786 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7787 floating point or vector of floating point type. Not all targets support
7792 declare float @llvm.pow.f32(float %Val, float %Power)
7793 declare double @llvm.pow.f64(double %Val, double %Power)
7794 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7795 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7796 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7801 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7802 specified (positive or negative) power.
7807 The second argument is a floating point power, and the first is a value
7808 to raise to that power.
7813 This function returns the first value raised to the second power,
7814 returning the same values as the libm ``pow`` functions would, and
7815 handles error conditions in the same way.
7817 '``llvm.exp.*``' Intrinsic
7818 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7823 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7824 floating point or vector of floating point type. Not all targets support
7829 declare float @llvm.exp.f32(float %Val)
7830 declare double @llvm.exp.f64(double %Val)
7831 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7832 declare fp128 @llvm.exp.f128(fp128 %Val)
7833 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7838 The '``llvm.exp.*``' intrinsics perform the exp function.
7843 The argument and return value are floating point numbers of the same
7849 This function returns the same values as the libm ``exp`` functions
7850 would, and handles error conditions in the same way.
7852 '``llvm.exp2.*``' Intrinsic
7853 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7858 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7859 floating point or vector of floating point type. Not all targets support
7864 declare float @llvm.exp2.f32(float %Val)
7865 declare double @llvm.exp2.f64(double %Val)
7866 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7867 declare fp128 @llvm.exp2.f128(fp128 %Val)
7868 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7873 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7878 The argument and return value are floating point numbers of the same
7884 This function returns the same values as the libm ``exp2`` functions
7885 would, and handles error conditions in the same way.
7887 '``llvm.log.*``' Intrinsic
7888 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7893 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7894 floating point or vector of floating point type. Not all targets support
7899 declare float @llvm.log.f32(float %Val)
7900 declare double @llvm.log.f64(double %Val)
7901 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7902 declare fp128 @llvm.log.f128(fp128 %Val)
7903 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7908 The '``llvm.log.*``' intrinsics perform the log function.
7913 The argument and return value are floating point numbers of the same
7919 This function returns the same values as the libm ``log`` functions
7920 would, and handles error conditions in the same way.
7922 '``llvm.log10.*``' Intrinsic
7923 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7928 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7929 floating point or vector of floating point type. Not all targets support
7934 declare float @llvm.log10.f32(float %Val)
7935 declare double @llvm.log10.f64(double %Val)
7936 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7937 declare fp128 @llvm.log10.f128(fp128 %Val)
7938 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7943 The '``llvm.log10.*``' intrinsics perform the log10 function.
7948 The argument and return value are floating point numbers of the same
7954 This function returns the same values as the libm ``log10`` functions
7955 would, and handles error conditions in the same way.
7957 '``llvm.log2.*``' Intrinsic
7958 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7963 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7964 floating point or vector of floating point type. Not all targets support
7969 declare float @llvm.log2.f32(float %Val)
7970 declare double @llvm.log2.f64(double %Val)
7971 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7972 declare fp128 @llvm.log2.f128(fp128 %Val)
7973 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7978 The '``llvm.log2.*``' intrinsics perform the log2 function.
7983 The argument and return value are floating point numbers of the same
7989 This function returns the same values as the libm ``log2`` functions
7990 would, and handles error conditions in the same way.
7992 '``llvm.fma.*``' Intrinsic
7993 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7998 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7999 floating point or vector of floating point type. Not all targets support
8004 declare float @llvm.fma.f32(float %a, float %b, float %c)
8005 declare double @llvm.fma.f64(double %a, double %b, double %c)
8006 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
8007 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
8008 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
8013 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
8019 The argument and return value are floating point numbers of the same
8025 This function returns the same values as the libm ``fma`` functions
8026 would, and does not set errno.
8028 '``llvm.fabs.*``' Intrinsic
8029 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8034 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
8035 floating point or vector of floating point type. Not all targets support
8040 declare float @llvm.fabs.f32(float %Val)
8041 declare double @llvm.fabs.f64(double %Val)
8042 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
8043 declare fp128 @llvm.fabs.f128(fp128 %Val)
8044 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
8049 The '``llvm.fabs.*``' intrinsics return the absolute value of the
8055 The argument and return value are floating point numbers of the same
8061 This function returns the same values as the libm ``fabs`` functions
8062 would, and handles error conditions in the same way.
