1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][a-zA-Z$._][a-zA-Z$._0-9]*``'. Identifiers which require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves.
83 #. Unnamed values are represented as an unsigned numeric value with
84 their prefix. For example, ``%12``, ``@2``, ``%44``.
85 #. Constants, which are described in the section Constants_ below.
87 LLVM requires that values start with a prefix for two reasons: Compilers
88 don't need to worry about name clashes with reserved words, and the set
89 of reserved words may be expanded in the future without penalty.
90 Additionally, unnamed identifiers allow a compiler to quickly come up
91 with a temporary variable without having to avoid symbol table
94 Reserved words in LLVM are very similar to reserved words in other
95 languages. There are keywords for different opcodes ('``add``',
96 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
97 '``i32``', etc...), and others. These reserved words cannot conflict
98 with variable names, because none of them start with a prefix character
101 Here is an example of LLVM code to multiply the integer variable
108 %result = mul i32 %X, 8
110 After strength reduction:
114 %result = shl i32 %X, 3
120 %0 = add i32 %X, %X ; yields {i32}:%0
121 %1 = add i32 %0, %0 ; yields {i32}:%1
122 %result = add i32 %1, %1
124 This last way of multiplying ``%X`` by 8 illustrates several important
125 lexical features of LLVM:
127 #. Comments are delimited with a '``;``' and go until the end of line.
128 #. Unnamed temporaries are created when the result of a computation is
129 not assigned to a named value.
130 #. Unnamed temporaries are numbered sequentially (using a per-function
131 incrementing counter, starting with 0). Note that basic blocks are
132 included in this numbering. For example, if the entry basic block is not
133 given a label name, then it will get number 0.
135 It also shows a convention that we follow in this document. When
136 demonstrating instructions, we will follow an instruction with a comment
137 that defines the type and name of value produced.
145 LLVM programs are composed of ``Module``'s, each of which is a
146 translation unit of the input programs. Each module consists of
147 functions, global variables, and symbol table entries. Modules may be
148 combined together with the LLVM linker, which merges function (and
149 global variable) definitions, resolves forward declarations, and merges
150 symbol table entries. Here is an example of the "hello world" module:
154 ; Declare the string constant as a global constant.
155 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
157 ; External declaration of the puts function
158 declare i32 @puts(i8* nocapture) nounwind
160 ; Definition of main function
161 define i32 @main() { ; i32()*
162 ; Convert [13 x i8]* to i8 *...
163 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
165 ; Call puts function to write out the string to stdout.
166 call i32 @puts(i8* %cast210)
171 !1 = metadata !{i32 42}
174 This example is made up of a :ref:`global variable <globalvars>` named
175 "``.str``", an external declaration of the "``puts``" function, a
176 :ref:`function definition <functionstructure>` for "``main``" and
177 :ref:`named metadata <namedmetadatastructure>` "``foo``".
179 In general, a module is made up of a list of global values (where both
180 functions and global variables are global values). Global values are
181 represented by a pointer to a memory location (in this case, a pointer
182 to an array of char, and a pointer to a function), and have one of the
183 following :ref:`linkage types <linkage>`.
190 All Global Variables and Functions have one of the following types of
194 Global values with "``private``" linkage are only directly
195 accessible by objects in the current module. In particular, linking
196 code into a module with an private global value may cause the
197 private to be renamed as necessary to avoid collisions. Because the
198 symbol is private to the module, all references can be updated. This
199 doesn't show up in any symbol table in the object file.
201 Similar to ``private``, but the symbol is passed through the
202 assembler and evaluated by the linker. Unlike normal strong symbols,
203 they are removed by the linker from the final linked image
204 (executable or dynamic library).
205 ``linker_private_weak``
206 Similar to "``linker_private``", but the symbol is weak. Note that
207 ``linker_private_weak`` symbols are subject to coalescing by the
208 linker. The symbols are removed by the linker from the final linked
209 image (executable or dynamic library).
211 Similar to private, but the value shows as a local symbol
212 (``STB_LOCAL`` in the case of ELF) in the object file. This
213 corresponds to the notion of the '``static``' keyword in C.
214 ``available_externally``
215 Globals with "``available_externally``" linkage are never emitted
216 into the object file corresponding to the LLVM module. They exist to
217 allow inlining and other optimizations to take place given knowledge
218 of the definition of the global, which is known to be somewhere
219 outside the module. Globals with ``available_externally`` linkage
220 are allowed to be discarded at will, and are otherwise the same as
221 ``linkonce_odr``. This linkage type is only allowed on definitions,
224 Globals with "``linkonce``" linkage are merged with other globals of
225 the same name when linkage occurs. This can be used to implement
226 some forms of inline functions, templates, or other code which must
227 be generated in each translation unit that uses it, but where the
228 body may be overridden with a more definitive definition later.
229 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
230 that ``linkonce`` linkage does not actually allow the optimizer to
231 inline the body of this function into callers because it doesn't
232 know if this definition of the function is the definitive definition
233 within the program or whether it will be overridden by a stronger
234 definition. To enable inlining and other optimizations, use
235 "``linkonce_odr``" linkage.
237 "``weak``" linkage has the same merging semantics as ``linkonce``
238 linkage, except that unreferenced globals with ``weak`` linkage may
239 not be discarded. This is used for globals that are declared "weak"
242 "``common``" linkage is most similar to "``weak``" linkage, but they
243 are used for tentative definitions in C, such as "``int X;``" at
244 global scope. Symbols with "``common``" linkage are merged in the
245 same way as ``weak symbols``, and they may not be deleted if
246 unreferenced. ``common`` symbols may not have an explicit section,
247 must have a zero initializer, and may not be marked
248 ':ref:`constant <globalvars>`'. Functions and aliases may not have
251 .. _linkage_appending:
254 "``appending``" linkage may only be applied to global variables of
255 pointer to array type. When two global variables with appending
256 linkage are linked together, the two global arrays are appended
257 together. This is the LLVM, typesafe, equivalent of having the
258 system linker append together "sections" with identical names when
261 The semantics of this linkage follow the ELF object file model: the
262 symbol is weak until linked, if not linked, the symbol becomes null
263 instead of being an undefined reference.
