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.
470 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
471 types <t_struct>`. Literal types are uniqued structurally, but identified types
472 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
473 to forward declare a type which is not yet available.
475 An example of a identified structure specification is:
479 %mytype = type { %mytype*, i32 }
481 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
482 literal types are uniqued in recent versions of LLVM.
489 Global variables define regions of memory allocated at compilation time
492 Global variables definitions must be initialized, may have an explicit section
493 to be placed in, and may have an optional explicit alignment specified.
495 Global variables in other translation units can also be declared, in which
496 case they don't have an initializer.
498 A variable may be defined as ``thread_local``, which means that it will
499 not be shared by threads (each thread will have a separated copy of the
500 variable). Not all targets support thread-local variables. Optionally, a
501 TLS model may be specified:
504 For variables that are only used within the current shared library.
506 For variables in modules that will not be loaded dynamically.
508 For variables defined in the executable and only used within it.
510 The models correspond to the ELF TLS models; see `ELF Handling For
511 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
512 more information on under which circumstances the different models may
513 be used. The target may choose a different TLS model if the specified
514 model is not supported, or if a better choice of model can be made.
516 A variable may be defined as a global ``constant``, which indicates that
517 the contents of the variable will **never** be modified (enabling better
518 optimization, allowing the global data to be placed in the read-only
519 section of an executable, etc). Note that variables that need runtime
520 initialization cannot be marked ``constant`` as there is a store to the
523 LLVM explicitly allows *declarations* of global variables to be marked
524 constant, even if the final definition of the global is not. This
525 capability can be used to enable slightly better optimization of the
526 program, but requires the language definition to guarantee that
527 optimizations based on the 'constantness' are valid for the translation
528 units that do not include the definition.
530 As SSA values, global variables define pointer values that are in scope
531 (i.e. they dominate) all basic blocks in the program. Global variables
532 always define a pointer to their "content" type because they describe a
533 region of memory, and all memory objects in LLVM are accessed through
536 Global variables can be marked with ``unnamed_addr`` which indicates
537 that the address is not significant, only the content. Constants marked
538 like this can be merged with other constants if they have the same
539 initializer. Note that a constant with significant address *can* be
540 merged with a ``unnamed_addr`` constant, the result being a constant
541 whose address is significant.
543 A global variable may be declared to reside in a target-specific
544 numbered address space. For targets that support them, address spaces
545 may affect how optimizations are performed and/or what target
546 instructions are used to access the variable. The default address space
547 is zero. The address space qualifier must precede any other attributes.
549 LLVM allows an explicit section to be specified for globals. If the
550 target supports it, it will emit globals to the section specified.
552 By default, global initializers are optimized by assuming that global
553 variables defined within the module are not modified from their
554 initial values before the start of the global initializer. This is
555 true even for variables potentially accessible from outside the
556 module, including those with external linkage or appearing in
557 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
558 by marking the variable with ``externally_initialized``.
560 An explicit alignment may be specified for a global, which must be a
561 power of 2. If not present, or if the alignment is set to zero, the
562 alignment of the global is set by the target to whatever it feels
563 convenient. If an explicit alignment is specified, the global is forced
564 to have exactly that alignment. Targets and optimizers are not allowed
565 to over-align the global if the global has an assigned section. In this
566 case, the extra alignment could be observable: for example, code could
567 assume that the globals are densely packed in their section and try to
568 iterate over them as an array, alignment padding would break this
571 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
575 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
576 [AddrSpace] [unnamed_addr] [ExternallyInitialized]
577 <global | constant> <Type>
578 [, section "name"] [, align <Alignment>]
580 For example, the following defines a global in a numbered address space
581 with an initializer, section, and alignment:
585 @G = addrspace(5) constant float 1.0, section "foo", align 4
587 The following example just declares a global variable
591 @G = external global i32
593 The following example defines a thread-local global with the
594 ``initialexec`` TLS model:
598 @G = thread_local(initialexec) global i32 0, align 4
600 .. _functionstructure:
605 LLVM function definitions consist of the "``define``" keyword, an
606 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
607 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
608 an optional :ref:`calling convention <callingconv>`,
609 an optional ``unnamed_addr`` attribute, a return type, an optional
610 :ref:`parameter attribute <paramattrs>` for the return type, a function
611 name, a (possibly empty) argument list (each with optional :ref:`parameter
612 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
613 an optional section, an optional alignment, an optional :ref:`garbage
614 collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
615 curly brace, a list of basic blocks, and a closing curly brace.
617 LLVM function declarations consist of the "``declare``" keyword, an
618 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
619 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
620 an optional :ref:`calling convention <callingconv>`,
621 an optional ``unnamed_addr`` attribute, a return type, an optional
622 :ref:`parameter attribute <paramattrs>` for the return type, a function
623 name, a possibly empty list of arguments, an optional alignment, an optional
624 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
626 A function definition contains a list of basic blocks, forming the CFG (Control
627 Flow Graph) for the function. Each basic block may optionally start with a label
628 (giving the basic block a symbol table entry), contains a list of instructions,
629 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
630 function return). If an explicit label is not provided, a block is assigned an
631 implicit numbered label, using the next value from the same counter as used for
632 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
633 entry block does not have an explicit label, it will be assigned label "%0",
634 then the first unnamed temporary in that block will be "%1", etc.
636 The first basic block in a function is special in two ways: it is
637 immediately executed on entrance to the function, and it is not allowed
638 to have predecessor basic blocks (i.e. there can not be any branches to
639 the entry block of a function). Because the block can have no
640 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
642 LLVM allows an explicit section to be specified for functions. If the
643 target supports it, it will emit functions to the section specified.
645 An explicit alignment may be specified for a function. If not present,
646 or if the alignment is set to zero, the alignment of the function is set
647 by the target to whatever it feels convenient. If an explicit alignment
648 is specified, the function is forced to have at least that much
649 alignment. All alignments must be a power of 2.
651 If the ``unnamed_addr`` attribute is given, the address is know to not
652 be significant and two identical functions can be merged.
656 define [linkage] [visibility] [DLLStorageClass]
658 <ResultType> @<FunctionName> ([argument list])
659 [unnamed_addr] [fn Attrs] [section "name"] [align N]
660 [gc] [prefix Constant] { ... }
667 Aliases act as "second name" for the aliasee value (which can be either
668 function, global variable, another alias or bitcast of global value).
669 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
670 :ref:`visibility style <visibility>`, and an optional :ref:`DLL storage class
675 @<Name> = [Visibility] [DLLStorageClass] alias [Linkage] <AliaseeTy> @<Aliasee>
677 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
678 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
679 might not correctly handle dropping a weak symbol that is aliased by a non-weak
682 Alias that are not ``unnamed_addr`` are guaranteed to have the same address as
685 The aliasee must be a definition.
687 Aliases are not allowed to point to aliases with linkages that can be
688 overridden. Since they are only a second name, the possibility of the
689 intermediate alias being overridden cannot be represented in an object file.
691 .. _namedmetadatastructure:
696 Named metadata is a collection of metadata. :ref:`Metadata
697 nodes <metadata>` (but not metadata strings) are the only valid
698 operands for a named metadata.
702 ; Some unnamed metadata nodes, which are referenced by the named metadata.
703 !0 = metadata !{metadata !"zero"}
704 !1 = metadata !{metadata !"one"}
705 !2 = metadata !{metadata !"two"}
707 !name = !{!0, !1, !2}
714 The return type and each parameter of a function type may have a set of
715 *parameter attributes* associated with them. Parameter attributes are
716 used to communicate additional information about the result or
717 parameters of a function. Parameter attributes are considered to be part
718 of the function, not of the function type, so functions with different
719 parameter attributes can have the same function type.
721 Parameter attributes are simple keywords that follow the type specified.
722 If multiple parameter attributes are needed, they are space separated.
727 declare i32 @printf(i8* noalias nocapture, ...)
728 declare i32 @atoi(i8 zeroext)
729 declare signext i8 @returns_signed_char()
731 Note that any attributes for the function result (``nounwind``,
732 ``readonly``) come immediately after the argument list.
734 Currently, only the following parameter attributes are defined:
737 This indicates to the code generator that the parameter or return
738 value should be zero-extended to the extent required by the target's
739 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
740 the caller (for a parameter) or the callee (for a return value).
742 This indicates to the code generator that the parameter or return
743 value should be sign-extended to the extent required by the target's
744 ABI (which is usually 32-bits) by the caller (for a parameter) or
745 the callee (for a return value).
747 This indicates that this parameter or return value should be treated
748 in a special target-dependent fashion during while emitting code for
749 a function call or return (usually, by putting it in a register as
750 opposed to memory, though some targets use it to distinguish between
751 two different kinds of registers). Use of this attribute is
754 This indicates that the pointer parameter should really be passed by
755 value to the function. The attribute implies that a hidden copy of
756 the pointee is made between the caller and the callee, so the callee
757 is unable to modify the value in the caller. This attribute is only
758 valid on LLVM pointer arguments. It is generally used to pass
759 structs and arrays by value, but is also valid on pointers to
760 scalars. The copy is considered to belong to the caller not the
761 callee (for example, ``readonly`` functions should not write to
762 ``byval`` parameters). This is not a valid attribute for return
765 The byval attribute also supports specifying an alignment with the
766 align attribute. It indicates the alignment of the stack slot to
767 form and the known alignment of the pointer specified to the call
768 site. If the alignment is not specified, then the code generator
769 makes a target-specific assumption.
775 The ``inalloca`` argument attribute allows the caller to take the
776 address of outgoing stack arguments. An ``inalloca`` argument must
777 be a pointer to stack memory produced by an ``alloca`` instruction.
778 The alloca, or argument allocation, must also be tagged with the
779 inalloca keyword. Only the past argument may have the ``inalloca``
780 attribute, and that argument is guaranteed to be passed in memory.
782 An argument allocation may be used by a call at most once because
783 the call may deallocate it. The ``inalloca`` attribute cannot be
784 used in conjunction with other attributes that affect argument
785 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
786 ``inalloca`` attribute also disables LLVM's implicit lowering of
787 large aggregate return values, which means that frontend authors
788 must lower them with ``sret`` pointers.
790 When the call site is reached, the argument allocation must have
791 been the most recent stack allocation that is still live, or the
792 results are undefined. It is possible to allocate additional stack
793 space after an argument allocation and before its call site, but it
794 must be cleared off with :ref:`llvm.stackrestore
797 See :doc:`InAlloca` for more information on how to use this
801 This indicates that the pointer parameter specifies the address of a
802 structure that is the return value of the function in the source
803 program. This pointer must be guaranteed by the caller to be valid:
804 loads and stores to the structure may be assumed by the callee
805 not to trap and to be properly aligned. This may only be applied to
806 the first parameter. This is not a valid attribute for return
812 This indicates that pointer values :ref:`based <pointeraliasing>` on
813 the argument or return value do not alias pointer values which are
814 not *based* on it, ignoring certain "irrelevant" dependencies. For a
815 call to the parent function, dependencies between memory references
816 from before or after the call and from those during the call are
817 "irrelevant" to the ``noalias`` keyword for the arguments and return
818 value used in that call. The caller shares the responsibility with
819 the callee for ensuring that these requirements are met. For further
820 details, please see the discussion of the NoAlias response in :ref:`alias
821 analysis <Must, May, or No>`.
823 Note that this definition of ``noalias`` is intentionally similar
824 to the definition of ``restrict`` in C99 for function arguments,
825 though it is slightly weaker.
827 For function return values, C99's ``restrict`` is not meaningful,
828 while LLVM's ``noalias`` is.
830 This indicates that the callee does not make any copies of the
831 pointer that outlive the callee itself. This is not a valid
832 attribute for return values.
837 This indicates that the pointer parameter can be excised using the
838 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
839 attribute for return values and can only be applied to one parameter.
842 This indicates that the function always returns the argument as its return
843 value. This is an optimization hint to the code generator when generating
844 the caller, allowing tail call optimization and omission of register saves
845 and restores in some cases; it is not checked or enforced when generating
846 the callee. The parameter and the function return type must be valid
847 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
848 valid attribute for return values and can only be applied to one parameter.
851 This indicates that the parameter or return pointer is not null. This
852 attribute may only be applied to pointer typed parameters. This is not
853 checked or enforced by LLVM, the caller must ensure that the pointer
854 passed in is non-null, or the callee must ensure that the returned pointer
859 Garbage Collector Names
860 -----------------------
862 Each function may specify a garbage collector name, which is simply a
867 define void @f() gc "name" { ... }
869 The compiler declares the supported values of *name*. Specifying a
870 collector which will cause the compiler to alter its output in order to
871 support the named garbage collection algorithm.
878 Prefix data is data associated with a function which the code generator
879 will emit immediately before the function body. The purpose of this feature
880 is to allow frontends to associate language-specific runtime metadata with
881 specific functions and make it available through the function pointer while
882 still allowing the function pointer to be called. To access the data for a
883 given function, a program may bitcast the function pointer to a pointer to
884 the constant's type. This implies that the IR symbol points to the start
887 To maintain the semantics of ordinary function calls, the prefix data must
888 have a particular format. Specifically, it must begin with a sequence of
889 bytes which decode to a sequence of machine instructions, valid for the
890 module's target, which transfer control to the point immediately succeeding
891 the prefix data, without performing any other visible action. This allows
892 the inliner and other passes to reason about the semantics of the function
893 definition without needing to reason about the prefix data. Obviously this
894 makes the format of the prefix data highly target dependent.
896 Prefix data is laid out as if it were an initializer for a global variable
897 of the prefix data's type. No padding is automatically placed between the
898 prefix data and the function body. If padding is required, it must be part
901 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
902 which encodes the ``nop`` instruction:
906 define void @f() prefix i8 144 { ... }
908 Generally prefix data can be formed by encoding a relative branch instruction
909 which skips the metadata, as in this example of valid prefix data for the
910 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
914 %0 = type <{ i8, i8, i8* }>
916 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
918 A function may have prefix data but no body. This has similar semantics
919 to the ``available_externally`` linkage in that the data may be used by the
920 optimizers but will not be emitted in the object file.
927 Attribute groups are groups of attributes that are referenced by objects within
928 the IR. They are important for keeping ``.ll`` files readable, because a lot of
929 functions will use the same set of attributes. In the degenerative case of a
930 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
931 group will capture the important command line flags used to build that file.
933 An attribute group is a module-level object. To use an attribute group, an
934 object references the attribute group's ID (e.g. ``#37``). An object may refer
935 to more than one attribute group. In that situation, the attributes from the
936 different groups are merged.
938 Here is an example of attribute groups for a function that should always be
939 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
943 ; Target-independent attributes:
944 attributes #0 = { alwaysinline alignstack=4 }
946 ; Target-dependent attributes:
947 attributes #1 = { "no-sse" }
949 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
950 define void @f() #0 #1 { ... }
957 Function attributes are set to communicate additional information about
958 a function. Function attributes are considered to be part of the
959 function, not of the function type, so functions with different function
960 attributes can have the same function type.
962 Function attributes are simple keywords that follow the type specified.
963 If multiple attributes are needed, they are space separated. For
968 define void @f() noinline { ... }
969 define void @f() alwaysinline { ... }
970 define void @f() alwaysinline optsize { ... }
971 define void @f() optsize { ... }
974 This attribute indicates that, when emitting the prologue and
975 epilogue, the backend should forcibly align the stack pointer.
976 Specify the desired alignment, which must be a power of two, in
979 This attribute indicates that the inliner should attempt to inline
980 this function into callers whenever possible, ignoring any active
981 inlining size threshold for this caller.
983 This indicates that the callee function at a call site should be
984 recognized as a built-in function, even though the function's declaration
985 uses the ``nobuiltin`` attribute. This is only valid at call sites for
986 direct calls to functions which are declared with the ``nobuiltin``
989 This attribute indicates that this function is rarely called. When
990 computing edge weights, basic blocks post-dominated by a cold
991 function call are also considered to be cold; and, thus, given low
994 This attribute indicates that the source code contained a hint that
995 inlining this function is desirable (such as the "inline" keyword in
996 C/C++). It is just a hint; it imposes no requirements on the
999 This attribute suggests that optimization passes and code generator
1000 passes make choices that keep the code size of this function as small
1001 as possible and perform optimizations that may sacrifice runtime
1002 performance in order to minimize the size of the generated code.
1004 This attribute disables prologue / epilogue emission for the
1005 function. This can have very system-specific consequences.
1007 This indicates that the callee function at a call site is not recognized as
1008 a built-in function. LLVM will retain the original call and not replace it
1009 with equivalent code based on the semantics of the built-in function, unless
1010 the call site uses the ``builtin`` attribute. This is valid at call sites
1011 and on function declarations and definitions.
1013 This attribute indicates that calls to the function cannot be
1014 duplicated. A call to a ``noduplicate`` function may be moved
1015 within its parent function, but may not be duplicated within
1016 its parent function.
1018 A function containing a ``noduplicate`` call may still
1019 be an inlining candidate, provided that the call is not
1020 duplicated by inlining. That implies that the function has
1021 internal linkage and only has one call site, so the original
1022 call is dead after inlining.
1024 This attributes disables implicit floating point instructions.
1026 This attribute indicates that the inliner should never inline this
1027 function in any situation. This attribute may not be used together
1028 with the ``alwaysinline`` attribute.
1030 This attribute suppresses lazy symbol binding for the function. This
1031 may make calls to the function faster, at the cost of extra program
1032 startup time if the function is not called during program startup.
1034 This attribute indicates that the code generator should not use a
1035 red zone, even if the target-specific ABI normally permits it.
1037 This function attribute indicates that the function never returns
1038 normally. This produces undefined behavior at runtime if the
1039 function ever does dynamically return.
1041 This function attribute indicates that the function never returns
1042 with an unwind or exceptional control flow. If the function does
1043 unwind, its runtime behavior is undefined.
1045 This function attribute indicates that the function is not optimized
1046 by any optimization or code generator passes with the
1047 exception of interprocedural optimization passes.
1048 This attribute cannot be used together with the ``alwaysinline``
1049 attribute; this attribute is also incompatible
1050 with the ``minsize`` attribute and the ``optsize`` attribute.
1052 This attribute requires the ``noinline`` attribute to be specified on
1053 the function as well, so the function is never inlined into any caller.
1054 Only functions with the ``alwaysinline`` attribute are valid
1055 candidates for inlining into the body of this function.
1057 This attribute suggests that optimization passes and code generator
1058 passes make choices that keep the code size of this function low,
1059 and otherwise do optimizations specifically to reduce code size as
1060 long as they do not significantly impact runtime performance.
1062 On a function, this attribute indicates that the function computes its
1063 result (or decides to unwind an exception) based strictly on its arguments,
1064 without dereferencing any pointer arguments or otherwise accessing
1065 any mutable state (e.g. memory, control registers, etc) visible to
1066 caller functions. It does not write through any pointer arguments
1067 (including ``byval`` arguments) and never changes any state visible
1068 to callers. This means that it cannot unwind exceptions by calling
1069 the ``C++`` exception throwing methods.
1071 On an argument, this attribute indicates that the function does not
1072 dereference that pointer argument, even though it may read or write the
1073 memory that the pointer points to if accessed through other pointers.
1075 On a function, this attribute indicates that the function does not write
1076 through any pointer arguments (including ``byval`` arguments) or otherwise
1077 modify any state (e.g. memory, control registers, etc) visible to
1078 caller functions. It may dereference pointer arguments and read
1079 state that may be set in the caller. A readonly function always
1080 returns the same value (or unwinds an exception identically) when
1081 called with the same set of arguments and global state. It cannot
1082 unwind an exception by calling the ``C++`` exception throwing
1085 On an argument, this attribute indicates that the function does not write
1086 through this pointer argument, even though it may write to the memory that
1087 the pointer points to.
1089 This attribute indicates that this function can return twice. The C
1090 ``setjmp`` is an example of such a function. The compiler disables
1091 some optimizations (like tail calls) in the caller of these
1093 ``sanitize_address``
1094 This attribute indicates that AddressSanitizer checks
1095 (dynamic address safety analysis) are enabled for this function.
1097 This attribute indicates that MemorySanitizer checks (dynamic detection
1098 of accesses to uninitialized memory) are enabled for this function.
1100 This attribute indicates that ThreadSanitizer checks
1101 (dynamic thread safety analysis) are enabled for this function.
1103 This attribute indicates that the function should emit a stack
1104 smashing protector. It is in the form of a "canary" --- a random value
1105 placed on the stack before the local variables that's checked upon
1106 return from the function to see if it has been overwritten. A
1107 heuristic is used to determine if a function needs stack protectors
1108 or not. The heuristic used will enable protectors for functions with:
1110 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1111 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1112 - Calls to alloca() with variable sizes or constant sizes greater than
1113 ``ssp-buffer-size``.
1115 Variables that are identified as requiring a protector will be arranged
1116 on the stack such that they are adjacent to the stack protector guard.
1118 If a function that has an ``ssp`` attribute is inlined into a
1119 function that doesn't have an ``ssp`` attribute, then the resulting
1120 function will have an ``ssp`` attribute.
1122 This attribute indicates that the function should *always* emit a
1123 stack smashing protector. This overrides the ``ssp`` function
1126 Variables that are identified as requiring a protector will be arranged
1127 on the stack such that they are adjacent to the stack protector guard.
1128 The specific layout rules are:
1130 #. Large arrays and structures containing large arrays
1131 (``>= ssp-buffer-size``) are closest to the stack protector.
1132 #. Small arrays and structures containing small arrays
1133 (``< ssp-buffer-size``) are 2nd closest to the protector.
1134 #. Variables that have had their address taken are 3rd closest to the
1137 If a function that has an ``sspreq`` attribute is inlined into a
1138 function that doesn't have an ``sspreq`` attribute or which has an
1139 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1140 an ``sspreq`` attribute.
1142 This attribute indicates that the function should emit a stack smashing
1143 protector. This attribute causes a strong heuristic to be used when
1144 determining if a function needs stack protectors. The strong heuristic
1145 will enable protectors for functions with:
1147 - Arrays of any size and type
1148 - Aggregates containing an array of any size and type.
1149 - Calls to alloca().
1150 - Local variables that have had their address taken.
1152 Variables that are identified as requiring a protector will be arranged
1153 on the stack such that they are adjacent to the stack protector guard.
1154 The specific layout rules are:
1156 #. Large arrays and structures containing large arrays
1157 (``>= ssp-buffer-size``) are closest to the stack protector.
1158 #. Small arrays and structures containing small arrays
1159 (``< ssp-buffer-size``) are 2nd closest to the protector.
1160 #. Variables that have had their address taken are 3rd closest to the
1163 This overrides the ``ssp`` function attribute.
1165 If a function that has an ``sspstrong`` attribute is inlined into a
1166 function that doesn't have an ``sspstrong`` attribute, then the
1167 resulting function will have an ``sspstrong`` attribute.
1169 This attribute indicates that the ABI being targeted requires that
1170 an unwind table entry be produce for this function even if we can
1171 show that no exceptions passes by it. This is normally the case for
1172 the ELF x86-64 abi, but it can be disabled for some compilation
1177 Module-Level Inline Assembly
1178 ----------------------------
1180 Modules may contain "module-level inline asm" blocks, which corresponds
1181 to the GCC "file scope inline asm" blocks. These blocks are internally
1182 concatenated by LLVM and treated as a single unit, but may be separated
1183 in the ``.ll`` file if desired. The syntax is very simple:
1185 .. code-block:: llvm
1187 module asm "inline asm code goes here"
1188 module asm "more can go here"
1190 The strings can contain any character by escaping non-printable
1191 characters. The escape sequence used is simply "\\xx" where "xx" is the
1192 two digit hex code for the number.
1194 The inline asm code is simply printed to the machine code .s file when
1195 assembly code is generated.
1197 .. _langref_datalayout:
1202 A module may specify a target specific data layout string that specifies
1203 how data is to be laid out in memory. The syntax for the data layout is
1206 .. code-block:: llvm
1208 target datalayout = "layout specification"
1210 The *layout specification* consists of a list of specifications
1211 separated by the minus sign character ('-'). Each specification starts
1212 with a letter and may include other information after the letter to
1213 define some aspect of the data layout. The specifications accepted are
1217 Specifies that the target lays out data in big-endian form. That is,
1218 the bits with the most significance have the lowest address
1221 Specifies that the target lays out data in little-endian form. That
1222 is, the bits with the least significance have the lowest address
1225 Specifies the natural alignment of the stack in bits. Alignment
1226 promotion of stack variables is limited to the natural stack
1227 alignment to avoid dynamic stack realignment. The stack alignment
1228 must be a multiple of 8-bits. If omitted, the natural stack
1229 alignment defaults to "unspecified", which does not prevent any
1230 alignment promotions.