8064 '``llvm.minnum.*``' Intrinsic
8065 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8070 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
8071 floating point or vector of floating point type. Not all targets support
8076 declare float @llvm.minnum.f32(float %Val0, float %Val1)
8077 declare double @llvm.minnum.f64(double %Val0, double %Val1)
8078 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8079 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
8080 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8085 The '``llvm.minnum.*``' intrinsics return the minimum of the two
8092 The arguments and return value are floating point numbers of the same
8098 Follows the IEEE-754 semantics for minNum, which also match for libm's
8101 If either operand is a NaN, returns the other non-NaN operand. Returns
8102 NaN only if both operands are NaN. If the operands compare equal,
8103 returns a value that compares equal to both operands. This means that
8104 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8106 '``llvm.maxnum.*``' Intrinsic
8107 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8112 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
8113 floating point or vector of floating point type. Not all targets support
8118 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
8119 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
8120 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8121 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
8122 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8127 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
8134 The arguments and return value are floating point numbers of the same
8139 Follows the IEEE-754 semantics for maxNum, which also match for libm's
8142 If either operand is a NaN, returns the other non-NaN operand. Returns
8143 NaN only if both operands are NaN. If the operands compare equal,
8144 returns a value that compares equal to both operands. This means that
8145 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8147 '``llvm.copysign.*``' Intrinsic
8148 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8153 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
8154 floating point or vector of floating point type. Not all targets support
8159 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
8160 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
8161 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
8162 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
8163 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
8168 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
8169 first operand and the sign of the second operand.
8174 The arguments and return value are floating point numbers of the same
8180 This function returns the same values as the libm ``copysign``
8181 functions would, and handles error conditions in the same way.
8183 '``llvm.floor.*``' Intrinsic
8184 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8189 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
8190 floating point or vector of floating point type. Not all targets support
8195 declare float @llvm.floor.f32(float %Val)
8196 declare double @llvm.floor.f64(double %Val)
8197 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
8198 declare fp128 @llvm.floor.f128(fp128 %Val)
8199 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
8204 The '``llvm.floor.*``' intrinsics return the floor of the operand.
8209 The argument and return value are floating point numbers of the same
8215 This function returns the same values as the libm ``floor`` functions
8216 would, and handles error conditions in the same way.
8218 '``llvm.ceil.*``' Intrinsic
8219 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8224 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
8225 floating point or vector of floating point type. Not all targets support
8230 declare float @llvm.ceil.f32(float %Val)
8231 declare double @llvm.ceil.f64(double %Val)
8232 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
8233 declare fp128 @llvm.ceil.f128(fp128 %Val)
8234 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
8239 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
8244 The argument and return value are floating point numbers of the same
8250 This function returns the same values as the libm ``ceil`` functions
8251 would, and handles error conditions in the same way.
8253 '``llvm.trunc.*``' Intrinsic
8254 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8259 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
8260 floating point or vector of floating point type. Not all targets support
8265 declare float @llvm.trunc.f32(float %Val)
8266 declare double @llvm.trunc.f64(double %Val)
8267 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
8268 declare fp128 @llvm.trunc.f128(fp128 %Val)
8269 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
8274 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
8275 nearest integer not larger in magnitude than the operand.
8280 The argument and return value are floating point numbers of the same
8286 This function returns the same values as the libm ``trunc`` functions
8287 would, and handles error conditions in the same way.
8289 '``llvm.rint.*``' Intrinsic
8290 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8295 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
8296 floating point or vector of floating point type. Not all targets support
8301 declare float @llvm.rint.f32(float %Val)
8302 declare double @llvm.rint.f64(double %Val)
8303 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
8304 declare fp128 @llvm.rint.f128(fp128 %Val)
8305 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
8310 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
8311 nearest integer. It may raise an inexact floating-point exception if the
8312 operand isn't an integer.
8317 The argument and return value are floating point numbers of the same
8323 This function returns the same values as the libm ``rint`` functions
8324 would, and handles error conditions in the same way.