264 ``linkonce_odr``, ``weak_odr``
265 Some languages allow differing globals to be merged, such as two
266 functions with different semantics. Other languages, such as
267 ``C++``, ensure that only equivalent globals are ever merged (the
268 "one definition rule" --- "ODR"). Such languages can use the
269 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
270 global will only be merged with equivalent globals. These linkage
271 types are otherwise the same as their non-``odr`` versions.
273 If none of the above identifiers are used, the global is externally
274 visible, meaning that it participates in linkage and can be used to
275 resolve external symbol references.
277 It is illegal for a function *declaration* to have any linkage type
278 other than ``external`` or ``extern_weak``.
285 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
286 :ref:`invokes <i_invoke>` can all have an optional calling convention
287 specified for the call. The calling convention of any pair of dynamic
288 caller/callee must match, or the behavior of the program is undefined.
289 The following calling conventions are supported by LLVM, and more may be
292 "``ccc``" - The C calling convention
293 This calling convention (the default if no other calling convention
294 is specified) matches the target C calling conventions. This calling
295 convention supports varargs function calls and tolerates some
296 mismatch in the declared prototype and implemented declaration of
297 the function (as does normal C).
298 "``fastcc``" - The fast calling convention
299 This calling convention attempts to make calls as fast as possible
300 (e.g. by passing things in registers). This calling convention
301 allows the target to use whatever tricks it wants to produce fast
302 code for the target, without having to conform to an externally
303 specified ABI (Application Binary Interface). `Tail calls can only
304 be optimized when this, the GHC or the HiPE convention is
305 used. <CodeGenerator.html#id80>`_ This calling convention does not
306 support varargs and requires the prototype of all callees to exactly
307 match the prototype of the function definition.
308 "``coldcc``" - The cold calling convention
309 This calling convention attempts to make code in the caller as
310 efficient as possible under the assumption that the call is not
311 commonly executed. As such, these calls often preserve all registers
312 so that the call does not break any live ranges in the caller side.
313 This calling convention does not support varargs and requires the
314 prototype of all callees to exactly match the prototype of the
315 function definition. Furthermore the inliner doesn't consider such function
317 "``cc 10``" - GHC convention
318 This calling convention has been implemented specifically for use by
319 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
320 It passes everything in registers, going to extremes to achieve this
321 by disabling callee save registers. This calling convention should
322 not be used lightly but only for specific situations such as an
323 alternative to the *register pinning* performance technique often
324 used when implementing functional programming languages. At the
325 moment only X86 supports this convention and it has the following
328 - On *X86-32* only supports up to 4 bit type parameters. No
329 floating point types are supported.
330 - On *X86-64* only supports up to 10 bit type parameters and 6
331 floating point parameters.
333 This calling convention supports `tail call
334 optimization <CodeGenerator.html#id80>`_ but requires both the
335 caller and callee are using it.
336 "``cc 11``" - The HiPE calling convention
337 This calling convention has been implemented specifically for use by
338 the `High-Performance Erlang
339 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
340 native code compiler of the `Ericsson's Open Source Erlang/OTP
341 system <http://www.erlang.org/download.shtml>`_. It uses more
342 registers for argument passing than the ordinary C calling
343 convention and defines no callee-saved registers. The calling
344 convention properly supports `tail call
345 optimization <CodeGenerator.html#id80>`_ but requires that both the
346 caller and the callee use it. It uses a *register pinning*
347 mechanism, similar to GHC's convention, for keeping frequently
348 accessed runtime components pinned to specific hardware registers.
349 At the moment only X86 supports this convention (both 32 and 64
351 "``webkit_jscc``" - WebKit's JavaScript calling convention
352 This calling convention has been implemented for `WebKit FTL JIT
353 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
354 stack right to left (as cdecl does), and returns a value in the
355 platform's customary return register.
356 "``anyregcc``" - Dynamic calling convention for code patching
357 This is a special convention that supports patching an arbitrary code
358 sequence in place of a call site. This convention forces the call
359 arguments into registers but allows them to be dynamcially
360 allocated. This can currently only be used with calls to
361 llvm.experimental.patchpoint because only this intrinsic records
362 the location of its arguments in a side table. See :doc:`StackMaps`.
363 "``preserve_mostcc``" - The `PreserveMost` calling convention
364 This calling convention attempts to make the code in the caller as little
365 intrusive as possible. This calling convention behaves identical to the `C`
366 calling convention on how arguments and return values are passed, but it
367 uses a different set of caller/callee-saved registers. This alleviates the
368 burden of saving and recovering a large register set before and after the
369 call in the caller. If the arguments are passed in callee-saved registers,
370 then they will be preserved by the callee across the call. This doesn't
371 apply for values returned in callee-saved registers.
373 - On X86-64 the callee preserves all general purpose registers, except for
374 R11. R11 can be used as a scratch register. Floating-point registers
375 (XMMs/YMMs) are not preserved and need to be saved by the caller.
377 The idea behind this convention is to support calls to runtime functions
378 that have a hot path and a cold path. The hot path is usually a small piece
379 of code that doesn't many registers. The cold path might need to call out to
380 another function and therefore only needs to preserve the caller-saved
381 registers, which haven't already been saved by the caller. The
382 `PreserveMost` calling convention is very similar to the `cold` calling
383 convention in terms of caller/callee-saved registers, but they are used for
384 different types of function calls. `coldcc` is for function calls that are
385 rarely executed, whereas `preserve_mostcc` function calls are intended to be
386 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
387 doesn't prevent the inliner from inlining the function call.
