1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][a-zA-Z$._][a-zA-Z$._0-9]*``'. Identifiers which require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves.
83 #. Unnamed values are represented as an unsigned numeric value with
84 their prefix. For example, ``%12``, ``@2``, ``%44``.
85 #. Constants, which are described in the section Constants_ below.
87 LLVM requires that values start with a prefix for two reasons: Compilers
88 don't need to worry about name clashes with reserved words, and the set
89 of reserved words may be expanded in the future without penalty.
90 Additionally, unnamed identifiers allow a compiler to quickly come up
91 with a temporary variable without having to avoid symbol table
94 Reserved words in LLVM are very similar to reserved words in other
95 languages. There are keywords for different opcodes ('``add``',
96 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
97 '``i32``', etc...), and others. These reserved words cannot conflict
98 with variable names, because none of them start with a prefix character
101 Here is an example of LLVM code to multiply the integer variable
108 %result = mul i32 %X, 8
110 After strength reduction:
114 %result = shl i32 %X, 3
120 %0 = add i32 %X, %X ; yields i32:%0
121 %1 = add i32 %0, %0 ; yields i32:%1
122 %result = add i32 %1, %1
124 This last way of multiplying ``%X`` by 8 illustrates several important
125 lexical features of LLVM:
127 #. Comments are delimited with a '``;``' and go until the end of line.
128 #. Unnamed temporaries are created when the result of a computation is
129 not assigned to a named value.
130 #. Unnamed temporaries are numbered sequentially (using a per-function
131 incrementing counter, starting with 0). Note that basic blocks are
132 included in this numbering. For example, if the entry basic block is not
133 given a label name, then it will get number 0.
135 It also shows a convention that we follow in this document. When
136 demonstrating instructions, we will follow an instruction with a comment
137 that defines the type and name of value produced.
145 LLVM programs are composed of ``Module``'s, each of which is a
146 translation unit of the input programs. Each module consists of
147 functions, global variables, and symbol table entries. Modules may be
148 combined together with the LLVM linker, which merges function (and
149 global variable) definitions, resolves forward declarations, and merges
150 symbol table entries. Here is an example of the "hello world" module:
154 ; Declare the string constant as a global constant.
155 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
157 ; External declaration of the puts function
158 declare i32 @puts(i8* nocapture) nounwind
160 ; Definition of main function
161 define i32 @main() { ; i32()*
162 ; Convert [13 x i8]* to i8 *...
163 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
165 ; Call puts function to write out the string to stdout.
166 call i32 @puts(i8* %cast210)
171 !1 = metadata !{i32 42}
174 This example is made up of a :ref:`global variable <globalvars>` named
175 "``.str``", an external declaration of the "``puts``" function, a
176 :ref:`function definition <functionstructure>` for "``main``" and
177 :ref:`named metadata <namedmetadatastructure>` "``foo``".
179 In general, a module is made up of a list of global values (where both
180 functions and global variables are global values). Global values are
181 represented by a pointer to a memory location (in this case, a pointer
182 to an array of char, and a pointer to a function), and have one of the
183 following :ref:`linkage types <linkage>`.
190 All Global Variables and Functions have one of the following types of
194 Global values with "``private``" linkage are only directly
195 accessible by objects in the current module. In particular, linking
196 code into a module with an private global value may cause the
197 private to be renamed as necessary to avoid collisions. Because the
198 symbol is private to the module, all references can be updated. This
199 doesn't show up in any symbol table in the object file.
201 Similar to private, but the value shows as a local symbol
202 (``STB_LOCAL`` in the case of ELF) in the object file. This
203 corresponds to the notion of the '``static``' keyword in C.
204 ``available_externally``
205 Globals with "``available_externally``" linkage are never emitted
206 into the object file corresponding to the LLVM module. They exist to
207 allow inlining and other optimizations to take place given knowledge
208 of the definition of the global, which is known to be somewhere
209 outside the module. Globals with ``available_externally`` linkage
210 are allowed to be discarded at will, and are otherwise the same as
211 ``linkonce_odr``. This linkage type is only allowed on definitions,
214 Globals with "``linkonce``" linkage are merged with other globals of
215 the same name when linkage occurs. This can be used to implement
216 some forms of inline functions, templates, or other code which must
217 be generated in each translation unit that uses it, but where the
218 body may be overridden with a more definitive definition later.
219 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
220 that ``linkonce`` linkage does not actually allow the optimizer to
221 inline the body of this function into callers because it doesn't
222 know if this definition of the function is the definitive definition
223 within the program or whether it will be overridden by a stronger
224 definition. To enable inlining and other optimizations, use
225 "``linkonce_odr``" linkage.
227 "``weak``" linkage has the same merging semantics as ``linkonce``
228 linkage, except that unreferenced globals with ``weak`` linkage may
229 not be discarded. This is used for globals that are declared "weak"
232 "``common``" linkage is most similar to "``weak``" linkage, but they
233 are used for tentative definitions in C, such as "``int X;``" at
234 global scope. Symbols with "``common``" linkage are merged in the
235 same way as ``weak symbols``, and they may not be deleted if
236 unreferenced. ``common`` symbols may not have an explicit section,
237 must have a zero initializer, and may not be marked
238 ':ref:`constant <globalvars>`'. Functions and aliases may not have
241 .. _linkage_appending:
244 "``appending``" linkage may only be applied to global variables of
245 pointer to array type. When two global variables with appending
246 linkage are linked together, the two global arrays are appended
247 together. This is the LLVM, typesafe, equivalent of having the
248 system linker append together "sections" with identical names when
251 The semantics of this linkage follow the ELF object file model: the
252 symbol is weak until linked, if not linked, the symbol becomes null
253 instead of being an undefined reference.
254 ``linkonce_odr``, ``weak_odr``
255 Some languages allow differing globals to be merged, such as two
256 functions with different semantics. Other languages, such as
257 ``C++``, ensure that only equivalent globals are ever merged (the
258 "one definition rule" --- "ODR"). Such languages can use the
259 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
260 global will only be merged with equivalent globals. These linkage
261 types are otherwise the same as their non-``odr`` versions.
263 If none of the above identifiers are used, the global is externally
264 visible, meaning that it participates in linkage and can be used to
265 resolve external symbol references.
267 It is illegal for a function *declaration* to have any linkage type
268 other than ``external`` or ``extern_weak``.
275 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
276 :ref:`invokes <i_invoke>` can all have an optional calling convention
277 specified for the call. The calling convention of any pair of dynamic
278 caller/callee must match, or the behavior of the program is undefined.
279 The following calling conventions are supported by LLVM, and more may be
282 "``ccc``" - The C calling convention
283 This calling convention (the default if no other calling convention
284 is specified) matches the target C calling conventions. This calling
285 convention supports varargs function calls and tolerates some
286 mismatch in the declared prototype and implemented declaration of
287 the function (as does normal C).
288 "``fastcc``" - The fast calling convention
289 This calling convention attempts to make calls as fast as possible
290 (e.g. by passing things in registers). This calling convention
291 allows the target to use whatever tricks it wants to produce fast
292 code for the target, without having to conform to an externally
293 specified ABI (Application Binary Interface). `Tail calls can only
294 be optimized when this, the GHC or the HiPE convention is
295 used. <CodeGenerator.html#id80>`_ This calling convention does not
296 support varargs and requires the prototype of all callees to exactly
297 match the prototype of the function definition.
298 "``coldcc``" - The cold calling convention
299 This calling convention attempts to make code in the caller as
300 efficient as possible under the assumption that the call is not
301 commonly executed. As such, these calls often preserve all registers
302 so that the call does not break any live ranges in the caller side.
303 This calling convention does not support varargs and requires the
304 prototype of all callees to exactly match the prototype of the
305 function definition. Furthermore the inliner doesn't consider such function
307 "``cc 10``" - GHC convention
308 This calling convention has been implemented specifically for use by
309 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
310 It passes everything in registers, going to extremes to achieve this
311 by disabling callee save registers. This calling convention should
312 not be used lightly but only for specific situations such as an
313 alternative to the *register pinning* performance technique often
314 used when implementing functional programming languages. At the
315 moment only X86 supports this convention and it has the following
318 - On *X86-32* only supports up to 4 bit type parameters. No
319 floating point types are supported.
320 - On *X86-64* only supports up to 10 bit type parameters and 6
321 floating point parameters.
323 This calling convention supports `tail call
324 optimization <CodeGenerator.html#id80>`_ but requires both the
325 caller and callee are using it.
326 "``cc 11``" - The HiPE calling convention
327 This calling convention has been implemented specifically for use by
328 the `High-Performance Erlang
329 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
330 native code compiler of the `Ericsson's Open Source Erlang/OTP
331 system <http://www.erlang.org/download.shtml>`_. It uses more
332 registers for argument passing than the ordinary C calling
333 convention and defines no callee-saved registers. The calling
334 convention properly supports `tail call
335 optimization <CodeGenerator.html#id80>`_ but requires that both the
336 caller and the callee use it. It uses a *register pinning*
337 mechanism, similar to GHC's convention, for keeping frequently
338 accessed runtime components pinned to specific hardware registers.
339 At the moment only X86 supports this convention (both 32 and 64
341 "``webkit_jscc``" - WebKit's JavaScript calling convention
342 This calling convention has been implemented for `WebKit FTL JIT
343 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
344 stack right to left (as cdecl does), and returns a value in the
345 platform's customary return register.
346 "``anyregcc``" - Dynamic calling convention for code patching
347 This is a special convention that supports patching an arbitrary code
348 sequence in place of a call site. This convention forces the call
349 arguments into registers but allows them to be dynamcially
350 allocated. This can currently only be used with calls to
351 llvm.experimental.patchpoint because only this intrinsic records
352 the location of its arguments in a side table. See :doc:`StackMaps`.
353 "``preserve_mostcc``" - The `PreserveMost` calling convention
354 This calling convention attempts to make the code in the caller as little
355 intrusive as possible. This calling convention behaves identical to the `C`
356 calling convention on how arguments and return values are passed, but it
357 uses a different set of caller/callee-saved registers. This alleviates the
358 burden of saving and recovering a large register set before and after the
359 call in the caller. If the arguments are passed in callee-saved registers,
360 then they will be preserved by the callee across the call. This doesn't
361 apply for values returned in callee-saved registers.
363 - On X86-64 the callee preserves all general purpose registers, except for
364 R11. R11 can be used as a scratch register. Floating-point registers
365 (XMMs/YMMs) are not preserved and need to be saved by the caller.
367 The idea behind this convention is to support calls to runtime functions
368 that have a hot path and a cold path. The hot path is usually a small piece
369 of code that doesn't many registers. The cold path might need to call out to
370 another function and therefore only needs to preserve the caller-saved
371 registers, which haven't already been saved by the caller. The
372 `PreserveMost` calling convention is very similar to the `cold` calling
373 convention in terms of caller/callee-saved registers, but they are used for
374 different types of function calls. `coldcc` is for function calls that are
375 rarely executed, whereas `preserve_mostcc` function calls are intended to be
376 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
377 doesn't prevent the inliner from inlining the function call.
379 This calling convention will be used by a future version of the ObjectiveC
380 runtime and should therefore still be considered experimental at this time.
381 Although this convention was created to optimize certain runtime calls to
382 the ObjectiveC runtime, it is not limited to this runtime and might be used
383 by other runtimes in the future too. The current implementation only
384 supports X86-64, but the intention is to support more architectures in the
386 "``preserve_allcc``" - The `PreserveAll` calling convention
387 This calling convention attempts to make the code in the caller even less
388 intrusive than the `PreserveMost` calling convention. This calling
389 convention also behaves identical to the `C` calling convention on how
390 arguments and return values are passed, but it uses a different set of
391 caller/callee-saved registers. This removes the burden of saving and
392 recovering a large register set before and after the call in the caller. If
393 the arguments are passed in callee-saved registers, then they will be
394 preserved by the callee across the call. This doesn't apply for values
395 returned in callee-saved registers.
397 - On X86-64 the callee preserves all general purpose registers, except for
398 R11. R11 can be used as a scratch register. Furthermore it also preserves
399 all floating-point registers (XMMs/YMMs).
401 The idea behind this convention is to support calls to runtime functions
402 that don't need to call out to any other functions.
404 This calling convention, like the `PreserveMost` calling convention, will be
405 used by a future version of the ObjectiveC runtime and should be considered
406 experimental at this time.
407 "``cc <n>``" - Numbered convention
408 Any calling convention may be specified by number, allowing
409 target-specific calling conventions to be used. Target specific
410 calling conventions start at 64.
412 More calling conventions can be added/defined on an as-needed basis, to
413 support Pascal conventions or any other well-known target-independent
416 .. _visibilitystyles:
421 All Global Variables and Functions have one of the following visibility
424 "``default``" - Default style
425 On targets that use the ELF object file format, default visibility
426 means that the declaration is visible to other modules and, in
427 shared libraries, means that the declared entity may be overridden.
428 On Darwin, default visibility means that the declaration is visible
429 to other modules. Default visibility corresponds to "external
430 linkage" in the language.
431 "``hidden``" - Hidden style
432 Two declarations of an object with hidden visibility refer to the
433 same object if they are in the same shared object. Usually, hidden
434 visibility indicates that the symbol will not be placed into the
435 dynamic symbol table, so no other module (executable or shared
436 library) can reference it directly.
437 "``protected``" - Protected style
438 On ELF, protected visibility indicates that the symbol will be
439 placed in the dynamic symbol table, but that references within the
440 defining module will bind to the local symbol. That is, the symbol
441 cannot be overridden by another module.
443 A symbol with ``internal`` or ``private`` linkage must have ``default``
451 All Global Variables, Functions and Aliases can have one of the following
455 "``dllimport``" causes the compiler to reference a function or variable via
456 a global pointer to a pointer that is set up by the DLL exporting the
457 symbol. On Microsoft Windows targets, the pointer name is formed by
458 combining ``__imp_`` and the function or variable name.
460 "``dllexport``" causes the compiler to provide a global pointer to a pointer
461 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
462 Microsoft Windows targets, the pointer name is formed by combining
463 ``__imp_`` and the function or variable name. Since this storage class
464 exists for defining a dll interface, the compiler, assembler and linker know
465 it is externally referenced and must refrain from deleting the symbol.
469 Thread Local Storage Models
470 ---------------------------
472 A variable may be defined as ``thread_local``, which means that it will
473 not be shared by threads (each thread will have a separated copy of the
474 variable). Not all targets support thread-local variables. Optionally, a
475 TLS model may be specified:
478 For variables that are only used within the current shared library.
480 For variables in modules that will not be loaded dynamically.
482 For variables defined in the executable and only used within it.
484 If no explicit model is given, the "general dynamic" model is used.
486 The models correspond to the ELF TLS models; see `ELF Handling For
487 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
488 more information on under which circumstances the different models may
489 be used. The target may choose a different TLS model if the specified
490 model is not supported, or if a better choice of model can be made.
492 A model can also be specified in a alias, but then it only governs how
493 the alias is accessed. It will not have any effect in the aliasee.
500 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
501 types <t_struct>`. Literal types are uniqued structurally, but identified types
502 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
503 to forward declare a type which is not yet available.
505 An example of a identified structure specification is:
509 %mytype = type { %mytype*, i32 }
511 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
512 literal types are uniqued in recent versions of LLVM.
519 Global variables define regions of memory allocated at compilation time
522 Global variables definitions must be initialized.
524 Global variables in other translation units can also be declared, in which
525 case they don't have an initializer.
527 Either global variable definitions or declarations may have an explicit section
528 to be placed in and may have an optional explicit alignment specified.
530 A variable may be defined as a global ``constant``, which indicates that
531 the contents of the variable will **never** be modified (enabling better
532 optimization, allowing the global data to be placed in the read-only
533 section of an executable, etc). Note that variables that need runtime
534 initialization cannot be marked ``constant`` as there is a store to the
537 LLVM explicitly allows *declarations* of global variables to be marked
538 constant, even if the final definition of the global is not. This
539 capability can be used to enable slightly better optimization of the
540 program, but requires the language definition to guarantee that
541 optimizations based on the 'constantness' are valid for the translation
542 units that do not include the definition.
544 As SSA values, global variables define pointer values that are in scope
545 (i.e. they dominate) all basic blocks in the program. Global variables
546 always define a pointer to their "content" type because they describe a
547 region of memory, and all memory objects in LLVM are accessed through
550 Global variables can be marked with ``unnamed_addr`` which indicates
551 that the address is not significant, only the content. Constants marked
552 like this can be merged with other constants if they have the same
553 initializer. Note that a constant with significant address *can* be
554 merged with a ``unnamed_addr`` constant, the result being a constant
555 whose address is significant.
557 A global variable may be declared to reside in a target-specific
558 numbered address space. For targets that support them, address spaces
559 may affect how optimizations are performed and/or what target
560 instructions are used to access the variable. The default address space
561 is zero. The address space qualifier must precede any other attributes.
563 LLVM allows an explicit section to be specified for globals. If the
564 target supports it, it will emit globals to the section specified.
565 Additionally, the global can placed in a comdat if the target has the necessary
568 By default, global initializers are optimized by assuming that global
569 variables defined within the module are not modified from their
570 initial values before the start of the global initializer. This is
571 true even for variables potentially accessible from outside the
572 module, including those with external linkage or appearing in
573 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
574 by marking the variable with ``externally_initialized``.
576 An explicit alignment may be specified for a global, which must be a
577 power of 2. If not present, or if the alignment is set to zero, the
578 alignment of the global is set by the target to whatever it feels
579 convenient. If an explicit alignment is specified, the global is forced
580 to have exactly that alignment. Targets and optimizers are not allowed
581 to over-align the global if the global has an assigned section. In this
582 case, the extra alignment could be observable: for example, code could
583 assume that the globals are densely packed in their section and try to
584 iterate over them as an array, alignment padding would break this
585 iteration. The maximum alignment is ``1 << 29``.
587 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
589 Variables and aliasaes can have a
590 :ref:`Thread Local Storage Model <tls_model>`.
594 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
595 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
596 <global | constant> <Type> [<InitializerConstant>]
597 [, section "name"] [, align <Alignment>]
599 For example, the following defines a global in a numbered address space
600 with an initializer, section, and alignment:
604 @G = addrspace(5) constant float 1.0, section "foo", align 4
606 The following example just declares a global variable
610 @G = external global i32
612 The following example defines a thread-local global with the
613 ``initialexec`` TLS model:
617 @G = thread_local(initialexec) global i32 0, align 4
619 .. _functionstructure:
624 LLVM function definitions consist of the "``define``" keyword, an
625 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
626 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
627 an optional :ref:`calling convention <callingconv>`,
628 an optional ``unnamed_addr`` attribute, a return type, an optional
629 :ref:`parameter attribute <paramattrs>` for the return type, a function
630 name, a (possibly empty) argument list (each with optional :ref:`parameter
631 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
632 an optional section, an optional alignment,
633 an optional :ref:`comdat <langref_comdats>`,
634 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
635 curly brace, a list of basic blocks, and a closing curly brace.
637 LLVM function declarations consist of the "``declare``" keyword, an
638 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
639 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
640 an optional :ref:`calling convention <callingconv>`,
641 an optional ``unnamed_addr`` attribute, a return type, an optional
642 :ref:`parameter attribute <paramattrs>` for the return type, a function
643 name, a possibly empty list of arguments, an optional alignment, an optional
644 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
646 A function definition contains a list of basic blocks, forming the CFG (Control
647 Flow Graph) for the function. Each basic block may optionally start with a label
648 (giving the basic block a symbol table entry), contains a list of instructions,
649 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
650 function return). If an explicit label is not provided, a block is assigned an
651 implicit numbered label, using the next value from the same counter as used for
652 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
653 entry block does not have an explicit label, it will be assigned label "%0",
654 then the first unnamed temporary in that block will be "%1", etc.
656 The first basic block in a function is special in two ways: it is
657 immediately executed on entrance to the function, and it is not allowed
658 to have predecessor basic blocks (i.e. there can not be any branches to
659 the entry block of a function). Because the block can have no
660 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
662 LLVM allows an explicit section to be specified for functions. If the
663 target supports it, it will emit functions to the section specified.
664 Additionally, the function can placed in a COMDAT.
666 An explicit alignment may be specified for a function. If not present,
667 or if the alignment is set to zero, the alignment of the function is set
668 by the target to whatever it feels convenient. If an explicit alignment
669 is specified, the function is forced to have at least that much
670 alignment. All alignments must be a power of 2.
672 If the ``unnamed_addr`` attribute is given, the address is know to not
673 be significant and two identical functions can be merged.
677 define [linkage] [visibility] [DLLStorageClass]
679 <ResultType> @<FunctionName> ([argument list])
680 [unnamed_addr] [fn Attrs] [section "name"] [comdat $<ComdatName>]
681 [align N] [gc] [prefix Constant] { ... }
688 Aliases, unlike function or variables, don't create any new data. They
689 are just a new symbol and metadata for an existing position.
691 Aliases have a name and an aliasee that is either a global value or a
694 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
695 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
696 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
700 @<Name> = [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias [Linkage] <AliaseeTy> @<Aliasee>
702 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
703 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
704 might not correctly handle dropping a weak symbol that is aliased.
706 Alias that are not ``unnamed_addr`` are guaranteed to have the same address as
707 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
710 Since aliases are only a second name, some restrictions apply, of which
711 some can only be checked when producing an object file:
713 * The expression defining the aliasee must be computable at assembly
714 time. Since it is just a name, no relocations can be used.
716 * No alias in the expression can be weak as the possibility of the
717 intermediate alias being overridden cannot be represented in an
720 * No global value in the expression can be a declaration, since that
721 would require a relocation, which is not possible.
728 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
730 Comdats have a name which represents the COMDAT key. All global objects which
731 specify this key will only end up in the final object file if the linker chooses
732 that key over some other key. Aliases are placed in the same COMDAT that their
733 aliasee computes to, if any.
735 Comdats have a selection kind to provide input on how the linker should
736 choose between keys in two different object files.
740 $<Name> = comdat SelectionKind
742 The selection kind must be one of the following:
745 The linker may choose any COMDAT key, the choice is arbitrary.
747 The linker may choose any COMDAT key but the sections must contain the
750 The linker will choose the section containing the largest COMDAT key.
752 The linker requires that only section with this COMDAT key exist.
754 The linker may choose any COMDAT key but the sections must contain the
757 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
758 ``any`` as a selection kind.
760 Here is an example of a COMDAT group where a function will only be selected if
761 the COMDAT key's section is the largest:
765 $foo = comdat largest
766 @foo = global i32 2, comdat $foo
768 define void @bar() comdat $foo {
772 In a COFF object file, this will create a COMDAT section with selection kind
773 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
774 and another COMDAT section with selection kind
775 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
776 section and contains the contents of the ``@baz`` symbol.
778 There are some restrictions on the properties of the global object.
779 It, or an alias to it, must have the same name as the COMDAT group when
781 The contents and size of this object may be used during link-time to determine
782 which COMDAT groups get selected depending on the selection kind.
783 Because the name of the object must match the name of the COMDAT group, the
784 linkage of the global object must not be local; local symbols can get renamed
785 if a collision occurs in the symbol table.
787 The combined use of COMDATS and section attributes may yield surprising results.
794 @g1 = global i32 42, section "sec", comdat $foo
795 @g2 = global i32 42, section "sec", comdat $bar
797 From the object file perspective, this requires the creation of two sections
798 with the same name. This is necessary because both globals belong to different
799 COMDAT groups and COMDATs, at the object file level, are represented by
802 Note that certain IR constructs like global variables and functions may create
803 COMDATs in the object file in addition to any which are specified using COMDAT
804 IR. This arises, for example, when a global variable has linkonce_odr linkage.
806 .. _namedmetadatastructure:
811 Named metadata is a collection of metadata. :ref:`Metadata
812 nodes <metadata>` (but not metadata strings) are the only valid
813 operands for a named metadata.
817 ; Some unnamed metadata nodes, which are referenced by the named metadata.
818 !0 = metadata !{metadata !"zero"}
819 !1 = metadata !{metadata !"one"}
820 !2 = metadata !{metadata !"two"}
822 !name = !{!0, !1, !2}
829 The return type and each parameter of a function type may have a set of
830 *parameter attributes* associated with them. Parameter attributes are
831 used to communicate additional information about the result or
832 parameters of a function. Parameter attributes are considered to be part
833 of the function, not of the function type, so functions with different
834 parameter attributes can have the same function type.
836 Parameter attributes are simple keywords that follow the type specified.
837 If multiple parameter attributes are needed, they are space separated.
842 declare i32 @printf(i8* noalias nocapture, ...)
843 declare i32 @atoi(i8 zeroext)
844 declare signext i8 @returns_signed_char()
846 Note that any attributes for the function result (``nounwind``,
847 ``readonly``) come immediately after the argument list.
849 Currently, only the following parameter attributes are defined:
852 This indicates to the code generator that the parameter or return
853 value should be zero-extended to the extent required by the target's
854 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
855 the caller (for a parameter) or the callee (for a return value).
857 This indicates to the code generator that the parameter or return
858 value should be sign-extended to the extent required by the target's
859 ABI (which is usually 32-bits) by the caller (for a parameter) or
860 the callee (for a return value).
862 This indicates that this parameter or return value should be treated
863 in a special target-dependent fashion during while emitting code for
864 a function call or return (usually, by putting it in a register as
865 opposed to memory, though some targets use it to distinguish between
866 two different kinds of registers). Use of this attribute is
869 This indicates that the pointer parameter should really be passed by
870 value to the function. The attribute implies that a hidden copy of
871 the pointee is made between the caller and the callee, so the callee
872 is unable to modify the value in the caller. This attribute is only
873 valid on LLVM pointer arguments. It is generally used to pass
874 structs and arrays by value, but is also valid on pointers to
875 scalars. The copy is considered to belong to the caller not the
876 callee (for example, ``readonly`` functions should not write to
877 ``byval`` parameters). This is not a valid attribute for return
880 The byval attribute also supports specifying an alignment with the
881 align attribute. It indicates the alignment of the stack slot to
882 form and the known alignment of the pointer specified to the call
883 site. If the alignment is not specified, then the code generator
884 makes a target-specific assumption.
890 The ``inalloca`` argument attribute allows the caller to take the
891 address of outgoing stack arguments. An ``inalloca`` argument must
892 be a pointer to stack memory produced by an ``alloca`` instruction.
893 The alloca, or argument allocation, must also be tagged with the
894 inalloca keyword. Only the last argument may have the ``inalloca``
895 attribute, and that argument is guaranteed to be passed in memory.
897 An argument allocation may be used by a call at most once because
898 the call may deallocate it. The ``inalloca`` attribute cannot be
899 used in conjunction with other attributes that affect argument
900 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
901 ``inalloca`` attribute also disables LLVM's implicit lowering of
902 large aggregate return values, which means that frontend authors
903 must lower them with ``sret`` pointers.