8326 '``llvm.nearbyint.*``' Intrinsic
8327 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8332 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
8333 floating point or vector of floating point type. Not all targets support
8338 declare float @llvm.nearbyint.f32(float %Val)
8339 declare double @llvm.nearbyint.f64(double %Val)
8340 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
8341 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
8342 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
8347 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
8353 The argument and return value are floating point numbers of the same
8359 This function returns the same values as the libm ``nearbyint``
8360 functions would, and handles error conditions in the same way.
8362 '``llvm.round.*``' Intrinsic
8363 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8368 This is an overloaded intrinsic. You can use ``llvm.round`` on any
8369 floating point or vector of floating point type. Not all targets support
8374 declare float @llvm.round.f32(float %Val)
8375 declare double @llvm.round.f64(double %Val)
8376 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
8377 declare fp128 @llvm.round.f128(fp128 %Val)
8378 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
8383 The '``llvm.round.*``' intrinsics returns the operand rounded to the
8389 The argument and return value are floating point numbers of the same
8395 This function returns the same values as the libm ``round``
8396 functions would, and handles error conditions in the same way.
8398 Bit Manipulation Intrinsics
8399 ---------------------------
8401 LLVM provides intrinsics for a few important bit manipulation
8402 operations. These allow efficient code generation for some algorithms.
8404 '``llvm.bswap.*``' Intrinsics
8405 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8410 This is an overloaded intrinsic function. You can use bswap on any
8411 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
8415 declare i16 @llvm.bswap.i16(i16 <id>)
8416 declare i32 @llvm.bswap.i32(i32 <id>)
8417 declare i64 @llvm.bswap.i64(i64 <id>)
8422 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
8423 values with an even number of bytes (positive multiple of 16 bits).
8424 These are useful for performing operations on data that is not in the
8425 target's native byte order.
8430 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
8431 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
8432 intrinsic returns an i32 value that has the four bytes of the input i32
8433 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
8434 returned i32 will have its bytes in 3, 2, 1, 0 order. The
8435 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
8436 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
8439 '``llvm.ctpop.*``' Intrinsic
8440 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8445 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
8446 bit width, or on any vector with integer elements. Not all targets
8447 support all bit widths or vector types, however.
8451 declare i8 @llvm.ctpop.i8(i8 <src>)
8452 declare i16 @llvm.ctpop.i16(i16 <src>)
8453 declare i32 @llvm.ctpop.i32(i32 <src>)
8454 declare i64 @llvm.ctpop.i64(i64 <src>)
8455 declare i256 @llvm.ctpop.i256(i256 <src>)
8456 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
8461 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
8467 The only argument is the value to be counted. The argument may be of any
8468 integer type, or a vector with integer elements. The return type must
8469 match the argument type.
8474 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
8475 each element of a vector.
8477 '``llvm.ctlz.*``' Intrinsic
8478 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8483 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
8484 integer bit width, or any vector whose elements are integers. Not all
8485 targets support all bit widths or vector types, however.
8489 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
8490 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
8491 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
8492 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
8493 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
8494 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8499 The '``llvm.ctlz``' family of intrinsic functions counts the number of
8500 leading zeros in a variable.
8505 The first argument is the value to be counted. This argument may be of
8506 any integer type, or a vectory with integer element type. The return
8507 type must match the first argument type.
8509 The second argument must be a constant and is a flag to indicate whether
8510 the intrinsic should ensure that a zero as the first argument produces a
8511 defined result. Historically some architectures did not provide a
8512 defined result for zero values as efficiently, and many algorithms are
8513 now predicated on avoiding zero-value inputs.
8518 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
8519 zeros in a variable, or within each element of the vector. If
8520 ``src == 0`` then the result is the size in bits of the type of ``src``
8521 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8522 ``llvm.ctlz(i32 2) = 30``.
8524 '``llvm.cttz.*``' Intrinsic
8525 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8530 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
8531 integer bit width, or any vector of integer elements. Not all targets
8532 support all bit widths or vector types, however.