389 This calling convention will be used by a future version of the ObjectiveC
390 runtime and should therefore still be considered experimental at this time.
391 Although this convention was created to optimize certain runtime calls to
392 the ObjectiveC runtime, it is not limited to this runtime and might be used
393 by other runtimes in the future too. The current implementation only
394 supports X86-64, but the intention is to support more architectures in the
396 "``preserve_allcc``" - The `PreserveAll` calling convention
397 This calling convention attempts to make the code in the caller even less
398 intrusive than the `PreserveMost` calling convention. This calling
399 convention also behaves identical to the `C` calling convention on how
400 arguments and return values are passed, but it uses a different set of
401 caller/callee-saved registers. This removes the burden of saving and
402 recovering a large register set before and after the call in the caller. If
403 the arguments are passed in callee-saved registers, then they will be
404 preserved by the callee across the call. This doesn't apply for values
405 returned in callee-saved registers.
407 - On X86-64 the callee preserves all general purpose registers, except for
408 R11. R11 can be used as a scratch register. Furthermore it also preserves
409 all floating-point registers (XMMs/YMMs).
411 The idea behind this convention is to support calls to runtime functions
412 that don't need to call out to any other functions.
414 This calling convention, like the `PreserveMost` calling convention, will be
415 used by a future version of the ObjectiveC runtime and should be considered
416 experimental at this time.
417 "``cc <n>``" - Numbered convention
418 Any calling convention may be specified by number, allowing
419 target-specific calling conventions to be used. Target specific
420 calling conventions start at 64.
422 More calling conventions can be added/defined on an as-needed basis, to
423 support Pascal conventions or any other well-known target-independent
426 .. _visibilitystyles:
431 All Global Variables and Functions have one of the following visibility
434 "``default``" - Default style
435 On targets that use the ELF object file format, default visibility
436 means that the declaration is visible to other modules and, in
437 shared libraries, means that the declared entity may be overridden.
438 On Darwin, default visibility means that the declaration is visible
439 to other modules. Default visibility corresponds to "external
440 linkage" in the language.
441 "``hidden``" - Hidden style
442 Two declarations of an object with hidden visibility refer to the
443 same object if they are in the same shared object. Usually, hidden
444 visibility indicates that the symbol will not be placed into the
445 dynamic symbol table, so no other module (executable or shared
446 library) can reference it directly.
447 "``protected``" - Protected style
448 On ELF, protected visibility indicates that the symbol will be
449 placed in the dynamic symbol table, but that references within the
450 defining module will bind to the local symbol. That is, the symbol
451 cannot be overridden by another module.
458 All Global Variables, Functions and Aliases can have one of the following
462 "``dllimport``" causes the compiler to reference a function or variable via
463 a global pointer to a pointer that is set up by the DLL exporting the
464 symbol. On Microsoft Windows targets, the pointer name is formed by
465 combining ``__imp_`` and the function or variable name.
467 "``dllexport``" causes the compiler to provide a global pointer to a pointer
468 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
469 Microsoft Windows targets, the pointer name is formed by combining
470 ``__imp_`` and the function or variable name. Since this storage class
471 exists for defining a dll interface, the compiler, assembler and linker know
472 it is externally referenced and must refrain from deleting the symbol.
477 LLVM IR allows you to specify name aliases for certain types. This can
478 make it easier to read the IR and make the IR more condensed
479 (particularly when recursive types are involved). An example of a name
484 %mytype = type { %mytype*, i32 }
486 You may give a name to any :ref:`type <typesystem>` except
487 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
488 expected with the syntax "%mytype".
490 Note that type names are aliases for the structural type that they
491 indicate, and that you can therefore specify multiple names for the same
492 type. This often leads to confusing behavior when dumping out a .ll
493 file. Since LLVM IR uses structural typing, the name is not part of the
494 type. When printing out LLVM IR, the printer will pick *one name* to
495 render all types of a particular shape. This means that if you have code
496 where two different source types end up having the same LLVM type, that
497 the dumper will sometimes print the "wrong" or unexpected type. This is
498 an important design point and isn't going to change.
505 Global variables define regions of memory allocated at compilation time
508 Global variables definitions must be initialized, may have an explicit section
509 to be placed in, and may have an optional explicit alignment specified.
511 Global variables in other translation units can also be declared, in which
512 case they don't have an initializer.
514 A variable may be defined as ``thread_local``, which means that it will
515 not be shared by threads (each thread will have a separated copy of the
516 variable). Not all targets support thread-local variables. Optionally, a
517 TLS model may be specified:
520 For variables that are only used within the current shared library.
522 For variables in modules that will not be loaded dynamically.
524 For variables defined in the executable and only used within it.
526 The models correspond to the ELF TLS models; see `ELF Handling For
527 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
528 more information on under which circumstances the different models may
529 be used. The target may choose a different TLS model if the specified
530 model is not supported, or if a better choice of model can be made.
532 A variable may be defined as a global ``constant``, which indicates that
533 the contents of the variable will **never** be modified (enabling better
534 optimization, allowing the global data to be placed in the read-only
535 section of an executable, etc). Note that variables that need runtime
536 initialization cannot be marked ``constant`` as there is a store to the
539 LLVM explicitly allows *declarations* of global variables to be marked
540 constant, even if the final definition of the global is not. This
541 capability can be used to enable slightly better optimization of the
542 program, but requires the language definition to guarantee that
543 optimizations based on the 'constantness' are valid for the translation
544 units that do not include the definition.
546 As SSA values, global variables define pointer values that are in scope
547 (i.e. they dominate) all basic blocks in the program. Global variables
548 always define a pointer to their "content" type because they describe a
549 region of memory, and all memory objects in LLVM are accessed through
552 Global variables can be marked with ``unnamed_addr`` which indicates
553 that the address is not significant, only the content. Constants marked
554 like this can be merged with other constants if they have the same
555 initializer. Note that a constant with significant address *can* be
556 merged with a ``unnamed_addr`` constant, the result being a constant
557 whose address is significant.
