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, may have an explicit section
523 to be placed in, and may have an optional explicit alignment specified.
525 Global variables in other translation units can also be declared, in which
526 case they don't have an initializer.
528 A variable may be defined as a global ``constant``, which indicates that
529 the contents of the variable will **never** be modified (enabling better
530 optimization, allowing the global data to be placed in the read-only
531 section of an executable, etc). Note that variables that need runtime
532 initialization cannot be marked ``constant`` as there is a store to the
535 LLVM explicitly allows *declarations* of global variables to be marked
536 constant, even if the final definition of the global is not. This
537 capability can be used to enable slightly better optimization of the
538 program, but requires the language definition to guarantee that
539 optimizations based on the 'constantness' are valid for the translation
540 units that do not include the definition.
542 As SSA values, global variables define pointer values that are in scope
543 (i.e. they dominate) all basic blocks in the program. Global variables
544 always define a pointer to their "content" type because they describe a
545 region of memory, and all memory objects in LLVM are accessed through
548 Global variables can be marked with ``unnamed_addr`` which indicates
549 that the address is not significant, only the content. Constants marked
550 like this can be merged with other constants if they have the same
551 initializer. Note that a constant with significant address *can* be
552 merged with a ``unnamed_addr`` constant, the result being a constant
553 whose address is significant.
555 A global variable may be declared to reside in a target-specific
556 numbered address space. For targets that support them, address spaces
557 may affect how optimizations are performed and/or what target
558 instructions are used to access the variable. The default address space
559 is zero. The address space qualifier must precede any other attributes.
561 LLVM allows an explicit section to be specified for globals. If the
562 target supports it, it will emit globals to the section specified.
564 By default, global initializers are optimized by assuming that global
565 variables defined within the module are not modified from their
566 initial values before the start of the global initializer. This is
567 true even for variables potentially accessible from outside the
568 module, including those with external linkage or appearing in
569 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
570 by marking the variable with ``externally_initialized``.
572 An explicit alignment may be specified for a global, which must be a
573 power of 2. If not present, or if the alignment is set to zero, the
574 alignment of the global is set by the target to whatever it feels
575 convenient. If an explicit alignment is specified, the global is forced
576 to have exactly that alignment. Targets and optimizers are not allowed
577 to over-align the global if the global has an assigned section. In this
578 case, the extra alignment could be observable: for example, code could
579 assume that the globals are densely packed in their section and try to
580 iterate over them as an array, alignment padding would break this
583 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
585 Variables and aliasaes can have a
586 :ref:`Thread Local Storage Model <tls_model>`.
590 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
591 [AddrSpace] [unnamed_addr] [ExternallyInitialized]
592 <global | constant> <Type>
593 [, section "name"] [, align <Alignment>]
595 For example, the following defines a global in a numbered address space
596 with an initializer, section, and alignment:
600 @G = addrspace(5) constant float 1.0, section "foo", align 4
602 The following example just declares a global variable
606 @G = external global i32
608 The following example defines a thread-local global with the
609 ``initialexec`` TLS model:
613 @G = thread_local(initialexec) global i32 0, align 4
615 .. _functionstructure:
620 LLVM function definitions consist of the "``define``" keyword, an
621 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
622 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
623 an optional :ref:`calling convention <callingconv>`,
624 an optional ``unnamed_addr`` attribute, a return type, an optional
625 :ref:`parameter attribute <paramattrs>` for the return type, a function
626 name, a (possibly empty) argument list (each with optional :ref:`parameter
627 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
628 an optional section, an optional alignment, an optional :ref:`garbage
629 collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
630 curly brace, a list of basic blocks, and a closing curly brace.
632 LLVM function declarations consist of the "``declare``" keyword, an
633 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
634 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
635 an optional :ref:`calling convention <callingconv>`,
636 an optional ``unnamed_addr`` attribute, a return type, an optional
637 :ref:`parameter attribute <paramattrs>` for the return type, a function
638 name, a possibly empty list of arguments, an optional alignment, an optional
639 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
641 A function definition contains a list of basic blocks, forming the CFG (Control
642 Flow Graph) for the function. Each basic block may optionally start with a label
643 (giving the basic block a symbol table entry), contains a list of instructions,
644 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
645 function return). If an explicit label is not provided, a block is assigned an
646 implicit numbered label, using the next value from the same counter as used for
647 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
648 entry block does not have an explicit label, it will be assigned label "%0",
649 then the first unnamed temporary in that block will be "%1", etc.
651 The first basic block in a function is special in two ways: it is
652 immediately executed on entrance to the function, and it is not allowed
653 to have predecessor basic blocks (i.e. there can not be any branches to
654 the entry block of a function). Because the block can have no
655 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
657 LLVM allows an explicit section to be specified for functions. If the
658 target supports it, it will emit functions to the section specified.
660 An explicit alignment may be specified for a function. If not present,
661 or if the alignment is set to zero, the alignment of the function is set
662 by the target to whatever it feels convenient. If an explicit alignment
663 is specified, the function is forced to have at least that much
664 alignment. All alignments must be a power of 2.
666 If the ``unnamed_addr`` attribute is given, the address is know to not
667 be significant and two identical functions can be merged.
671 define [linkage] [visibility] [DLLStorageClass]
673 <ResultType> @<FunctionName> ([argument list])
674 [unnamed_addr] [fn Attrs] [section "name"] [align N]
675 [gc] [prefix Constant] { ... }
682 Aliases, unlike function or variables, don't create any new data. They
683 are just a new symbol and metadata for an existing position.
685 Aliases have a name and an aliasee that is either a global value or a
688 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
689 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
690 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
694 @<Name> = [Visibility] [DLLStorageClass] [ThreadLocal] alias [Linkage] <AliaseeTy> @<Aliasee>
696 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
697 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
698 might not correctly handle dropping a weak symbol that is aliased.
700 Alias that are not ``unnamed_addr`` are guaranteed to have the same address as
703 Since aliases are only a second name, some restrictions apply, of which
704 some can only be checked when producing an object file:
706 * The expression defining the aliasee must be computable at assembly
707 time. Since it is just a name, no relocations can be used.
709 * No alias in the expression can be weak as the possibility of the
710 intermediate alias being overridden cannot be represented in an
713 * No global value in the expression can be a declaration, since that
714 would require a relocation, which is not possible.
716 .. _namedmetadatastructure:
721 Named metadata is a collection of metadata. :ref:`Metadata
722 nodes <metadata>` (but not metadata strings) are the only valid
723 operands for a named metadata.
727 ; Some unnamed metadata nodes, which are referenced by the named metadata.
728 !0 = metadata !{metadata !"zero"}
729 !1 = metadata !{metadata !"one"}
730 !2 = metadata !{metadata !"two"}
732 !name = !{!0, !1, !2}
739 The return type and each parameter of a function type may have a set of
740 *parameter attributes* associated with them. Parameter attributes are
741 used to communicate additional information about the result or
742 parameters of a function. Parameter attributes are considered to be part
743 of the function, not of the function type, so functions with different
744 parameter attributes can have the same function type.
746 Parameter attributes are simple keywords that follow the type specified.
747 If multiple parameter attributes are needed, they are space separated.
752 declare i32 @printf(i8* noalias nocapture, ...)
753 declare i32 @atoi(i8 zeroext)
754 declare signext i8 @returns_signed_char()
756 Note that any attributes for the function result (``nounwind``,
757 ``readonly``) come immediately after the argument list.
759 Currently, only the following parameter attributes are defined:
762 This indicates to the code generator that the parameter or return
763 value should be zero-extended to the extent required by the target's
764 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
765 the caller (for a parameter) or the callee (for a return value).
767 This indicates to the code generator that the parameter or return
768 value should be sign-extended to the extent required by the target's
769 ABI (which is usually 32-bits) by the caller (for a parameter) or
770 the callee (for a return value).
772 This indicates that this parameter or return value should be treated
773 in a special target-dependent fashion during while emitting code for
774 a function call or return (usually, by putting it in a register as
775 opposed to memory, though some targets use it to distinguish between
776 two different kinds of registers). Use of this attribute is
779 This indicates that the pointer parameter should really be passed by
780 value to the function. The attribute implies that a hidden copy of
781 the pointee is made between the caller and the callee, so the callee
782 is unable to modify the value in the caller. This attribute is only
783 valid on LLVM pointer arguments. It is generally used to pass
784 structs and arrays by value, but is also valid on pointers to
785 scalars. The copy is considered to belong to the caller not the
786 callee (for example, ``readonly`` functions should not write to
787 ``byval`` parameters). This is not a valid attribute for return
790 The byval attribute also supports specifying an alignment with the
791 align attribute. It indicates the alignment of the stack slot to
792 form and the known alignment of the pointer specified to the call
793 site. If the alignment is not specified, then the code generator
794 makes a target-specific assumption.
800 The ``inalloca`` argument attribute allows the caller to take the
801 address of outgoing stack arguments. An ``inalloca`` argument must
802 be a pointer to stack memory produced by an ``alloca`` instruction.
803 The alloca, or argument allocation, must also be tagged with the
804 inalloca keyword. Only the past argument may have the ``inalloca``
805 attribute, and that argument is guaranteed to be passed in memory.
807 An argument allocation may be used by a call at most once because
808 the call may deallocate it. The ``inalloca`` attribute cannot be
809 used in conjunction with other attributes that affect argument
810 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
811 ``inalloca`` attribute also disables LLVM's implicit lowering of
812 large aggregate return values, which means that frontend authors
813 must lower them with ``sret`` pointers.
815 When the call site is reached, the argument allocation must have
816 been the most recent stack allocation that is still live, or the
817 results are undefined. It is possible to allocate additional stack
818 space after an argument allocation and before its call site, but it
819 must be cleared off with :ref:`llvm.stackrestore
822 See :doc:`InAlloca` for more information on how to use this
826 This indicates that the pointer parameter specifies the address of a
827 structure that is the return value of the function in the source
828 program. This pointer must be guaranteed by the caller to be valid:
829 loads and stores to the structure may be assumed by the callee
830 not to trap and to be properly aligned. This may only be applied to
831 the first parameter. This is not a valid attribute for return
837 This indicates that pointer values :ref:`based <pointeraliasing>` on
838 the argument or return value do not alias pointer values which are
839 not *based* on it, ignoring certain "irrelevant" dependencies. For a
840 call to the parent function, dependencies between memory references
841 from before or after the call and from those during the call are
842 "irrelevant" to the ``noalias`` keyword for the arguments and return
843 value used in that call. The caller shares the responsibility with
844 the callee for ensuring that these requirements are met. For further
845 details, please see the discussion of the NoAlias response in :ref:`alias
846 analysis <Must, May, or No>`.
848 Note that this definition of ``noalias`` is intentionally similar
849 to the definition of ``restrict`` in C99 for function arguments,
850 though it is slightly weaker.
852 For function return values, C99's ``restrict`` is not meaningful,
853 while LLVM's ``noalias`` is.
855 This indicates that the callee does not make any copies of the
856 pointer that outlive the callee itself. This is not a valid
857 attribute for return values.
862 This indicates that the pointer parameter can be excised using the
863 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
864 attribute for return values and can only be applied to one parameter.
867 This indicates that the function always returns the argument as its return
868 value. This is an optimization hint to the code generator when generating
869 the caller, allowing tail call optimization and omission of register saves
870 and restores in some cases; it is not checked or enforced when generating
871 the callee. The parameter and the function return type must be valid
872 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
873 valid attribute for return values and can only be applied to one parameter.
876 This indicates that the parameter or return pointer is not null. This
877 attribute may only be applied to pointer typed parameters. This is not
878 checked or enforced by LLVM, the caller must ensure that the pointer
879 passed in is non-null, or the callee must ensure that the returned pointer
884 Garbage Collector Names
885 -----------------------
887 Each function may specify a garbage collector name, which is simply a
892 define void @f() gc "name" { ... }
894 The compiler declares the supported values of *name*. Specifying a
895 collector which will cause the compiler to alter its output in order to
896 support the named garbage collection algorithm.
903 Prefix data is data associated with a function which the code generator
904 will emit immediately before the function body. The purpose of this feature
905 is to allow frontends to associate language-specific runtime metadata with
906 specific functions and make it available through the function pointer while
907 still allowing the function pointer to be called. To access the data for a
908 given function, a program may bitcast the function pointer to a pointer to
909 the constant's type. This implies that the IR symbol points to the start
912 To maintain the semantics of ordinary function calls, the prefix data must
913 have a particular format. Specifically, it must begin with a sequence of
914 bytes which decode to a sequence of machine instructions, valid for the
915 module's target, which transfer control to the point immediately succeeding
916 the prefix data, without performing any other visible action. This allows
917 the inliner and other passes to reason about the semantics of the function
918 definition without needing to reason about the prefix data. Obviously this
919 makes the format of the prefix data highly target dependent.
921 Prefix data is laid out as if it were an initializer for a global variable
922 of the prefix data's type. No padding is automatically placed between the
923 prefix data and the function body. If padding is required, it must be part
926 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
927 which encodes the ``nop`` instruction:
931 define void @f() prefix i8 144 { ... }
933 Generally prefix data can be formed by encoding a relative branch instruction
934 which skips the metadata, as in this example of valid prefix data for the
935 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
939 %0 = type <{ i8, i8, i8* }>
941 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
943 A function may have prefix data but no body. This has similar semantics
944 to the ``available_externally`` linkage in that the data may be used by the
945 optimizers but will not be emitted in the object file.
952 Attribute groups are groups of attributes that are referenced by objects within
953 the IR. They are important for keeping ``.ll`` files readable, because a lot of
954 functions will use the same set of attributes. In the degenerative case of a
955 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
956 group will capture the important command line flags used to build that file.
958 An attribute group is a module-level object. To use an attribute group, an
959 object references the attribute group's ID (e.g. ``#37``). An object may refer
960 to more than one attribute group. In that situation, the attributes from the
961 different groups are merged.
963 Here is an example of attribute groups for a function that should always be
964 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
968 ; Target-independent attributes:
969 attributes #0 = { alwaysinline alignstack=4 }
971 ; Target-dependent attributes:
972 attributes #1 = { "no-sse" }
974 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
975 define void @f() #0 #1 { ... }
982 Function attributes are set to communicate additional information about
983 a function. Function attributes are considered to be part of the
984 function, not of the function type, so functions with different function
985 attributes can have the same function type.
987 Function attributes are simple keywords that follow the type specified.
988 If multiple attributes are needed, they are space separated. For
993 define void @f() noinline { ... }
994 define void @f() alwaysinline { ... }
995 define void @f() alwaysinline optsize { ... }
996 define void @f() optsize { ... }
999 This attribute indicates that, when emitting the prologue and
1000 epilogue, the backend should forcibly align the stack pointer.
1001 Specify the desired alignment, which must be a power of two, in
1004 This attribute indicates that the inliner should attempt to inline
1005 this function into callers whenever possible, ignoring any active
1006 inlining size threshold for this caller.
1008 This indicates that the callee function at a call site should be
1009 recognized as a built-in function, even though the function's declaration
1010 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1011 direct calls to functions which are declared with the ``nobuiltin``
1014 This attribute indicates that this function is rarely called. When
1015 computing edge weights, basic blocks post-dominated by a cold
1016 function call are also considered to be cold; and, thus, given low
1019 This attribute indicates that the source code contained a hint that
1020 inlining this function is desirable (such as the "inline" keyword in
1021 C/C++). It is just a hint; it imposes no requirements on the
1024 This attribute suggests that optimization passes and code generator
1025 passes make choices that keep the code size of this function as small
1026 as possible and perform optimizations that may sacrifice runtime
1027 performance in order to minimize the size of the generated code.
1029 This attribute disables prologue / epilogue emission for the
1030 function. This can have very system-specific consequences.
1032 This indicates that the callee function at a call site is not recognized as
1033 a built-in function. LLVM will retain the original call and not replace it
1034 with equivalent code based on the semantics of the built-in function, unless
1035 the call site uses the ``builtin`` attribute. This is valid at call sites
1036 and on function declarations and definitions.
1038 This attribute indicates that calls to the function cannot be
1039 duplicated. A call to a ``noduplicate`` function may be moved
1040 within its parent function, but may not be duplicated within
1041 its parent function.
1043 A function containing a ``noduplicate`` call may still
1044 be an inlining candidate, provided that the call is not
1045 duplicated by inlining. That implies that the function has
1046 internal linkage and only has one call site, so the original
1047 call is dead after inlining.
1049 This attributes disables implicit floating point instructions.
1051 This attribute indicates that the inliner should never inline this
1052 function in any situation. This attribute may not be used together
1053 with the ``alwaysinline`` attribute.
1055 This attribute suppresses lazy symbol binding for the function. This
1056 may make calls to the function faster, at the cost of extra program
1057 startup time if the function is not called during program startup.
1059 This attribute indicates that the code generator should not use a
1060 red zone, even if the target-specific ABI normally permits it.
1062 This function attribute indicates that the function never returns
1063 normally. This produces undefined behavior at runtime if the
1064 function ever does dynamically return.
1066 This function attribute indicates that the function never returns
1067 with an unwind or exceptional control flow. If the function does
1068 unwind, its runtime behavior is undefined.
1070 This function attribute indicates that the function is not optimized
1071 by any optimization or code generator passes with the
1072 exception of interprocedural optimization passes.
1073 This attribute cannot be used together with the ``alwaysinline``
1074 attribute; this attribute is also incompatible
1075 with the ``minsize`` attribute and the ``optsize`` attribute.
1077 This attribute requires the ``noinline`` attribute to be specified on
1078 the function as well, so the function is never inlined into any caller.
1079 Only functions with the ``alwaysinline`` attribute are valid
1080 candidates for inlining into the body of this function.
1082 This attribute suggests that optimization passes and code generator
1083 passes make choices that keep the code size of this function low,
1084 and otherwise do optimizations specifically to reduce code size as
1085 long as they do not significantly impact runtime performance.
1087 On a function, this attribute indicates that the function computes its
1088 result (or decides to unwind an exception) based strictly on its arguments,
1089 without dereferencing any pointer arguments or otherwise accessing
1090 any mutable state (e.g. memory, control registers, etc) visible to
1091 caller functions. It does not write through any pointer arguments
1092 (including ``byval`` arguments) and never changes any state visible
1093 to callers. This means that it cannot unwind exceptions by calling
1094 the ``C++`` exception throwing methods.
1096 On an argument, this attribute indicates that the function does not
1097 dereference that pointer argument, even though it may read or write the
1098 memory that the pointer points to if accessed through other pointers.
1100 On a function, this attribute indicates that the function does not write
1101 through any pointer arguments (including ``byval`` arguments) or otherwise
1102 modify any state (e.g. memory, control registers, etc) visible to
1103 caller functions. It may dereference pointer arguments and read
1104 state that may be set in the caller. A readonly function always
1105 returns the same value (or unwinds an exception identically) when
1106 called with the same set of arguments and global state. It cannot
1107 unwind an exception by calling the ``C++`` exception throwing
1110 On an argument, this attribute indicates that the function does not write
1111 through this pointer argument, even though it may write to the memory that
1112 the pointer points to.
1114 This attribute indicates that this function can return twice. The C
1115 ``setjmp`` is an example of such a function. The compiler disables
1116 some optimizations (like tail calls) in the caller of these
1118 ``sanitize_address``
1119 This attribute indicates that AddressSanitizer checks
1120 (dynamic address safety analysis) are enabled for this function.
1122 This attribute indicates that MemorySanitizer checks (dynamic detection
1123 of accesses to uninitialized memory) are enabled for this function.
1125 This attribute indicates that ThreadSanitizer checks
1126 (dynamic thread safety analysis) are enabled for this function.
1128 This attribute indicates that the function should emit a stack
1129 smashing protector. It is in the form of a "canary" --- a random value
1130 placed on the stack before the local variables that's checked upon
1131 return from the function to see if it has been overwritten. A
1132 heuristic is used to determine if a function needs stack protectors
1133 or not. The heuristic used will enable protectors for functions with:
1135 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1136 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1137 - Calls to alloca() with variable sizes or constant sizes greater than
1138 ``ssp-buffer-size``.
1140 Variables that are identified as requiring a protector will be arranged
1141 on the stack such that they are adjacent to the stack protector guard.
1143 If a function that has an ``ssp`` attribute is inlined into a
1144 function that doesn't have an ``ssp`` attribute, then the resulting
1145 function will have an ``ssp`` attribute.
1147 This attribute indicates that the function should *always* emit a
1148 stack smashing protector. This overrides the ``ssp`` function
1151 Variables that are identified as requiring a protector will be arranged
1152 on the stack such that they are adjacent to the stack protector guard.
1153 The specific layout rules are:
1155 #. Large arrays and structures containing large arrays
1156 (``>= ssp-buffer-size``) are closest to the stack protector.
1157 #. Small arrays and structures containing small arrays
1158 (``< ssp-buffer-size``) are 2nd closest to the protector.
1159 #. Variables that have had their address taken are 3rd closest to the
1162 If a function that has an ``sspreq`` attribute is inlined into a
1163 function that doesn't have an ``sspreq`` attribute or which has an
1164 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1165 an ``sspreq`` attribute.
1167 This attribute indicates that the function should emit a stack smashing
1168 protector. This attribute causes a strong heuristic to be used when
1169 determining if a function needs stack protectors. The strong heuristic
1170 will enable protectors for functions with:
1172 - Arrays of any size and type
1173 - Aggregates containing an array of any size and type.
1174 - Calls to alloca().
1175 - Local variables that have had their address taken.
1177 Variables that are identified as requiring a protector will be arranged
1178 on the stack such that they are adjacent to the stack protector guard.
1179 The specific layout rules are:
1181 #. Large arrays and structures containing large arrays
1182 (``>= ssp-buffer-size``) are closest to the stack protector.
1183 #. Small arrays and structures containing small arrays
1184 (``< ssp-buffer-size``) are 2nd closest to the protector.
1185 #. Variables that have had their address taken are 3rd closest to the
1188 This overrides the ``ssp`` function attribute.
1190 If a function that has an ``sspstrong`` attribute is inlined into a
1191 function that doesn't have an ``sspstrong`` attribute, then the
1192 resulting function will have an ``sspstrong`` attribute.
1194 This attribute indicates that the ABI being targeted requires that
1195 an unwind table entry be produce for this function even if we can
1196 show that no exceptions passes by it. This is normally the case for
1197 the ELF x86-64 abi, but it can be disabled for some compilation
1202 Module-Level Inline Assembly
1203 ----------------------------
1205 Modules may contain "module-level inline asm" blocks, which corresponds
1206 to the GCC "file scope inline asm" blocks. These blocks are internally
1207 concatenated by LLVM and treated as a single unit, but may be separated
1208 in the ``.ll`` file if desired. The syntax is very simple:
1210 .. code-block:: llvm
1212 module asm "inline asm code goes here"
1213 module asm "more can go here"
1215 The strings can contain any character by escaping non-printable
1216 characters. The escape sequence used is simply "\\xx" where "xx" is the
1217 two digit hex code for the number.
1219 The inline asm code is simply printed to the machine code .s file when
1220 assembly code is generated.
1222 .. _langref_datalayout:
1227 A module may specify a target specific data layout string that specifies
1228 how data is to be laid out in memory. The syntax for the data layout is
1231 .. code-block:: llvm
1233 target datalayout = "layout specification"
1235 The *layout specification* consists of a list of specifications
1236 separated by the minus sign character ('-'). Each specification starts
1237 with a letter and may include other information after the letter to
1238 define some aspect of the data layout. The specifications accepted are
1242 Specifies that the target lays out data in big-endian form. That is,
1243 the bits with the most significance have the lowest address
1246 Specifies that the target lays out data in little-endian form. That
1247 is, the bits with the least significance have the lowest address
1250 Specifies the natural alignment of the stack in bits. Alignment
1251 promotion of stack variables is limited to the natural stack
1252 alignment to avoid dynamic stack realignment. The stack alignment
1253 must be a multiple of 8-bits. If omitted, the natural stack
1254 alignment defaults to "unspecified", which does not prevent any
1255 alignment promotions.