8536 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
8537 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
8538 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
8539 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
8540 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
8541 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8546 The '``llvm.cttz``' family of intrinsic functions counts the number of
8552 The first argument is the value to be counted. This argument may be of
8553 any integer type, or a vectory with integer element type. The return
8554 type must match the first argument type.
8556 The second argument must be a constant and is a flag to indicate whether
8557 the intrinsic should ensure that a zero as the first argument produces a
8558 defined result. Historically some architectures did not provide a
8559 defined result for zero values as efficiently, and many algorithms are
8560 now predicated on avoiding zero-value inputs.
8565 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
8566 zeros in a variable, or within each element of a vector. If ``src == 0``
8567 then the result is the size in bits of the type of ``src`` if
8568 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8569 ``llvm.cttz(2) = 1``.
8571 Arithmetic with Overflow Intrinsics
8572 -----------------------------------
8574 LLVM provides intrinsics for some arithmetic with overflow operations.
8576 '``llvm.sadd.with.overflow.*``' Intrinsics
8577 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8582 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
8583 on any integer bit width.
8587 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8588 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8589 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8594 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8595 a signed addition of the two arguments, and indicate whether an overflow
8596 occurred during the signed summation.
8601 The arguments (%a and %b) and the first element of the result structure
8602 may be of integer types of any bit width, but they must have the same
8603 bit width. The second element of the result structure must be of type
8604 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8610 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8611 a signed addition of the two variables. They return a structure --- the
8612 first element of which is the signed summation, and the second element
8613 of which is a bit specifying if the signed summation resulted in an
8619 .. code-block:: llvm
8621 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8622 %sum = extractvalue {i32, i1} %res, 0
8623 %obit = extractvalue {i32, i1} %res, 1
8624 br i1 %obit, label %overflow, label %normal
8626 '``llvm.uadd.with.overflow.*``' Intrinsics
8627 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8632 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8633 on any integer bit width.
8637 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8638 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8639 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8644 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8645 an unsigned addition of the two arguments, and indicate whether a carry
8646 occurred during the unsigned summation.
8651 The arguments (%a and %b) and the first element of the result structure
8652 may be of integer types of any bit width, but they must have the same
8653 bit width. The second element of the result structure must be of type
8654 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8660 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8661 an unsigned addition of the two arguments. They return a structure --- the
8662 first element of which is the sum, and the second element of which is a
8663 bit specifying if the unsigned summation resulted in a carry.
8668 .. code-block:: llvm
8670 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8671 %sum = extractvalue {i32, i1} %res, 0
8672 %obit = extractvalue {i32, i1} %res, 1
8673 br i1 %obit, label %carry, label %normal
8675 '``llvm.ssub.with.overflow.*``' Intrinsics
8676 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8681 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8682 on any integer bit width.
8686 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8687 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8688 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8693 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8694 a signed subtraction of the two arguments, and indicate whether an
8695 overflow occurred during the signed subtraction.
8700 The arguments (%a and %b) and the first element of the result structure
8701 may be of integer types of any bit width, but they must have the same
8702 bit width. The second element of the result structure must be of type
8703 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8709 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8710 a signed subtraction of the two arguments. They return a structure --- the
8711 first element of which is the subtraction, and the second element of
8712 which is a bit specifying if the signed subtraction resulted in an
8718 .. code-block:: llvm
8720 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8721 %sum = extractvalue {i32, i1} %res, 0
8722 %obit = extractvalue {i32, i1} %res, 1
8723 br i1 %obit, label %overflow, label %normal
8725 '``llvm.usub.with.overflow.*``' Intrinsics
8726 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8731 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8732 on any integer bit width.
8736 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8737 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8738 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8743 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8744 an unsigned subtraction of the two arguments, and indicate whether an
8745 overflow occurred during the unsigned subtraction.
8750 The arguments (%a and %b) and the first element of the result structure
8751 may be of integer types of any bit width, but they must have the same
8752 bit width. The second element of the result structure must be of type
8753 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8759 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8760 an unsigned subtraction of the two arguments. They return a structure ---
8761 the first element of which is the subtraction, and the second element of
8762 which is a bit specifying if the unsigned subtraction resulted in an
8768 .. code-block:: llvm
8770 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8771 %sum = extractvalue {i32, i1} %res, 0
8772 %obit = extractvalue {i32, i1} %res, 1
8773 br i1 %obit, label %overflow, label %normal
8775 '``llvm.smul.with.overflow.*``' Intrinsics
8776 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8781 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8782 on any integer bit width.