905 When the call site is reached, the argument allocation must have
906 been the most recent stack allocation that is still live, or the
907 results are undefined. It is possible to allocate additional stack
908 space after an argument allocation and before its call site, but it
909 must be cleared off with :ref:`llvm.stackrestore
912 See :doc:`InAlloca` for more information on how to use this
916 This indicates that the pointer parameter specifies the address of a
917 structure that is the return value of the function in the source
918 program. This pointer must be guaranteed by the caller to be valid:
919 loads and stores to the structure may be assumed by the callee
920 not to trap and to be properly aligned. This may only be applied to
921 the first parameter. This is not a valid attribute for return
927 This indicates that pointer values :ref:`based <pointeraliasing>` on
928 the argument or return value do not alias pointer values which are
929 not *based* on it, ignoring certain "irrelevant" dependencies. For a
930 call to the parent function, dependencies between memory references
931 from before or after the call and from those during the call are
932 "irrelevant" to the ``noalias`` keyword for the arguments and return
933 value used in that call. The caller shares the responsibility with
934 the callee for ensuring that these requirements are met. For further
935 details, please see the discussion of the NoAlias response in :ref:`alias
936 analysis <Must, May, or No>`.
938 Note that this definition of ``noalias`` is intentionally similar
939 to the definition of ``restrict`` in C99 for function arguments,
940 though it is slightly weaker.
942 For function return values, C99's ``restrict`` is not meaningful,
943 while LLVM's ``noalias`` is.
945 This indicates that the callee does not make any copies of the
946 pointer that outlive the callee itself. This is not a valid
947 attribute for return values.
952 This indicates that the pointer parameter can be excised using the
953 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
954 attribute for return values and can only be applied to one parameter.
957 This indicates that the function always returns the argument as its return
958 value. This is an optimization hint to the code generator when generating
959 the caller, allowing tail call optimization and omission of register saves
960 and restores in some cases; it is not checked or enforced when generating
961 the callee. The parameter and the function return type must be valid
962 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
963 valid attribute for return values and can only be applied to one parameter.
966 This indicates that the parameter or return pointer is not null. This
967 attribute may only be applied to pointer typed parameters. This is not
968 checked or enforced by LLVM, the caller must ensure that the pointer
969 passed in is non-null, or the callee must ensure that the returned pointer
972 ``dereferenceable(<n>)``
973 This indicates that the parameter or return pointer is dereferenceable. This
974 attribute may only be applied to pointer typed parameters. A pointer that
975 is dereferenceable can be loaded from speculatively without a risk of
976 trapping. The number of bytes known to be dereferenceable must be provided
977 in parentheses. It is legal for the number of bytes to be less than the
978 size of the pointee type. The ``nonnull`` attribute does not imply
979 dereferenceability (consider a pointer to one element past the end of an
980 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
981 ``addrspace(0)`` (which is the default address space).
985 Garbage Collector Names
986 -----------------------
988 Each function may specify a garbage collector name, which is simply a
993 define void @f() gc "name" { ... }
995 The compiler declares the supported values of *name*. Specifying a
996 collector which will cause the compiler to alter its output in order to
997 support the named garbage collection algorithm.
1004 Prefix data is data associated with a function which the code generator
1005 will emit immediately before the function body. The purpose of this feature
1006 is to allow frontends to associate language-specific runtime metadata with
1007 specific functions and make it available through the function pointer while
1008 still allowing the function pointer to be called. To access the data for a
1009 given function, a program may bitcast the function pointer to a pointer to
1010 the constant's type. This implies that the IR symbol points to the start
1013 To maintain the semantics of ordinary function calls, the prefix data must
1014 have a particular format. Specifically, it must begin with a sequence of
1015 bytes which decode to a sequence of machine instructions, valid for the
1016 module's target, which transfer control to the point immediately succeeding
1017 the prefix data, without performing any other visible action. This allows
1018 the inliner and other passes to reason about the semantics of the function
1019 definition without needing to reason about the prefix data. Obviously this
1020 makes the format of the prefix data highly target dependent.
1022 Prefix data is laid out as if it were an initializer for a global variable
1023 of the prefix data's type. No padding is automatically placed between the
1024 prefix data and the function body. If padding is required, it must be part
1027 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
1028 which encodes the ``nop`` instruction:
1030 .. code-block:: llvm
1032 define void @f() prefix i8 144 { ... }
1034 Generally prefix data can be formed by encoding a relative branch instruction
1035 which skips the metadata, as in this example of valid prefix data for the
1036 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1038 .. code-block:: llvm
1040 %0 = type <{ i8, i8, i8* }>
1042 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
1044 A function may have prefix data but no body. This has similar semantics
1045 to the ``available_externally`` linkage in that the data may be used by the
1046 optimizers but will not be emitted in the object file.
1053 Attribute groups are groups of attributes that are referenced by objects within
1054 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1055 functions will use the same set of attributes. In the degenerative case of a
1056 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1057 group will capture the important command line flags used to build that file.
1059 An attribute group is a module-level object. To use an attribute group, an
1060 object references the attribute group's ID (e.g. ``#37``). An object may refer
1061 to more than one attribute group. In that situation, the attributes from the
1062 different groups are merged.
1064 Here is an example of attribute groups for a function that should always be
1065 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1067 .. code-block:: llvm
1069 ; Target-independent attributes:
1070 attributes #0 = { alwaysinline alignstack=4 }
1072 ; Target-dependent attributes:
1073 attributes #1 = { "no-sse" }
1075 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1076 define void @f() #0 #1 { ... }
1083 Function attributes are set to communicate additional information about
1084 a function. Function attributes are considered to be part of the
1085 function, not of the function type, so functions with different function
1086 attributes can have the same function type.
1088 Function attributes are simple keywords that follow the type specified.
1089 If multiple attributes are needed, they are space separated. For
1092 .. code-block:: llvm
1094 define void @f() noinline { ... }
1095 define void @f() alwaysinline { ... }
1096 define void @f() alwaysinline optsize { ... }
1097 define void @f() optsize { ... }
1100 This attribute indicates that, when emitting the prologue and
1101 epilogue, the backend should forcibly align the stack pointer.
1102 Specify the desired alignment, which must be a power of two, in
1105 This attribute indicates that the inliner should attempt to inline
1106 this function into callers whenever possible, ignoring any active
1107 inlining size threshold for this caller.
1109 This indicates that the callee function at a call site should be
1110 recognized as a built-in function, even though the function's declaration
1111 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1112 direct calls to functions which are declared with the ``nobuiltin``
1115 This attribute indicates that this function is rarely called. When
1116 computing edge weights, basic blocks post-dominated by a cold
1117 function call are also considered to be cold; and, thus, given low
1120 This attribute indicates that the source code contained a hint that
1121 inlining this function is desirable (such as the "inline" keyword in
1122 C/C++). It is just a hint; it imposes no requirements on the
1125 This attribute indicates that the function should be added to a
1126 jump-instruction table at code-generation time, and that all address-taken
1127 references to this function should be replaced with a reference to the
1128 appropriate jump-instruction-table function pointer. Note that this creates
1129 a new pointer for the original function, which means that code that depends
1130 on function-pointer identity can break. So, any function annotated with
1131 ``jumptable`` must also be ``unnamed_addr``.
1133 This attribute suggests that optimization passes and code generator
1134 passes make choices that keep the code size of this function as small
1135 as possible and perform optimizations that may sacrifice runtime
1136 performance in order to minimize the size of the generated code.
1138 This attribute disables prologue / epilogue emission for the
1139 function. This can have very system-specific consequences.
1141 This indicates that the callee function at a call site is not recognized as
1142 a built-in function. LLVM will retain the original call and not replace it
1143 with equivalent code based on the semantics of the built-in function, unless
1144 the call site uses the ``builtin`` attribute. This is valid at call sites
1145 and on function declarations and definitions.
1147 This attribute indicates that calls to the function cannot be
1148 duplicated. A call to a ``noduplicate`` function may be moved
1149 within its parent function, but may not be duplicated within
1150 its parent function.
1152 A function containing a ``noduplicate`` call may still
1153 be an inlining candidate, provided that the call is not
1154 duplicated by inlining. That implies that the function has
1155 internal linkage and only has one call site, so the original
1156 call is dead after inlining.
1158 This attributes disables implicit floating point instructions.
1160 This attribute indicates that the inliner should never inline this
1161 function in any situation. This attribute may not be used together
1162 with the ``alwaysinline`` attribute.
1164 This attribute suppresses lazy symbol binding for the function. This
1165 may make calls to the function faster, at the cost of extra program
1166 startup time if the function is not called during program startup.
1168 This attribute indicates that the code generator should not use a
1169 red zone, even if the target-specific ABI normally permits it.
1171 This function attribute indicates that the function never returns
1172 normally. This produces undefined behavior at runtime if the
1173 function ever does dynamically return.
1175 This function attribute indicates that the function never returns
1176 with an unwind or exceptional control flow. If the function does
1177 unwind, its runtime behavior is undefined.
1179 This function attribute indicates that the function is not optimized
1180 by any optimization or code generator passes with the
1181 exception of interprocedural optimization passes.
1182 This attribute cannot be used together with the ``alwaysinline``
1183 attribute; this attribute is also incompatible
1184 with the ``minsize`` attribute and the ``optsize`` attribute.
1186 This attribute requires the ``noinline`` attribute to be specified on
1187 the function as well, so the function is never inlined into any caller.
1188 Only functions with the ``alwaysinline`` attribute are valid
1189 candidates for inlining into the body of this function.
1191 This attribute suggests that optimization passes and code generator
1192 passes make choices that keep the code size of this function low,
1193 and otherwise do optimizations specifically to reduce code size as
1194 long as they do not significantly impact runtime performance.
1196 On a function, this attribute indicates that the function computes its
1197 result (or decides to unwind an exception) based strictly on its arguments,
1198 without dereferencing any pointer arguments or otherwise accessing
1199 any mutable state (e.g. memory, control registers, etc) visible to
1200 caller functions. It does not write through any pointer arguments
1201 (including ``byval`` arguments) and never changes any state visible
1202 to callers. This means that it cannot unwind exceptions by calling
1203 the ``C++`` exception throwing methods.
1205 On an argument, this attribute indicates that the function does not
1206 dereference that pointer argument, even though it may read or write the
1207 memory that the pointer points to if accessed through other pointers.
1209 On a function, this attribute indicates that the function does not write
1210 through any pointer arguments (including ``byval`` arguments) or otherwise
1211 modify any state (e.g. memory, control registers, etc) visible to
1212 caller functions. It may dereference pointer arguments and read
1213 state that may be set in the caller. A readonly function always
1214 returns the same value (or unwinds an exception identically) when
1215 called with the same set of arguments and global state. It cannot
1216 unwind an exception by calling the ``C++`` exception throwing
1219 On an argument, this attribute indicates that the function does not write
1220 through this pointer argument, even though it may write to the memory that
1221 the pointer points to.
1223 This attribute indicates that this function can return twice. The C
1224 ``setjmp`` is an example of such a function. The compiler disables
1225 some optimizations (like tail calls) in the caller of these
1227 ``sanitize_address``
1228 This attribute indicates that AddressSanitizer checks
1229 (dynamic address safety analysis) are enabled for this function.
1231 This attribute indicates that MemorySanitizer checks (dynamic detection
1232 of accesses to uninitialized memory) are enabled for this function.
1234 This attribute indicates that ThreadSanitizer checks
1235 (dynamic thread safety analysis) are enabled for this function.
1237 This attribute indicates that the function should emit a stack
1238 smashing protector. It is in the form of a "canary" --- a random value
1239 placed on the stack before the local variables that's checked upon
1240 return from the function to see if it has been overwritten. A
1241 heuristic is used to determine if a function needs stack protectors
1242 or not. The heuristic used will enable protectors for functions with:
1244 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1245 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1246 - Calls to alloca() with variable sizes or constant sizes greater than
1247 ``ssp-buffer-size``.
1249 Variables that are identified as requiring a protector will be arranged
1250 on the stack such that they are adjacent to the stack protector guard.
1252 If a function that has an ``ssp`` attribute is inlined into a
1253 function that doesn't have an ``ssp`` attribute, then the resulting
1254 function will have an ``ssp`` attribute.
1256 This attribute indicates that the function should *always* emit a
1257 stack smashing protector. This overrides the ``ssp`` function
1260 Variables that are identified as requiring a protector will be arranged
1261 on the stack such that they are adjacent to the stack protector guard.
1262 The specific layout rules are:
1264 #. Large arrays and structures containing large arrays
1265 (``>= ssp-buffer-size``) are closest to the stack protector.
1266 #. Small arrays and structures containing small arrays
1267 (``< ssp-buffer-size``) are 2nd closest to the protector.
1268 #. Variables that have had their address taken are 3rd closest to the
1271 If a function that has an ``sspreq`` attribute is inlined into a
1272 function that doesn't have an ``sspreq`` attribute or which has an
1273 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1274 an ``sspreq`` attribute.
1276 This attribute indicates that the function should emit a stack smashing
1277 protector. This attribute causes a strong heuristic to be used when
1278 determining if a function needs stack protectors. The strong heuristic
1279 will enable protectors for functions with:
1281 - Arrays of any size and type
1282 - Aggregates containing an array of any size and type.
1283 - Calls to alloca().
1284 - Local variables that have had their address taken.
1286 Variables that are identified as requiring a protector will be arranged
1287 on the stack such that they are adjacent to the stack protector guard.
1288 The specific layout rules are:
1290 #. Large arrays and structures containing large arrays
1291 (``>= ssp-buffer-size``) are closest to the stack protector.
1292 #. Small arrays and structures containing small arrays
1293 (``< ssp-buffer-size``) are 2nd closest to the protector.
1294 #. Variables that have had their address taken are 3rd closest to the
1297 This overrides the ``ssp`` function attribute.
1299 If a function that has an ``sspstrong`` attribute is inlined into a
1300 function that doesn't have an ``sspstrong`` attribute, then the
1301 resulting function will have an ``sspstrong`` attribute.
1303 This attribute indicates that the ABI being targeted requires that
1304 an unwind table entry be produce for this function even if we can
1305 show that no exceptions passes by it. This is normally the case for
1306 the ELF x86-64 abi, but it can be disabled for some compilation
1311 Module-Level Inline Assembly
1312 ----------------------------
1314 Modules may contain "module-level inline asm" blocks, which corresponds
1315 to the GCC "file scope inline asm" blocks. These blocks are internally
1316 concatenated by LLVM and treated as a single unit, but may be separated
1317 in the ``.ll`` file if desired. The syntax is very simple:
1319 .. code-block:: llvm
1321 module asm "inline asm code goes here"
1322 module asm "more can go here"
1324 The strings can contain any character by escaping non-printable
1325 characters. The escape sequence used is simply "\\xx" where "xx" is the
1326 two digit hex code for the number.
1328 The inline asm code is simply printed to the machine code .s file when
1329 assembly code is generated.
1331 .. _langref_datalayout:
1336 A module may specify a target specific data layout string that specifies
1337 how data is to be laid out in memory. The syntax for the data layout is
1340 .. code-block:: llvm
1342 target datalayout = "layout specification"
1344 The *layout specification* consists of a list of specifications
1345 separated by the minus sign character ('-'). Each specification starts
1346 with a letter and may include other information after the letter to
1347 define some aspect of the data layout. The specifications accepted are
1351 Specifies that the target lays out data in big-endian form. That is,
1352 the bits with the most significance have the lowest address
1355 Specifies that the target lays out data in little-endian form. That
1356 is, the bits with the least significance have the lowest address
1359 Specifies the natural alignment of the stack in bits. Alignment
1360 promotion of stack variables is limited to the natural stack
1361 alignment to avoid dynamic stack realignment. The stack alignment
1362 must be a multiple of 8-bits. If omitted, the natural stack
1363 alignment defaults to "unspecified", which does not prevent any
1364 alignment promotions.
1365 ``p[n]:<size>:<abi>:<pref>``
1366 This specifies the *size* of a pointer and its ``<abi>`` and
1367 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1368 bits. The address space, ``n`` is optional, and if not specified,
1369 denotes the default address space 0. The value of ``n`` must be
1370 in the range [1,2^23).
1371 ``i<size>:<abi>:<pref>``
1372 This specifies the alignment for an integer type of a given bit
1373 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1374 ``v<size>:<abi>:<pref>``
1375 This specifies the alignment for a vector type of a given bit
1377 ``f<size>:<abi>:<pref>``
1378 This specifies the alignment for a floating point type of a given bit
1379 ``<size>``. Only values of ``<size>`` that are supported by the target
1380 will work. 32 (float) and 64 (double) are supported on all targets; 80
1381 or 128 (different flavors of long double) are also supported on some
1384 This specifies the alignment for an object of aggregate type.
1386 If present, specifies that llvm names are mangled in the output. The
1389 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1390 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1391 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1392 symbols get a ``_`` prefix.
1393 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1394 functions also get a suffix based on the frame size.
1395 ``n<size1>:<size2>:<size3>...``
1396 This specifies a set of native integer widths for the target CPU in
1397 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1398 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1399 this set are considered to support most general arithmetic operations
1402 On every specification that takes a ``<abi>:<pref>``, specifying the
1403 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1404 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1406 When constructing the data layout for a given target, LLVM starts with a
1407 default set of specifications which are then (possibly) overridden by
1408 the specifications in the ``datalayout`` keyword. The default
1409 specifications are given in this list:
1411 - ``E`` - big endian
1412 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1413 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1414 same as the default address space.
1415 - ``S0`` - natural stack alignment is unspecified
1416 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1417 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1418 - ``i16:16:16`` - i16 is 16-bit aligned
1419 - ``i32:32:32`` - i32 is 32-bit aligned
1420 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1421 alignment of 64-bits
1422 - ``f16:16:16`` - half is 16-bit aligned
1423 - ``f32:32:32`` - float is 32-bit aligned
1424 - ``f64:64:64`` - double is 64-bit aligned
1425 - ``f128:128:128`` - quad is 128-bit aligned
1426 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1427 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1428 - ``a:0:64`` - aggregates are 64-bit aligned
1430 When LLVM is determining the alignment for a given type, it uses the
1433 #. If the type sought is an exact match for one of the specifications,
1434 that specification is used.
1435 #. If no match is found, and the type sought is an integer type, then
1436 the smallest integer type that is larger than the bitwidth of the
1437 sought type is used. If none of the specifications are larger than
1438 the bitwidth then the largest integer type is used. For example,
1439 given the default specifications above, the i7 type will use the
1440 alignment of i8 (next largest) while both i65 and i256 will use the
1441 alignment of i64 (largest specified).
1442 #. If no match is found, and the type sought is a vector type, then the
1443 largest vector type that is smaller than the sought vector type will
1444 be used as a fall back. This happens because <128 x double> can be
1445 implemented in terms of 64 <2 x double>, for example.
1447 The function of the data layout string may not be what you expect.
1448 Notably, this is not a specification from the frontend of what alignment
1449 the code generator should use.
1451 Instead, if specified, the target data layout is required to match what
1452 the ultimate *code generator* expects. This string is used by the
1453 mid-level optimizers to improve code, and this only works if it matches
1454 what the ultimate code generator uses. If you would like to generate IR
1455 that does not embed this target-specific detail into the IR, then you
1456 don't have to specify the string. This will disable some optimizations
1457 that require precise layout information, but this also prevents those
1458 optimizations from introducing target specificity into the IR.
1465 A module may specify a target triple string that describes the target
1466 host. The syntax for the target triple is simply:
1468 .. code-block:: llvm
1470 target triple = "x86_64-apple-macosx10.7.0"
1472 The *target triple* string consists of a series of identifiers delimited
1473 by the minus sign character ('-'). The canonical forms are:
1477 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1478 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1480 This information is passed along to the backend so that it generates
1481 code for the proper architecture. It's possible to override this on the
1482 command line with the ``-mtriple`` command line option.
1484 .. _pointeraliasing:
1486 Pointer Aliasing Rules
1487 ----------------------
1489 Any memory access must be done through a pointer value associated with
1490 an address range of the memory access, otherwise the behavior is
1491 undefined. Pointer values are associated with address ranges according
1492 to the following rules:
1494 - A pointer value is associated with the addresses associated with any
1495 value it is *based* on.
1496 - An address of a global variable is associated with the address range
1497 of the variable's storage.
1498 - The result value of an allocation instruction is associated with the
1499 address range of the allocated storage.
1500 - A null pointer in the default address-space is associated with no
1502 - An integer constant other than zero or a pointer value returned from
1503 a function not defined within LLVM may be associated with address
1504 ranges allocated through mechanisms other than those provided by
1505 LLVM. Such ranges shall not overlap with any ranges of addresses
1506 allocated by mechanisms provided by LLVM.
1508 A pointer value is *based* on another pointer value according to the
1511 - A pointer value formed from a ``getelementptr`` operation is *based*
1512 on the first operand of the ``getelementptr``.
1513 - The result value of a ``bitcast`` is *based* on the operand of the
1515 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1516 values that contribute (directly or indirectly) to the computation of
1517 the pointer's value.
1518 - The "*based* on" relationship is transitive.
1520 Note that this definition of *"based"* is intentionally similar to the
1521 definition of *"based"* in C99, though it is slightly weaker.
1523 LLVM IR does not associate types with memory. The result type of a
1524 ``load`` merely indicates the size and alignment of the memory from
1525 which to load, as well as the interpretation of the value. The first
1526 operand type of a ``store`` similarly only indicates the size and
1527 alignment of the store.
1529 Consequently, type-based alias analysis, aka TBAA, aka
1530 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1531 :ref:`Metadata <metadata>` may be used to encode additional information
1532 which specialized optimization passes may use to implement type-based
1537 Volatile Memory Accesses
1538 ------------------------
1540 Certain memory accesses, such as :ref:`load <i_load>`'s,
1541 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1542 marked ``volatile``. The optimizers must not change the number of
1543 volatile operations or change their order of execution relative to other
1544 volatile operations. The optimizers *may* change the order of volatile
1545 operations relative to non-volatile operations. This is not Java's
1546 "volatile" and has no cross-thread synchronization behavior.
1548 IR-level volatile loads and stores cannot safely be optimized into
1549 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1550 flagged volatile. Likewise, the backend should never split or merge
1551 target-legal volatile load/store instructions.
1553 .. admonition:: Rationale
1555 Platforms may rely on volatile loads and stores of natively supported
1556 data width to be executed as single instruction. For example, in C
1557 this holds for an l-value of volatile primitive type with native
1558 hardware support, but not necessarily for aggregate types. The
1559 frontend upholds these expectations, which are intentionally
1560 unspecified in the IR. The rules above ensure that IR transformation
1561 do not violate the frontend's contract with the language.
1565 Memory Model for Concurrent Operations
1566 --------------------------------------
1568 The LLVM IR does not define any way to start parallel threads of
1569 execution or to register signal handlers. Nonetheless, there are
1570 platform-specific ways to create them, and we define LLVM IR's behavior
1571 in their presence. This model is inspired by the C++0x memory model.
1573 For a more informal introduction to this model, see the :doc:`Atomics`.
1575 We define a *happens-before* partial order as the least partial order
1578 - Is a superset of single-thread program order, and
1579 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1580 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1581 techniques, like pthread locks, thread creation, thread joining,
1582 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1583 Constraints <ordering>`).
1585 Note that program order does not introduce *happens-before* edges
1586 between a thread and signals executing inside that thread.
1588 Every (defined) read operation (load instructions, memcpy, atomic
1589 loads/read-modify-writes, etc.) R reads a series of bytes written by
1590 (defined) write operations (store instructions, atomic
1591 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1592 section, initialized globals are considered to have a write of the
1593 initializer which is atomic and happens before any other read or write
1594 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1595 may see any write to the same byte, except:
1597 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1598 write\ :sub:`2` happens before R\ :sub:`byte`, then
1599 R\ :sub:`byte` does not see write\ :sub:`1`.
1600 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1601 R\ :sub:`byte` does not see write\ :sub:`3`.
1603 Given that definition, R\ :sub:`byte` is defined as follows:
1605 - If R is volatile, the result is target-dependent. (Volatile is
1606 supposed to give guarantees which can support ``sig_atomic_t`` in
1607 C/C++, and may be used for accesses to addresses which do not behave
1608 like normal memory. It does not generally provide cross-thread
1610 - Otherwise, if there is no write to the same byte that happens before
1611 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1612 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1613 R\ :sub:`byte` returns the value written by that write.
1614 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1615 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1616 Memory Ordering Constraints <ordering>` section for additional
1617 constraints on how the choice is made.
1618 - Otherwise R\ :sub:`byte` returns ``undef``.
1620 R returns the value composed of the series of bytes it read. This
1621 implies that some bytes within the value may be ``undef`` **without**
1622 the entire value being ``undef``. Note that this only defines the
1623 semantics of the operation; it doesn't mean that targets will emit more
1624 than one instruction to read the series of bytes.
1626 Note that in cases where none of the atomic intrinsics are used, this
1627 model places only one restriction on IR transformations on top of what
1628 is required for single-threaded execution: introducing a store to a byte
1629 which might not otherwise be stored is not allowed in general.
1630 (Specifically, in the case where another thread might write to and read
1631 from an address, introducing a store can change a load that may see
1632 exactly one write into a load that may see multiple writes.)
1636 Atomic Memory Ordering Constraints
1637 ----------------------------------
1639 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1640 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1641 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1642 ordering parameters that determine which other atomic instructions on
1643 the same address they *synchronize with*. These semantics are borrowed
1644 from Java and C++0x, but are somewhat more colloquial. If these
1645 descriptions aren't precise enough, check those specs (see spec
1646 references in the :doc:`atomics guide <Atomics>`).
1647 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1648 differently since they don't take an address. See that instruction's
1649 documentation for details.
1651 For a simpler introduction to the ordering constraints, see the
1655 The set of values that can be read is governed by the happens-before
1656 partial order. A value cannot be read unless some operation wrote
1657 it. This is intended to provide a guarantee strong enough to model
1658 Java's non-volatile shared variables. This ordering cannot be
1659 specified for read-modify-write operations; it is not strong enough
1660 to make them atomic in any interesting way.
1662 In addition to the guarantees of ``unordered``, there is a single
1663 total order for modifications by ``monotonic`` operations on each
1664 address. All modification orders must be compatible with the
1665 happens-before order. There is no guarantee that the modification
1666 orders can be combined to a global total order for the whole program
1667 (and this often will not be possible). The read in an atomic
1668 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1669 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1670 order immediately before the value it writes. If one atomic read
1671 happens before another atomic read of the same address, the later
1672 read must see the same value or a later value in the address's
1673 modification order. This disallows reordering of ``monotonic`` (or
1674 stronger) operations on the same address. If an address is written
1675 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1676 read that address repeatedly, the other threads must eventually see
1677 the write. This corresponds to the C++0x/C1x
1678 ``memory_order_relaxed``.
1680 In addition to the guarantees of ``monotonic``, a
1681 *synchronizes-with* edge may be formed with a ``release`` operation.
1682 This is intended to model C++'s ``memory_order_acquire``.
1684 In addition to the guarantees of ``monotonic``, if this operation
1685 writes a value which is subsequently read by an ``acquire``
1686 operation, it *synchronizes-with* that operation. (This isn't a
1687 complete description; see the C++0x definition of a release
1688 sequence.) This corresponds to the C++0x/C1x
1689 ``memory_order_release``.