1231 ``p[n]:<size>:<abi>:<pref>``
1232 This specifies the *size* of a pointer and its ``<abi>`` and
1233 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1234 bits. The address space, ``n`` is optional, and if not specified,
1235 denotes the default address space 0. The value of ``n`` must be
1236 in the range [1,2^23).
1237 ``i<size>:<abi>:<pref>``
1238 This specifies the alignment for an integer type of a given bit
1239 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1240 ``v<size>:<abi>:<pref>``
1241 This specifies the alignment for a vector type of a given bit
1243 ``f<size>:<abi>:<pref>``
1244 This specifies the alignment for a floating point type of a given bit
1245 ``<size>``. Only values of ``<size>`` that are supported by the target
1246 will work. 32 (float) and 64 (double) are supported on all targets; 80
1247 or 128 (different flavors of long double) are also supported on some
1250 This specifies the alignment for an object of aggregate type.
1252 If present, specifies that llvm names are mangled in the output. The
1255 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1256 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1257 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1258 symbols get a ``_`` prefix.
1259 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1260 functions also get a suffix based on the frame size.
1261 ``n<size1>:<size2>:<size3>...``
1262 This specifies a set of native integer widths for the target CPU in
1263 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1264 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1265 this set are considered to support most general arithmetic operations
1268 On every specification that takes a ``<abi>:<pref>``, specifying the
1269 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1270 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1272 When constructing the data layout for a given target, LLVM starts with a
1273 default set of specifications which are then (possibly) overridden by
1274 the specifications in the ``datalayout`` keyword. The default
1275 specifications are given in this list:
1277 - ``E`` - big endian
1278 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1279 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1280 same as the default address space.
1281 - ``S0`` - natural stack alignment is unspecified
1282 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1283 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1284 - ``i16:16:16`` - i16 is 16-bit aligned
1285 - ``i32:32:32`` - i32 is 32-bit aligned
1286 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1287 alignment of 64-bits
1288 - ``f16:16:16`` - half is 16-bit aligned
1289 - ``f32:32:32`` - float is 32-bit aligned
1290 - ``f64:64:64`` - double is 64-bit aligned
1291 - ``f128:128:128`` - quad is 128-bit aligned
1292 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1293 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1294 - ``a:0:64`` - aggregates are 64-bit aligned
1296 When LLVM is determining the alignment for a given type, it uses the
1299 #. If the type sought is an exact match for one of the specifications,
1300 that specification is used.
1301 #. If no match is found, and the type sought is an integer type, then
1302 the smallest integer type that is larger than the bitwidth of the
1303 sought type is used. If none of the specifications are larger than
1304 the bitwidth then the largest integer type is used. For example,
1305 given the default specifications above, the i7 type will use the
1306 alignment of i8 (next largest) while both i65 and i256 will use the
1307 alignment of i64 (largest specified).
1308 #. If no match is found, and the type sought is a vector type, then the
1309 largest vector type that is smaller than the sought vector type will
1310 be used as a fall back. This happens because <128 x double> can be
1311 implemented in terms of 64 <2 x double>, for example.
1313 The function of the data layout string may not be what you expect.
1314 Notably, this is not a specification from the frontend of what alignment
1315 the code generator should use.
1317 Instead, if specified, the target data layout is required to match what
1318 the ultimate *code generator* expects. This string is used by the
1319 mid-level optimizers to improve code, and this only works if it matches
1320 what the ultimate code generator uses. If you would like to generate IR
1321 that does not embed this target-specific detail into the IR, then you
1322 don't have to specify the string. This will disable some optimizations
1323 that require precise layout information, but this also prevents those
1324 optimizations from introducing target specificity into the IR.
1331 A module may specify a target triple string that describes the target
1332 host. The syntax for the target triple is simply:
1334 .. code-block:: llvm
1336 target triple = "x86_64-apple-macosx10.7.0"
1338 The *target triple* string consists of a series of identifiers delimited
1339 by the minus sign character ('-'). The canonical forms are:
1343 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1344 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1346 This information is passed along to the backend so that it generates
1347 code for the proper architecture. It's possible to override this on the
1348 command line with the ``-mtriple`` command line option.
1350 .. _pointeraliasing:
1352 Pointer Aliasing Rules
1353 ----------------------
1355 Any memory access must be done through a pointer value associated with
1356 an address range of the memory access, otherwise the behavior is
1357 undefined. Pointer values are associated with address ranges according
1358 to the following rules:
1360 - A pointer value is associated with the addresses associated with any
1361 value it is *based* on.
1362 - An address of a global variable is associated with the address range
1363 of the variable's storage.
1364 - The result value of an allocation instruction is associated with the
1365 address range of the allocated storage.
1366 - A null pointer in the default address-space is associated with no
1368 - An integer constant other than zero or a pointer value returned from
1369 a function not defined within LLVM may be associated with address
1370 ranges allocated through mechanisms other than those provided by
1371 LLVM. Such ranges shall not overlap with any ranges of addresses
1372 allocated by mechanisms provided by LLVM.
1374 A pointer value is *based* on another pointer value according to the
1377 - A pointer value formed from a ``getelementptr`` operation is *based*
1378 on the first operand of the ``getelementptr``.
1379 - The result value of a ``bitcast`` is *based* on the operand of the
1381 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1382 values that contribute (directly or indirectly) to the computation of
1383 the pointer's value.
1384 - The "*based* on" relationship is transitive.
1386 Note that this definition of *"based"* is intentionally similar to the
1387 definition of *"based"* in C99, though it is slightly weaker.
1389 LLVM IR does not associate types with memory. The result type of a
1390 ``load`` merely indicates the size and alignment of the memory from
1391 which to load, as well as the interpretation of the value. The first
1392 operand type of a ``store`` similarly only indicates the size and
1393 alignment of the store.
1395 Consequently, type-based alias analysis, aka TBAA, aka
1396 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1397 :ref:`Metadata <metadata>` may be used to encode additional information
1398 which specialized optimization passes may use to implement type-based
1403 Volatile Memory Accesses
1404 ------------------------
1406 Certain memory accesses, such as :ref:`load <i_load>`'s,
1407 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1408 marked ``volatile``. The optimizers must not change the number of
1409 volatile operations or change their order of execution relative to other
1410 volatile operations. The optimizers *may* change the order of volatile
1411 operations relative to non-volatile operations. This is not Java's
1412 "volatile" and has no cross-thread synchronization behavior.
1414 IR-level volatile loads and stores cannot safely be optimized into
1415 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1416 flagged volatile. Likewise, the backend should never split or merge
1417 target-legal volatile load/store instructions.
1419 .. admonition:: Rationale
1421 Platforms may rely on volatile loads and stores of natively supported
1422 data width to be executed as single instruction. For example, in C
1423 this holds for an l-value of volatile primitive type with native
1424 hardware support, but not necessarily for aggregate types. The
1425 frontend upholds these expectations, which are intentionally
1426 unspecified in the IR. The rules above ensure that IR transformation
1427 do not violate the frontend's contract with the language.
1431 Memory Model for Concurrent Operations
1432 --------------------------------------
1434 The LLVM IR does not define any way to start parallel threads of
1435 execution or to register signal handlers. Nonetheless, there are
1436 platform-specific ways to create them, and we define LLVM IR's behavior
1437 in their presence. This model is inspired by the C++0x memory model.
1439 For a more informal introduction to this model, see the :doc:`Atomics`.
1441 We define a *happens-before* partial order as the least partial order
1444 - Is a superset of single-thread program order, and
1445 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1446 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1447 techniques, like pthread locks, thread creation, thread joining,
1448 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1449 Constraints <ordering>`).
1451 Note that program order does not introduce *happens-before* edges
1452 between a thread and signals executing inside that thread.
1454 Every (defined) read operation (load instructions, memcpy, atomic
1455 loads/read-modify-writes, etc.) R reads a series of bytes written by
1456 (defined) write operations (store instructions, atomic
1457 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1458 section, initialized globals are considered to have a write of the
1459 initializer which is atomic and happens before any other read or write
1460 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1461 may see any write to the same byte, except:
1463 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1464 write\ :sub:`2` happens before R\ :sub:`byte`, then
1465 R\ :sub:`byte` does not see write\ :sub:`1`.
1466 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1467 R\ :sub:`byte` does not see write\ :sub:`3`.
1469 Given that definition, R\ :sub:`byte` is defined as follows:
1471 - If R is volatile, the result is target-dependent. (Volatile is
1472 supposed to give guarantees which can support ``sig_atomic_t`` in
1473 C/C++, and may be used for accesses to addresses which do not behave
1474 like normal memory. It does not generally provide cross-thread
1476 - Otherwise, if there is no write to the same byte that happens before
1477 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1478 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1479 R\ :sub:`byte` returns the value written by that write.
1480 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1481 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1482 Memory Ordering Constraints <ordering>` section for additional
1483 constraints on how the choice is made.
1484 - Otherwise R\ :sub:`byte` returns ``undef``.
1486 R returns the value composed of the series of bytes it read. This
1487 implies that some bytes within the value may be ``undef`` **without**
1488 the entire value being ``undef``. Note that this only defines the
1489 semantics of the operation; it doesn't mean that targets will emit more
1490 than one instruction to read the series of bytes.
1492 Note that in cases where none of the atomic intrinsics are used, this
1493 model places only one restriction on IR transformations on top of what
1494 is required for single-threaded execution: introducing a store to a byte
1495 which might not otherwise be stored is not allowed in general.
1496 (Specifically, in the case where another thread might write to and read
1497 from an address, introducing a store can change a load that may see
1498 exactly one write into a load that may see multiple writes.)
1502 Atomic Memory Ordering Constraints
1503 ----------------------------------
1505 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1506 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1507 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1508 ordering parameters that determine which other atomic instructions on
1509 the same address they *synchronize with*. These semantics are borrowed
1510 from Java and C++0x, but are somewhat more colloquial. If these
1511 descriptions aren't precise enough, check those specs (see spec
1512 references in the :doc:`atomics guide <Atomics>`).
1513 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1514 differently since they don't take an address. See that instruction's
1515 documentation for details.
1517 For a simpler introduction to the ordering constraints, see the
1521 The set of values that can be read is governed by the happens-before
1522 partial order. A value cannot be read unless some operation wrote
1523 it. This is intended to provide a guarantee strong enough to model
1524 Java's non-volatile shared variables. This ordering cannot be
1525 specified for read-modify-write operations; it is not strong enough
1526 to make them atomic in any interesting way.
1528 In addition to the guarantees of ``unordered``, there is a single
1529 total order for modifications by ``monotonic`` operations on each
1530 address. All modification orders must be compatible with the
1531 happens-before order. There is no guarantee that the modification
1532 orders can be combined to a global total order for the whole program
1533 (and this often will not be possible). The read in an atomic
1534 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1535 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1536 order immediately before the value it writes. If one atomic read
1537 happens before another atomic read of the same address, the later
1538 read must see the same value or a later value in the address's
1539 modification order. This disallows reordering of ``monotonic`` (or
1540 stronger) operations on the same address. If an address is written
1541 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1542 read that address repeatedly, the other threads must eventually see
1543 the write. This corresponds to the C++0x/C1x
1544 ``memory_order_relaxed``.
1546 In addition to the guarantees of ``monotonic``, a
1547 *synchronizes-with* edge may be formed with a ``release`` operation.
1548 This is intended to model C++'s ``memory_order_acquire``.
1550 In addition to the guarantees of ``monotonic``, if this operation
1551 writes a value which is subsequently read by an ``acquire``
1552 operation, it *synchronizes-with* that operation. (This isn't a
1553 complete description; see the C++0x definition of a release
1554 sequence.) This corresponds to the C++0x/C1x
1555 ``memory_order_release``.
1556 ``acq_rel`` (acquire+release)
1557 Acts as both an ``acquire`` and ``release`` operation on its
1558 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1559 ``seq_cst`` (sequentially consistent)
1560 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1561 operation which only reads, ``release`` for an operation which only
1562 writes), there is a global total order on all
1563 sequentially-consistent operations on all addresses, which is
1564 consistent with the *happens-before* partial order and with the
1565 modification orders of all the affected addresses. Each
1566 sequentially-consistent read sees the last preceding write to the
1567 same address in this global order. This corresponds to the C++0x/C1x
1568 ``memory_order_seq_cst`` and Java volatile.
1572 If an atomic operation is marked ``singlethread``, it only *synchronizes
1573 with* or participates in modification and seq\_cst total orderings with
1574 other operations running in the same thread (for example, in signal
1582 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1583 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1584 :ref:`frem <i_frem>`) have the following flags that can set to enable
1585 otherwise unsafe floating point operations
1588 No NaNs - Allow optimizations to assume the arguments and result are not
1589 NaN. Such optimizations are required to retain defined behavior over
1590 NaNs, but the value of the result is undefined.
1593 No Infs - Allow optimizations to assume the arguments and result are not
1594 +/-Inf. Such optimizations are required to retain defined behavior over
1595 +/-Inf, but the value of the result is undefined.
1598 No Signed Zeros - Allow optimizations to treat the sign of a zero
1599 argument or result as insignificant.
1602 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1603 argument rather than perform division.
1606 Fast - Allow algebraically equivalent transformations that may
1607 dramatically change results in floating point (e.g. reassociate). This
1608 flag implies all the others.
1615 The LLVM type system is one of the most important features of the
1616 intermediate representation. Being typed enables a number of
1617 optimizations to be performed on the intermediate representation
1618 directly, without having to do extra analyses on the side before the
1619 transformation. A strong type system makes it easier to read the
1620 generated code and enables novel analyses and transformations that are
1621 not feasible to perform on normal three address code representations.
1631 The void type does not represent any value and has no size.
1649 The function type can be thought of as a function signature. It consists of a
1650 return type and a list of formal parameter types. The return type of a function
1651 type is a void type or first class type --- except for :ref:`label <t_label>`
1652 and :ref:`metadata <t_metadata>` types.
1658 <returntype> (<parameter list>)
1660 ...where '``<parameter list>``' is a comma-separated list of type
1661 specifiers. Optionally, the parameter list may include a type ``...``, which
1662 indicates that the function takes a variable number of arguments. Variable
1663 argument functions can access their arguments with the :ref:`variable argument
1664 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1665 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1669 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1670 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1671 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1672 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1673 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1674 | ``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. |
1675 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1676 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1677 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1684 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1685 Values of these types are the only ones which can be produced by
1693 These are the types that are valid in registers from CodeGen's perspective.
1702 The integer type is a very simple type that simply specifies an
1703 arbitrary bit width for the integer type desired. Any bit width from 1
1704 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1712 The number of bits the integer will occupy is specified by the ``N``
1718 +----------------+------------------------------------------------+
1719 | ``i1`` | a single-bit integer. |
1720 +----------------+------------------------------------------------+
1721 | ``i32`` | a 32-bit integer. |
1722 +----------------+------------------------------------------------+
1723 | ``i1942652`` | a really big integer of over 1 million bits. |
1724 +----------------+------------------------------------------------+
1728 Floating Point Types
1729 """"""""""""""""""""
1738 - 16-bit floating point value
1741 - 32-bit floating point value
1744 - 64-bit floating point value
1747 - 128-bit floating point value (112-bit mantissa)
1750 - 80-bit floating point value (X87)
1753 - 128-bit floating point value (two 64-bits)
1760 The x86_mmx type represents a value held in an MMX register on an x86
1761 machine. The operations allowed on it are quite limited: parameters and
1762 return values, load and store, and bitcast. User-specified MMX
1763 instructions are represented as intrinsic or asm calls with arguments
1764 and/or results of this type. There are no arrays, vectors or constants
1781 The pointer type is used to specify memory locations. Pointers are
1782 commonly used to reference objects in memory.
1784 Pointer types may have an optional address space attribute defining the
1785 numbered address space where the pointed-to object resides. The default
1786 address space is number zero. The semantics of non-zero address spaces
1787 are target-specific.
1789 Note that LLVM does not permit pointers to void (``void*``) nor does it
1790 permit pointers to labels (``label*``). Use ``i8*`` instead.
1800 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1801 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1802 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1803 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1804 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1805 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1806 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1815 A vector type is a simple derived type that represents a vector of
1816 elements. Vector types are used when multiple primitive data are
1817 operated in parallel using a single instruction (SIMD). A vector type
1818 requires a size (number of elements) and an underlying primitive data
1819 type. Vector types are considered :ref:`first class <t_firstclass>`.
1825 < <# elements> x <elementtype> >
1827 The number of elements is a constant integer value larger than 0;
1828 elementtype may be any integer or floating point type, or a pointer to
1829 these types. Vectors of size zero are not allowed.
1833 +-------------------+--------------------------------------------------+
1834 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1835 +-------------------+--------------------------------------------------+
1836 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1837 +-------------------+--------------------------------------------------+
1838 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1839 +-------------------+--------------------------------------------------+
1840 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1841 +-------------------+--------------------------------------------------+
1850 The label type represents code labels.
1865 The metadata type represents embedded metadata. No derived types may be
1866 created from metadata except for :ref:`function <t_function>` arguments.
1879 Aggregate Types are a subset of derived types that can contain multiple
1880 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1881 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1891 The array type is a very simple derived type that arranges elements
1892 sequentially in memory. The array type requires a size (number of
1893 elements) and an underlying data type.
1899 [<# elements> x <elementtype>]
1901 The number of elements is a constant integer value; ``elementtype`` may
1902 be any type with a size.
1906 +------------------+--------------------------------------+
1907 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1908 +------------------+--------------------------------------+
1909 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1910 +------------------+--------------------------------------+
1911 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1912 +------------------+--------------------------------------+
1914 Here are some examples of multidimensional arrays:
1916 +-----------------------------+----------------------------------------------------------+
1917 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1918 +-----------------------------+----------------------------------------------------------+
1919 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1920 +-----------------------------+----------------------------------------------------------+
1921 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1922 +-----------------------------+----------------------------------------------------------+
1924 There is no restriction on indexing beyond the end of the array implied
1925 by a static type (though there are restrictions on indexing beyond the
1926 bounds of an allocated object in some cases). This means that
1927 single-dimension 'variable sized array' addressing can be implemented in
1928 LLVM with a zero length array type. An implementation of 'pascal style
1929 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1939 The structure type is used to represent a collection of data members
1940 together in memory. The elements of a structure may be any type that has
1943 Structures in memory are accessed using '``load``' and '``store``' by
1944 getting a pointer to a field with the '``getelementptr``' instruction.
1945 Structures in registers are accessed using the '``extractvalue``' and
1946 '``insertvalue``' instructions.
1948 Structures may optionally be "packed" structures, which indicate that
1949 the alignment of the struct is one byte, and that there is no padding
1950 between the elements. In non-packed structs, padding between field types
1951 is inserted as defined by the DataLayout string in the module, which is
1952 required to match what the underlying code generator expects.
1954 Structures can either be "literal" or "identified". A literal structure
1955 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1956 identified types are always defined at the top level with a name.
1957 Literal types are uniqued by their contents and can never be recursive
1958 or opaque since there is no way to write one. Identified types can be
1959 recursive, can be opaqued, and are never uniqued.
1965 %T1 = type { <type list> } ; Identified normal struct type
1966 %T2 = type <{ <type list> }> ; Identified packed struct type
1970 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1971 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1972 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1973 | ``{ 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``. |
1974 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1975 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1976 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1980 Opaque Structure Types
1981 """"""""""""""""""""""
1985 Opaque structure types are used to represent named structure types that
1986 do not have a body specified. This corresponds (for example) to the C
1987 notion of a forward declared structure.
1998 +--------------+-------------------+
1999 | ``opaque`` | An opaque type. |
2000 +--------------+-------------------+
2007 LLVM has several different basic types of constants. This section
2008 describes them all and their syntax.
2013 **Boolean constants**
2014 The two strings '``true``' and '``false``' are both valid constants
2016 **Integer constants**
2017 Standard integers (such as '4') are constants of the
2018 :ref:`integer <t_integer>` type. Negative numbers may be used with
2020 **Floating point constants**
2021 Floating point constants use standard decimal notation (e.g.
2022 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2023 hexadecimal notation (see below). The assembler requires the exact
2024 decimal value of a floating-point constant. For example, the
2025 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2026 decimal in binary. Floating point constants must have a :ref:`floating
2027 point <t_floating>` type.
2028 **Null pointer constants**
2029 The identifier '``null``' is recognized as a null pointer constant
2030 and must be of :ref:`pointer type <t_pointer>`.
2032 The one non-intuitive notation for constants is the hexadecimal form of
2033 floating point constants. For example, the form
2034 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2035 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2036 constants are required (and the only time that they are generated by the
2037 disassembler) is when a floating point constant must be emitted but it
2038 cannot be represented as a decimal floating point number in a reasonable
2039 number of digits. For example, NaN's, infinities, and other special
2040 values are represented in their IEEE hexadecimal format so that assembly
2041 and disassembly do not cause any bits to change in the constants.
2043 When using the hexadecimal form, constants of types half, float, and
2044 double are represented using the 16-digit form shown above (which
2045 matches the IEEE754 representation for double); half and float values
2046 must, however, be exactly representable as IEEE 754 half and single
2047 precision, respectively. Hexadecimal format is always used for long
2048 double, and there are three forms of long double. The 80-bit format used
2049 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2050 128-bit format used by PowerPC (two adjacent doubles) is represented by
2051 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2052 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2053 will only work if they match the long double format on your target.
2054 The IEEE 16-bit format (half precision) is represented by ``0xH``
2055 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2056 (sign bit at the left).
2058 There are no constants of type x86_mmx.
2060 .. _complexconstants:
2065 Complex constants are a (potentially recursive) combination of simple
2066 constants and smaller complex constants.
2068 **Structure constants**
2069 Structure constants are represented with notation similar to
2070 structure type definitions (a comma separated list of elements,
2071 surrounded by braces (``{}``)). For example:
2072 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2073 "``@G = external global i32``". Structure constants must have
2074 :ref:`structure type <t_struct>`, and the number and types of elements
2075 must match those specified by the type.
2077 Array constants are represented with notation similar to array type
2078 definitions (a comma separated list of elements, surrounded by
2079 square brackets (``[]``)). For example:
2080 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2081 :ref:`array type <t_array>`, and the number and types of elements must
2082 match those specified by the type.
2083 **Vector constants**
2084 Vector constants are represented with notation similar to vector
2085 type definitions (a comma separated list of elements, surrounded by
2086 less-than/greater-than's (``<>``)). For example:
2087 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2088 must have :ref:`vector type <t_vector>`, and the number and types of
2089 elements must match those specified by the type.
2090 **Zero initialization**
2091 The string '``zeroinitializer``' can be used to zero initialize a
2092 value to zero of *any* type, including scalar and
2093 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2094 having to print large zero initializers (e.g. for large arrays) and
2095 is always exactly equivalent to using explicit zero initializers.
2097 A metadata node is a structure-like constant with :ref:`metadata
2098 type <t_metadata>`. For example:
2099 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2100 constants that are meant to be interpreted as part of the
2101 instruction stream, metadata is a place to attach additional
2102 information such as debug info.
2104 Global Variable and Function Addresses
2105 --------------------------------------
2107 The addresses of :ref:`global variables <globalvars>` and
2108 :ref:`functions <functionstructure>` are always implicitly valid
2109 (link-time) constants. These constants are explicitly referenced when
2110 the :ref:`identifier for the global <identifiers>` is used and always have
2111 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2114 .. code-block:: llvm
2118 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2125 The string '``undef``' can be used anywhere a constant is expected, and
2126 indicates that the user of the value may receive an unspecified
2127 bit-pattern. Undefined values may be of any type (other than '``label``'
2128 or '``void``') and be used anywhere a constant is permitted.