8786 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8787 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8788 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8793 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8794 a signed multiplication of the two arguments, and indicate whether an
8795 overflow occurred during the signed multiplication.
8800 The arguments (%a and %b) and the first element of the result structure
8801 may be of integer types of any bit width, but they must have the same
8802 bit width. The second element of the result structure must be of type
8803 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8809 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8810 a signed multiplication of the two arguments. They return a structure ---
8811 the first element of which is the multiplication, and the second element
8812 of which is a bit specifying if the signed multiplication resulted in an
8818 .. code-block:: llvm
8820 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8821 %sum = extractvalue {i32, i1} %res, 0
8822 %obit = extractvalue {i32, i1} %res, 1
8823 br i1 %obit, label %overflow, label %normal
8825 '``llvm.umul.with.overflow.*``' Intrinsics
8826 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8831 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8832 on any integer bit width.
8836 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8837 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8838 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8843 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8844 a unsigned multiplication of the two arguments, and indicate whether an
8845 overflow occurred during the unsigned multiplication.
8850 The arguments (%a and %b) and the first element of the result structure
8851 may be of integer types of any bit width, but they must have the same
8852 bit width. The second element of the result structure must be of type
8853 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8859 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8860 an unsigned multiplication of the two arguments. They return a structure ---
8861 the first element of which is the multiplication, and the second
8862 element of which is a bit specifying if the unsigned multiplication
8863 resulted in an overflow.
8868 .. code-block:: llvm
8870 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8871 %sum = extractvalue {i32, i1} %res, 0
8872 %obit = extractvalue {i32, i1} %res, 1
8873 br i1 %obit, label %overflow, label %normal
8875 Specialised Arithmetic Intrinsics
8876 ---------------------------------
8878 '``llvm.fmuladd.*``' Intrinsic
8879 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8886 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8887 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8892 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8893 expressions that can be fused if the code generator determines that (a) the
8894 target instruction set has support for a fused operation, and (b) that the
8895 fused operation is more efficient than the equivalent, separate pair of mul
8896 and add instructions.
8901 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8902 multiplicands, a and b, and an addend c.
8911 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8913 is equivalent to the expression a \* b + c, except that rounding will
8914 not be performed between the multiplication and addition steps if the
8915 code generator fuses the operations. Fusion is not guaranteed, even if
8916 the target platform supports it. If a fused multiply-add is required the
8917 corresponding llvm.fma.\* intrinsic function should be used
8918 instead. This never sets errno, just as '``llvm.fma.*``'.
8923 .. code-block:: llvm
8925 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
8927 Half Precision Floating Point Intrinsics
8928 ----------------------------------------
8930 For most target platforms, half precision floating point is a
8931 storage-only format. This means that it is a dense encoding (in memory)
8932 but does not support computation in the format.
8934 This means that code must first load the half-precision floating point
8935 value as an i16, then convert it to float with
8936 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8937 then be performed on the float value (including extending to double
8938 etc). To store the value back to memory, it is first converted to float
8939 if needed, then converted to i16 with
8940 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8943 .. _int_convert_to_fp16:
8945 '``llvm.convert.to.fp16``' Intrinsic
8946 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8953 declare i16 @llvm.convert.to.fp16.f32(float %a)
8954 declare i16 @llvm.convert.to.fp16.f64(double %a)
8959 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
8960 conventional floating point type to half precision floating point format.
8965 The intrinsic function contains single argument - the value to be
8971 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
8972 conventional floating point format to half precision floating point format. The
8973 return value is an ``i16`` which contains the converted number.
8978 .. code-block:: llvm
8980 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
8981 store i16 %res, i16* @x, align 2
8983 .. _int_convert_from_fp16:
8985 '``llvm.convert.from.fp16``' Intrinsic
8986 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8993 declare float @llvm.convert.from.fp16.f32(i16 %a)
8994 declare double @llvm.convert.from.fp16.f64(i16 %a)
8999 The '``llvm.convert.from.fp16``' intrinsic function performs a
9000 conversion from half precision floating point format to single precision
9001 floating point format.