1256 ``p[n]:<size>:<abi>:<pref>``
1257 This specifies the *size* of a pointer and its ``<abi>`` and
1258 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1259 bits. The address space, ``n`` is optional, and if not specified,
1260 denotes the default address space 0. The value of ``n`` must be
1261 in the range [1,2^23).
1262 ``i<size>:<abi>:<pref>``
1263 This specifies the alignment for an integer type of a given bit
1264 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1265 ``v<size>:<abi>:<pref>``
1266 This specifies the alignment for a vector type of a given bit
1268 ``f<size>:<abi>:<pref>``
1269 This specifies the alignment for a floating point type of a given bit
1270 ``<size>``. Only values of ``<size>`` that are supported by the target
1271 will work. 32 (float) and 64 (double) are supported on all targets; 80
1272 or 128 (different flavors of long double) are also supported on some
1275 This specifies the alignment for an object of aggregate type.
1277 If present, specifies that llvm names are mangled in the output. The
1280 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1281 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1282 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1283 symbols get a ``_`` prefix.
1284 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1285 functions also get a suffix based on the frame size.
1286 ``n<size1>:<size2>:<size3>...``
1287 This specifies a set of native integer widths for the target CPU in
1288 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1289 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1290 this set are considered to support most general arithmetic operations
1293 On every specification that takes a ``<abi>:<pref>``, specifying the
1294 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1295 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1297 When constructing the data layout for a given target, LLVM starts with a
1298 default set of specifications which are then (possibly) overridden by
1299 the specifications in the ``datalayout`` keyword. The default
1300 specifications are given in this list:
1302 - ``E`` - big endian
1303 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1304 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1305 same as the default address space.
1306 - ``S0`` - natural stack alignment is unspecified
1307 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1308 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1309 - ``i16:16:16`` - i16 is 16-bit aligned
1310 - ``i32:32:32`` - i32 is 32-bit aligned
1311 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1312 alignment of 64-bits
1313 - ``f16:16:16`` - half is 16-bit aligned
1314 - ``f32:32:32`` - float is 32-bit aligned
1315 - ``f64:64:64`` - double is 64-bit aligned
1316 - ``f128:128:128`` - quad is 128-bit aligned
1317 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1318 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1319 - ``a:0:64`` - aggregates are 64-bit aligned
1321 When LLVM is determining the alignment for a given type, it uses the
1324 #. If the type sought is an exact match for one of the specifications,
1325 that specification is used.
1326 #. If no match is found, and the type sought is an integer type, then
1327 the smallest integer type that is larger than the bitwidth of the
1328 sought type is used. If none of the specifications are larger than
1329 the bitwidth then the largest integer type is used. For example,
1330 given the default specifications above, the i7 type will use the
1331 alignment of i8 (next largest) while both i65 and i256 will use the
1332 alignment of i64 (largest specified).
1333 #. If no match is found, and the type sought is a vector type, then the
1334 largest vector type that is smaller than the sought vector type will
1335 be used as a fall back. This happens because <128 x double> can be
1336 implemented in terms of 64 <2 x double>, for example.
1338 The function of the data layout string may not be what you expect.
1339 Notably, this is not a specification from the frontend of what alignment
1340 the code generator should use.
1342 Instead, if specified, the target data layout is required to match what
1343 the ultimate *code generator* expects. This string is used by the
1344 mid-level optimizers to improve code, and this only works if it matches
1345 what the ultimate code generator uses. If you would like to generate IR
1346 that does not embed this target-specific detail into the IR, then you
1347 don't have to specify the string. This will disable some optimizations
1348 that require precise layout information, but this also prevents those
1349 optimizations from introducing target specificity into the IR.
1356 A module may specify a target triple string that describes the target
1357 host. The syntax for the target triple is simply:
1359 .. code-block:: llvm
1361 target triple = "x86_64-apple-macosx10.7.0"
1363 The *target triple* string consists of a series of identifiers delimited
1364 by the minus sign character ('-'). The canonical forms are:
1368 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1369 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1371 This information is passed along to the backend so that it generates
1372 code for the proper architecture. It's possible to override this on the
1373 command line with the ``-mtriple`` command line option.
1375 .. _pointeraliasing:
1377 Pointer Aliasing Rules
1378 ----------------------
1380 Any memory access must be done through a pointer value associated with
1381 an address range of the memory access, otherwise the behavior is
1382 undefined. Pointer values are associated with address ranges according
1383 to the following rules:
1385 - A pointer value is associated with the addresses associated with any
1386 value it is *based* on.
1387 - An address of a global variable is associated with the address range
1388 of the variable's storage.
1389 - The result value of an allocation instruction is associated with the
1390 address range of the allocated storage.
1391 - A null pointer in the default address-space is associated with no
1393 - An integer constant other than zero or a pointer value returned from
1394 a function not defined within LLVM may be associated with address
1395 ranges allocated through mechanisms other than those provided by
1396 LLVM. Such ranges shall not overlap with any ranges of addresses
1397 allocated by mechanisms provided by LLVM.
1399 A pointer value is *based* on another pointer value according to the
1402 - A pointer value formed from a ``getelementptr`` operation is *based*
1403 on the first operand of the ``getelementptr``.
1404 - The result value of a ``bitcast`` is *based* on the operand of the
1406 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1407 values that contribute (directly or indirectly) to the computation of
1408 the pointer's value.
1409 - The "*based* on" relationship is transitive.
1411 Note that this definition of *"based"* is intentionally similar to the
1412 definition of *"based"* in C99, though it is slightly weaker.
1414 LLVM IR does not associate types with memory. The result type of a
1415 ``load`` merely indicates the size and alignment of the memory from
1416 which to load, as well as the interpretation of the value. The first
1417 operand type of a ``store`` similarly only indicates the size and
1418 alignment of the store.
1420 Consequently, type-based alias analysis, aka TBAA, aka
1421 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1422 :ref:`Metadata <metadata>` may be used to encode additional information
1423 which specialized optimization passes may use to implement type-based
1428 Volatile Memory Accesses
1429 ------------------------
1431 Certain memory accesses, such as :ref:`load <i_load>`'s,
1432 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1433 marked ``volatile``. The optimizers must not change the number of
1434 volatile operations or change their order of execution relative to other
1435 volatile operations. The optimizers *may* change the order of volatile
1436 operations relative to non-volatile operations. This is not Java's
1437 "volatile" and has no cross-thread synchronization behavior.
1439 IR-level volatile loads and stores cannot safely be optimized into
1440 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1441 flagged volatile. Likewise, the backend should never split or merge
1442 target-legal volatile load/store instructions.
1444 .. admonition:: Rationale
1446 Platforms may rely on volatile loads and stores of natively supported
1447 data width to be executed as single instruction. For example, in C
1448 this holds for an l-value of volatile primitive type with native
1449 hardware support, but not necessarily for aggregate types. The
1450 frontend upholds these expectations, which are intentionally
1451 unspecified in the IR. The rules above ensure that IR transformation
1452 do not violate the frontend's contract with the language.
1456 Memory Model for Concurrent Operations
1457 --------------------------------------
1459 The LLVM IR does not define any way to start parallel threads of
1460 execution or to register signal handlers. Nonetheless, there are
1461 platform-specific ways to create them, and we define LLVM IR's behavior
1462 in their presence. This model is inspired by the C++0x memory model.
1464 For a more informal introduction to this model, see the :doc:`Atomics`.
1466 We define a *happens-before* partial order as the least partial order
1469 - Is a superset of single-thread program order, and
1470 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1471 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1472 techniques, like pthread locks, thread creation, thread joining,
1473 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1474 Constraints <ordering>`).
1476 Note that program order does not introduce *happens-before* edges
1477 between a thread and signals executing inside that thread.
1479 Every (defined) read operation (load instructions, memcpy, atomic
1480 loads/read-modify-writes, etc.) R reads a series of bytes written by
1481 (defined) write operations (store instructions, atomic
1482 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1483 section, initialized globals are considered to have a write of the
1484 initializer which is atomic and happens before any other read or write
1485 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1486 may see any write to the same byte, except:
1488 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1489 write\ :sub:`2` happens before R\ :sub:`byte`, then
1490 R\ :sub:`byte` does not see write\ :sub:`1`.
1491 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1492 R\ :sub:`byte` does not see write\ :sub:`3`.
1494 Given that definition, R\ :sub:`byte` is defined as follows:
1496 - If R is volatile, the result is target-dependent. (Volatile is
1497 supposed to give guarantees which can support ``sig_atomic_t`` in
1498 C/C++, and may be used for accesses to addresses which do not behave
1499 like normal memory. It does not generally provide cross-thread
1501 - Otherwise, if there is no write to the same byte that happens before
1502 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1503 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1504 R\ :sub:`byte` returns the value written by that write.
1505 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1506 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1507 Memory Ordering Constraints <ordering>` section for additional
1508 constraints on how the choice is made.
1509 - Otherwise R\ :sub:`byte` returns ``undef``.
1511 R returns the value composed of the series of bytes it read. This
1512 implies that some bytes within the value may be ``undef`` **without**
1513 the entire value being ``undef``. Note that this only defines the
1514 semantics of the operation; it doesn't mean that targets will emit more
1515 than one instruction to read the series of bytes.
1517 Note that in cases where none of the atomic intrinsics are used, this
1518 model places only one restriction on IR transformations on top of what
1519 is required for single-threaded execution: introducing a store to a byte
1520 which might not otherwise be stored is not allowed in general.
1521 (Specifically, in the case where another thread might write to and read
1522 from an address, introducing a store can change a load that may see
1523 exactly one write into a load that may see multiple writes.)
1527 Atomic Memory Ordering Constraints
1528 ----------------------------------
1530 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1531 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1532 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1533 ordering parameters that determine which other atomic instructions on
1534 the same address they *synchronize with*. These semantics are borrowed
1535 from Java and C++0x, but are somewhat more colloquial. If these
1536 descriptions aren't precise enough, check those specs (see spec
1537 references in the :doc:`atomics guide <Atomics>`).
1538 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1539 differently since they don't take an address. See that instruction's
1540 documentation for details.
1542 For a simpler introduction to the ordering constraints, see the
1546 The set of values that can be read is governed by the happens-before
1547 partial order. A value cannot be read unless some operation wrote
1548 it. This is intended to provide a guarantee strong enough to model
1549 Java's non-volatile shared variables. This ordering cannot be
1550 specified for read-modify-write operations; it is not strong enough
1551 to make them atomic in any interesting way.
1553 In addition to the guarantees of ``unordered``, there is a single
1554 total order for modifications by ``monotonic`` operations on each
1555 address. All modification orders must be compatible with the
1556 happens-before order. There is no guarantee that the modification
1557 orders can be combined to a global total order for the whole program
1558 (and this often will not be possible). The read in an atomic
1559 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1560 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1561 order immediately before the value it writes. If one atomic read
1562 happens before another atomic read of the same address, the later
1563 read must see the same value or a later value in the address's
1564 modification order. This disallows reordering of ``monotonic`` (or
1565 stronger) operations on the same address. If an address is written
1566 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1567 read that address repeatedly, the other threads must eventually see
1568 the write. This corresponds to the C++0x/C1x
1569 ``memory_order_relaxed``.
1571 In addition to the guarantees of ``monotonic``, a
1572 *synchronizes-with* edge may be formed with a ``release`` operation.
1573 This is intended to model C++'s ``memory_order_acquire``.
1575 In addition to the guarantees of ``monotonic``, if this operation
1576 writes a value which is subsequently read by an ``acquire``
1577 operation, it *synchronizes-with* that operation. (This isn't a
1578 complete description; see the C++0x definition of a release
1579 sequence.) This corresponds to the C++0x/C1x
1580 ``memory_order_release``.
1581 ``acq_rel`` (acquire+release)
1582 Acts as both an ``acquire`` and ``release`` operation on its
1583 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1584 ``seq_cst`` (sequentially consistent)
1585 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1586 operation which only reads, ``release`` for an operation which only
1587 writes), there is a global total order on all
1588 sequentially-consistent operations on all addresses, which is
1589 consistent with the *happens-before* partial order and with the
1590 modification orders of all the affected addresses. Each
1591 sequentially-consistent read sees the last preceding write to the
1592 same address in this global order. This corresponds to the C++0x/C1x
1593 ``memory_order_seq_cst`` and Java volatile.
1597 If an atomic operation is marked ``singlethread``, it only *synchronizes
1598 with* or participates in modification and seq\_cst total orderings with
1599 other operations running in the same thread (for example, in signal
1607 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1608 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1609 :ref:`frem <i_frem>`) have the following flags that can set to enable
1610 otherwise unsafe floating point operations
1613 No NaNs - Allow optimizations to assume the arguments and result are not
1614 NaN. Such optimizations are required to retain defined behavior over
1615 NaNs, but the value of the result is undefined.
1618 No Infs - Allow optimizations to assume the arguments and result are not
1619 +/-Inf. Such optimizations are required to retain defined behavior over
1620 +/-Inf, but the value of the result is undefined.
1623 No Signed Zeros - Allow optimizations to treat the sign of a zero
1624 argument or result as insignificant.
1627 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1628 argument rather than perform division.
1631 Fast - Allow algebraically equivalent transformations that may
1632 dramatically change results in floating point (e.g. reassociate). This
1633 flag implies all the others.
1640 The LLVM type system is one of the most important features of the
1641 intermediate representation. Being typed enables a number of
1642 optimizations to be performed on the intermediate representation
1643 directly, without having to do extra analyses on the side before the
1644 transformation. A strong type system makes it easier to read the
1645 generated code and enables novel analyses and transformations that are
1646 not feasible to perform on normal three address code representations.
1656 The void type does not represent any value and has no size.
1674 The function type can be thought of as a function signature. It consists of a
1675 return type and a list of formal parameter types. The return type of a function
1676 type is a void type or first class type --- except for :ref:`label <t_label>`
1677 and :ref:`metadata <t_metadata>` types.
1683 <returntype> (<parameter list>)
1685 ...where '``<parameter list>``' is a comma-separated list of type
1686 specifiers. Optionally, the parameter list may include a type ``...``, which
1687 indicates that the function takes a variable number of arguments. Variable
1688 argument functions can access their arguments with the :ref:`variable argument
1689 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1690 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1694 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1695 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1696 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1697 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1698 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1699 | ``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. |
1700 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1701 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1702 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1709 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1710 Values of these types are the only ones which can be produced by
1718 These are the types that are valid in registers from CodeGen's perspective.
1727 The integer type is a very simple type that simply specifies an
1728 arbitrary bit width for the integer type desired. Any bit width from 1
1729 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1737 The number of bits the integer will occupy is specified by the ``N``
1743 +----------------+------------------------------------------------+
1744 | ``i1`` | a single-bit integer. |
1745 +----------------+------------------------------------------------+
1746 | ``i32`` | a 32-bit integer. |
1747 +----------------+------------------------------------------------+
1748 | ``i1942652`` | a really big integer of over 1 million bits. |
1749 +----------------+------------------------------------------------+
1753 Floating Point Types
1754 """"""""""""""""""""
1763 - 16-bit floating point value
1766 - 32-bit floating point value
1769 - 64-bit floating point value
1772 - 128-bit floating point value (112-bit mantissa)
1775 - 80-bit floating point value (X87)
1778 - 128-bit floating point value (two 64-bits)
1785 The x86_mmx type represents a value held in an MMX register on an x86
1786 machine. The operations allowed on it are quite limited: parameters and
1787 return values, load and store, and bitcast. User-specified MMX
1788 instructions are represented as intrinsic or asm calls with arguments
1789 and/or results of this type. There are no arrays, vectors or constants
1806 The pointer type is used to specify memory locations. Pointers are
1807 commonly used to reference objects in memory.
1809 Pointer types may have an optional address space attribute defining the
1810 numbered address space where the pointed-to object resides. The default
1811 address space is number zero. The semantics of non-zero address spaces
1812 are target-specific.
1814 Note that LLVM does not permit pointers to void (``void*``) nor does it
1815 permit pointers to labels (``label*``). Use ``i8*`` instead.
1825 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1826 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1827 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1828 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1829 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1830 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1831 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1840 A vector type is a simple derived type that represents a vector of
1841 elements. Vector types are used when multiple primitive data are
1842 operated in parallel using a single instruction (SIMD). A vector type
1843 requires a size (number of elements) and an underlying primitive data
1844 type. Vector types are considered :ref:`first class <t_firstclass>`.
1850 < <# elements> x <elementtype> >
1852 The number of elements is a constant integer value larger than 0;
1853 elementtype may be any integer or floating point type, or a pointer to
1854 these types. Vectors of size zero are not allowed.
1858 +-------------------+--------------------------------------------------+
1859 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1860 +-------------------+--------------------------------------------------+
1861 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1862 +-------------------+--------------------------------------------------+
1863 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1864 +-------------------+--------------------------------------------------+
1865 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1866 +-------------------+--------------------------------------------------+
1875 The label type represents code labels.
1890 The metadata type represents embedded metadata. No derived types may be
1891 created from metadata except for :ref:`function <t_function>` arguments.
1904 Aggregate Types are a subset of derived types that can contain multiple
1905 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1906 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1916 The array type is a very simple derived type that arranges elements
1917 sequentially in memory. The array type requires a size (number of
1918 elements) and an underlying data type.
1924 [<# elements> x <elementtype>]
1926 The number of elements is a constant integer value; ``elementtype`` may
1927 be any type with a size.
1931 +------------------+--------------------------------------+
1932 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1933 +------------------+--------------------------------------+
1934 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1935 +------------------+--------------------------------------+
1936 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1937 +------------------+--------------------------------------+
1939 Here are some examples of multidimensional arrays:
1941 +-----------------------------+----------------------------------------------------------+
1942 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1943 +-----------------------------+----------------------------------------------------------+
1944 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1945 +-----------------------------+----------------------------------------------------------+
1946 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1947 +-----------------------------+----------------------------------------------------------+
1949 There is no restriction on indexing beyond the end of the array implied
1950 by a static type (though there are restrictions on indexing beyond the
1951 bounds of an allocated object in some cases). This means that
1952 single-dimension 'variable sized array' addressing can be implemented in
1953 LLVM with a zero length array type. An implementation of 'pascal style
1954 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1964 The structure type is used to represent a collection of data members
1965 together in memory. The elements of a structure may be any type that has
1968 Structures in memory are accessed using '``load``' and '``store``' by
1969 getting a pointer to a field with the '``getelementptr``' instruction.
1970 Structures in registers are accessed using the '``extractvalue``' and
1971 '``insertvalue``' instructions.
1973 Structures may optionally be "packed" structures, which indicate that
1974 the alignment of the struct is one byte, and that there is no padding
1975 between the elements. In non-packed structs, padding between field types
1976 is inserted as defined by the DataLayout string in the module, which is
1977 required to match what the underlying code generator expects.
1979 Structures can either be "literal" or "identified". A literal structure
1980 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1981 identified types are always defined at the top level with a name.
1982 Literal types are uniqued by their contents and can never be recursive
1983 or opaque since there is no way to write one. Identified types can be
1984 recursive, can be opaqued, and are never uniqued.
1990 %T1 = type { <type list> } ; Identified normal struct type
1991 %T2 = type <{ <type list> }> ; Identified packed struct type
1995 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1996 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1997 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1998 | ``{ 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``. |
1999 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2000 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2001 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2005 Opaque Structure Types
2006 """"""""""""""""""""""
2010 Opaque structure types are used to represent named structure types that
2011 do not have a body specified. This corresponds (for example) to the C
2012 notion of a forward declared structure.
2023 +--------------+-------------------+
2024 | ``opaque`` | An opaque type. |
2025 +--------------+-------------------+
2032 LLVM has several different basic types of constants. This section
2033 describes them all and their syntax.
2038 **Boolean constants**
2039 The two strings '``true``' and '``false``' are both valid constants
2041 **Integer constants**
2042 Standard integers (such as '4') are constants of the
2043 :ref:`integer <t_integer>` type. Negative numbers may be used with
2045 **Floating point constants**
2046 Floating point constants use standard decimal notation (e.g.
2047 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2048 hexadecimal notation (see below). The assembler requires the exact
2049 decimal value of a floating-point constant. For example, the
2050 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2051 decimal in binary. Floating point constants must have a :ref:`floating
2052 point <t_floating>` type.
2053 **Null pointer constants**
2054 The identifier '``null``' is recognized as a null pointer constant
2055 and must be of :ref:`pointer type <t_pointer>`.
2057 The one non-intuitive notation for constants is the hexadecimal form of
2058 floating point constants. For example, the form
2059 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2060 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2061 constants are required (and the only time that they are generated by the
2062 disassembler) is when a floating point constant must be emitted but it
2063 cannot be represented as a decimal floating point number in a reasonable
2064 number of digits. For example, NaN's, infinities, and other special
2065 values are represented in their IEEE hexadecimal format so that assembly
2066 and disassembly do not cause any bits to change in the constants.
2068 When using the hexadecimal form, constants of types half, float, and
2069 double are represented using the 16-digit form shown above (which
2070 matches the IEEE754 representation for double); half and float values
2071 must, however, be exactly representable as IEEE 754 half and single
2072 precision, respectively. Hexadecimal format is always used for long
2073 double, and there are three forms of long double. The 80-bit format used
2074 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2075 128-bit format used by PowerPC (two adjacent doubles) is represented by
2076 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2077 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2078 will only work if they match the long double format on your target.
2079 The IEEE 16-bit format (half precision) is represented by ``0xH``
2080 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2081 (sign bit at the left).
2083 There are no constants of type x86_mmx.
2085 .. _complexconstants:
2090 Complex constants are a (potentially recursive) combination of simple
2091 constants and smaller complex constants.
2093 **Structure constants**
2094 Structure constants are represented with notation similar to
2095 structure type definitions (a comma separated list of elements,
2096 surrounded by braces (``{}``)). For example:
2097 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2098 "``@G = external global i32``". Structure constants must have
2099 :ref:`structure type <t_struct>`, and the number and types of elements
2100 must match those specified by the type.
2102 Array constants are represented with notation similar to array type
2103 definitions (a comma separated list of elements, surrounded by
2104 square brackets (``[]``)). For example:
2105 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2106 :ref:`array type <t_array>`, and the number and types of elements must
2107 match those specified by the type.
2108 **Vector constants**
2109 Vector constants are represented with notation similar to vector
2110 type definitions (a comma separated list of elements, surrounded by
2111 less-than/greater-than's (``<>``)). For example:
2112 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2113 must have :ref:`vector type <t_vector>`, and the number and types of
2114 elements must match those specified by the type.
2115 **Zero initialization**
2116 The string '``zeroinitializer``' can be used to zero initialize a
2117 value to zero of *any* type, including scalar and
2118 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2119 having to print large zero initializers (e.g. for large arrays) and
2120 is always exactly equivalent to using explicit zero initializers.
2122 A metadata node is a structure-like constant with :ref:`metadata
2123 type <t_metadata>`. For example:
2124 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2125 constants that are meant to be interpreted as part of the
2126 instruction stream, metadata is a place to attach additional
2127 information such as debug info.
2129 Global Variable and Function Addresses
2130 --------------------------------------
2132 The addresses of :ref:`global variables <globalvars>` and
2133 :ref:`functions <functionstructure>` are always implicitly valid
2134 (link-time) constants. These constants are explicitly referenced when
2135 the :ref:`identifier for the global <identifiers>` is used and always have
2136 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2139 .. code-block:: llvm
2143 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2150 The string '``undef``' can be used anywhere a constant is expected, and
2151 indicates that the user of the value may receive an unspecified
2152 bit-pattern. Undefined values may be of any type (other than '``label``'
2153 or '``void``') and be used anywhere a constant is permitted.
2155 Undefined values are useful because they indicate to the compiler that
2156 the program is well defined no matter what value is used. This gives the
2157 compiler more freedom to optimize. Here are some examples of
2158 (potentially surprising) transformations that are valid (in pseudo IR):
2160 .. code-block:: llvm
2170 This is safe because all of the output bits are affected by the undef
2171 bits. Any output bit can have a zero or one depending on the input bits.
2173 .. code-block:: llvm
2184 These logical operations have bits that are not always affected by the
2185 input. For example, if ``%X`` has a zero bit, then the output of the
2186 '``and``' operation will always be a zero for that bit, no matter what
2187 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2188 optimize or assume that the result of the '``and``' is '``undef``'.