2130 Undefined values are useful because they indicate to the compiler that
2131 the program is well defined no matter what value is used. This gives the
2132 compiler more freedom to optimize. Here are some examples of
2133 (potentially surprising) transformations that are valid (in pseudo IR):
2135 .. code-block:: llvm
2145 This is safe because all of the output bits are affected by the undef
2146 bits. Any output bit can have a zero or one depending on the input bits.
2148 .. code-block:: llvm
2159 These logical operations have bits that are not always affected by the
2160 input. For example, if ``%X`` has a zero bit, then the output of the
2161 '``and``' operation will always be a zero for that bit, no matter what
2162 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2163 optimize or assume that the result of the '``and``' is '``undef``'.
2164 However, it is safe to assume that all bits of the '``undef``' could be
2165 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2166 all the bits of the '``undef``' operand to the '``or``' could be set,
2167 allowing the '``or``' to be folded to -1.
2169 .. code-block:: llvm
2171 %A = select undef, %X, %Y
2172 %B = select undef, 42, %Y
2173 %C = select %X, %Y, undef
2183 This set of examples shows that undefined '``select``' (and conditional
2184 branch) conditions can go *either way*, but they have to come from one
2185 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2186 both known to have a clear low bit, then ``%A`` would have to have a
2187 cleared low bit. However, in the ``%C`` example, the optimizer is
2188 allowed to assume that the '``undef``' operand could be the same as
2189 ``%Y``, allowing the whole '``select``' to be eliminated.
2191 .. code-block:: llvm
2193 %A = xor undef, undef
2210 This example points out that two '``undef``' operands are not
2211 necessarily the same. This can be surprising to people (and also matches
2212 C semantics) where they assume that "``X^X``" is always zero, even if
2213 ``X`` is undefined. This isn't true for a number of reasons, but the
2214 short answer is that an '``undef``' "variable" can arbitrarily change
2215 its value over its "live range". This is true because the variable
2216 doesn't actually *have a live range*. Instead, the value is logically
2217 read from arbitrary registers that happen to be around when needed, so
2218 the value is not necessarily consistent over time. In fact, ``%A`` and
2219 ``%C`` need to have the same semantics or the core LLVM "replace all
2220 uses with" concept would not hold.
2222 .. code-block:: llvm
2230 These examples show the crucial difference between an *undefined value*
2231 and *undefined behavior*. An undefined value (like '``undef``') is
2232 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2233 operation can be constant folded to '``undef``', because the '``undef``'
2234 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2235 However, in the second example, we can make a more aggressive
2236 assumption: because the ``undef`` is allowed to be an arbitrary value,
2237 we are allowed to assume that it could be zero. Since a divide by zero
2238 has *undefined behavior*, we are allowed to assume that the operation
2239 does not execute at all. This allows us to delete the divide and all
2240 code after it. Because the undefined operation "can't happen", the
2241 optimizer can assume that it occurs in dead code.
2243 .. code-block:: llvm
2245 a: store undef -> %X
2246 b: store %X -> undef
2251 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2252 value can be assumed to not have any effect; we can assume that the
2253 value is overwritten with bits that happen to match what was already
2254 there. However, a store *to* an undefined location could clobber
2255 arbitrary memory, therefore, it has undefined behavior.
2262 Poison values are similar to :ref:`undef values <undefvalues>`, however
2263 they also represent the fact that an instruction or constant expression
2264 which cannot evoke side effects has nevertheless detected a condition
2265 which results in undefined behavior.
2267 There is currently no way of representing a poison value in the IR; they
2268 only exist when produced by operations such as :ref:`add <i_add>` with
2271 Poison value behavior is defined in terms of value *dependence*:
2273 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2274 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2275 their dynamic predecessor basic block.
2276 - Function arguments depend on the corresponding actual argument values
2277 in the dynamic callers of their functions.
2278 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2279 instructions that dynamically transfer control back to them.
2280 - :ref:`Invoke <i_invoke>` instructions depend on the
2281 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2282 call instructions that dynamically transfer control back to them.
2283 - Non-volatile loads and stores depend on the most recent stores to all
2284 of the referenced memory addresses, following the order in the IR
2285 (including loads and stores implied by intrinsics such as
2286 :ref:`@llvm.memcpy <int_memcpy>`.)
2287 - An instruction with externally visible side effects depends on the
2288 most recent preceding instruction with externally visible side
2289 effects, following the order in the IR. (This includes :ref:`volatile
2290 operations <volatile>`.)
2291 - An instruction *control-depends* on a :ref:`terminator
2292 instruction <terminators>` if the terminator instruction has
2293 multiple successors and the instruction is always executed when
2294 control transfers to one of the successors, and may not be executed
2295 when control is transferred to another.
2296 - Additionally, an instruction also *control-depends* on a terminator
2297 instruction if the set of instructions it otherwise depends on would
2298 be different if the terminator had transferred control to a different
2300 - Dependence is transitive.
2302 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2303 with the additional affect that any instruction which has a *dependence*
2304 on a poison value has undefined behavior.
2306 Here are some examples:
2308 .. code-block:: llvm
2311 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2312 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2313 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2314 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2316 store i32 %poison, i32* @g ; Poison value stored to memory.
2317 %poison2 = load i32* @g ; Poison value loaded back from memory.
2319 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2321 %narrowaddr = bitcast i32* @g to i16*
2322 %wideaddr = bitcast i32* @g to i64*
2323 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2324 %poison4 = load i64* %wideaddr ; Returns a poison value.
2326 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2327 br i1 %cmp, label %true, label %end ; Branch to either destination.
2330 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2331 ; it has undefined behavior.
2335 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2336 ; Both edges into this PHI are
2337 ; control-dependent on %cmp, so this
2338 ; always results in a poison value.
2340 store volatile i32 0, i32* @g ; This would depend on the store in %true
2341 ; if %cmp is true, or the store in %entry
2342 ; otherwise, so this is undefined behavior.
2344 br i1 %cmp, label %second_true, label %second_end
2345 ; The same branch again, but this time the
2346 ; true block doesn't have side effects.
2353 store volatile i32 0, i32* @g ; This time, the instruction always depends
2354 ; on the store in %end. Also, it is
2355 ; control-equivalent to %end, so this is
2356 ; well-defined (ignoring earlier undefined
2357 ; behavior in this example).
2361 Addresses of Basic Blocks
2362 -------------------------
2364 ``blockaddress(@function, %block)``
2366 The '``blockaddress``' constant computes the address of the specified
2367 basic block in the specified function, and always has an ``i8*`` type.
2368 Taking the address of the entry block is illegal.
2370 This value only has defined behavior when used as an operand to the
2371 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2372 against null. Pointer equality tests between labels addresses results in
2373 undefined behavior --- though, again, comparison against null is ok, and
2374 no label is equal to the null pointer. This may be passed around as an
2375 opaque pointer sized value as long as the bits are not inspected. This
2376 allows ``ptrtoint`` and arithmetic to be performed on these values so
2377 long as the original value is reconstituted before the ``indirectbr``
2380 Finally, some targets may provide defined semantics when using the value
2381 as the operand to an inline assembly, but that is target specific.
2385 Constant Expressions
2386 --------------------
2388 Constant expressions are used to allow expressions involving other
2389 constants to be used as constants. Constant expressions may be of any
2390 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2391 that does not have side effects (e.g. load and call are not supported).
2392 The following is the syntax for constant expressions:
2394 ``trunc (CST to TYPE)``
2395 Truncate a constant to another type. The bit size of CST must be
2396 larger than the bit size of TYPE. Both types must be integers.
2397 ``zext (CST to TYPE)``
2398 Zero extend a constant to another type. The bit size of CST must be
2399 smaller than the bit size of TYPE. Both types must be integers.
2400 ``sext (CST to TYPE)``
2401 Sign extend a constant to another type. The bit size of CST must be
2402 smaller than the bit size of TYPE. Both types must be integers.
2403 ``fptrunc (CST to TYPE)``
2404 Truncate a floating point constant to another floating point type.
2405 The size of CST must be larger than the size of TYPE. Both types
2406 must be floating point.
2407 ``fpext (CST to TYPE)``
2408 Floating point extend a constant to another type. The size of CST
2409 must be smaller or equal to the size of TYPE. Both types must be
2411 ``fptoui (CST to TYPE)``
2412 Convert a floating point constant to the corresponding unsigned
2413 integer constant. TYPE must be a scalar or vector integer type. CST
2414 must be of scalar or vector floating point type. Both CST and TYPE
2415 must be scalars, or vectors of the same number of elements. If the
2416 value won't fit in the integer type, the results are undefined.
2417 ``fptosi (CST to TYPE)``
2418 Convert a floating point constant to the corresponding signed
2419 integer constant. TYPE must be a scalar or vector integer type. CST
2420 must be of scalar or vector floating point type. Both CST and TYPE
2421 must be scalars, or vectors of the same number of elements. If the
2422 value won't fit in the integer type, the results are undefined.
2423 ``uitofp (CST to TYPE)``
2424 Convert an unsigned integer constant to the corresponding floating
2425 point constant. TYPE must be a scalar or vector floating point type.
2426 CST must be of scalar or vector integer type. Both CST and TYPE must
2427 be scalars, or vectors of the same number of elements. If the value
2428 won't fit in the floating point type, the results are undefined.
2429 ``sitofp (CST to TYPE)``
2430 Convert a signed integer constant to the corresponding floating
2431 point constant. TYPE must be a scalar or vector floating point type.
2432 CST must be of scalar or vector integer type. Both CST and TYPE must
2433 be scalars, or vectors of the same number of elements. If the value
2434 won't fit in the floating point type, the results are undefined.
2435 ``ptrtoint (CST to TYPE)``
2436 Convert a pointer typed constant to the corresponding integer
2437 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2438 pointer type. The ``CST`` value is zero extended, truncated, or
2439 unchanged to make it fit in ``TYPE``.
2440 ``inttoptr (CST to TYPE)``
2441 Convert an integer constant to a pointer constant. TYPE must be a
2442 pointer type. CST must be of integer type. The CST value is zero
2443 extended, truncated, or unchanged to make it fit in a pointer size.
2444 This one is *really* dangerous!
2445 ``bitcast (CST to TYPE)``
2446 Convert a constant, CST, to another TYPE. The constraints of the
2447 operands are the same as those for the :ref:`bitcast
2448 instruction <i_bitcast>`.
2449 ``addrspacecast (CST to TYPE)``
2450 Convert a constant pointer or constant vector of pointer, CST, to another
2451 TYPE in a different address space. The constraints of the operands are the
2452 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2453 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2454 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2455 constants. As with the :ref:`getelementptr <i_getelementptr>`
2456 instruction, the index list may have zero or more indexes, which are
2457 required to make sense for the type of "CSTPTR".
2458 ``select (COND, VAL1, VAL2)``
2459 Perform the :ref:`select operation <i_select>` on constants.
2460 ``icmp COND (VAL1, VAL2)``
2461 Performs the :ref:`icmp operation <i_icmp>` on constants.
2462 ``fcmp COND (VAL1, VAL2)``
2463 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2464 ``extractelement (VAL, IDX)``
2465 Perform the :ref:`extractelement operation <i_extractelement>` on
2467 ``insertelement (VAL, ELT, IDX)``
2468 Perform the :ref:`insertelement operation <i_insertelement>` on
2470 ``shufflevector (VEC1, VEC2, IDXMASK)``
2471 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2473 ``extractvalue (VAL, IDX0, IDX1, ...)``
2474 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2475 constants. The index list is interpreted in a similar manner as
2476 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2477 least one index value must be specified.
2478 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2479 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2480 The index list is interpreted in a similar manner as indices in a
2481 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2482 value must be specified.
2483 ``OPCODE (LHS, RHS)``
2484 Perform the specified operation of the LHS and RHS constants. OPCODE
2485 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2486 binary <bitwiseops>` operations. The constraints on operands are
2487 the same as those for the corresponding instruction (e.g. no bitwise
2488 operations on floating point values are allowed).
2495 Inline Assembler Expressions
2496 ----------------------------
2498 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2499 Inline Assembly <moduleasm>`) through the use of a special value. This
2500 value represents the inline assembler as a string (containing the
2501 instructions to emit), a list of operand constraints (stored as a
2502 string), a flag that indicates whether or not the inline asm expression
2503 has side effects, and a flag indicating whether the function containing
2504 the asm needs to align its stack conservatively. An example inline
2505 assembler expression is:
2507 .. code-block:: llvm
2509 i32 (i32) asm "bswap $0", "=r,r"
2511 Inline assembler expressions may **only** be used as the callee operand
2512 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2513 Thus, typically we have:
2515 .. code-block:: llvm
2517 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2519 Inline asms with side effects not visible in the constraint list must be
2520 marked as having side effects. This is done through the use of the
2521 '``sideeffect``' keyword, like so:
2523 .. code-block:: llvm
2525 call void asm sideeffect "eieio", ""()
2527 In some cases inline asms will contain code that will not work unless
2528 the stack is aligned in some way, such as calls or SSE instructions on
2529 x86, yet will not contain code that does that alignment within the asm.
2530 The compiler should make conservative assumptions about what the asm
2531 might contain and should generate its usual stack alignment code in the
2532 prologue if the '``alignstack``' keyword is present:
2534 .. code-block:: llvm
2536 call void asm alignstack "eieio", ""()
2538 Inline asms also support using non-standard assembly dialects. The
2539 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2540 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2541 the only supported dialects. An example is:
2543 .. code-block:: llvm
2545 call void asm inteldialect "eieio", ""()
2547 If multiple keywords appear the '``sideeffect``' keyword must come
2548 first, the '``alignstack``' keyword second and the '``inteldialect``'
2554 The call instructions that wrap inline asm nodes may have a
2555 "``!srcloc``" MDNode attached to it that contains a list of constant
2556 integers. If present, the code generator will use the integer as the
2557 location cookie value when report errors through the ``LLVMContext``
2558 error reporting mechanisms. This allows a front-end to correlate backend
2559 errors that occur with inline asm back to the source code that produced
2562 .. code-block:: llvm
2564 call void asm sideeffect "something bad", ""(), !srcloc !42
2566 !42 = !{ i32 1234567 }
2568 It is up to the front-end to make sense of the magic numbers it places
2569 in the IR. If the MDNode contains multiple constants, the code generator
2570 will use the one that corresponds to the line of the asm that the error
2575 Metadata Nodes and Metadata Strings
2576 -----------------------------------
2578 LLVM IR allows metadata to be attached to instructions in the program
2579 that can convey extra information about the code to the optimizers and
2580 code generator. One example application of metadata is source-level
2581 debug information. There are two metadata primitives: strings and nodes.
2582 All metadata has the ``metadata`` type and is identified in syntax by a
2583 preceding exclamation point ('``!``').
2585 A metadata string is a string surrounded by double quotes. It can
2586 contain any character by escaping non-printable characters with
2587 "``\xx``" where "``xx``" is the two digit hex code. For example:
2590 Metadata nodes are represented with notation similar to structure
2591 constants (a comma separated list of elements, surrounded by braces and
2592 preceded by an exclamation point). Metadata nodes can have any values as
2593 their operand. For example:
2595 .. code-block:: llvm
2597 !{ metadata !"test\00", i32 10}
2599 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2600 metadata nodes, which can be looked up in the module symbol table. For
2603 .. code-block:: llvm
2605 !foo = metadata !{!4, !3}
2607 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2608 function is using two metadata arguments:
2610 .. code-block:: llvm
2612 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2614 Metadata can be attached with an instruction. Here metadata ``!21`` is
2615 attached to the ``add`` instruction using the ``!dbg`` identifier:
2617 .. code-block:: llvm
2619 %indvar.next = add i64 %indvar, 1, !dbg !21
2621 More information about specific metadata nodes recognized by the
2622 optimizers and code generator is found below.
2627 In LLVM IR, memory does not have types, so LLVM's own type system is not
2628 suitable for doing TBAA. Instead, metadata is added to the IR to
2629 describe a type system of a higher level language. This can be used to
2630 implement typical C/C++ TBAA, but it can also be used to implement
2631 custom alias analysis behavior for other languages.
2633 The current metadata format is very simple. TBAA metadata nodes have up
2634 to three fields, e.g.:
2636 .. code-block:: llvm
2638 !0 = metadata !{ metadata !"an example type tree" }
2639 !1 = metadata !{ metadata !"int", metadata !0 }
2640 !2 = metadata !{ metadata !"float", metadata !0 }
2641 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2643 The first field is an identity field. It can be any value, usually a
2644 metadata string, which uniquely identifies the type. The most important
2645 name in the tree is the name of the root node. Two trees with different
2646 root node names are entirely disjoint, even if they have leaves with
2649 The second field identifies the type's parent node in the tree, or is
2650 null or omitted for a root node. A type is considered to alias all of
2651 its descendants and all of its ancestors in the tree. Also, a type is
2652 considered to alias all types in other trees, so that bitcode produced
2653 from multiple front-ends is handled conservatively.
2655 If the third field is present, it's an integer which if equal to 1
2656 indicates that the type is "constant" (meaning
2657 ``pointsToConstantMemory`` should return true; see `other useful
2658 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2660 '``tbaa.struct``' Metadata
2661 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2663 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2664 aggregate assignment operations in C and similar languages, however it
2665 is defined to copy a contiguous region of memory, which is more than
2666 strictly necessary for aggregate types which contain holes due to
2667 padding. Also, it doesn't contain any TBAA information about the fields
2670 ``!tbaa.struct`` metadata can describe which memory subregions in a
2671 memcpy are padding and what the TBAA tags of the struct are.
2673 The current metadata format is very simple. ``!tbaa.struct`` metadata
2674 nodes are a list of operands which are in conceptual groups of three.
2675 For each group of three, the first operand gives the byte offset of a
2676 field in bytes, the second gives its size in bytes, and the third gives
2679 .. code-block:: llvm
2681 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2683 This describes a struct with two fields. The first is at offset 0 bytes
2684 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2685 and has size 4 bytes and has tbaa tag !2.
2687 Note that the fields need not be contiguous. In this example, there is a
2688 4 byte gap between the two fields. This gap represents padding which
2689 does not carry useful data and need not be preserved.
2691 '``fpmath``' Metadata
2692 ^^^^^^^^^^^^^^^^^^^^^
2694 ``fpmath`` metadata may be attached to any instruction of floating point
2695 type. It can be used to express the maximum acceptable error in the
2696 result of that instruction, in ULPs, thus potentially allowing the
2697 compiler to use a more efficient but less accurate method of computing
2698 it. ULP is defined as follows:
2700 If ``x`` is a real number that lies between two finite consecutive
2701 floating-point numbers ``a`` and ``b``, without being equal to one
2702 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2703 distance between the two non-equal finite floating-point numbers
2704 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2706 The metadata node shall consist of a single positive floating point
2707 number representing the maximum relative error, for example:
2709 .. code-block:: llvm
2711 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2713 '``range``' Metadata
2714 ^^^^^^^^^^^^^^^^^^^^
2716 ``range`` metadata may be attached only to loads of integer types. It
2717 expresses the possible ranges the loaded value is in. The ranges are
2718 represented with a flattened list of integers. The loaded value is known
2719 to be in the union of the ranges defined by each consecutive pair. Each
2720 pair has the following properties:
2722 - The type must match the type loaded by the instruction.
2723 - The pair ``a,b`` represents the range ``[a,b)``.
2724 - Both ``a`` and ``b`` are constants.
2725 - The range is allowed to wrap.
2726 - The range should not represent the full or empty set. That is,
2729 In addition, the pairs must be in signed order of the lower bound and
2730 they must be non-contiguous.
2734 .. code-block:: llvm
2736 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2737 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2738 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2739 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2741 !0 = metadata !{ i8 0, i8 2 }
2742 !1 = metadata !{ i8 255, i8 2 }
2743 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2744 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2749 It is sometimes useful to attach information to loop constructs. Currently,
2750 loop metadata is implemented as metadata attached to the branch instruction
2751 in the loop latch block. This type of metadata refer to a metadata node that is
2752 guaranteed to be separate for each loop. The loop identifier metadata is
2753 specified with the name ``llvm.loop``.
2755 The loop identifier metadata is implemented using a metadata that refers to
2756 itself to avoid merging it with any other identifier metadata, e.g.,
2757 during module linkage or function inlining. That is, each loop should refer
2758 to their own identification metadata even if they reside in separate functions.
2759 The following example contains loop identifier metadata for two separate loop
2762 .. code-block:: llvm
2764 !0 = metadata !{ metadata !0 }
2765 !1 = metadata !{ metadata !1 }
2767 The loop identifier metadata can be used to specify additional per-loop
2768 metadata. Any operands after the first operand can be treated as user-defined
2769 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2770 by the loop vectorizer to indicate how many times to unroll the loop:
2772 .. code-block:: llvm
2774 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2776 !0 = metadata !{ metadata !0, metadata !1 }
2777 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2782 Metadata types used to annotate memory accesses with information helpful
2783 for optimizations are prefixed with ``llvm.mem``.
2785 '``llvm.mem.parallel_loop_access``' Metadata
2786 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2788 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
2789 or metadata containing a list of loop identifiers for nested loops.
2790 The metadata is attached to memory accessing instructions and denotes that
2791 no loop carried memory dependence exist between it and other instructions denoted
2792 with the same loop identifier.
2794 Precisely, given two instructions ``m1`` and ``m2`` that both have the
2795 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
2796 set of loops associated with that metadata, respectively, then there is no loop
2797 carried dependence between ``m1`` and ``m2`` for loops ``L1`` or
2800 As a special case, if all memory accessing instructions in a loop have
2801 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
2802 loop has no loop carried memory dependences and is considered to be a parallel
2805 Note that if not all memory access instructions have such metadata referring to
2806 the loop, then the loop is considered not being trivially parallel. Additional
2807 memory dependence analysis is required to make that determination. As a fail
2808 safe mechanism, this causes loops that were originally parallel to be considered
2809 sequential (if optimization passes that are unaware of the parallel semantics
2810 insert new memory instructions into the loop body).
2812 Example of a loop that is considered parallel due to its correct use of
2813 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2814 metadata types that refer to the same loop identifier metadata.
2816 .. code-block:: llvm
2820 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
2822 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2824 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2828 !0 = metadata !{ metadata !0 }
2830 It is also possible to have nested parallel loops. In that case the
2831 memory accesses refer to a list of loop identifier metadata nodes instead of
2832 the loop identifier metadata node directly:
2834 .. code-block:: llvm
2838 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
2840 br label %inner.for.body
2844 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2846 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
2848 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2852 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
2854 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2856 outer.for.end: ; preds = %for.body
2858 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2859 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2860 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2862 '``llvm.vectorizer``'
2863 ^^^^^^^^^^^^^^^^^^^^^
2865 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2866 vectorization parameters such as vectorization factor and unroll factor.
2868 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2869 loop identification metadata.
2871 '``llvm.vectorizer.unroll``' Metadata
2872 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2874 This metadata instructs the loop vectorizer to unroll the specified
2875 loop exactly ``N`` times.
2877 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2878 operand is an integer specifying the unroll factor. For example:
2880 .. code-block:: llvm
2882 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2884 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2887 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2888 determined automatically.
2890 '``llvm.vectorizer.width``' Metadata
2891 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2893 This metadata sets the target width of the vectorizer to ``N``. Without
2894 this metadata, the vectorizer will choose a width automatically.
2895 Regardless of this metadata, the vectorizer will only vectorize loops if
2896 it believes it is valid to do so.
2898 The first operand is the string ``llvm.vectorizer.width`` and the second
2899 operand is an integer specifying the width. For example:
2901 .. code-block:: llvm
2903 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2905 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2908 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2911 Module Flags Metadata
2912 =====================
2914 Information about the module as a whole is difficult to convey to LLVM's
2915 subsystems. The LLVM IR isn't sufficient to transmit this information.
2916 The ``llvm.module.flags`` named metadata exists in order to facilitate
2917 this. These flags are in the form of key / value pairs --- much like a
2918 dictionary --- making it easy for any subsystem who cares about a flag to
2921 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2922 Each triplet has the following form:
2924 - The first element is a *behavior* flag, which specifies the behavior
2925 when two (or more) modules are merged together, and it encounters two
2926 (or more) metadata with the same ID. The supported behaviors are
2928 - The second element is a metadata string that is a unique ID for the
2929 metadata. Each module may only have one flag entry for each unique ID (not
2930 including entries with the **Require** behavior).