1690 ``acq_rel`` (acquire+release)
1691 Acts as both an ``acquire`` and ``release`` operation on its
1692 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1693 ``seq_cst`` (sequentially consistent)
1694 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1695 operation which only reads, ``release`` for an operation which only
1696 writes), there is a global total order on all
1697 sequentially-consistent operations on all addresses, which is
1698 consistent with the *happens-before* partial order and with the
1699 modification orders of all the affected addresses. Each
1700 sequentially-consistent read sees the last preceding write to the
1701 same address in this global order. This corresponds to the C++0x/C1x
1702 ``memory_order_seq_cst`` and Java volatile.
1706 If an atomic operation is marked ``singlethread``, it only *synchronizes
1707 with* or participates in modification and seq\_cst total orderings with
1708 other operations running in the same thread (for example, in signal
1716 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1717 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1718 :ref:`frem <i_frem>`) have the following flags that can set to enable
1719 otherwise unsafe floating point operations
1722 No NaNs - Allow optimizations to assume the arguments and result are not
1723 NaN. Such optimizations are required to retain defined behavior over
1724 NaNs, but the value of the result is undefined.
1727 No Infs - Allow optimizations to assume the arguments and result are not
1728 +/-Inf. Such optimizations are required to retain defined behavior over
1729 +/-Inf, but the value of the result is undefined.
1732 No Signed Zeros - Allow optimizations to treat the sign of a zero
1733 argument or result as insignificant.
1736 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1737 argument rather than perform division.
1740 Fast - Allow algebraically equivalent transformations that may
1741 dramatically change results in floating point (e.g. reassociate). This
1742 flag implies all the others.
1749 The LLVM type system is one of the most important features of the
1750 intermediate representation. Being typed enables a number of
1751 optimizations to be performed on the intermediate representation
1752 directly, without having to do extra analyses on the side before the
1753 transformation. A strong type system makes it easier to read the
1754 generated code and enables novel analyses and transformations that are
1755 not feasible to perform on normal three address code representations.
1765 The void type does not represent any value and has no size.
1783 The function type can be thought of as a function signature. It consists of a
1784 return type and a list of formal parameter types. The return type of a function
1785 type is a void type or first class type --- except for :ref:`label <t_label>`
1786 and :ref:`metadata <t_metadata>` types.
1792 <returntype> (<parameter list>)
1794 ...where '``<parameter list>``' is a comma-separated list of type
1795 specifiers. Optionally, the parameter list may include a type ``...``, which
1796 indicates that the function takes a variable number of arguments. Variable
1797 argument functions can access their arguments with the :ref:`variable argument
1798 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1799 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1803 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1804 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1805 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1806 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1807 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1808 | ``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. |
1809 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1810 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1811 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1818 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1819 Values of these types are the only ones which can be produced by
1827 These are the types that are valid in registers from CodeGen's perspective.
1836 The integer type is a very simple type that simply specifies an
1837 arbitrary bit width for the integer type desired. Any bit width from 1
1838 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1846 The number of bits the integer will occupy is specified by the ``N``
1852 +----------------+------------------------------------------------+
1853 | ``i1`` | a single-bit integer. |
1854 +----------------+------------------------------------------------+
1855 | ``i32`` | a 32-bit integer. |
1856 +----------------+------------------------------------------------+
1857 | ``i1942652`` | a really big integer of over 1 million bits. |
1858 +----------------+------------------------------------------------+
1862 Floating Point Types
1863 """"""""""""""""""""
1872 - 16-bit floating point value
1875 - 32-bit floating point value
1878 - 64-bit floating point value
1881 - 128-bit floating point value (112-bit mantissa)
1884 - 80-bit floating point value (X87)
1887 - 128-bit floating point value (two 64-bits)
1894 The x86_mmx type represents a value held in an MMX register on an x86
1895 machine. The operations allowed on it are quite limited: parameters and
1896 return values, load and store, and bitcast. User-specified MMX
1897 instructions are represented as intrinsic or asm calls with arguments
1898 and/or results of this type. There are no arrays, vectors or constants
1915 The pointer type is used to specify memory locations. Pointers are
1916 commonly used to reference objects in memory.
1918 Pointer types may have an optional address space attribute defining the
1919 numbered address space where the pointed-to object resides. The default
1920 address space is number zero. The semantics of non-zero address spaces
1921 are target-specific.
1923 Note that LLVM does not permit pointers to void (``void*``) nor does it
1924 permit pointers to labels (``label*``). Use ``i8*`` instead.
1934 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1935 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1936 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1937 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1938 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1939 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1940 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1949 A vector type is a simple derived type that represents a vector of
1950 elements. Vector types are used when multiple primitive data are
1951 operated in parallel using a single instruction (SIMD). A vector type
1952 requires a size (number of elements) and an underlying primitive data
1953 type. Vector types are considered :ref:`first class <t_firstclass>`.
1959 < <# elements> x <elementtype> >
1961 The number of elements is a constant integer value larger than 0;
1962 elementtype may be any integer or floating point type, or a pointer to
1963 these types. Vectors of size zero are not allowed.
1967 +-------------------+--------------------------------------------------+
1968 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1969 +-------------------+--------------------------------------------------+
1970 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1971 +-------------------+--------------------------------------------------+
1972 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1973 +-------------------+--------------------------------------------------+
1974 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1975 +-------------------+--------------------------------------------------+
1984 The label type represents code labels.
1999 The metadata type represents embedded metadata. No derived types may be
2000 created from metadata except for :ref:`function <t_function>` arguments.
2013 Aggregate Types are a subset of derived types that can contain multiple
2014 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2015 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2025 The array type is a very simple derived type that arranges elements
2026 sequentially in memory. The array type requires a size (number of
2027 elements) and an underlying data type.
2033 [<# elements> x <elementtype>]
2035 The number of elements is a constant integer value; ``elementtype`` may
2036 be any type with a size.
2040 +------------------+--------------------------------------+
2041 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2042 +------------------+--------------------------------------+
2043 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2044 +------------------+--------------------------------------+
2045 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2046 +------------------+--------------------------------------+
2048 Here are some examples of multidimensional arrays:
2050 +-----------------------------+----------------------------------------------------------+
2051 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2052 +-----------------------------+----------------------------------------------------------+
2053 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2054 +-----------------------------+----------------------------------------------------------+
2055 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2056 +-----------------------------+----------------------------------------------------------+
2058 There is no restriction on indexing beyond the end of the array implied
2059 by a static type (though there are restrictions on indexing beyond the
2060 bounds of an allocated object in some cases). This means that
2061 single-dimension 'variable sized array' addressing can be implemented in
2062 LLVM with a zero length array type. An implementation of 'pascal style
2063 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2073 The structure type is used to represent a collection of data members
2074 together in memory. The elements of a structure may be any type that has
2077 Structures in memory are accessed using '``load``' and '``store``' by
2078 getting a pointer to a field with the '``getelementptr``' instruction.
2079 Structures in registers are accessed using the '``extractvalue``' and
2080 '``insertvalue``' instructions.
2082 Structures may optionally be "packed" structures, which indicate that
2083 the alignment of the struct is one byte, and that there is no padding
2084 between the elements. In non-packed structs, padding between field types
2085 is inserted as defined by the DataLayout string in the module, which is
2086 required to match what the underlying code generator expects.
2088 Structures can either be "literal" or "identified". A literal structure
2089 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2090 identified types are always defined at the top level with a name.
2091 Literal types are uniqued by their contents and can never be recursive
2092 or opaque since there is no way to write one. Identified types can be
2093 recursive, can be opaqued, and are never uniqued.
2099 %T1 = type { <type list> } ; Identified normal struct type
2100 %T2 = type <{ <type list> }> ; Identified packed struct type
2104 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2105 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2106 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2107 | ``{ 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``. |
2108 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2109 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2110 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2114 Opaque Structure Types
2115 """"""""""""""""""""""
2119 Opaque structure types are used to represent named structure types that
2120 do not have a body specified. This corresponds (for example) to the C
2121 notion of a forward declared structure.
2132 +--------------+-------------------+
2133 | ``opaque`` | An opaque type. |
2134 +--------------+-------------------+
2141 LLVM has several different basic types of constants. This section
2142 describes them all and their syntax.
2147 **Boolean constants**
2148 The two strings '``true``' and '``false``' are both valid constants
2150 **Integer constants**
2151 Standard integers (such as '4') are constants of the
2152 :ref:`integer <t_integer>` type. Negative numbers may be used with
2154 **Floating point constants**
2155 Floating point constants use standard decimal notation (e.g.
2156 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2157 hexadecimal notation (see below). The assembler requires the exact
2158 decimal value of a floating-point constant. For example, the
2159 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2160 decimal in binary. Floating point constants must have a :ref:`floating
2161 point <t_floating>` type.
2162 **Null pointer constants**
2163 The identifier '``null``' is recognized as a null pointer constant
2164 and must be of :ref:`pointer type <t_pointer>`.
2166 The one non-intuitive notation for constants is the hexadecimal form of
2167 floating point constants. For example, the form
2168 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2169 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2170 constants are required (and the only time that they are generated by the
2171 disassembler) is when a floating point constant must be emitted but it
2172 cannot be represented as a decimal floating point number in a reasonable
2173 number of digits. For example, NaN's, infinities, and other special
2174 values are represented in their IEEE hexadecimal format so that assembly
2175 and disassembly do not cause any bits to change in the constants.
2177 When using the hexadecimal form, constants of types half, float, and
2178 double are represented using the 16-digit form shown above (which
2179 matches the IEEE754 representation for double); half and float values
2180 must, however, be exactly representable as IEEE 754 half and single
2181 precision, respectively. Hexadecimal format is always used for long
2182 double, and there are three forms of long double. The 80-bit format used
2183 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2184 128-bit format used by PowerPC (two adjacent doubles) is represented by
2185 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2186 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2187 will only work if they match the long double format on your target.
2188 The IEEE 16-bit format (half precision) is represented by ``0xH``
2189 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2190 (sign bit at the left).
2192 There are no constants of type x86_mmx.
2194 .. _complexconstants:
2199 Complex constants are a (potentially recursive) combination of simple
2200 constants and smaller complex constants.
2202 **Structure constants**
2203 Structure constants are represented with notation similar to
2204 structure type definitions (a comma separated list of elements,
2205 surrounded by braces (``{}``)). For example:
2206 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2207 "``@G = external global i32``". Structure constants must have
2208 :ref:`structure type <t_struct>`, and the number and types of elements
2209 must match those specified by the type.
2211 Array constants are represented with notation similar to array type
2212 definitions (a comma separated list of elements, surrounded by
2213 square brackets (``[]``)). For example:
2214 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2215 :ref:`array type <t_array>`, and the number and types of elements must
2216 match those specified by the type.
2217 **Vector constants**
2218 Vector constants are represented with notation similar to vector
2219 type definitions (a comma separated list of elements, surrounded by
2220 less-than/greater-than's (``<>``)). For example:
2221 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2222 must have :ref:`vector type <t_vector>`, and the number and types of
2223 elements must match those specified by the type.
2224 **Zero initialization**
2225 The string '``zeroinitializer``' can be used to zero initialize a
2226 value to zero of *any* type, including scalar and
2227 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2228 having to print large zero initializers (e.g. for large arrays) and
2229 is always exactly equivalent to using explicit zero initializers.
2231 A metadata node is a structure-like constant with :ref:`metadata
2232 type <t_metadata>`. For example:
2233 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2234 constants that are meant to be interpreted as part of the
2235 instruction stream, metadata is a place to attach additional
2236 information such as debug info.
2238 Global Variable and Function Addresses
2239 --------------------------------------
2241 The addresses of :ref:`global variables <globalvars>` and
2242 :ref:`functions <functionstructure>` are always implicitly valid
2243 (link-time) constants. These constants are explicitly referenced when
2244 the :ref:`identifier for the global <identifiers>` is used and always have
2245 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2248 .. code-block:: llvm
2252 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2259 The string '``undef``' can be used anywhere a constant is expected, and
2260 indicates that the user of the value may receive an unspecified
2261 bit-pattern. Undefined values may be of any type (other than '``label``'
2262 or '``void``') and be used anywhere a constant is permitted.
2264 Undefined values are useful because they indicate to the compiler that
2265 the program is well defined no matter what value is used. This gives the
2266 compiler more freedom to optimize. Here are some examples of
2267 (potentially surprising) transformations that are valid (in pseudo IR):
2269 .. code-block:: llvm
2279 This is safe because all of the output bits are affected by the undef
2280 bits. Any output bit can have a zero or one depending on the input bits.
2282 .. code-block:: llvm
2293 These logical operations have bits that are not always affected by the
2294 input. For example, if ``%X`` has a zero bit, then the output of the
2295 '``and``' operation will always be a zero for that bit, no matter what
2296 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2297 optimize or assume that the result of the '``and``' is '``undef``'.
2298 However, it is safe to assume that all bits of the '``undef``' could be
2299 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2300 all the bits of the '``undef``' operand to the '``or``' could be set,
2301 allowing the '``or``' to be folded to -1.
2303 .. code-block:: llvm
2305 %A = select undef, %X, %Y
2306 %B = select undef, 42, %Y
2307 %C = select %X, %Y, undef
2317 This set of examples shows that undefined '``select``' (and conditional
2318 branch) conditions can go *either way*, but they have to come from one
2319 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2320 both known to have a clear low bit, then ``%A`` would have to have a
2321 cleared low bit. However, in the ``%C`` example, the optimizer is
2322 allowed to assume that the '``undef``' operand could be the same as
2323 ``%Y``, allowing the whole '``select``' to be eliminated.
2325 .. code-block:: llvm
2327 %A = xor undef, undef
2344 This example points out that two '``undef``' operands are not
2345 necessarily the same. This can be surprising to people (and also matches
2346 C semantics) where they assume that "``X^X``" is always zero, even if
2347 ``X`` is undefined. This isn't true for a number of reasons, but the
2348 short answer is that an '``undef``' "variable" can arbitrarily change
2349 its value over its "live range". This is true because the variable
2350 doesn't actually *have a live range*. Instead, the value is logically
2351 read from arbitrary registers that happen to be around when needed, so
2352 the value is not necessarily consistent over time. In fact, ``%A`` and
2353 ``%C`` need to have the same semantics or the core LLVM "replace all
2354 uses with" concept would not hold.
2356 .. code-block:: llvm
2364 These examples show the crucial difference between an *undefined value*
2365 and *undefined behavior*. An undefined value (like '``undef``') is
2366 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2367 operation can be constant folded to '``undef``', because the '``undef``'
2368 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2369 However, in the second example, we can make a more aggressive
2370 assumption: because the ``undef`` is allowed to be an arbitrary value,
2371 we are allowed to assume that it could be zero. Since a divide by zero
2372 has *undefined behavior*, we are allowed to assume that the operation
2373 does not execute at all. This allows us to delete the divide and all
2374 code after it. Because the undefined operation "can't happen", the
2375 optimizer can assume that it occurs in dead code.
2377 .. code-block:: llvm
2379 a: store undef -> %X
2380 b: store %X -> undef
2385 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2386 value can be assumed to not have any effect; we can assume that the
2387 value is overwritten with bits that happen to match what was already
2388 there. However, a store *to* an undefined location could clobber
2389 arbitrary memory, therefore, it has undefined behavior.
2396 Poison values are similar to :ref:`undef values <undefvalues>`, however
2397 they also represent the fact that an instruction or constant expression
2398 which cannot evoke side effects has nevertheless detected a condition
2399 which results in undefined behavior.
2401 There is currently no way of representing a poison value in the IR; they
2402 only exist when produced by operations such as :ref:`add <i_add>` with
2405 Poison value behavior is defined in terms of value *dependence*:
2407 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2408 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2409 their dynamic predecessor basic block.
2410 - Function arguments depend on the corresponding actual argument values
2411 in the dynamic callers of their functions.
2412 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2413 instructions that dynamically transfer control back to them.
2414 - :ref:`Invoke <i_invoke>` instructions depend on the
2415 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2416 call instructions that dynamically transfer control back to them.
2417 - Non-volatile loads and stores depend on the most recent stores to all
2418 of the referenced memory addresses, following the order in the IR
2419 (including loads and stores implied by intrinsics such as
2420 :ref:`@llvm.memcpy <int_memcpy>`.)
2421 - An instruction with externally visible side effects depends on the
2422 most recent preceding instruction with externally visible side
2423 effects, following the order in the IR. (This includes :ref:`volatile
2424 operations <volatile>`.)
2425 - An instruction *control-depends* on a :ref:`terminator
2426 instruction <terminators>` if the terminator instruction has
2427 multiple successors and the instruction is always executed when
2428 control transfers to one of the successors, and may not be executed
2429 when control is transferred to another.
2430 - Additionally, an instruction also *control-depends* on a terminator
2431 instruction if the set of instructions it otherwise depends on would
2432 be different if the terminator had transferred control to a different
2434 - Dependence is transitive.
2436 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2437 with the additional affect that any instruction which has a *dependence*
2438 on a poison value has undefined behavior.
2440 Here are some examples:
2442 .. code-block:: llvm
2445 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2446 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2447 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2448 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2450 store i32 %poison, i32* @g ; Poison value stored to memory.
2451 %poison2 = load i32* @g ; Poison value loaded back from memory.
2453 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2455 %narrowaddr = bitcast i32* @g to i16*
2456 %wideaddr = bitcast i32* @g to i64*
2457 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2458 %poison4 = load i64* %wideaddr ; Returns a poison value.
2460 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2461 br i1 %cmp, label %true, label %end ; Branch to either destination.
2464 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2465 ; it has undefined behavior.
2469 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2470 ; Both edges into this PHI are
2471 ; control-dependent on %cmp, so this
2472 ; always results in a poison value.
2474 store volatile i32 0, i32* @g ; This would depend on the store in %true
2475 ; if %cmp is true, or the store in %entry
2476 ; otherwise, so this is undefined behavior.
2478 br i1 %cmp, label %second_true, label %second_end
2479 ; The same branch again, but this time the
2480 ; true block doesn't have side effects.
2487 store volatile i32 0, i32* @g ; This time, the instruction always depends
2488 ; on the store in %end. Also, it is
2489 ; control-equivalent to %end, so this is
2490 ; well-defined (ignoring earlier undefined
2491 ; behavior in this example).
2495 Addresses of Basic Blocks
2496 -------------------------
2498 ``blockaddress(@function, %block)``
2500 The '``blockaddress``' constant computes the address of the specified
2501 basic block in the specified function, and always has an ``i8*`` type.
2502 Taking the address of the entry block is illegal.
2504 This value only has defined behavior when used as an operand to the
2505 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2506 against null. Pointer equality tests between labels addresses results in
2507 undefined behavior --- though, again, comparison against null is ok, and
2508 no label is equal to the null pointer. This may be passed around as an
2509 opaque pointer sized value as long as the bits are not inspected. This
2510 allows ``ptrtoint`` and arithmetic to be performed on these values so
2511 long as the original value is reconstituted before the ``indirectbr``
2514 Finally, some targets may provide defined semantics when using the value
2515 as the operand to an inline assembly, but that is target specific.
2519 Constant Expressions
2520 --------------------
2522 Constant expressions are used to allow expressions involving other
2523 constants to be used as constants. Constant expressions may be of any
2524 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2525 that does not have side effects (e.g. load and call are not supported).
2526 The following is the syntax for constant expressions:
2528 ``trunc (CST to TYPE)``
2529 Truncate a constant to another type. The bit size of CST must be
2530 larger than the bit size of TYPE. Both types must be integers.
2531 ``zext (CST to TYPE)``
2532 Zero extend a constant to another type. The bit size of CST must be
2533 smaller than the bit size of TYPE. Both types must be integers.
2534 ``sext (CST to TYPE)``
2535 Sign extend a constant to another type. The bit size of CST must be
2536 smaller than the bit size of TYPE. Both types must be integers.
2537 ``fptrunc (CST to TYPE)``
2538 Truncate a floating point constant to another floating point type.
2539 The size of CST must be larger than the size of TYPE. Both types
2540 must be floating point.
2541 ``fpext (CST to TYPE)``
2542 Floating point extend a constant to another type. The size of CST
2543 must be smaller or equal to the size of TYPE. Both types must be
2545 ``fptoui (CST to TYPE)``
2546 Convert a floating point constant to the corresponding unsigned
2547 integer constant. TYPE must be a scalar or vector integer type. CST
2548 must be of scalar or vector floating point type. Both CST and TYPE
2549 must be scalars, or vectors of the same number of elements. If the
2550 value won't fit in the integer type, the results are undefined.
2551 ``fptosi (CST to TYPE)``
2552 Convert a floating point constant to the corresponding signed
2553 integer constant. TYPE must be a scalar or vector integer type. CST
2554 must be of scalar or vector floating point type. Both CST and TYPE
2555 must be scalars, or vectors of the same number of elements. If the
2556 value won't fit in the integer type, the results are undefined.
2557 ``uitofp (CST to TYPE)``
2558 Convert an unsigned integer constant to the corresponding floating
2559 point constant. TYPE must be a scalar or vector floating point type.
2560 CST must be of scalar or vector integer type. Both CST and TYPE must
2561 be scalars, or vectors of the same number of elements. If the value
2562 won't fit in the floating point type, the results are undefined.
2563 ``sitofp (CST to TYPE)``
2564 Convert a signed integer constant to the corresponding floating
2565 point constant. TYPE must be a scalar or vector floating point type.
2566 CST must be of scalar or vector integer type. Both CST and TYPE must
2567 be scalars, or vectors of the same number of elements. If the value
2568 won't fit in the floating point type, the results are undefined.
2569 ``ptrtoint (CST to TYPE)``
2570 Convert a pointer typed constant to the corresponding integer
2571 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2572 pointer type. The ``CST`` value is zero extended, truncated, or
2573 unchanged to make it fit in ``TYPE``.
2574 ``inttoptr (CST to TYPE)``
2575 Convert an integer constant to a pointer constant. TYPE must be a
2576 pointer type. CST must be of integer type. The CST value is zero
2577 extended, truncated, or unchanged to make it fit in a pointer size.
2578 This one is *really* dangerous!
2579 ``bitcast (CST to TYPE)``
2580 Convert a constant, CST, to another TYPE. The constraints of the
2581 operands are the same as those for the :ref:`bitcast
2582 instruction <i_bitcast>`.
2583 ``addrspacecast (CST to TYPE)``
2584 Convert a constant pointer or constant vector of pointer, CST, to another
2585 TYPE in a different address space. The constraints of the operands are the
2586 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2587 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2588 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2589 constants. As with the :ref:`getelementptr <i_getelementptr>`
2590 instruction, the index list may have zero or more indexes, which are
2591 required to make sense for the type of "CSTPTR".
2592 ``select (COND, VAL1, VAL2)``
2593 Perform the :ref:`select operation <i_select>` on constants.
2594 ``icmp COND (VAL1, VAL2)``
2595 Performs the :ref:`icmp operation <i_icmp>` on constants.
2596 ``fcmp COND (VAL1, VAL2)``
2597 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2598 ``extractelement (VAL, IDX)``
2599 Perform the :ref:`extractelement operation <i_extractelement>` on
2601 ``insertelement (VAL, ELT, IDX)``
2602 Perform the :ref:`insertelement operation <i_insertelement>` on
2604 ``shufflevector (VEC1, VEC2, IDXMASK)``
2605 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2607 ``extractvalue (VAL, IDX0, IDX1, ...)``
2608 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2609 constants. The index list is interpreted in a similar manner as
2610 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2611 least one index value must be specified.
2612 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2613 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2614 The index list is interpreted in a similar manner as indices in a
2615 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2616 value must be specified.
2617 ``OPCODE (LHS, RHS)``
2618 Perform the specified operation of the LHS and RHS constants. OPCODE
2619 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2620 binary <bitwiseops>` operations. The constraints on operands are
2621 the same as those for the corresponding instruction (e.g. no bitwise
2622 operations on floating point values are allowed).
2629 Inline Assembler Expressions
2630 ----------------------------
2632 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2633 Inline Assembly <moduleasm>`) through the use of a special value. This
2634 value represents the inline assembler as a string (containing the
2635 instructions to emit), a list of operand constraints (stored as a
2636 string), a flag that indicates whether or not the inline asm expression
2637 has side effects, and a flag indicating whether the function containing
2638 the asm needs to align its stack conservatively. An example inline
2639 assembler expression is:
2641 .. code-block:: llvm
2643 i32 (i32) asm "bswap $0", "=r,r"
2645 Inline assembler expressions may **only** be used as the callee operand
2646 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2647 Thus, typically we have:
2649 .. code-block:: llvm
2651 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2653 Inline asms with side effects not visible in the constraint list must be
2654 marked as having side effects. This is done through the use of the
2655 '``sideeffect``' keyword, like so:
2657 .. code-block:: llvm
2659 call void asm sideeffect "eieio", ""()
2661 In some cases inline asms will contain code that will not work unless
2662 the stack is aligned in some way, such as calls or SSE instructions on
2663 x86, yet will not contain code that does that alignment within the asm.
2664 The compiler should make conservative assumptions about what the asm
2665 might contain and should generate its usual stack alignment code in the
2666 prologue if the '``alignstack``' keyword is present:
2668 .. code-block:: llvm
2670 call void asm alignstack "eieio", ""()
2672 Inline asms also support using non-standard assembly dialects. The
2673 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2674 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2675 the only supported dialects. An example is:
2677 .. code-block:: llvm
2679 call void asm inteldialect "eieio", ""()
2681 If multiple keywords appear the '``sideeffect``' keyword must come
2682 first, the '``alignstack``' keyword second and the '``inteldialect``'
2688 The call instructions that wrap inline asm nodes may have a
2689 "``!srcloc``" MDNode attached to it that contains a list of constant
2690 integers. If present, the code generator will use the integer as the
2691 location cookie value when report errors through the ``LLVMContext``
2692 error reporting mechanisms. This allows a front-end to correlate backend
2693 errors that occur with inline asm back to the source code that produced
2696 .. code-block:: llvm
2698 call void asm sideeffect "something bad", ""(), !srcloc !42
2700 !42 = !{ i32 1234567 }
2702 It is up to the front-end to make sense of the magic numbers it places
2703 in the IR. If the MDNode contains multiple constants, the code generator
2704 will use the one that corresponds to the line of the asm that the error
2709 Metadata Nodes and Metadata Strings
2710 -----------------------------------
2712 LLVM IR allows metadata to be attached to instructions in the program
2713 that can convey extra information about the code to the optimizers and
2714 code generator. One example application of metadata is source-level
2715 debug information. There are two metadata primitives: strings and nodes.
2716 All metadata has the ``metadata`` type and is identified in syntax by a
2717 preceding exclamation point ('``!``').