559 A global variable may be declared to reside in a target-specific
560 numbered address space. For targets that support them, address spaces
561 may affect how optimizations are performed and/or what target
562 instructions are used to access the variable. The default address space
563 is zero. The address space qualifier must precede any other attributes.
565 LLVM allows an explicit section to be specified for globals. If the
566 target supports it, it will emit globals to the section specified.
568 By default, global initializers are optimized by assuming that global
569 variables defined within the module are not modified from their
570 initial values before the start of the global initializer. This is
571 true even for variables potentially accessible from outside the
572 module, including those with external linkage or appearing in
573 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
574 by marking the variable with ``externally_initialized``.
576 An explicit alignment may be specified for a global, which must be a
577 power of 2. If not present, or if the alignment is set to zero, the
578 alignment of the global is set by the target to whatever it feels
579 convenient. If an explicit alignment is specified, the global is forced
580 to have exactly that alignment. Targets and optimizers are not allowed
581 to over-align the global if the global has an assigned section. In this
582 case, the extra alignment could be observable: for example, code could
583 assume that the globals are densely packed in their section and try to
584 iterate over them as an array, alignment padding would break this
587 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
591 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
592 [AddrSpace] [unnamed_addr] [ExternallyInitialized]
593 <global | constant> <Type>
594 [, section "name"] [, align <Alignment>]
596 For example, the following defines a global in a numbered address space
597 with an initializer, section, and alignment:
601 @G = addrspace(5) constant float 1.0, section "foo", align 4
603 The following example just declares a global variable
607 @G = external global i32
609 The following example defines a thread-local global with the
610 ``initialexec`` TLS model:
614 @G = thread_local(initialexec) global i32 0, align 4
616 .. _functionstructure:
621 LLVM function definitions consist of the "``define``" keyword, an
622 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
623 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
624 an optional :ref:`calling convention <callingconv>`,
625 an optional ``unnamed_addr`` attribute, a return type, an optional
626 :ref:`parameter attribute <paramattrs>` for the return type, a function
627 name, a (possibly empty) argument list (each with optional :ref:`parameter
628 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
629 an optional section, an optional alignment, an optional :ref:`garbage
630 collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
631 curly brace, a list of basic blocks, and a closing curly brace.
633 LLVM function declarations consist of the "``declare``" keyword, an
634 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
635 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
636 an optional :ref:`calling convention <callingconv>`,
637 an optional ``unnamed_addr`` attribute, a return type, an optional
638 :ref:`parameter attribute <paramattrs>` for the return type, a function
639 name, a possibly empty list of arguments, an optional alignment, an optional
640 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
642 A function definition contains a list of basic blocks, forming the CFG (Control
643 Flow Graph) for the function. Each basic block may optionally start with a label
644 (giving the basic block a symbol table entry), contains a list of instructions,
645 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
646 function return). If an explicit label is not provided, a block is assigned an
647 implicit numbered label, using the next value from the same counter as used for
648 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
649 entry block does not have an explicit label, it will be assigned label "%0",
650 then the first unnamed temporary in that block will be "%1", etc.
652 The first basic block in a function is special in two ways: it is
653 immediately executed on entrance to the function, and it is not allowed
654 to have predecessor basic blocks (i.e. there can not be any branches to
655 the entry block of a function). Because the block can have no
656 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
658 LLVM allows an explicit section to be specified for functions. If the
659 target supports it, it will emit functions to the section specified.
661 An explicit alignment may be specified for a function. If not present,
662 or if the alignment is set to zero, the alignment of the function is set
663 by the target to whatever it feels convenient. If an explicit alignment
664 is specified, the function is forced to have at least that much
665 alignment. All alignments must be a power of 2.
667 If the ``unnamed_addr`` attribute is given, the address is know to not
668 be significant and two identical functions can be merged.
672 define [linkage] [visibility] [DLLStorageClass]
674 <ResultType> @<FunctionName> ([argument list])
675 [fn Attrs] [section "name"] [align N]
676 [gc] [prefix Constant] { ... }
683 Aliases act as "second name" for the aliasee value (which can be either
684 function, global variable, another alias or bitcast of global value).
685 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
686 :ref:`visibility style <visibility>`, and an optional :ref:`DLL storage class
691 @<Name> = [Visibility] [DLLStorageClass] alias [Linkage] <AliaseeTy> @<Aliasee>
693 The linkage must be one of ``private``, ``linker_private``,
694 ``linker_private_weak``, ``internal``, ``linkonce``, ``weak``,
695 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
696 might not correctly handle dropping a weak symbol that is aliased by a non-weak
699 .. _namedmetadatastructure:
704 Named metadata is a collection of metadata. :ref:`Metadata
705 nodes <metadata>` (but not metadata strings) are the only valid
706 operands for a named metadata.
710 ; Some unnamed metadata nodes, which are referenced by the named metadata.
711 !0 = metadata !{metadata !"zero"}
712 !1 = metadata !{metadata !"one"}
713 !2 = metadata !{metadata !"two"}
715 !name = !{!0, !1, !2}
722 The return type and each parameter of a function type may have a set of
723 *parameter attributes* associated with them. Parameter attributes are
724 used to communicate additional information about the result or
725 parameters of a function. Parameter attributes are considered to be part
726 of the function, not of the function type, so functions with different
727 parameter attributes can have the same function type.
729 Parameter attributes are simple keywords that follow the type specified.
730 If multiple parameter attributes are needed, they are space separated.
735 declare i32 @printf(i8* noalias nocapture, ...)
736 declare i32 @atoi(i8 zeroext)
737 declare signext i8 @returns_signed_char()
739 Note that any attributes for the function result (``nounwind``,
740 ``readonly``) come immediately after the argument list.