2931 - The third element is the value of the flag.
2933 When two (or more) modules are merged together, the resulting
2934 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2935 each unique metadata ID string, there will be exactly one entry in the merged
2936 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2937 be determined by the merge behavior flag, as described below. The only exception
2938 is that entries with the *Require* behavior are always preserved.
2940 The following behaviors are supported:
2951 Emits an error if two values disagree, otherwise the resulting value
2952 is that of the operands.
2956 Emits a warning if two values disagree. The result value will be the
2957 operand for the flag from the first module being linked.
2961 Adds a requirement that another module flag be present and have a
2962 specified value after linking is performed. The value must be a
2963 metadata pair, where the first element of the pair is the ID of the
2964 module flag to be restricted, and the second element of the pair is
2965 the value the module flag should be restricted to. This behavior can
2966 be used to restrict the allowable results (via triggering of an
2967 error) of linking IDs with the **Override** behavior.
2971 Uses the specified value, regardless of the behavior or value of the
2972 other module. If both modules specify **Override**, but the values
2973 differ, an error will be emitted.
2977 Appends the two values, which are required to be metadata nodes.
2981 Appends the two values, which are required to be metadata
2982 nodes. However, duplicate entries in the second list are dropped
2983 during the append operation.
2985 It is an error for a particular unique flag ID to have multiple behaviors,
2986 except in the case of **Require** (which adds restrictions on another metadata
2987 value) or **Override**.
2989 An example of module flags:
2991 .. code-block:: llvm
2993 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2994 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2995 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2996 !3 = metadata !{ i32 3, metadata !"qux",
2998 metadata !"foo", i32 1
3001 !llvm.module.flags = !{ !0, !1, !2, !3 }
3003 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3004 if two or more ``!"foo"`` flags are seen is to emit an error if their
3005 values are not equal.
3007 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3008 behavior if two or more ``!"bar"`` flags are seen is to use the value
3011 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3012 behavior if two or more ``!"qux"`` flags are seen is to emit a
3013 warning if their values are not equal.
3015 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3019 metadata !{ metadata !"foo", i32 1 }
3021 The behavior is to emit an error if the ``llvm.module.flags`` does not
3022 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3025 Objective-C Garbage Collection Module Flags Metadata
3026 ----------------------------------------------------
3028 On the Mach-O platform, Objective-C stores metadata about garbage
3029 collection in a special section called "image info". The metadata
3030 consists of a version number and a bitmask specifying what types of
3031 garbage collection are supported (if any) by the file. If two or more
3032 modules are linked together their garbage collection metadata needs to
3033 be merged rather than appended together.
3035 The Objective-C garbage collection module flags metadata consists of the
3036 following key-value pairs:
3045 * - ``Objective-C Version``
3046 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3048 * - ``Objective-C Image Info Version``
3049 - **[Required]** --- The version of the image info section. Currently
3052 * - ``Objective-C Image Info Section``
3053 - **[Required]** --- The section to place the metadata. Valid values are
3054 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3055 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3056 Objective-C ABI version 2.
3058 * - ``Objective-C Garbage Collection``
3059 - **[Required]** --- Specifies whether garbage collection is supported or
3060 not. Valid values are 0, for no garbage collection, and 2, for garbage
3061 collection supported.
3063 * - ``Objective-C GC Only``
3064 - **[Optional]** --- Specifies that only garbage collection is supported.
3065 If present, its value must be 6. This flag requires that the
3066 ``Objective-C Garbage Collection`` flag have the value 2.
3068 Some important flag interactions:
3070 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3071 merged with a module with ``Objective-C Garbage Collection`` set to
3072 2, then the resulting module has the
3073 ``Objective-C Garbage Collection`` flag set to 0.
3074 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3075 merged with a module with ``Objective-C GC Only`` set to 6.
3077 Automatic Linker Flags Module Flags Metadata
3078 --------------------------------------------
3080 Some targets support embedding flags to the linker inside individual object
3081 files. Typically this is used in conjunction with language extensions which
3082 allow source files to explicitly declare the libraries they depend on, and have
3083 these automatically be transmitted to the linker via object files.
3085 These flags are encoded in the IR using metadata in the module flags section,
3086 using the ``Linker Options`` key. The merge behavior for this flag is required
3087 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3088 node which should be a list of other metadata nodes, each of which should be a
3089 list of metadata strings defining linker options.
3091 For example, the following metadata section specifies two separate sets of
3092 linker options, presumably to link against ``libz`` and the ``Cocoa``
3095 !0 = metadata !{ i32 6, metadata !"Linker Options",
3097 metadata !{ metadata !"-lz" },
3098 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3099 !llvm.module.flags = !{ !0 }
3101 The metadata encoding as lists of lists of options, as opposed to a collapsed
3102 list of options, is chosen so that the IR encoding can use multiple option
3103 strings to specify e.g., a single library, while still having that specifier be
3104 preserved as an atomic element that can be recognized by a target specific
3105 assembly writer or object file emitter.
3107 Each individual option is required to be either a valid option for the target's
3108 linker, or an option that is reserved by the target specific assembly writer or
3109 object file emitter. No other aspect of these options is defined by the IR.
3111 .. _intrinsicglobalvariables:
3113 Intrinsic Global Variables
3114 ==========================
3116 LLVM has a number of "magic" global variables that contain data that
3117 affect code generation or other IR semantics. These are documented here.
3118 All globals of this sort should have a section specified as
3119 "``llvm.metadata``". This section and all globals that start with
3120 "``llvm.``" are reserved for use by LLVM.
3124 The '``llvm.used``' Global Variable
3125 -----------------------------------
3127 The ``@llvm.used`` global is an array which has
3128 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3129 pointers to named global variables, functions and aliases which may optionally
3130 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3133 .. code-block:: llvm
3138 @llvm.used = appending global [2 x i8*] [
3140 i8* bitcast (i32* @Y to i8*)
3141 ], section "llvm.metadata"
3143 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3144 and linker are required to treat the symbol as if there is a reference to the
3145 symbol that it cannot see (which is why they have to be named). For example, if
3146 a variable has internal linkage and no references other than that from the
3147 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3148 references from inline asms and other things the compiler cannot "see", and
3149 corresponds to "``attribute((used))``" in GNU C.
3151 On some targets, the code generator must emit a directive to the
3152 assembler or object file to prevent the assembler and linker from
3153 molesting the symbol.
3155 .. _gv_llvmcompilerused:
3157 The '``llvm.compiler.used``' Global Variable
3158 --------------------------------------------
3160 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3161 directive, except that it only prevents the compiler from touching the
3162 symbol. On targets that support it, this allows an intelligent linker to
3163 optimize references to the symbol without being impeded as it would be
3166 This is a rare construct that should only be used in rare circumstances,
3167 and should not be exposed to source languages.
3169 .. _gv_llvmglobalctors:
3171 The '``llvm.global_ctors``' Global Variable
3172 -------------------------------------------
3174 .. code-block:: llvm
3176 %0 = type { i32, void ()*, i8* }
3177 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
3179 The ``@llvm.global_ctors`` array contains a list of constructor
3180 functions, priorities, and an optional associated global or function.
3181 The functions referenced by this array will be called in ascending order
3182 of priority (i.e. lowest first) when the module is loaded. The order of
3183 functions with the same priority is not defined.
3185 If the third field is present, non-null, and points to a global variable
3186 or function, the initializer function will only run if the associated
3187 data from the current module is not discarded.
3189 .. _llvmglobaldtors:
3191 The '``llvm.global_dtors``' Global Variable
3192 -------------------------------------------
3194 .. code-block:: llvm
3196 %0 = type { i32, void ()*, i8* }
3197 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
3199 The ``@llvm.global_dtors`` array contains a list of destructor
3200 functions, priorities, and an optional associated global or function.
3201 The functions referenced by this array will be called in descending
3202 order of priority (i.e. highest first) when the module is loaded. The
3203 order of functions with the same priority is not defined.
3205 If the third field is present, non-null, and points to a global variable
3206 or function, the destructor function will only run if the associated
3207 data from the current module is not discarded.
3209 Instruction Reference
3210 =====================
3212 The LLVM instruction set consists of several different classifications
3213 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3214 instructions <binaryops>`, :ref:`bitwise binary
3215 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3216 :ref:`other instructions <otherops>`.
3220 Terminator Instructions
3221 -----------------------
3223 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3224 program ends with a "Terminator" instruction, which indicates which
3225 block should be executed after the current block is finished. These
3226 terminator instructions typically yield a '``void``' value: they produce
3227 control flow, not values (the one exception being the
3228 ':ref:`invoke <i_invoke>`' instruction).
3230 The terminator instructions are: ':ref:`ret <i_ret>`',
3231 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3232 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3233 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3237 '``ret``' Instruction
3238 ^^^^^^^^^^^^^^^^^^^^^
3245 ret <type> <value> ; Return a value from a non-void function
3246 ret void ; Return from void function
3251 The '``ret``' instruction is used to return control flow (and optionally
3252 a value) from a function back to the caller.
3254 There are two forms of the '``ret``' instruction: one that returns a
3255 value and then causes control flow, and one that just causes control
3261 The '``ret``' instruction optionally accepts a single argument, the
3262 return value. The type of the return value must be a ':ref:`first
3263 class <t_firstclass>`' type.
3265 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3266 return type and contains a '``ret``' instruction with no return value or
3267 a return value with a type that does not match its type, or if it has a
3268 void return type and contains a '``ret``' instruction with a return
3274 When the '``ret``' instruction is executed, control flow returns back to
3275 the calling function's context. If the caller is a
3276 ":ref:`call <i_call>`" instruction, execution continues at the
3277 instruction after the call. If the caller was an
3278 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3279 beginning of the "normal" destination block. If the instruction returns
3280 a value, that value shall set the call or invoke instruction's return
3286 .. code-block:: llvm
3288 ret i32 5 ; Return an integer value of 5
3289 ret void ; Return from a void function
3290 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3294 '``br``' Instruction
3295 ^^^^^^^^^^^^^^^^^^^^
3302 br i1 <cond>, label <iftrue>, label <iffalse>
3303 br label <dest> ; Unconditional branch
3308 The '``br``' instruction is used to cause control flow to transfer to a
3309 different basic block in the current function. There are two forms of
3310 this instruction, corresponding to a conditional branch and an
3311 unconditional branch.
3316 The conditional branch form of the '``br``' instruction takes a single
3317 '``i1``' value and two '``label``' values. The unconditional form of the
3318 '``br``' instruction takes a single '``label``' value as a target.
3323 Upon execution of a conditional '``br``' instruction, the '``i1``'
3324 argument is evaluated. If the value is ``true``, control flows to the
3325 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3326 to the '``iffalse``' ``label`` argument.
3331 .. code-block:: llvm
3334 %cond = icmp eq i32 %a, %b
3335 br i1 %cond, label %IfEqual, label %IfUnequal
3343 '``switch``' Instruction
3344 ^^^^^^^^^^^^^^^^^^^^^^^^
3351 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3356 The '``switch``' instruction is used to transfer control flow to one of
3357 several different places. It is a generalization of the '``br``'
3358 instruction, allowing a branch to occur to one of many possible
3364 The '``switch``' instruction uses three parameters: an integer
3365 comparison value '``value``', a default '``label``' destination, and an
3366 array of pairs of comparison value constants and '``label``'s. The table
3367 is not allowed to contain duplicate constant entries.
3372 The ``switch`` instruction specifies a table of values and destinations.
3373 When the '``switch``' instruction is executed, this table is searched
3374 for the given value. If the value is found, control flow is transferred
3375 to the corresponding destination; otherwise, control flow is transferred
3376 to the default destination.
3381 Depending on properties of the target machine and the particular
3382 ``switch`` instruction, this instruction may be code generated in
3383 different ways. For example, it could be generated as a series of
3384 chained conditional branches or with a lookup table.
3389 .. code-block:: llvm
3391 ; Emulate a conditional br instruction
3392 %Val = zext i1 %value to i32
3393 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3395 ; Emulate an unconditional br instruction
3396 switch i32 0, label %dest [ ]
3398 ; Implement a jump table:
3399 switch i32 %val, label %otherwise [ i32 0, label %onzero
3401 i32 2, label %ontwo ]
3405 '``indirectbr``' Instruction
3406 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3413 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3418 The '``indirectbr``' instruction implements an indirect branch to a
3419 label within the current function, whose address is specified by
3420 "``address``". Address must be derived from a
3421 :ref:`blockaddress <blockaddress>` constant.
3426 The '``address``' argument is the address of the label to jump to. The
3427 rest of the arguments indicate the full set of possible destinations
3428 that the address may point to. Blocks are allowed to occur multiple
3429 times in the destination list, though this isn't particularly useful.
3431 This destination list is required so that dataflow analysis has an
3432 accurate understanding of the CFG.
3437 Control transfers to the block specified in the address argument. All
3438 possible destination blocks must be listed in the label list, otherwise
3439 this instruction has undefined behavior. This implies that jumps to
3440 labels defined in other functions have undefined behavior as well.
3445 This is typically implemented with a jump through a register.
3450 .. code-block:: llvm
3452 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3456 '``invoke``' Instruction
3457 ^^^^^^^^^^^^^^^^^^^^^^^^
3464 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3465 to label <normal label> unwind label <exception label>
3470 The '``invoke``' instruction causes control to transfer to a specified
3471 function, with the possibility of control flow transfer to either the
3472 '``normal``' label or the '``exception``' label. If the callee function
3473 returns with the "``ret``" instruction, control flow will return to the
3474 "normal" label. If the callee (or any indirect callees) returns via the
3475 ":ref:`resume <i_resume>`" instruction or other exception handling
3476 mechanism, control is interrupted and continued at the dynamically
3477 nearest "exception" label.
3479 The '``exception``' label is a `landing
3480 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3481 '``exception``' label is required to have the
3482 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3483 information about the behavior of the program after unwinding happens,
3484 as its first non-PHI instruction. The restrictions on the
3485 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3486 instruction, so that the important information contained within the
3487 "``landingpad``" instruction can't be lost through normal code motion.
3492 This instruction requires several arguments:
3494 #. The optional "cconv" marker indicates which :ref:`calling
3495 convention <callingconv>` the call should use. If none is
3496 specified, the call defaults to using C calling conventions.
3497 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3498 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3500 #. '``ptr to function ty``': shall be the signature of the pointer to
3501 function value being invoked. In most cases, this is a direct
3502 function invocation, but indirect ``invoke``'s are just as possible,
3503 branching off an arbitrary pointer to function value.
3504 #. '``function ptr val``': An LLVM value containing a pointer to a
3505 function to be invoked.
3506 #. '``function args``': argument list whose types match the function
3507 signature argument types and parameter attributes. All arguments must
3508 be of :ref:`first class <t_firstclass>` type. If the function signature
3509 indicates the function accepts a variable number of arguments, the
3510 extra arguments can be specified.
3511 #. '``normal label``': the label reached when the called function
3512 executes a '``ret``' instruction.
3513 #. '``exception label``': the label reached when a callee returns via
3514 the :ref:`resume <i_resume>` instruction or other exception handling
3516 #. The optional :ref:`function attributes <fnattrs>` list. Only
3517 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3518 attributes are valid here.
3523 This instruction is designed to operate as a standard '``call``'
3524 instruction in most regards. The primary difference is that it
3525 establishes an association with a label, which is used by the runtime
3526 library to unwind the stack.
3528 This instruction is used in languages with destructors to ensure that
3529 proper cleanup is performed in the case of either a ``longjmp`` or a
3530 thrown exception. Additionally, this is important for implementation of
3531 '``catch``' clauses in high-level languages that support them.
3533 For the purposes of the SSA form, the definition of the value returned
3534 by the '``invoke``' instruction is deemed to occur on the edge from the
3535 current block to the "normal" label. If the callee unwinds then no
3536 return value is available.
3541 .. code-block:: llvm
3543 %retval = invoke i32 @Test(i32 15) to label %Continue
3544 unwind label %TestCleanup ; {i32}:retval set
3545 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3546 unwind label %TestCleanup ; {i32}:retval set
3550 '``resume``' Instruction
3551 ^^^^^^^^^^^^^^^^^^^^^^^^
3558 resume <type> <value>
3563 The '``resume``' instruction is a terminator instruction that has no
3569 The '``resume``' instruction requires one argument, which must have the
3570 same type as the result of any '``landingpad``' instruction in the same
3576 The '``resume``' instruction resumes propagation of an existing
3577 (in-flight) exception whose unwinding was interrupted with a
3578 :ref:`landingpad <i_landingpad>` instruction.
3583 .. code-block:: llvm
3585 resume { i8*, i32 } %exn
3589 '``unreachable``' Instruction
3590 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3602 The '``unreachable``' instruction has no defined semantics. This
3603 instruction is used to inform the optimizer that a particular portion of
3604 the code is not reachable. This can be used to indicate that the code
3605 after a no-return function cannot be reached, and other facts.
3610 The '``unreachable``' instruction has no defined semantics.
3617 Binary operators are used to do most of the computation in a program.
3618 They require two operands of the same type, execute an operation on
3619 them, and produce a single value. The operands might represent multiple
3620 data, as is the case with the :ref:`vector <t_vector>` data type. The
3621 result value has the same type as its operands.
3623 There are several different binary operators:
3627 '``add``' Instruction
3628 ^^^^^^^^^^^^^^^^^^^^^
3635 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3636 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3637 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3638 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3643 The '``add``' instruction returns the sum of its two operands.
3648 The two arguments to the '``add``' instruction must be
3649 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3650 arguments must have identical types.
3655 The value produced is the integer sum of the two operands.
3657 If the sum has unsigned overflow, the result returned is the
3658 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3661 Because LLVM integers use a two's complement representation, this
3662 instruction is appropriate for both signed and unsigned integers.
3664 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3665 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3666 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3667 unsigned and/or signed overflow, respectively, occurs.
3672 .. code-block:: llvm
3674 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3678 '``fadd``' Instruction
3679 ^^^^^^^^^^^^^^^^^^^^^^
3686 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3691 The '``fadd``' instruction returns the sum of its two operands.
3696 The two arguments to the '``fadd``' instruction must be :ref:`floating
3697 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3698 Both arguments must have identical types.
3703 The value produced is the floating point sum of the two operands. This
3704 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3705 which are optimization hints to enable otherwise unsafe floating point
3711 .. code-block:: llvm
3713 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3715 '``sub``' Instruction
3716 ^^^^^^^^^^^^^^^^^^^^^
3723 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3724 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3725 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3726 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3731 The '``sub``' instruction returns the difference of its two operands.
3733 Note that the '``sub``' instruction is used to represent the '``neg``'
3734 instruction present in most other intermediate representations.
3739 The two arguments to the '``sub``' instruction must be
3740 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3741 arguments must have identical types.
3746 The value produced is the integer difference of the two operands.
3748 If the difference has unsigned overflow, the result returned is the
3749 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3752 Because LLVM integers use a two's complement representation, this
3753 instruction is appropriate for both signed and unsigned integers.
3755 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3756 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3757 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3758 unsigned and/or signed overflow, respectively, occurs.
3763 .. code-block:: llvm
3765 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3766 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3770 '``fsub``' Instruction
3771 ^^^^^^^^^^^^^^^^^^^^^^
3778 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3783 The '``fsub``' instruction returns the difference of its two operands.
3785 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3786 instruction present in most other intermediate representations.
3791 The two arguments to the '``fsub``' instruction must be :ref:`floating
3792 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3793 Both arguments must have identical types.
3798 The value produced is the floating point difference of the two operands.
3799 This instruction can also take any number of :ref:`fast-math
3800 flags <fastmath>`, which are optimization hints to enable otherwise
3801 unsafe floating point optimizations:
3806 .. code-block:: llvm
3808 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3809 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3811 '``mul``' Instruction
3812 ^^^^^^^^^^^^^^^^^^^^^
3819 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3820 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3821 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3822 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3827 The '``mul``' instruction returns the product of its two operands.
3832 The two arguments to the '``mul``' instruction must be
3833 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3834 arguments must have identical types.
3839 The value produced is the integer product of the two operands.
3841 If the result of the multiplication has unsigned overflow, the result
3842 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3843 bit width of the result.
3845 Because LLVM integers use a two's complement representation, and the
3846 result is the same width as the operands, this instruction returns the
3847 correct result for both signed and unsigned integers. If a full product
3848 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3849 sign-extended or zero-extended as appropriate to the width of the full
3852 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3853 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3854 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3855 unsigned and/or signed overflow, respectively, occurs.
3860 .. code-block:: llvm
3862 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3866 '``fmul``' Instruction
3867 ^^^^^^^^^^^^^^^^^^^^^^
3874 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3879 The '``fmul``' instruction returns the product of its two operands.
3884 The two arguments to the '``fmul``' instruction must be :ref:`floating
3885 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3886 Both arguments must have identical types.
3891 The value produced is the floating point product of the two operands.
3892 This instruction can also take any number of :ref:`fast-math
3893 flags <fastmath>`, which are optimization hints to enable otherwise
3894 unsafe floating point optimizations:
3899 .. code-block:: llvm
3901 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3903 '``udiv``' Instruction
3904 ^^^^^^^^^^^^^^^^^^^^^^
3911 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3912 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3917 The '``udiv``' instruction returns the quotient of its two operands.
3922 The two arguments to the '``udiv``' instruction must be
3923 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3924 arguments must have identical types.
3929 The value produced is the unsigned integer quotient of the two operands.
3931 Note that unsigned integer division and signed integer division are
3932 distinct operations; for signed integer division, use '``sdiv``'.
3934 Division by zero leads to undefined behavior.
3936 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3937 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3938 such, "((a udiv exact b) mul b) == a").
3943 .. code-block:: llvm
3945 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3947 '``sdiv``' Instruction
3948 ^^^^^^^^^^^^^^^^^^^^^^
3955 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3956 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3961 The '``sdiv``' instruction returns the quotient of its two operands.
3966 The two arguments to the '``sdiv``' instruction must be
3967 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3968 arguments must have identical types.
3973 The value produced is the signed integer quotient of the two operands
3974 rounded towards zero.
3976 Note that signed integer division and unsigned integer division are
3977 distinct operations; for unsigned integer division, use '``udiv``'.
3979 Division by zero leads to undefined behavior. Overflow also leads to
3980 undefined behavior; this is a rare case, but can occur, for example, by
3981 doing a 32-bit division of -2147483648 by -1.
3983 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3984 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3989 .. code-block:: llvm
3991 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3995 '``fdiv``' Instruction
3996 ^^^^^^^^^^^^^^^^^^^^^^
4003 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4008 The '``fdiv``' instruction returns the quotient of its two operands.
4013 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4014 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4015 Both arguments must have identical types.
4020 The value produced is the floating point quotient of the two operands.
4021 This instruction can also take any number of :ref:`fast-math
4022 flags <fastmath>`, which are optimization hints to enable otherwise
4023 unsafe floating point optimizations:
4028 .. code-block:: llvm
4030 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
4032 '``urem``' Instruction
4033 ^^^^^^^^^^^^^^^^^^^^^^
4040 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
4045 The '``urem``' instruction returns the remainder from the unsigned
4046 division of its two arguments.
4051 The two arguments to the '``urem``' instruction must be
4052 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4053 arguments must have identical types.
4058 This instruction returns the unsigned integer *remainder* of a division.
4059 This instruction always performs an unsigned division to get the
4062 Note that unsigned integer remainder and signed integer remainder are
4063 distinct operations; for signed integer remainder, use '``srem``'.
4065 Taking the remainder of a division by zero leads to undefined behavior.
4070 .. code-block:: llvm
4072 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
4074 '``srem``' Instruction
4075 ^^^^^^^^^^^^^^^^^^^^^^
4082 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
4087 The '``srem``' instruction returns the remainder from the signed
4088 division of its two operands. This instruction can also take
4089 :ref:`vector <t_vector>` versions of the values in which case the elements
4095 The two arguments to the '``srem``' instruction must be
4096 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4097 arguments must have identical types.
4102 This instruction returns the *remainder* of a division (where the result
4103 is either zero or has the same sign as the dividend, ``op1``), not the
4104 *modulo* operator (where the result is either zero or has the same sign
4105 as the divisor, ``op2``) of a value. For more information about the
4106 difference, see `The Math
4107 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4108 table of how this is implemented in various languages, please see
4110 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4112 Note that signed integer remainder and unsigned integer remainder are
4113 distinct operations; for unsigned integer remainder, use '``urem``'.