2719 A metadata string is a string surrounded by double quotes. It can
2720 contain any character by escaping non-printable characters with
2721 "``\xx``" where "``xx``" is the two digit hex code. For example:
2724 Metadata nodes are represented with notation similar to structure
2725 constants (a comma separated list of elements, surrounded by braces and
2726 preceded by an exclamation point). Metadata nodes can have any values as
2727 their operand. For example:
2729 .. code-block:: llvm
2731 !{ metadata !"test\00", i32 10}
2733 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2734 metadata nodes, which can be looked up in the module symbol table. For
2737 .. code-block:: llvm
2739 !foo = metadata !{!4, !3}
2741 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2742 function is using two metadata arguments:
2744 .. code-block:: llvm
2746 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2748 Metadata can be attached with an instruction. Here metadata ``!21`` is
2749 attached to the ``add`` instruction using the ``!dbg`` identifier:
2751 .. code-block:: llvm
2753 %indvar.next = add i64 %indvar, 1, !dbg !21
2755 More information about specific metadata nodes recognized by the
2756 optimizers and code generator is found below.
2761 In LLVM IR, memory does not have types, so LLVM's own type system is not
2762 suitable for doing TBAA. Instead, metadata is added to the IR to
2763 describe a type system of a higher level language. This can be used to
2764 implement typical C/C++ TBAA, but it can also be used to implement
2765 custom alias analysis behavior for other languages.
2767 The current metadata format is very simple. TBAA metadata nodes have up
2768 to three fields, e.g.:
2770 .. code-block:: llvm
2772 !0 = metadata !{ metadata !"an example type tree" }
2773 !1 = metadata !{ metadata !"int", metadata !0 }
2774 !2 = metadata !{ metadata !"float", metadata !0 }
2775 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2777 The first field is an identity field. It can be any value, usually a
2778 metadata string, which uniquely identifies the type. The most important
2779 name in the tree is the name of the root node. Two trees with different
2780 root node names are entirely disjoint, even if they have leaves with
2783 The second field identifies the type's parent node in the tree, or is
2784 null or omitted for a root node. A type is considered to alias all of
2785 its descendants and all of its ancestors in the tree. Also, a type is
2786 considered to alias all types in other trees, so that bitcode produced
2787 from multiple front-ends is handled conservatively.
2789 If the third field is present, it's an integer which if equal to 1
2790 indicates that the type is "constant" (meaning
2791 ``pointsToConstantMemory`` should return true; see `other useful
2792 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2794 '``tbaa.struct``' Metadata
2795 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2797 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2798 aggregate assignment operations in C and similar languages, however it
2799 is defined to copy a contiguous region of memory, which is more than
2800 strictly necessary for aggregate types which contain holes due to
2801 padding. Also, it doesn't contain any TBAA information about the fields
2804 ``!tbaa.struct`` metadata can describe which memory subregions in a
2805 memcpy are padding and what the TBAA tags of the struct are.
2807 The current metadata format is very simple. ``!tbaa.struct`` metadata
2808 nodes are a list of operands which are in conceptual groups of three.
2809 For each group of three, the first operand gives the byte offset of a
2810 field in bytes, the second gives its size in bytes, and the third gives
2813 .. code-block:: llvm
2815 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2817 This describes a struct with two fields. The first is at offset 0 bytes
2818 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2819 and has size 4 bytes and has tbaa tag !2.
2821 Note that the fields need not be contiguous. In this example, there is a
2822 4 byte gap between the two fields. This gap represents padding which
2823 does not carry useful data and need not be preserved.
2825 '``fpmath``' Metadata
2826 ^^^^^^^^^^^^^^^^^^^^^
2828 ``fpmath`` metadata may be attached to any instruction of floating point
2829 type. It can be used to express the maximum acceptable error in the
2830 result of that instruction, in ULPs, thus potentially allowing the
2831 compiler to use a more efficient but less accurate method of computing
2832 it. ULP is defined as follows:
2834 If ``x`` is a real number that lies between two finite consecutive
2835 floating-point numbers ``a`` and ``b``, without being equal to one
2836 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2837 distance between the two non-equal finite floating-point numbers
2838 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2840 The metadata node shall consist of a single positive floating point
2841 number representing the maximum relative error, for example:
2843 .. code-block:: llvm
2845 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2847 '``range``' Metadata
2848 ^^^^^^^^^^^^^^^^^^^^
2850 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
2851 integer types. It expresses the possible ranges the loaded value or the value
2852 returned by the called function at this call site is in. The ranges are
2853 represented with a flattened list of integers. The loaded value or the value
2854 returned is known to be in the union of the ranges defined by each consecutive
2855 pair. Each pair has the following properties:
2857 - The type must match the type loaded by the instruction.
2858 - The pair ``a,b`` represents the range ``[a,b)``.
2859 - Both ``a`` and ``b`` are constants.
2860 - The range is allowed to wrap.
2861 - The range should not represent the full or empty set. That is,
2864 In addition, the pairs must be in signed order of the lower bound and
2865 they must be non-contiguous.
2869 .. code-block:: llvm
2871 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2872 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2873 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
2874 %d = invoke i8 @bar() to label %cont
2875 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
2877 !0 = metadata !{ i8 0, i8 2 }
2878 !1 = metadata !{ i8 255, i8 2 }
2879 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2880 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2885 It is sometimes useful to attach information to loop constructs. Currently,
2886 loop metadata is implemented as metadata attached to the branch instruction
2887 in the loop latch block. This type of metadata refer to a metadata node that is
2888 guaranteed to be separate for each loop. The loop identifier metadata is
2889 specified with the name ``llvm.loop``.
2891 The loop identifier metadata is implemented using a metadata that refers to
2892 itself to avoid merging it with any other identifier metadata, e.g.,
2893 during module linkage or function inlining. That is, each loop should refer
2894 to their own identification metadata even if they reside in separate functions.
2895 The following example contains loop identifier metadata for two separate loop
2898 .. code-block:: llvm
2900 !0 = metadata !{ metadata !0 }
2901 !1 = metadata !{ metadata !1 }
2903 The loop identifier metadata can be used to specify additional per-loop
2904 metadata. Any operands after the first operand can be treated as user-defined
2905 metadata. For example the ``llvm.loop.vectorize.unroll`` metadata is understood
2906 by the loop vectorizer to indicate how many times to unroll the loop:
2908 .. code-block:: llvm
2910 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2912 !0 = metadata !{ metadata !0, metadata !1 }
2913 !1 = metadata !{ metadata !"llvm.loop.vectorize.unroll", i32 2 }
2918 Metadata types used to annotate memory accesses with information helpful
2919 for optimizations are prefixed with ``llvm.mem``.
2921 '``llvm.mem.parallel_loop_access``' Metadata
2922 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2924 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
2925 or metadata containing a list of loop identifiers for nested loops.
2926 The metadata is attached to memory accessing instructions and denotes that
2927 no loop carried memory dependence exist between it and other instructions denoted
2928 with the same loop identifier.
2930 Precisely, given two instructions ``m1`` and ``m2`` that both have the
2931 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
2932 set of loops associated with that metadata, respectively, then there is no loop
2933 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
2936 As a special case, if all memory accessing instructions in a loop have
2937 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
2938 loop has no loop carried memory dependences and is considered to be a parallel
2941 Note that if not all memory access instructions have such metadata referring to
2942 the loop, then the loop is considered not being trivially parallel. Additional
2943 memory dependence analysis is required to make that determination. As a fail
2944 safe mechanism, this causes loops that were originally parallel to be considered
2945 sequential (if optimization passes that are unaware of the parallel semantics
2946 insert new memory instructions into the loop body).
2948 Example of a loop that is considered parallel due to its correct use of
2949 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2950 metadata types that refer to the same loop identifier metadata.
2952 .. code-block:: llvm
2956 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
2958 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2960 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2964 !0 = metadata !{ metadata !0 }
2966 It is also possible to have nested parallel loops. In that case the
2967 memory accesses refer to a list of loop identifier metadata nodes instead of
2968 the loop identifier metadata node directly:
2970 .. code-block:: llvm
2974 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
2976 br label %inner.for.body
2980 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2982 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
2984 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2988 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
2990 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2992 outer.for.end: ; preds = %for.body
2994 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2995 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2996 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2998 '``llvm.loop.vectorize``'
2999 ^^^^^^^^^^^^^^^^^^^^^^^^^
3001 Metadata prefixed with ``llvm.loop.vectorize`` is used to control per-loop
3002 vectorization parameters such as vectorization factor and unroll factor.
3004 ``llvm.loop.vectorize`` metadata should be used in conjunction with
3005 ``llvm.loop`` loop identification metadata.
3007 '``llvm.loop.vectorize.unroll``' Metadata
3008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3010 This metadata instructs the loop vectorizer to unroll the specified
3011 loop exactly ``N`` times.
3013 The first operand is the string ``llvm.loop.vectorize.unroll`` and the second
3014 operand is an integer specifying the unroll factor. For example:
3016 .. code-block:: llvm
3018 !0 = metadata !{ metadata !"llvm.loop.vectorize.unroll", i32 4 }
3020 Note that setting ``llvm.loop.vectorize.unroll`` to 1 disables
3021 unrolling of the loop.
3023 If ``llvm.loop.vectorize.unroll`` is set to 0 then the amount of
3024 unrolling will be determined automatically.
3026 '``llvm.loop.vectorize.width``' Metadata
3027 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3029 This metadata sets the target width of the vectorizer to ``N``. Without
3030 this metadata, the vectorizer will choose a width automatically.
3031 Regardless of this metadata, the vectorizer will only vectorize loops if
3032 it believes it is valid to do so.
3034 The first operand is the string ``llvm.loop.vectorize.width`` and the
3035 second operand is an integer specifying the width. For example:
3037 .. code-block:: llvm
3039 !0 = metadata !{ metadata !"llvm.loop.vectorize.width", i32 4 }
3041 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
3042 vectorization of the loop.
3044 If ``llvm.loop.vectorize.width`` is set to 0 then the width will be
3045 determined automatically.
3047 Module Flags Metadata
3048 =====================
3050 Information about the module as a whole is difficult to convey to LLVM's
3051 subsystems. The LLVM IR isn't sufficient to transmit this information.
3052 The ``llvm.module.flags`` named metadata exists in order to facilitate
3053 this. These flags are in the form of key / value pairs --- much like a
3054 dictionary --- making it easy for any subsystem who cares about a flag to
3057 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
3058 Each triplet has the following form:
3060 - The first element is a *behavior* flag, which specifies the behavior
3061 when two (or more) modules are merged together, and it encounters two
3062 (or more) metadata with the same ID. The supported behaviors are
3064 - The second element is a metadata string that is a unique ID for the
3065 metadata. Each module may only have one flag entry for each unique ID (not
3066 including entries with the **Require** behavior).
3067 - The third element is the value of the flag.
3069 When two (or more) modules are merged together, the resulting
3070 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
3071 each unique metadata ID string, there will be exactly one entry in the merged
3072 modules ``llvm.module.flags`` metadata table, and the value for that entry will
3073 be determined by the merge behavior flag, as described below. The only exception
3074 is that entries with the *Require* behavior are always preserved.
3076 The following behaviors are supported:
3087 Emits an error if two values disagree, otherwise the resulting value
3088 is that of the operands.
3092 Emits a warning if two values disagree. The result value will be the
3093 operand for the flag from the first module being linked.
3097 Adds a requirement that another module flag be present and have a
3098 specified value after linking is performed. The value must be a
3099 metadata pair, where the first element of the pair is the ID of the
3100 module flag to be restricted, and the second element of the pair is
3101 the value the module flag should be restricted to. This behavior can
3102 be used to restrict the allowable results (via triggering of an
3103 error) of linking IDs with the **Override** behavior.
3107 Uses the specified value, regardless of the behavior or value of the
3108 other module. If both modules specify **Override**, but the values
3109 differ, an error will be emitted.
3113 Appends the two values, which are required to be metadata nodes.
3117 Appends the two values, which are required to be metadata
3118 nodes. However, duplicate entries in the second list are dropped
3119 during the append operation.
3121 It is an error for a particular unique flag ID to have multiple behaviors,
3122 except in the case of **Require** (which adds restrictions on another metadata
3123 value) or **Override**.
3125 An example of module flags:
3127 .. code-block:: llvm
3129 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3130 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3131 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3132 !3 = metadata !{ i32 3, metadata !"qux",
3134 metadata !"foo", i32 1
3137 !llvm.module.flags = !{ !0, !1, !2, !3 }
3139 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3140 if two or more ``!"foo"`` flags are seen is to emit an error if their
3141 values are not equal.
3143 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3144 behavior if two or more ``!"bar"`` flags are seen is to use the value
3147 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3148 behavior if two or more ``!"qux"`` flags are seen is to emit a
3149 warning if their values are not equal.
3151 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3155 metadata !{ metadata !"foo", i32 1 }
3157 The behavior is to emit an error if the ``llvm.module.flags`` does not
3158 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3161 Objective-C Garbage Collection Module Flags Metadata
3162 ----------------------------------------------------
3164 On the Mach-O platform, Objective-C stores metadata about garbage
3165 collection in a special section called "image info". The metadata
3166 consists of a version number and a bitmask specifying what types of
3167 garbage collection are supported (if any) by the file. If two or more
3168 modules are linked together their garbage collection metadata needs to
3169 be merged rather than appended together.
3171 The Objective-C garbage collection module flags metadata consists of the
3172 following key-value pairs:
3181 * - ``Objective-C Version``
3182 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3184 * - ``Objective-C Image Info Version``
3185 - **[Required]** --- The version of the image info section. Currently
3188 * - ``Objective-C Image Info Section``
3189 - **[Required]** --- The section to place the metadata. Valid values are
3190 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3191 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3192 Objective-C ABI version 2.
3194 * - ``Objective-C Garbage Collection``
3195 - **[Required]** --- Specifies whether garbage collection is supported or
3196 not. Valid values are 0, for no garbage collection, and 2, for garbage
3197 collection supported.
3199 * - ``Objective-C GC Only``
3200 - **[Optional]** --- Specifies that only garbage collection is supported.
3201 If present, its value must be 6. This flag requires that the
3202 ``Objective-C Garbage Collection`` flag have the value 2.
3204 Some important flag interactions:
3206 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3207 merged with a module with ``Objective-C Garbage Collection`` set to
3208 2, then the resulting module has the
3209 ``Objective-C Garbage Collection`` flag set to 0.
3210 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3211 merged with a module with ``Objective-C GC Only`` set to 6.
3213 Automatic Linker Flags Module Flags Metadata
3214 --------------------------------------------
3216 Some targets support embedding flags to the linker inside individual object
3217 files. Typically this is used in conjunction with language extensions which
3218 allow source files to explicitly declare the libraries they depend on, and have
3219 these automatically be transmitted to the linker via object files.
3221 These flags are encoded in the IR using metadata in the module flags section,
3222 using the ``Linker Options`` key. The merge behavior for this flag is required
3223 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3224 node which should be a list of other metadata nodes, each of which should be a
3225 list of metadata strings defining linker options.
3227 For example, the following metadata section specifies two separate sets of
3228 linker options, presumably to link against ``libz`` and the ``Cocoa``
3231 !0 = metadata !{ i32 6, metadata !"Linker Options",
3233 metadata !{ metadata !"-lz" },
3234 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3235 !llvm.module.flags = !{ !0 }
3237 The metadata encoding as lists of lists of options, as opposed to a collapsed
3238 list of options, is chosen so that the IR encoding can use multiple option
3239 strings to specify e.g., a single library, while still having that specifier be
3240 preserved as an atomic element that can be recognized by a target specific
3241 assembly writer or object file emitter.
3243 Each individual option is required to be either a valid option for the target's
3244 linker, or an option that is reserved by the target specific assembly writer or
3245 object file emitter. No other aspect of these options is defined by the IR.
3247 C type width Module Flags Metadata
3248 ----------------------------------
3250 The ARM backend emits a section into each generated object file describing the
3251 options that it was compiled with (in a compiler-independent way) to prevent
3252 linking incompatible objects, and to allow automatic library selection. Some
3253 of these options are not visible at the IR level, namely wchar_t width and enum
3256 To pass this information to the backend, these options are encoded in module
3257 flags metadata, using the following key-value pairs:
3267 - * 0 --- sizeof(wchar_t) == 4
3268 * 1 --- sizeof(wchar_t) == 2
3271 - * 0 --- Enums are at least as large as an ``int``.
3272 * 1 --- Enums are stored in the smallest integer type which can
3273 represent all of its values.
3275 For example, the following metadata section specifies that the module was
3276 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
3277 enum is the smallest type which can represent all of its values::
3279 !llvm.module.flags = !{!0, !1}
3280 !0 = metadata !{i32 1, metadata !"short_wchar", i32 1}
3281 !1 = metadata !{i32 1, metadata !"short_enum", i32 0}
3283 .. _intrinsicglobalvariables:
3285 Intrinsic Global Variables
3286 ==========================
3288 LLVM has a number of "magic" global variables that contain data that
3289 affect code generation or other IR semantics. These are documented here.
3290 All globals of this sort should have a section specified as
3291 "``llvm.metadata``". This section and all globals that start with
3292 "``llvm.``" are reserved for use by LLVM.
3296 The '``llvm.used``' Global Variable
3297 -----------------------------------
3299 The ``@llvm.used`` global is an array which has
3300 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3301 pointers to named global variables, functions and aliases which may optionally
3302 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3305 .. code-block:: llvm
3310 @llvm.used = appending global [2 x i8*] [
3312 i8* bitcast (i32* @Y to i8*)
3313 ], section "llvm.metadata"
3315 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3316 and linker are required to treat the symbol as if there is a reference to the
3317 symbol that it cannot see (which is why they have to be named). For example, if
3318 a variable has internal linkage and no references other than that from the
3319 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3320 references from inline asms and other things the compiler cannot "see", and
3321 corresponds to "``attribute((used))``" in GNU C.
3323 On some targets, the code generator must emit a directive to the
3324 assembler or object file to prevent the assembler and linker from
3325 molesting the symbol.
3327 .. _gv_llvmcompilerused:
3329 The '``llvm.compiler.used``' Global Variable
3330 --------------------------------------------
3332 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3333 directive, except that it only prevents the compiler from touching the
3334 symbol. On targets that support it, this allows an intelligent linker to
3335 optimize references to the symbol without being impeded as it would be
3338 This is a rare construct that should only be used in rare circumstances,
3339 and should not be exposed to source languages.
3341 .. _gv_llvmglobalctors:
3343 The '``llvm.global_ctors``' Global Variable
3344 -------------------------------------------
3346 .. code-block:: llvm
3348 %0 = type { i32, void ()*, i8* }
3349 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
3351 The ``@llvm.global_ctors`` array contains a list of constructor
3352 functions, priorities, and an optional associated global or function.
3353 The functions referenced by this array will be called in ascending order
3354 of priority (i.e. lowest first) when the module is loaded. The order of
3355 functions with the same priority is not defined.
3357 If the third field is present, non-null, and points to a global variable
3358 or function, the initializer function will only run if the associated
3359 data from the current module is not discarded.
3361 .. _llvmglobaldtors:
3363 The '``llvm.global_dtors``' Global Variable
3364 -------------------------------------------
3366 .. code-block:: llvm
3368 %0 = type { i32, void ()*, i8* }
3369 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
3371 The ``@llvm.global_dtors`` array contains a list of destructor
3372 functions, priorities, and an optional associated global or function.
3373 The functions referenced by this array will be called in descending
3374 order of priority (i.e. highest first) when the module is unloaded. The
3375 order of functions with the same priority is not defined.
3377 If the third field is present, non-null, and points to a global variable
3378 or function, the destructor function will only run if the associated
3379 data from the current module is not discarded.
3381 Instruction Reference
3382 =====================
3384 The LLVM instruction set consists of several different classifications
3385 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3386 instructions <binaryops>`, :ref:`bitwise binary
3387 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3388 :ref:`other instructions <otherops>`.
3392 Terminator Instructions
3393 -----------------------
3395 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3396 program ends with a "Terminator" instruction, which indicates which
3397 block should be executed after the current block is finished. These
3398 terminator instructions typically yield a '``void``' value: they produce
3399 control flow, not values (the one exception being the
3400 ':ref:`invoke <i_invoke>`' instruction).
3402 The terminator instructions are: ':ref:`ret <i_ret>`',
3403 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3404 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3405 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3409 '``ret``' Instruction
3410 ^^^^^^^^^^^^^^^^^^^^^
3417 ret <type> <value> ; Return a value from a non-void function
3418 ret void ; Return from void function
3423 The '``ret``' instruction is used to return control flow (and optionally
3424 a value) from a function back to the caller.
3426 There are two forms of the '``ret``' instruction: one that returns a
3427 value and then causes control flow, and one that just causes control
3433 The '``ret``' instruction optionally accepts a single argument, the
3434 return value. The type of the return value must be a ':ref:`first
3435 class <t_firstclass>`' type.
3437 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3438 return type and contains a '``ret``' instruction with no return value or
3439 a return value with a type that does not match its type, or if it has a
3440 void return type and contains a '``ret``' instruction with a return
3446 When the '``ret``' instruction is executed, control flow returns back to
3447 the calling function's context. If the caller is a
3448 ":ref:`call <i_call>`" instruction, execution continues at the
3449 instruction after the call. If the caller was an
3450 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3451 beginning of the "normal" destination block. If the instruction returns
3452 a value, that value shall set the call or invoke instruction's return
3458 .. code-block:: llvm
3460 ret i32 5 ; Return an integer value of 5
3461 ret void ; Return from a void function
3462 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3466 '``br``' Instruction
3467 ^^^^^^^^^^^^^^^^^^^^
3474 br i1 <cond>, label <iftrue>, label <iffalse>
3475 br label <dest> ; Unconditional branch
3480 The '``br``' instruction is used to cause control flow to transfer to a
3481 different basic block in the current function. There are two forms of
3482 this instruction, corresponding to a conditional branch and an
3483 unconditional branch.
3488 The conditional branch form of the '``br``' instruction takes a single
3489 '``i1``' value and two '``label``' values. The unconditional form of the
3490 '``br``' instruction takes a single '``label``' value as a target.
3495 Upon execution of a conditional '``br``' instruction, the '``i1``'
3496 argument is evaluated. If the value is ``true``, control flows to the
3497 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3498 to the '``iffalse``' ``label`` argument.
3503 .. code-block:: llvm
3506 %cond = icmp eq i32 %a, %b
3507 br i1 %cond, label %IfEqual, label %IfUnequal
3515 '``switch``' Instruction
3516 ^^^^^^^^^^^^^^^^^^^^^^^^
3523 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3528 The '``switch``' instruction is used to transfer control flow to one of
3529 several different places. It is a generalization of the '``br``'
3530 instruction, allowing a branch to occur to one of many possible
3536 The '``switch``' instruction uses three parameters: an integer
3537 comparison value '``value``', a default '``label``' destination, and an
3538 array of pairs of comparison value constants and '``label``'s. The table
3539 is not allowed to contain duplicate constant entries.
3544 The ``switch`` instruction specifies a table of values and destinations.
3545 When the '``switch``' instruction is executed, this table is searched
3546 for the given value. If the value is found, control flow is transferred
3547 to the corresponding destination; otherwise, control flow is transferred
3548 to the default destination.
3553 Depending on properties of the target machine and the particular
3554 ``switch`` instruction, this instruction may be code generated in
3555 different ways. For example, it could be generated as a series of
3556 chained conditional branches or with a lookup table.
3561 .. code-block:: llvm
3563 ; Emulate a conditional br instruction
3564 %Val = zext i1 %value to i32
3565 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3567 ; Emulate an unconditional br instruction
3568 switch i32 0, label %dest [ ]
3570 ; Implement a jump table:
3571 switch i32 %val, label %otherwise [ i32 0, label %onzero
3573 i32 2, label %ontwo ]
3577 '``indirectbr``' Instruction
3578 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3585 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3590 The '``indirectbr``' instruction implements an indirect branch to a
3591 label within the current function, whose address is specified by
3592 "``address``". Address must be derived from a
3593 :ref:`blockaddress <blockaddress>` constant.
3598 The '``address``' argument is the address of the label to jump to. The
3599 rest of the arguments indicate the full set of possible destinations
3600 that the address may point to. Blocks are allowed to occur multiple
3601 times in the destination list, though this isn't particularly useful.
3603 This destination list is required so that dataflow analysis has an
3604 accurate understanding of the CFG.
3609 Control transfers to the block specified in the address argument. All
3610 possible destination blocks must be listed in the label list, otherwise
3611 this instruction has undefined behavior. This implies that jumps to
3612 labels defined in other functions have undefined behavior as well.
3617 This is typically implemented with a jump through a register.
3622 .. code-block:: llvm
3624 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3628 '``invoke``' Instruction
3629 ^^^^^^^^^^^^^^^^^^^^^^^^
3636 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3637 to label <normal label> unwind label <exception label>
3642 The '``invoke``' instruction causes control to transfer to a specified
3643 function, with the possibility of control flow transfer to either the
3644 '``normal``' label or the '``exception``' label. If the callee function
3645 returns with the "``ret``" instruction, control flow will return to the
3646 "normal" label. If the callee (or any indirect callees) returns via the
3647 ":ref:`resume <i_resume>`" instruction or other exception handling
3648 mechanism, control is interrupted and continued at the dynamically
3649 nearest "exception" label.
3651 The '``exception``' label is a `landing
3652 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3653 '``exception``' label is required to have the
3654 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3655 information about the behavior of the program after unwinding happens,
3656 as its first non-PHI instruction. The restrictions on the
3657 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3658 instruction, so that the important information contained within the
3659 "``landingpad``" instruction can't be lost through normal code motion.
3664 This instruction requires several arguments:
3666 #. The optional "cconv" marker indicates which :ref:`calling
3667 convention <callingconv>` the call should use. If none is
3668 specified, the call defaults to using C calling conventions.
3669 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3670 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3672 #. '``ptr to function ty``': shall be the signature of the pointer to
3673 function value being invoked. In most cases, this is a direct
3674 function invocation, but indirect ``invoke``'s are just as possible,
3675 branching off an arbitrary pointer to function value.
3676 #. '``function ptr val``': An LLVM value containing a pointer to a
3677 function to be invoked.
3678 #. '``function args``': argument list whose types match the function
3679 signature argument types and parameter attributes. All arguments must
3680 be of :ref:`first class <t_firstclass>` type. If the function signature
3681 indicates the function accepts a variable number of arguments, the
3682 extra arguments can be specified.
3683 #. '``normal label``': the label reached when the called function
3684 executes a '``ret``' instruction.
3685 #. '``exception label``': the label reached when a callee returns via
3686 the :ref:`resume <i_resume>` instruction or other exception handling
3688 #. The optional :ref:`function attributes <fnattrs>` list. Only
3689 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3690 attributes are valid here.
3695 This instruction is designed to operate as a standard '``call``'
3696 instruction in most regards. The primary difference is that it
3697 establishes an association with a label, which is used by the runtime
3698 library to unwind the stack.
3700 This instruction is used in languages with destructors to ensure that
3701 proper cleanup is performed in the case of either a ``longjmp`` or a
3702 thrown exception. Additionally, this is important for implementation of
3703 '``catch``' clauses in high-level languages that support them.
3705 For the purposes of the SSA form, the definition of the value returned
3706 by the '``invoke``' instruction is deemed to occur on the edge from the
3707 current block to the "normal" label. If the callee unwinds then no
3708 return value is available.