742 Currently, only the following parameter attributes are defined:
745 This indicates to the code generator that the parameter or return
746 value should be zero-extended to the extent required by the target's
747 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
748 the caller (for a parameter) or the callee (for a return value).
750 This indicates to the code generator that the parameter or return
751 value should be sign-extended to the extent required by the target's
752 ABI (which is usually 32-bits) by the caller (for a parameter) or
753 the callee (for a return value).
755 This indicates that this parameter or return value should be treated
756 in a special target-dependent fashion during while emitting code for
757 a function call or return (usually, by putting it in a register as
758 opposed to memory, though some targets use it to distinguish between
759 two different kinds of registers). Use of this attribute is
762 This indicates that the pointer parameter should really be passed by
763 value to the function. The attribute implies that a hidden copy of
764 the pointee is made between the caller and the callee, so the callee
765 is unable to modify the value in the caller. This attribute is only
766 valid on LLVM pointer arguments. It is generally used to pass
767 structs and arrays by value, but is also valid on pointers to
768 scalars. The copy is considered to belong to the caller not the
769 callee (for example, ``readonly`` functions should not write to
770 ``byval`` parameters). This is not a valid attribute for return
773 The byval attribute also supports specifying an alignment with the
774 align attribute. It indicates the alignment of the stack slot to
775 form and the known alignment of the pointer specified to the call
776 site. If the alignment is not specified, then the code generator
777 makes a target-specific assumption.
783 .. Warning:: This feature is unstable and not fully implemented.
785 The ``inalloca`` argument attribute allows the caller to take the
786 address of outgoing stack arguments. An ``inalloca`` argument must
787 be a pointer to stack memory produced by an ``alloca`` instruction.
788 The alloca, or argument allocation, must also be tagged with the
789 inalloca keyword. Only the past argument may have the ``inalloca``
790 attribute, and that argument is guaranteed to be passed in memory.
792 An argument allocation may be used by a call at most once because
793 the call may deallocate it. The ``inalloca`` attribute cannot be
794 used in conjunction with other attributes that affect argument
795 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
796 ``inalloca`` attribute also disables LLVM's implicit lowering of
797 large aggregate return values, which means that frontend authors
798 must lower them with ``sret`` pointers.
800 When the call site is reached, the argument allocation must have
801 been the most recent stack allocation that is still live, or the
802 results are undefined. It is possible to allocate additional stack
803 space after an argument allocation and before its call site, but it
804 must be cleared off with :ref:`llvm.stackrestore
807 See :doc:`InAlloca` for more information on how to use this
811 This indicates that the pointer parameter specifies the address of a
812 structure that is the return value of the function in the source
813 program. This pointer must be guaranteed by the caller to be valid:
814 loads and stores to the structure may be assumed by the callee
815 not to trap and to be properly aligned. This may only be applied to
816 the first parameter. This is not a valid attribute for return
819 This indicates that pointer values :ref:`based <pointeraliasing>` on
820 the argument or return value do not alias pointer values which are
821 not *based* on it, ignoring certain "irrelevant" dependencies. For a
822 call to the parent function, dependencies between memory references
823 from before or after the call and from those during the call are
824 "irrelevant" to the ``noalias`` keyword for the arguments and return
825 value used in that call. The caller shares the responsibility with
826 the callee for ensuring that these requirements are met. For further
827 details, please see the discussion of the NoAlias response in `alias
828 analysis <AliasAnalysis.html#MustMayNo>`_.
830 Note that this definition of ``noalias`` is intentionally similar
831 to the definition of ``restrict`` in C99 for function arguments,
832 though it is slightly weaker.
834 For function return values, C99's ``restrict`` is not meaningful,
835 while LLVM's ``noalias`` is.
837 This indicates that the callee does not make any copies of the
838 pointer that outlive the callee itself. This is not a valid
839 attribute for return values.
844 This indicates that the pointer parameter can be excised using the
845 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
846 attribute for return values and can only be applied to one parameter.
849 This indicates that the function always returns the argument as its return
850 value. This is an optimization hint to the code generator when generating
851 the caller, allowing tail call optimization and omission of register saves
852 and restores in some cases; it is not checked or enforced when generating
853 the callee. The parameter and the function return type must be valid
854 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
855 valid attribute for return values and can only be applied to one parameter.
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 If a function that has an ``ssp`` attribute is inlined into a
1116 function that doesn't have an ``ssp`` attribute, then the resulting
1117 function will have an ``ssp`` attribute.
1119 This attribute indicates that the function should *always* emit a
1120 stack smashing protector. This overrides the ``ssp`` function
1123 If a function that has an ``sspreq`` attribute is inlined into a
1124 function that doesn't have an ``sspreq`` attribute or which has an
1125 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1126 an ``sspreq`` attribute.
1128 This attribute indicates that the function should emit a stack smashing
1129 protector. This attribute causes a strong heuristic to be used when
1130 determining if a function needs stack protectors. The strong heuristic
1131 will enable protectors for functions with:
1133 - Arrays of any size and type
1134 - Aggregates containing an array of any size and type.
1135 - Calls to alloca().
1136 - Local variables that have had their address taken.
1138 This overrides the ``ssp`` function attribute.
1140 If a function that has an ``sspstrong`` attribute is inlined into a
1141 function that doesn't have an ``sspstrong`` attribute, then the
1142 resulting function will have an ``sspstrong`` attribute.
1144 This attribute indicates that the ABI being targeted requires that
1145 an unwind table entry be produce for this function even if we can
1146 show that no exceptions passes by it. This is normally the case for
1147 the ELF x86-64 abi, but it can be disabled for some compilation
1152 Module-Level Inline Assembly
1153 ----------------------------
1155 Modules may contain "module-level inline asm" blocks, which corresponds
1156 to the GCC "file scope inline asm" blocks. These blocks are internally
1157 concatenated by LLVM and treated as a single unit, but may be separated
1158 in the ``.ll`` file if desired. The syntax is very simple:
1160 .. code-block:: llvm
1162 module asm "inline asm code goes here"
1163 module asm "more can go here"
1165 The strings can contain any character by escaping non-printable
1166 characters. The escape sequence used is simply "\\xx" where "xx" is the
1167 two digit hex code for the number.