2189 However, it is safe to assume that all bits of the '``undef``' could be
2190 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2191 all the bits of the '``undef``' operand to the '``or``' could be set,
2192 allowing the '``or``' to be folded to -1.
2194 .. code-block:: llvm
2196 %A = select undef, %X, %Y
2197 %B = select undef, 42, %Y
2198 %C = select %X, %Y, undef
2208 This set of examples shows that undefined '``select``' (and conditional
2209 branch) conditions can go *either way*, but they have to come from one
2210 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2211 both known to have a clear low bit, then ``%A`` would have to have a
2212 cleared low bit. However, in the ``%C`` example, the optimizer is
2213 allowed to assume that the '``undef``' operand could be the same as
2214 ``%Y``, allowing the whole '``select``' to be eliminated.
2216 .. code-block:: llvm
2218 %A = xor undef, undef
2235 This example points out that two '``undef``' operands are not
2236 necessarily the same. This can be surprising to people (and also matches
2237 C semantics) where they assume that "``X^X``" is always zero, even if
2238 ``X`` is undefined. This isn't true for a number of reasons, but the
2239 short answer is that an '``undef``' "variable" can arbitrarily change
2240 its value over its "live range". This is true because the variable
2241 doesn't actually *have a live range*. Instead, the value is logically
2242 read from arbitrary registers that happen to be around when needed, so
2243 the value is not necessarily consistent over time. In fact, ``%A`` and
2244 ``%C`` need to have the same semantics or the core LLVM "replace all
2245 uses with" concept would not hold.
2247 .. code-block:: llvm
2255 These examples show the crucial difference between an *undefined value*
2256 and *undefined behavior*. An undefined value (like '``undef``') is
2257 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2258 operation can be constant folded to '``undef``', because the '``undef``'
2259 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2260 However, in the second example, we can make a more aggressive
2261 assumption: because the ``undef`` is allowed to be an arbitrary value,
2262 we are allowed to assume that it could be zero. Since a divide by zero
2263 has *undefined behavior*, we are allowed to assume that the operation
2264 does not execute at all. This allows us to delete the divide and all
2265 code after it. Because the undefined operation "can't happen", the
2266 optimizer can assume that it occurs in dead code.
2268 .. code-block:: llvm
2270 a: store undef -> %X
2271 b: store %X -> undef
2276 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2277 value can be assumed to not have any effect; we can assume that the
2278 value is overwritten with bits that happen to match what was already
2279 there. However, a store *to* an undefined location could clobber
2280 arbitrary memory, therefore, it has undefined behavior.
2287 Poison values are similar to :ref:`undef values <undefvalues>`, however
2288 they also represent the fact that an instruction or constant expression
2289 which cannot evoke side effects has nevertheless detected a condition
2290 which results in undefined behavior.
2292 There is currently no way of representing a poison value in the IR; they
2293 only exist when produced by operations such as :ref:`add <i_add>` with
2296 Poison value behavior is defined in terms of value *dependence*:
2298 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2299 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2300 their dynamic predecessor basic block.
2301 - Function arguments depend on the corresponding actual argument values
2302 in the dynamic callers of their functions.
2303 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2304 instructions that dynamically transfer control back to them.
2305 - :ref:`Invoke <i_invoke>` instructions depend on the
2306 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2307 call instructions that dynamically transfer control back to them.
2308 - Non-volatile loads and stores depend on the most recent stores to all
2309 of the referenced memory addresses, following the order in the IR
2310 (including loads and stores implied by intrinsics such as
2311 :ref:`@llvm.memcpy <int_memcpy>`.)
2312 - An instruction with externally visible side effects depends on the
2313 most recent preceding instruction with externally visible side
2314 effects, following the order in the IR. (This includes :ref:`volatile
2315 operations <volatile>`.)
2316 - An instruction *control-depends* on a :ref:`terminator
2317 instruction <terminators>` if the terminator instruction has
2318 multiple successors and the instruction is always executed when
2319 control transfers to one of the successors, and may not be executed
2320 when control is transferred to another.
2321 - Additionally, an instruction also *control-depends* on a terminator
2322 instruction if the set of instructions it otherwise depends on would
2323 be different if the terminator had transferred control to a different
2325 - Dependence is transitive.
2327 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2328 with the additional affect that any instruction which has a *dependence*
2329 on a poison value has undefined behavior.
2331 Here are some examples:
2333 .. code-block:: llvm
2336 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2337 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2338 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2339 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2341 store i32 %poison, i32* @g ; Poison value stored to memory.
2342 %poison2 = load i32* @g ; Poison value loaded back from memory.
2344 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2346 %narrowaddr = bitcast i32* @g to i16*
2347 %wideaddr = bitcast i32* @g to i64*
2348 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2349 %poison4 = load i64* %wideaddr ; Returns a poison value.
2351 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2352 br i1 %cmp, label %true, label %end ; Branch to either destination.
2355 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2356 ; it has undefined behavior.
2360 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2361 ; Both edges into this PHI are
2362 ; control-dependent on %cmp, so this
2363 ; always results in a poison value.
2365 store volatile i32 0, i32* @g ; This would depend on the store in %true
2366 ; if %cmp is true, or the store in %entry
2367 ; otherwise, so this is undefined behavior.
2369 br i1 %cmp, label %second_true, label %second_end
2370 ; The same branch again, but this time the
2371 ; true block doesn't have side effects.
2378 store volatile i32 0, i32* @g ; This time, the instruction always depends
2379 ; on the store in %end. Also, it is
2380 ; control-equivalent to %end, so this is
2381 ; well-defined (ignoring earlier undefined
2382 ; behavior in this example).
2386 Addresses of Basic Blocks
2387 -------------------------
2389 ``blockaddress(@function, %block)``
2391 The '``blockaddress``' constant computes the address of the specified
2392 basic block in the specified function, and always has an ``i8*`` type.
2393 Taking the address of the entry block is illegal.
2395 This value only has defined behavior when used as an operand to the
2396 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2397 against null. Pointer equality tests between labels addresses results in
2398 undefined behavior --- though, again, comparison against null is ok, and
2399 no label is equal to the null pointer. This may be passed around as an
2400 opaque pointer sized value as long as the bits are not inspected. This
2401 allows ``ptrtoint`` and arithmetic to be performed on these values so
2402 long as the original value is reconstituted before the ``indirectbr``
2405 Finally, some targets may provide defined semantics when using the value
2406 as the operand to an inline assembly, but that is target specific.
2410 Constant Expressions
2411 --------------------
2413 Constant expressions are used to allow expressions involving other
2414 constants to be used as constants. Constant expressions may be of any
2415 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2416 that does not have side effects (e.g. load and call are not supported).
2417 The following is the syntax for constant expressions:
2419 ``trunc (CST to TYPE)``
2420 Truncate a constant to another type. The bit size of CST must be
2421 larger than the bit size of TYPE. Both types must be integers.
2422 ``zext (CST to TYPE)``
2423 Zero extend a constant to another type. The bit size of CST must be
2424 smaller than the bit size of TYPE. Both types must be integers.
2425 ``sext (CST to TYPE)``
2426 Sign extend a constant to another type. The bit size of CST must be
2427 smaller than the bit size of TYPE. Both types must be integers.
2428 ``fptrunc (CST to TYPE)``
2429 Truncate a floating point constant to another floating point type.
2430 The size of CST must be larger than the size of TYPE. Both types
2431 must be floating point.
2432 ``fpext (CST to TYPE)``
2433 Floating point extend a constant to another type. The size of CST
2434 must be smaller or equal to the size of TYPE. Both types must be
2436 ``fptoui (CST to TYPE)``
2437 Convert a floating point constant to the corresponding unsigned
2438 integer constant. TYPE must be a scalar or vector integer type. CST
2439 must be of scalar or vector floating point type. Both CST and TYPE
2440 must be scalars, or vectors of the same number of elements. If the
2441 value won't fit in the integer type, the results are undefined.
2442 ``fptosi (CST to TYPE)``
2443 Convert a floating point constant to the corresponding signed
2444 integer constant. TYPE must be a scalar or vector integer type. CST
2445 must be of scalar or vector floating point type. Both CST and TYPE
2446 must be scalars, or vectors of the same number of elements. If the
2447 value won't fit in the integer type, the results are undefined.
2448 ``uitofp (CST to TYPE)``
2449 Convert an unsigned integer constant to the corresponding floating
2450 point constant. TYPE must be a scalar or vector floating point type.
2451 CST must be of scalar or vector integer type. Both CST and TYPE must
2452 be scalars, or vectors of the same number of elements. If the value
2453 won't fit in the floating point type, the results are undefined.
2454 ``sitofp (CST to TYPE)``
2455 Convert a signed integer constant to the corresponding floating
2456 point constant. TYPE must be a scalar or vector floating point type.
2457 CST must be of scalar or vector integer type. Both CST and TYPE must
2458 be scalars, or vectors of the same number of elements. If the value
2459 won't fit in the floating point type, the results are undefined.
2460 ``ptrtoint (CST to TYPE)``
2461 Convert a pointer typed constant to the corresponding integer
2462 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2463 pointer type. The ``CST`` value is zero extended, truncated, or
2464 unchanged to make it fit in ``TYPE``.
2465 ``inttoptr (CST to TYPE)``
2466 Convert an integer constant to a pointer constant. TYPE must be a
2467 pointer type. CST must be of integer type. The CST value is zero
2468 extended, truncated, or unchanged to make it fit in a pointer size.
2469 This one is *really* dangerous!
2470 ``bitcast (CST to TYPE)``
2471 Convert a constant, CST, to another TYPE. The constraints of the
2472 operands are the same as those for the :ref:`bitcast
2473 instruction <i_bitcast>`.
2474 ``addrspacecast (CST to TYPE)``
2475 Convert a constant pointer or constant vector of pointer, CST, to another
2476 TYPE in a different address space. The constraints of the operands are the
2477 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2478 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2479 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2480 constants. As with the :ref:`getelementptr <i_getelementptr>`
2481 instruction, the index list may have zero or more indexes, which are
2482 required to make sense for the type of "CSTPTR".
2483 ``select (COND, VAL1, VAL2)``
2484 Perform the :ref:`select operation <i_select>` on constants.
2485 ``icmp COND (VAL1, VAL2)``
2486 Performs the :ref:`icmp operation <i_icmp>` on constants.
2487 ``fcmp COND (VAL1, VAL2)``
2488 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2489 ``extractelement (VAL, IDX)``
2490 Perform the :ref:`extractelement operation <i_extractelement>` on
2492 ``insertelement (VAL, ELT, IDX)``
2493 Perform the :ref:`insertelement operation <i_insertelement>` on
2495 ``shufflevector (VEC1, VEC2, IDXMASK)``
2496 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2498 ``extractvalue (VAL, IDX0, IDX1, ...)``
2499 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2500 constants. The index list is interpreted in a similar manner as
2501 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2502 least one index value must be specified.
2503 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2504 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2505 The index list is interpreted in a similar manner as indices in a
2506 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2507 value must be specified.
2508 ``OPCODE (LHS, RHS)``
2509 Perform the specified operation of the LHS and RHS constants. OPCODE
2510 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2511 binary <bitwiseops>` operations. The constraints on operands are
2512 the same as those for the corresponding instruction (e.g. no bitwise
2513 operations on floating point values are allowed).
2520 Inline Assembler Expressions
2521 ----------------------------
2523 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2524 Inline Assembly <moduleasm>`) through the use of a special value. This
2525 value represents the inline assembler as a string (containing the
2526 instructions to emit), a list of operand constraints (stored as a
2527 string), a flag that indicates whether or not the inline asm expression
2528 has side effects, and a flag indicating whether the function containing
2529 the asm needs to align its stack conservatively. An example inline
2530 assembler expression is:
2532 .. code-block:: llvm
2534 i32 (i32) asm "bswap $0", "=r,r"
2536 Inline assembler expressions may **only** be used as the callee operand
2537 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2538 Thus, typically we have:
2540 .. code-block:: llvm
2542 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2544 Inline asms with side effects not visible in the constraint list must be
2545 marked as having side effects. This is done through the use of the
2546 '``sideeffect``' keyword, like so:
2548 .. code-block:: llvm
2550 call void asm sideeffect "eieio", ""()
2552 In some cases inline asms will contain code that will not work unless
2553 the stack is aligned in some way, such as calls or SSE instructions on
2554 x86, yet will not contain code that does that alignment within the asm.
2555 The compiler should make conservative assumptions about what the asm
2556 might contain and should generate its usual stack alignment code in the
2557 prologue if the '``alignstack``' keyword is present:
2559 .. code-block:: llvm
2561 call void asm alignstack "eieio", ""()
2563 Inline asms also support using non-standard assembly dialects. The
2564 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2565 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2566 the only supported dialects. An example is:
2568 .. code-block:: llvm
2570 call void asm inteldialect "eieio", ""()
2572 If multiple keywords appear the '``sideeffect``' keyword must come
2573 first, the '``alignstack``' keyword second and the '``inteldialect``'
2579 The call instructions that wrap inline asm nodes may have a
2580 "``!srcloc``" MDNode attached to it that contains a list of constant
2581 integers. If present, the code generator will use the integer as the
2582 location cookie value when report errors through the ``LLVMContext``
2583 error reporting mechanisms. This allows a front-end to correlate backend
2584 errors that occur with inline asm back to the source code that produced
2587 .. code-block:: llvm
2589 call void asm sideeffect "something bad", ""(), !srcloc !42
2591 !42 = !{ i32 1234567 }
2593 It is up to the front-end to make sense of the magic numbers it places
2594 in the IR. If the MDNode contains multiple constants, the code generator
2595 will use the one that corresponds to the line of the asm that the error
2600 Metadata Nodes and Metadata Strings
2601 -----------------------------------
2603 LLVM IR allows metadata to be attached to instructions in the program
2604 that can convey extra information about the code to the optimizers and
2605 code generator. One example application of metadata is source-level
2606 debug information. There are two metadata primitives: strings and nodes.
2607 All metadata has the ``metadata`` type and is identified in syntax by a
2608 preceding exclamation point ('``!``').
2610 A metadata string is a string surrounded by double quotes. It can
2611 contain any character by escaping non-printable characters with
2612 "``\xx``" where "``xx``" is the two digit hex code. For example:
2615 Metadata nodes are represented with notation similar to structure
2616 constants (a comma separated list of elements, surrounded by braces and
2617 preceded by an exclamation point). Metadata nodes can have any values as
2618 their operand. For example:
2620 .. code-block:: llvm
2622 !{ metadata !"test\00", i32 10}
2624 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2625 metadata nodes, which can be looked up in the module symbol table. For
2628 .. code-block:: llvm
2630 !foo = metadata !{!4, !3}
2632 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2633 function is using two metadata arguments:
2635 .. code-block:: llvm
2637 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2639 Metadata can be attached with an instruction. Here metadata ``!21`` is
2640 attached to the ``add`` instruction using the ``!dbg`` identifier:
2642 .. code-block:: llvm
2644 %indvar.next = add i64 %indvar, 1, !dbg !21
2646 More information about specific metadata nodes recognized by the
2647 optimizers and code generator is found below.
2652 In LLVM IR, memory does not have types, so LLVM's own type system is not
2653 suitable for doing TBAA. Instead, metadata is added to the IR to
2654 describe a type system of a higher level language. This can be used to
2655 implement typical C/C++ TBAA, but it can also be used to implement
2656 custom alias analysis behavior for other languages.
2658 The current metadata format is very simple. TBAA metadata nodes have up
2659 to three fields, e.g.:
2661 .. code-block:: llvm
2663 !0 = metadata !{ metadata !"an example type tree" }
2664 !1 = metadata !{ metadata !"int", metadata !0 }
2665 !2 = metadata !{ metadata !"float", metadata !0 }
2666 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2668 The first field is an identity field. It can be any value, usually a
2669 metadata string, which uniquely identifies the type. The most important
2670 name in the tree is the name of the root node. Two trees with different
2671 root node names are entirely disjoint, even if they have leaves with
2674 The second field identifies the type's parent node in the tree, or is
2675 null or omitted for a root node. A type is considered to alias all of
2676 its descendants and all of its ancestors in the tree. Also, a type is
2677 considered to alias all types in other trees, so that bitcode produced
2678 from multiple front-ends is handled conservatively.
2680 If the third field is present, it's an integer which if equal to 1
2681 indicates that the type is "constant" (meaning
2682 ``pointsToConstantMemory`` should return true; see `other useful
2683 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2685 '``tbaa.struct``' Metadata
2686 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2688 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2689 aggregate assignment operations in C and similar languages, however it
2690 is defined to copy a contiguous region of memory, which is more than
2691 strictly necessary for aggregate types which contain holes due to
2692 padding. Also, it doesn't contain any TBAA information about the fields
2695 ``!tbaa.struct`` metadata can describe which memory subregions in a
2696 memcpy are padding and what the TBAA tags of the struct are.
2698 The current metadata format is very simple. ``!tbaa.struct`` metadata
2699 nodes are a list of operands which are in conceptual groups of three.
2700 For each group of three, the first operand gives the byte offset of a
2701 field in bytes, the second gives its size in bytes, and the third gives
2704 .. code-block:: llvm
2706 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2708 This describes a struct with two fields. The first is at offset 0 bytes
2709 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2710 and has size 4 bytes and has tbaa tag !2.
2712 Note that the fields need not be contiguous. In this example, there is a
2713 4 byte gap between the two fields. This gap represents padding which
2714 does not carry useful data and need not be preserved.
2716 '``fpmath``' Metadata
2717 ^^^^^^^^^^^^^^^^^^^^^
2719 ``fpmath`` metadata may be attached to any instruction of floating point
2720 type. It can be used to express the maximum acceptable error in the
2721 result of that instruction, in ULPs, thus potentially allowing the
2722 compiler to use a more efficient but less accurate method of computing
2723 it. ULP is defined as follows:
2725 If ``x`` is a real number that lies between two finite consecutive
2726 floating-point numbers ``a`` and ``b``, without being equal to one
2727 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2728 distance between the two non-equal finite floating-point numbers
2729 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2731 The metadata node shall consist of a single positive floating point
2732 number representing the maximum relative error, for example:
2734 .. code-block:: llvm
2736 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2738 '``range``' Metadata
2739 ^^^^^^^^^^^^^^^^^^^^
2741 ``range`` metadata may be attached only to loads of integer types. It
2742 expresses the possible ranges the loaded value is in. The ranges are
2743 represented with a flattened list of integers. The loaded value is known
2744 to be in the union of the ranges defined by each consecutive pair. Each
2745 pair has the following properties:
2747 - The type must match the type loaded by the instruction.
2748 - The pair ``a,b`` represents the range ``[a,b)``.
2749 - Both ``a`` and ``b`` are constants.
2750 - The range is allowed to wrap.
2751 - The range should not represent the full or empty set. That is,
2754 In addition, the pairs must be in signed order of the lower bound and
2755 they must be non-contiguous.
2759 .. code-block:: llvm
2761 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2762 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2763 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2764 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2766 !0 = metadata !{ i8 0, i8 2 }
2767 !1 = metadata !{ i8 255, i8 2 }
2768 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2769 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2774 It is sometimes useful to attach information to loop constructs. Currently,
2775 loop metadata is implemented as metadata attached to the branch instruction
2776 in the loop latch block. This type of metadata refer to a metadata node that is
2777 guaranteed to be separate for each loop. The loop identifier metadata is
2778 specified with the name ``llvm.loop``.
2780 The loop identifier metadata is implemented using a metadata that refers to
2781 itself to avoid merging it with any other identifier metadata, e.g.,
2782 during module linkage or function inlining. That is, each loop should refer
2783 to their own identification metadata even if they reside in separate functions.
2784 The following example contains loop identifier metadata for two separate loop
2787 .. code-block:: llvm
2789 !0 = metadata !{ metadata !0 }
2790 !1 = metadata !{ metadata !1 }
2792 The loop identifier metadata can be used to specify additional per-loop
2793 metadata. Any operands after the first operand can be treated as user-defined
2794 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2795 by the loop vectorizer to indicate how many times to unroll the loop:
2797 .. code-block:: llvm
2799 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2801 !0 = metadata !{ metadata !0, metadata !1 }
2802 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2807 Metadata types used to annotate memory accesses with information helpful
2808 for optimizations are prefixed with ``llvm.mem``.
2810 '``llvm.mem.parallel_loop_access``' Metadata
2811 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2813 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
2814 or metadata containing a list of loop identifiers for nested loops.
2815 The metadata is attached to memory accessing instructions and denotes that
2816 no loop carried memory dependence exist between it and other instructions denoted
2817 with the same loop identifier.
2819 Precisely, given two instructions ``m1`` and ``m2`` that both have the
2820 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
2821 set of loops associated with that metadata, respectively, then there is no loop
2822 carried dependence between ``m1`` and ``m2`` for loops ``L1`` or
2825 As a special case, if all memory accessing instructions in a loop have
2826 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
2827 loop has no loop carried memory dependences and is considered to be a parallel
2830 Note that if not all memory access instructions have such metadata referring to
2831 the loop, then the loop is considered not being trivially parallel. Additional
2832 memory dependence analysis is required to make that determination. As a fail
2833 safe mechanism, this causes loops that were originally parallel to be considered
2834 sequential (if optimization passes that are unaware of the parallel semantics
2835 insert new memory instructions into the loop body).
2837 Example of a loop that is considered parallel due to its correct use of
2838 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2839 metadata types that refer to the same loop identifier metadata.
2841 .. code-block:: llvm
2845 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
2847 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2849 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2853 !0 = metadata !{ metadata !0 }
2855 It is also possible to have nested parallel loops. In that case the
2856 memory accesses refer to a list of loop identifier metadata nodes instead of
2857 the loop identifier metadata node directly:
2859 .. code-block:: llvm
2863 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
2865 br label %inner.for.body
2869 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2871 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
2873 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2877 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
2879 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2881 outer.for.end: ; preds = %for.body
2883 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2884 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2885 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2887 '``llvm.vectorizer``'
2888 ^^^^^^^^^^^^^^^^^^^^^
2890 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2891 vectorization parameters such as vectorization factor and unroll factor.
2893 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2894 loop identification metadata.
2896 '``llvm.vectorizer.unroll``' Metadata
2897 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2899 This metadata instructs the loop vectorizer to unroll the specified
2900 loop exactly ``N`` times.
2902 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2903 operand is an integer specifying the unroll factor. For example:
2905 .. code-block:: llvm
2907 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2909 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2912 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2913 determined automatically.
2915 '``llvm.vectorizer.width``' Metadata
2916 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2918 This metadata sets the target width of the vectorizer to ``N``. Without
2919 this metadata, the vectorizer will choose a width automatically.
2920 Regardless of this metadata, the vectorizer will only vectorize loops if
2921 it believes it is valid to do so.
2923 The first operand is the string ``llvm.vectorizer.width`` and the second
2924 operand is an integer specifying the width. For example:
2926 .. code-block:: llvm
2928 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2930 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2933 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2936 Module Flags Metadata
2937 =====================
2939 Information about the module as a whole is difficult to convey to LLVM's
2940 subsystems. The LLVM IR isn't sufficient to transmit this information.
2941 The ``llvm.module.flags`` named metadata exists in order to facilitate
2942 this. These flags are in the form of key / value pairs --- much like a
2943 dictionary --- making it easy for any subsystem who cares about a flag to
2946 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2947 Each triplet has the following form:
2949 - The first element is a *behavior* flag, which specifies the behavior
2950 when two (or more) modules are merged together, and it encounters two
2951 (or more) metadata with the same ID. The supported behaviors are
2953 - The second element is a metadata string that is a unique ID for the
2954 metadata. Each module may only have one flag entry for each unique ID (not
2955 including entries with the **Require** behavior).
2956 - The third element is the value of the flag.
2958 When two (or more) modules are merged together, the resulting
2959 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2960 each unique metadata ID string, there will be exactly one entry in the merged
2961 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2962 be determined by the merge behavior flag, as described below. The only exception
2963 is that entries with the *Require* behavior are always preserved.
2965 The following behaviors are supported:
2976 Emits an error if two values disagree, otherwise the resulting value
2977 is that of the operands.