3713 .. code-block:: llvm
3715 %retval = invoke i32 @Test(i32 15) to label %Continue
3716 unwind label %TestCleanup ; i32:retval set
3717 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3718 unwind label %TestCleanup ; i32:retval set
3722 '``resume``' Instruction
3723 ^^^^^^^^^^^^^^^^^^^^^^^^
3730 resume <type> <value>
3735 The '``resume``' instruction is a terminator instruction that has no
3741 The '``resume``' instruction requires one argument, which must have the
3742 same type as the result of any '``landingpad``' instruction in the same
3748 The '``resume``' instruction resumes propagation of an existing
3749 (in-flight) exception whose unwinding was interrupted with a
3750 :ref:`landingpad <i_landingpad>` instruction.
3755 .. code-block:: llvm
3757 resume { i8*, i32 } %exn
3761 '``unreachable``' Instruction
3762 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3774 The '``unreachable``' instruction has no defined semantics. This
3775 instruction is used to inform the optimizer that a particular portion of
3776 the code is not reachable. This can be used to indicate that the code
3777 after a no-return function cannot be reached, and other facts.
3782 The '``unreachable``' instruction has no defined semantics.
3789 Binary operators are used to do most of the computation in a program.
3790 They require two operands of the same type, execute an operation on
3791 them, and produce a single value. The operands might represent multiple
3792 data, as is the case with the :ref:`vector <t_vector>` data type. The
3793 result value has the same type as its operands.
3795 There are several different binary operators:
3799 '``add``' Instruction
3800 ^^^^^^^^^^^^^^^^^^^^^
3807 <result> = add <ty> <op1>, <op2> ; yields ty:result
3808 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
3809 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
3810 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
3815 The '``add``' instruction returns the sum of its two operands.
3820 The two arguments to the '``add``' instruction must be
3821 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3822 arguments must have identical types.
3827 The value produced is the integer sum of the two operands.
3829 If the sum has unsigned overflow, the result returned is the
3830 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3833 Because LLVM integers use a two's complement representation, this
3834 instruction is appropriate for both signed and unsigned integers.
3836 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3837 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3838 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3839 unsigned and/or signed overflow, respectively, occurs.
3844 .. code-block:: llvm
3846 <result> = add i32 4, %var ; yields i32:result = 4 + %var
3850 '``fadd``' Instruction
3851 ^^^^^^^^^^^^^^^^^^^^^^
3858 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
3863 The '``fadd``' instruction returns the sum of its two operands.
3868 The two arguments to the '``fadd``' instruction must be :ref:`floating
3869 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3870 Both arguments must have identical types.
3875 The value produced is the floating point sum of the two operands. This
3876 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3877 which are optimization hints to enable otherwise unsafe floating point
3883 .. code-block:: llvm
3885 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
3887 '``sub``' Instruction
3888 ^^^^^^^^^^^^^^^^^^^^^
3895 <result> = sub <ty> <op1>, <op2> ; yields ty:result
3896 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
3897 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
3898 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
3903 The '``sub``' instruction returns the difference of its two operands.
3905 Note that the '``sub``' instruction is used to represent the '``neg``'
3906 instruction present in most other intermediate representations.
3911 The two arguments to the '``sub``' instruction must be
3912 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3913 arguments must have identical types.
3918 The value produced is the integer difference of the two operands.
3920 If the difference has unsigned overflow, the result returned is the
3921 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3924 Because LLVM integers use a two's complement representation, this
3925 instruction is appropriate for both signed and unsigned integers.
3927 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3928 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3929 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3930 unsigned and/or signed overflow, respectively, occurs.
3935 .. code-block:: llvm
3937 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
3938 <result> = sub i32 0, %val ; yields i32:result = -%var
3942 '``fsub``' Instruction
3943 ^^^^^^^^^^^^^^^^^^^^^^
3950 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
3955 The '``fsub``' instruction returns the difference of its two operands.
3957 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3958 instruction present in most other intermediate representations.
3963 The two arguments to the '``fsub``' instruction must be :ref:`floating
3964 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3965 Both arguments must have identical types.
3970 The value produced is the floating point difference of the two operands.
3971 This instruction can also take any number of :ref:`fast-math
3972 flags <fastmath>`, which are optimization hints to enable otherwise
3973 unsafe floating point optimizations:
3978 .. code-block:: llvm
3980 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
3981 <result> = fsub float -0.0, %val ; yields float:result = -%var
3983 '``mul``' Instruction
3984 ^^^^^^^^^^^^^^^^^^^^^
3991 <result> = mul <ty> <op1>, <op2> ; yields ty:result
3992 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
3993 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
3994 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
3999 The '``mul``' instruction returns the product of its two operands.
4004 The two arguments to the '``mul``' 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 product of the two operands.
4013 If the result of the multiplication has unsigned overflow, the result
4014 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
4015 bit width of the result.
4017 Because LLVM integers use a two's complement representation, and the
4018 result is the same width as the operands, this instruction returns the
4019 correct result for both signed and unsigned integers. If a full product
4020 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
4021 sign-extended or zero-extended as appropriate to the width of the full
4024 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4025 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4026 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
4027 unsigned and/or signed overflow, respectively, occurs.
4032 .. code-block:: llvm
4034 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
4038 '``fmul``' Instruction
4039 ^^^^^^^^^^^^^^^^^^^^^^
4046 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4051 The '``fmul``' instruction returns the product of its two operands.
4056 The two arguments to the '``fmul``' instruction must be :ref:`floating
4057 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4058 Both arguments must have identical types.
4063 The value produced is the floating point product of the two operands.
4064 This instruction can also take any number of :ref:`fast-math
4065 flags <fastmath>`, which are optimization hints to enable otherwise
4066 unsafe floating point optimizations:
4071 .. code-block:: llvm
4073 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
4075 '``udiv``' Instruction
4076 ^^^^^^^^^^^^^^^^^^^^^^
4083 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
4084 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
4089 The '``udiv``' instruction returns the quotient of its two operands.
4094 The two arguments to the '``udiv``' instruction must be
4095 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4096 arguments must have identical types.
4101 The value produced is the unsigned integer quotient of the two operands.
4103 Note that unsigned integer division and signed integer division are
4104 distinct operations; for signed integer division, use '``sdiv``'.
4106 Division by zero leads to undefined behavior.
4108 If the ``exact`` keyword is present, the result value of the ``udiv`` is
4109 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
4110 such, "((a udiv exact b) mul b) == a").
4115 .. code-block:: llvm
4117 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
4119 '``sdiv``' Instruction
4120 ^^^^^^^^^^^^^^^^^^^^^^
4127 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
4128 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
4133 The '``sdiv``' instruction returns the quotient of its two operands.
4138 The two arguments to the '``sdiv``' instruction must be
4139 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4140 arguments must have identical types.
4145 The value produced is the signed integer quotient of the two operands
4146 rounded towards zero.
4148 Note that signed integer division and unsigned integer division are
4149 distinct operations; for unsigned integer division, use '``udiv``'.
4151 Division by zero leads to undefined behavior. Overflow also leads to
4152 undefined behavior; this is a rare case, but can occur, for example, by
4153 doing a 32-bit division of -2147483648 by -1.
4155 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4156 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4161 .. code-block:: llvm
4163 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
4167 '``fdiv``' Instruction
4168 ^^^^^^^^^^^^^^^^^^^^^^
4175 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4180 The '``fdiv``' instruction returns the quotient of its two operands.
4185 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4186 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4187 Both arguments must have identical types.
4192 The value produced is the floating point quotient of the two operands.
4193 This instruction can also take any number of :ref:`fast-math
4194 flags <fastmath>`, which are optimization hints to enable otherwise
4195 unsafe floating point optimizations:
4200 .. code-block:: llvm
4202 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
4204 '``urem``' Instruction
4205 ^^^^^^^^^^^^^^^^^^^^^^
4212 <result> = urem <ty> <op1>, <op2> ; yields ty:result
4217 The '``urem``' instruction returns the remainder from the unsigned
4218 division of its two arguments.
4223 The two arguments to the '``urem``' instruction must be
4224 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4225 arguments must have identical types.
4230 This instruction returns the unsigned integer *remainder* of a division.
4231 This instruction always performs an unsigned division to get the
4234 Note that unsigned integer remainder and signed integer remainder are
4235 distinct operations; for signed integer remainder, use '``srem``'.
4237 Taking the remainder of a division by zero leads to undefined behavior.
4242 .. code-block:: llvm
4244 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
4246 '``srem``' Instruction
4247 ^^^^^^^^^^^^^^^^^^^^^^
4254 <result> = srem <ty> <op1>, <op2> ; yields ty:result
4259 The '``srem``' instruction returns the remainder from the signed
4260 division of its two operands. This instruction can also take
4261 :ref:`vector <t_vector>` versions of the values in which case the elements
4267 The two arguments to the '``srem``' instruction must be
4268 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4269 arguments must have identical types.
4274 This instruction returns the *remainder* of a division (where the result
4275 is either zero or has the same sign as the dividend, ``op1``), not the
4276 *modulo* operator (where the result is either zero or has the same sign
4277 as the divisor, ``op2``) of a value. For more information about the
4278 difference, see `The Math
4279 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4280 table of how this is implemented in various languages, please see
4282 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4284 Note that signed integer remainder and unsigned integer remainder are
4285 distinct operations; for unsigned integer remainder, use '``urem``'.
4287 Taking the remainder of a division by zero leads to undefined behavior.
4288 Overflow also leads to undefined behavior; this is a rare case, but can
4289 occur, for example, by taking the remainder of a 32-bit division of
4290 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4291 rule lets srem be implemented using instructions that return both the
4292 result of the division and the remainder.)
4297 .. code-block:: llvm
4299 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
4303 '``frem``' Instruction
4304 ^^^^^^^^^^^^^^^^^^^^^^
4311 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4316 The '``frem``' instruction returns the remainder from the division of
4322 The two arguments to the '``frem``' instruction must be :ref:`floating
4323 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4324 Both arguments must have identical types.
4329 This instruction returns the *remainder* of a division. The remainder
4330 has the same sign as the dividend. This instruction can also take any
4331 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4332 to enable otherwise unsafe floating point optimizations:
4337 .. code-block:: llvm
4339 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
4343 Bitwise Binary Operations
4344 -------------------------
4346 Bitwise binary operators are used to do various forms of bit-twiddling
4347 in a program. They are generally very efficient instructions and can
4348 commonly be strength reduced from other instructions. They require two
4349 operands of the same type, execute an operation on them, and produce a
4350 single value. The resulting value is the same type as its operands.
4352 '``shl``' Instruction
4353 ^^^^^^^^^^^^^^^^^^^^^
4360 <result> = shl <ty> <op1>, <op2> ; yields ty:result
4361 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
4362 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
4363 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
4368 The '``shl``' instruction returns the first operand shifted to the left
4369 a specified number of bits.
4374 Both arguments to the '``shl``' instruction must be the same
4375 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4376 '``op2``' is treated as an unsigned value.
4381 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4382 where ``n`` is the width of the result. If ``op2`` is (statically or
4383 dynamically) negative or equal to or larger than the number of bits in
4384 ``op1``, the result is undefined. If the arguments are vectors, each
4385 vector element of ``op1`` is shifted by the corresponding shift amount
4388 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4389 value <poisonvalues>` if it shifts out any non-zero bits. If the
4390 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4391 value <poisonvalues>` if it shifts out any bits that disagree with the
4392 resultant sign bit. As such, NUW/NSW have the same semantics as they
4393 would if the shift were expressed as a mul instruction with the same
4394 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4399 .. code-block:: llvm
4401 <result> = shl i32 4, %var ; yields i32: 4 << %var
4402 <result> = shl i32 4, 2 ; yields i32: 16
4403 <result> = shl i32 1, 10 ; yields i32: 1024
4404 <result> = shl i32 1, 32 ; undefined
4405 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4407 '``lshr``' Instruction
4408 ^^^^^^^^^^^^^^^^^^^^^^
4415 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
4416 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
4421 The '``lshr``' instruction (logical shift right) returns the first
4422 operand shifted to the right a specified number of bits with zero fill.
4427 Both arguments to the '``lshr``' instruction must be the same
4428 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4429 '``op2``' is treated as an unsigned value.
4434 This instruction always performs a logical shift right operation. The
4435 most significant bits of the result will be filled with zero bits after
4436 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4437 than the number of bits in ``op1``, the result is undefined. If the
4438 arguments are vectors, each vector element of ``op1`` is shifted by the
4439 corresponding shift amount in ``op2``.
4441 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4442 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4448 .. code-block:: llvm
4450 <result> = lshr i32 4, 1 ; yields i32:result = 2
4451 <result> = lshr i32 4, 2 ; yields i32:result = 1
4452 <result> = lshr i8 4, 3 ; yields i8:result = 0
4453 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
4454 <result> = lshr i32 1, 32 ; undefined
4455 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4457 '``ashr``' Instruction
4458 ^^^^^^^^^^^^^^^^^^^^^^
4465 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
4466 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
4471 The '``ashr``' instruction (arithmetic shift right) returns the first
4472 operand shifted to the right a specified number of bits with sign
4478 Both arguments to the '``ashr``' instruction must be the same
4479 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4480 '``op2``' is treated as an unsigned value.
4485 This instruction always performs an arithmetic shift right operation,
4486 The most significant bits of the result will be filled with the sign bit
4487 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4488 than the number of bits in ``op1``, the result is undefined. If the
4489 arguments are vectors, each vector element of ``op1`` is shifted by the
4490 corresponding shift amount in ``op2``.
4492 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4493 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4499 .. code-block:: llvm
4501 <result> = ashr i32 4, 1 ; yields i32:result = 2
4502 <result> = ashr i32 4, 2 ; yields i32:result = 1
4503 <result> = ashr i8 4, 3 ; yields i8:result = 0
4504 <result> = ashr i8 -2, 1 ; yields i8:result = -1
4505 <result> = ashr i32 1, 32 ; undefined
4506 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4508 '``and``' Instruction
4509 ^^^^^^^^^^^^^^^^^^^^^
4516 <result> = and <ty> <op1>, <op2> ; yields ty:result
4521 The '``and``' instruction returns the bitwise logical and of its two
4527 The two arguments to the '``and``' instruction must be
4528 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4529 arguments must have identical types.
4534 The truth table used for the '``and``' instruction is:
4551 .. code-block:: llvm
4553 <result> = and i32 4, %var ; yields i32:result = 4 & %var
4554 <result> = and i32 15, 40 ; yields i32:result = 8
4555 <result> = and i32 4, 8 ; yields i32:result = 0
4557 '``or``' Instruction
4558 ^^^^^^^^^^^^^^^^^^^^
4565 <result> = or <ty> <op1>, <op2> ; yields ty:result
4570 The '``or``' instruction returns the bitwise logical inclusive or of its
4576 The two arguments to the '``or``' instruction must be
4577 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4578 arguments must have identical types.
4583 The truth table used for the '``or``' instruction is:
4602 <result> = or i32 4, %var ; yields i32:result = 4 | %var
4603 <result> = or i32 15, 40 ; yields i32:result = 47
4604 <result> = or i32 4, 8 ; yields i32:result = 12
4606 '``xor``' Instruction
4607 ^^^^^^^^^^^^^^^^^^^^^
4614 <result> = xor <ty> <op1>, <op2> ; yields ty:result
4619 The '``xor``' instruction returns the bitwise logical exclusive or of
4620 its two operands. The ``xor`` is used to implement the "one's
4621 complement" operation, which is the "~" operator in C.
4626 The two arguments to the '``xor``' instruction must be
4627 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4628 arguments must have identical types.
4633 The truth table used for the '``xor``' instruction is:
4650 .. code-block:: llvm
4652 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
4653 <result> = xor i32 15, 40 ; yields i32:result = 39
4654 <result> = xor i32 4, 8 ; yields i32:result = 12
4655 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
4660 LLVM supports several instructions to represent vector operations in a
4661 target-independent manner. These instructions cover the element-access
4662 and vector-specific operations needed to process vectors effectively.
4663 While LLVM does directly support these vector operations, many
4664 sophisticated algorithms will want to use target-specific intrinsics to
4665 take full advantage of a specific target.
4667 .. _i_extractelement:
4669 '``extractelement``' Instruction
4670 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4677 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
4682 The '``extractelement``' instruction extracts a single scalar element
4683 from a vector at a specified index.
4688 The first operand of an '``extractelement``' instruction is a value of
4689 :ref:`vector <t_vector>` type. The second operand is an index indicating
4690 the position from which to extract the element. The index may be a
4691 variable of any integer type.
4696 The result is a scalar of the same type as the element type of ``val``.
4697 Its value is the value at position ``idx`` of ``val``. If ``idx``
4698 exceeds the length of ``val``, the results are undefined.
4703 .. code-block:: llvm
4705 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4707 .. _i_insertelement:
4709 '``insertelement``' Instruction
4710 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4717 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
4722 The '``insertelement``' instruction inserts a scalar element into a
4723 vector at a specified index.
4728 The first operand of an '``insertelement``' instruction is a value of
4729 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4730 type must equal the element type of the first operand. The third operand
4731 is an index indicating the position at which to insert the value. The
4732 index may be a variable of any integer type.
4737 The result is a vector of the same type as ``val``. Its element values
4738 are those of ``val`` except at position ``idx``, where it gets the value
4739 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4745 .. code-block:: llvm
4747 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4749 .. _i_shufflevector:
4751 '``shufflevector``' Instruction
4752 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4759 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4764 The '``shufflevector``' instruction constructs a permutation of elements
4765 from two input vectors, returning a vector with the same element type as
4766 the input and length that is the same as the shuffle mask.
4771 The first two operands of a '``shufflevector``' instruction are vectors
4772 with the same type. The third argument is a shuffle mask whose element
4773 type is always 'i32'. The result of the instruction is a vector whose
4774 length is the same as the shuffle mask and whose element type is the
4775 same as the element type of the first two operands.
4777 The shuffle mask operand is required to be a constant vector with either
4778 constant integer or undef values.
4783 The elements of the two input vectors are numbered from left to right
4784 across both of the vectors. The shuffle mask operand specifies, for each
4785 element of the result vector, which element of the two input vectors the
4786 result element gets. The element selector may be undef (meaning "don't
4787 care") and the second operand may be undef if performing a shuffle from
4793 .. code-block:: llvm
4795 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4796 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4797 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4798 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4799 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4800 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4801 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4802 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4804 Aggregate Operations
4805 --------------------
4807 LLVM supports several instructions for working with
4808 :ref:`aggregate <t_aggregate>` values.
4812 '``extractvalue``' Instruction
4813 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4820 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4825 The '``extractvalue``' instruction extracts the value of a member field
4826 from an :ref:`aggregate <t_aggregate>` value.
4831 The first operand of an '``extractvalue``' instruction is a value of
4832 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4833 constant indices to specify which value to extract in a similar manner
4834 as indices in a '``getelementptr``' instruction.
4836 The major differences to ``getelementptr`` indexing are:
4838 - Since the value being indexed is not a pointer, the first index is
4839 omitted and assumed to be zero.
4840 - At least one index must be specified.
4841 - Not only struct indices but also array indices must be in bounds.
4846 The result is the value at the position in the aggregate specified by
4852 .. code-block:: llvm
4854 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4858 '``insertvalue``' Instruction
4859 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4866 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4871 The '``insertvalue``' instruction inserts a value into a member field in
4872 an :ref:`aggregate <t_aggregate>` value.
4877 The first operand of an '``insertvalue``' instruction is a value of
4878 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4879 a first-class value to insert. The following operands are constant
4880 indices indicating the position at which to insert the value in a
4881 similar manner as indices in a '``extractvalue``' instruction. The value
4882 to insert must have the same type as the value identified by the
4888 The result is an aggregate of the same type as ``val``. Its value is
4889 that of ``val`` except that the value at the position specified by the
4890 indices is that of ``elt``.
4895 .. code-block:: llvm
4897 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4898 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4899 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4903 Memory Access and Addressing Operations
4904 ---------------------------------------
4906 A key design point of an SSA-based representation is how it represents
4907 memory. In LLVM, no memory locations are in SSA form, which makes things
4908 very simple. This section describes how to read, write, and allocate
4913 '``alloca``' Instruction
4914 ^^^^^^^^^^^^^^^^^^^^^^^^
4921 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
4926 The '``alloca``' instruction allocates memory on the stack frame of the
4927 currently executing function, to be automatically released when this
4928 function returns to its caller. The object is always allocated in the
4929 generic address space (address space zero).
4934 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4935 bytes of memory on the runtime stack, returning a pointer of the
4936 appropriate type to the program. If "NumElements" is specified, it is
4937 the number of elements allocated, otherwise "NumElements" is defaulted
4938 to be one. If a constant alignment is specified, the value result of the
4939 allocation is guaranteed to be aligned to at least that boundary. The
4940 alignment may not be greater than ``1 << 29``. If not specified, or if
4941 zero, the target can choose to align the allocation on any convenient
4942 boundary compatible with the type.
4944 '``type``' may be any sized type.
4949 Memory is allocated; a pointer is returned. The operation is undefined
4950 if there is insufficient stack space for the allocation. '``alloca``'d
4951 memory is automatically released when the function returns. The
4952 '``alloca``' instruction is commonly used to represent automatic
4953 variables that must have an address available. When the function returns
4954 (either with the ``ret`` or ``resume`` instructions), the memory is
4955 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4956 The order in which memory is allocated (ie., which way the stack grows)
4962 .. code-block:: llvm
4964 %ptr = alloca i32 ; yields i32*:ptr
4965 %ptr = alloca i32, i32 4 ; yields i32*:ptr
4966 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
4967 %ptr = alloca i32, align 1024 ; yields i32*:ptr
4971 '``load``' Instruction
4972 ^^^^^^^^^^^^^^^^^^^^^^
4979 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4980 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4981 !<index> = !{ i32 1 }
4986 The '``load``' instruction is used to read from memory.
4991 The argument to the ``load`` instruction specifies the memory address
4992 from which to load. The pointer must point to a :ref:`first
4993 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4994 then the optimizer is not allowed to modify the number or order of
4995 execution of this ``load`` with other :ref:`volatile
4996 operations <volatile>`.
4998 If the ``load`` is marked as ``atomic``, it takes an extra
4999 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5000 ``release`` and ``acq_rel`` orderings are not valid on ``load``
5001 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5002 when they may see multiple atomic stores. The type of the pointee must
5003 be an integer type whose bit width is a power of two greater than or
5004 equal to eight and less than or equal to a target-specific size limit.
5005 ``align`` must be explicitly specified on atomic loads, and the load has
5006 undefined behavior if the alignment is not set to a value which is at
5007 least the size in bytes of the pointee. ``!nontemporal`` does not have
5008 any defined semantics for atomic loads.
5010 The optional constant ``align`` argument specifies the alignment of the
5011 operation (that is, the alignment of the memory address). A value of 0
5012 or an omitted ``align`` argument means that the operation has the ABI
5013 alignment for the target. It is the responsibility of the code emitter
5014 to ensure that the alignment information is correct. Overestimating the
5015 alignment results in undefined behavior. Underestimating the alignment
5016 may produce less efficient code. An alignment of 1 is always safe. The
5017 maximum possible alignment is ``1 << 29``.
5019 The optional ``!nontemporal`` metadata must reference a single
5020 metadata name ``<index>`` corresponding to a metadata node with one
5021 ``i32`` entry of value 1. The existence of the ``!nontemporal``
5022 metadata on the instruction tells the optimizer and code generator
5023 that this load is not expected to be reused in the cache. The code
5024 generator may select special instructions to save cache bandwidth, such
5025 as the ``MOVNT`` instruction on x86.
5027 The optional ``!invariant.load`` metadata must reference a single
5028 metadata name ``<index>`` corresponding to a metadata node with no
5029 entries. The existence of the ``!invariant.load`` metadata on the
5030 instruction tells the optimizer and code generator that this load
5031 address points to memory which does not change value during program
5032 execution. The optimizer may then move this load around, for example, by
5033 hoisting it out of loops using loop invariant code motion.
5038 The location of memory pointed to is loaded. If the value being loaded
5039 is of scalar type then the number of bytes read does not exceed the
5040 minimum number of bytes needed to hold all bits of the type. For
5041 example, loading an ``i24`` reads at most three bytes. When loading a
5042 value of a type like ``i20`` with a size that is not an integral number
5043 of bytes, the result is undefined if the value was not originally
5044 written using a store of the same type.
5049 .. code-block:: llvm
5051 %ptr = alloca i32 ; yields i32*:ptr
5052 store i32 3, i32* %ptr ; yields void
5053 %val = load i32* %ptr ; yields i32:val = i32 3
5057 '``store``' Instruction
5058 ^^^^^^^^^^^^^^^^^^^^^^^
5065 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
5066 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
5071 The '``store``' instruction is used to write to memory.
5076 There are two arguments to the ``store`` instruction: a value to store
5077 and an address at which to store it. The type of the ``<pointer>``
5078 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
5079 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
5080 then the optimizer is not allowed to modify the number or order of
5081 execution of this ``store`` with other :ref:`volatile
5082 operations <volatile>`.
5084 If the ``store`` is marked as ``atomic``, it takes an extra
5085 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5086 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
5087 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5088 when they may see multiple atomic stores. The type of the pointee must
5089 be an integer type whose bit width is a power of two greater than or
5090 equal to eight and less than or equal to a target-specific size limit.
5091 ``align`` must be explicitly specified on atomic stores, and the store
5092 has undefined behavior if the alignment is not set to a value which is
5093 at least the size in bytes of the pointee. ``!nontemporal`` does not
5094 have any defined semantics for atomic stores.
5096 The optional constant ``align`` argument specifies the alignment of the
5097 operation (that is, the alignment of the memory address). A value of 0
5098 or an omitted ``align`` argument means that the operation has the ABI
5099 alignment for the target. It is the responsibility of the code emitter
5100 to ensure that the alignment information is correct. Overestimating the
5101 alignment results in undefined behavior. Underestimating the
5102 alignment may produce less efficient code. An alignment of 1 is always
5103 safe. The maximum possible alignment is ``1 << 29``.
5105 The optional ``!nontemporal`` metadata must reference a single metadata
5106 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
5107 value 1. The existence of the ``!nontemporal`` metadata on the instruction
5108 tells the optimizer and code generator that this load is not expected to
5109 be reused in the cache. The code generator may select special
5110 instructions to save cache bandwidth, such as the MOVNT instruction on
5116 The contents of memory are updated to contain ``<value>`` at the
5117 location specified by the ``<pointer>`` operand. If ``<value>`` is
5118 of scalar type then the number of bytes written does not exceed the
5119 minimum number of bytes needed to hold all bits of the type. For
5120 example, storing an ``i24`` writes at most three bytes. When writing a
5121 value of a type like ``i20`` with a size that is not an integral number
5122 of bytes, it is unspecified what happens to the extra bits that do not
5123 belong to the type, but they will typically be overwritten.