4115 Taking the remainder of a division by zero leads to undefined behavior.
4116 Overflow also leads to undefined behavior; this is a rare case, but can
4117 occur, for example, by taking the remainder of a 32-bit division of
4118 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4119 rule lets srem be implemented using instructions that return both the
4120 result of the division and the remainder.)
4125 .. code-block:: llvm
4127 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4131 '``frem``' Instruction
4132 ^^^^^^^^^^^^^^^^^^^^^^
4139 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4144 The '``frem``' instruction returns the remainder from the division of
4150 The two arguments to the '``frem``' instruction must be :ref:`floating
4151 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4152 Both arguments must have identical types.
4157 This instruction returns the *remainder* of a division. The remainder
4158 has the same sign as the dividend. This instruction can also take any
4159 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4160 to enable otherwise unsafe floating point optimizations:
4165 .. code-block:: llvm
4167 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4171 Bitwise Binary Operations
4172 -------------------------
4174 Bitwise binary operators are used to do various forms of bit-twiddling
4175 in a program. They are generally very efficient instructions and can
4176 commonly be strength reduced from other instructions. They require two
4177 operands of the same type, execute an operation on them, and produce a
4178 single value. The resulting value is the same type as its operands.
4180 '``shl``' Instruction
4181 ^^^^^^^^^^^^^^^^^^^^^
4188 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4189 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4190 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4191 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4196 The '``shl``' instruction returns the first operand shifted to the left
4197 a specified number of bits.
4202 Both arguments to the '``shl``' instruction must be the same
4203 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4204 '``op2``' is treated as an unsigned value.
4209 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4210 where ``n`` is the width of the result. If ``op2`` is (statically or
4211 dynamically) negative or equal to or larger than the number of bits in
4212 ``op1``, the result is undefined. If the arguments are vectors, each
4213 vector element of ``op1`` is shifted by the corresponding shift amount
4216 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4217 value <poisonvalues>` if it shifts out any non-zero bits. If the
4218 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4219 value <poisonvalues>` if it shifts out any bits that disagree with the
4220 resultant sign bit. As such, NUW/NSW have the same semantics as they
4221 would if the shift were expressed as a mul instruction with the same
4222 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4227 .. code-block:: llvm
4229 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4230 <result> = shl i32 4, 2 ; yields {i32}: 16
4231 <result> = shl i32 1, 10 ; yields {i32}: 1024
4232 <result> = shl i32 1, 32 ; undefined
4233 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4235 '``lshr``' Instruction
4236 ^^^^^^^^^^^^^^^^^^^^^^
4243 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4244 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4249 The '``lshr``' instruction (logical shift right) returns the first
4250 operand shifted to the right a specified number of bits with zero fill.
4255 Both arguments to the '``lshr``' instruction must be the same
4256 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4257 '``op2``' is treated as an unsigned value.
4262 This instruction always performs a logical shift right operation. The
4263 most significant bits of the result will be filled with zero bits after
4264 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4265 than the number of bits in ``op1``, the result is undefined. If the
4266 arguments are vectors, each vector element of ``op1`` is shifted by the
4267 corresponding shift amount in ``op2``.
4269 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4270 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4276 .. code-block:: llvm
4278 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4279 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4280 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4281 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4282 <result> = lshr i32 1, 32 ; undefined
4283 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4285 '``ashr``' Instruction
4286 ^^^^^^^^^^^^^^^^^^^^^^
4293 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4294 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4299 The '``ashr``' instruction (arithmetic shift right) returns the first
4300 operand shifted to the right a specified number of bits with sign
4306 Both arguments to the '``ashr``' instruction must be the same
4307 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4308 '``op2``' is treated as an unsigned value.
4313 This instruction always performs an arithmetic shift right operation,
4314 The most significant bits of the result will be filled with the sign bit
4315 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4316 than the number of bits in ``op1``, the result is undefined. If the
4317 arguments are vectors, each vector element of ``op1`` is shifted by the
4318 corresponding shift amount in ``op2``.
4320 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4321 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4327 .. code-block:: llvm
4329 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4330 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4331 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4332 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4333 <result> = ashr i32 1, 32 ; undefined
4334 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4336 '``and``' Instruction
4337 ^^^^^^^^^^^^^^^^^^^^^
4344 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4349 The '``and``' instruction returns the bitwise logical and of its two
4355 The two arguments to the '``and``' instruction must be
4356 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4357 arguments must have identical types.
4362 The truth table used for the '``and``' instruction is:
4379 .. code-block:: llvm
4381 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4382 <result> = and i32 15, 40 ; yields {i32}:result = 8
4383 <result> = and i32 4, 8 ; yields {i32}:result = 0
4385 '``or``' Instruction
4386 ^^^^^^^^^^^^^^^^^^^^
4393 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4398 The '``or``' instruction returns the bitwise logical inclusive or of its
4404 The two arguments to the '``or``' instruction must be
4405 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4406 arguments must have identical types.
4411 The truth table used for the '``or``' instruction is:
4430 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4431 <result> = or i32 15, 40 ; yields {i32}:result = 47
4432 <result> = or i32 4, 8 ; yields {i32}:result = 12
4434 '``xor``' Instruction
4435 ^^^^^^^^^^^^^^^^^^^^^
4442 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4447 The '``xor``' instruction returns the bitwise logical exclusive or of
4448 its two operands. The ``xor`` is used to implement the "one's
4449 complement" operation, which is the "~" operator in C.
4454 The two arguments to the '``xor``' instruction must be
4455 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4456 arguments must have identical types.
4461 The truth table used for the '``xor``' instruction is:
4478 .. code-block:: llvm
4480 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4481 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4482 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4483 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4488 LLVM supports several instructions to represent vector operations in a
4489 target-independent manner. These instructions cover the element-access
4490 and vector-specific operations needed to process vectors effectively.
4491 While LLVM does directly support these vector operations, many
4492 sophisticated algorithms will want to use target-specific intrinsics to
4493 take full advantage of a specific target.
4495 .. _i_extractelement:
4497 '``extractelement``' Instruction
4498 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4505 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
4510 The '``extractelement``' instruction extracts a single scalar element
4511 from a vector at a specified index.
4516 The first operand of an '``extractelement``' instruction is a value of
4517 :ref:`vector <t_vector>` type. The second operand is an index indicating
4518 the position from which to extract the element. The index may be a
4519 variable of any integer type.
4524 The result is a scalar of the same type as the element type of ``val``.
4525 Its value is the value at position ``idx`` of ``val``. If ``idx``
4526 exceeds the length of ``val``, the results are undefined.
4531 .. code-block:: llvm
4533 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4535 .. _i_insertelement:
4537 '``insertelement``' Instruction
4538 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4545 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
4550 The '``insertelement``' instruction inserts a scalar element into a
4551 vector at a specified index.
4556 The first operand of an '``insertelement``' instruction is a value of
4557 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4558 type must equal the element type of the first operand. The third operand
4559 is an index indicating the position at which to insert the value. The
4560 index may be a variable of any integer type.
4565 The result is a vector of the same type as ``val``. Its element values
4566 are those of ``val`` except at position ``idx``, where it gets the value
4567 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4573 .. code-block:: llvm
4575 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4577 .. _i_shufflevector:
4579 '``shufflevector``' Instruction
4580 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4587 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4592 The '``shufflevector``' instruction constructs a permutation of elements
4593 from two input vectors, returning a vector with the same element type as
4594 the input and length that is the same as the shuffle mask.
4599 The first two operands of a '``shufflevector``' instruction are vectors
4600 with the same type. The third argument is a shuffle mask whose element
4601 type is always 'i32'. The result of the instruction is a vector whose
4602 length is the same as the shuffle mask and whose element type is the
4603 same as the element type of the first two operands.
4605 The shuffle mask operand is required to be a constant vector with either
4606 constant integer or undef values.
4611 The elements of the two input vectors are numbered from left to right
4612 across both of the vectors. The shuffle mask operand specifies, for each
4613 element of the result vector, which element of the two input vectors the
4614 result element gets. The element selector may be undef (meaning "don't
4615 care") and the second operand may be undef if performing a shuffle from
4621 .. code-block:: llvm
4623 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4624 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4625 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4626 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4627 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4628 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4629 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4630 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4632 Aggregate Operations
4633 --------------------
4635 LLVM supports several instructions for working with
4636 :ref:`aggregate <t_aggregate>` values.
4640 '``extractvalue``' Instruction
4641 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4648 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4653 The '``extractvalue``' instruction extracts the value of a member field
4654 from an :ref:`aggregate <t_aggregate>` value.
4659 The first operand of an '``extractvalue``' instruction is a value of
4660 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4661 constant indices to specify which value to extract in a similar manner
4662 as indices in a '``getelementptr``' instruction.
4664 The major differences to ``getelementptr`` indexing are:
4666 - Since the value being indexed is not a pointer, the first index is
4667 omitted and assumed to be zero.
4668 - At least one index must be specified.
4669 - Not only struct indices but also array indices must be in bounds.
4674 The result is the value at the position in the aggregate specified by
4680 .. code-block:: llvm
4682 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4686 '``insertvalue``' Instruction
4687 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4694 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4699 The '``insertvalue``' instruction inserts a value into a member field in
4700 an :ref:`aggregate <t_aggregate>` value.
4705 The first operand of an '``insertvalue``' instruction is a value of
4706 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4707 a first-class value to insert. The following operands are constant
4708 indices indicating the position at which to insert the value in a
4709 similar manner as indices in a '``extractvalue``' instruction. The value
4710 to insert must have the same type as the value identified by the
4716 The result is an aggregate of the same type as ``val``. Its value is
4717 that of ``val`` except that the value at the position specified by the
4718 indices is that of ``elt``.
4723 .. code-block:: llvm
4725 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4726 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4727 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4731 Memory Access and Addressing Operations
4732 ---------------------------------------
4734 A key design point of an SSA-based representation is how it represents
4735 memory. In LLVM, no memory locations are in SSA form, which makes things
4736 very simple. This section describes how to read, write, and allocate
4741 '``alloca``' Instruction
4742 ^^^^^^^^^^^^^^^^^^^^^^^^
4749 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields {type*}:result
4754 The '``alloca``' instruction allocates memory on the stack frame of the
4755 currently executing function, to be automatically released when this
4756 function returns to its caller. The object is always allocated in the
4757 generic address space (address space zero).
4762 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4763 bytes of memory on the runtime stack, returning a pointer of the
4764 appropriate type to the program. If "NumElements" is specified, it is
4765 the number of elements allocated, otherwise "NumElements" is defaulted
4766 to be one. If a constant alignment is specified, the value result of the
4767 allocation is guaranteed to be aligned to at least that boundary. If not
4768 specified, or if zero, the target can choose to align the allocation on
4769 any convenient boundary compatible with the type.
4771 '``type``' may be any sized type.
4776 Memory is allocated; a pointer is returned. The operation is undefined
4777 if there is insufficient stack space for the allocation. '``alloca``'d
4778 memory is automatically released when the function returns. The
4779 '``alloca``' instruction is commonly used to represent automatic
4780 variables that must have an address available. When the function returns
4781 (either with the ``ret`` or ``resume`` instructions), the memory is
4782 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4783 The order in which memory is allocated (ie., which way the stack grows)
4789 .. code-block:: llvm
4791 %ptr = alloca i32 ; yields {i32*}:ptr
4792 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4793 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4794 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4798 '``load``' Instruction
4799 ^^^^^^^^^^^^^^^^^^^^^^
4806 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4807 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4808 !<index> = !{ i32 1 }
4813 The '``load``' instruction is used to read from memory.
4818 The argument to the ``load`` instruction specifies the memory address
4819 from which to load. The pointer must point to a :ref:`first
4820 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4821 then the optimizer is not allowed to modify the number or order of
4822 execution of this ``load`` with other :ref:`volatile
4823 operations <volatile>`.
4825 If the ``load`` is marked as ``atomic``, it takes an extra
4826 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4827 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4828 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4829 when they may see multiple atomic stores. The type of the pointee must
4830 be an integer type whose bit width is a power of two greater than or
4831 equal to eight and less than or equal to a target-specific size limit.
4832 ``align`` must be explicitly specified on atomic loads, and the load has
4833 undefined behavior if the alignment is not set to a value which is at
4834 least the size in bytes of the pointee. ``!nontemporal`` does not have
4835 any defined semantics for atomic loads.
4837 The optional constant ``align`` argument specifies the alignment of the
4838 operation (that is, the alignment of the memory address). A value of 0
4839 or an omitted ``align`` argument means that the operation has the ABI
4840 alignment for the target. It is the responsibility of the code emitter
4841 to ensure that the alignment information is correct. Overestimating the
4842 alignment results in undefined behavior. Underestimating the alignment
4843 may produce less efficient code. An alignment of 1 is always safe.
4845 The optional ``!nontemporal`` metadata must reference a single
4846 metadata name ``<index>`` corresponding to a metadata node with one
4847 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4848 metadata on the instruction tells the optimizer and code generator
4849 that this load is not expected to be reused in the cache. The code
4850 generator may select special instructions to save cache bandwidth, such
4851 as the ``MOVNT`` instruction on x86.
4853 The optional ``!invariant.load`` metadata must reference a single
4854 metadata name ``<index>`` corresponding to a metadata node with no
4855 entries. The existence of the ``!invariant.load`` metadata on the
4856 instruction tells the optimizer and code generator that this load
4857 address points to memory which does not change value during program
4858 execution. The optimizer may then move this load around, for example, by
4859 hoisting it out of loops using loop invariant code motion.
4864 The location of memory pointed to is loaded. If the value being loaded
4865 is of scalar type then the number of bytes read does not exceed the
4866 minimum number of bytes needed to hold all bits of the type. For
4867 example, loading an ``i24`` reads at most three bytes. When loading a
4868 value of a type like ``i20`` with a size that is not an integral number
4869 of bytes, the result is undefined if the value was not originally
4870 written using a store of the same type.
4875 .. code-block:: llvm
4877 %ptr = alloca i32 ; yields {i32*}:ptr
4878 store i32 3, i32* %ptr ; yields {void}
4879 %val = load i32* %ptr ; yields {i32}:val = i32 3
4883 '``store``' Instruction
4884 ^^^^^^^^^^^^^^^^^^^^^^^
4891 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4892 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4897 The '``store``' instruction is used to write to memory.
4902 There are two arguments to the ``store`` instruction: a value to store
4903 and an address at which to store it. The type of the ``<pointer>``
4904 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4905 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4906 then the optimizer is not allowed to modify the number or order of
4907 execution of this ``store`` with other :ref:`volatile
4908 operations <volatile>`.
4910 If the ``store`` is marked as ``atomic``, it takes an extra
4911 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4912 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4913 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4914 when they may see multiple atomic stores. The type of the pointee must
4915 be an integer type whose bit width is a power of two greater than or
4916 equal to eight and less than or equal to a target-specific size limit.
4917 ``align`` must be explicitly specified on atomic stores, and the store
4918 has undefined behavior if the alignment is not set to a value which is
4919 at least the size in bytes of the pointee. ``!nontemporal`` does not
4920 have any defined semantics for atomic stores.
4922 The optional constant ``align`` argument specifies the alignment of the
4923 operation (that is, the alignment of the memory address). A value of 0
4924 or an omitted ``align`` argument means that the operation has the ABI
4925 alignment for the target. It is the responsibility of the code emitter
4926 to ensure that the alignment information is correct. Overestimating the
4927 alignment results in undefined behavior. Underestimating the
4928 alignment may produce less efficient code. An alignment of 1 is always
4931 The optional ``!nontemporal`` metadata must reference a single metadata
4932 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4933 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4934 tells the optimizer and code generator that this load is not expected to
4935 be reused in the cache. The code generator may select special
4936 instructions to save cache bandwidth, such as the MOVNT instruction on
4942 The contents of memory are updated to contain ``<value>`` at the
4943 location specified by the ``<pointer>`` operand. If ``<value>`` is
4944 of scalar type then the number of bytes written does not exceed the
4945 minimum number of bytes needed to hold all bits of the type. For
4946 example, storing an ``i24`` writes at most three bytes. When writing a
4947 value of a type like ``i20`` with a size that is not an integral number
4948 of bytes, it is unspecified what happens to the extra bits that do not
4949 belong to the type, but they will typically be overwritten.
4954 .. code-block:: llvm
4956 %ptr = alloca i32 ; yields {i32*}:ptr
4957 store i32 3, i32* %ptr ; yields {void}
4958 %val = load i32* %ptr ; yields {i32}:val = i32 3
4962 '``fence``' Instruction
4963 ^^^^^^^^^^^^^^^^^^^^^^^
4970 fence [singlethread] <ordering> ; yields {void}
4975 The '``fence``' instruction is used to introduce happens-before edges
4981 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4982 defines what *synchronizes-with* edges they add. They can only be given
4983 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4988 A fence A which has (at least) ``release`` ordering semantics
4989 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4990 semantics if and only if there exist atomic operations X and Y, both
4991 operating on some atomic object M, such that A is sequenced before X, X
4992 modifies M (either directly or through some side effect of a sequence
4993 headed by X), Y is sequenced before B, and Y observes M. This provides a
4994 *happens-before* dependency between A and B. Rather than an explicit
4995 ``fence``, one (but not both) of the atomic operations X or Y might
4996 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4997 still *synchronize-with* the explicit ``fence`` and establish the
4998 *happens-before* edge.
5000 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5001 ``acquire`` and ``release`` semantics specified above, participates in
5002 the global program order of other ``seq_cst`` operations and/or fences.
5004 The optional ":ref:`singlethread <singlethread>`" argument specifies
5005 that the fence only synchronizes with other fences in the same thread.
5006 (This is useful for interacting with signal handlers.)
5011 .. code-block:: llvm
5013 fence acquire ; yields {void}
5014 fence singlethread seq_cst ; yields {void}
5018 '``cmpxchg``' Instruction
5019 ^^^^^^^^^^^^^^^^^^^^^^^^^
5026 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields {ty}
5031 The '``cmpxchg``' instruction is used to atomically modify memory. It
5032 loads a value in memory and compares it to a given value. If they are
5033 equal, it stores a new value into the memory.
5038 There are three arguments to the '``cmpxchg``' instruction: an address
5039 to operate on, a value to compare to the value currently be at that
5040 address, and a new value to place at that address if the compared values
5041 are equal. The type of '<cmp>' must be an integer type whose bit width
5042 is a power of two greater than or equal to eight and less than or equal
5043 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5044 type, and the type of '<pointer>' must be a pointer to that type. If the
5045 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5046 to modify the number or order of execution of this ``cmpxchg`` with
5047 other :ref:`volatile operations <volatile>`.
5049 The success and failure :ref:`ordering <ordering>` arguments specify how this
5050 ``cmpxchg`` synchronizes with other atomic operations. The both ordering
5051 parameters must be at least ``monotonic``, the ordering constraint on failure
5052 must be no stronger than that on success, and the failure ordering cannot be
5053 either ``release`` or ``acq_rel``.
5055 The optional "``singlethread``" argument declares that the ``cmpxchg``
5056 is only atomic with respect to code (usually signal handlers) running in
5057 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5058 respect to all other code in the system.
5060 The pointer passed into cmpxchg must have alignment greater than or
5061 equal to the size in memory of the operand.
5066 The contents of memory at the location specified by the '``<pointer>``'
5067 operand is read and compared to '``<cmp>``'; if the read value is the
5068 equal, '``<new>``' is written. The original value at the location is
5071 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5072 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5073 load with an ordering parameter determined the second ordering parameter.
5078 .. code-block:: llvm
5081 %orig = atomic load i32* %ptr unordered ; yields {i32}
5085 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5086 %squared = mul i32 %cmp, %cmp
5087 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields {i32}
5088 %success = icmp eq i32 %cmp, %old
5089 br i1 %success, label %done, label %loop
5096 '``atomicrmw``' Instruction
5097 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5104 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
5109 The '``atomicrmw``' instruction is used to atomically modify memory.
5114 There are three arguments to the '``atomicrmw``' instruction: an
5115 operation to apply, an address whose value to modify, an argument to the
5116 operation. The operation must be one of the following keywords:
5130 The type of '<value>' must be an integer type whose bit width is a power
5131 of two greater than or equal to eight and less than or equal to a
5132 target-specific size limit. The type of the '``<pointer>``' operand must
5133 be a pointer to that type. If the ``atomicrmw`` is marked as
5134 ``volatile``, then the optimizer is not allowed to modify the number or
5135 order of execution of this ``atomicrmw`` with other :ref:`volatile
5136 operations <volatile>`.
5141 The contents of memory at the location specified by the '``<pointer>``'
5142 operand are atomically read, modified, and written back. The original
5143 value at the location is returned. The modification is specified by the
5146 - xchg: ``*ptr = val``
5147 - add: ``*ptr = *ptr + val``
5148 - sub: ``*ptr = *ptr - val``
5149 - and: ``*ptr = *ptr & val``
5150 - nand: ``*ptr = ~(*ptr & val)``
5151 - or: ``*ptr = *ptr | val``
5152 - xor: ``*ptr = *ptr ^ val``
5153 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5154 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5155 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5157 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5163 .. code-block:: llvm
5165 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5167 .. _i_getelementptr:
5169 '``getelementptr``' Instruction
5170 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5177 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5178 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5179 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5184 The '``getelementptr``' instruction is used to get the address of a
5185 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5186 address calculation only and does not access memory.
5191 The first argument is always a pointer or a vector of pointers, and
5192 forms the basis of the calculation. The remaining arguments are indices
5193 that indicate which of the elements of the aggregate object are indexed.
5194 The interpretation of each index is dependent on the type being indexed
5195 into. The first index always indexes the pointer value given as the
5196 first argument, the second index indexes a value of the type pointed to
5197 (not necessarily the value directly pointed to, since the first index
5198 can be non-zero), etc. The first type indexed into must be a pointer
5199 value, subsequent types can be arrays, vectors, and structs. Note that
5200 subsequent types being indexed into can never be pointers, since that
5201 would require loading the pointer before continuing calculation.
5203 The type of each index argument depends on the type it is indexing into.
5204 When indexing into a (optionally packed) structure, only ``i32`` integer
5205 **constants** are allowed (when using a vector of indices they must all
5206 be the **same** ``i32`` integer constant). When indexing into an array,
5207 pointer or vector, integers of any width are allowed, and they are not
5208 required to be constant. These integers are treated as signed values
5211 For example, let's consider a C code fragment and how it gets compiled
5227 int *foo(struct ST *s) {
5228 return &s[1].Z.B[5][13];
5231 The LLVM code generated by Clang is:
5233 .. code-block:: llvm
5235 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5236 %struct.ST = type { i32, double, %struct.RT }
5238 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5240 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5247 In the example above, the first index is indexing into the
5248 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5249 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5250 indexes into the third element of the structure, yielding a
5251 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5252 structure. The third index indexes into the second element of the
5253 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5254 dimensions of the array are subscripted into, yielding an '``i32``'
5255 type. The '``getelementptr``' instruction returns a pointer to this
5256 element, thus computing a value of '``i32*``' type.
5258 Note that it is perfectly legal to index partially through a structure,
5259 returning a pointer to an inner element. Because of this, the LLVM code
5260 for the given testcase is equivalent to:
5262 .. code-block:: llvm
5264 define i32* @foo(%struct.ST* %s) {
5265 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5266 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5267 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5268 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5269 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5273 If the ``inbounds`` keyword is present, the result value of the
5274 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5275 pointer is not an *in bounds* address of an allocated object, or if any
5276 of the addresses that would be formed by successive addition of the
5277 offsets implied by the indices to the base address with infinitely
5278 precise signed arithmetic are not an *in bounds* address of that
5279 allocated object. The *in bounds* addresses for an allocated object are
5280 all the addresses that point into the object, plus the address one byte
5281 past the end. In cases where the base is a vector of pointers the
5282 ``inbounds`` keyword applies to each of the computations element-wise.