2981 Emits a warning if two values disagree. The result value will be the
2982 operand for the flag from the first module being linked.
2986 Adds a requirement that another module flag be present and have a
2987 specified value after linking is performed. The value must be a
2988 metadata pair, where the first element of the pair is the ID of the
2989 module flag to be restricted, and the second element of the pair is
2990 the value the module flag should be restricted to. This behavior can
2991 be used to restrict the allowable results (via triggering of an
2992 error) of linking IDs with the **Override** behavior.
2996 Uses the specified value, regardless of the behavior or value of the
2997 other module. If both modules specify **Override**, but the values
2998 differ, an error will be emitted.
3002 Appends the two values, which are required to be metadata nodes.
3006 Appends the two values, which are required to be metadata
3007 nodes. However, duplicate entries in the second list are dropped
3008 during the append operation.
3010 It is an error for a particular unique flag ID to have multiple behaviors,
3011 except in the case of **Require** (which adds restrictions on another metadata
3012 value) or **Override**.
3014 An example of module flags:
3016 .. code-block:: llvm
3018 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3019 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3020 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3021 !3 = metadata !{ i32 3, metadata !"qux",
3023 metadata !"foo", i32 1
3026 !llvm.module.flags = !{ !0, !1, !2, !3 }
3028 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3029 if two or more ``!"foo"`` flags are seen is to emit an error if their
3030 values are not equal.
3032 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3033 behavior if two or more ``!"bar"`` flags are seen is to use the value
3036 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3037 behavior if two or more ``!"qux"`` flags are seen is to emit a
3038 warning if their values are not equal.
3040 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3044 metadata !{ metadata !"foo", i32 1 }
3046 The behavior is to emit an error if the ``llvm.module.flags`` does not
3047 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3050 Objective-C Garbage Collection Module Flags Metadata
3051 ----------------------------------------------------
3053 On the Mach-O platform, Objective-C stores metadata about garbage
3054 collection in a special section called "image info". The metadata
3055 consists of a version number and a bitmask specifying what types of
3056 garbage collection are supported (if any) by the file. If two or more
3057 modules are linked together their garbage collection metadata needs to
3058 be merged rather than appended together.
3060 The Objective-C garbage collection module flags metadata consists of the
3061 following key-value pairs:
3070 * - ``Objective-C Version``
3071 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3073 * - ``Objective-C Image Info Version``
3074 - **[Required]** --- The version of the image info section. Currently
3077 * - ``Objective-C Image Info Section``
3078 - **[Required]** --- The section to place the metadata. Valid values are
3079 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3080 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3081 Objective-C ABI version 2.
3083 * - ``Objective-C Garbage Collection``
3084 - **[Required]** --- Specifies whether garbage collection is supported or
3085 not. Valid values are 0, for no garbage collection, and 2, for garbage
3086 collection supported.
3088 * - ``Objective-C GC Only``
3089 - **[Optional]** --- Specifies that only garbage collection is supported.
3090 If present, its value must be 6. This flag requires that the
3091 ``Objective-C Garbage Collection`` flag have the value 2.
3093 Some important flag interactions:
3095 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3096 merged with a module with ``Objective-C Garbage Collection`` set to
3097 2, then the resulting module has the
3098 ``Objective-C Garbage Collection`` flag set to 0.
3099 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3100 merged with a module with ``Objective-C GC Only`` set to 6.
3102 Automatic Linker Flags Module Flags Metadata
3103 --------------------------------------------
3105 Some targets support embedding flags to the linker inside individual object
3106 files. Typically this is used in conjunction with language extensions which
3107 allow source files to explicitly declare the libraries they depend on, and have
3108 these automatically be transmitted to the linker via object files.
3110 These flags are encoded in the IR using metadata in the module flags section,
3111 using the ``Linker Options`` key. The merge behavior for this flag is required
3112 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3113 node which should be a list of other metadata nodes, each of which should be a
3114 list of metadata strings defining linker options.
3116 For example, the following metadata section specifies two separate sets of
3117 linker options, presumably to link against ``libz`` and the ``Cocoa``
3120 !0 = metadata !{ i32 6, metadata !"Linker Options",
3122 metadata !{ metadata !"-lz" },
3123 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3124 !llvm.module.flags = !{ !0 }
3126 The metadata encoding as lists of lists of options, as opposed to a collapsed
3127 list of options, is chosen so that the IR encoding can use multiple option
3128 strings to specify e.g., a single library, while still having that specifier be
3129 preserved as an atomic element that can be recognized by a target specific
3130 assembly writer or object file emitter.
3132 Each individual option is required to be either a valid option for the target's
3133 linker, or an option that is reserved by the target specific assembly writer or
3134 object file emitter. No other aspect of these options is defined by the IR.
3136 .. _intrinsicglobalvariables:
3138 Intrinsic Global Variables
3139 ==========================
3141 LLVM has a number of "magic" global variables that contain data that
3142 affect code generation or other IR semantics. These are documented here.
3143 All globals of this sort should have a section specified as
3144 "``llvm.metadata``". This section and all globals that start with
3145 "``llvm.``" are reserved for use by LLVM.
3149 The '``llvm.used``' Global Variable
3150 -----------------------------------
3152 The ``@llvm.used`` global is an array which has
3153 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3154 pointers to named global variables, functions and aliases which may optionally
3155 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3158 .. code-block:: llvm
3163 @llvm.used = appending global [2 x i8*] [
3165 i8* bitcast (i32* @Y to i8*)
3166 ], section "llvm.metadata"
3168 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3169 and linker are required to treat the symbol as if there is a reference to the
3170 symbol that it cannot see (which is why they have to be named). For example, if
3171 a variable has internal linkage and no references other than that from the
3172 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3173 references from inline asms and other things the compiler cannot "see", and
3174 corresponds to "``attribute((used))``" in GNU C.
3176 On some targets, the code generator must emit a directive to the
3177 assembler or object file to prevent the assembler and linker from
3178 molesting the symbol.
3180 .. _gv_llvmcompilerused:
3182 The '``llvm.compiler.used``' Global Variable
3183 --------------------------------------------
3185 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3186 directive, except that it only prevents the compiler from touching the
3187 symbol. On targets that support it, this allows an intelligent linker to
3188 optimize references to the symbol without being impeded as it would be
3191 This is a rare construct that should only be used in rare circumstances,
3192 and should not be exposed to source languages.
3194 .. _gv_llvmglobalctors:
3196 The '``llvm.global_ctors``' Global Variable
3197 -------------------------------------------
3199 .. code-block:: llvm
3201 %0 = type { i32, void ()*, i8* }
3202 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
3204 The ``@llvm.global_ctors`` array contains a list of constructor
3205 functions, priorities, and an optional associated global or function.
3206 The functions referenced by this array will be called in ascending order
3207 of priority (i.e. lowest first) when the module is loaded. The order of
3208 functions with the same priority is not defined.
3210 If the third field is present, non-null, and points to a global variable
3211 or function, the initializer function will only run if the associated
3212 data from the current module is not discarded.
3214 .. _llvmglobaldtors:
3216 The '``llvm.global_dtors``' Global Variable
3217 -------------------------------------------
3219 .. code-block:: llvm
3221 %0 = type { i32, void ()*, i8* }
3222 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
3224 The ``@llvm.global_dtors`` array contains a list of destructor
3225 functions, priorities, and an optional associated global or function.
3226 The functions referenced by this array will be called in descending
3227 order of priority (i.e. highest first) when the module is unloaded. The
3228 order of functions with the same priority is not defined.
3230 If the third field is present, non-null, and points to a global variable
3231 or function, the destructor function will only run if the associated
3232 data from the current module is not discarded.
3234 Instruction Reference
3235 =====================
3237 The LLVM instruction set consists of several different classifications
3238 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3239 instructions <binaryops>`, :ref:`bitwise binary
3240 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3241 :ref:`other instructions <otherops>`.
3245 Terminator Instructions
3246 -----------------------
3248 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3249 program ends with a "Terminator" instruction, which indicates which
3250 block should be executed after the current block is finished. These
3251 terminator instructions typically yield a '``void``' value: they produce
3252 control flow, not values (the one exception being the
3253 ':ref:`invoke <i_invoke>`' instruction).
3255 The terminator instructions are: ':ref:`ret <i_ret>`',
3256 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3257 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3258 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3262 '``ret``' Instruction
3263 ^^^^^^^^^^^^^^^^^^^^^
3270 ret <type> <value> ; Return a value from a non-void function
3271 ret void ; Return from void function
3276 The '``ret``' instruction is used to return control flow (and optionally
3277 a value) from a function back to the caller.
3279 There are two forms of the '``ret``' instruction: one that returns a
3280 value and then causes control flow, and one that just causes control
3286 The '``ret``' instruction optionally accepts a single argument, the
3287 return value. The type of the return value must be a ':ref:`first
3288 class <t_firstclass>`' type.
3290 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3291 return type and contains a '``ret``' instruction with no return value or
3292 a return value with a type that does not match its type, or if it has a
3293 void return type and contains a '``ret``' instruction with a return
3299 When the '``ret``' instruction is executed, control flow returns back to
3300 the calling function's context. If the caller is a
3301 ":ref:`call <i_call>`" instruction, execution continues at the
3302 instruction after the call. If the caller was an
3303 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3304 beginning of the "normal" destination block. If the instruction returns
3305 a value, that value shall set the call or invoke instruction's return
3311 .. code-block:: llvm
3313 ret i32 5 ; Return an integer value of 5
3314 ret void ; Return from a void function
3315 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3319 '``br``' Instruction
3320 ^^^^^^^^^^^^^^^^^^^^
3327 br i1 <cond>, label <iftrue>, label <iffalse>
3328 br label <dest> ; Unconditional branch
3333 The '``br``' instruction is used to cause control flow to transfer to a
3334 different basic block in the current function. There are two forms of
3335 this instruction, corresponding to a conditional branch and an
3336 unconditional branch.
3341 The conditional branch form of the '``br``' instruction takes a single
3342 '``i1``' value and two '``label``' values. The unconditional form of the
3343 '``br``' instruction takes a single '``label``' value as a target.
3348 Upon execution of a conditional '``br``' instruction, the '``i1``'
3349 argument is evaluated. If the value is ``true``, control flows to the
3350 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3351 to the '``iffalse``' ``label`` argument.
3356 .. code-block:: llvm
3359 %cond = icmp eq i32 %a, %b
3360 br i1 %cond, label %IfEqual, label %IfUnequal
3368 '``switch``' Instruction
3369 ^^^^^^^^^^^^^^^^^^^^^^^^
3376 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3381 The '``switch``' instruction is used to transfer control flow to one of
3382 several different places. It is a generalization of the '``br``'
3383 instruction, allowing a branch to occur to one of many possible
3389 The '``switch``' instruction uses three parameters: an integer
3390 comparison value '``value``', a default '``label``' destination, and an
3391 array of pairs of comparison value constants and '``label``'s. The table
3392 is not allowed to contain duplicate constant entries.
3397 The ``switch`` instruction specifies a table of values and destinations.
3398 When the '``switch``' instruction is executed, this table is searched
3399 for the given value. If the value is found, control flow is transferred
3400 to the corresponding destination; otherwise, control flow is transferred
3401 to the default destination.
3406 Depending on properties of the target machine and the particular
3407 ``switch`` instruction, this instruction may be code generated in
3408 different ways. For example, it could be generated as a series of
3409 chained conditional branches or with a lookup table.
3414 .. code-block:: llvm
3416 ; Emulate a conditional br instruction
3417 %Val = zext i1 %value to i32
3418 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3420 ; Emulate an unconditional br instruction
3421 switch i32 0, label %dest [ ]
3423 ; Implement a jump table:
3424 switch i32 %val, label %otherwise [ i32 0, label %onzero
3426 i32 2, label %ontwo ]
3430 '``indirectbr``' Instruction
3431 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3438 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3443 The '``indirectbr``' instruction implements an indirect branch to a
3444 label within the current function, whose address is specified by
3445 "``address``". Address must be derived from a
3446 :ref:`blockaddress <blockaddress>` constant.
3451 The '``address``' argument is the address of the label to jump to. The
3452 rest of the arguments indicate the full set of possible destinations
3453 that the address may point to. Blocks are allowed to occur multiple
3454 times in the destination list, though this isn't particularly useful.
3456 This destination list is required so that dataflow analysis has an
3457 accurate understanding of the CFG.
3462 Control transfers to the block specified in the address argument. All
3463 possible destination blocks must be listed in the label list, otherwise
3464 this instruction has undefined behavior. This implies that jumps to
3465 labels defined in other functions have undefined behavior as well.
3470 This is typically implemented with a jump through a register.
3475 .. code-block:: llvm
3477 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3481 '``invoke``' Instruction
3482 ^^^^^^^^^^^^^^^^^^^^^^^^
3489 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3490 to label <normal label> unwind label <exception label>
3495 The '``invoke``' instruction causes control to transfer to a specified
3496 function, with the possibility of control flow transfer to either the
3497 '``normal``' label or the '``exception``' label. If the callee function
3498 returns with the "``ret``" instruction, control flow will return to the
3499 "normal" label. If the callee (or any indirect callees) returns via the
3500 ":ref:`resume <i_resume>`" instruction or other exception handling
3501 mechanism, control is interrupted and continued at the dynamically
3502 nearest "exception" label.
3504 The '``exception``' label is a `landing
3505 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3506 '``exception``' label is required to have the
3507 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3508 information about the behavior of the program after unwinding happens,
3509 as its first non-PHI instruction. The restrictions on the
3510 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3511 instruction, so that the important information contained within the
3512 "``landingpad``" instruction can't be lost through normal code motion.
3517 This instruction requires several arguments:
3519 #. The optional "cconv" marker indicates which :ref:`calling
3520 convention <callingconv>` the call should use. If none is
3521 specified, the call defaults to using C calling conventions.
3522 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3523 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3525 #. '``ptr to function ty``': shall be the signature of the pointer to
3526 function value being invoked. In most cases, this is a direct
3527 function invocation, but indirect ``invoke``'s are just as possible,
3528 branching off an arbitrary pointer to function value.
3529 #. '``function ptr val``': An LLVM value containing a pointer to a
3530 function to be invoked.
3531 #. '``function args``': argument list whose types match the function
3532 signature argument types and parameter attributes. All arguments must
3533 be of :ref:`first class <t_firstclass>` type. If the function signature
3534 indicates the function accepts a variable number of arguments, the
3535 extra arguments can be specified.
3536 #. '``normal label``': the label reached when the called function
3537 executes a '``ret``' instruction.
3538 #. '``exception label``': the label reached when a callee returns via
3539 the :ref:`resume <i_resume>` instruction or other exception handling
3541 #. The optional :ref:`function attributes <fnattrs>` list. Only
3542 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3543 attributes are valid here.
3548 This instruction is designed to operate as a standard '``call``'
3549 instruction in most regards. The primary difference is that it
3550 establishes an association with a label, which is used by the runtime
3551 library to unwind the stack.
3553 This instruction is used in languages with destructors to ensure that
3554 proper cleanup is performed in the case of either a ``longjmp`` or a
3555 thrown exception. Additionally, this is important for implementation of
3556 '``catch``' clauses in high-level languages that support them.
3558 For the purposes of the SSA form, the definition of the value returned
3559 by the '``invoke``' instruction is deemed to occur on the edge from the
3560 current block to the "normal" label. If the callee unwinds then no
3561 return value is available.
3566 .. code-block:: llvm
3568 %retval = invoke i32 @Test(i32 15) to label %Continue
3569 unwind label %TestCleanup ; {i32}:retval set
3570 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3571 unwind label %TestCleanup ; {i32}:retval set
3575 '``resume``' Instruction
3576 ^^^^^^^^^^^^^^^^^^^^^^^^
3583 resume <type> <value>
3588 The '``resume``' instruction is a terminator instruction that has no
3594 The '``resume``' instruction requires one argument, which must have the
3595 same type as the result of any '``landingpad``' instruction in the same
3601 The '``resume``' instruction resumes propagation of an existing
3602 (in-flight) exception whose unwinding was interrupted with a
3603 :ref:`landingpad <i_landingpad>` instruction.
3608 .. code-block:: llvm
3610 resume { i8*, i32 } %exn
3614 '``unreachable``' Instruction
3615 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3627 The '``unreachable``' instruction has no defined semantics. This
3628 instruction is used to inform the optimizer that a particular portion of
3629 the code is not reachable. This can be used to indicate that the code
3630 after a no-return function cannot be reached, and other facts.
3635 The '``unreachable``' instruction has no defined semantics.
3642 Binary operators are used to do most of the computation in a program.
3643 They require two operands of the same type, execute an operation on
3644 them, and produce a single value. The operands might represent multiple
3645 data, as is the case with the :ref:`vector <t_vector>` data type. The
3646 result value has the same type as its operands.
3648 There are several different binary operators:
3652 '``add``' Instruction
3653 ^^^^^^^^^^^^^^^^^^^^^
3660 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3661 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3662 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3663 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3668 The '``add``' instruction returns the sum of its two operands.
3673 The two arguments to the '``add``' instruction must be
3674 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3675 arguments must have identical types.
3680 The value produced is the integer sum of the two operands.
3682 If the sum has unsigned overflow, the result returned is the
3683 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3686 Because LLVM integers use a two's complement representation, this
3687 instruction is appropriate for both signed and unsigned integers.
3689 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3690 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3691 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3692 unsigned and/or signed overflow, respectively, occurs.
3697 .. code-block:: llvm
3699 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3703 '``fadd``' Instruction
3704 ^^^^^^^^^^^^^^^^^^^^^^
3711 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3716 The '``fadd``' instruction returns the sum of its two operands.
3721 The two arguments to the '``fadd``' instruction must be :ref:`floating
3722 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3723 Both arguments must have identical types.
3728 The value produced is the floating point sum of the two operands. This
3729 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3730 which are optimization hints to enable otherwise unsafe floating point
3736 .. code-block:: llvm
3738 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3740 '``sub``' Instruction
3741 ^^^^^^^^^^^^^^^^^^^^^
3748 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3749 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3750 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3751 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3756 The '``sub``' instruction returns the difference of its two operands.
3758 Note that the '``sub``' instruction is used to represent the '``neg``'
3759 instruction present in most other intermediate representations.
3764 The two arguments to the '``sub``' instruction must be
3765 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3766 arguments must have identical types.
3771 The value produced is the integer difference of the two operands.
3773 If the difference has unsigned overflow, the result returned is the
3774 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3777 Because LLVM integers use a two's complement representation, this
3778 instruction is appropriate for both signed and unsigned integers.
3780 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3781 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3782 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3783 unsigned and/or signed overflow, respectively, occurs.
3788 .. code-block:: llvm
3790 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3791 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3795 '``fsub``' Instruction
3796 ^^^^^^^^^^^^^^^^^^^^^^
3803 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3808 The '``fsub``' instruction returns the difference of its two operands.
3810 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3811 instruction present in most other intermediate representations.
3816 The two arguments to the '``fsub``' instruction must be :ref:`floating
3817 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3818 Both arguments must have identical types.
3823 The value produced is the floating point difference of the two operands.
3824 This instruction can also take any number of :ref:`fast-math
3825 flags <fastmath>`, which are optimization hints to enable otherwise
3826 unsafe floating point optimizations:
3831 .. code-block:: llvm
3833 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3834 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3836 '``mul``' Instruction
3837 ^^^^^^^^^^^^^^^^^^^^^
3844 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3845 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3846 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3847 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3852 The '``mul``' instruction returns the product of its two operands.
3857 The two arguments to the '``mul``' instruction must be
3858 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3859 arguments must have identical types.
3864 The value produced is the integer product of the two operands.
3866 If the result of the multiplication has unsigned overflow, the result
3867 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3868 bit width of the result.
3870 Because LLVM integers use a two's complement representation, and the
3871 result is the same width as the operands, this instruction returns the
3872 correct result for both signed and unsigned integers. If a full product
3873 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3874 sign-extended or zero-extended as appropriate to the width of the full
3877 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3878 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3879 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3880 unsigned and/or signed overflow, respectively, occurs.
3885 .. code-block:: llvm
3887 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3891 '``fmul``' Instruction
3892 ^^^^^^^^^^^^^^^^^^^^^^
3899 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3904 The '``fmul``' instruction returns the product of its two operands.
3909 The two arguments to the '``fmul``' instruction must be :ref:`floating
3910 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3911 Both arguments must have identical types.
3916 The value produced is the floating point product of the two operands.
3917 This instruction can also take any number of :ref:`fast-math
3918 flags <fastmath>`, which are optimization hints to enable otherwise
3919 unsafe floating point optimizations:
3924 .. code-block:: llvm
3926 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3928 '``udiv``' Instruction
3929 ^^^^^^^^^^^^^^^^^^^^^^
3936 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3937 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3942 The '``udiv``' instruction returns the quotient of its two operands.
3947 The two arguments to the '``udiv``' instruction must be
3948 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3949 arguments must have identical types.
3954 The value produced is the unsigned integer quotient of the two operands.
3956 Note that unsigned integer division and signed integer division are
3957 distinct operations; for signed integer division, use '``sdiv``'.
3959 Division by zero leads to undefined behavior.
3961 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3962 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3963 such, "((a udiv exact b) mul b) == a").
3968 .. code-block:: llvm
3970 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3972 '``sdiv``' Instruction
3973 ^^^^^^^^^^^^^^^^^^^^^^
3980 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3981 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3986 The '``sdiv``' instruction returns the quotient of its two operands.
3991 The two arguments to the '``sdiv``' instruction must be
3992 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3993 arguments must have identical types.
3998 The value produced is the signed integer quotient of the two operands
3999 rounded towards zero.
4001 Note that signed integer division and unsigned integer division are
4002 distinct operations; for unsigned integer division, use '``udiv``'.
4004 Division by zero leads to undefined behavior. Overflow also leads to
4005 undefined behavior; this is a rare case, but can occur, for example, by
4006 doing a 32-bit division of -2147483648 by -1.
4008 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4009 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4014 .. code-block:: llvm
4016 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
4020 '``fdiv``' Instruction
4021 ^^^^^^^^^^^^^^^^^^^^^^
4028 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4033 The '``fdiv``' instruction returns the quotient of its two operands.
4038 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4039 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4040 Both arguments must have identical types.
4045 The value produced is the floating point quotient of the two operands.
4046 This instruction can also take any number of :ref:`fast-math
4047 flags <fastmath>`, which are optimization hints to enable otherwise
4048 unsafe floating point optimizations:
4053 .. code-block:: llvm
4055 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
4057 '``urem``' Instruction
4058 ^^^^^^^^^^^^^^^^^^^^^^
4065 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
4070 The '``urem``' instruction returns the remainder from the unsigned
4071 division of its two arguments.
4076 The two arguments to the '``urem``' instruction must be
4077 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4078 arguments must have identical types.
4083 This instruction returns the unsigned integer *remainder* of a division.
4084 This instruction always performs an unsigned division to get the
4087 Note that unsigned integer remainder and signed integer remainder are
4088 distinct operations; for signed integer remainder, use '``srem``'.
4090 Taking the remainder of a division by zero leads to undefined behavior.
4095 .. code-block:: llvm
4097 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
4099 '``srem``' Instruction
4100 ^^^^^^^^^^^^^^^^^^^^^^
4107 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
4112 The '``srem``' instruction returns the remainder from the signed
4113 division of its two operands. This instruction can also take
4114 :ref:`vector <t_vector>` versions of the values in which case the elements
4120 The two arguments to the '``srem``' instruction must be
4121 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4122 arguments must have identical types.
4127 This instruction returns the *remainder* of a division (where the result
4128 is either zero or has the same sign as the dividend, ``op1``), not the
4129 *modulo* operator (where the result is either zero or has the same sign
4130 as the divisor, ``op2``) of a value. For more information about the
4131 difference, see `The Math
4132 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4133 table of how this is implemented in various languages, please see
4135 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4137 Note that signed integer remainder and unsigned integer remainder are
4138 distinct operations; for unsigned integer remainder, use '``urem``'.