9006 The intrinsic function contains single argument - the value to be
9012 The '``llvm.convert.from.fp16``' intrinsic function performs a
9013 conversion from half single precision floating point format to single
9014 precision floating point format. The input half-float value is
9015 represented by an ``i16`` value.
9020 .. code-block:: llvm
9022 %a = load i16* @x, align 2
9023 %res = call float @llvm.convert.from.fp16(i16 %a)
9028 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
9029 prefix), are described in the `LLVM Source Level
9030 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
9033 Exception Handling Intrinsics
9034 -----------------------------
9036 The LLVM exception handling intrinsics (which all start with
9037 ``llvm.eh.`` prefix), are described in the `LLVM Exception
9038 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
9042 Trampoline Intrinsics
9043 ---------------------
9045 These intrinsics make it possible to excise one parameter, marked with
9046 the :ref:`nest <nest>` attribute, from a function. The result is a
9047 callable function pointer lacking the nest parameter - the caller does
9048 not need to provide a value for it. Instead, the value to use is stored
9049 in advance in a "trampoline", a block of memory usually allocated on the
9050 stack, which also contains code to splice the nest value into the
9051 argument list. This is used to implement the GCC nested function address
9054 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
9055 then the resulting function pointer has signature ``i32 (i32, i32)*``.
9056 It can be created as follows:
9058 .. code-block:: llvm
9060 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
9061 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
9062 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
9063 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
9064 %fp = bitcast i8* %p to i32 (i32, i32)*
9066 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
9067 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
9071 '``llvm.init.trampoline``' Intrinsic
9072 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9079 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
9084 This fills the memory pointed to by ``tramp`` with executable code,
9085 turning it into a trampoline.
9090 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
9091 pointers. The ``tramp`` argument must point to a sufficiently large and
9092 sufficiently aligned block of memory; this memory is written to by the
9093 intrinsic. Note that the size and the alignment are target-specific -
9094 LLVM currently provides no portable way of determining them, so a
9095 front-end that generates this intrinsic needs to have some
9096 target-specific knowledge. The ``func`` argument must hold a function
9097 bitcast to an ``i8*``.
9102 The block of memory pointed to by ``tramp`` is filled with target
9103 dependent code, turning it into a function. Then ``tramp`` needs to be
9104 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
9105 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
9106 function's signature is the same as that of ``func`` with any arguments
9107 marked with the ``nest`` attribute removed. At most one such ``nest``
9108 argument is allowed, and it must be of pointer type. Calling the new
9109 function is equivalent to calling ``func`` with the same argument list,
9110 but with ``nval`` used for the missing ``nest`` argument. If, after
9111 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
9112 modified, then the effect of any later call to the returned function
9113 pointer is undefined.
9117 '``llvm.adjust.trampoline``' Intrinsic
9118 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9125 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
9130 This performs any required machine-specific adjustment to the address of
9131 a trampoline (passed as ``tramp``).
9136 ``tramp`` must point to a block of memory which already has trampoline
9137 code filled in by a previous call to
9138 :ref:`llvm.init.trampoline <int_it>`.
9143 On some architectures the address of the code to be executed needs to be
9144 different than the address where the trampoline is actually stored. This
9145 intrinsic returns the executable address corresponding to ``tramp``
9146 after performing the required machine specific adjustments. The pointer
9147 returned can then be :ref:`bitcast and executed <int_trampoline>`.
9152 This class of intrinsics provides information about the lifetime of
9153 memory objects and ranges where variables are immutable.
9157 '``llvm.lifetime.start``' Intrinsic
9158 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9165 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
9170 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
9176 The first argument is a constant integer representing the size of the
9177 object, or -1 if it is variable sized. The second argument is a pointer
9183 This intrinsic indicates that before this point in the code, the value
9184 of the memory pointed to by ``ptr`` is dead. This means that it is known
9185 to never be used and has an undefined value. A load from the pointer
9186 that precedes this intrinsic can be replaced with ``'undef'``.