1169 The inline asm code is simply printed to the machine code .s file when
1170 assembly code is generated.
1172 .. _langref_datalayout:
1177 A module may specify a target specific data layout string that specifies
1178 how data is to be laid out in memory. The syntax for the data layout is
1181 .. code-block:: llvm
1183 target datalayout = "layout specification"
1185 The *layout specification* consists of a list of specifications
1186 separated by the minus sign character ('-'). Each specification starts
1187 with a letter and may include other information after the letter to
1188 define some aspect of the data layout. The specifications accepted are
1192 Specifies that the target lays out data in big-endian form. That is,
1193 the bits with the most significance have the lowest address
1196 Specifies that the target lays out data in little-endian form. That
1197 is, the bits with the least significance have the lowest address
1200 Specifies the natural alignment of the stack in bits. Alignment
1201 promotion of stack variables is limited to the natural stack
1202 alignment to avoid dynamic stack realignment. The stack alignment
1203 must be a multiple of 8-bits. If omitted, the natural stack
1204 alignment defaults to "unspecified", which does not prevent any
1205 alignment promotions.
1206 ``p[n]:<size>:<abi>:<pref>``
1207 This specifies the *size* of a pointer and its ``<abi>`` and
1208 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1209 bits. The address space, ``n`` is optional, and if not specified,
1210 denotes the default address space 0. The value of ``n`` must be
1211 in the range [1,2^23).
1212 ``i<size>:<abi>:<pref>``
1213 This specifies the alignment for an integer type of a given bit
1214 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1215 ``v<size>:<abi>:<pref>``
1216 This specifies the alignment for a vector type of a given bit
1218 ``f<size>:<abi>:<pref>``
1219 This specifies the alignment for a floating point type of a given bit
1220 ``<size>``. Only values of ``<size>`` that are supported by the target
1221 will work. 32 (float) and 64 (double) are supported on all targets; 80
1222 or 128 (different flavors of long double) are also supported on some
1225 This specifies the alignment for an object of aggregate type.
1227 If present, specifies that llvm names are mangled in the output. The
1230 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1231 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1232 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1233 symbols get a ``_`` prefix.
1234 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1235 functions also get a suffix based on the frame size.
1236 ``n<size1>:<size2>:<size3>...``
1237 This specifies a set of native integer widths for the target CPU in
1238 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1239 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1240 this set are considered to support most general arithmetic operations
1243 On every specification that takes a ``<abi>:<pref>``, specifying the
1244 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1245 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1247 When constructing the data layout for a given target, LLVM starts with a
1248 default set of specifications which are then (possibly) overridden by
1249 the specifications in the ``datalayout`` keyword. The default
1250 specifications are given in this list:
1252 - ``E`` - big endian
1253 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1254 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1255 same as the default address space.
1256 - ``S0`` - natural stack alignment is unspecified
1257 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1258 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1259 - ``i16:16:16`` - i16 is 16-bit aligned
1260 - ``i32:32:32`` - i32 is 32-bit aligned
1261 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1262 alignment of 64-bits
1263 - ``f16:16:16`` - half is 16-bit aligned
1264 - ``f32:32:32`` - float is 32-bit aligned
1265 - ``f64:64:64`` - double is 64-bit aligned
1266 - ``f128:128:128`` - quad is 128-bit aligned
1267 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1268 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1269 - ``a:0:64`` - aggregates are 64-bit aligned
1271 When LLVM is determining the alignment for a given type, it uses the
1274 #. If the type sought is an exact match for one of the specifications,
1275 that specification is used.
1276 #. If no match is found, and the type sought is an integer type, then
1277 the smallest integer type that is larger than the bitwidth of the
1278 sought type is used. If none of the specifications are larger than
1279 the bitwidth then the largest integer type is used. For example,
1280 given the default specifications above, the i7 type will use the
1281 alignment of i8 (next largest) while both i65 and i256 will use the
1282 alignment of i64 (largest specified).
1283 #. If no match is found, and the type sought is a vector type, then the
1284 largest vector type that is smaller than the sought vector type will
1285 be used as a fall back. This happens because <128 x double> can be
1286 implemented in terms of 64 <2 x double>, for example.
1288 The function of the data layout string may not be what you expect.
1289 Notably, this is not a specification from the frontend of what alignment
1290 the code generator should use.
1292 Instead, if specified, the target data layout is required to match what
1293 the ultimate *code generator* expects. This string is used by the
1294 mid-level optimizers to improve code, and this only works if it matches
1295 what the ultimate code generator uses. If you would like to generate IR
1296 that does not embed this target-specific detail into the IR, then you
1297 don't have to specify the string. This will disable some optimizations
1298 that require precise layout information, but this also prevents those
1299 optimizations from introducing target specificity into the IR.
1306 A module may specify a target triple string that describes the target
1307 host. The syntax for the target triple is simply:
1309 .. code-block:: llvm
1311 target triple = "x86_64-apple-macosx10.7.0"
1313 The *target triple* string consists of a series of identifiers delimited
1314 by the minus sign character ('-'). The canonical forms are:
1318 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1319 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1321 This information is passed along to the backend so that it generates
1322 code for the proper architecture. It's possible to override this on the
1323 command line with the ``-mtriple`` command line option.
1325 .. _pointeraliasing:
1327 Pointer Aliasing Rules
1328 ----------------------
1330 Any memory access must be done through a pointer value associated with
1331 an address range of the memory access, otherwise the behavior is
1332 undefined. Pointer values are associated with address ranges according
1333 to the following rules:
1335 - A pointer value is associated with the addresses associated with any
1336 value it is *based* on.