4140 Taking the remainder of a division by zero leads to undefined behavior.
4141 Overflow also leads to undefined behavior; this is a rare case, but can
4142 occur, for example, by taking the remainder of a 32-bit division of
4143 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4144 rule lets srem be implemented using instructions that return both the
4145 result of the division and the remainder.)
4150 .. code-block:: llvm
4152 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4156 '``frem``' Instruction
4157 ^^^^^^^^^^^^^^^^^^^^^^
4164 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4169 The '``frem``' instruction returns the remainder from the division of
4175 The two arguments to the '``frem``' instruction must be :ref:`floating
4176 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4177 Both arguments must have identical types.
4182 This instruction returns the *remainder* of a division. The remainder
4183 has the same sign as the dividend. This instruction can also take any
4184 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4185 to enable otherwise unsafe floating point optimizations:
4190 .. code-block:: llvm
4192 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4196 Bitwise Binary Operations
4197 -------------------------
4199 Bitwise binary operators are used to do various forms of bit-twiddling
4200 in a program. They are generally very efficient instructions and can
4201 commonly be strength reduced from other instructions. They require two
4202 operands of the same type, execute an operation on them, and produce a
4203 single value. The resulting value is the same type as its operands.
4205 '``shl``' Instruction
4206 ^^^^^^^^^^^^^^^^^^^^^
4213 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4214 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4215 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4216 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4221 The '``shl``' instruction returns the first operand shifted to the left
4222 a specified number of bits.
4227 Both arguments to the '``shl``' instruction must be the same
4228 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4229 '``op2``' is treated as an unsigned value.
4234 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4235 where ``n`` is the width of the result. If ``op2`` is (statically or
4236 dynamically) negative or equal to or larger than the number of bits in
4237 ``op1``, the result is undefined. If the arguments are vectors, each
4238 vector element of ``op1`` is shifted by the corresponding shift amount
4241 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4242 value <poisonvalues>` if it shifts out any non-zero bits. If the
4243 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4244 value <poisonvalues>` if it shifts out any bits that disagree with the
4245 resultant sign bit. As such, NUW/NSW have the same semantics as they
4246 would if the shift were expressed as a mul instruction with the same
4247 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4252 .. code-block:: llvm
4254 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4255 <result> = shl i32 4, 2 ; yields {i32}: 16
4256 <result> = shl i32 1, 10 ; yields {i32}: 1024
4257 <result> = shl i32 1, 32 ; undefined
4258 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4260 '``lshr``' Instruction
4261 ^^^^^^^^^^^^^^^^^^^^^^
4268 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4269 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4274 The '``lshr``' instruction (logical shift right) returns the first
4275 operand shifted to the right a specified number of bits with zero fill.
4280 Both arguments to the '``lshr``' instruction must be the same
4281 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4282 '``op2``' is treated as an unsigned value.
4287 This instruction always performs a logical shift right operation. The
4288 most significant bits of the result will be filled with zero bits after
4289 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4290 than the number of bits in ``op1``, the result is undefined. If the
4291 arguments are vectors, each vector element of ``op1`` is shifted by the
4292 corresponding shift amount in ``op2``.
4294 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4295 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4301 .. code-block:: llvm
4303 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4304 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4305 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4306 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4307 <result> = lshr i32 1, 32 ; undefined
4308 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4310 '``ashr``' Instruction
4311 ^^^^^^^^^^^^^^^^^^^^^^
4318 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4319 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4324 The '``ashr``' instruction (arithmetic shift right) returns the first
4325 operand shifted to the right a specified number of bits with sign
4331 Both arguments to the '``ashr``' instruction must be the same
4332 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4333 '``op2``' is treated as an unsigned value.
4338 This instruction always performs an arithmetic shift right operation,
4339 The most significant bits of the result will be filled with the sign bit
4340 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4341 than the number of bits in ``op1``, the result is undefined. If the
4342 arguments are vectors, each vector element of ``op1`` is shifted by the
4343 corresponding shift amount in ``op2``.
4345 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4346 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4352 .. code-block:: llvm
4354 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4355 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4356 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4357 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4358 <result> = ashr i32 1, 32 ; undefined
4359 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4361 '``and``' Instruction
4362 ^^^^^^^^^^^^^^^^^^^^^
4369 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4374 The '``and``' instruction returns the bitwise logical and of its two
4380 The two arguments to the '``and``' instruction must be
4381 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4382 arguments must have identical types.
4387 The truth table used for the '``and``' instruction is:
4404 .. code-block:: llvm
4406 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4407 <result> = and i32 15, 40 ; yields {i32}:result = 8
4408 <result> = and i32 4, 8 ; yields {i32}:result = 0
4410 '``or``' Instruction
4411 ^^^^^^^^^^^^^^^^^^^^
4418 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4423 The '``or``' instruction returns the bitwise logical inclusive or of its
4429 The two arguments to the '``or``' instruction must be
4430 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4431 arguments must have identical types.
4436 The truth table used for the '``or``' instruction is:
4455 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4456 <result> = or i32 15, 40 ; yields {i32}:result = 47
4457 <result> = or i32 4, 8 ; yields {i32}:result = 12
4459 '``xor``' Instruction
4460 ^^^^^^^^^^^^^^^^^^^^^
4467 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4472 The '``xor``' instruction returns the bitwise logical exclusive or of
4473 its two operands. The ``xor`` is used to implement the "one's
4474 complement" operation, which is the "~" operator in C.
4479 The two arguments to the '``xor``' instruction must be
4480 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4481 arguments must have identical types.
4486 The truth table used for the '``xor``' instruction is:
4503 .. code-block:: llvm
4505 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4506 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4507 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4508 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4513 LLVM supports several instructions to represent vector operations in a
4514 target-independent manner. These instructions cover the element-access
4515 and vector-specific operations needed to process vectors effectively.
4516 While LLVM does directly support these vector operations, many
4517 sophisticated algorithms will want to use target-specific intrinsics to
4518 take full advantage of a specific target.
4520 .. _i_extractelement:
4522 '``extractelement``' Instruction
4523 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4530 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
4535 The '``extractelement``' instruction extracts a single scalar element
4536 from a vector at a specified index.
4541 The first operand of an '``extractelement``' instruction is a value of
4542 :ref:`vector <t_vector>` type. The second operand is an index indicating
4543 the position from which to extract the element. The index may be a
4544 variable of any integer type.
4549 The result is a scalar of the same type as the element type of ``val``.
4550 Its value is the value at position ``idx`` of ``val``. If ``idx``
4551 exceeds the length of ``val``, the results are undefined.
4556 .. code-block:: llvm
4558 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4560 .. _i_insertelement:
4562 '``insertelement``' Instruction
4563 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4570 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
4575 The '``insertelement``' instruction inserts a scalar element into a
4576 vector at a specified index.
4581 The first operand of an '``insertelement``' instruction is a value of
4582 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4583 type must equal the element type of the first operand. The third operand
4584 is an index indicating the position at which to insert the value. The
4585 index may be a variable of any integer type.
4590 The result is a vector of the same type as ``val``. Its element values
4591 are those of ``val`` except at position ``idx``, where it gets the value
4592 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4598 .. code-block:: llvm
4600 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4602 .. _i_shufflevector:
4604 '``shufflevector``' Instruction
4605 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4612 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4617 The '``shufflevector``' instruction constructs a permutation of elements
4618 from two input vectors, returning a vector with the same element type as
4619 the input and length that is the same as the shuffle mask.
4624 The first two operands of a '``shufflevector``' instruction are vectors
4625 with the same type. The third argument is a shuffle mask whose element
4626 type is always 'i32'. The result of the instruction is a vector whose
4627 length is the same as the shuffle mask and whose element type is the
4628 same as the element type of the first two operands.
4630 The shuffle mask operand is required to be a constant vector with either
4631 constant integer or undef values.
4636 The elements of the two input vectors are numbered from left to right
4637 across both of the vectors. The shuffle mask operand specifies, for each
4638 element of the result vector, which element of the two input vectors the
4639 result element gets. The element selector may be undef (meaning "don't
4640 care") and the second operand may be undef if performing a shuffle from
4646 .. code-block:: llvm
4648 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4649 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4650 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4651 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4652 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4653 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4654 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4655 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4657 Aggregate Operations
4658 --------------------
4660 LLVM supports several instructions for working with
4661 :ref:`aggregate <t_aggregate>` values.
4665 '``extractvalue``' Instruction
4666 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4673 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4678 The '``extractvalue``' instruction extracts the value of a member field
4679 from an :ref:`aggregate <t_aggregate>` value.
4684 The first operand of an '``extractvalue``' instruction is a value of
4685 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4686 constant indices to specify which value to extract in a similar manner
4687 as indices in a '``getelementptr``' instruction.
4689 The major differences to ``getelementptr`` indexing are:
4691 - Since the value being indexed is not a pointer, the first index is
4692 omitted and assumed to be zero.
4693 - At least one index must be specified.
4694 - Not only struct indices but also array indices must be in bounds.
4699 The result is the value at the position in the aggregate specified by
4705 .. code-block:: llvm
4707 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4711 '``insertvalue``' Instruction
4712 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4719 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4724 The '``insertvalue``' instruction inserts a value into a member field in
4725 an :ref:`aggregate <t_aggregate>` value.
4730 The first operand of an '``insertvalue``' instruction is a value of
4731 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4732 a first-class value to insert. The following operands are constant
4733 indices indicating the position at which to insert the value in a
4734 similar manner as indices in a '``extractvalue``' instruction. The value
4735 to insert must have the same type as the value identified by the
4741 The result is an aggregate of the same type as ``val``. Its value is
4742 that of ``val`` except that the value at the position specified by the
4743 indices is that of ``elt``.
4748 .. code-block:: llvm
4750 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4751 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4752 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4756 Memory Access and Addressing Operations
4757 ---------------------------------------
4759 A key design point of an SSA-based representation is how it represents
4760 memory. In LLVM, no memory locations are in SSA form, which makes things
4761 very simple. This section describes how to read, write, and allocate
4766 '``alloca``' Instruction
4767 ^^^^^^^^^^^^^^^^^^^^^^^^
4774 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields {type*}:result
4779 The '``alloca``' instruction allocates memory on the stack frame of the
4780 currently executing function, to be automatically released when this
4781 function returns to its caller. The object is always allocated in the
4782 generic address space (address space zero).
4787 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4788 bytes of memory on the runtime stack, returning a pointer of the
4789 appropriate type to the program. If "NumElements" is specified, it is
4790 the number of elements allocated, otherwise "NumElements" is defaulted
4791 to be one. If a constant alignment is specified, the value result of the
4792 allocation is guaranteed to be aligned to at least that boundary. If not
4793 specified, or if zero, the target can choose to align the allocation on
4794 any convenient boundary compatible with the type.
4796 '``type``' may be any sized type.
4801 Memory is allocated; a pointer is returned. The operation is undefined
4802 if there is insufficient stack space for the allocation. '``alloca``'d
4803 memory is automatically released when the function returns. The
4804 '``alloca``' instruction is commonly used to represent automatic
4805 variables that must have an address available. When the function returns
4806 (either with the ``ret`` or ``resume`` instructions), the memory is
4807 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4808 The order in which memory is allocated (ie., which way the stack grows)
4814 .. code-block:: llvm
4816 %ptr = alloca i32 ; yields {i32*}:ptr
4817 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4818 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4819 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4823 '``load``' Instruction
4824 ^^^^^^^^^^^^^^^^^^^^^^
4831 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4832 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4833 !<index> = !{ i32 1 }
4838 The '``load``' instruction is used to read from memory.
4843 The argument to the ``load`` instruction specifies the memory address
4844 from which to load. The pointer must point to a :ref:`first
4845 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4846 then the optimizer is not allowed to modify the number or order of
4847 execution of this ``load`` with other :ref:`volatile
4848 operations <volatile>`.
4850 If the ``load`` is marked as ``atomic``, it takes an extra
4851 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4852 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4853 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4854 when they may see multiple atomic stores. The type of the pointee must
4855 be an integer type whose bit width is a power of two greater than or
4856 equal to eight and less than or equal to a target-specific size limit.
4857 ``align`` must be explicitly specified on atomic loads, and the load has
4858 undefined behavior if the alignment is not set to a value which is at
4859 least the size in bytes of the pointee. ``!nontemporal`` does not have
4860 any defined semantics for atomic loads.
4862 The optional constant ``align`` argument specifies the alignment of the
4863 operation (that is, the alignment of the memory address). A value of 0
4864 or an omitted ``align`` argument means that the operation has the ABI
4865 alignment for the target. It is the responsibility of the code emitter
4866 to ensure that the alignment information is correct. Overestimating the
4867 alignment results in undefined behavior. Underestimating the alignment
4868 may produce less efficient code. An alignment of 1 is always safe.
4870 The optional ``!nontemporal`` metadata must reference a single
4871 metadata name ``<index>`` corresponding to a metadata node with one
4872 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4873 metadata on the instruction tells the optimizer and code generator
4874 that this load is not expected to be reused in the cache. The code
4875 generator may select special instructions to save cache bandwidth, such
4876 as the ``MOVNT`` instruction on x86.
4878 The optional ``!invariant.load`` metadata must reference a single
4879 metadata name ``<index>`` corresponding to a metadata node with no
4880 entries. The existence of the ``!invariant.load`` metadata on the
4881 instruction tells the optimizer and code generator that this load
4882 address points to memory which does not change value during program
4883 execution. The optimizer may then move this load around, for example, by
4884 hoisting it out of loops using loop invariant code motion.
4889 The location of memory pointed to is loaded. If the value being loaded
4890 is of scalar type then the number of bytes read does not exceed the
4891 minimum number of bytes needed to hold all bits of the type. For
4892 example, loading an ``i24`` reads at most three bytes. When loading a
4893 value of a type like ``i20`` with a size that is not an integral number
4894 of bytes, the result is undefined if the value was not originally
4895 written using a store of the same type.
4900 .. code-block:: llvm
4902 %ptr = alloca i32 ; yields {i32*}:ptr
4903 store i32 3, i32* %ptr ; yields {void}
4904 %val = load i32* %ptr ; yields {i32}:val = i32 3
4908 '``store``' Instruction
4909 ^^^^^^^^^^^^^^^^^^^^^^^
4916 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4917 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4922 The '``store``' instruction is used to write to memory.
4927 There are two arguments to the ``store`` instruction: a value to store
4928 and an address at which to store it. The type of the ``<pointer>``
4929 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4930 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4931 then the optimizer is not allowed to modify the number or order of
4932 execution of this ``store`` with other :ref:`volatile
4933 operations <volatile>`.
4935 If the ``store`` is marked as ``atomic``, it takes an extra
4936 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4937 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4938 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4939 when they may see multiple atomic stores. The type of the pointee must
4940 be an integer type whose bit width is a power of two greater than or
4941 equal to eight and less than or equal to a target-specific size limit.
4942 ``align`` must be explicitly specified on atomic stores, and the store
4943 has undefined behavior if the alignment is not set to a value which is
4944 at least the size in bytes of the pointee. ``!nontemporal`` does not
4945 have any defined semantics for atomic stores.
4947 The optional constant ``align`` argument specifies the alignment of the
4948 operation (that is, the alignment of the memory address). A value of 0
4949 or an omitted ``align`` argument means that the operation has the ABI
4950 alignment for the target. It is the responsibility of the code emitter
4951 to ensure that the alignment information is correct. Overestimating the
4952 alignment results in undefined behavior. Underestimating the
4953 alignment may produce less efficient code. An alignment of 1 is always
4956 The optional ``!nontemporal`` metadata must reference a single metadata
4957 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4958 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4959 tells the optimizer and code generator that this load is not expected to
4960 be reused in the cache. The code generator may select special
4961 instructions to save cache bandwidth, such as the MOVNT instruction on
4967 The contents of memory are updated to contain ``<value>`` at the
4968 location specified by the ``<pointer>`` operand. If ``<value>`` is
4969 of scalar type then the number of bytes written does not exceed the
4970 minimum number of bytes needed to hold all bits of the type. For
4971 example, storing an ``i24`` writes at most three bytes. When writing a
4972 value of a type like ``i20`` with a size that is not an integral number
4973 of bytes, it is unspecified what happens to the extra bits that do not
4974 belong to the type, but they will typically be overwritten.
4979 .. code-block:: llvm
4981 %ptr = alloca i32 ; yields {i32*}:ptr
4982 store i32 3, i32* %ptr ; yields {void}
4983 %val = load i32* %ptr ; yields {i32}:val = i32 3
4987 '``fence``' Instruction
4988 ^^^^^^^^^^^^^^^^^^^^^^^
4995 fence [singlethread] <ordering> ; yields {void}
5000 The '``fence``' instruction is used to introduce happens-before edges
5006 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5007 defines what *synchronizes-with* edges they add. They can only be given
5008 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5013 A fence A which has (at least) ``release`` ordering semantics
5014 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5015 semantics if and only if there exist atomic operations X and Y, both
5016 operating on some atomic object M, such that A is sequenced before X, X
5017 modifies M (either directly or through some side effect of a sequence
5018 headed by X), Y is sequenced before B, and Y observes M. This provides a
5019 *happens-before* dependency between A and B. Rather than an explicit
5020 ``fence``, one (but not both) of the atomic operations X or Y might
5021 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5022 still *synchronize-with* the explicit ``fence`` and establish the
5023 *happens-before* edge.
5025 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5026 ``acquire`` and ``release`` semantics specified above, participates in
5027 the global program order of other ``seq_cst`` operations and/or fences.
5029 The optional ":ref:`singlethread <singlethread>`" argument specifies
5030 that the fence only synchronizes with other fences in the same thread.
5031 (This is useful for interacting with signal handlers.)
5036 .. code-block:: llvm
5038 fence acquire ; yields {void}
5039 fence singlethread seq_cst ; yields {void}
5043 '``cmpxchg``' Instruction
5044 ^^^^^^^^^^^^^^^^^^^^^^^^^
5051 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields {ty}
5056 The '``cmpxchg``' instruction is used to atomically modify memory. It
5057 loads a value in memory and compares it to a given value. If they are
5058 equal, it stores a new value into the memory.
5063 There are three arguments to the '``cmpxchg``' instruction: an address
5064 to operate on, a value to compare to the value currently be at that
5065 address, and a new value to place at that address if the compared values
5066 are equal. The type of '<cmp>' must be an integer type whose bit width
5067 is a power of two greater than or equal to eight and less than or equal
5068 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5069 type, and the type of '<pointer>' must be a pointer to that type. If the
5070 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5071 to modify the number or order of execution of this ``cmpxchg`` with
5072 other :ref:`volatile operations <volatile>`.
5074 The success and failure :ref:`ordering <ordering>` arguments specify how this
5075 ``cmpxchg`` synchronizes with other atomic operations. The both ordering
5076 parameters must be at least ``monotonic``, the ordering constraint on failure
5077 must be no stronger than that on success, and the failure ordering cannot be
5078 either ``release`` or ``acq_rel``.
5080 The optional "``singlethread``" argument declares that the ``cmpxchg``
5081 is only atomic with respect to code (usually signal handlers) running in
5082 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5083 respect to all other code in the system.
5085 The pointer passed into cmpxchg must have alignment greater than or
5086 equal to the size in memory of the operand.
5091 The contents of memory at the location specified by the '``<pointer>``'
5092 operand is read and compared to '``<cmp>``'; if the read value is the
5093 equal, '``<new>``' is written. The original value at the location is
5096 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5097 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5098 load with an ordering parameter determined the second ordering parameter.
5103 .. code-block:: llvm
5106 %orig = atomic load i32* %ptr unordered ; yields {i32}
5110 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5111 %squared = mul i32 %cmp, %cmp
5112 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields {i32}
5113 %success = icmp eq i32 %cmp, %old
5114 br i1 %success, label %done, label %loop
5121 '``atomicrmw``' Instruction
5122 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5129 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
5134 The '``atomicrmw``' instruction is used to atomically modify memory.
5139 There are three arguments to the '``atomicrmw``' instruction: an
5140 operation to apply, an address whose value to modify, an argument to the
5141 operation. The operation must be one of the following keywords:
5155 The type of '<value>' must be an integer type whose bit width is a power
5156 of two greater than or equal to eight and less than or equal to a
5157 target-specific size limit. The type of the '``<pointer>``' operand must
5158 be a pointer to that type. If the ``atomicrmw`` is marked as
5159 ``volatile``, then the optimizer is not allowed to modify the number or
5160 order of execution of this ``atomicrmw`` with other :ref:`volatile
5161 operations <volatile>`.
5166 The contents of memory at the location specified by the '``<pointer>``'
5167 operand are atomically read, modified, and written back. The original
5168 value at the location is returned. The modification is specified by the
5171 - xchg: ``*ptr = val``
5172 - add: ``*ptr = *ptr + val``
5173 - sub: ``*ptr = *ptr - val``
5174 - and: ``*ptr = *ptr & val``
5175 - nand: ``*ptr = ~(*ptr & val)``
5176 - or: ``*ptr = *ptr | val``
5177 - xor: ``*ptr = *ptr ^ val``
5178 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5179 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5180 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5182 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5188 .. code-block:: llvm
5190 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5192 .. _i_getelementptr:
5194 '``getelementptr``' Instruction
5195 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5202 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5203 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5204 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5209 The '``getelementptr``' instruction is used to get the address of a
5210 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5211 address calculation only and does not access memory.
5216 The first argument is always a pointer or a vector of pointers, and
5217 forms the basis of the calculation. The remaining arguments are indices
5218 that indicate which of the elements of the aggregate object are indexed.
5219 The interpretation of each index is dependent on the type being indexed
5220 into. The first index always indexes the pointer value given as the
5221 first argument, the second index indexes a value of the type pointed to
5222 (not necessarily the value directly pointed to, since the first index
5223 can be non-zero), etc. The first type indexed into must be a pointer
5224 value, subsequent types can be arrays, vectors, and structs. Note that
5225 subsequent types being indexed into can never be pointers, since that
5226 would require loading the pointer before continuing calculation.
5228 The type of each index argument depends on the type it is indexing into.
5229 When indexing into a (optionally packed) structure, only ``i32`` integer
5230 **constants** are allowed (when using a vector of indices they must all
5231 be the **same** ``i32`` integer constant). When indexing into an array,
5232 pointer or vector, integers of any width are allowed, and they are not
5233 required to be constant. These integers are treated as signed values
5236 For example, let's consider a C code fragment and how it gets compiled
5252 int *foo(struct ST *s) {
5253 return &s[1].Z.B[5][13];
5256 The LLVM code generated by Clang is:
5258 .. code-block:: llvm
5260 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5261 %struct.ST = type { i32, double, %struct.RT }
5263 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5265 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5272 In the example above, the first index is indexing into the
5273 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5274 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5275 indexes into the third element of the structure, yielding a
5276 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5277 structure. The third index indexes into the second element of the
5278 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5279 dimensions of the array are subscripted into, yielding an '``i32``'
5280 type. The '``getelementptr``' instruction returns a pointer to this
5281 element, thus computing a value of '``i32*``' type.
5283 Note that it is perfectly legal to index partially through a structure,
5284 returning a pointer to an inner element. Because of this, the LLVM code
5285 for the given testcase is equivalent to:
5287 .. code-block:: llvm
5289 define i32* @foo(%struct.ST* %s) {
5290 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5291 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5292 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5293 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5294 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5298 If the ``inbounds`` keyword is present, the result value of the
5299 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5300 pointer is not an *in bounds* address of an allocated object, or if any
5301 of the addresses that would be formed by successive addition of the
5302 offsets implied by the indices to the base address with infinitely
5303 precise signed arithmetic are not an *in bounds* address of that
5304 allocated object. The *in bounds* addresses for an allocated object are
5305 all the addresses that point into the object, plus the address one byte
5306 past the end. In cases where the base is a vector of pointers the
5307 ``inbounds`` keyword applies to each of the computations element-wise.