5284 If the ``inbounds`` keyword is not present, the offsets are added to the
5285 base address with silently-wrapping two's complement arithmetic. If the
5286 offsets have a different width from the pointer, they are sign-extended
5287 or truncated to the width of the pointer. The result value of the
5288 ``getelementptr`` may be outside the object pointed to by the base
5289 pointer. The result value may not necessarily be used to access memory
5290 though, even if it happens to point into allocated storage. See the
5291 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5294 The getelementptr instruction is often confusing. For some more insight
5295 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5300 .. code-block:: llvm
5302 ; yields [12 x i8]*:aptr
5303 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5305 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5307 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5309 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5311 In cases where the pointer argument is a vector of pointers, each index
5312 must be a vector with the same number of elements. For example:
5314 .. code-block:: llvm
5316 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5318 Conversion Operations
5319 ---------------------
5321 The instructions in this category are the conversion instructions
5322 (casting) which all take a single operand and a type. They perform
5323 various bit conversions on the operand.
5325 '``trunc .. to``' Instruction
5326 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5333 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5338 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5343 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5344 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5345 of the same number of integers. The bit size of the ``value`` must be
5346 larger than the bit size of the destination type, ``ty2``. Equal sized
5347 types are not allowed.
5352 The '``trunc``' instruction truncates the high order bits in ``value``
5353 and converts the remaining bits to ``ty2``. Since the source size must
5354 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5355 It will always truncate bits.
5360 .. code-block:: llvm
5362 %X = trunc i32 257 to i8 ; yields i8:1
5363 %Y = trunc i32 123 to i1 ; yields i1:true
5364 %Z = trunc i32 122 to i1 ; yields i1:false
5365 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5367 '``zext .. to``' Instruction
5368 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5375 <result> = zext <ty> <value> to <ty2> ; yields ty2
5380 The '``zext``' instruction zero extends its operand to type ``ty2``.
5385 The '``zext``' instruction takes a value to cast, and a type to cast it
5386 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5387 the same number of integers. The bit size of the ``value`` must be
5388 smaller than the bit size of the destination type, ``ty2``.
5393 The ``zext`` fills the high order bits of the ``value`` with zero bits
5394 until it reaches the size of the destination type, ``ty2``.
5396 When zero extending from i1, the result will always be either 0 or 1.
5401 .. code-block:: llvm
5403 %X = zext i32 257 to i64 ; yields i64:257
5404 %Y = zext i1 true to i32 ; yields i32:1
5405 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5407 '``sext .. to``' Instruction
5408 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5415 <result> = sext <ty> <value> to <ty2> ; yields ty2
5420 The '``sext``' sign extends ``value`` to the type ``ty2``.
5425 The '``sext``' instruction takes a value to cast, and a type to cast it
5426 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5427 the same number of integers. The bit size of the ``value`` must be
5428 smaller than the bit size of the destination type, ``ty2``.
5433 The '``sext``' instruction performs a sign extension by copying the sign
5434 bit (highest order bit) of the ``value`` until it reaches the bit size
5435 of the type ``ty2``.
5437 When sign extending from i1, the extension always results in -1 or 0.
5442 .. code-block:: llvm
5444 %X = sext i8 -1 to i16 ; yields i16 :65535
5445 %Y = sext i1 true to i32 ; yields i32:-1
5446 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5448 '``fptrunc .. to``' Instruction
5449 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5456 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5461 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5466 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5467 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5468 The size of ``value`` must be larger than the size of ``ty2``. This
5469 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5474 The '``fptrunc``' instruction truncates a ``value`` from a larger
5475 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5476 point <t_floating>` type. If the value cannot fit within the
5477 destination type, ``ty2``, then the results are undefined.
5482 .. code-block:: llvm
5484 %X = fptrunc double 123.0 to float ; yields float:123.0
5485 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5487 '``fpext .. to``' Instruction
5488 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5495 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5500 The '``fpext``' extends a floating point ``value`` to a larger floating
5506 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5507 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5508 to. The source type must be smaller than the destination type.
5513 The '``fpext``' instruction extends the ``value`` from a smaller
5514 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5515 point <t_floating>` type. The ``fpext`` cannot be used to make a
5516 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5517 *no-op cast* for a floating point cast.
5522 .. code-block:: llvm
5524 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5525 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5527 '``fptoui .. to``' Instruction
5528 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5535 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5540 The '``fptoui``' converts a floating point ``value`` to its unsigned
5541 integer equivalent of type ``ty2``.
5546 The '``fptoui``' instruction takes a value to cast, which must be a
5547 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5548 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5549 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5550 type with the same number of elements as ``ty``
5555 The '``fptoui``' instruction converts its :ref:`floating
5556 point <t_floating>` operand into the nearest (rounding towards zero)
5557 unsigned integer value. If the value cannot fit in ``ty2``, the results
5563 .. code-block:: llvm
5565 %X = fptoui double 123.0 to i32 ; yields i32:123
5566 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5567 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5569 '``fptosi .. to``' Instruction
5570 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5577 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5582 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5583 ``value`` to type ``ty2``.
5588 The '``fptosi``' instruction takes a value to cast, which must be a
5589 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5590 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5591 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5592 type with the same number of elements as ``ty``
5597 The '``fptosi``' instruction converts its :ref:`floating
5598 point <t_floating>` operand into the nearest (rounding towards zero)
5599 signed integer value. If the value cannot fit in ``ty2``, the results
5605 .. code-block:: llvm
5607 %X = fptosi double -123.0 to i32 ; yields i32:-123
5608 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5609 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5611 '``uitofp .. to``' Instruction
5612 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5619 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5624 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5625 and converts that value to the ``ty2`` type.
5630 The '``uitofp``' instruction takes a value to cast, which must be a
5631 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5632 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5633 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5634 type with the same number of elements as ``ty``
5639 The '``uitofp``' instruction interprets its operand as an unsigned
5640 integer quantity and converts it to the corresponding floating point
5641 value. If the value cannot fit in the floating point value, the results
5647 .. code-block:: llvm
5649 %X = uitofp i32 257 to float ; yields float:257.0
5650 %Y = uitofp i8 -1 to double ; yields double:255.0
5652 '``sitofp .. to``' Instruction
5653 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5660 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5665 The '``sitofp``' instruction regards ``value`` as a signed integer and
5666 converts that value to the ``ty2`` type.
5671 The '``sitofp``' instruction takes a value to cast, which must be a
5672 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5673 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5674 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5675 type with the same number of elements as ``ty``
5680 The '``sitofp``' instruction interprets its operand as a signed integer
5681 quantity and converts it to the corresponding floating point value. If
5682 the value cannot fit in the floating point value, the results are
5688 .. code-block:: llvm
5690 %X = sitofp i32 257 to float ; yields float:257.0
5691 %Y = sitofp i8 -1 to double ; yields double:-1.0
5695 '``ptrtoint .. to``' Instruction
5696 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5703 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5708 The '``ptrtoint``' instruction converts the pointer or a vector of
5709 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5714 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5715 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5716 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5717 a vector of integers type.
5722 The '``ptrtoint``' instruction converts ``value`` to integer type
5723 ``ty2`` by interpreting the pointer value as an integer and either
5724 truncating or zero extending that value to the size of the integer type.
5725 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5726 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5727 the same size, then nothing is done (*no-op cast*) other than a type
5733 .. code-block:: llvm
5735 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5736 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5737 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5741 '``inttoptr .. to``' Instruction
5742 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5749 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5754 The '``inttoptr``' instruction converts an integer ``value`` to a
5755 pointer type, ``ty2``.
5760 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5761 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5767 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5768 applying either a zero extension or a truncation depending on the size
5769 of the integer ``value``. If ``value`` is larger than the size of a
5770 pointer then a truncation is done. If ``value`` is smaller than the size
5771 of a pointer then a zero extension is done. If they are the same size,
5772 nothing is done (*no-op cast*).
5777 .. code-block:: llvm
5779 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5780 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5781 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5782 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5786 '``bitcast .. to``' Instruction
5787 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5794 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5799 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5805 The '``bitcast``' instruction takes a value to cast, which must be a
5806 non-aggregate first class value, and a type to cast it to, which must
5807 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5808 bit sizes of ``value`` and the destination type, ``ty2``, must be
5809 identical. If the source type is a pointer, the destination type must
5810 also be a pointer of the same size. This instruction supports bitwise
5811 conversion of vectors to integers and to vectors of other types (as
5812 long as they have the same size).
5817 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5818 is always a *no-op cast* because no bits change with this
5819 conversion. The conversion is done as if the ``value`` had been stored
5820 to memory and read back as type ``ty2``. Pointer (or vector of
5821 pointers) types may only be converted to other pointer (or vector of
5822 pointers) types with the same address space through this instruction.
5823 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5824 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5829 .. code-block:: llvm
5831 %X = bitcast i8 255 to i8 ; yields i8 :-1
5832 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5833 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5834 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5836 .. _i_addrspacecast:
5838 '``addrspacecast .. to``' Instruction
5839 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5846 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5851 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5852 address space ``n`` to type ``pty2`` in address space ``m``.
5857 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5858 to cast and a pointer type to cast it to, which must have a different
5864 The '``addrspacecast``' instruction converts the pointer value
5865 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5866 value modification, depending on the target and the address space
5867 pair. Pointer conversions within the same address space must be
5868 performed with the ``bitcast`` instruction. Note that if the address space
5869 conversion is legal then both result and operand refer to the same memory
5875 .. code-block:: llvm
5877 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5878 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5879 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5886 The instructions in this category are the "miscellaneous" instructions,
5887 which defy better classification.
5891 '``icmp``' Instruction
5892 ^^^^^^^^^^^^^^^^^^^^^^
5899 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5904 The '``icmp``' instruction returns a boolean value or a vector of
5905 boolean values based on comparison of its two integer, integer vector,
5906 pointer, or pointer vector operands.
5911 The '``icmp``' instruction takes three operands. The first operand is
5912 the condition code indicating the kind of comparison to perform. It is
5913 not a value, just a keyword. The possible condition code are:
5916 #. ``ne``: not equal
5917 #. ``ugt``: unsigned greater than
5918 #. ``uge``: unsigned greater or equal
5919 #. ``ult``: unsigned less than
5920 #. ``ule``: unsigned less or equal
5921 #. ``sgt``: signed greater than
5922 #. ``sge``: signed greater or equal
5923 #. ``slt``: signed less than
5924 #. ``sle``: signed less or equal
5926 The remaining two arguments must be :ref:`integer <t_integer>` or
5927 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5928 must also be identical types.
5933 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5934 code given as ``cond``. The comparison performed always yields either an
5935 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5937 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5938 otherwise. No sign interpretation is necessary or performed.
5939 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5940 otherwise. No sign interpretation is necessary or performed.
5941 #. ``ugt``: interprets the operands as unsigned values and yields
5942 ``true`` if ``op1`` is greater than ``op2``.
5943 #. ``uge``: interprets the operands as unsigned values and yields
5944 ``true`` if ``op1`` is greater than or equal to ``op2``.
5945 #. ``ult``: interprets the operands as unsigned values and yields
5946 ``true`` if ``op1`` is less than ``op2``.
5947 #. ``ule``: interprets the operands as unsigned values and yields
5948 ``true`` if ``op1`` is less than or equal to ``op2``.
5949 #. ``sgt``: interprets the operands as signed values and yields ``true``
5950 if ``op1`` is greater than ``op2``.
5951 #. ``sge``: interprets the operands as signed values and yields ``true``
5952 if ``op1`` is greater than or equal to ``op2``.
5953 #. ``slt``: interprets the operands as signed values and yields ``true``
5954 if ``op1`` is less than ``op2``.
5955 #. ``sle``: interprets the operands as signed values and yields ``true``
5956 if ``op1`` is less than or equal to ``op2``.
5958 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5959 are compared as if they were integers.
5961 If the operands are integer vectors, then they are compared element by
5962 element. The result is an ``i1`` vector with the same number of elements
5963 as the values being compared. Otherwise, the result is an ``i1``.
5968 .. code-block:: llvm
5970 <result> = icmp eq i32 4, 5 ; yields: result=false
5971 <result> = icmp ne float* %X, %X ; yields: result=false
5972 <result> = icmp ult i16 4, 5 ; yields: result=true
5973 <result> = icmp sgt i16 4, 5 ; yields: result=false
5974 <result> = icmp ule i16 -4, 5 ; yields: result=false
5975 <result> = icmp sge i16 4, 5 ; yields: result=false
5977 Note that the code generator does not yet support vector types with the
5978 ``icmp`` instruction.
5982 '``fcmp``' Instruction
5983 ^^^^^^^^^^^^^^^^^^^^^^
5990 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5995 The '``fcmp``' instruction returns a boolean value or vector of boolean
5996 values based on comparison of its operands.
5998 If the operands are floating point scalars, then the result type is a
5999 boolean (:ref:`i1 <t_integer>`).
6001 If the operands are floating point vectors, then the result type is a
6002 vector of boolean with the same number of elements as the operands being
6008 The '``fcmp``' instruction takes three operands. The first operand is
6009 the condition code indicating the kind of comparison to perform. It is
6010 not a value, just a keyword. The possible condition code are:
6012 #. ``false``: no comparison, always returns false
6013 #. ``oeq``: ordered and equal
6014 #. ``ogt``: ordered and greater than
6015 #. ``oge``: ordered and greater than or equal
6016 #. ``olt``: ordered and less than
6017 #. ``ole``: ordered and less than or equal
6018 #. ``one``: ordered and not equal
6019 #. ``ord``: ordered (no nans)
6020 #. ``ueq``: unordered or equal
6021 #. ``ugt``: unordered or greater than
6022 #. ``uge``: unordered or greater than or equal
6023 #. ``ult``: unordered or less than
6024 #. ``ule``: unordered or less than or equal
6025 #. ``une``: unordered or not equal
6026 #. ``uno``: unordered (either nans)
6027 #. ``true``: no comparison, always returns true
6029 *Ordered* means that neither operand is a QNAN while *unordered* means
6030 that either operand may be a QNAN.
6032 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6033 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6034 type. They must have identical types.
6039 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6040 condition code given as ``cond``. If the operands are vectors, then the
6041 vectors are compared element by element. Each comparison performed
6042 always yields an :ref:`i1 <t_integer>` result, as follows:
6044 #. ``false``: always yields ``false``, regardless of operands.
6045 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6046 is equal to ``op2``.
6047 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6048 is greater than ``op2``.
6049 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6050 is greater than or equal to ``op2``.
6051 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6052 is less than ``op2``.
6053 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6054 is less than or equal to ``op2``.
6055 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6056 is not equal to ``op2``.
6057 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6058 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6060 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6061 greater than ``op2``.
6062 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6063 greater than or equal to ``op2``.
6064 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6066 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6067 less than or equal to ``op2``.
6068 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6069 not equal to ``op2``.
6070 #. ``uno``: yields ``true`` if either operand is a QNAN.
6071 #. ``true``: always yields ``true``, regardless of operands.
6076 .. code-block:: llvm
6078 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6079 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6080 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6081 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6083 Note that the code generator does not yet support vector types with the
6084 ``fcmp`` instruction.
6088 '``phi``' Instruction
6089 ^^^^^^^^^^^^^^^^^^^^^
6096 <result> = phi <ty> [ <val0>, <label0>], ...
6101 The '``phi``' instruction is used to implement the φ node in the SSA
6102 graph representing the function.
6107 The type of the incoming values is specified with the first type field.
6108 After this, the '``phi``' instruction takes a list of pairs as
6109 arguments, with one pair for each predecessor basic block of the current
6110 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6111 the value arguments to the PHI node. Only labels may be used as the
6114 There must be no non-phi instructions between the start of a basic block
6115 and the PHI instructions: i.e. PHI instructions must be first in a basic
6118 For the purposes of the SSA form, the use of each incoming value is
6119 deemed to occur on the edge from the corresponding predecessor block to
6120 the current block (but after any definition of an '``invoke``'
6121 instruction's return value on the same edge).
6126 At runtime, the '``phi``' instruction logically takes on the value
6127 specified by the pair corresponding to the predecessor basic block that
6128 executed just prior to the current block.
6133 .. code-block:: llvm
6135 Loop: ; Infinite loop that counts from 0 on up...
6136 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6137 %nextindvar = add i32 %indvar, 1
6142 '``select``' Instruction
6143 ^^^^^^^^^^^^^^^^^^^^^^^^
6150 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6152 selty is either i1 or {<N x i1>}
6157 The '``select``' instruction is used to choose one value based on a
6158 condition, without IR-level branching.
6163 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6164 values indicating the condition, and two values of the same :ref:`first
6165 class <t_firstclass>` type. If the val1/val2 are vectors and the
6166 condition is a scalar, then entire vectors are selected, not individual
6172 If the condition is an i1 and it evaluates to 1, the instruction returns
6173 the first value argument; otherwise, it returns the second value
6176 If the condition is a vector of i1, then the value arguments must be
6177 vectors of the same size, and the selection is done element by element.
6182 .. code-block:: llvm
6184 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6188 '``call``' Instruction
6189 ^^^^^^^^^^^^^^^^^^^^^^
6196 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6201 The '``call``' instruction represents a simple function call.
6206 This instruction requires several arguments:
6208 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
6209 should perform tail call optimization. The ``tail`` marker is a hint that
6210 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
6211 means that the call must be tail call optimized in order for the program to
6212 be correct. The ``musttail`` marker provides these guarantees:
6214 #. The call will not cause unbounded stack growth if it is part of a
6215 recursive cycle in the call graph.
6216 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
6219 Both markers imply that the callee does not access allocas or varargs from
6220 the caller. Calls marked ``musttail`` must obey the following additional
6223 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
6224 or a pointer bitcast followed by a ret instruction.
6225 - The ret instruction must return the (possibly bitcasted) value
6226 produced by the call or void.
6227 - The caller and callee prototypes must match. Pointer types of
6228 parameters or return types may differ in pointee type, but not
6230 - The calling conventions of the caller and callee must match.
6231 - All ABI-impacting function attributes, such as sret, byval, inreg,
6232 returned, and inalloca, must match.
6234 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
6235 the following conditions are met:
6237 - Caller and callee both have the calling convention ``fastcc``.
6238 - The call is in tail position (ret immediately follows call and ret
6239 uses value of call or is void).
6240 - Option ``-tailcallopt`` is enabled, or
6241 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6242 - `Platform specific constraints are
6243 met. <CodeGenerator.html#tailcallopt>`_
6245 #. The optional "cconv" marker indicates which :ref:`calling
6246 convention <callingconv>` the call should use. If none is
6247 specified, the call defaults to using C calling conventions. The
6248 calling convention of the call must match the calling convention of
6249 the target function, or else the behavior is undefined.
6250 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6251 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6253 #. '``ty``': the type of the call instruction itself which is also the
6254 type of the return value. Functions that return no value are marked
6256 #. '``fnty``': shall be the signature of the pointer to function value
6257 being invoked. The argument types must match the types implied by
6258 this signature. This type can be omitted if the function is not
6259 varargs and if the function type does not return a pointer to a
6261 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6262 be invoked. In most cases, this is a direct function invocation, but
6263 indirect ``call``'s are just as possible, calling an arbitrary pointer
6265 #. '``function args``': argument list whose types match the function
6266 signature argument types and parameter attributes. All arguments must
6267 be of :ref:`first class <t_firstclass>` type. If the function signature
6268 indicates the function accepts a variable number of arguments, the
6269 extra arguments can be specified.
6270 #. The optional :ref:`function attributes <fnattrs>` list. Only
6271 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6272 attributes are valid here.
6277 The '``call``' instruction is used to cause control flow to transfer to
6278 a specified function, with its incoming arguments bound to the specified
6279 values. Upon a '``ret``' instruction in the called function, control
6280 flow continues with the instruction after the function call, and the
6281 return value of the function is bound to the result argument.
6286 .. code-block:: llvm
6288 %retval = call i32 @test(i32 %argc)
6289 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6290 %X = tail call i32 @foo() ; yields i32
6291 %Y = tail call fastcc i32 @foo() ; yields i32
6292 call void %foo(i8 97 signext)
6294 %struct.A = type { i32, i8 }
6295 %r = call %struct.A @foo() ; yields { 32, i8 }
6296 %gr = extractvalue %struct.A %r, 0 ; yields i32
6297 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6298 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6299 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6301 llvm treats calls to some functions with names and arguments that match
6302 the standard C99 library as being the C99 library functions, and may
6303 perform optimizations or generate code for them under that assumption.
6304 This is something we'd like to change in the future to provide better
6305 support for freestanding environments and non-C-based languages.
6309 '``va_arg``' Instruction
6310 ^^^^^^^^^^^^^^^^^^^^^^^^
6317 <resultval> = va_arg <va_list*> <arglist>, <argty>
6322 The '``va_arg``' instruction is used to access arguments passed through
6323 the "variable argument" area of a function call. It is used to implement
6324 the ``va_arg`` macro in C.
6329 This instruction takes a ``va_list*`` value and the type of the
6330 argument. It returns a value of the specified argument type and
6331 increments the ``va_list`` to point to the next argument. The actual
6332 type of ``va_list`` is target specific.
6337 The '``va_arg``' instruction loads an argument of the specified type
6338 from the specified ``va_list`` and causes the ``va_list`` to point to
6339 the next argument. For more information, see the variable argument
6340 handling :ref:`Intrinsic Functions <int_varargs>`.
6342 It is legal for this instruction to be called in a function which does
6343 not take a variable number of arguments, for example, the ``vfprintf``
6346 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6347 function <intrinsics>` because it takes a type as an argument.
6352 See the :ref:`variable argument processing <int_varargs>` section.
6354 Note that the code generator does not yet fully support va\_arg on many
6355 targets. Also, it does not currently support va\_arg with aggregate
6356 types on any target.
6360 '``landingpad``' Instruction
6361 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6368 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6369 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6371 <clause> := catch <type> <value>
6372 <clause> := filter <array constant type> <array constant>
6377 The '``landingpad``' instruction is used by `LLVM's exception handling
6378 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6379 is a landing pad --- one where the exception lands, and corresponds to the
6380 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6381 defines values supplied by the personality function (``pers_fn``) upon
6382 re-entry to the function. The ``resultval`` has the type ``resultty``.
6387 This instruction takes a ``pers_fn`` value. This is the personality
6388 function associated with the unwinding mechanism. The optional
6389 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6391 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6392 contains the global variable representing the "type" that may be caught
6393 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6394 clause takes an array constant as its argument. Use
6395 "``[0 x i8**] undef``" for a filter which cannot throw. The
6396 '``landingpad``' instruction must contain *at least* one ``clause`` or
6397 the ``cleanup`` flag.
6402 The '``landingpad``' instruction defines the values which are set by the
6403 personality function (``pers_fn``) upon re-entry to the function, and
6404 therefore the "result type" of the ``landingpad`` instruction. As with
6405 calling conventions, how the personality function results are
6406 represented in LLVM IR is target specific.
6408 The clauses are applied in order from top to bottom. If two
6409 ``landingpad`` instructions are merged together through inlining, the
6410 clauses from the calling function are appended to the list of clauses.
6411 When the call stack is being unwound due to an exception being thrown,
6412 the exception is compared against each ``clause`` in turn. If it doesn't
6413 match any of the clauses, and the ``cleanup`` flag is not set, then
6414 unwinding continues further up the call stack.
6416 The ``landingpad`` instruction has several restrictions:
6418 - A landing pad block is a basic block which is the unwind destination
6419 of an '``invoke``' instruction.
6420 - A landing pad block must have a '``landingpad``' instruction as its
6421 first non-PHI instruction.
6422 - There can be only one '``landingpad``' instruction within the landing
6424 - A basic block that is not a landing pad block may not include a
6425 '``landingpad``' instruction.
6426 - All '``landingpad``' instructions in a function must have the same
6427 personality function.
6432 .. code-block:: llvm
6434 ;; A landing pad which can catch an integer.
6435 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6437 ;; A landing pad that is a cleanup.
6438 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6440 ;; A landing pad which can catch an integer and can only throw a double.
6441 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6443 filter [1 x i8**] [@_ZTId]
6450 LLVM supports the notion of an "intrinsic function". These functions
6451 have well known names and semantics and are required to follow certain
6452 restrictions. Overall, these intrinsics represent an extension mechanism
6453 for the LLVM language that does not require changing all of the
6454 transformations in LLVM when adding to the language (or the bitcode
6455 reader/writer, the parser, etc...).