5128 .. code-block:: llvm
5130 %ptr = alloca i32 ; yields i32*:ptr
5131 store i32 3, i32* %ptr ; yields void
5132 %val = load i32* %ptr ; yields i32:val = i32 3
5136 '``fence``' Instruction
5137 ^^^^^^^^^^^^^^^^^^^^^^^
5144 fence [singlethread] <ordering> ; yields void
5149 The '``fence``' instruction is used to introduce happens-before edges
5155 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5156 defines what *synchronizes-with* edges they add. They can only be given
5157 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5162 A fence A which has (at least) ``release`` ordering semantics
5163 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5164 semantics if and only if there exist atomic operations X and Y, both
5165 operating on some atomic object M, such that A is sequenced before X, X
5166 modifies M (either directly or through some side effect of a sequence
5167 headed by X), Y is sequenced before B, and Y observes M. This provides a
5168 *happens-before* dependency between A and B. Rather than an explicit
5169 ``fence``, one (but not both) of the atomic operations X or Y might
5170 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5171 still *synchronize-with* the explicit ``fence`` and establish the
5172 *happens-before* edge.
5174 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5175 ``acquire`` and ``release`` semantics specified above, participates in
5176 the global program order of other ``seq_cst`` operations and/or fences.
5178 The optional ":ref:`singlethread <singlethread>`" argument specifies
5179 that the fence only synchronizes with other fences in the same thread.
5180 (This is useful for interacting with signal handlers.)
5185 .. code-block:: llvm
5187 fence acquire ; yields void
5188 fence singlethread seq_cst ; yields void
5192 '``cmpxchg``' Instruction
5193 ^^^^^^^^^^^^^^^^^^^^^^^^^
5200 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
5205 The '``cmpxchg``' instruction is used to atomically modify memory. It
5206 loads a value in memory and compares it to a given value. If they are
5207 equal, it tries to store a new value into the memory.
5212 There are three arguments to the '``cmpxchg``' instruction: an address
5213 to operate on, a value to compare to the value currently be at that
5214 address, and a new value to place at that address if the compared values
5215 are equal. The type of '<cmp>' must be an integer type whose bit width
5216 is a power of two greater than or equal to eight and less than or equal
5217 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5218 type, and the type of '<pointer>' must be a pointer to that type. If the
5219 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5220 to modify the number or order of execution of this ``cmpxchg`` with
5221 other :ref:`volatile operations <volatile>`.
5223 The success and failure :ref:`ordering <ordering>` arguments specify how this
5224 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
5225 must be at least ``monotonic``, the ordering constraint on failure must be no
5226 stronger than that on success, and the failure ordering cannot be either
5227 ``release`` or ``acq_rel``.
5229 The optional "``singlethread``" argument declares that the ``cmpxchg``
5230 is only atomic with respect to code (usually signal handlers) running in
5231 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5232 respect to all other code in the system.
5234 The pointer passed into cmpxchg must have alignment greater than or
5235 equal to the size in memory of the operand.
5240 The contents of memory at the location specified by the '``<pointer>``' operand
5241 is read and compared to '``<cmp>``'; if the read value is the equal, the
5242 '``<new>``' is written. The original value at the location is returned, together
5243 with a flag indicating success (true) or failure (false).
5245 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
5246 permitted: the operation may not write ``<new>`` even if the comparison
5249 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
5250 if the value loaded equals ``cmp``.
5252 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5253 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5254 load with an ordering parameter determined the second ordering parameter.
5259 .. code-block:: llvm
5262 %orig = atomic load i32* %ptr unordered ; yields i32
5266 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5267 %squared = mul i32 %cmp, %cmp
5268 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
5269 %value_loaded = extractvalue { i32, i1 } %val_success, 0
5270 %success = extractvalue { i32, i1 } %val_success, 1
5271 br i1 %success, label %done, label %loop
5278 '``atomicrmw``' Instruction
5279 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5286 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
5291 The '``atomicrmw``' instruction is used to atomically modify memory.
5296 There are three arguments to the '``atomicrmw``' instruction: an
5297 operation to apply, an address whose value to modify, an argument to the
5298 operation. The operation must be one of the following keywords:
5312 The type of '<value>' must be an integer type whose bit width is a power
5313 of two greater than or equal to eight and less than or equal to a
5314 target-specific size limit. The type of the '``<pointer>``' operand must
5315 be a pointer to that type. If the ``atomicrmw`` is marked as
5316 ``volatile``, then the optimizer is not allowed to modify the number or
5317 order of execution of this ``atomicrmw`` with other :ref:`volatile
5318 operations <volatile>`.
5323 The contents of memory at the location specified by the '``<pointer>``'
5324 operand are atomically read, modified, and written back. The original
5325 value at the location is returned. The modification is specified by the
5328 - xchg: ``*ptr = val``
5329 - add: ``*ptr = *ptr + val``
5330 - sub: ``*ptr = *ptr - val``
5331 - and: ``*ptr = *ptr & val``
5332 - nand: ``*ptr = ~(*ptr & val)``
5333 - or: ``*ptr = *ptr | val``
5334 - xor: ``*ptr = *ptr ^ val``
5335 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5336 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5337 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5339 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5345 .. code-block:: llvm
5347 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
5349 .. _i_getelementptr:
5351 '``getelementptr``' Instruction
5352 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5359 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5360 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5361 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5366 The '``getelementptr``' instruction is used to get the address of a
5367 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5368 address calculation only and does not access memory.
5373 The first argument is always a pointer or a vector of pointers, and
5374 forms the basis of the calculation. The remaining arguments are indices
5375 that indicate which of the elements of the aggregate object are indexed.
5376 The interpretation of each index is dependent on the type being indexed
5377 into. The first index always indexes the pointer value given as the
5378 first argument, the second index indexes a value of the type pointed to
5379 (not necessarily the value directly pointed to, since the first index
5380 can be non-zero), etc. The first type indexed into must be a pointer
5381 value, subsequent types can be arrays, vectors, and structs. Note that
5382 subsequent types being indexed into can never be pointers, since that
5383 would require loading the pointer before continuing calculation.
5385 The type of each index argument depends on the type it is indexing into.
5386 When indexing into a (optionally packed) structure, only ``i32`` integer
5387 **constants** are allowed (when using a vector of indices they must all
5388 be the **same** ``i32`` integer constant). When indexing into an array,
5389 pointer or vector, integers of any width are allowed, and they are not
5390 required to be constant. These integers are treated as signed values
5393 For example, let's consider a C code fragment and how it gets compiled
5409 int *foo(struct ST *s) {
5410 return &s[1].Z.B[5][13];
5413 The LLVM code generated by Clang is:
5415 .. code-block:: llvm
5417 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5418 %struct.ST = type { i32, double, %struct.RT }
5420 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5422 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5429 In the example above, the first index is indexing into the
5430 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5431 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5432 indexes into the third element of the structure, yielding a
5433 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5434 structure. The third index indexes into the second element of the
5435 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5436 dimensions of the array are subscripted into, yielding an '``i32``'
5437 type. The '``getelementptr``' instruction returns a pointer to this
5438 element, thus computing a value of '``i32*``' type.
5440 Note that it is perfectly legal to index partially through a structure,
5441 returning a pointer to an inner element. Because of this, the LLVM code
5442 for the given testcase is equivalent to:
5444 .. code-block:: llvm
5446 define i32* @foo(%struct.ST* %s) {
5447 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5448 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5449 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5450 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5451 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5455 If the ``inbounds`` keyword is present, the result value of the
5456 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5457 pointer is not an *in bounds* address of an allocated object, or if any
5458 of the addresses that would be formed by successive addition of the
5459 offsets implied by the indices to the base address with infinitely
5460 precise signed arithmetic are not an *in bounds* address of that
5461 allocated object. The *in bounds* addresses for an allocated object are
5462 all the addresses that point into the object, plus the address one byte
5463 past the end. In cases where the base is a vector of pointers the
5464 ``inbounds`` keyword applies to each of the computations element-wise.
5466 If the ``inbounds`` keyword is not present, the offsets are added to the
5467 base address with silently-wrapping two's complement arithmetic. If the
5468 offsets have a different width from the pointer, they are sign-extended
5469 or truncated to the width of the pointer. The result value of the
5470 ``getelementptr`` may be outside the object pointed to by the base
5471 pointer. The result value may not necessarily be used to access memory
5472 though, even if it happens to point into allocated storage. See the
5473 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5476 The getelementptr instruction is often confusing. For some more insight
5477 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5482 .. code-block:: llvm
5484 ; yields [12 x i8]*:aptr
5485 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5487 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5489 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5491 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5493 In cases where the pointer argument is a vector of pointers, each index
5494 must be a vector with the same number of elements. For example:
5496 .. code-block:: llvm
5498 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5500 Conversion Operations
5501 ---------------------
5503 The instructions in this category are the conversion instructions
5504 (casting) which all take a single operand and a type. They perform
5505 various bit conversions on the operand.
5507 '``trunc .. to``' Instruction
5508 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5515 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5520 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5525 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5526 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5527 of the same number of integers. The bit size of the ``value`` must be
5528 larger than the bit size of the destination type, ``ty2``. Equal sized
5529 types are not allowed.
5534 The '``trunc``' instruction truncates the high order bits in ``value``
5535 and converts the remaining bits to ``ty2``. Since the source size must
5536 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5537 It will always truncate bits.
5542 .. code-block:: llvm
5544 %X = trunc i32 257 to i8 ; yields i8:1
5545 %Y = trunc i32 123 to i1 ; yields i1:true
5546 %Z = trunc i32 122 to i1 ; yields i1:false
5547 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5549 '``zext .. to``' Instruction
5550 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5557 <result> = zext <ty> <value> to <ty2> ; yields ty2
5562 The '``zext``' instruction zero extends its operand to type ``ty2``.
5567 The '``zext``' instruction takes a value to cast, and a type to cast it
5568 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5569 the same number of integers. The bit size of the ``value`` must be
5570 smaller than the bit size of the destination type, ``ty2``.
5575 The ``zext`` fills the high order bits of the ``value`` with zero bits
5576 until it reaches the size of the destination type, ``ty2``.
5578 When zero extending from i1, the result will always be either 0 or 1.
5583 .. code-block:: llvm
5585 %X = zext i32 257 to i64 ; yields i64:257
5586 %Y = zext i1 true to i32 ; yields i32:1
5587 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5589 '``sext .. to``' Instruction
5590 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5597 <result> = sext <ty> <value> to <ty2> ; yields ty2
5602 The '``sext``' sign extends ``value`` to the type ``ty2``.
5607 The '``sext``' instruction takes a value to cast, and a type to cast it
5608 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5609 the same number of integers. The bit size of the ``value`` must be
5610 smaller than the bit size of the destination type, ``ty2``.
5615 The '``sext``' instruction performs a sign extension by copying the sign
5616 bit (highest order bit) of the ``value`` until it reaches the bit size
5617 of the type ``ty2``.
5619 When sign extending from i1, the extension always results in -1 or 0.
5624 .. code-block:: llvm
5626 %X = sext i8 -1 to i16 ; yields i16 :65535
5627 %Y = sext i1 true to i32 ; yields i32:-1
5628 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5630 '``fptrunc .. to``' Instruction
5631 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5638 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5643 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5648 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5649 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5650 The size of ``value`` must be larger than the size of ``ty2``. This
5651 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5656 The '``fptrunc``' instruction truncates a ``value`` from a larger
5657 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5658 point <t_floating>` type. If the value cannot fit within the
5659 destination type, ``ty2``, then the results are undefined.
5664 .. code-block:: llvm
5666 %X = fptrunc double 123.0 to float ; yields float:123.0
5667 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5669 '``fpext .. to``' Instruction
5670 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5677 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5682 The '``fpext``' extends a floating point ``value`` to a larger floating
5688 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5689 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5690 to. The source type must be smaller than the destination type.
5695 The '``fpext``' instruction extends the ``value`` from a smaller
5696 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5697 point <t_floating>` type. The ``fpext`` cannot be used to make a
5698 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5699 *no-op cast* for a floating point cast.
5704 .. code-block:: llvm
5706 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5707 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5709 '``fptoui .. to``' Instruction
5710 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5717 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5722 The '``fptoui``' converts a floating point ``value`` to its unsigned
5723 integer equivalent of type ``ty2``.
5728 The '``fptoui``' instruction takes a value to cast, which must be a
5729 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5730 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5731 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5732 type with the same number of elements as ``ty``
5737 The '``fptoui``' instruction converts its :ref:`floating
5738 point <t_floating>` operand into the nearest (rounding towards zero)
5739 unsigned integer value. If the value cannot fit in ``ty2``, the results
5745 .. code-block:: llvm
5747 %X = fptoui double 123.0 to i32 ; yields i32:123
5748 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5749 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5751 '``fptosi .. to``' Instruction
5752 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5759 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5764 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5765 ``value`` to type ``ty2``.
5770 The '``fptosi``' instruction takes a value to cast, which must be a
5771 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5772 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5773 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5774 type with the same number of elements as ``ty``
5779 The '``fptosi``' instruction converts its :ref:`floating
5780 point <t_floating>` operand into the nearest (rounding towards zero)
5781 signed integer value. If the value cannot fit in ``ty2``, the results
5787 .. code-block:: llvm
5789 %X = fptosi double -123.0 to i32 ; yields i32:-123
5790 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5791 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5793 '``uitofp .. to``' Instruction
5794 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5801 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5806 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5807 and converts that value to the ``ty2`` type.
5812 The '``uitofp``' instruction takes a value to cast, which must be a
5813 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5814 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5815 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5816 type with the same number of elements as ``ty``
5821 The '``uitofp``' instruction interprets its operand as an unsigned
5822 integer quantity and converts it to the corresponding floating point
5823 value. If the value cannot fit in the floating point value, the results
5829 .. code-block:: llvm
5831 %X = uitofp i32 257 to float ; yields float:257.0
5832 %Y = uitofp i8 -1 to double ; yields double:255.0
5834 '``sitofp .. to``' Instruction
5835 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5842 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5847 The '``sitofp``' instruction regards ``value`` as a signed integer and
5848 converts that value to the ``ty2`` type.
5853 The '``sitofp``' instruction takes a value to cast, which must be a
5854 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5855 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5856 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5857 type with the same number of elements as ``ty``
5862 The '``sitofp``' instruction interprets its operand as a signed integer
5863 quantity and converts it to the corresponding floating point value. If
5864 the value cannot fit in the floating point value, the results are
5870 .. code-block:: llvm
5872 %X = sitofp i32 257 to float ; yields float:257.0
5873 %Y = sitofp i8 -1 to double ; yields double:-1.0
5877 '``ptrtoint .. to``' Instruction
5878 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5885 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5890 The '``ptrtoint``' instruction converts the pointer or a vector of
5891 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5896 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5897 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5898 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5899 a vector of integers type.
5904 The '``ptrtoint``' instruction converts ``value`` to integer type
5905 ``ty2`` by interpreting the pointer value as an integer and either
5906 truncating or zero extending that value to the size of the integer type.
5907 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5908 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5909 the same size, then nothing is done (*no-op cast*) other than a type
5915 .. code-block:: llvm
5917 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5918 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5919 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5923 '``inttoptr .. to``' Instruction
5924 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5931 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5936 The '``inttoptr``' instruction converts an integer ``value`` to a
5937 pointer type, ``ty2``.
5942 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5943 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5949 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5950 applying either a zero extension or a truncation depending on the size
5951 of the integer ``value``. If ``value`` is larger than the size of a
5952 pointer then a truncation is done. If ``value`` is smaller than the size
5953 of a pointer then a zero extension is done. If they are the same size,
5954 nothing is done (*no-op cast*).
5959 .. code-block:: llvm
5961 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5962 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5963 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5964 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5968 '``bitcast .. to``' Instruction
5969 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5976 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5981 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5987 The '``bitcast``' instruction takes a value to cast, which must be a
5988 non-aggregate first class value, and a type to cast it to, which must
5989 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5990 bit sizes of ``value`` and the destination type, ``ty2``, must be
5991 identical. If the source type is a pointer, the destination type must
5992 also be a pointer of the same size. This instruction supports bitwise
5993 conversion of vectors to integers and to vectors of other types (as
5994 long as they have the same size).
5999 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
6000 is always a *no-op cast* because no bits change with this
6001 conversion. The conversion is done as if the ``value`` had been stored
6002 to memory and read back as type ``ty2``. Pointer (or vector of
6003 pointers) types may only be converted to other pointer (or vector of
6004 pointers) types with the same address space through this instruction.
6005 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
6006 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
6011 .. code-block:: llvm
6013 %X = bitcast i8 255 to i8 ; yields i8 :-1
6014 %Y = bitcast i32* %x to sint* ; yields sint*:%x
6015 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
6016 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
6018 .. _i_addrspacecast:
6020 '``addrspacecast .. to``' Instruction
6021 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6028 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
6033 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
6034 address space ``n`` to type ``pty2`` in address space ``m``.
6039 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
6040 to cast and a pointer type to cast it to, which must have a different
6046 The '``addrspacecast``' instruction converts the pointer value
6047 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
6048 value modification, depending on the target and the address space
6049 pair. Pointer conversions within the same address space must be
6050 performed with the ``bitcast`` instruction. Note that if the address space
6051 conversion is legal then both result and operand refer to the same memory
6057 .. code-block:: llvm
6059 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
6060 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
6061 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
6068 The instructions in this category are the "miscellaneous" instructions,
6069 which defy better classification.
6073 '``icmp``' Instruction
6074 ^^^^^^^^^^^^^^^^^^^^^^
6081 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6086 The '``icmp``' instruction returns a boolean value or a vector of
6087 boolean values based on comparison of its two integer, integer vector,
6088 pointer, or pointer vector operands.
6093 The '``icmp``' instruction takes three operands. The first operand is
6094 the condition code indicating the kind of comparison to perform. It is
6095 not a value, just a keyword. The possible condition code are:
6098 #. ``ne``: not equal
6099 #. ``ugt``: unsigned greater than
6100 #. ``uge``: unsigned greater or equal
6101 #. ``ult``: unsigned less than
6102 #. ``ule``: unsigned less or equal
6103 #. ``sgt``: signed greater than
6104 #. ``sge``: signed greater or equal
6105 #. ``slt``: signed less than
6106 #. ``sle``: signed less or equal
6108 The remaining two arguments must be :ref:`integer <t_integer>` or
6109 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
6110 must also be identical types.
6115 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
6116 code given as ``cond``. The comparison performed always yields either an
6117 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
6119 #. ``eq``: yields ``true`` if the operands are equal, ``false``
6120 otherwise. No sign interpretation is necessary or performed.
6121 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
6122 otherwise. No sign interpretation is necessary or performed.
6123 #. ``ugt``: interprets the operands as unsigned values and yields
6124 ``true`` if ``op1`` is greater than ``op2``.
6125 #. ``uge``: interprets the operands as unsigned values and yields
6126 ``true`` if ``op1`` is greater than or equal to ``op2``.
6127 #. ``ult``: interprets the operands as unsigned values and yields
6128 ``true`` if ``op1`` is less than ``op2``.
6129 #. ``ule``: interprets the operands as unsigned values and yields
6130 ``true`` if ``op1`` is less than or equal to ``op2``.
6131 #. ``sgt``: interprets the operands as signed values and yields ``true``
6132 if ``op1`` is greater than ``op2``.
6133 #. ``sge``: interprets the operands as signed values and yields ``true``
6134 if ``op1`` is greater than or equal to ``op2``.
6135 #. ``slt``: interprets the operands as signed values and yields ``true``
6136 if ``op1`` is less than ``op2``.
6137 #. ``sle``: interprets the operands as signed values and yields ``true``
6138 if ``op1`` is less than or equal to ``op2``.
6140 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
6141 are compared as if they were integers.
6143 If the operands are integer vectors, then they are compared element by
6144 element. The result is an ``i1`` vector with the same number of elements
6145 as the values being compared. Otherwise, the result is an ``i1``.
6150 .. code-block:: llvm
6152 <result> = icmp eq i32 4, 5 ; yields: result=false
6153 <result> = icmp ne float* %X, %X ; yields: result=false
6154 <result> = icmp ult i16 4, 5 ; yields: result=true
6155 <result> = icmp sgt i16 4, 5 ; yields: result=false
6156 <result> = icmp ule i16 -4, 5 ; yields: result=false
6157 <result> = icmp sge i16 4, 5 ; yields: result=false
6159 Note that the code generator does not yet support vector types with the
6160 ``icmp`` instruction.
6164 '``fcmp``' Instruction
6165 ^^^^^^^^^^^^^^^^^^^^^^
6172 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6177 The '``fcmp``' instruction returns a boolean value or vector of boolean
6178 values based on comparison of its operands.
6180 If the operands are floating point scalars, then the result type is a
6181 boolean (:ref:`i1 <t_integer>`).
6183 If the operands are floating point vectors, then the result type is a
6184 vector of boolean with the same number of elements as the operands being
6190 The '``fcmp``' instruction takes three operands. The first operand is
6191 the condition code indicating the kind of comparison to perform. It is
6192 not a value, just a keyword. The possible condition code are:
6194 #. ``false``: no comparison, always returns false
6195 #. ``oeq``: ordered and equal
6196 #. ``ogt``: ordered and greater than
6197 #. ``oge``: ordered and greater than or equal
6198 #. ``olt``: ordered and less than
6199 #. ``ole``: ordered and less than or equal
6200 #. ``one``: ordered and not equal
6201 #. ``ord``: ordered (no nans)
6202 #. ``ueq``: unordered or equal
6203 #. ``ugt``: unordered or greater than
6204 #. ``uge``: unordered or greater than or equal
6205 #. ``ult``: unordered or less than
6206 #. ``ule``: unordered or less than or equal
6207 #. ``une``: unordered or not equal
6208 #. ``uno``: unordered (either nans)
6209 #. ``true``: no comparison, always returns true
6211 *Ordered* means that neither operand is a QNAN while *unordered* means
6212 that either operand may be a QNAN.
6214 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6215 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6216 type. They must have identical types.
6221 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6222 condition code given as ``cond``. If the operands are vectors, then the
6223 vectors are compared element by element. Each comparison performed
6224 always yields an :ref:`i1 <t_integer>` result, as follows:
6226 #. ``false``: always yields ``false``, regardless of operands.
6227 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6228 is equal to ``op2``.
6229 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6230 is greater than ``op2``.
6231 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6232 is greater than or equal to ``op2``.
6233 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6234 is less than ``op2``.
6235 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6236 is less than or equal to ``op2``.
6237 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6238 is not equal to ``op2``.
6239 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6240 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6242 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6243 greater than ``op2``.
6244 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6245 greater than or equal to ``op2``.
6246 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6248 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6249 less than or equal to ``op2``.
6250 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6251 not equal to ``op2``.
6252 #. ``uno``: yields ``true`` if either operand is a QNAN.
6253 #. ``true``: always yields ``true``, regardless of operands.
6258 .. code-block:: llvm
6260 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6261 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6262 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6263 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6265 Note that the code generator does not yet support vector types with the
6266 ``fcmp`` instruction.
6270 '``phi``' Instruction
6271 ^^^^^^^^^^^^^^^^^^^^^
6278 <result> = phi <ty> [ <val0>, <label0>], ...
6283 The '``phi``' instruction is used to implement the φ node in the SSA
6284 graph representing the function.
6289 The type of the incoming values is specified with the first type field.
6290 After this, the '``phi``' instruction takes a list of pairs as
6291 arguments, with one pair for each predecessor basic block of the current
6292 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6293 the value arguments to the PHI node. Only labels may be used as the
6296 There must be no non-phi instructions between the start of a basic block
6297 and the PHI instructions: i.e. PHI instructions must be first in a basic
6300 For the purposes of the SSA form, the use of each incoming value is
6301 deemed to occur on the edge from the corresponding predecessor block to
6302 the current block (but after any definition of an '``invoke``'
6303 instruction's return value on the same edge).
6308 At runtime, the '``phi``' instruction logically takes on the value
6309 specified by the pair corresponding to the predecessor basic block that
6310 executed just prior to the current block.
6315 .. code-block:: llvm
6317 Loop: ; Infinite loop that counts from 0 on up...
6318 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6319 %nextindvar = add i32 %indvar, 1
6324 '``select``' Instruction
6325 ^^^^^^^^^^^^^^^^^^^^^^^^
6332 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6334 selty is either i1 or {<N x i1>}
6339 The '``select``' instruction is used to choose one value based on a
6340 condition, without IR-level branching.
6345 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6346 values indicating the condition, and two values of the same :ref:`first
6347 class <t_firstclass>` type. If the val1/val2 are vectors and the
6348 condition is a scalar, then entire vectors are selected, not individual
6354 If the condition is an i1 and it evaluates to 1, the instruction returns
6355 the first value argument; otherwise, it returns the second value
6358 If the condition is a vector of i1, then the value arguments must be
6359 vectors of the same size, and the selection is done element by element.
6364 .. code-block:: llvm
6366 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6370 '``call``' Instruction
6371 ^^^^^^^^^^^^^^^^^^^^^^
6378 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6383 The '``call``' instruction represents a simple function call.
6388 This instruction requires several arguments:
6390 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
6391 should perform tail call optimization. The ``tail`` marker is a hint that
6392 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
6393 means that the call must be tail call optimized in order for the program to
6394 be correct. The ``musttail`` marker provides these guarantees:
6396 #. The call will not cause unbounded stack growth if it is part of a
6397 recursive cycle in the call graph.
6398 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
6401 Both markers imply that the callee does not access allocas or varargs from
6402 the caller. Calls marked ``musttail`` must obey the following additional
6405 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
6406 or a pointer bitcast followed by a ret instruction.
6407 - The ret instruction must return the (possibly bitcasted) value
6408 produced by the call or void.
6409 - The caller and callee prototypes must match. Pointer types of
6410 parameters or return types may differ in pointee type, but not
6412 - The calling conventions of the caller and callee must match.
6413 - All ABI-impacting function attributes, such as sret, byval, inreg,
6414 returned, and inalloca, must match.
6416 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
6417 the following conditions are met:
6419 - Caller and callee both have the calling convention ``fastcc``.
6420 - The call is in tail position (ret immediately follows call and ret
6421 uses value of call or is void).
6422 - Option ``-tailcallopt`` is enabled, or
6423 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6424 - `Platform-specific constraints are
6425 met. <CodeGenerator.html#tailcallopt>`_
6427 #. The optional "cconv" marker indicates which :ref:`calling
6428 convention <callingconv>` the call should use. If none is
6429 specified, the call defaults to using C calling conventions. The
6430 calling convention of the call must match the calling convention of
6431 the target function, or else the behavior is undefined.
6432 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6433 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6435 #. '``ty``': the type of the call instruction itself which is also the
6436 type of the return value. Functions that return no value are marked
6438 #. '``fnty``': shall be the signature of the pointer to function value
6439 being invoked. The argument types must match the types implied by
6440 this signature. This type can be omitted if the function is not
6441 varargs and if the function type does not return a pointer to a
6443 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6444 be invoked. In most cases, this is a direct function invocation, but
6445 indirect ``call``'s are just as possible, calling an arbitrary pointer
6447 #. '``function args``': argument list whose types match the function
6448 signature argument types and parameter attributes. All arguments must
6449 be of :ref:`first class <t_firstclass>` type. If the function signature
6450 indicates the function accepts a variable number of arguments, the
6451 extra arguments can be specified.