5309 If the ``inbounds`` keyword is not present, the offsets are added to the
5310 base address with silently-wrapping two's complement arithmetic. If the
5311 offsets have a different width from the pointer, they are sign-extended
5312 or truncated to the width of the pointer. The result value of the
5313 ``getelementptr`` may be outside the object pointed to by the base
5314 pointer. The result value may not necessarily be used to access memory
5315 though, even if it happens to point into allocated storage. See the
5316 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5319 The getelementptr instruction is often confusing. For some more insight
5320 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5325 .. code-block:: llvm
5327 ; yields [12 x i8]*:aptr
5328 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5330 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5332 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5334 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5336 In cases where the pointer argument is a vector of pointers, each index
5337 must be a vector with the same number of elements. For example:
5339 .. code-block:: llvm
5341 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5343 Conversion Operations
5344 ---------------------
5346 The instructions in this category are the conversion instructions
5347 (casting) which all take a single operand and a type. They perform
5348 various bit conversions on the operand.
5350 '``trunc .. to``' Instruction
5351 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5358 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5363 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5368 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5369 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5370 of the same number of integers. The bit size of the ``value`` must be
5371 larger than the bit size of the destination type, ``ty2``. Equal sized
5372 types are not allowed.
5377 The '``trunc``' instruction truncates the high order bits in ``value``
5378 and converts the remaining bits to ``ty2``. Since the source size must
5379 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5380 It will always truncate bits.
5385 .. code-block:: llvm
5387 %X = trunc i32 257 to i8 ; yields i8:1
5388 %Y = trunc i32 123 to i1 ; yields i1:true
5389 %Z = trunc i32 122 to i1 ; yields i1:false
5390 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5392 '``zext .. to``' Instruction
5393 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5400 <result> = zext <ty> <value> to <ty2> ; yields ty2
5405 The '``zext``' instruction zero extends its operand to type ``ty2``.
5410 The '``zext``' instruction takes a value to cast, and a type to cast it
5411 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5412 the same number of integers. The bit size of the ``value`` must be
5413 smaller than the bit size of the destination type, ``ty2``.
5418 The ``zext`` fills the high order bits of the ``value`` with zero bits
5419 until it reaches the size of the destination type, ``ty2``.
5421 When zero extending from i1, the result will always be either 0 or 1.
5426 .. code-block:: llvm
5428 %X = zext i32 257 to i64 ; yields i64:257
5429 %Y = zext i1 true to i32 ; yields i32:1
5430 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5432 '``sext .. to``' Instruction
5433 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5440 <result> = sext <ty> <value> to <ty2> ; yields ty2
5445 The '``sext``' sign extends ``value`` to the type ``ty2``.
5450 The '``sext``' instruction takes a value to cast, and a type to cast it
5451 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5452 the same number of integers. The bit size of the ``value`` must be
5453 smaller than the bit size of the destination type, ``ty2``.
5458 The '``sext``' instruction performs a sign extension by copying the sign
5459 bit (highest order bit) of the ``value`` until it reaches the bit size
5460 of the type ``ty2``.
5462 When sign extending from i1, the extension always results in -1 or 0.
5467 .. code-block:: llvm
5469 %X = sext i8 -1 to i16 ; yields i16 :65535
5470 %Y = sext i1 true to i32 ; yields i32:-1
5471 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5473 '``fptrunc .. to``' Instruction
5474 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5481 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5486 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5491 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5492 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5493 The size of ``value`` must be larger than the size of ``ty2``. This
5494 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5499 The '``fptrunc``' instruction truncates a ``value`` from a larger
5500 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5501 point <t_floating>` type. If the value cannot fit within the
5502 destination type, ``ty2``, then the results are undefined.
5507 .. code-block:: llvm
5509 %X = fptrunc double 123.0 to float ; yields float:123.0
5510 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5512 '``fpext .. to``' Instruction
5513 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5520 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5525 The '``fpext``' extends a floating point ``value`` to a larger floating
5531 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5532 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5533 to. The source type must be smaller than the destination type.
5538 The '``fpext``' instruction extends the ``value`` from a smaller
5539 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5540 point <t_floating>` type. The ``fpext`` cannot be used to make a
5541 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5542 *no-op cast* for a floating point cast.
5547 .. code-block:: llvm
5549 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5550 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5552 '``fptoui .. to``' Instruction
5553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5560 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5565 The '``fptoui``' converts a floating point ``value`` to its unsigned
5566 integer equivalent of type ``ty2``.
5571 The '``fptoui``' instruction takes a value to cast, which must be a
5572 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5573 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5574 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5575 type with the same number of elements as ``ty``
5580 The '``fptoui``' instruction converts its :ref:`floating
5581 point <t_floating>` operand into the nearest (rounding towards zero)
5582 unsigned integer value. If the value cannot fit in ``ty2``, the results
5588 .. code-block:: llvm
5590 %X = fptoui double 123.0 to i32 ; yields i32:123
5591 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5592 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5594 '``fptosi .. to``' Instruction
5595 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5602 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5607 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5608 ``value`` to type ``ty2``.
5613 The '``fptosi``' instruction takes a value to cast, which must be a
5614 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5615 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5616 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5617 type with the same number of elements as ``ty``
5622 The '``fptosi``' instruction converts its :ref:`floating
5623 point <t_floating>` operand into the nearest (rounding towards zero)
5624 signed integer value. If the value cannot fit in ``ty2``, the results
5630 .. code-block:: llvm
5632 %X = fptosi double -123.0 to i32 ; yields i32:-123
5633 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5634 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5636 '``uitofp .. to``' Instruction
5637 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5644 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5649 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5650 and converts that value to the ``ty2`` type.
5655 The '``uitofp``' instruction takes a value to cast, which must be a
5656 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5657 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5658 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5659 type with the same number of elements as ``ty``
5664 The '``uitofp``' instruction interprets its operand as an unsigned
5665 integer quantity and converts it to the corresponding floating point
5666 value. If the value cannot fit in the floating point value, the results
5672 .. code-block:: llvm
5674 %X = uitofp i32 257 to float ; yields float:257.0
5675 %Y = uitofp i8 -1 to double ; yields double:255.0
5677 '``sitofp .. to``' Instruction
5678 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5685 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5690 The '``sitofp``' instruction regards ``value`` as a signed integer and
5691 converts that value to the ``ty2`` type.
5696 The '``sitofp``' instruction takes a value to cast, which must be a
5697 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5698 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5699 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5700 type with the same number of elements as ``ty``
5705 The '``sitofp``' instruction interprets its operand as a signed integer
5706 quantity and converts it to the corresponding floating point value. If
5707 the value cannot fit in the floating point value, the results are
5713 .. code-block:: llvm
5715 %X = sitofp i32 257 to float ; yields float:257.0
5716 %Y = sitofp i8 -1 to double ; yields double:-1.0
5720 '``ptrtoint .. to``' Instruction
5721 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5728 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5733 The '``ptrtoint``' instruction converts the pointer or a vector of
5734 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5739 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5740 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5741 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5742 a vector of integers type.
5747 The '``ptrtoint``' instruction converts ``value`` to integer type
5748 ``ty2`` by interpreting the pointer value as an integer and either
5749 truncating or zero extending that value to the size of the integer type.
5750 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5751 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5752 the same size, then nothing is done (*no-op cast*) other than a type
5758 .. code-block:: llvm
5760 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5761 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5762 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5766 '``inttoptr .. to``' Instruction
5767 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5774 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5779 The '``inttoptr``' instruction converts an integer ``value`` to a
5780 pointer type, ``ty2``.
5785 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5786 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5792 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5793 applying either a zero extension or a truncation depending on the size
5794 of the integer ``value``. If ``value`` is larger than the size of a
5795 pointer then a truncation is done. If ``value`` is smaller than the size
5796 of a pointer then a zero extension is done. If they are the same size,
5797 nothing is done (*no-op cast*).
5802 .. code-block:: llvm
5804 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5805 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5806 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5807 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5811 '``bitcast .. to``' Instruction
5812 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5819 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5824 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5830 The '``bitcast``' instruction takes a value to cast, which must be a
5831 non-aggregate first class value, and a type to cast it to, which must
5832 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5833 bit sizes of ``value`` and the destination type, ``ty2``, must be
5834 identical. If the source type is a pointer, the destination type must
5835 also be a pointer of the same size. This instruction supports bitwise
5836 conversion of vectors to integers and to vectors of other types (as
5837 long as they have the same size).
5842 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5843 is always a *no-op cast* because no bits change with this
5844 conversion. The conversion is done as if the ``value`` had been stored
5845 to memory and read back as type ``ty2``. Pointer (or vector of
5846 pointers) types may only be converted to other pointer (or vector of
5847 pointers) types with the same address space through this instruction.
5848 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5849 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5854 .. code-block:: llvm
5856 %X = bitcast i8 255 to i8 ; yields i8 :-1
5857 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5858 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5859 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5861 .. _i_addrspacecast:
5863 '``addrspacecast .. to``' Instruction
5864 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5871 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5876 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5877 address space ``n`` to type ``pty2`` in address space ``m``.
5882 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5883 to cast and a pointer type to cast it to, which must have a different
5889 The '``addrspacecast``' instruction converts the pointer value
5890 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5891 value modification, depending on the target and the address space
5892 pair. Pointer conversions within the same address space must be
5893 performed with the ``bitcast`` instruction. Note that if the address space
5894 conversion is legal then both result and operand refer to the same memory
5900 .. code-block:: llvm
5902 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5903 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5904 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5911 The instructions in this category are the "miscellaneous" instructions,
5912 which defy better classification.
5916 '``icmp``' Instruction
5917 ^^^^^^^^^^^^^^^^^^^^^^
5924 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5929 The '``icmp``' instruction returns a boolean value or a vector of
5930 boolean values based on comparison of its two integer, integer vector,
5931 pointer, or pointer vector operands.
5936 The '``icmp``' instruction takes three operands. The first operand is
5937 the condition code indicating the kind of comparison to perform. It is
5938 not a value, just a keyword. The possible condition code are:
5941 #. ``ne``: not equal
5942 #. ``ugt``: unsigned greater than
5943 #. ``uge``: unsigned greater or equal
5944 #. ``ult``: unsigned less than
5945 #. ``ule``: unsigned less or equal
5946 #. ``sgt``: signed greater than
5947 #. ``sge``: signed greater or equal
5948 #. ``slt``: signed less than
5949 #. ``sle``: signed less or equal
5951 The remaining two arguments must be :ref:`integer <t_integer>` or
5952 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5953 must also be identical types.
5958 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5959 code given as ``cond``. The comparison performed always yields either an
5960 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5962 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5963 otherwise. No sign interpretation is necessary or performed.
5964 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5965 otherwise. No sign interpretation is necessary or performed.
5966 #. ``ugt``: interprets the operands as unsigned values and yields
5967 ``true`` if ``op1`` is greater than ``op2``.
5968 #. ``uge``: interprets the operands as unsigned values and yields
5969 ``true`` if ``op1`` is greater than or equal to ``op2``.
5970 #. ``ult``: interprets the operands as unsigned values and yields
5971 ``true`` if ``op1`` is less than ``op2``.
5972 #. ``ule``: interprets the operands as unsigned values and yields
5973 ``true`` if ``op1`` is less than or equal to ``op2``.
5974 #. ``sgt``: interprets the operands as signed values and yields ``true``
5975 if ``op1`` is greater than ``op2``.
5976 #. ``sge``: interprets the operands as signed values and yields ``true``
5977 if ``op1`` is greater than or equal to ``op2``.
5978 #. ``slt``: interprets the operands as signed values and yields ``true``
5979 if ``op1`` is less than ``op2``.
5980 #. ``sle``: interprets the operands as signed values and yields ``true``
5981 if ``op1`` is less than or equal to ``op2``.
5983 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5984 are compared as if they were integers.
5986 If the operands are integer vectors, then they are compared element by
5987 element. The result is an ``i1`` vector with the same number of elements
5988 as the values being compared. Otherwise, the result is an ``i1``.
5993 .. code-block:: llvm
5995 <result> = icmp eq i32 4, 5 ; yields: result=false
5996 <result> = icmp ne float* %X, %X ; yields: result=false
5997 <result> = icmp ult i16 4, 5 ; yields: result=true
5998 <result> = icmp sgt i16 4, 5 ; yields: result=false
5999 <result> = icmp ule i16 -4, 5 ; yields: result=false
6000 <result> = icmp sge i16 4, 5 ; yields: result=false
6002 Note that the code generator does not yet support vector types with the
6003 ``icmp`` instruction.
6007 '``fcmp``' Instruction
6008 ^^^^^^^^^^^^^^^^^^^^^^
6015 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
6020 The '``fcmp``' instruction returns a boolean value or vector of boolean
6021 values based on comparison of its operands.
6023 If the operands are floating point scalars, then the result type is a
6024 boolean (:ref:`i1 <t_integer>`).
6026 If the operands are floating point vectors, then the result type is a
6027 vector of boolean with the same number of elements as the operands being
6033 The '``fcmp``' instruction takes three operands. The first operand is
6034 the condition code indicating the kind of comparison to perform. It is
6035 not a value, just a keyword. The possible condition code are:
6037 #. ``false``: no comparison, always returns false
6038 #. ``oeq``: ordered and equal
6039 #. ``ogt``: ordered and greater than
6040 #. ``oge``: ordered and greater than or equal
6041 #. ``olt``: ordered and less than
6042 #. ``ole``: ordered and less than or equal
6043 #. ``one``: ordered and not equal
6044 #. ``ord``: ordered (no nans)
6045 #. ``ueq``: unordered or equal
6046 #. ``ugt``: unordered or greater than
6047 #. ``uge``: unordered or greater than or equal
6048 #. ``ult``: unordered or less than
6049 #. ``ule``: unordered or less than or equal
6050 #. ``une``: unordered or not equal
6051 #. ``uno``: unordered (either nans)
6052 #. ``true``: no comparison, always returns true
6054 *Ordered* means that neither operand is a QNAN while *unordered* means
6055 that either operand may be a QNAN.
6057 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6058 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6059 type. They must have identical types.
6064 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6065 condition code given as ``cond``. If the operands are vectors, then the
6066 vectors are compared element by element. Each comparison performed
6067 always yields an :ref:`i1 <t_integer>` result, as follows:
6069 #. ``false``: always yields ``false``, regardless of operands.
6070 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6071 is equal to ``op2``.
6072 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6073 is greater than ``op2``.
6074 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6075 is greater than or equal to ``op2``.
6076 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6077 is less than ``op2``.
6078 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6079 is less than or equal to ``op2``.
6080 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6081 is not equal to ``op2``.
6082 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6083 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6085 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6086 greater than ``op2``.
6087 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6088 greater than or equal to ``op2``.
6089 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6091 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6092 less than or equal to ``op2``.
6093 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6094 not equal to ``op2``.
6095 #. ``uno``: yields ``true`` if either operand is a QNAN.
6096 #. ``true``: always yields ``true``, regardless of operands.
6101 .. code-block:: llvm
6103 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6104 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6105 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6106 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6108 Note that the code generator does not yet support vector types with the
6109 ``fcmp`` instruction.
6113 '``phi``' Instruction
6114 ^^^^^^^^^^^^^^^^^^^^^
6121 <result> = phi <ty> [ <val0>, <label0>], ...
6126 The '``phi``' instruction is used to implement the φ node in the SSA
6127 graph representing the function.
6132 The type of the incoming values is specified with the first type field.
6133 After this, the '``phi``' instruction takes a list of pairs as
6134 arguments, with one pair for each predecessor basic block of the current
6135 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6136 the value arguments to the PHI node. Only labels may be used as the
6139 There must be no non-phi instructions between the start of a basic block
6140 and the PHI instructions: i.e. PHI instructions must be first in a basic
6143 For the purposes of the SSA form, the use of each incoming value is
6144 deemed to occur on the edge from the corresponding predecessor block to
6145 the current block (but after any definition of an '``invoke``'
6146 instruction's return value on the same edge).
6151 At runtime, the '``phi``' instruction logically takes on the value
6152 specified by the pair corresponding to the predecessor basic block that
6153 executed just prior to the current block.
6158 .. code-block:: llvm
6160 Loop: ; Infinite loop that counts from 0 on up...
6161 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6162 %nextindvar = add i32 %indvar, 1
6167 '``select``' Instruction
6168 ^^^^^^^^^^^^^^^^^^^^^^^^
6175 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6177 selty is either i1 or {<N x i1>}
6182 The '``select``' instruction is used to choose one value based on a
6183 condition, without IR-level branching.
6188 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6189 values indicating the condition, and two values of the same :ref:`first
6190 class <t_firstclass>` type. If the val1/val2 are vectors and the
6191 condition is a scalar, then entire vectors are selected, not individual
6197 If the condition is an i1 and it evaluates to 1, the instruction returns
6198 the first value argument; otherwise, it returns the second value
6201 If the condition is a vector of i1, then the value arguments must be
6202 vectors of the same size, and the selection is done element by element.
6207 .. code-block:: llvm
6209 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6213 '``call``' Instruction
6214 ^^^^^^^^^^^^^^^^^^^^^^
6221 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6226 The '``call``' instruction represents a simple function call.
6231 This instruction requires several arguments:
6233 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
6234 should perform tail call optimization. The ``tail`` marker is a hint that
6235 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
6236 means that the call must be tail call optimized in order for the program to
6237 be correct. The ``musttail`` marker provides these guarantees:
6239 #. The call will not cause unbounded stack growth if it is part of a
6240 recursive cycle in the call graph.
6241 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
6244 Both markers imply that the callee does not access allocas or varargs from
6245 the caller. Calls marked ``musttail`` must obey the following additional
6248 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
6249 or a pointer bitcast followed by a ret instruction.
6250 - The ret instruction must return the (possibly bitcasted) value
6251 produced by the call or void.
6252 - The caller and callee prototypes must match. Pointer types of
6253 parameters or return types may differ in pointee type, but not
6255 - The calling conventions of the caller and callee must match.
6256 - All ABI-impacting function attributes, such as sret, byval, inreg,
6257 returned, and inalloca, must match.
6259 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
6260 the following conditions are met:
6262 - Caller and callee both have the calling convention ``fastcc``.
6263 - The call is in tail position (ret immediately follows call and ret
6264 uses value of call or is void).
6265 - Option ``-tailcallopt`` is enabled, or
6266 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6267 - `Platform specific constraints are
6268 met. <CodeGenerator.html#tailcallopt>`_
6270 #. The optional "cconv" marker indicates which :ref:`calling
6271 convention <callingconv>` the call should use. If none is
6272 specified, the call defaults to using C calling conventions. The
6273 calling convention of the call must match the calling convention of
6274 the target function, or else the behavior is undefined.
6275 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6276 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6278 #. '``ty``': the type of the call instruction itself which is also the
6279 type of the return value. Functions that return no value are marked
6281 #. '``fnty``': shall be the signature of the pointer to function value
6282 being invoked. The argument types must match the types implied by
6283 this signature. This type can be omitted if the function is not
6284 varargs and if the function type does not return a pointer to a
6286 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6287 be invoked. In most cases, this is a direct function invocation, but
6288 indirect ``call``'s are just as possible, calling an arbitrary pointer
6290 #. '``function args``': argument list whose types match the function
6291 signature argument types and parameter attributes. All arguments must
6292 be of :ref:`first class <t_firstclass>` type. If the function signature
6293 indicates the function accepts a variable number of arguments, the
6294 extra arguments can be specified.
6295 #. The optional :ref:`function attributes <fnattrs>` list. Only
6296 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6297 attributes are valid here.
6302 The '``call``' instruction is used to cause control flow to transfer to
6303 a specified function, with its incoming arguments bound to the specified
6304 values. Upon a '``ret``' instruction in the called function, control
6305 flow continues with the instruction after the function call, and the
6306 return value of the function is bound to the result argument.
6311 .. code-block:: llvm
6313 %retval = call i32 @test(i32 %argc)
6314 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6315 %X = tail call i32 @foo() ; yields i32
6316 %Y = tail call fastcc i32 @foo() ; yields i32
6317 call void %foo(i8 97 signext)
6319 %struct.A = type { i32, i8 }
6320 %r = call %struct.A @foo() ; yields { 32, i8 }
6321 %gr = extractvalue %struct.A %r, 0 ; yields i32
6322 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6323 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6324 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6326 llvm treats calls to some functions with names and arguments that match
6327 the standard C99 library as being the C99 library functions, and may
6328 perform optimizations or generate code for them under that assumption.
6329 This is something we'd like to change in the future to provide better
6330 support for freestanding environments and non-C-based languages.
6334 '``va_arg``' Instruction
6335 ^^^^^^^^^^^^^^^^^^^^^^^^
6342 <resultval> = va_arg <va_list*> <arglist>, <argty>
6347 The '``va_arg``' instruction is used to access arguments passed through
6348 the "variable argument" area of a function call. It is used to implement
6349 the ``va_arg`` macro in C.
6354 This instruction takes a ``va_list*`` value and the type of the
6355 argument. It returns a value of the specified argument type and
6356 increments the ``va_list`` to point to the next argument. The actual
6357 type of ``va_list`` is target specific.
6362 The '``va_arg``' instruction loads an argument of the specified type
6363 from the specified ``va_list`` and causes the ``va_list`` to point to
6364 the next argument. For more information, see the variable argument
6365 handling :ref:`Intrinsic Functions <int_varargs>`.
6367 It is legal for this instruction to be called in a function which does
6368 not take a variable number of arguments, for example, the ``vfprintf``
6371 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6372 function <intrinsics>` because it takes a type as an argument.
6377 See the :ref:`variable argument processing <int_varargs>` section.
6379 Note that the code generator does not yet fully support va\_arg on many
6380 targets. Also, it does not currently support va\_arg with aggregate
6381 types on any target.
6385 '``landingpad``' Instruction
6386 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6393 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6394 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6396 <clause> := catch <type> <value>
6397 <clause> := filter <array constant type> <array constant>
6402 The '``landingpad``' instruction is used by `LLVM's exception handling
6403 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6404 is a landing pad --- one where the exception lands, and corresponds to the
6405 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6406 defines values supplied by the personality function (``pers_fn``) upon
6407 re-entry to the function. The ``resultval`` has the type ``resultty``.
6412 This instruction takes a ``pers_fn`` value. This is the personality
6413 function associated with the unwinding mechanism. The optional
6414 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6416 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6417 contains the global variable representing the "type" that may be caught
6418 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6419 clause takes an array constant as its argument. Use
6420 "``[0 x i8**] undef``" for a filter which cannot throw. The
6421 '``landingpad``' instruction must contain *at least* one ``clause`` or
6422 the ``cleanup`` flag.
6427 The '``landingpad``' instruction defines the values which are set by the
6428 personality function (``pers_fn``) upon re-entry to the function, and
6429 therefore the "result type" of the ``landingpad`` instruction. As with
6430 calling conventions, how the personality function results are
6431 represented in LLVM IR is target specific.
6433 The clauses are applied in order from top to bottom. If two
6434 ``landingpad`` instructions are merged together through inlining, the
6435 clauses from the calling function are appended to the list of clauses.
6436 When the call stack is being unwound due to an exception being thrown,
6437 the exception is compared against each ``clause`` in turn. If it doesn't
6438 match any of the clauses, and the ``cleanup`` flag is not set, then
6439 unwinding continues further up the call stack.
6441 The ``landingpad`` instruction has several restrictions:
6443 - A landing pad block is a basic block which is the unwind destination
6444 of an '``invoke``' instruction.
6445 - A landing pad block must have a '``landingpad``' instruction as its
6446 first non-PHI instruction.
6447 - There can be only one '``landingpad``' instruction within the landing
6449 - A basic block that is not a landing pad block may not include a
6450 '``landingpad``' instruction.
6451 - All '``landingpad``' instructions in a function must have the same
6452 personality function.
6457 .. code-block:: llvm
6459 ;; A landing pad which can catch an integer.
6460 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6462 ;; A landing pad that is a cleanup.
6463 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6465 ;; A landing pad which can catch an integer and can only throw a double.
6466 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6468 filter [1 x i8**] [@_ZTId]
6475 LLVM supports the notion of an "intrinsic function". These functions
6476 have well known names and semantics and are required to follow certain
6477 restrictions. Overall, these intrinsics represent an extension mechanism
6478 for the LLVM language that does not require changing all of the
6479 transformations in LLVM when adding to the language (or the bitcode
6480 reader/writer, the parser, etc...).