9190 '``llvm.lifetime.end``' Intrinsic
9191 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9198 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
9203 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
9209 The first argument is a constant integer representing the size of the
9210 object, or -1 if it is variable sized. The second argument is a pointer
9216 This intrinsic indicates that after this point in the code, the value of
9217 the memory pointed to by ``ptr`` is dead. This means that it is known to
9218 never be used and has an undefined value. Any stores into the memory
9219 object following this intrinsic may be removed as dead.
9221 '``llvm.invariant.start``' Intrinsic
9222 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9229 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
9234 The '``llvm.invariant.start``' intrinsic specifies that the contents of
9235 a memory object will not change.
9240 The first argument is a constant integer representing the size of the
9241 object, or -1 if it is variable sized. The second argument is a pointer
9247 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
9248 the return value, the referenced memory location is constant and
9251 '``llvm.invariant.end``' Intrinsic
9252 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9259 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
9264 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
9265 memory object are mutable.
9270 The first argument is the matching ``llvm.invariant.start`` intrinsic.
9271 The second argument is a constant integer representing the size of the
9272 object, or -1 if it is variable sized and the third argument is a
9273 pointer to the object.
9278 This intrinsic indicates that the memory is mutable again.
9283 This class of intrinsics is designed to be generic and has no specific
9286 '``llvm.var.annotation``' Intrinsic
9287 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9294 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9299 The '``llvm.var.annotation``' intrinsic.
9304 The first argument is a pointer to a value, the second is a pointer to a
9305 global string, the third is a pointer to a global string which is the
9306 source file name, and the last argument is the line number.
9311 This intrinsic allows annotation of local variables with arbitrary
9312 strings. This can be useful for special purpose optimizations that want
9313 to look for these annotations. These have no other defined use; they are
9314 ignored by code generation and optimization.
9316 '``llvm.ptr.annotation.*``' Intrinsic
9317 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9322 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
9323 pointer to an integer of any width. *NOTE* you must specify an address space for
9324 the pointer. The identifier for the default address space is the integer
9329 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9330 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
9331 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
9332 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
9333 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
9338 The '``llvm.ptr.annotation``' intrinsic.
9343 The first argument is a pointer to an integer value of arbitrary bitwidth
9344 (result of some expression), the second is a pointer to a global string, the
9345 third is a pointer to a global string which is the source file name, and the
9346 last argument is the line number. It returns the value of the first argument.
9351 This intrinsic allows annotation of a pointer to an integer with arbitrary
9352 strings. This can be useful for special purpose optimizations that want to look
9353 for these annotations. These have no other defined use; they are ignored by code
9354 generation and optimization.
9356 '``llvm.annotation.*``' Intrinsic
9357 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9362 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
9363 any integer bit width.
9367 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
9368 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
9369 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
9370 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
9371 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
9376 The '``llvm.annotation``' intrinsic.
9381 The first argument is an integer value (result of some expression), the
9382 second is a pointer to a global string, the third is a pointer to a
9383 global string which is the source file name, and the last argument is
9384 the line number. It returns the value of the first argument.
9389 This intrinsic allows annotations to be put on arbitrary expressions
9390 with arbitrary strings. This can be useful for special purpose
9391 optimizations that want to look for these annotations. These have no
9392 other defined use; they are ignored by code generation and optimization.
9394 '``llvm.trap``' Intrinsic
9395 ^^^^^^^^^^^^^^^^^^^^^^^^^
9402 declare void @llvm.trap() noreturn nounwind
9407 The '``llvm.trap``' intrinsic.
9417 This intrinsic is lowered to the target dependent trap instruction. If
9418 the target does not have a trap instruction, this intrinsic will be
9419 lowered to a call of the ``abort()`` function.
9421 '``llvm.debugtrap``' Intrinsic
9422 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9429 declare void @llvm.debugtrap() nounwind
9434 The '``llvm.debugtrap``' intrinsic.
9444 This intrinsic is lowered to code which is intended to cause an
9445 execution trap with the intention of requesting the attention of a
9448 '``llvm.stackprotector``' Intrinsic
9449 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9456 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
9461 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
9462 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
9463 is placed on the stack before local variables.