6457 Intrinsic function names must all start with an "``llvm.``" prefix. This
6458 prefix is reserved in LLVM for intrinsic names; thus, function names may
6459 not begin with this prefix. Intrinsic functions must always be external
6460 functions: you cannot define the body of intrinsic functions. Intrinsic
6461 functions may only be used in call or invoke instructions: it is illegal
6462 to take the address of an intrinsic function. Additionally, because
6463 intrinsic functions are part of the LLVM language, it is required if any
6464 are added that they be documented here.
6466 Some intrinsic functions can be overloaded, i.e., the intrinsic
6467 represents a family of functions that perform the same operation but on
6468 different data types. Because LLVM can represent over 8 million
6469 different integer types, overloading is used commonly to allow an
6470 intrinsic function to operate on any integer type. One or more of the
6471 argument types or the result type can be overloaded to accept any
6472 integer type. Argument types may also be defined as exactly matching a
6473 previous argument's type or the result type. This allows an intrinsic
6474 function which accepts multiple arguments, but needs all of them to be
6475 of the same type, to only be overloaded with respect to a single
6476 argument or the result.
6478 Overloaded intrinsics will have the names of its overloaded argument
6479 types encoded into its function name, each preceded by a period. Only
6480 those types which are overloaded result in a name suffix. Arguments
6481 whose type is matched against another type do not. For example, the
6482 ``llvm.ctpop`` function can take an integer of any width and returns an
6483 integer of exactly the same integer width. This leads to a family of
6484 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6485 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6486 overloaded, and only one type suffix is required. Because the argument's
6487 type is matched against the return type, it does not require its own
6490 To learn how to add an intrinsic function, please see the `Extending
6491 LLVM Guide <ExtendingLLVM.html>`_.
6495 Variable Argument Handling Intrinsics
6496 -------------------------------------
6498 Variable argument support is defined in LLVM with the
6499 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6500 functions. These functions are related to the similarly named macros
6501 defined in the ``<stdarg.h>`` header file.
6503 All of these functions operate on arguments that use a target-specific
6504 value type "``va_list``". The LLVM assembly language reference manual
6505 does not define what this type is, so all transformations should be
6506 prepared to handle these functions regardless of the type used.
6508 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6509 variable argument handling intrinsic functions are used.
6511 .. code-block:: llvm
6513 define i32 @test(i32 %X, ...) {
6514 ; Initialize variable argument processing
6516 %ap2 = bitcast i8** %ap to i8*
6517 call void @llvm.va_start(i8* %ap2)
6519 ; Read a single integer argument
6520 %tmp = va_arg i8** %ap, i32
6522 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6524 %aq2 = bitcast i8** %aq to i8*
6525 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6526 call void @llvm.va_end(i8* %aq2)
6528 ; Stop processing of arguments.
6529 call void @llvm.va_end(i8* %ap2)
6533 declare void @llvm.va_start(i8*)
6534 declare void @llvm.va_copy(i8*, i8*)
6535 declare void @llvm.va_end(i8*)
6539 '``llvm.va_start``' Intrinsic
6540 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6547 declare void @llvm.va_start(i8* <arglist>)
6552 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6553 subsequent use by ``va_arg``.
6558 The argument is a pointer to a ``va_list`` element to initialize.
6563 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6564 available in C. In a target-dependent way, it initializes the
6565 ``va_list`` element to which the argument points, so that the next call
6566 to ``va_arg`` will produce the first variable argument passed to the
6567 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6568 to know the last argument of the function as the compiler can figure
6571 '``llvm.va_end``' Intrinsic
6572 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6579 declare void @llvm.va_end(i8* <arglist>)
6584 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6585 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6590 The argument is a pointer to a ``va_list`` to destroy.
6595 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6596 available in C. In a target-dependent way, it destroys the ``va_list``
6597 element to which the argument points. Calls to
6598 :ref:`llvm.va_start <int_va_start>` and
6599 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6604 '``llvm.va_copy``' Intrinsic
6605 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6612 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6617 The '``llvm.va_copy``' intrinsic copies the current argument position
6618 from the source argument list to the destination argument list.
6623 The first argument is a pointer to a ``va_list`` element to initialize.
6624 The second argument is a pointer to a ``va_list`` element to copy from.
6629 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6630 available in C. In a target-dependent way, it copies the source
6631 ``va_list`` element into the destination ``va_list`` element. This
6632 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6633 arbitrarily complex and require, for example, memory allocation.
6635 Accurate Garbage Collection Intrinsics
6636 --------------------------------------
6638 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6639 (GC) requires the implementation and generation of these intrinsics.
6640 These intrinsics allow identification of :ref:`GC roots on the
6641 stack <int_gcroot>`, as well as garbage collector implementations that
6642 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6643 Front-ends for type-safe garbage collected languages should generate
6644 these intrinsics to make use of the LLVM garbage collectors. For more
6645 details, see `Accurate Garbage Collection with
6646 LLVM <GarbageCollection.html>`_.
6648 The garbage collection intrinsics only operate on objects in the generic
6649 address space (address space zero).
6653 '``llvm.gcroot``' Intrinsic
6654 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6661 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6666 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6667 the code generator, and allows some metadata to be associated with it.
6672 The first argument specifies the address of a stack object that contains
6673 the root pointer. The second pointer (which must be either a constant or
6674 a global value address) contains the meta-data to be associated with the
6680 At runtime, a call to this intrinsic stores a null pointer into the
6681 "ptrloc" location. At compile-time, the code generator generates
6682 information to allow the runtime to find the pointer at GC safe points.
6683 The '``llvm.gcroot``' intrinsic may only be used in a function which
6684 :ref:`specifies a GC algorithm <gc>`.
6688 '``llvm.gcread``' Intrinsic
6689 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6696 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6701 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6702 locations, allowing garbage collector implementations that require read
6708 The second argument is the address to read from, which should be an
6709 address allocated from the garbage collector. The first object is a
6710 pointer to the start of the referenced object, if needed by the language
6711 runtime (otherwise null).
6716 The '``llvm.gcread``' intrinsic has the same semantics as a load
6717 instruction, but may be replaced with substantially more complex code by
6718 the garbage collector runtime, as needed. The '``llvm.gcread``'
6719 intrinsic may only be used in a function which :ref:`specifies a GC
6724 '``llvm.gcwrite``' Intrinsic
6725 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6732 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6737 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6738 locations, allowing garbage collector implementations that require write
6739 barriers (such as generational or reference counting collectors).
6744 The first argument is the reference to store, the second is the start of
6745 the object to store it to, and the third is the address of the field of
6746 Obj to store to. If the runtime does not require a pointer to the
6747 object, Obj may be null.
6752 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6753 instruction, but may be replaced with substantially more complex code by
6754 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6755 intrinsic may only be used in a function which :ref:`specifies a GC
6758 Code Generator Intrinsics
6759 -------------------------
6761 These intrinsics are provided by LLVM to expose special features that
6762 may only be implemented with code generator support.
6764 '``llvm.returnaddress``' Intrinsic
6765 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6772 declare i8 *@llvm.returnaddress(i32 <level>)
6777 The '``llvm.returnaddress``' intrinsic attempts to compute a
6778 target-specific value indicating the return address of the current
6779 function or one of its callers.
6784 The argument to this intrinsic indicates which function to return the
6785 address for. Zero indicates the calling function, one indicates its
6786 caller, etc. The argument is **required** to be a constant integer
6792 The '``llvm.returnaddress``' intrinsic either returns a pointer
6793 indicating the return address of the specified call frame, or zero if it
6794 cannot be identified. The value returned by this intrinsic is likely to
6795 be incorrect or 0 for arguments other than zero, so it should only be
6796 used for debugging purposes.
6798 Note that calling this intrinsic does not prevent function inlining or
6799 other aggressive transformations, so the value returned may not be that
6800 of the obvious source-language caller.
6802 '``llvm.frameaddress``' Intrinsic
6803 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6810 declare i8* @llvm.frameaddress(i32 <level>)
6815 The '``llvm.frameaddress``' intrinsic attempts to return the
6816 target-specific frame pointer value for the specified stack frame.
6821 The argument to this intrinsic indicates which function to return the
6822 frame pointer for. Zero indicates the calling function, one indicates
6823 its caller, etc. The argument is **required** to be a constant integer
6829 The '``llvm.frameaddress``' intrinsic either returns a pointer
6830 indicating the frame address of the specified call frame, or zero if it
6831 cannot be identified. The value returned by this intrinsic is likely to
6832 be incorrect or 0 for arguments other than zero, so it should only be
6833 used for debugging purposes.
6835 Note that calling this intrinsic does not prevent function inlining or
6836 other aggressive transformations, so the value returned may not be that
6837 of the obvious source-language caller.
6839 .. _int_read_register:
6840 .. _int_write_register:
6842 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
6843 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6850 declare i32 @llvm.read_register.i32(metadata)
6851 declare i64 @llvm.read_register.i64(metadata)
6852 declare void @llvm.write_register.i32(metadata, i32 @value)
6853 declare void @llvm.write_register.i64(metadata, i64 @value)
6854 !0 = metadata !{metadata !"sp\00"}
6859 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
6860 provides access to the named register. The register must be valid on
6861 the architecture being compiled to. The type needs to be compatible
6862 with the register being read.
6867 The '``llvm.read_register``' intrinsic returns the current value of the
6868 register, where possible. The '``llvm.write_register``' intrinsic sets
6869 the current value of the register, where possible.
6871 This is useful to implement named register global variables that need
6872 to always be mapped to a specific register, as is common practice on
6873 bare-metal programs including OS kernels.
6875 The compiler doesn't check for register availability or use of the used
6876 register in surrounding code, including inline assembly. Because of that,
6877 allocatable registers are not supported.
6879 Warning: So far it only works with the stack pointer on selected
6880 architectures (ARM, ARM64, AArch64, PowerPC and x86_64). Significant amount of
6881 work is needed to support other registers and even more so, allocatable
6886 '``llvm.stacksave``' Intrinsic
6887 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6894 declare i8* @llvm.stacksave()
6899 The '``llvm.stacksave``' intrinsic is used to remember the current state
6900 of the function stack, for use with
6901 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6902 implementing language features like scoped automatic variable sized
6908 This intrinsic returns a opaque pointer value that can be passed to
6909 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6910 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6911 ``llvm.stacksave``, it effectively restores the state of the stack to
6912 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6913 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6914 were allocated after the ``llvm.stacksave`` was executed.
6916 .. _int_stackrestore:
6918 '``llvm.stackrestore``' Intrinsic
6919 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6926 declare void @llvm.stackrestore(i8* %ptr)
6931 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6932 the function stack to the state it was in when the corresponding
6933 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6934 useful for implementing language features like scoped automatic variable
6935 sized arrays in C99.
6940 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6942 '``llvm.prefetch``' Intrinsic
6943 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6950 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6955 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6956 insert a prefetch instruction if supported; otherwise, it is a noop.
6957 Prefetches have no effect on the behavior of the program but can change
6958 its performance characteristics.
6963 ``address`` is the address to be prefetched, ``rw`` is the specifier
6964 determining if the fetch should be for a read (0) or write (1), and
6965 ``locality`` is a temporal locality specifier ranging from (0) - no
6966 locality, to (3) - extremely local keep in cache. The ``cache type``
6967 specifies whether the prefetch is performed on the data (1) or
6968 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6969 arguments must be constant integers.
6974 This intrinsic does not modify the behavior of the program. In
6975 particular, prefetches cannot trap and do not produce a value. On
6976 targets that support this intrinsic, the prefetch can provide hints to
6977 the processor cache for better performance.
6979 '``llvm.pcmarker``' Intrinsic
6980 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6987 declare void @llvm.pcmarker(i32 <id>)
6992 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6993 Counter (PC) in a region of code to simulators and other tools. The
6994 method is target specific, but it is expected that the marker will use
6995 exported symbols to transmit the PC of the marker. The marker makes no
6996 guarantees that it will remain with any specific instruction after
6997 optimizations. It is possible that the presence of a marker will inhibit
6998 optimizations. The intended use is to be inserted after optimizations to
6999 allow correlations of simulation runs.
7004 ``id`` is a numerical id identifying the marker.
7009 This intrinsic does not modify the behavior of the program. Backends
7010 that do not support this intrinsic may ignore it.
7012 '``llvm.readcyclecounter``' Intrinsic
7013 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7020 declare i64 @llvm.readcyclecounter()
7025 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
7026 counter register (or similar low latency, high accuracy clocks) on those
7027 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7028 should map to RPCC. As the backing counters overflow quickly (on the
7029 order of 9 seconds on alpha), this should only be used for small
7035 When directly supported, reading the cycle counter should not modify any
7036 memory. Implementations are allowed to either return a application
7037 specific value or a system wide value. On backends without support, this
7038 is lowered to a constant 0.
7040 Note that runtime support may be conditional on the privilege-level code is
7041 running at and the host platform.
7043 '``llvm.clear_cache``' Intrinsic
7044 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7051 declare void @llvm.clear_cache(i8*, i8*)
7056 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
7057 in the specified range to the execution unit of the processor. On
7058 targets with non-unified instruction and data cache, the implementation
7059 flushes the instruction cache.
7064 On platforms with coherent instruction and data caches (e.g. x86), this
7065 intrinsic is a nop. On platforms with non-coherent instruction and data
7066 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
7067 instructions or a system call, if cache flushing requires special
7070 The default behavior is to emit a call to ``__clear_cache`` from the run
7073 This instrinsic does *not* empty the instruction pipeline. Modifications
7074 of the current function are outside the scope of the intrinsic.
7076 Standard C Library Intrinsics
7077 -----------------------------
7079 LLVM provides intrinsics for a few important standard C library
7080 functions. These intrinsics allow source-language front-ends to pass
7081 information about the alignment of the pointer arguments to the code
7082 generator, providing opportunity for more efficient code generation.
7086 '``llvm.memcpy``' Intrinsic
7087 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7092 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
7093 integer bit width and for different address spaces. Not all targets
7094 support all bit widths however.
7098 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7099 i32 <len>, i32 <align>, i1 <isvolatile>)
7100 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7101 i64 <len>, i32 <align>, i1 <isvolatile>)
7106 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7107 source location to the destination location.
7109 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
7110 intrinsics do not return a value, takes extra alignment/isvolatile
7111 arguments and the pointers can be in specified address spaces.
7116 The first argument is a pointer to the destination, the second is a
7117 pointer to the source. The third argument is an integer argument
7118 specifying the number of bytes to copy, the fourth argument is the
7119 alignment of the source and destination locations, and the fifth is a
7120 boolean indicating a volatile access.
7122 If the call to this intrinsic has an alignment value that is not 0 or 1,
7123 then the caller guarantees that both the source and destination pointers
7124 are aligned to that boundary.
7126 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
7127 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7128 very cleanly specified and it is unwise to depend on it.
7133 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7134 source location to the destination location, which are not allowed to
7135 overlap. It copies "len" bytes of memory over. If the argument is known
7136 to be aligned to some boundary, this can be specified as the fourth
7137 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
7139 '``llvm.memmove``' Intrinsic
7140 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7145 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7146 bit width and for different address space. Not all targets support all
7151 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7152 i32 <len>, i32 <align>, i1 <isvolatile>)
7153 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7154 i64 <len>, i32 <align>, i1 <isvolatile>)
7159 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7160 source location to the destination location. It is similar to the
7161 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7164 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7165 intrinsics do not return a value, takes extra alignment/isvolatile
7166 arguments and the pointers can be in specified address spaces.
7171 The first argument is a pointer to the destination, the second is a
7172 pointer to the source. The third argument is an integer argument
7173 specifying the number of bytes to copy, the fourth argument is the
7174 alignment of the source and destination locations, and the fifth is a
7175 boolean indicating a volatile access.
7177 If the call to this intrinsic has an alignment value that is not 0 or 1,
7178 then the caller guarantees that the source and destination pointers are
7179 aligned to that boundary.
7181 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7182 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7183 not very cleanly specified and it is unwise to depend on it.
7188 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7189 source location to the destination location, which may overlap. It
7190 copies "len" bytes of memory over. If the argument is known to be
7191 aligned to some boundary, this can be specified as the fourth argument,
7192 otherwise it should be set to 0 or 1 (both meaning no alignment).
7194 '``llvm.memset.*``' Intrinsics
7195 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7200 This is an overloaded intrinsic. You can use llvm.memset on any integer
7201 bit width and for different address spaces. However, not all targets
7202 support all bit widths.
7206 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7207 i32 <len>, i32 <align>, i1 <isvolatile>)
7208 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7209 i64 <len>, i32 <align>, i1 <isvolatile>)
7214 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7215 particular byte value.
7217 Note that, unlike the standard libc function, the ``llvm.memset``
7218 intrinsic does not return a value and takes extra alignment/volatile
7219 arguments. Also, the destination can be in an arbitrary address space.
7224 The first argument is a pointer to the destination to fill, the second
7225 is the byte value with which to fill it, the third argument is an
7226 integer argument specifying the number of bytes to fill, and the fourth
7227 argument is the known alignment of the destination location.
7229 If the call to this intrinsic has an alignment value that is not 0 or 1,
7230 then the caller guarantees that the destination pointer is aligned to
7233 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7234 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7235 very cleanly specified and it is unwise to depend on it.
7240 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7241 at the destination location. If the argument is known to be aligned to
7242 some boundary, this can be specified as the fourth argument, otherwise
7243 it should be set to 0 or 1 (both meaning no alignment).
7245 '``llvm.sqrt.*``' Intrinsic
7246 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7251 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7252 floating point or vector of floating point type. Not all targets support
7257 declare float @llvm.sqrt.f32(float %Val)
7258 declare double @llvm.sqrt.f64(double %Val)
7259 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7260 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7261 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7266 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7267 returning the same value as the libm '``sqrt``' functions would. Unlike
7268 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7269 negative numbers other than -0.0 (which allows for better optimization,
7270 because there is no need to worry about errno being set).
7271 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7276 The argument and return value are floating point numbers of the same
7282 This function returns the sqrt of the specified operand if it is a
7283 nonnegative floating point number.
7285 '``llvm.powi.*``' Intrinsic
7286 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7291 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7292 floating point or vector of floating point type. Not all targets support
7297 declare float @llvm.powi.f32(float %Val, i32 %power)
7298 declare double @llvm.powi.f64(double %Val, i32 %power)
7299 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7300 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7301 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7306 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7307 specified (positive or negative) power. The order of evaluation of
7308 multiplications is not defined. When a vector of floating point type is
7309 used, the second argument remains a scalar integer value.
7314 The second argument is an integer power, and the first is a value to
7315 raise to that power.
7320 This function returns the first value raised to the second power with an
7321 unspecified sequence of rounding operations.
7323 '``llvm.sin.*``' Intrinsic
7324 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7329 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7330 floating point or vector of floating point type. Not all targets support
7335 declare float @llvm.sin.f32(float %Val)
7336 declare double @llvm.sin.f64(double %Val)
7337 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7338 declare fp128 @llvm.sin.f128(fp128 %Val)
7339 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7344 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7349 The argument and return value are floating point numbers of the same
7355 This function returns the sine of the specified operand, returning the
7356 same values as the libm ``sin`` functions would, and handles error
7357 conditions in the same way.
7359 '``llvm.cos.*``' Intrinsic
7360 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7365 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7366 floating point or vector of floating point type. Not all targets support
7371 declare float @llvm.cos.f32(float %Val)
7372 declare double @llvm.cos.f64(double %Val)
7373 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7374 declare fp128 @llvm.cos.f128(fp128 %Val)
7375 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7380 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7385 The argument and return value are floating point numbers of the same
7391 This function returns the cosine of the specified operand, returning the
7392 same values as the libm ``cos`` functions would, and handles error
7393 conditions in the same way.
7395 '``llvm.pow.*``' Intrinsic
7396 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7401 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7402 floating point or vector of floating point type. Not all targets support
7407 declare float @llvm.pow.f32(float %Val, float %Power)
7408 declare double @llvm.pow.f64(double %Val, double %Power)
7409 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7410 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7411 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7416 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7417 specified (positive or negative) power.
7422 The second argument is a floating point power, and the first is a value
7423 to raise to that power.
7428 This function returns the first value raised to the second power,
7429 returning the same values as the libm ``pow`` functions would, and
7430 handles error conditions in the same way.
7432 '``llvm.exp.*``' Intrinsic
7433 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7438 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7439 floating point or vector of floating point type. Not all targets support
7444 declare float @llvm.exp.f32(float %Val)
7445 declare double @llvm.exp.f64(double %Val)
7446 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7447 declare fp128 @llvm.exp.f128(fp128 %Val)
7448 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7453 The '``llvm.exp.*``' intrinsics perform the exp function.
7458 The argument and return value are floating point numbers of the same
7464 This function returns the same values as the libm ``exp`` functions
7465 would, and handles error conditions in the same way.
7467 '``llvm.exp2.*``' Intrinsic
7468 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7473 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7474 floating point or vector of floating point type. Not all targets support
7479 declare float @llvm.exp2.f32(float %Val)
7480 declare double @llvm.exp2.f64(double %Val)
7481 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7482 declare fp128 @llvm.exp2.f128(fp128 %Val)
7483 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7488 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7493 The argument and return value are floating point numbers of the same
7499 This function returns the same values as the libm ``exp2`` functions
7500 would, and handles error conditions in the same way.
7502 '``llvm.log.*``' Intrinsic
7503 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7508 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7509 floating point or vector of floating point type. Not all targets support
7514 declare float @llvm.log.f32(float %Val)
7515 declare double @llvm.log.f64(double %Val)
7516 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7517 declare fp128 @llvm.log.f128(fp128 %Val)
7518 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7523 The '``llvm.log.*``' intrinsics perform the log function.
7528 The argument and return value are floating point numbers of the same
7534 This function returns the same values as the libm ``log`` functions
7535 would, and handles error conditions in the same way.
7537 '``llvm.log10.*``' Intrinsic
7538 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7543 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7544 floating point or vector of floating point type. Not all targets support
7549 declare float @llvm.log10.f32(float %Val)
7550 declare double @llvm.log10.f64(double %Val)
7551 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7552 declare fp128 @llvm.log10.f128(fp128 %Val)
7553 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7558 The '``llvm.log10.*``' intrinsics perform the log10 function.
7563 The argument and return value are floating point numbers of the same
7569 This function returns the same values as the libm ``log10`` functions
7570 would, and handles error conditions in the same way.
7572 '``llvm.log2.*``' Intrinsic
7573 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7578 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7579 floating point or vector of floating point type. Not all targets support
7584 declare float @llvm.log2.f32(float %Val)
7585 declare double @llvm.log2.f64(double %Val)
7586 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7587 declare fp128 @llvm.log2.f128(fp128 %Val)
7588 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7593 The '``llvm.log2.*``' intrinsics perform the log2 function.
7598 The argument and return value are floating point numbers of the same
7604 This function returns the same values as the libm ``log2`` functions
7605 would, and handles error conditions in the same way.
7607 '``llvm.fma.*``' Intrinsic
7608 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7613 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7614 floating point or vector of floating point type. Not all targets support
7619 declare float @llvm.fma.f32(float %a, float %b, float %c)
7620 declare double @llvm.fma.f64(double %a, double %b, double %c)
7621 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7622 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7623 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7628 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7634 The argument and return value are floating point numbers of the same
7640 This function returns the same values as the libm ``fma`` functions
7641 would, and does not set errno.
7643 '``llvm.fabs.*``' Intrinsic
7644 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7649 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7650 floating point or vector of floating point type. Not all targets support
7655 declare float @llvm.fabs.f32(float %Val)
7656 declare double @llvm.fabs.f64(double %Val)
7657 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7658 declare fp128 @llvm.fabs.f128(fp128 %Val)
7659 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7664 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7670 The argument and return value are floating point numbers of the same
7676 This function returns the same values as the libm ``fabs`` functions
7677 would, and handles error conditions in the same way.
7679 '``llvm.copysign.*``' Intrinsic
7680 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7685 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7686 floating point or vector of floating point type. Not all targets support
7691 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7692 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7693 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7694 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7695 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7700 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7701 first operand and the sign of the second operand.
7706 The arguments and return value are floating point numbers of the same
7712 This function returns the same values as the libm ``copysign``
7713 functions would, and handles error conditions in the same way.