6482 Intrinsic function names must all start with an "``llvm.``" prefix. This
6483 prefix is reserved in LLVM for intrinsic names; thus, function names may
6484 not begin with this prefix. Intrinsic functions must always be external
6485 functions: you cannot define the body of intrinsic functions. Intrinsic
6486 functions may only be used in call or invoke instructions: it is illegal
6487 to take the address of an intrinsic function. Additionally, because
6488 intrinsic functions are part of the LLVM language, it is required if any
6489 are added that they be documented here.
6491 Some intrinsic functions can be overloaded, i.e., the intrinsic
6492 represents a family of functions that perform the same operation but on
6493 different data types. Because LLVM can represent over 8 million
6494 different integer types, overloading is used commonly to allow an
6495 intrinsic function to operate on any integer type. One or more of the
6496 argument types or the result type can be overloaded to accept any
6497 integer type. Argument types may also be defined as exactly matching a
6498 previous argument's type or the result type. This allows an intrinsic
6499 function which accepts multiple arguments, but needs all of them to be
6500 of the same type, to only be overloaded with respect to a single
6501 argument or the result.
6503 Overloaded intrinsics will have the names of its overloaded argument
6504 types encoded into its function name, each preceded by a period. Only
6505 those types which are overloaded result in a name suffix. Arguments
6506 whose type is matched against another type do not. For example, the
6507 ``llvm.ctpop`` function can take an integer of any width and returns an
6508 integer of exactly the same integer width. This leads to a family of
6509 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6510 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6511 overloaded, and only one type suffix is required. Because the argument's
6512 type is matched against the return type, it does not require its own
6515 To learn how to add an intrinsic function, please see the `Extending
6516 LLVM Guide <ExtendingLLVM.html>`_.
6520 Variable Argument Handling Intrinsics
6521 -------------------------------------
6523 Variable argument support is defined in LLVM with the
6524 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6525 functions. These functions are related to the similarly named macros
6526 defined in the ``<stdarg.h>`` header file.
6528 All of these functions operate on arguments that use a target-specific
6529 value type "``va_list``". The LLVM assembly language reference manual
6530 does not define what this type is, so all transformations should be
6531 prepared to handle these functions regardless of the type used.
6533 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6534 variable argument handling intrinsic functions are used.
6536 .. code-block:: llvm
6538 define i32 @test(i32 %X, ...) {
6539 ; Initialize variable argument processing
6541 %ap2 = bitcast i8** %ap to i8*
6542 call void @llvm.va_start(i8* %ap2)
6544 ; Read a single integer argument
6545 %tmp = va_arg i8** %ap, i32
6547 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6549 %aq2 = bitcast i8** %aq to i8*
6550 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6551 call void @llvm.va_end(i8* %aq2)
6553 ; Stop processing of arguments.
6554 call void @llvm.va_end(i8* %ap2)
6558 declare void @llvm.va_start(i8*)
6559 declare void @llvm.va_copy(i8*, i8*)
6560 declare void @llvm.va_end(i8*)
6564 '``llvm.va_start``' Intrinsic
6565 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6572 declare void @llvm.va_start(i8* <arglist>)
6577 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6578 subsequent use by ``va_arg``.
6583 The argument is a pointer to a ``va_list`` element to initialize.
6588 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6589 available in C. In a target-dependent way, it initializes the
6590 ``va_list`` element to which the argument points, so that the next call
6591 to ``va_arg`` will produce the first variable argument passed to the
6592 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6593 to know the last argument of the function as the compiler can figure
6596 '``llvm.va_end``' Intrinsic
6597 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6604 declare void @llvm.va_end(i8* <arglist>)
6609 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6610 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6615 The argument is a pointer to a ``va_list`` to destroy.
6620 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6621 available in C. In a target-dependent way, it destroys the ``va_list``
6622 element to which the argument points. Calls to
6623 :ref:`llvm.va_start <int_va_start>` and
6624 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6629 '``llvm.va_copy``' Intrinsic
6630 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6637 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6642 The '``llvm.va_copy``' intrinsic copies the current argument position
6643 from the source argument list to the destination argument list.
6648 The first argument is a pointer to a ``va_list`` element to initialize.
6649 The second argument is a pointer to a ``va_list`` element to copy from.
6654 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6655 available in C. In a target-dependent way, it copies the source
6656 ``va_list`` element into the destination ``va_list`` element. This
6657 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6658 arbitrarily complex and require, for example, memory allocation.
6660 Accurate Garbage Collection Intrinsics
6661 --------------------------------------
6663 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6664 (GC) requires the implementation and generation of these intrinsics.
6665 These intrinsics allow identification of :ref:`GC roots on the
6666 stack <int_gcroot>`, as well as garbage collector implementations that
6667 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6668 Front-ends for type-safe garbage collected languages should generate
6669 these intrinsics to make use of the LLVM garbage collectors. For more
6670 details, see `Accurate Garbage Collection with
6671 LLVM <GarbageCollection.html>`_.
6673 The garbage collection intrinsics only operate on objects in the generic
6674 address space (address space zero).
6678 '``llvm.gcroot``' Intrinsic
6679 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6686 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6691 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6692 the code generator, and allows some metadata to be associated with it.
6697 The first argument specifies the address of a stack object that contains
6698 the root pointer. The second pointer (which must be either a constant or
6699 a global value address) contains the meta-data to be associated with the
6705 At runtime, a call to this intrinsic stores a null pointer into the
6706 "ptrloc" location. At compile-time, the code generator generates
6707 information to allow the runtime to find the pointer at GC safe points.
6708 The '``llvm.gcroot``' intrinsic may only be used in a function which
6709 :ref:`specifies a GC algorithm <gc>`.
6713 '``llvm.gcread``' Intrinsic
6714 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6721 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6726 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6727 locations, allowing garbage collector implementations that require read
6733 The second argument is the address to read from, which should be an
6734 address allocated from the garbage collector. The first object is a
6735 pointer to the start of the referenced object, if needed by the language
6736 runtime (otherwise null).
6741 The '``llvm.gcread``' intrinsic has the same semantics as a load
6742 instruction, but may be replaced with substantially more complex code by
6743 the garbage collector runtime, as needed. The '``llvm.gcread``'
6744 intrinsic may only be used in a function which :ref:`specifies a GC
6749 '``llvm.gcwrite``' Intrinsic
6750 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6757 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6762 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6763 locations, allowing garbage collector implementations that require write
6764 barriers (such as generational or reference counting collectors).
6769 The first argument is the reference to store, the second is the start of
6770 the object to store it to, and the third is the address of the field of
6771 Obj to store to. If the runtime does not require a pointer to the
6772 object, Obj may be null.
6777 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6778 instruction, but may be replaced with substantially more complex code by
6779 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6780 intrinsic may only be used in a function which :ref:`specifies a GC
6783 Code Generator Intrinsics
6784 -------------------------
6786 These intrinsics are provided by LLVM to expose special features that
6787 may only be implemented with code generator support.
6789 '``llvm.returnaddress``' Intrinsic
6790 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6797 declare i8 *@llvm.returnaddress(i32 <level>)
6802 The '``llvm.returnaddress``' intrinsic attempts to compute a
6803 target-specific value indicating the return address of the current
6804 function or one of its callers.
6809 The argument to this intrinsic indicates which function to return the
6810 address for. Zero indicates the calling function, one indicates its
6811 caller, etc. The argument is **required** to be a constant integer
6817 The '``llvm.returnaddress``' intrinsic either returns a pointer
6818 indicating the return address of the specified call frame, or zero if it
6819 cannot be identified. The value returned by this intrinsic is likely to
6820 be incorrect or 0 for arguments other than zero, so it should only be
6821 used for debugging purposes.
6823 Note that calling this intrinsic does not prevent function inlining or
6824 other aggressive transformations, so the value returned may not be that
6825 of the obvious source-language caller.
6827 '``llvm.frameaddress``' Intrinsic
6828 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6835 declare i8* @llvm.frameaddress(i32 <level>)
6840 The '``llvm.frameaddress``' intrinsic attempts to return the
6841 target-specific frame pointer value for the specified stack frame.
6846 The argument to this intrinsic indicates which function to return the
6847 frame pointer for. Zero indicates the calling function, one indicates
6848 its caller, etc. The argument is **required** to be a constant integer
6854 The '``llvm.frameaddress``' intrinsic either returns a pointer
6855 indicating the frame address of the specified call frame, or zero if it
6856 cannot be identified. The value returned by this intrinsic is likely to
6857 be incorrect or 0 for arguments other than zero, so it should only be
6858 used for debugging purposes.
6860 Note that calling this intrinsic does not prevent function inlining or
6861 other aggressive transformations, so the value returned may not be that
6862 of the obvious source-language caller.
6864 .. _int_read_register:
6865 .. _int_write_register:
6867 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
6868 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6875 declare i32 @llvm.read_register.i32(metadata)
6876 declare i64 @llvm.read_register.i64(metadata)
6877 declare void @llvm.write_register.i32(metadata, i32 @value)
6878 declare void @llvm.write_register.i64(metadata, i64 @value)
6879 !0 = metadata !{metadata !"sp\00"}
6884 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
6885 provides access to the named register. The register must be valid on
6886 the architecture being compiled to. The type needs to be compatible
6887 with the register being read.
6892 The '``llvm.read_register``' intrinsic returns the current value of the
6893 register, where possible. The '``llvm.write_register``' intrinsic sets
6894 the current value of the register, where possible.
6896 This is useful to implement named register global variables that need
6897 to always be mapped to a specific register, as is common practice on
6898 bare-metal programs including OS kernels.
6900 The compiler doesn't check for register availability or use of the used
6901 register in surrounding code, including inline assembly. Because of that,
6902 allocatable registers are not supported.
6904 Warning: So far it only works with the stack pointer on selected
6905 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
6906 work is needed to support other registers and even more so, allocatable
6911 '``llvm.stacksave``' Intrinsic
6912 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6919 declare i8* @llvm.stacksave()
6924 The '``llvm.stacksave``' intrinsic is used to remember the current state
6925 of the function stack, for use with
6926 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6927 implementing language features like scoped automatic variable sized
6933 This intrinsic returns a opaque pointer value that can be passed to
6934 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6935 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6936 ``llvm.stacksave``, it effectively restores the state of the stack to
6937 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6938 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6939 were allocated after the ``llvm.stacksave`` was executed.
6941 .. _int_stackrestore:
6943 '``llvm.stackrestore``' Intrinsic
6944 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6951 declare void @llvm.stackrestore(i8* %ptr)
6956 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6957 the function stack to the state it was in when the corresponding
6958 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6959 useful for implementing language features like scoped automatic variable
6960 sized arrays in C99.
6965 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6967 '``llvm.prefetch``' Intrinsic
6968 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6975 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6980 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6981 insert a prefetch instruction if supported; otherwise, it is a noop.
6982 Prefetches have no effect on the behavior of the program but can change
6983 its performance characteristics.
6988 ``address`` is the address to be prefetched, ``rw`` is the specifier
6989 determining if the fetch should be for a read (0) or write (1), and
6990 ``locality`` is a temporal locality specifier ranging from (0) - no
6991 locality, to (3) - extremely local keep in cache. The ``cache type``
6992 specifies whether the prefetch is performed on the data (1) or
6993 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6994 arguments must be constant integers.
6999 This intrinsic does not modify the behavior of the program. In
7000 particular, prefetches cannot trap and do not produce a value. On
7001 targets that support this intrinsic, the prefetch can provide hints to
7002 the processor cache for better performance.
7004 '``llvm.pcmarker``' Intrinsic
7005 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7012 declare void @llvm.pcmarker(i32 <id>)
7017 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7018 Counter (PC) in a region of code to simulators and other tools. The
7019 method is target specific, but it is expected that the marker will use
7020 exported symbols to transmit the PC of the marker. The marker makes no
7021 guarantees that it will remain with any specific instruction after
7022 optimizations. It is possible that the presence of a marker will inhibit
7023 optimizations. The intended use is to be inserted after optimizations to
7024 allow correlations of simulation runs.
7029 ``id`` is a numerical id identifying the marker.
7034 This intrinsic does not modify the behavior of the program. Backends
7035 that do not support this intrinsic may ignore it.
7037 '``llvm.readcyclecounter``' Intrinsic
7038 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7045 declare i64 @llvm.readcyclecounter()
7050 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
7051 counter register (or similar low latency, high accuracy clocks) on those
7052 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7053 should map to RPCC. As the backing counters overflow quickly (on the
7054 order of 9 seconds on alpha), this should only be used for small
7060 When directly supported, reading the cycle counter should not modify any
7061 memory. Implementations are allowed to either return a application
7062 specific value or a system wide value. On backends without support, this
7063 is lowered to a constant 0.
7065 Note that runtime support may be conditional on the privilege-level code is
7066 running at and the host platform.
7068 '``llvm.clear_cache``' Intrinsic
7069 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7076 declare void @llvm.clear_cache(i8*, i8*)
7081 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
7082 in the specified range to the execution unit of the processor. On
7083 targets with non-unified instruction and data cache, the implementation
7084 flushes the instruction cache.
7089 On platforms with coherent instruction and data caches (e.g. x86), this
7090 intrinsic is a nop. On platforms with non-coherent instruction and data
7091 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
7092 instructions or a system call, if cache flushing requires special
7095 The default behavior is to emit a call to ``__clear_cache`` from the run
7098 This instrinsic does *not* empty the instruction pipeline. Modifications
7099 of the current function are outside the scope of the intrinsic.
7101 Standard C Library Intrinsics
7102 -----------------------------
7104 LLVM provides intrinsics for a few important standard C library
7105 functions. These intrinsics allow source-language front-ends to pass
7106 information about the alignment of the pointer arguments to the code
7107 generator, providing opportunity for more efficient code generation.
7111 '``llvm.memcpy``' Intrinsic
7112 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7117 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
7118 integer bit width and for different address spaces. Not all targets
7119 support all bit widths however.
7123 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7124 i32 <len>, i32 <align>, i1 <isvolatile>)
7125 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7126 i64 <len>, i32 <align>, i1 <isvolatile>)
7131 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7132 source location to the destination location.
7134 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
7135 intrinsics do not return a value, takes extra alignment/isvolatile
7136 arguments and the pointers can be in specified address spaces.
7141 The first argument is a pointer to the destination, the second is a
7142 pointer to the source. The third argument is an integer argument
7143 specifying the number of bytes to copy, the fourth argument is the
7144 alignment of the source and destination locations, and the fifth is a
7145 boolean indicating a volatile access.
7147 If the call to this intrinsic has an alignment value that is not 0 or 1,
7148 then the caller guarantees that both the source and destination pointers
7149 are aligned to that boundary.
7151 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
7152 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7153 very cleanly specified and it is unwise to depend on it.
7158 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7159 source location to the destination location, which are not allowed to
7160 overlap. It copies "len" bytes of memory over. If the argument is known
7161 to be aligned to some boundary, this can be specified as the fourth
7162 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
7164 '``llvm.memmove``' Intrinsic
7165 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7170 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7171 bit width and for different address space. Not all targets support all
7176 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7177 i32 <len>, i32 <align>, i1 <isvolatile>)
7178 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7179 i64 <len>, i32 <align>, i1 <isvolatile>)
7184 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7185 source location to the destination location. It is similar to the
7186 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7189 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7190 intrinsics do not return a value, takes extra alignment/isvolatile
7191 arguments and the pointers can be in specified address spaces.
7196 The first argument is a pointer to the destination, the second is a
7197 pointer to the source. The third argument is an integer argument
7198 specifying the number of bytes to copy, the fourth argument is the
7199 alignment of the source and destination locations, and the fifth is a
7200 boolean indicating a volatile access.
7202 If the call to this intrinsic has an alignment value that is not 0 or 1,
7203 then the caller guarantees that the source and destination pointers are
7204 aligned to that boundary.
7206 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7207 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7208 not very cleanly specified and it is unwise to depend on it.
7213 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7214 source location to the destination location, which may overlap. It
7215 copies "len" bytes of memory over. If the argument is known to be
7216 aligned to some boundary, this can be specified as the fourth argument,
7217 otherwise it should be set to 0 or 1 (both meaning no alignment).
7219 '``llvm.memset.*``' Intrinsics
7220 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7225 This is an overloaded intrinsic. You can use llvm.memset on any integer
7226 bit width and for different address spaces. However, not all targets
7227 support all bit widths.
7231 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7232 i32 <len>, i32 <align>, i1 <isvolatile>)
7233 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7234 i64 <len>, i32 <align>, i1 <isvolatile>)
7239 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7240 particular byte value.
7242 Note that, unlike the standard libc function, the ``llvm.memset``
7243 intrinsic does not return a value and takes extra alignment/volatile
7244 arguments. Also, the destination can be in an arbitrary address space.
7249 The first argument is a pointer to the destination to fill, the second
7250 is the byte value with which to fill it, the third argument is an
7251 integer argument specifying the number of bytes to fill, and the fourth
7252 argument is the known alignment of the destination location.
7254 If the call to this intrinsic has an alignment value that is not 0 or 1,
7255 then the caller guarantees that the destination pointer is aligned to
7258 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7259 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7260 very cleanly specified and it is unwise to depend on it.
7265 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7266 at the destination location. If the argument is known to be aligned to
7267 some boundary, this can be specified as the fourth argument, otherwise
7268 it should be set to 0 or 1 (both meaning no alignment).
7270 '``llvm.sqrt.*``' Intrinsic
7271 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7276 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7277 floating point or vector of floating point type. Not all targets support
7282 declare float @llvm.sqrt.f32(float %Val)
7283 declare double @llvm.sqrt.f64(double %Val)
7284 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7285 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7286 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7291 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7292 returning the same value as the libm '``sqrt``' functions would. Unlike
7293 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7294 negative numbers other than -0.0 (which allows for better optimization,
7295 because there is no need to worry about errno being set).
7296 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7301 The argument and return value are floating point numbers of the same
7307 This function returns the sqrt of the specified operand if it is a
7308 nonnegative floating point number.
7310 '``llvm.powi.*``' Intrinsic
7311 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7316 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7317 floating point or vector of floating point type. Not all targets support
7322 declare float @llvm.powi.f32(float %Val, i32 %power)
7323 declare double @llvm.powi.f64(double %Val, i32 %power)
7324 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7325 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7326 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7331 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7332 specified (positive or negative) power. The order of evaluation of
7333 multiplications is not defined. When a vector of floating point type is
7334 used, the second argument remains a scalar integer value.
7339 The second argument is an integer power, and the first is a value to
7340 raise to that power.
7345 This function returns the first value raised to the second power with an
7346 unspecified sequence of rounding operations.
7348 '``llvm.sin.*``' Intrinsic
7349 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7354 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7355 floating point or vector of floating point type. Not all targets support
7360 declare float @llvm.sin.f32(float %Val)
7361 declare double @llvm.sin.f64(double %Val)
7362 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7363 declare fp128 @llvm.sin.f128(fp128 %Val)
7364 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7369 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7374 The argument and return value are floating point numbers of the same
7380 This function returns the sine of the specified operand, returning the
7381 same values as the libm ``sin`` functions would, and handles error
7382 conditions in the same way.
7384 '``llvm.cos.*``' Intrinsic
7385 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7390 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7391 floating point or vector of floating point type. Not all targets support
7396 declare float @llvm.cos.f32(float %Val)
7397 declare double @llvm.cos.f64(double %Val)
7398 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7399 declare fp128 @llvm.cos.f128(fp128 %Val)
7400 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7405 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7410 The argument and return value are floating point numbers of the same
7416 This function returns the cosine of the specified operand, returning the
7417 same values as the libm ``cos`` functions would, and handles error
7418 conditions in the same way.
7420 '``llvm.pow.*``' Intrinsic
7421 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7426 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7427 floating point or vector of floating point type. Not all targets support
7432 declare float @llvm.pow.f32(float %Val, float %Power)
7433 declare double @llvm.pow.f64(double %Val, double %Power)
7434 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7435 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7436 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7441 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7442 specified (positive or negative) power.
7447 The second argument is a floating point power, and the first is a value
7448 to raise to that power.
7453 This function returns the first value raised to the second power,
7454 returning the same values as the libm ``pow`` functions would, and
7455 handles error conditions in the same way.
7457 '``llvm.exp.*``' Intrinsic
7458 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7463 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7464 floating point or vector of floating point type. Not all targets support
7469 declare float @llvm.exp.f32(float %Val)
7470 declare double @llvm.exp.f64(double %Val)
7471 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7472 declare fp128 @llvm.exp.f128(fp128 %Val)
7473 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7478 The '``llvm.exp.*``' intrinsics perform the exp function.
7483 The argument and return value are floating point numbers of the same
7489 This function returns the same values as the libm ``exp`` functions
7490 would, and handles error conditions in the same way.
7492 '``llvm.exp2.*``' Intrinsic
7493 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7498 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7499 floating point or vector of floating point type. Not all targets support
7504 declare float @llvm.exp2.f32(float %Val)
7505 declare double @llvm.exp2.f64(double %Val)
7506 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7507 declare fp128 @llvm.exp2.f128(fp128 %Val)
7508 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7513 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7518 The argument and return value are floating point numbers of the same
7524 This function returns the same values as the libm ``exp2`` functions
7525 would, and handles error conditions in the same way.
7527 '``llvm.log.*``' Intrinsic
7528 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7533 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7534 floating point or vector of floating point type. Not all targets support
7539 declare float @llvm.log.f32(float %Val)
7540 declare double @llvm.log.f64(double %Val)
7541 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7542 declare fp128 @llvm.log.f128(fp128 %Val)
7543 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7548 The '``llvm.log.*``' intrinsics perform the log function.
7553 The argument and return value are floating point numbers of the same
7559 This function returns the same values as the libm ``log`` functions
7560 would, and handles error conditions in the same way.
7562 '``llvm.log10.*``' Intrinsic
7563 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7568 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7569 floating point or vector of floating point type. Not all targets support
7574 declare float @llvm.log10.f32(float %Val)
7575 declare double @llvm.log10.f64(double %Val)
7576 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7577 declare fp128 @llvm.log10.f128(fp128 %Val)
7578 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7583 The '``llvm.log10.*``' intrinsics perform the log10 function.
7588 The argument and return value are floating point numbers of the same
7594 This function returns the same values as the libm ``log10`` functions
7595 would, and handles error conditions in the same way.
7597 '``llvm.log2.*``' Intrinsic
7598 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7603 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7604 floating point or vector of floating point type. Not all targets support
7609 declare float @llvm.log2.f32(float %Val)
7610 declare double @llvm.log2.f64(double %Val)
7611 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7612 declare fp128 @llvm.log2.f128(fp128 %Val)
7613 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7618 The '``llvm.log2.*``' intrinsics perform the log2 function.
7623 The argument and return value are floating point numbers of the same
7629 This function returns the same values as the libm ``log2`` functions
7630 would, and handles error conditions in the same way.
7632 '``llvm.fma.*``' Intrinsic
7633 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7638 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7639 floating point or vector of floating point type. Not all targets support
7644 declare float @llvm.fma.f32(float %a, float %b, float %c)
7645 declare double @llvm.fma.f64(double %a, double %b, double %c)
7646 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7647 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7648 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7653 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7659 The argument and return value are floating point numbers of the same
7665 This function returns the same values as the libm ``fma`` functions
7666 would, and does not set errno.
7668 '``llvm.fabs.*``' Intrinsic
7669 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7674 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7675 floating point or vector of floating point type. Not all targets support
7680 declare float @llvm.fabs.f32(float %Val)
7681 declare double @llvm.fabs.f64(double %Val)
7682 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7683 declare fp128 @llvm.fabs.f128(fp128 %Val)
7684 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7689 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7695 The argument and return value are floating point numbers of the same
7701 This function returns the same values as the libm ``fabs`` functions
7702 would, and handles error conditions in the same way.
7704 '``llvm.copysign.*``' Intrinsic
7705 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7710 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7711 floating point or vector of floating point type. Not all targets support
7716 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7717 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7718 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7719 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7720 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7725 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7726 first operand and the sign of the second operand.
7731 The arguments and return value are floating point numbers of the same
7737 This function returns the same values as the libm ``copysign``
7738 functions would, and handles error conditions in the same way.