9468 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
9469 The first argument is the value loaded from the stack guard
9470 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
9471 enough space to hold the value of the guard.
9476 This intrinsic causes the prologue/epilogue inserter to force the position of
9477 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
9478 to ensure that if a local variable on the stack is overwritten, it will destroy
9479 the value of the guard. When the function exits, the guard on the stack is
9480 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
9481 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
9482 calling the ``__stack_chk_fail()`` function.
9484 '``llvm.stackprotectorcheck``' Intrinsic
9485 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9492 declare void @llvm.stackprotectorcheck(i8** <guard>)
9497 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
9498 created stack protector and if they are not equal calls the
9499 ``__stack_chk_fail()`` function.
9504 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
9505 the variable ``@__stack_chk_guard``.
9510 This intrinsic is provided to perform the stack protector check by comparing
9511 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
9512 values do not match call the ``__stack_chk_fail()`` function.
9514 The reason to provide this as an IR level intrinsic instead of implementing it
9515 via other IR operations is that in order to perform this operation at the IR
9516 level without an intrinsic, one would need to create additional basic blocks to
9517 handle the success/failure cases. This makes it difficult to stop the stack
9518 protector check from disrupting sibling tail calls in Codegen. With this
9519 intrinsic, we are able to generate the stack protector basic blocks late in
9520 codegen after the tail call decision has occurred.
9522 '``llvm.objectsize``' Intrinsic
9523 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9530 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
9531 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
9536 The ``llvm.objectsize`` intrinsic is designed to provide information to
9537 the optimizers to determine at compile time whether a) an operation
9538 (like memcpy) will overflow a buffer that corresponds to an object, or
9539 b) that a runtime check for overflow isn't necessary. An object in this
9540 context means an allocation of a specific class, structure, array, or
9546 The ``llvm.objectsize`` intrinsic takes two arguments. The first
9547 argument is a pointer to or into the ``object``. The second argument is
9548 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
9549 or -1 (if false) when the object size is unknown. The second argument
9550 only accepts constants.
9555 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
9556 the size of the object concerned. If the size cannot be determined at
9557 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
9558 on the ``min`` argument).
9560 '``llvm.expect``' Intrinsic
9561 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9566 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
9571 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
9572 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9573 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9578 The ``llvm.expect`` intrinsic provides information about expected (the
9579 most probable) value of ``val``, which can be used by optimizers.
9584 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
9585 a value. The second argument is an expected value, this needs to be a
9586 constant value, variables are not allowed.
9591 This intrinsic is lowered to the ``val``.
9593 '``llvm.assume``' Intrinsic
9594 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9601 declare void @llvm.assume(i1 %cond)
9606 The ``llvm.assume`` allows the optimizer to assume that the provided
9607 condition is true. This information can then be used in simplifying other parts
9613 The condition which the optimizer may assume is always true.
9618 The intrinsic allows the optimizer to assume that the provided condition is
9619 always true whenever the control flow reaches the intrinsic call. No code is
9620 generated for this intrinsic, and instructions that contribute only to the
9621 provided condition are not used for code generation. If the condition is
9622 violated during execution, the behavior is undefined.
9624 Please note that optimizer might limit the transformations performed on values
9625 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
9626 only used to form the intrinsic's input argument. This might prove undesirable
9627 if the extra information provided by the ``llvm.assume`` intrinsic does cause
9628 sufficient overall improvement in code quality. For this reason,
9629 ``llvm.assume`` should not be used to document basic mathematical invariants
9630 that the optimizer can otherwise deduce or facts that are of little use to the
9633 '``llvm.donothing``' Intrinsic
9634 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9641 declare void @llvm.donothing() nounwind readnone
9646 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
9647 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
9648 with an invoke instruction.
9658 This intrinsic does nothing, and it's removed by optimizers and ignored
9661 Stack Map Intrinsics
9662 --------------------
9664 LLVM provides experimental intrinsics to support runtime patching
9665 mechanisms commonly desired in dynamic language JITs. These intrinsics
9666 are described in :doc:`StackMaps`.