7715 '``llvm.floor.*``' Intrinsic
7716 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7721 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7722 floating point or vector of floating point type. Not all targets support
7727 declare float @llvm.floor.f32(float %Val)
7728 declare double @llvm.floor.f64(double %Val)
7729 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7730 declare fp128 @llvm.floor.f128(fp128 %Val)
7731 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7736 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7741 The argument and return value are floating point numbers of the same
7747 This function returns the same values as the libm ``floor`` functions
7748 would, and handles error conditions in the same way.
7750 '``llvm.ceil.*``' Intrinsic
7751 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7756 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7757 floating point or vector of floating point type. Not all targets support
7762 declare float @llvm.ceil.f32(float %Val)
7763 declare double @llvm.ceil.f64(double %Val)
7764 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7765 declare fp128 @llvm.ceil.f128(fp128 %Val)
7766 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7771 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7776 The argument and return value are floating point numbers of the same
7782 This function returns the same values as the libm ``ceil`` functions
7783 would, and handles error conditions in the same way.
7785 '``llvm.trunc.*``' Intrinsic
7786 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7791 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7792 floating point or vector of floating point type. Not all targets support
7797 declare float @llvm.trunc.f32(float %Val)
7798 declare double @llvm.trunc.f64(double %Val)
7799 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7800 declare fp128 @llvm.trunc.f128(fp128 %Val)
7801 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7806 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7807 nearest integer not larger in magnitude than the operand.
7812 The argument and return value are floating point numbers of the same
7818 This function returns the same values as the libm ``trunc`` functions
7819 would, and handles error conditions in the same way.
7821 '``llvm.rint.*``' Intrinsic
7822 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7827 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7828 floating point or vector of floating point type. Not all targets support
7833 declare float @llvm.rint.f32(float %Val)
7834 declare double @llvm.rint.f64(double %Val)
7835 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7836 declare fp128 @llvm.rint.f128(fp128 %Val)
7837 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7842 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7843 nearest integer. It may raise an inexact floating-point exception if the
7844 operand isn't an integer.
7849 The argument and return value are floating point numbers of the same
7855 This function returns the same values as the libm ``rint`` functions
7856 would, and handles error conditions in the same way.
7858 '``llvm.nearbyint.*``' Intrinsic
7859 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7864 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7865 floating point or vector of floating point type. Not all targets support
7870 declare float @llvm.nearbyint.f32(float %Val)
7871 declare double @llvm.nearbyint.f64(double %Val)
7872 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7873 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7874 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7879 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7885 The argument and return value are floating point numbers of the same
7891 This function returns the same values as the libm ``nearbyint``
7892 functions would, and handles error conditions in the same way.
7894 '``llvm.round.*``' Intrinsic
7895 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7900 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7901 floating point or vector of floating point type. Not all targets support
7906 declare float @llvm.round.f32(float %Val)
7907 declare double @llvm.round.f64(double %Val)
7908 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7909 declare fp128 @llvm.round.f128(fp128 %Val)
7910 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7915 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7921 The argument and return value are floating point numbers of the same
7927 This function returns the same values as the libm ``round``
7928 functions would, and handles error conditions in the same way.
7930 Bit Manipulation Intrinsics
7931 ---------------------------
7933 LLVM provides intrinsics for a few important bit manipulation
7934 operations. These allow efficient code generation for some algorithms.
7936 '``llvm.bswap.*``' Intrinsics
7937 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7942 This is an overloaded intrinsic function. You can use bswap on any
7943 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7947 declare i16 @llvm.bswap.i16(i16 <id>)
7948 declare i32 @llvm.bswap.i32(i32 <id>)
7949 declare i64 @llvm.bswap.i64(i64 <id>)
7954 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7955 values with an even number of bytes (positive multiple of 16 bits).
7956 These are useful for performing operations on data that is not in the
7957 target's native byte order.
7962 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7963 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7964 intrinsic returns an i32 value that has the four bytes of the input i32
7965 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7966 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7967 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7968 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7971 '``llvm.ctpop.*``' Intrinsic
7972 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7977 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7978 bit width, or on any vector with integer elements. Not all targets
7979 support all bit widths or vector types, however.
7983 declare i8 @llvm.ctpop.i8(i8 <src>)
7984 declare i16 @llvm.ctpop.i16(i16 <src>)
7985 declare i32 @llvm.ctpop.i32(i32 <src>)
7986 declare i64 @llvm.ctpop.i64(i64 <src>)
7987 declare i256 @llvm.ctpop.i256(i256 <src>)
7988 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7993 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7999 The only argument is the value to be counted. The argument may be of any
8000 integer type, or a vector with integer elements. The return type must
8001 match the argument type.
8006 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
8007 each element of a vector.
8009 '``llvm.ctlz.*``' Intrinsic
8010 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8015 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
8016 integer bit width, or any vector whose elements are integers. Not all
8017 targets support all bit widths or vector types, however.
8021 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
8022 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
8023 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
8024 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
8025 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
8026 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8031 The '``llvm.ctlz``' family of intrinsic functions counts the number of
8032 leading zeros in a variable.
8037 The first argument is the value to be counted. This argument may be of
8038 any integer type, or a vectory with integer element type. The return
8039 type must match the first argument type.
8041 The second argument must be a constant and is a flag to indicate whether
8042 the intrinsic should ensure that a zero as the first argument produces a
8043 defined result. Historically some architectures did not provide a
8044 defined result for zero values as efficiently, and many algorithms are
8045 now predicated on avoiding zero-value inputs.
8050 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
8051 zeros in a variable, or within each element of the vector. If
8052 ``src == 0`` then the result is the size in bits of the type of ``src``
8053 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8054 ``llvm.ctlz(i32 2) = 30``.
8056 '``llvm.cttz.*``' Intrinsic
8057 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8062 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
8063 integer bit width, or any vector of integer elements. Not all targets
8064 support all bit widths or vector types, however.
8068 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
8069 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
8070 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
8071 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
8072 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
8073 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8078 The '``llvm.cttz``' family of intrinsic functions counts the number of
8084 The first argument is the value to be counted. This argument may be of
8085 any integer type, or a vectory with integer element type. The return
8086 type must match the first argument type.
8088 The second argument must be a constant and is a flag to indicate whether
8089 the intrinsic should ensure that a zero as the first argument produces a
8090 defined result. Historically some architectures did not provide a
8091 defined result for zero values as efficiently, and many algorithms are
8092 now predicated on avoiding zero-value inputs.
8097 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
8098 zeros in a variable, or within each element of a vector. If ``src == 0``
8099 then the result is the size in bits of the type of ``src`` if
8100 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8101 ``llvm.cttz(2) = 1``.
8103 Arithmetic with Overflow Intrinsics
8104 -----------------------------------
8106 LLVM provides intrinsics for some arithmetic with overflow operations.
8108 '``llvm.sadd.with.overflow.*``' Intrinsics
8109 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8114 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
8115 on any integer bit width.
8119 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8120 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8121 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8126 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8127 a signed addition of the two arguments, and indicate whether an overflow
8128 occurred during the signed summation.
8133 The arguments (%a and %b) and the first element of the result structure
8134 may be of integer types of any bit width, but they must have the same
8135 bit width. The second element of the result structure must be of type
8136 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8142 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8143 a signed addition of the two variables. They return a structure --- the
8144 first element of which is the signed summation, and the second element
8145 of which is a bit specifying if the signed summation resulted in an
8151 .. code-block:: llvm
8153 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8154 %sum = extractvalue {i32, i1} %res, 0
8155 %obit = extractvalue {i32, i1} %res, 1
8156 br i1 %obit, label %overflow, label %normal
8158 '``llvm.uadd.with.overflow.*``' Intrinsics
8159 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8164 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8165 on any integer bit width.
8169 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8170 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8171 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8176 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8177 an unsigned addition of the two arguments, and indicate whether a carry
8178 occurred during the unsigned summation.
8183 The arguments (%a and %b) and the first element of the result structure
8184 may be of integer types of any bit width, but they must have the same
8185 bit width. The second element of the result structure must be of type
8186 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8192 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8193 an unsigned addition of the two arguments. They return a structure --- the
8194 first element of which is the sum, and the second element of which is a
8195 bit specifying if the unsigned summation resulted in a carry.
8200 .. code-block:: llvm
8202 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8203 %sum = extractvalue {i32, i1} %res, 0
8204 %obit = extractvalue {i32, i1} %res, 1
8205 br i1 %obit, label %carry, label %normal
8207 '``llvm.ssub.with.overflow.*``' Intrinsics
8208 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8213 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8214 on any integer bit width.
8218 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8219 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8220 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8225 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8226 a signed subtraction of the two arguments, and indicate whether an
8227 overflow occurred during the signed subtraction.
8232 The arguments (%a and %b) and the first element of the result structure
8233 may be of integer types of any bit width, but they must have the same
8234 bit width. The second element of the result structure must be of type
8235 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8241 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8242 a signed subtraction of the two arguments. They return a structure --- the
8243 first element of which is the subtraction, and the second element of
8244 which is a bit specifying if the signed subtraction resulted in an
8250 .. code-block:: llvm
8252 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8253 %sum = extractvalue {i32, i1} %res, 0
8254 %obit = extractvalue {i32, i1} %res, 1
8255 br i1 %obit, label %overflow, label %normal
8257 '``llvm.usub.with.overflow.*``' Intrinsics
8258 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8263 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8264 on any integer bit width.
8268 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8269 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8270 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8275 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8276 an unsigned subtraction of the two arguments, and indicate whether an
8277 overflow occurred during the unsigned subtraction.
8282 The arguments (%a and %b) and the first element of the result structure
8283 may be of integer types of any bit width, but they must have the same
8284 bit width. The second element of the result structure must be of type
8285 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8291 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8292 an unsigned subtraction of the two arguments. They return a structure ---
8293 the first element of which is the subtraction, and the second element of
8294 which is a bit specifying if the unsigned subtraction resulted in an
8300 .. code-block:: llvm
8302 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8303 %sum = extractvalue {i32, i1} %res, 0
8304 %obit = extractvalue {i32, i1} %res, 1
8305 br i1 %obit, label %overflow, label %normal
8307 '``llvm.smul.with.overflow.*``' Intrinsics
8308 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8313 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8314 on any integer bit width.
8318 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8319 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8320 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8325 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8326 a signed multiplication of the two arguments, and indicate whether an
8327 overflow occurred during the signed multiplication.
8332 The arguments (%a and %b) and the first element of the result structure
8333 may be of integer types of any bit width, but they must have the same
8334 bit width. The second element of the result structure must be of type
8335 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8341 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8342 a signed multiplication of the two arguments. They return a structure ---
8343 the first element of which is the multiplication, and the second element
8344 of which is a bit specifying if the signed multiplication resulted in an
8350 .. code-block:: llvm
8352 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8353 %sum = extractvalue {i32, i1} %res, 0
8354 %obit = extractvalue {i32, i1} %res, 1
8355 br i1 %obit, label %overflow, label %normal
8357 '``llvm.umul.with.overflow.*``' Intrinsics
8358 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8363 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8364 on any integer bit width.
8368 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8369 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8370 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8375 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8376 a unsigned multiplication of the two arguments, and indicate whether an
8377 overflow occurred during the unsigned multiplication.
8382 The arguments (%a and %b) and the first element of the result structure
8383 may be of integer types of any bit width, but they must have the same
8384 bit width. The second element of the result structure must be of type
8385 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8391 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8392 an unsigned multiplication of the two arguments. They return a structure ---
8393 the first element of which is the multiplication, and the second
8394 element of which is a bit specifying if the unsigned multiplication
8395 resulted in an overflow.
8400 .. code-block:: llvm
8402 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8403 %sum = extractvalue {i32, i1} %res, 0
8404 %obit = extractvalue {i32, i1} %res, 1
8405 br i1 %obit, label %overflow, label %normal
8407 Specialised Arithmetic Intrinsics
8408 ---------------------------------
8410 '``llvm.fmuladd.*``' Intrinsic
8411 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8418 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8419 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8424 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8425 expressions that can be fused if the code generator determines that (a) the
8426 target instruction set has support for a fused operation, and (b) that the
8427 fused operation is more efficient than the equivalent, separate pair of mul
8428 and add instructions.
8433 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8434 multiplicands, a and b, and an addend c.
8443 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8445 is equivalent to the expression a \* b + c, except that rounding will
8446 not be performed between the multiplication and addition steps if the
8447 code generator fuses the operations. Fusion is not guaranteed, even if
8448 the target platform supports it. If a fused multiply-add is required the
8449 corresponding llvm.fma.\* intrinsic function should be used
8450 instead. This never sets errno, just as '``llvm.fma.*``'.
8455 .. code-block:: llvm
8457 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8459 Half Precision Floating Point Intrinsics
8460 ----------------------------------------
8462 For most target platforms, half precision floating point is a
8463 storage-only format. This means that it is a dense encoding (in memory)
8464 but does not support computation in the format.
8466 This means that code must first load the half-precision floating point
8467 value as an i16, then convert it to float with
8468 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8469 then be performed on the float value (including extending to double
8470 etc). To store the value back to memory, it is first converted to float
8471 if needed, then converted to i16 with
8472 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8475 .. _int_convert_to_fp16:
8477 '``llvm.convert.to.fp16``' Intrinsic
8478 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8485 declare i16 @llvm.convert.to.fp16(f32 %a)
8490 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8491 from single precision floating point format to half precision floating
8497 The intrinsic function contains single argument - the value to be
8503 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8504 from single precision floating point format to half precision floating
8505 point format. The return value is an ``i16`` which contains the
8511 .. code-block:: llvm
8513 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8514 store i16 %res, i16* @x, align 2
8516 .. _int_convert_from_fp16:
8518 '``llvm.convert.from.fp16``' Intrinsic
8519 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8526 declare f32 @llvm.convert.from.fp16(i16 %a)
8531 The '``llvm.convert.from.fp16``' intrinsic function performs a
8532 conversion from half precision floating point format to single precision
8533 floating point format.
8538 The intrinsic function contains single argument - the value to be
8544 The '``llvm.convert.from.fp16``' intrinsic function performs a
8545 conversion from half single precision floating point format to single
8546 precision floating point format. The input half-float value is
8547 represented by an ``i16`` value.
8552 .. code-block:: llvm
8554 %a = load i16* @x, align 2
8555 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8560 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8561 prefix), are described in the `LLVM Source Level
8562 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8565 Exception Handling Intrinsics
8566 -----------------------------
8568 The LLVM exception handling intrinsics (which all start with
8569 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8570 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8574 Trampoline Intrinsics
8575 ---------------------
8577 These intrinsics make it possible to excise one parameter, marked with
8578 the :ref:`nest <nest>` attribute, from a function. The result is a
8579 callable function pointer lacking the nest parameter - the caller does
8580 not need to provide a value for it. Instead, the value to use is stored
8581 in advance in a "trampoline", a block of memory usually allocated on the
8582 stack, which also contains code to splice the nest value into the
8583 argument list. This is used to implement the GCC nested function address
8586 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8587 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8588 It can be created as follows:
8590 .. code-block:: llvm
8592 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8593 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8594 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8595 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8596 %fp = bitcast i8* %p to i32 (i32, i32)*
8598 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8599 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8603 '``llvm.init.trampoline``' Intrinsic
8604 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8611 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8616 This fills the memory pointed to by ``tramp`` with executable code,
8617 turning it into a trampoline.
8622 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8623 pointers. The ``tramp`` argument must point to a sufficiently large and
8624 sufficiently aligned block of memory; this memory is written to by the
8625 intrinsic. Note that the size and the alignment are target-specific -
8626 LLVM currently provides no portable way of determining them, so a
8627 front-end that generates this intrinsic needs to have some
8628 target-specific knowledge. The ``func`` argument must hold a function
8629 bitcast to an ``i8*``.
8634 The block of memory pointed to by ``tramp`` is filled with target
8635 dependent code, turning it into a function. Then ``tramp`` needs to be
8636 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8637 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8638 function's signature is the same as that of ``func`` with any arguments
8639 marked with the ``nest`` attribute removed. At most one such ``nest``
8640 argument is allowed, and it must be of pointer type. Calling the new
8641 function is equivalent to calling ``func`` with the same argument list,
8642 but with ``nval`` used for the missing ``nest`` argument. If, after
8643 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8644 modified, then the effect of any later call to the returned function
8645 pointer is undefined.
8649 '``llvm.adjust.trampoline``' Intrinsic
8650 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8657 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8662 This performs any required machine-specific adjustment to the address of
8663 a trampoline (passed as ``tramp``).
8668 ``tramp`` must point to a block of memory which already has trampoline
8669 code filled in by a previous call to
8670 :ref:`llvm.init.trampoline <int_it>`.
8675 On some architectures the address of the code to be executed needs to be
8676 different to the address where the trampoline is actually stored. This
8677 intrinsic returns the executable address corresponding to ``tramp``
8678 after performing the required machine specific adjustments. The pointer
8679 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8684 This class of intrinsics exists to information about the lifetime of
8685 memory objects and ranges where variables are immutable.
8689 '``llvm.lifetime.start``' Intrinsic
8690 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8697 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8702 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8708 The first argument is a constant integer representing the size of the
8709 object, or -1 if it is variable sized. The second argument is a pointer
8715 This intrinsic indicates that before this point in the code, the value
8716 of the memory pointed to by ``ptr`` is dead. This means that it is known
8717 to never be used and has an undefined value. A load from the pointer
8718 that precedes this intrinsic can be replaced with ``'undef'``.
8722 '``llvm.lifetime.end``' Intrinsic
8723 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8730 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8735 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8741 The first argument is a constant integer representing the size of the
8742 object, or -1 if it is variable sized. The second argument is a pointer
8748 This intrinsic indicates that after this point in the code, the value of
8749 the memory pointed to by ``ptr`` is dead. This means that it is known to
8750 never be used and has an undefined value. Any stores into the memory
8751 object following this intrinsic may be removed as dead.
8753 '``llvm.invariant.start``' Intrinsic
8754 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8761 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8766 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8767 a memory object will not change.
8772 The first argument is a constant integer representing the size of the
8773 object, or -1 if it is variable sized. The second argument is a pointer
8779 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8780 the return value, the referenced memory location is constant and
8783 '``llvm.invariant.end``' Intrinsic
8784 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8791 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8796 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8797 memory object are mutable.
8802 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8803 The second argument is a constant integer representing the size of the
8804 object, or -1 if it is variable sized and the third argument is a
8805 pointer to the object.
8810 This intrinsic indicates that the memory is mutable again.
8815 This class of intrinsics is designed to be generic and has no specific
8818 '``llvm.var.annotation``' Intrinsic
8819 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8826 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8831 The '``llvm.var.annotation``' intrinsic.
8836 The first argument is a pointer to a value, the second is a pointer to a
8837 global string, the third is a pointer to a global string which is the
8838 source file name, and the last argument is the line number.
8843 This intrinsic allows annotation of local variables with arbitrary
8844 strings. This can be useful for special purpose optimizations that want
8845 to look for these annotations. These have no other defined use; they are
8846 ignored by code generation and optimization.
8848 '``llvm.ptr.annotation.*``' Intrinsic
8849 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8854 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8855 pointer to an integer of any width. *NOTE* you must specify an address space for
8856 the pointer. The identifier for the default address space is the integer
8861 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8862 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8863 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8864 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8865 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8870 The '``llvm.ptr.annotation``' intrinsic.
8875 The first argument is a pointer to an integer value of arbitrary bitwidth
8876 (result of some expression), the second is a pointer to a global string, the
8877 third is a pointer to a global string which is the source file name, and the
8878 last argument is the line number. It returns the value of the first argument.
8883 This intrinsic allows annotation of a pointer to an integer with arbitrary
8884 strings. This can be useful for special purpose optimizations that want to look
8885 for these annotations. These have no other defined use; they are ignored by code
8886 generation and optimization.
8888 '``llvm.annotation.*``' Intrinsic
8889 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8894 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8895 any integer bit width.
8899 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8900 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8901 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8902 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8903 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8908 The '``llvm.annotation``' intrinsic.
8913 The first argument is an integer value (result of some expression), the
8914 second is a pointer to a global string, the third is a pointer to a
8915 global string which is the source file name, and the last argument is
8916 the line number. It returns the value of the first argument.
8921 This intrinsic allows annotations to be put on arbitrary expressions
8922 with arbitrary strings. This can be useful for special purpose
8923 optimizations that want to look for these annotations. These have no
8924 other defined use; they are ignored by code generation and optimization.
8926 '``llvm.trap``' Intrinsic
8927 ^^^^^^^^^^^^^^^^^^^^^^^^^
8934 declare void @llvm.trap() noreturn nounwind
8939 The '``llvm.trap``' intrinsic.
8949 This intrinsic is lowered to the target dependent trap instruction. If
8950 the target does not have a trap instruction, this intrinsic will be
8951 lowered to a call of the ``abort()`` function.
8953 '``llvm.debugtrap``' Intrinsic
8954 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8961 declare void @llvm.debugtrap() nounwind
8966 The '``llvm.debugtrap``' intrinsic.
8976 This intrinsic is lowered to code which is intended to cause an
8977 execution trap with the intention of requesting the attention of a
8980 '``llvm.stackprotector``' Intrinsic
8981 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8988 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8993 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8994 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8995 is placed on the stack before local variables.
9000 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
9001 The first argument is the value loaded from the stack guard
9002 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
9003 enough space to hold the value of the guard.
9008 This intrinsic causes the prologue/epilogue inserter to force the position of
9009 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
9010 to ensure that if a local variable on the stack is overwritten, it will destroy
9011 the value of the guard. When the function exits, the guard on the stack is
9012 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
9013 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
9014 calling the ``__stack_chk_fail()`` function.
9016 '``llvm.stackprotectorcheck``' Intrinsic
9017 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9024 declare void @llvm.stackprotectorcheck(i8** <guard>)
9029 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
9030 created stack protector and if they are not equal calls the
9031 ``__stack_chk_fail()`` function.
9036 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
9037 the variable ``@__stack_chk_guard``.
9042 This intrinsic is provided to perform the stack protector check by comparing
9043 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
9044 values do not match call the ``__stack_chk_fail()`` function.
9046 The reason to provide this as an IR level intrinsic instead of implementing it
9047 via other IR operations is that in order to perform this operation at the IR
9048 level without an intrinsic, one would need to create additional basic blocks to
9049 handle the success/failure cases. This makes it difficult to stop the stack
9050 protector check from disrupting sibling tail calls in Codegen. With this
9051 intrinsic, we are able to generate the stack protector basic blocks late in
9052 codegen after the tail call decision has occurred.
9054 '``llvm.objectsize``' Intrinsic
9055 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9062 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
9063 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
9068 The ``llvm.objectsize`` intrinsic is designed to provide information to
9069 the optimizers to determine at compile time whether a) an operation
9070 (like memcpy) will overflow a buffer that corresponds to an object, or
9071 b) that a runtime check for overflow isn't necessary. An object in this
9072 context means an allocation of a specific class, structure, array, or
9078 The ``llvm.objectsize`` intrinsic takes two arguments. The first
9079 argument is a pointer to or into the ``object``. The second argument is
9080 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
9081 or -1 (if false) when the object size is unknown. The second argument
9082 only accepts constants.
9087 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
9088 the size of the object concerned. If the size cannot be determined at
9089 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
9090 on the ``min`` argument).
9092 '``llvm.expect``' Intrinsic
9093 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9098 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
9103 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
9104 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9105 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9110 The ``llvm.expect`` intrinsic provides information about expected (the
9111 most probable) value of ``val``, which can be used by optimizers.
9116 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
9117 a value. The second argument is an expected value, this needs to be a
9118 constant value, variables are not allowed.
9123 This intrinsic is lowered to the ``val``.
9125 '``llvm.donothing``' Intrinsic
9126 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9133 declare void @llvm.donothing() nounwind readnone
9138 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
9139 only intrinsic that can be called with an invoke instruction.
9149 This intrinsic does nothing, and it's removed by optimizers and ignored
9152 Stack Map Intrinsics
9153 --------------------
9155 LLVM provides experimental intrinsics to support runtime patching
9156 mechanisms commonly desired in dynamic language JITs. These intrinsics
9157 are described in :doc:`StackMaps`.