6452 #. The optional :ref:`function attributes <fnattrs>` list. Only
6453 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6454 attributes are valid here.
6459 The '``call``' instruction is used to cause control flow to transfer to
6460 a specified function, with its incoming arguments bound to the specified
6461 values. Upon a '``ret``' instruction in the called function, control
6462 flow continues with the instruction after the function call, and the
6463 return value of the function is bound to the result argument.
6468 .. code-block:: llvm
6470 %retval = call i32 @test(i32 %argc)
6471 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6472 %X = tail call i32 @foo() ; yields i32
6473 %Y = tail call fastcc i32 @foo() ; yields i32
6474 call void %foo(i8 97 signext)
6476 %struct.A = type { i32, i8 }
6477 %r = call %struct.A @foo() ; yields { i32, i8 }
6478 %gr = extractvalue %struct.A %r, 0 ; yields i32
6479 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6480 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6481 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6483 llvm treats calls to some functions with names and arguments that match
6484 the standard C99 library as being the C99 library functions, and may
6485 perform optimizations or generate code for them under that assumption.
6486 This is something we'd like to change in the future to provide better
6487 support for freestanding environments and non-C-based languages.
6491 '``va_arg``' Instruction
6492 ^^^^^^^^^^^^^^^^^^^^^^^^
6499 <resultval> = va_arg <va_list*> <arglist>, <argty>
6504 The '``va_arg``' instruction is used to access arguments passed through
6505 the "variable argument" area of a function call. It is used to implement
6506 the ``va_arg`` macro in C.
6511 This instruction takes a ``va_list*`` value and the type of the
6512 argument. It returns a value of the specified argument type and
6513 increments the ``va_list`` to point to the next argument. The actual
6514 type of ``va_list`` is target specific.
6519 The '``va_arg``' instruction loads an argument of the specified type
6520 from the specified ``va_list`` and causes the ``va_list`` to point to
6521 the next argument. For more information, see the variable argument
6522 handling :ref:`Intrinsic Functions <int_varargs>`.
6524 It is legal for this instruction to be called in a function which does
6525 not take a variable number of arguments, for example, the ``vfprintf``
6528 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6529 function <intrinsics>` because it takes a type as an argument.
6534 See the :ref:`variable argument processing <int_varargs>` section.
6536 Note that the code generator does not yet fully support va\_arg on many
6537 targets. Also, it does not currently support va\_arg with aggregate
6538 types on any target.
6542 '``landingpad``' Instruction
6543 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6550 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6551 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6553 <clause> := catch <type> <value>
6554 <clause> := filter <array constant type> <array constant>
6559 The '``landingpad``' instruction is used by `LLVM's exception handling
6560 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6561 is a landing pad --- one where the exception lands, and corresponds to the
6562 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6563 defines values supplied by the personality function (``pers_fn``) upon
6564 re-entry to the function. The ``resultval`` has the type ``resultty``.
6569 This instruction takes a ``pers_fn`` value. This is the personality
6570 function associated with the unwinding mechanism. The optional
6571 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6573 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6574 contains the global variable representing the "type" that may be caught
6575 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6576 clause takes an array constant as its argument. Use
6577 "``[0 x i8**] undef``" for a filter which cannot throw. The
6578 '``landingpad``' instruction must contain *at least* one ``clause`` or
6579 the ``cleanup`` flag.
6584 The '``landingpad``' instruction defines the values which are set by the
6585 personality function (``pers_fn``) upon re-entry to the function, and
6586 therefore the "result type" of the ``landingpad`` instruction. As with
6587 calling conventions, how the personality function results are
6588 represented in LLVM IR is target specific.
6590 The clauses are applied in order from top to bottom. If two
6591 ``landingpad`` instructions are merged together through inlining, the
6592 clauses from the calling function are appended to the list of clauses.
6593 When the call stack is being unwound due to an exception being thrown,
6594 the exception is compared against each ``clause`` in turn. If it doesn't
6595 match any of the clauses, and the ``cleanup`` flag is not set, then
6596 unwinding continues further up the call stack.
6598 The ``landingpad`` instruction has several restrictions:
6600 - A landing pad block is a basic block which is the unwind destination
6601 of an '``invoke``' instruction.
6602 - A landing pad block must have a '``landingpad``' instruction as its
6603 first non-PHI instruction.
6604 - There can be only one '``landingpad``' instruction within the landing
6606 - A basic block that is not a landing pad block may not include a
6607 '``landingpad``' instruction.
6608 - All '``landingpad``' instructions in a function must have the same
6609 personality function.
6614 .. code-block:: llvm
6616 ;; A landing pad which can catch an integer.
6617 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6619 ;; A landing pad that is a cleanup.
6620 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6622 ;; A landing pad which can catch an integer and can only throw a double.
6623 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6625 filter [1 x i8**] [@_ZTId]
6632 LLVM supports the notion of an "intrinsic function". These functions
6633 have well known names and semantics and are required to follow certain
6634 restrictions. Overall, these intrinsics represent an extension mechanism
6635 for the LLVM language that does not require changing all of the
6636 transformations in LLVM when adding to the language (or the bitcode
6637 reader/writer, the parser, etc...).
6639 Intrinsic function names must all start with an "``llvm.``" prefix. This
6640 prefix is reserved in LLVM for intrinsic names; thus, function names may
6641 not begin with this prefix. Intrinsic functions must always be external
6642 functions: you cannot define the body of intrinsic functions. Intrinsic
6643 functions may only be used in call or invoke instructions: it is illegal
6644 to take the address of an intrinsic function. Additionally, because
6645 intrinsic functions are part of the LLVM language, it is required if any
6646 are added that they be documented here.
6648 Some intrinsic functions can be overloaded, i.e., the intrinsic
6649 represents a family of functions that perform the same operation but on
6650 different data types. Because LLVM can represent over 8 million
6651 different integer types, overloading is used commonly to allow an
6652 intrinsic function to operate on any integer type. One or more of the
6653 argument types or the result type can be overloaded to accept any
6654 integer type. Argument types may also be defined as exactly matching a
6655 previous argument's type or the result type. This allows an intrinsic
6656 function which accepts multiple arguments, but needs all of them to be
6657 of the same type, to only be overloaded with respect to a single
6658 argument or the result.
6660 Overloaded intrinsics will have the names of its overloaded argument
6661 types encoded into its function name, each preceded by a period. Only
6662 those types which are overloaded result in a name suffix. Arguments
6663 whose type is matched against another type do not. For example, the
6664 ``llvm.ctpop`` function can take an integer of any width and returns an
6665 integer of exactly the same integer width. This leads to a family of
6666 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6667 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6668 overloaded, and only one type suffix is required. Because the argument's
6669 type is matched against the return type, it does not require its own
6672 To learn how to add an intrinsic function, please see the `Extending
6673 LLVM Guide <ExtendingLLVM.html>`_.
6677 Variable Argument Handling Intrinsics
6678 -------------------------------------
6680 Variable argument support is defined in LLVM with the
6681 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6682 functions. These functions are related to the similarly named macros
6683 defined in the ``<stdarg.h>`` header file.
6685 All of these functions operate on arguments that use a target-specific
6686 value type "``va_list``". The LLVM assembly language reference manual
6687 does not define what this type is, so all transformations should be
6688 prepared to handle these functions regardless of the type used.
6690 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6691 variable argument handling intrinsic functions are used.
6693 .. code-block:: llvm
6695 define i32 @test(i32 %X, ...) {
6696 ; Initialize variable argument processing
6698 %ap2 = bitcast i8** %ap to i8*
6699 call void @llvm.va_start(i8* %ap2)
6701 ; Read a single integer argument
6702 %tmp = va_arg i8** %ap, i32
6704 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6706 %aq2 = bitcast i8** %aq to i8*
6707 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6708 call void @llvm.va_end(i8* %aq2)
6710 ; Stop processing of arguments.
6711 call void @llvm.va_end(i8* %ap2)
6715 declare void @llvm.va_start(i8*)
6716 declare void @llvm.va_copy(i8*, i8*)
6717 declare void @llvm.va_end(i8*)
6721 '``llvm.va_start``' Intrinsic
6722 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6729 declare void @llvm.va_start(i8* <arglist>)
6734 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6735 subsequent use by ``va_arg``.
6740 The argument is a pointer to a ``va_list`` element to initialize.
6745 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6746 available in C. In a target-dependent way, it initializes the
6747 ``va_list`` element to which the argument points, so that the next call
6748 to ``va_arg`` will produce the first variable argument passed to the
6749 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6750 to know the last argument of the function as the compiler can figure
6753 '``llvm.va_end``' Intrinsic
6754 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6761 declare void @llvm.va_end(i8* <arglist>)
6766 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6767 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6772 The argument is a pointer to a ``va_list`` to destroy.
6777 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6778 available in C. In a target-dependent way, it destroys the ``va_list``
6779 element to which the argument points. Calls to
6780 :ref:`llvm.va_start <int_va_start>` and
6781 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6786 '``llvm.va_copy``' Intrinsic
6787 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6794 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6799 The '``llvm.va_copy``' intrinsic copies the current argument position
6800 from the source argument list to the destination argument list.
6805 The first argument is a pointer to a ``va_list`` element to initialize.
6806 The second argument is a pointer to a ``va_list`` element to copy from.
6811 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6812 available in C. In a target-dependent way, it copies the source
6813 ``va_list`` element into the destination ``va_list`` element. This
6814 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6815 arbitrarily complex and require, for example, memory allocation.
6817 Accurate Garbage Collection Intrinsics
6818 --------------------------------------
6820 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6821 (GC) requires the implementation and generation of these intrinsics.
6822 These intrinsics allow identification of :ref:`GC roots on the
6823 stack <int_gcroot>`, as well as garbage collector implementations that
6824 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6825 Front-ends for type-safe garbage collected languages should generate
6826 these intrinsics to make use of the LLVM garbage collectors. For more
6827 details, see `Accurate Garbage Collection with
6828 LLVM <GarbageCollection.html>`_.
6830 The garbage collection intrinsics only operate on objects in the generic
6831 address space (address space zero).
6835 '``llvm.gcroot``' Intrinsic
6836 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6843 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6848 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6849 the code generator, and allows some metadata to be associated with it.
6854 The first argument specifies the address of a stack object that contains
6855 the root pointer. The second pointer (which must be either a constant or
6856 a global value address) contains the meta-data to be associated with the
6862 At runtime, a call to this intrinsic stores a null pointer into the
6863 "ptrloc" location. At compile-time, the code generator generates
6864 information to allow the runtime to find the pointer at GC safe points.
6865 The '``llvm.gcroot``' intrinsic may only be used in a function which
6866 :ref:`specifies a GC algorithm <gc>`.
6870 '``llvm.gcread``' Intrinsic
6871 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6878 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6883 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6884 locations, allowing garbage collector implementations that require read
6890 The second argument is the address to read from, which should be an
6891 address allocated from the garbage collector. The first object is a
6892 pointer to the start of the referenced object, if needed by the language
6893 runtime (otherwise null).
6898 The '``llvm.gcread``' intrinsic has the same semantics as a load
6899 instruction, but may be replaced with substantially more complex code by
6900 the garbage collector runtime, as needed. The '``llvm.gcread``'
6901 intrinsic may only be used in a function which :ref:`specifies a GC
6906 '``llvm.gcwrite``' Intrinsic
6907 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6914 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6919 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6920 locations, allowing garbage collector implementations that require write
6921 barriers (such as generational or reference counting collectors).
6926 The first argument is the reference to store, the second is the start of
6927 the object to store it to, and the third is the address of the field of
6928 Obj to store to. If the runtime does not require a pointer to the
6929 object, Obj may be null.
6934 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6935 instruction, but may be replaced with substantially more complex code by
6936 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6937 intrinsic may only be used in a function which :ref:`specifies a GC
6940 Code Generator Intrinsics
6941 -------------------------
6943 These intrinsics are provided by LLVM to expose special features that
6944 may only be implemented with code generator support.
6946 '``llvm.returnaddress``' Intrinsic
6947 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6954 declare i8 *@llvm.returnaddress(i32 <level>)
6959 The '``llvm.returnaddress``' intrinsic attempts to compute a
6960 target-specific value indicating the return address of the current
6961 function or one of its callers.
6966 The argument to this intrinsic indicates which function to return the
6967 address for. Zero indicates the calling function, one indicates its
6968 caller, etc. The argument is **required** to be a constant integer
6974 The '``llvm.returnaddress``' intrinsic either returns a pointer
6975 indicating the return address of the specified call frame, or zero if it
6976 cannot be identified. The value returned by this intrinsic is likely to
6977 be incorrect or 0 for arguments other than zero, so it should only be
6978 used for debugging purposes.
6980 Note that calling this intrinsic does not prevent function inlining or
6981 other aggressive transformations, so the value returned may not be that
6982 of the obvious source-language caller.
6984 '``llvm.frameaddress``' Intrinsic
6985 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6992 declare i8* @llvm.frameaddress(i32 <level>)
6997 The '``llvm.frameaddress``' intrinsic attempts to return the
6998 target-specific frame pointer value for the specified stack frame.
7003 The argument to this intrinsic indicates which function to return the
7004 frame pointer for. Zero indicates the calling function, one indicates
7005 its caller, etc. The argument is **required** to be a constant integer
7011 The '``llvm.frameaddress``' intrinsic either returns a pointer
7012 indicating the frame address of the specified call frame, or zero if it
7013 cannot be identified. The value returned by this intrinsic is likely to
7014 be incorrect or 0 for arguments other than zero, so it should only be
7015 used for debugging purposes.
7017 Note that calling this intrinsic does not prevent function inlining or
7018 other aggressive transformations, so the value returned may not be that
7019 of the obvious source-language caller.
7021 .. _int_read_register:
7022 .. _int_write_register:
7024 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
7025 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7032 declare i32 @llvm.read_register.i32(metadata)
7033 declare i64 @llvm.read_register.i64(metadata)
7034 declare void @llvm.write_register.i32(metadata, i32 @value)
7035 declare void @llvm.write_register.i64(metadata, i64 @value)
7036 !0 = metadata !{metadata !"sp\00"}
7041 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
7042 provides access to the named register. The register must be valid on
7043 the architecture being compiled to. The type needs to be compatible
7044 with the register being read.
7049 The '``llvm.read_register``' intrinsic returns the current value of the
7050 register, where possible. The '``llvm.write_register``' intrinsic sets
7051 the current value of the register, where possible.
7053 This is useful to implement named register global variables that need
7054 to always be mapped to a specific register, as is common practice on
7055 bare-metal programs including OS kernels.
7057 The compiler doesn't check for register availability or use of the used
7058 register in surrounding code, including inline assembly. Because of that,
7059 allocatable registers are not supported.
7061 Warning: So far it only works with the stack pointer on selected
7062 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
7063 work is needed to support other registers and even more so, allocatable
7068 '``llvm.stacksave``' Intrinsic
7069 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7076 declare i8* @llvm.stacksave()
7081 The '``llvm.stacksave``' intrinsic is used to remember the current state
7082 of the function stack, for use with
7083 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
7084 implementing language features like scoped automatic variable sized
7090 This intrinsic returns a opaque pointer value that can be passed to
7091 :ref:`llvm.stackrestore <int_stackrestore>`. When an
7092 ``llvm.stackrestore`` intrinsic is executed with a value saved from
7093 ``llvm.stacksave``, it effectively restores the state of the stack to
7094 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
7095 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
7096 were allocated after the ``llvm.stacksave`` was executed.
7098 .. _int_stackrestore:
7100 '``llvm.stackrestore``' Intrinsic
7101 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7108 declare void @llvm.stackrestore(i8* %ptr)
7113 The '``llvm.stackrestore``' intrinsic is used to restore the state of
7114 the function stack to the state it was in when the corresponding
7115 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
7116 useful for implementing language features like scoped automatic variable
7117 sized arrays in C99.
7122 See the description for :ref:`llvm.stacksave <int_stacksave>`.
7124 '``llvm.prefetch``' Intrinsic
7125 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7132 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7137 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
7138 insert a prefetch instruction if supported; otherwise, it is a noop.
7139 Prefetches have no effect on the behavior of the program but can change
7140 its performance characteristics.
7145 ``address`` is the address to be prefetched, ``rw`` is the specifier
7146 determining if the fetch should be for a read (0) or write (1), and
7147 ``locality`` is a temporal locality specifier ranging from (0) - no
7148 locality, to (3) - extremely local keep in cache. The ``cache type``
7149 specifies whether the prefetch is performed on the data (1) or
7150 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
7151 arguments must be constant integers.
7156 This intrinsic does not modify the behavior of the program. In
7157 particular, prefetches cannot trap and do not produce a value. On
7158 targets that support this intrinsic, the prefetch can provide hints to
7159 the processor cache for better performance.
7161 '``llvm.pcmarker``' Intrinsic
7162 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7169 declare void @llvm.pcmarker(i32 <id>)
7174 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7175 Counter (PC) in a region of code to simulators and other tools. The
7176 method is target specific, but it is expected that the marker will use
7177 exported symbols to transmit the PC of the marker. The marker makes no
7178 guarantees that it will remain with any specific instruction after
7179 optimizations. It is possible that the presence of a marker will inhibit
7180 optimizations. The intended use is to be inserted after optimizations to
7181 allow correlations of simulation runs.
7186 ``id`` is a numerical id identifying the marker.
7191 This intrinsic does not modify the behavior of the program. Backends
7192 that do not support this intrinsic may ignore it.
7194 '``llvm.readcyclecounter``' Intrinsic
7195 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7202 declare i64 @llvm.readcyclecounter()
7207 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
7208 counter register (or similar low latency, high accuracy clocks) on those
7209 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7210 should map to RPCC. As the backing counters overflow quickly (on the
7211 order of 9 seconds on alpha), this should only be used for small
7217 When directly supported, reading the cycle counter should not modify any
7218 memory. Implementations are allowed to either return a application
7219 specific value or a system wide value. On backends without support, this
7220 is lowered to a constant 0.
7222 Note that runtime support may be conditional on the privilege-level code is
7223 running at and the host platform.
7225 '``llvm.clear_cache``' Intrinsic
7226 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7233 declare void @llvm.clear_cache(i8*, i8*)
7238 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
7239 in the specified range to the execution unit of the processor. On
7240 targets with non-unified instruction and data cache, the implementation
7241 flushes the instruction cache.
7246 On platforms with coherent instruction and data caches (e.g. x86), this
7247 intrinsic is a nop. On platforms with non-coherent instruction and data
7248 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
7249 instructions or a system call, if cache flushing requires special
7252 The default behavior is to emit a call to ``__clear_cache`` from the run
7255 This instrinsic does *not* empty the instruction pipeline. Modifications
7256 of the current function are outside the scope of the intrinsic.
7258 Standard C Library Intrinsics
7259 -----------------------------
7261 LLVM provides intrinsics for a few important standard C library
7262 functions. These intrinsics allow source-language front-ends to pass
7263 information about the alignment of the pointer arguments to the code
7264 generator, providing opportunity for more efficient code generation.
7268 '``llvm.memcpy``' Intrinsic
7269 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7274 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
7275 integer bit width and for different address spaces. Not all targets
7276 support all bit widths however.
7280 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7281 i32 <len>, i32 <align>, i1 <isvolatile>)
7282 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7283 i64 <len>, i32 <align>, i1 <isvolatile>)
7288 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7289 source location to the destination location.
7291 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
7292 intrinsics do not return a value, takes extra alignment/isvolatile
7293 arguments and the pointers can be in specified address spaces.
7298 The first argument is a pointer to the destination, the second is a
7299 pointer to the source. The third argument is an integer argument
7300 specifying the number of bytes to copy, the fourth argument is the
7301 alignment of the source and destination locations, and the fifth is a
7302 boolean indicating a volatile access.
7304 If the call to this intrinsic has an alignment value that is not 0 or 1,
7305 then the caller guarantees that both the source and destination pointers
7306 are aligned to that boundary.
7308 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
7309 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7310 very cleanly specified and it is unwise to depend on it.
7315 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7316 source location to the destination location, which are not allowed to
7317 overlap. It copies "len" bytes of memory over. If the argument is known
7318 to be aligned to some boundary, this can be specified as the fourth
7319 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
7321 '``llvm.memmove``' Intrinsic
7322 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7327 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7328 bit width and for different address space. Not all targets support all
7333 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7334 i32 <len>, i32 <align>, i1 <isvolatile>)
7335 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7336 i64 <len>, i32 <align>, i1 <isvolatile>)
7341 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7342 source location to the destination location. It is similar to the
7343 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7346 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7347 intrinsics do not return a value, takes extra alignment/isvolatile
7348 arguments and the pointers can be in specified address spaces.
7353 The first argument is a pointer to the destination, the second is a
7354 pointer to the source. The third argument is an integer argument
7355 specifying the number of bytes to copy, the fourth argument is the
7356 alignment of the source and destination locations, and the fifth is a
7357 boolean indicating a volatile access.
7359 If the call to this intrinsic has an alignment value that is not 0 or 1,
7360 then the caller guarantees that the source and destination pointers are
7361 aligned to that boundary.
7363 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7364 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7365 not very cleanly specified and it is unwise to depend on it.
7370 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7371 source location to the destination location, which may overlap. It
7372 copies "len" bytes of memory over. If the argument is known to be
7373 aligned to some boundary, this can be specified as the fourth argument,
7374 otherwise it should be set to 0 or 1 (both meaning no alignment).
7376 '``llvm.memset.*``' Intrinsics
7377 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7382 This is an overloaded intrinsic. You can use llvm.memset on any integer
7383 bit width and for different address spaces. However, not all targets
7384 support all bit widths.
7388 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7389 i32 <len>, i32 <align>, i1 <isvolatile>)
7390 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7391 i64 <len>, i32 <align>, i1 <isvolatile>)
7396 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7397 particular byte value.
7399 Note that, unlike the standard libc function, the ``llvm.memset``
7400 intrinsic does not return a value and takes extra alignment/volatile
7401 arguments. Also, the destination can be in an arbitrary address space.
7406 The first argument is a pointer to the destination to fill, the second
7407 is the byte value with which to fill it, the third argument is an
7408 integer argument specifying the number of bytes to fill, and the fourth
7409 argument is the known alignment of the destination location.
7411 If the call to this intrinsic has an alignment value that is not 0 or 1,
7412 then the caller guarantees that the destination pointer is aligned to
7415 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7416 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7417 very cleanly specified and it is unwise to depend on it.
7422 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7423 at the destination location. If the argument is known to be aligned to
7424 some boundary, this can be specified as the fourth argument, otherwise
7425 it should be set to 0 or 1 (both meaning no alignment).
7427 '``llvm.sqrt.*``' Intrinsic
7428 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7433 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7434 floating point or vector of floating point type. Not all targets support
7439 declare float @llvm.sqrt.f32(float %Val)
7440 declare double @llvm.sqrt.f64(double %Val)
7441 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7442 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7443 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7448 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7449 returning the same value as the libm '``sqrt``' functions would. Unlike
7450 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7451 negative numbers other than -0.0 (which allows for better optimization,
7452 because there is no need to worry about errno being set).
7453 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7458 The argument and return value are floating point numbers of the same
7464 This function returns the sqrt of the specified operand if it is a
7465 nonnegative floating point number.
7467 '``llvm.powi.*``' Intrinsic
7468 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7473 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7474 floating point or vector of floating point type. Not all targets support
7479 declare float @llvm.powi.f32(float %Val, i32 %power)
7480 declare double @llvm.powi.f64(double %Val, i32 %power)
7481 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7482 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7483 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7488 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7489 specified (positive or negative) power. The order of evaluation of
7490 multiplications is not defined. When a vector of floating point type is
7491 used, the second argument remains a scalar integer value.
7496 The second argument is an integer power, and the first is a value to
7497 raise to that power.
7502 This function returns the first value raised to the second power with an
7503 unspecified sequence of rounding operations.
7505 '``llvm.sin.*``' Intrinsic
7506 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7511 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7512 floating point or vector of floating point type. Not all targets support
7517 declare float @llvm.sin.f32(float %Val)
7518 declare double @llvm.sin.f64(double %Val)
7519 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7520 declare fp128 @llvm.sin.f128(fp128 %Val)
7521 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7526 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7531 The argument and return value are floating point numbers of the same
7537 This function returns the sine of the specified operand, returning the
7538 same values as the libm ``sin`` functions would, and handles error
7539 conditions in the same way.
7541 '``llvm.cos.*``' Intrinsic
7542 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7547 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7548 floating point or vector of floating point type. Not all targets support
7553 declare float @llvm.cos.f32(float %Val)
7554 declare double @llvm.cos.f64(double %Val)
7555 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7556 declare fp128 @llvm.cos.f128(fp128 %Val)
7557 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7562 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7567 The argument and return value are floating point numbers of the same
7573 This function returns the cosine of the specified operand, returning the
7574 same values as the libm ``cos`` functions would, and handles error
7575 conditions in the same way.
7577 '``llvm.pow.*``' Intrinsic
7578 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7583 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7584 floating point or vector of floating point type. Not all targets support
7589 declare float @llvm.pow.f32(float %Val, float %Power)
7590 declare double @llvm.pow.f64(double %Val, double %Power)
7591 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7592 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7593 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7598 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7599 specified (positive or negative) power.
7604 The second argument is a floating point power, and the first is a value
7605 to raise to that power.
7610 This function returns the first value raised to the second power,
7611 returning the same values as the libm ``pow`` functions would, and
7612 handles error conditions in the same way.
7614 '``llvm.exp.*``' Intrinsic
7615 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7620 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7621 floating point or vector of floating point type. Not all targets support
7626 declare float @llvm.exp.f32(float %Val)
7627 declare double @llvm.exp.f64(double %Val)
7628 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7629 declare fp128 @llvm.exp.f128(fp128 %Val)
7630 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7635 The '``llvm.exp.*``' intrinsics perform the exp function.
7640 The argument and return value are floating point numbers of the same
7646 This function returns the same values as the libm ``exp`` functions
7647 would, and handles error conditions in the same way.
7649 '``llvm.exp2.*``' Intrinsic
7650 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7655 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7656 floating point or vector of floating point type. Not all targets support
7661 declare float @llvm.exp2.f32(float %Val)
7662 declare double @llvm.exp2.f64(double %Val)
7663 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7664 declare fp128 @llvm.exp2.f128(fp128 %Val)
7665 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7670 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7675 The argument and return value are floating point numbers of the same
7681 This function returns the same values as the libm ``exp2`` functions
7682 would, and handles error conditions in the same way.
7684 '``llvm.log.*``' Intrinsic
7685 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7690 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7691 floating point or vector of floating point type. Not all targets support
7696 declare float @llvm.log.f32(float %Val)
7697 declare double @llvm.log.f64(double %Val)
7698 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7699 declare fp128 @llvm.log.f128(fp128 %Val)
7700 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7705 The '``llvm.log.*``' intrinsics perform the log function.
7710 The argument and return value are floating point numbers of the same
7716 This function returns the same values as the libm ``log`` functions
7717 would, and handles error conditions in the same way.