7740 '``llvm.floor.*``' Intrinsic
7741 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7746 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7747 floating point or vector of floating point type. Not all targets support
7752 declare float @llvm.floor.f32(float %Val)
7753 declare double @llvm.floor.f64(double %Val)
7754 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7755 declare fp128 @llvm.floor.f128(fp128 %Val)
7756 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7761 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7766 The argument and return value are floating point numbers of the same
7772 This function returns the same values as the libm ``floor`` functions
7773 would, and handles error conditions in the same way.
7775 '``llvm.ceil.*``' Intrinsic
7776 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7781 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7782 floating point or vector of floating point type. Not all targets support
7787 declare float @llvm.ceil.f32(float %Val)
7788 declare double @llvm.ceil.f64(double %Val)
7789 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7790 declare fp128 @llvm.ceil.f128(fp128 %Val)
7791 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7796 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7801 The argument and return value are floating point numbers of the same
7807 This function returns the same values as the libm ``ceil`` functions
7808 would, and handles error conditions in the same way.
7810 '``llvm.trunc.*``' Intrinsic
7811 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7816 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7817 floating point or vector of floating point type. Not all targets support
7822 declare float @llvm.trunc.f32(float %Val)
7823 declare double @llvm.trunc.f64(double %Val)
7824 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7825 declare fp128 @llvm.trunc.f128(fp128 %Val)
7826 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7831 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7832 nearest integer not larger in magnitude than the operand.
7837 The argument and return value are floating point numbers of the same
7843 This function returns the same values as the libm ``trunc`` functions
7844 would, and handles error conditions in the same way.
7846 '``llvm.rint.*``' Intrinsic
7847 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7852 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7853 floating point or vector of floating point type. Not all targets support
7858 declare float @llvm.rint.f32(float %Val)
7859 declare double @llvm.rint.f64(double %Val)
7860 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7861 declare fp128 @llvm.rint.f128(fp128 %Val)
7862 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7867 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7868 nearest integer. It may raise an inexact floating-point exception if the
7869 operand isn't an integer.
7874 The argument and return value are floating point numbers of the same
7880 This function returns the same values as the libm ``rint`` functions
7881 would, and handles error conditions in the same way.
7883 '``llvm.nearbyint.*``' Intrinsic
7884 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7889 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7890 floating point or vector of floating point type. Not all targets support
7895 declare float @llvm.nearbyint.f32(float %Val)
7896 declare double @llvm.nearbyint.f64(double %Val)
7897 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7898 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7899 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7904 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7910 The argument and return value are floating point numbers of the same
7916 This function returns the same values as the libm ``nearbyint``
7917 functions would, and handles error conditions in the same way.
7919 '``llvm.round.*``' Intrinsic
7920 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7925 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7926 floating point or vector of floating point type. Not all targets support
7931 declare float @llvm.round.f32(float %Val)
7932 declare double @llvm.round.f64(double %Val)
7933 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7934 declare fp128 @llvm.round.f128(fp128 %Val)
7935 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7940 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7946 The argument and return value are floating point numbers of the same
7952 This function returns the same values as the libm ``round``
7953 functions would, and handles error conditions in the same way.
7955 Bit Manipulation Intrinsics
7956 ---------------------------
7958 LLVM provides intrinsics for a few important bit manipulation
7959 operations. These allow efficient code generation for some algorithms.
7961 '``llvm.bswap.*``' Intrinsics
7962 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7967 This is an overloaded intrinsic function. You can use bswap on any
7968 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7972 declare i16 @llvm.bswap.i16(i16 <id>)
7973 declare i32 @llvm.bswap.i32(i32 <id>)
7974 declare i64 @llvm.bswap.i64(i64 <id>)
7979 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7980 values with an even number of bytes (positive multiple of 16 bits).
7981 These are useful for performing operations on data that is not in the
7982 target's native byte order.
7987 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7988 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7989 intrinsic returns an i32 value that has the four bytes of the input i32
7990 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7991 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7992 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7993 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7996 '``llvm.ctpop.*``' Intrinsic
7997 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8002 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
8003 bit width, or on any vector with integer elements. Not all targets
8004 support all bit widths or vector types, however.
8008 declare i8 @llvm.ctpop.i8(i8 <src>)
8009 declare i16 @llvm.ctpop.i16(i16 <src>)
8010 declare i32 @llvm.ctpop.i32(i32 <src>)
8011 declare i64 @llvm.ctpop.i64(i64 <src>)
8012 declare i256 @llvm.ctpop.i256(i256 <src>)
8013 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
8018 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
8024 The only argument is the value to be counted. The argument may be of any
8025 integer type, or a vector with integer elements. The return type must
8026 match the argument type.
8031 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
8032 each element of a vector.
8034 '``llvm.ctlz.*``' Intrinsic
8035 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8040 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
8041 integer bit width, or any vector whose elements are integers. Not all
8042 targets support all bit widths or vector types, however.
8046 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
8047 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
8048 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
8049 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
8050 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
8051 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8056 The '``llvm.ctlz``' family of intrinsic functions counts the number of
8057 leading zeros in a variable.
8062 The first argument is the value to be counted. This argument may be of
8063 any integer type, or a vectory with integer element type. The return
8064 type must match the first argument type.
8066 The second argument must be a constant and is a flag to indicate whether
8067 the intrinsic should ensure that a zero as the first argument produces a
8068 defined result. Historically some architectures did not provide a
8069 defined result for zero values as efficiently, and many algorithms are
8070 now predicated on avoiding zero-value inputs.
8075 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
8076 zeros in a variable, or within each element of the vector. If
8077 ``src == 0`` then the result is the size in bits of the type of ``src``
8078 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8079 ``llvm.ctlz(i32 2) = 30``.
8081 '``llvm.cttz.*``' Intrinsic
8082 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8087 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
8088 integer bit width, or any vector of integer elements. Not all targets
8089 support all bit widths or vector types, however.
8093 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
8094 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
8095 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
8096 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
8097 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
8098 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8103 The '``llvm.cttz``' family of intrinsic functions counts the number of
8109 The first argument is the value to be counted. This argument may be of
8110 any integer type, or a vectory with integer element type. The return
8111 type must match the first argument type.
8113 The second argument must be a constant and is a flag to indicate whether
8114 the intrinsic should ensure that a zero as the first argument produces a
8115 defined result. Historically some architectures did not provide a
8116 defined result for zero values as efficiently, and many algorithms are
8117 now predicated on avoiding zero-value inputs.
8122 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
8123 zeros in a variable, or within each element of a vector. If ``src == 0``
8124 then the result is the size in bits of the type of ``src`` if
8125 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8126 ``llvm.cttz(2) = 1``.
8128 Arithmetic with Overflow Intrinsics
8129 -----------------------------------
8131 LLVM provides intrinsics for some arithmetic with overflow operations.
8133 '``llvm.sadd.with.overflow.*``' Intrinsics
8134 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8139 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
8140 on any integer bit width.
8144 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8145 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8146 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8151 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8152 a signed addition of the two arguments, and indicate whether an overflow
8153 occurred during the signed summation.
8158 The arguments (%a and %b) and the first element of the result structure
8159 may be of integer types of any bit width, but they must have the same
8160 bit width. The second element of the result structure must be of type
8161 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8167 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8168 a signed addition of the two variables. They return a structure --- the
8169 first element of which is the signed summation, and the second element
8170 of which is a bit specifying if the signed summation resulted in an
8176 .. code-block:: llvm
8178 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8179 %sum = extractvalue {i32, i1} %res, 0
8180 %obit = extractvalue {i32, i1} %res, 1
8181 br i1 %obit, label %overflow, label %normal
8183 '``llvm.uadd.with.overflow.*``' Intrinsics
8184 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8189 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8190 on any integer bit width.
8194 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8195 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8196 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8201 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8202 an unsigned addition of the two arguments, and indicate whether a carry
8203 occurred during the unsigned summation.
8208 The arguments (%a and %b) and the first element of the result structure
8209 may be of integer types of any bit width, but they must have the same
8210 bit width. The second element of the result structure must be of type
8211 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8217 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8218 an unsigned addition of the two arguments. They return a structure --- the
8219 first element of which is the sum, and the second element of which is a
8220 bit specifying if the unsigned summation resulted in a carry.
8225 .. code-block:: llvm
8227 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8228 %sum = extractvalue {i32, i1} %res, 0
8229 %obit = extractvalue {i32, i1} %res, 1
8230 br i1 %obit, label %carry, label %normal
8232 '``llvm.ssub.with.overflow.*``' Intrinsics
8233 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8238 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8239 on any integer bit width.
8243 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8244 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8245 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8250 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8251 a signed subtraction of the two arguments, and indicate whether an
8252 overflow occurred during the signed subtraction.
8257 The arguments (%a and %b) and the first element of the result structure
8258 may be of integer types of any bit width, but they must have the same
8259 bit width. The second element of the result structure must be of type
8260 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8266 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8267 a signed subtraction of the two arguments. They return a structure --- the
8268 first element of which is the subtraction, and the second element of
8269 which is a bit specifying if the signed subtraction resulted in an
8275 .. code-block:: llvm
8277 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8278 %sum = extractvalue {i32, i1} %res, 0
8279 %obit = extractvalue {i32, i1} %res, 1
8280 br i1 %obit, label %overflow, label %normal
8282 '``llvm.usub.with.overflow.*``' Intrinsics
8283 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8288 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8289 on any integer bit width.
8293 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8294 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8295 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8300 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8301 an unsigned subtraction of the two arguments, and indicate whether an
8302 overflow occurred during the unsigned subtraction.
8307 The arguments (%a and %b) and the first element of the result structure
8308 may be of integer types of any bit width, but they must have the same
8309 bit width. The second element of the result structure must be of type
8310 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8316 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8317 an unsigned subtraction of the two arguments. They return a structure ---
8318 the first element of which is the subtraction, and the second element of
8319 which is a bit specifying if the unsigned subtraction resulted in an
8325 .. code-block:: llvm
8327 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8328 %sum = extractvalue {i32, i1} %res, 0
8329 %obit = extractvalue {i32, i1} %res, 1
8330 br i1 %obit, label %overflow, label %normal
8332 '``llvm.smul.with.overflow.*``' Intrinsics
8333 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8338 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8339 on any integer bit width.
8343 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8344 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8345 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8350 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8351 a signed multiplication of the two arguments, and indicate whether an
8352 overflow occurred during the signed multiplication.
8357 The arguments (%a and %b) and the first element of the result structure
8358 may be of integer types of any bit width, but they must have the same
8359 bit width. The second element of the result structure must be of type
8360 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8366 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8367 a signed multiplication of the two arguments. They return a structure ---
8368 the first element of which is the multiplication, and the second element
8369 of which is a bit specifying if the signed multiplication resulted in an
8375 .. code-block:: llvm
8377 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8378 %sum = extractvalue {i32, i1} %res, 0
8379 %obit = extractvalue {i32, i1} %res, 1
8380 br i1 %obit, label %overflow, label %normal
8382 '``llvm.umul.with.overflow.*``' Intrinsics
8383 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8388 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8389 on any integer bit width.
8393 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8394 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8395 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8400 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8401 a unsigned multiplication of the two arguments, and indicate whether an
8402 overflow occurred during the unsigned multiplication.
8407 The arguments (%a and %b) and the first element of the result structure
8408 may be of integer types of any bit width, but they must have the same
8409 bit width. The second element of the result structure must be of type
8410 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8416 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8417 an unsigned multiplication of the two arguments. They return a structure ---
8418 the first element of which is the multiplication, and the second
8419 element of which is a bit specifying if the unsigned multiplication
8420 resulted in an overflow.
8425 .. code-block:: llvm
8427 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8428 %sum = extractvalue {i32, i1} %res, 0
8429 %obit = extractvalue {i32, i1} %res, 1
8430 br i1 %obit, label %overflow, label %normal
8432 Specialised Arithmetic Intrinsics
8433 ---------------------------------
8435 '``llvm.fmuladd.*``' Intrinsic
8436 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8443 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8444 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8449 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8450 expressions that can be fused if the code generator determines that (a) the
8451 target instruction set has support for a fused operation, and (b) that the
8452 fused operation is more efficient than the equivalent, separate pair of mul
8453 and add instructions.
8458 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8459 multiplicands, a and b, and an addend c.
8468 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8470 is equivalent to the expression a \* b + c, except that rounding will
8471 not be performed between the multiplication and addition steps if the
8472 code generator fuses the operations. Fusion is not guaranteed, even if
8473 the target platform supports it. If a fused multiply-add is required the
8474 corresponding llvm.fma.\* intrinsic function should be used
8475 instead. This never sets errno, just as '``llvm.fma.*``'.
8480 .. code-block:: llvm
8482 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8484 Half Precision Floating Point Intrinsics
8485 ----------------------------------------
8487 For most target platforms, half precision floating point is a
8488 storage-only format. This means that it is a dense encoding (in memory)
8489 but does not support computation in the format.
8491 This means that code must first load the half-precision floating point
8492 value as an i16, then convert it to float with
8493 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8494 then be performed on the float value (including extending to double
8495 etc). To store the value back to memory, it is first converted to float
8496 if needed, then converted to i16 with
8497 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8500 .. _int_convert_to_fp16:
8502 '``llvm.convert.to.fp16``' Intrinsic
8503 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8510 declare i16 @llvm.convert.to.fp16(f32 %a)
8515 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8516 from single precision floating point format to half precision floating
8522 The intrinsic function contains single argument - the value to be
8528 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8529 from single precision floating point format to half precision floating
8530 point format. The return value is an ``i16`` which contains the
8536 .. code-block:: llvm
8538 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8539 store i16 %res, i16* @x, align 2
8541 .. _int_convert_from_fp16:
8543 '``llvm.convert.from.fp16``' Intrinsic
8544 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8551 declare f32 @llvm.convert.from.fp16(i16 %a)
8556 The '``llvm.convert.from.fp16``' intrinsic function performs a
8557 conversion from half precision floating point format to single precision
8558 floating point format.
8563 The intrinsic function contains single argument - the value to be
8569 The '``llvm.convert.from.fp16``' intrinsic function performs a
8570 conversion from half single precision floating point format to single
8571 precision floating point format. The input half-float value is
8572 represented by an ``i16`` value.
8577 .. code-block:: llvm
8579 %a = load i16* @x, align 2
8580 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8585 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8586 prefix), are described in the `LLVM Source Level
8587 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8590 Exception Handling Intrinsics
8591 -----------------------------
8593 The LLVM exception handling intrinsics (which all start with
8594 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8595 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8599 Trampoline Intrinsics
8600 ---------------------
8602 These intrinsics make it possible to excise one parameter, marked with
8603 the :ref:`nest <nest>` attribute, from a function. The result is a
8604 callable function pointer lacking the nest parameter - the caller does
8605 not need to provide a value for it. Instead, the value to use is stored
8606 in advance in a "trampoline", a block of memory usually allocated on the
8607 stack, which also contains code to splice the nest value into the
8608 argument list. This is used to implement the GCC nested function address
8611 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8612 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8613 It can be created as follows:
8615 .. code-block:: llvm
8617 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8618 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8619 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8620 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8621 %fp = bitcast i8* %p to i32 (i32, i32)*
8623 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8624 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8628 '``llvm.init.trampoline``' Intrinsic
8629 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8636 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8641 This fills the memory pointed to by ``tramp`` with executable code,
8642 turning it into a trampoline.
8647 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8648 pointers. The ``tramp`` argument must point to a sufficiently large and
8649 sufficiently aligned block of memory; this memory is written to by the
8650 intrinsic. Note that the size and the alignment are target-specific -
8651 LLVM currently provides no portable way of determining them, so a
8652 front-end that generates this intrinsic needs to have some
8653 target-specific knowledge. The ``func`` argument must hold a function
8654 bitcast to an ``i8*``.
8659 The block of memory pointed to by ``tramp`` is filled with target
8660 dependent code, turning it into a function. Then ``tramp`` needs to be
8661 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8662 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8663 function's signature is the same as that of ``func`` with any arguments
8664 marked with the ``nest`` attribute removed. At most one such ``nest``
8665 argument is allowed, and it must be of pointer type. Calling the new
8666 function is equivalent to calling ``func`` with the same argument list,
8667 but with ``nval`` used for the missing ``nest`` argument. If, after
8668 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8669 modified, then the effect of any later call to the returned function
8670 pointer is undefined.
8674 '``llvm.adjust.trampoline``' Intrinsic
8675 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8682 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8687 This performs any required machine-specific adjustment to the address of
8688 a trampoline (passed as ``tramp``).
8693 ``tramp`` must point to a block of memory which already has trampoline
8694 code filled in by a previous call to
8695 :ref:`llvm.init.trampoline <int_it>`.
8700 On some architectures the address of the code to be executed needs to be
8701 different to the address where the trampoline is actually stored. This
8702 intrinsic returns the executable address corresponding to ``tramp``
8703 after performing the required machine specific adjustments. The pointer
8704 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8709 This class of intrinsics exists to information about the lifetime of
8710 memory objects and ranges where variables are immutable.
8714 '``llvm.lifetime.start``' Intrinsic
8715 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8722 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8727 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8733 The first argument is a constant integer representing the size of the
8734 object, or -1 if it is variable sized. The second argument is a pointer
8740 This intrinsic indicates that before this point in the code, the value
8741 of the memory pointed to by ``ptr`` is dead. This means that it is known
8742 to never be used and has an undefined value. A load from the pointer
8743 that precedes this intrinsic can be replaced with ``'undef'``.
8747 '``llvm.lifetime.end``' Intrinsic
8748 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8755 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8760 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8766 The first argument is a constant integer representing the size of the
8767 object, or -1 if it is variable sized. The second argument is a pointer
8773 This intrinsic indicates that after this point in the code, the value of
8774 the memory pointed to by ``ptr`` is dead. This means that it is known to
8775 never be used and has an undefined value. Any stores into the memory
8776 object following this intrinsic may be removed as dead.
8778 '``llvm.invariant.start``' Intrinsic
8779 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8786 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8791 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8792 a memory object will not change.
8797 The first argument is a constant integer representing the size of the
8798 object, or -1 if it is variable sized. The second argument is a pointer
8804 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8805 the return value, the referenced memory location is constant and
8808 '``llvm.invariant.end``' Intrinsic
8809 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8816 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8821 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8822 memory object are mutable.
8827 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8828 The second argument is a constant integer representing the size of the
8829 object, or -1 if it is variable sized and the third argument is a
8830 pointer to the object.
8835 This intrinsic indicates that the memory is mutable again.
8840 This class of intrinsics is designed to be generic and has no specific
8843 '``llvm.var.annotation``' Intrinsic
8844 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8851 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8856 The '``llvm.var.annotation``' intrinsic.
8861 The first argument is a pointer to a value, the second is a pointer to a
8862 global string, the third is a pointer to a global string which is the
8863 source file name, and the last argument is the line number.
8868 This intrinsic allows annotation of local variables with arbitrary
8869 strings. This can be useful for special purpose optimizations that want
8870 to look for these annotations. These have no other defined use; they are
8871 ignored by code generation and optimization.
8873 '``llvm.ptr.annotation.*``' Intrinsic
8874 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8879 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8880 pointer to an integer of any width. *NOTE* you must specify an address space for
8881 the pointer. The identifier for the default address space is the integer
8886 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8887 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8888 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8889 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8890 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8895 The '``llvm.ptr.annotation``' intrinsic.
8900 The first argument is a pointer to an integer value of arbitrary bitwidth
8901 (result of some expression), the second is a pointer to a global string, the
8902 third is a pointer to a global string which is the source file name, and the
8903 last argument is the line number. It returns the value of the first argument.
8908 This intrinsic allows annotation of a pointer to an integer with arbitrary
8909 strings. This can be useful for special purpose optimizations that want to look
8910 for these annotations. These have no other defined use; they are ignored by code
8911 generation and optimization.
8913 '``llvm.annotation.*``' Intrinsic
8914 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8919 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8920 any integer bit width.
8924 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8925 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8926 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8927 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8928 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8933 The '``llvm.annotation``' intrinsic.
8938 The first argument is an integer value (result of some expression), the
8939 second is a pointer to a global string, the third is a pointer to a
8940 global string which is the source file name, and the last argument is
8941 the line number. It returns the value of the first argument.
8946 This intrinsic allows annotations to be put on arbitrary expressions
8947 with arbitrary strings. This can be useful for special purpose
8948 optimizations that want to look for these annotations. These have no
8949 other defined use; they are ignored by code generation and optimization.
8951 '``llvm.trap``' Intrinsic
8952 ^^^^^^^^^^^^^^^^^^^^^^^^^
8959 declare void @llvm.trap() noreturn nounwind
8964 The '``llvm.trap``' intrinsic.
8974 This intrinsic is lowered to the target dependent trap instruction. If
8975 the target does not have a trap instruction, this intrinsic will be
8976 lowered to a call of the ``abort()`` function.
8978 '``llvm.debugtrap``' Intrinsic
8979 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8986 declare void @llvm.debugtrap() nounwind
8991 The '``llvm.debugtrap``' intrinsic.
9001 This intrinsic is lowered to code which is intended to cause an
9002 execution trap with the intention of requesting the attention of a
9005 '``llvm.stackprotector``' Intrinsic
9006 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9013 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
9018 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
9019 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
9020 is placed on the stack before local variables.
9025 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
9026 The first argument is the value loaded from the stack guard
9027 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
9028 enough space to hold the value of the guard.
9033 This intrinsic causes the prologue/epilogue inserter to force the position of
9034 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
9035 to ensure that if a local variable on the stack is overwritten, it will destroy
9036 the value of the guard. When the function exits, the guard on the stack is
9037 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
9038 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
9039 calling the ``__stack_chk_fail()`` function.
9041 '``llvm.stackprotectorcheck``' Intrinsic
9042 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9049 declare void @llvm.stackprotectorcheck(i8** <guard>)
9054 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
9055 created stack protector and if they are not equal calls the
9056 ``__stack_chk_fail()`` function.
9061 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
9062 the variable ``@__stack_chk_guard``.
9067 This intrinsic is provided to perform the stack protector check by comparing
9068 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
9069 values do not match call the ``__stack_chk_fail()`` function.
9071 The reason to provide this as an IR level intrinsic instead of implementing it
9072 via other IR operations is that in order to perform this operation at the IR
9073 level without an intrinsic, one would need to create additional basic blocks to
9074 handle the success/failure cases. This makes it difficult to stop the stack
9075 protector check from disrupting sibling tail calls in Codegen. With this
9076 intrinsic, we are able to generate the stack protector basic blocks late in
9077 codegen after the tail call decision has occurred.
9079 '``llvm.objectsize``' Intrinsic
9080 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9087 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
9088 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
9093 The ``llvm.objectsize`` intrinsic is designed to provide information to
9094 the optimizers to determine at compile time whether a) an operation
9095 (like memcpy) will overflow a buffer that corresponds to an object, or
9096 b) that a runtime check for overflow isn't necessary. An object in this
9097 context means an allocation of a specific class, structure, array, or
9103 The ``llvm.objectsize`` intrinsic takes two arguments. The first
9104 argument is a pointer to or into the ``object``. The second argument is
9105 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
9106 or -1 (if false) when the object size is unknown. The second argument
9107 only accepts constants.
9112 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
9113 the size of the object concerned. If the size cannot be determined at
9114 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
9115 on the ``min`` argument).
9117 '``llvm.expect``' Intrinsic
9118 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9123 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
9128 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
9129 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9130 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9135 The ``llvm.expect`` intrinsic provides information about expected (the
9136 most probable) value of ``val``, which can be used by optimizers.
9141 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
9142 a value. The second argument is an expected value, this needs to be a
9143 constant value, variables are not allowed.
9148 This intrinsic is lowered to the ``val``.
9150 '``llvm.donothing``' Intrinsic
9151 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9158 declare void @llvm.donothing() nounwind readnone
9163 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
9164 only intrinsic that can be called with an invoke instruction.
9174 This intrinsic does nothing, and it's removed by optimizers and ignored
9177 Stack Map Intrinsics
9178 --------------------
9180 LLVM provides experimental intrinsics to support runtime patching
9181 mechanisms commonly desired in dynamic language JITs. These intrinsics
9182 are described in :doc:`StackMaps`.