1337 - An address of a global variable is associated with the address range
1338 of the variable's storage.
1339 - The result value of an allocation instruction is associated with the
1340 address range of the allocated storage.
1341 - A null pointer in the default address-space is associated with no
1343 - An integer constant other than zero or a pointer value returned from
1344 a function not defined within LLVM may be associated with address
1345 ranges allocated through mechanisms other than those provided by
1346 LLVM. Such ranges shall not overlap with any ranges of addresses
1347 allocated by mechanisms provided by LLVM.
1349 A pointer value is *based* on another pointer value according to the
1352 - A pointer value formed from a ``getelementptr`` operation is *based*
1353 on the first operand of the ``getelementptr``.
1354 - The result value of a ``bitcast`` is *based* on the operand of the
1356 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1357 values that contribute (directly or indirectly) to the computation of
1358 the pointer's value.
1359 - The "*based* on" relationship is transitive.
1361 Note that this definition of *"based"* is intentionally similar to the
1362 definition of *"based"* in C99, though it is slightly weaker.
1364 LLVM IR does not associate types with memory. The result type of a
1365 ``load`` merely indicates the size and alignment of the memory from
1366 which to load, as well as the interpretation of the value. The first
1367 operand type of a ``store`` similarly only indicates the size and
1368 alignment of the store.
1370 Consequently, type-based alias analysis, aka TBAA, aka
1371 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1372 :ref:`Metadata <metadata>` may be used to encode additional information
1373 which specialized optimization passes may use to implement type-based
1378 Volatile Memory Accesses
1379 ------------------------
1381 Certain memory accesses, such as :ref:`load <i_load>`'s,
1382 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1383 marked ``volatile``. The optimizers must not change the number of
1384 volatile operations or change their order of execution relative to other
1385 volatile operations. The optimizers *may* change the order of volatile
1386 operations relative to non-volatile operations. This is not Java's
1387 "volatile" and has no cross-thread synchronization behavior.
1389 IR-level volatile loads and stores cannot safely be optimized into
1390 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1391 flagged volatile. Likewise, the backend should never split or merge
1392 target-legal volatile load/store instructions.
1394 .. admonition:: Rationale
1396 Platforms may rely on volatile loads and stores of natively supported
1397 data width to be executed as single instruction. For example, in C
1398 this holds for an l-value of volatile primitive type with native
1399 hardware support, but not necessarily for aggregate types. The
1400 frontend upholds these expectations, which are intentionally
1401 unspecified in the IR. The rules above ensure that IR transformation
1402 do not violate the frontend's contract with the language.
1406 Memory Model for Concurrent Operations
1407 --------------------------------------
1409 The LLVM IR does not define any way to start parallel threads of
1410 execution or to register signal handlers. Nonetheless, there are
1411 platform-specific ways to create them, and we define LLVM IR's behavior
1412 in their presence. This model is inspired by the C++0x memory model.
1414 For a more informal introduction to this model, see the :doc:`Atomics`.
1416 We define a *happens-before* partial order as the least partial order
1419 - Is a superset of single-thread program order, and
1420 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1421 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1422 techniques, like pthread locks, thread creation, thread joining,
1423 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1424 Constraints <ordering>`).
1426 Note that program order does not introduce *happens-before* edges
1427 between a thread and signals executing inside that thread.
1429 Every (defined) read operation (load instructions, memcpy, atomic
1430 loads/read-modify-writes, etc.) R reads a series of bytes written by
1431 (defined) write operations (store instructions, atomic
1432 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1433 section, initialized globals are considered to have a write of the
1434 initializer which is atomic and happens before any other read or write
1435 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1436 may see any write to the same byte, except:
1438 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1439 write\ :sub:`2` happens before R\ :sub:`byte`, then
1440 R\ :sub:`byte` does not see write\ :sub:`1`.
1441 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1442 R\ :sub:`byte` does not see write\ :sub:`3`.
1444 Given that definition, R\ :sub:`byte` is defined as follows:
1446 - If R is volatile, the result is target-dependent. (Volatile is
1447 supposed to give guarantees which can support ``sig_atomic_t`` in
1448 C/C++, and may be used for accesses to addresses which do not behave
1449 like normal memory. It does not generally provide cross-thread
1451 - Otherwise, if there is no write to the same byte that happens before
1452 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1453 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1454 R\ :sub:`byte` returns the value written by that write.
1455 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1456 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1457 Memory Ordering Constraints <ordering>` section for additional
1458 constraints on how the choice is made.
1459 - Otherwise R\ :sub:`byte` returns ``undef``.
1461 R returns the value composed of the series of bytes it read. This
1462 implies that some bytes within the value may be ``undef`` **without**
1463 the entire value being ``undef``. Note that this only defines the
1464 semantics of the operation; it doesn't mean that targets will emit more
1465 than one instruction to read the series of bytes.
1467 Note that in cases where none of the atomic intrinsics are used, this
1468 model places only one restriction on IR transformations on top of what
1469 is required for single-threaded execution: introducing a store to a byte
1470 which might not otherwise be stored is not allowed in general.
1471 (Specifically, in the case where another thread might write to and read
1472 from an address, introducing a store can change a load that may see
1473 exactly one write into a load that may see multiple writes.)
1477 Atomic Memory Ordering Constraints
1478 ----------------------------------
1480 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1481 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1482 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1483 an ordering parameter that determines which other atomic instructions on
1484 the same address they *synchronize with*. These semantics are borrowed
1485 from Java and C++0x, but are somewhat more colloquial. If these
1486 descriptions aren't precise enough, check those specs (see spec
1487 references in the :doc:`atomics guide <Atomics>`).
1488 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1489 differently since they don't take an address. See that instruction's
1490 documentation for details.