7719 '``llvm.log10.*``' Intrinsic
7720 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7725 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7726 floating point or vector of floating point type. Not all targets support
7731 declare float @llvm.log10.f32(float %Val)
7732 declare double @llvm.log10.f64(double %Val)
7733 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7734 declare fp128 @llvm.log10.f128(fp128 %Val)
7735 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7740 The '``llvm.log10.*``' intrinsics perform the log10 function.
7745 The argument and return value are floating point numbers of the same
7751 This function returns the same values as the libm ``log10`` functions
7752 would, and handles error conditions in the same way.
7754 '``llvm.log2.*``' Intrinsic
7755 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7760 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7761 floating point or vector of floating point type. Not all targets support
7766 declare float @llvm.log2.f32(float %Val)
7767 declare double @llvm.log2.f64(double %Val)
7768 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7769 declare fp128 @llvm.log2.f128(fp128 %Val)
7770 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7775 The '``llvm.log2.*``' intrinsics perform the log2 function.
7780 The argument and return value are floating point numbers of the same
7786 This function returns the same values as the libm ``log2`` functions
7787 would, and handles error conditions in the same way.
7789 '``llvm.fma.*``' Intrinsic
7790 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7795 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7796 floating point or vector of floating point type. Not all targets support
7801 declare float @llvm.fma.f32(float %a, float %b, float %c)
7802 declare double @llvm.fma.f64(double %a, double %b, double %c)
7803 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7804 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7805 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7810 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7816 The argument and return value are floating point numbers of the same
7822 This function returns the same values as the libm ``fma`` functions
7823 would, and does not set errno.
7825 '``llvm.fabs.*``' Intrinsic
7826 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7831 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7832 floating point or vector of floating point type. Not all targets support
7837 declare float @llvm.fabs.f32(float %Val)
7838 declare double @llvm.fabs.f64(double %Val)
7839 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7840 declare fp128 @llvm.fabs.f128(fp128 %Val)
7841 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7846 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7852 The argument and return value are floating point numbers of the same
7858 This function returns the same values as the libm ``fabs`` functions
7859 would, and handles error conditions in the same way.
7861 '``llvm.copysign.*``' Intrinsic
7862 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7867 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7868 floating point or vector of floating point type. Not all targets support
7873 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7874 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7875 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7876 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7877 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7882 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7883 first operand and the sign of the second operand.
7888 The arguments and return value are floating point numbers of the same
7894 This function returns the same values as the libm ``copysign``
7895 functions would, and handles error conditions in the same way.
7897 '``llvm.floor.*``' Intrinsic
7898 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7903 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7904 floating point or vector of floating point type. Not all targets support
7909 declare float @llvm.floor.f32(float %Val)
7910 declare double @llvm.floor.f64(double %Val)
7911 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7912 declare fp128 @llvm.floor.f128(fp128 %Val)
7913 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7918 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7923 The argument and return value are floating point numbers of the same
7929 This function returns the same values as the libm ``floor`` functions
7930 would, and handles error conditions in the same way.
7932 '``llvm.ceil.*``' Intrinsic
7933 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7938 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7939 floating point or vector of floating point type. Not all targets support
7944 declare float @llvm.ceil.f32(float %Val)
7945 declare double @llvm.ceil.f64(double %Val)
7946 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7947 declare fp128 @llvm.ceil.f128(fp128 %Val)
7948 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7953 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7958 The argument and return value are floating point numbers of the same
7964 This function returns the same values as the libm ``ceil`` functions
7965 would, and handles error conditions in the same way.
7967 '``llvm.trunc.*``' Intrinsic
7968 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7973 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7974 floating point or vector of floating point type. Not all targets support
7979 declare float @llvm.trunc.f32(float %Val)
7980 declare double @llvm.trunc.f64(double %Val)
7981 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7982 declare fp128 @llvm.trunc.f128(fp128 %Val)
7983 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7988 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7989 nearest integer not larger in magnitude than the operand.
7994 The argument and return value are floating point numbers of the same
8000 This function returns the same values as the libm ``trunc`` functions
8001 would, and handles error conditions in the same way.
8003 '``llvm.rint.*``' Intrinsic
8004 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8009 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
8010 floating point or vector of floating point type. Not all targets support
8015 declare float @llvm.rint.f32(float %Val)
8016 declare double @llvm.rint.f64(double %Val)
8017 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
8018 declare fp128 @llvm.rint.f128(fp128 %Val)
8019 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
8024 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
8025 nearest integer. It may raise an inexact floating-point exception if the
8026 operand isn't an integer.
8031 The argument and return value are floating point numbers of the same
8037 This function returns the same values as the libm ``rint`` functions
8038 would, and handles error conditions in the same way.
8040 '``llvm.nearbyint.*``' Intrinsic
8041 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8046 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
8047 floating point or vector of floating point type. Not all targets support
8052 declare float @llvm.nearbyint.f32(float %Val)
8053 declare double @llvm.nearbyint.f64(double %Val)
8054 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
8055 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
8056 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
8061 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
8067 The argument and return value are floating point numbers of the same
8073 This function returns the same values as the libm ``nearbyint``
8074 functions would, and handles error conditions in the same way.
8076 '``llvm.round.*``' Intrinsic
8077 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8082 This is an overloaded intrinsic. You can use ``llvm.round`` on any
8083 floating point or vector of floating point type. Not all targets support
8088 declare float @llvm.round.f32(float %Val)
8089 declare double @llvm.round.f64(double %Val)
8090 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
8091 declare fp128 @llvm.round.f128(fp128 %Val)
8092 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
8097 The '``llvm.round.*``' intrinsics returns the operand rounded to the
8103 The argument and return value are floating point numbers of the same
8109 This function returns the same values as the libm ``round``
8110 functions would, and handles error conditions in the same way.
8112 Bit Manipulation Intrinsics
8113 ---------------------------
8115 LLVM provides intrinsics for a few important bit manipulation
8116 operations. These allow efficient code generation for some algorithms.
8118 '``llvm.bswap.*``' Intrinsics
8119 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8124 This is an overloaded intrinsic function. You can use bswap on any
8125 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
8129 declare i16 @llvm.bswap.i16(i16 <id>)
8130 declare i32 @llvm.bswap.i32(i32 <id>)
8131 declare i64 @llvm.bswap.i64(i64 <id>)
8136 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
8137 values with an even number of bytes (positive multiple of 16 bits).
8138 These are useful for performing operations on data that is not in the
8139 target's native byte order.
8144 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
8145 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
8146 intrinsic returns an i32 value that has the four bytes of the input i32
8147 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
8148 returned i32 will have its bytes in 3, 2, 1, 0 order. The
8149 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
8150 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
8153 '``llvm.ctpop.*``' Intrinsic
8154 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8159 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
8160 bit width, or on any vector with integer elements. Not all targets
8161 support all bit widths or vector types, however.
8165 declare i8 @llvm.ctpop.i8(i8 <src>)
8166 declare i16 @llvm.ctpop.i16(i16 <src>)
8167 declare i32 @llvm.ctpop.i32(i32 <src>)
8168 declare i64 @llvm.ctpop.i64(i64 <src>)
8169 declare i256 @llvm.ctpop.i256(i256 <src>)
8170 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
8175 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
8181 The only argument is the value to be counted. The argument may be of any
8182 integer type, or a vector with integer elements. The return type must
8183 match the argument type.
8188 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
8189 each element of a vector.
8191 '``llvm.ctlz.*``' Intrinsic
8192 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8197 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
8198 integer bit width, or any vector whose elements are integers. Not all
8199 targets support all bit widths or vector types, however.
8203 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
8204 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
8205 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
8206 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
8207 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
8208 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8213 The '``llvm.ctlz``' family of intrinsic functions counts the number of
8214 leading zeros in a variable.
8219 The first argument is the value to be counted. This argument may be of
8220 any integer type, or a vectory with integer element type. The return
8221 type must match the first argument type.
8223 The second argument must be a constant and is a flag to indicate whether
8224 the intrinsic should ensure that a zero as the first argument produces a
8225 defined result. Historically some architectures did not provide a
8226 defined result for zero values as efficiently, and many algorithms are
8227 now predicated on avoiding zero-value inputs.
8232 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
8233 zeros in a variable, or within each element of the vector. If
8234 ``src == 0`` then the result is the size in bits of the type of ``src``
8235 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8236 ``llvm.ctlz(i32 2) = 30``.
8238 '``llvm.cttz.*``' Intrinsic
8239 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8244 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
8245 integer bit width, or any vector of integer elements. Not all targets
8246 support all bit widths or vector types, however.
8250 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
8251 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
8252 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
8253 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
8254 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
8255 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8260 The '``llvm.cttz``' family of intrinsic functions counts the number of
8266 The first argument is the value to be counted. This argument may be of
8267 any integer type, or a vectory with integer element type. The return
8268 type must match the first argument type.
8270 The second argument must be a constant and is a flag to indicate whether
8271 the intrinsic should ensure that a zero as the first argument produces a
8272 defined result. Historically some architectures did not provide a
8273 defined result for zero values as efficiently, and many algorithms are
8274 now predicated on avoiding zero-value inputs.
8279 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
8280 zeros in a variable, or within each element of a vector. If ``src == 0``
8281 then the result is the size in bits of the type of ``src`` if
8282 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8283 ``llvm.cttz(2) = 1``.
8285 Arithmetic with Overflow Intrinsics
8286 -----------------------------------
8288 LLVM provides intrinsics for some arithmetic with overflow operations.
8290 '``llvm.sadd.with.overflow.*``' Intrinsics
8291 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8296 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
8297 on any integer bit width.
8301 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8302 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8303 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8308 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8309 a signed addition of the two arguments, and indicate whether an overflow
8310 occurred during the signed summation.
8315 The arguments (%a and %b) and the first element of the result structure
8316 may be of integer types of any bit width, but they must have the same
8317 bit width. The second element of the result structure must be of type
8318 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8324 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8325 a signed addition of the two variables. They return a structure --- the
8326 first element of which is the signed summation, and the second element
8327 of which is a bit specifying if the signed summation resulted in an
8333 .. code-block:: llvm
8335 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8336 %sum = extractvalue {i32, i1} %res, 0
8337 %obit = extractvalue {i32, i1} %res, 1
8338 br i1 %obit, label %overflow, label %normal
8340 '``llvm.uadd.with.overflow.*``' Intrinsics
8341 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8346 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8347 on any integer bit width.
8351 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8352 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8353 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8358 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8359 an unsigned addition of the two arguments, and indicate whether a carry
8360 occurred during the unsigned summation.
8365 The arguments (%a and %b) and the first element of the result structure
8366 may be of integer types of any bit width, but they must have the same
8367 bit width. The second element of the result structure must be of type
8368 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8374 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8375 an unsigned addition of the two arguments. They return a structure --- the
8376 first element of which is the sum, and the second element of which is a
8377 bit specifying if the unsigned summation resulted in a carry.
8382 .. code-block:: llvm
8384 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8385 %sum = extractvalue {i32, i1} %res, 0
8386 %obit = extractvalue {i32, i1} %res, 1
8387 br i1 %obit, label %carry, label %normal
8389 '``llvm.ssub.with.overflow.*``' Intrinsics
8390 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8395 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8396 on any integer bit width.
8400 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8401 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8402 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8407 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8408 a signed subtraction of the two arguments, and indicate whether an
8409 overflow occurred during the signed subtraction.
8414 The arguments (%a and %b) and the first element of the result structure
8415 may be of integer types of any bit width, but they must have the same
8416 bit width. The second element of the result structure must be of type
8417 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8423 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8424 a signed subtraction of the two arguments. They return a structure --- the
8425 first element of which is the subtraction, and the second element of
8426 which is a bit specifying if the signed subtraction resulted in an
8432 .. code-block:: llvm
8434 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8435 %sum = extractvalue {i32, i1} %res, 0
8436 %obit = extractvalue {i32, i1} %res, 1
8437 br i1 %obit, label %overflow, label %normal
8439 '``llvm.usub.with.overflow.*``' Intrinsics
8440 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8445 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8446 on any integer bit width.
8450 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8451 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8452 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8457 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8458 an unsigned subtraction of the two arguments, and indicate whether an
8459 overflow occurred during the unsigned subtraction.
8464 The arguments (%a and %b) and the first element of the result structure
8465 may be of integer types of any bit width, but they must have the same
8466 bit width. The second element of the result structure must be of type
8467 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8473 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8474 an unsigned subtraction of the two arguments. They return a structure ---
8475 the first element of which is the subtraction, and the second element of
8476 which is a bit specifying if the unsigned subtraction resulted in an
8482 .. code-block:: llvm
8484 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8485 %sum = extractvalue {i32, i1} %res, 0
8486 %obit = extractvalue {i32, i1} %res, 1
8487 br i1 %obit, label %overflow, label %normal
8489 '``llvm.smul.with.overflow.*``' Intrinsics
8490 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8495 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8496 on any integer bit width.
8500 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8501 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8502 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8507 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8508 a signed multiplication of the two arguments, and indicate whether an
8509 overflow occurred during the signed multiplication.
8514 The arguments (%a and %b) and the first element of the result structure
8515 may be of integer types of any bit width, but they must have the same
8516 bit width. The second element of the result structure must be of type
8517 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8523 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8524 a signed multiplication of the two arguments. They return a structure ---
8525 the first element of which is the multiplication, and the second element
8526 of which is a bit specifying if the signed multiplication resulted in an
8532 .. code-block:: llvm
8534 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8535 %sum = extractvalue {i32, i1} %res, 0
8536 %obit = extractvalue {i32, i1} %res, 1
8537 br i1 %obit, label %overflow, label %normal
8539 '``llvm.umul.with.overflow.*``' Intrinsics
8540 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8545 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8546 on any integer bit width.
8550 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8551 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8552 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8557 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8558 a unsigned multiplication of the two arguments, and indicate whether an
8559 overflow occurred during the unsigned multiplication.
8564 The arguments (%a and %b) and the first element of the result structure
8565 may be of integer types of any bit width, but they must have the same
8566 bit width. The second element of the result structure must be of type
8567 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8573 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8574 an unsigned multiplication of the two arguments. They return a structure ---
8575 the first element of which is the multiplication, and the second
8576 element of which is a bit specifying if the unsigned multiplication
8577 resulted in an overflow.
8582 .. code-block:: llvm
8584 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8585 %sum = extractvalue {i32, i1} %res, 0
8586 %obit = extractvalue {i32, i1} %res, 1
8587 br i1 %obit, label %overflow, label %normal
8589 Specialised Arithmetic Intrinsics
8590 ---------------------------------
8592 '``llvm.fmuladd.*``' Intrinsic
8593 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8600 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8601 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8606 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8607 expressions that can be fused if the code generator determines that (a) the
8608 target instruction set has support for a fused operation, and (b) that the
8609 fused operation is more efficient than the equivalent, separate pair of mul
8610 and add instructions.
8615 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8616 multiplicands, a and b, and an addend c.
8625 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8627 is equivalent to the expression a \* b + c, except that rounding will
8628 not be performed between the multiplication and addition steps if the
8629 code generator fuses the operations. Fusion is not guaranteed, even if
8630 the target platform supports it. If a fused multiply-add is required the
8631 corresponding llvm.fma.\* intrinsic function should be used
8632 instead. This never sets errno, just as '``llvm.fma.*``'.
8637 .. code-block:: llvm
8639 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
8641 Half Precision Floating Point Intrinsics
8642 ----------------------------------------
8644 For most target platforms, half precision floating point is a
8645 storage-only format. This means that it is a dense encoding (in memory)
8646 but does not support computation in the format.
8648 This means that code must first load the half-precision floating point
8649 value as an i16, then convert it to float with
8650 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8651 then be performed on the float value (including extending to double
8652 etc). To store the value back to memory, it is first converted to float
8653 if needed, then converted to i16 with
8654 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8657 .. _int_convert_to_fp16:
8659 '``llvm.convert.to.fp16``' Intrinsic
8660 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8667 declare i16 @llvm.convert.to.fp16.f32(float %a)
8668 declare i16 @llvm.convert.to.fp16.f64(double %a)
8673 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
8674 conventional floating point type to half precision floating point format.
8679 The intrinsic function contains single argument - the value to be
8685 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
8686 conventional floating point format to half precision floating point format. The
8687 return value is an ``i16`` which contains the converted number.
8692 .. code-block:: llvm
8694 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
8695 store i16 %res, i16* @x, align 2
8697 .. _int_convert_from_fp16:
8699 '``llvm.convert.from.fp16``' Intrinsic
8700 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8707 declare float @llvm.convert.from.fp16.f32(i16 %a)
8708 declare double @llvm.convert.from.fp16.f64(i16 %a)
8713 The '``llvm.convert.from.fp16``' intrinsic function performs a
8714 conversion from half precision floating point format to single precision
8715 floating point format.
8720 The intrinsic function contains single argument - the value to be
8726 The '``llvm.convert.from.fp16``' intrinsic function performs a
8727 conversion from half single precision floating point format to single
8728 precision floating point format. The input half-float value is
8729 represented by an ``i16`` value.
8734 .. code-block:: llvm
8736 %a = load i16* @x, align 2
8737 %res = call float @llvm.convert.from.fp16(i16 %a)
8742 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8743 prefix), are described in the `LLVM Source Level
8744 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8747 Exception Handling Intrinsics
8748 -----------------------------
8750 The LLVM exception handling intrinsics (which all start with
8751 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8752 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8756 Trampoline Intrinsics
8757 ---------------------
8759 These intrinsics make it possible to excise one parameter, marked with
8760 the :ref:`nest <nest>` attribute, from a function. The result is a
8761 callable function pointer lacking the nest parameter - the caller does
8762 not need to provide a value for it. Instead, the value to use is stored
8763 in advance in a "trampoline", a block of memory usually allocated on the
8764 stack, which also contains code to splice the nest value into the
8765 argument list. This is used to implement the GCC nested function address
8768 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8769 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8770 It can be created as follows:
8772 .. code-block:: llvm
8774 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8775 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8776 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8777 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8778 %fp = bitcast i8* %p to i32 (i32, i32)*
8780 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8781 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8785 '``llvm.init.trampoline``' Intrinsic
8786 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8793 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8798 This fills the memory pointed to by ``tramp`` with executable code,
8799 turning it into a trampoline.
8804 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8805 pointers. The ``tramp`` argument must point to a sufficiently large and
8806 sufficiently aligned block of memory; this memory is written to by the
8807 intrinsic. Note that the size and the alignment are target-specific -
8808 LLVM currently provides no portable way of determining them, so a
8809 front-end that generates this intrinsic needs to have some
8810 target-specific knowledge. The ``func`` argument must hold a function
8811 bitcast to an ``i8*``.
8816 The block of memory pointed to by ``tramp`` is filled with target
8817 dependent code, turning it into a function. Then ``tramp`` needs to be
8818 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8819 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8820 function's signature is the same as that of ``func`` with any arguments
8821 marked with the ``nest`` attribute removed. At most one such ``nest``
8822 argument is allowed, and it must be of pointer type. Calling the new
8823 function is equivalent to calling ``func`` with the same argument list,
8824 but with ``nval`` used for the missing ``nest`` argument. If, after
8825 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8826 modified, then the effect of any later call to the returned function
8827 pointer is undefined.
8831 '``llvm.adjust.trampoline``' Intrinsic
8832 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8839 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8844 This performs any required machine-specific adjustment to the address of
8845 a trampoline (passed as ``tramp``).
8850 ``tramp`` must point to a block of memory which already has trampoline
8851 code filled in by a previous call to
8852 :ref:`llvm.init.trampoline <int_it>`.
8857 On some architectures the address of the code to be executed needs to be
8858 different than the address where the trampoline is actually stored. This
8859 intrinsic returns the executable address corresponding to ``tramp``
8860 after performing the required machine specific adjustments. The pointer
8861 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8866 This class of intrinsics provides information about the lifetime of
8867 memory objects and ranges where variables are immutable.
8871 '``llvm.lifetime.start``' Intrinsic
8872 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8879 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8884 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8890 The first argument is a constant integer representing the size of the
8891 object, or -1 if it is variable sized. The second argument is a pointer
8897 This intrinsic indicates that before this point in the code, the value
8898 of the memory pointed to by ``ptr`` is dead. This means that it is known
8899 to never be used and has an undefined value. A load from the pointer
8900 that precedes this intrinsic can be replaced with ``'undef'``.
8904 '``llvm.lifetime.end``' Intrinsic
8905 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8912 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8917 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8923 The first argument is a constant integer representing the size of the
8924 object, or -1 if it is variable sized. The second argument is a pointer
8930 This intrinsic indicates that after this point in the code, the value of
8931 the memory pointed to by ``ptr`` is dead. This means that it is known to
8932 never be used and has an undefined value. Any stores into the memory
8933 object following this intrinsic may be removed as dead.
8935 '``llvm.invariant.start``' Intrinsic
8936 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8943 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8948 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8949 a memory object will not change.
8954 The first argument is a constant integer representing the size of the
8955 object, or -1 if it is variable sized. The second argument is a pointer
8961 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8962 the return value, the referenced memory location is constant and
8965 '``llvm.invariant.end``' Intrinsic
8966 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8973 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8978 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8979 memory object are mutable.
8984 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8985 The second argument is a constant integer representing the size of the
8986 object, or -1 if it is variable sized and the third argument is a
8987 pointer to the object.
8992 This intrinsic indicates that the memory is mutable again.
8997 This class of intrinsics is designed to be generic and has no specific
9000 '``llvm.var.annotation``' Intrinsic
9001 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9008 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9013 The '``llvm.var.annotation``' intrinsic.
9018 The first argument is a pointer to a value, the second is a pointer to a
9019 global string, the third is a pointer to a global string which is the
9020 source file name, and the last argument is the line number.
9025 This intrinsic allows annotation of local variables with arbitrary
9026 strings. This can be useful for special purpose optimizations that want
9027 to look for these annotations. These have no other defined use; they are
9028 ignored by code generation and optimization.
9030 '``llvm.ptr.annotation.*``' Intrinsic
9031 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9036 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
9037 pointer to an integer of any width. *NOTE* you must specify an address space for
9038 the pointer. The identifier for the default address space is the integer
9043 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9044 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
9045 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
9046 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
9047 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
9052 The '``llvm.ptr.annotation``' intrinsic.
9057 The first argument is a pointer to an integer value of arbitrary bitwidth
9058 (result of some expression), the second is a pointer to a global string, the
9059 third is a pointer to a global string which is the source file name, and the
9060 last argument is the line number. It returns the value of the first argument.
9065 This intrinsic allows annotation of a pointer to an integer with arbitrary
9066 strings. This can be useful for special purpose optimizations that want to look
9067 for these annotations. These have no other defined use; they are ignored by code
9068 generation and optimization.
9070 '``llvm.annotation.*``' Intrinsic
9071 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9076 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
9077 any integer bit width.
9081 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
9082 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
9083 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
9084 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
9085 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
9090 The '``llvm.annotation``' intrinsic.
9095 The first argument is an integer value (result of some expression), the
9096 second is a pointer to a global string, the third is a pointer to a
9097 global string which is the source file name, and the last argument is
9098 the line number. It returns the value of the first argument.
9103 This intrinsic allows annotations to be put on arbitrary expressions
9104 with arbitrary strings. This can be useful for special purpose
9105 optimizations that want to look for these annotations. These have no
9106 other defined use; they are ignored by code generation and optimization.
9108 '``llvm.trap``' Intrinsic
9109 ^^^^^^^^^^^^^^^^^^^^^^^^^
9116 declare void @llvm.trap() noreturn nounwind
9121 The '``llvm.trap``' intrinsic.
9131 This intrinsic is lowered to the target dependent trap instruction. If
9132 the target does not have a trap instruction, this intrinsic will be
9133 lowered to a call of the ``abort()`` function.
9135 '``llvm.debugtrap``' Intrinsic
9136 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9143 declare void @llvm.debugtrap() nounwind
9148 The '``llvm.debugtrap``' intrinsic.
9158 This intrinsic is lowered to code which is intended to cause an
9159 execution trap with the intention of requesting the attention of a
9162 '``llvm.stackprotector``' Intrinsic
9163 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9170 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
9175 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
9176 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
9177 is placed on the stack before local variables.
9182 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
9183 The first argument is the value loaded from the stack guard
9184 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
9185 enough space to hold the value of the guard.
9190 This intrinsic causes the prologue/epilogue inserter to force the position of
9191 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
9192 to ensure that if a local variable on the stack is overwritten, it will destroy
9193 the value of the guard. When the function exits, the guard on the stack is
9194 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
9195 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
9196 calling the ``__stack_chk_fail()`` function.
9198 '``llvm.stackprotectorcheck``' Intrinsic
9199 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9206 declare void @llvm.stackprotectorcheck(i8** <guard>)
9211 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
9212 created stack protector and if they are not equal calls the
9213 ``__stack_chk_fail()`` function.
9218 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
9219 the variable ``@__stack_chk_guard``.
9224 This intrinsic is provided to perform the stack protector check by comparing
9225 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
9226 values do not match call the ``__stack_chk_fail()`` function.
9228 The reason to provide this as an IR level intrinsic instead of implementing it
9229 via other IR operations is that in order to perform this operation at the IR
9230 level without an intrinsic, one would need to create additional basic blocks to
9231 handle the success/failure cases. This makes it difficult to stop the stack
9232 protector check from disrupting sibling tail calls in Codegen. With this
9233 intrinsic, we are able to generate the stack protector basic blocks late in
9234 codegen after the tail call decision has occurred.
9236 '``llvm.objectsize``' Intrinsic
9237 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9244 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
9245 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
9250 The ``llvm.objectsize`` intrinsic is designed to provide information to
9251 the optimizers to determine at compile time whether a) an operation
9252 (like memcpy) will overflow a buffer that corresponds to an object, or
9253 b) that a runtime check for overflow isn't necessary. An object in this
9254 context means an allocation of a specific class, structure, array, or
9260 The ``llvm.objectsize`` intrinsic takes two arguments. The first
9261 argument is a pointer to or into the ``object``. The second argument is
9262 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
9263 or -1 (if false) when the object size is unknown. The second argument
9264 only accepts constants.
9269 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
9270 the size of the object concerned. If the size cannot be determined at
9271 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
9272 on the ``min`` argument).
9274 '``llvm.expect``' Intrinsic
9275 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9280 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
9285 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
9286 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9287 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9292 The ``llvm.expect`` intrinsic provides information about expected (the
9293 most probable) value of ``val``, which can be used by optimizers.
9298 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
9299 a value. The second argument is an expected value, this needs to be a
9300 constant value, variables are not allowed.
9305 This intrinsic is lowered to the ``val``.
9307 '``llvm.donothing``' Intrinsic
9308 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9315 declare void @llvm.donothing() nounwind readnone
9320 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
9321 only intrinsic that can be called with an invoke instruction.
9331 This intrinsic does nothing, and it's removed by optimizers and ignored
9334 Stack Map Intrinsics
9335 --------------------
9337 LLVM provides experimental intrinsics to support runtime patching
9338 mechanisms commonly desired in dynamic language JITs. These intrinsics
9339 are described in :doc:`StackMaps`.