1492 For a simpler introduction to the ordering constraints, see the
1496 The set of values that can be read is governed by the happens-before
1497 partial order. A value cannot be read unless some operation wrote
1498 it. This is intended to provide a guarantee strong enough to model
1499 Java's non-volatile shared variables. This ordering cannot be
1500 specified for read-modify-write operations; it is not strong enough
1501 to make them atomic in any interesting way.
1503 In addition to the guarantees of ``unordered``, there is a single
1504 total order for modifications by ``monotonic`` operations on each
1505 address. All modification orders must be compatible with the
1506 happens-before order. There is no guarantee that the modification
1507 orders can be combined to a global total order for the whole program
1508 (and this often will not be possible). The read in an atomic
1509 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1510 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1511 order immediately before the value it writes. If one atomic read
1512 happens before another atomic read of the same address, the later
1513 read must see the same value or a later value in the address's
1514 modification order. This disallows reordering of ``monotonic`` (or
1515 stronger) operations on the same address. If an address is written
1516 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1517 read that address repeatedly, the other threads must eventually see
1518 the write. This corresponds to the C++0x/C1x
1519 ``memory_order_relaxed``.
1521 In addition to the guarantees of ``monotonic``, a
1522 *synchronizes-with* edge may be formed with a ``release`` operation.
1523 This is intended to model C++'s ``memory_order_acquire``.
1525 In addition to the guarantees of ``monotonic``, if this operation
1526 writes a value which is subsequently read by an ``acquire``
1527 operation, it *synchronizes-with* that operation. (This isn't a
1528 complete description; see the C++0x definition of a release
1529 sequence.) This corresponds to the C++0x/C1x
1530 ``memory_order_release``.
1531 ``acq_rel`` (acquire+release)
1532 Acts as both an ``acquire`` and ``release`` operation on its
1533 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1534 ``seq_cst`` (sequentially consistent)
1535 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1536 operation which only reads, ``release`` for an operation which only
1537 writes), there is a global total order on all
1538 sequentially-consistent operations on all addresses, which is
1539 consistent with the *happens-before* partial order and with the
1540 modification orders of all the affected addresses. Each
1541 sequentially-consistent read sees the last preceding write to the
1542 same address in this global order. This corresponds to the C++0x/C1x
1543 ``memory_order_seq_cst`` and Java volatile.
1547 If an atomic operation is marked ``singlethread``, it only *synchronizes
1548 with* or participates in modification and seq\_cst total orderings with
1549 other operations running in the same thread (for example, in signal
1557 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1558 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1559 :ref:`frem <i_frem>`) have the following flags that can set to enable
1560 otherwise unsafe floating point operations
1563 No NaNs - Allow optimizations to assume the arguments and result are not
1564 NaN. Such optimizations are required to retain defined behavior over
1565 NaNs, but the value of the result is undefined.
1568 No Infs - Allow optimizations to assume the arguments and result are not
1569 +/-Inf. Such optimizations are required to retain defined behavior over
1570 +/-Inf, but the value of the result is undefined.
1573 No Signed Zeros - Allow optimizations to treat the sign of a zero
1574 argument or result as insignificant.
1577 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1578 argument rather than perform division.
1581 Fast - Allow algebraically equivalent transformations that may
1582 dramatically change results in floating point (e.g. reassociate). This
1583 flag implies all the others.
1590 The LLVM type system is one of the most important features of the
1591 intermediate representation. Being typed enables a number of
1592 optimizations to be performed on the intermediate representation
1593 directly, without having to do extra analyses on the side before the
1594 transformation. A strong type system makes it easier to read the
1595 generated code and enables novel analyses and transformations that are
1596 not feasible to perform on normal three address code representations.
1606 The void type does not represent any value and has no size.
1624 The function type can be thought of as a function signature. It consists of a
1625 return type and a list of formal parameter types. The return type of a function
1626 type is a void type or first class type --- except for :ref:`label <t_label>`
1627 and :ref:`metadata <t_metadata>` types.
1633 <returntype> (<parameter list>)
1635 ...where '``<parameter list>``' is a comma-separated list of type
1636 specifiers. Optionally, the parameter list may include a type ``...``, which
1637 indicates that the function takes a variable number of arguments. Variable
1638 argument functions can access their arguments with the :ref:`variable argument
1639 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1640 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1644 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1645 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1646 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1647 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1648 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1649 | ``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. |
1650 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1651 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1652 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1659 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1660 Values of these types are the only ones which can be produced by
1668 These are the types that are valid in registers from CodeGen's perspective.
1677 The integer type is a very simple type that simply specifies an
1678 arbitrary bit width for the integer type desired. Any bit width from 1
1679 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1687 The number of bits the integer will occupy is specified by the ``N``
1693 +----------------+------------------------------------------------+
1694 | ``i1`` | a single-bit integer. |
1695 +----------------+------------------------------------------------+
1696 | ``i32`` | a 32-bit integer. |
1697 +----------------+------------------------------------------------+
1698 | ``i1942652`` | a really big integer of over 1 million bits. |
1699 +----------------+------------------------------------------------+
1703 Floating Point Types
1704 """"""""""""""""""""
1713 - 16-bit floating point value
1716 - 32-bit floating point value
1719 - 64-bit floating point value
1722 - 128-bit floating point value (112-bit mantissa)
1725 - 80-bit floating point value (X87)
1728 - 128-bit floating point value (two 64-bits)
1737 The x86mmx type represents a value held in an MMX register on an x86
1738 machine. The operations allowed on it are quite limited: parameters and
1739 return values, load and store, and bitcast. User-specified MMX
1740 instructions are represented as intrinsic or asm calls with arguments
1741 and/or results of this type. There are no arrays, vectors or constants
1758 The pointer type is used to specify memory locations. Pointers are
1759 commonly used to reference objects in memory.
1761 Pointer types may have an optional address space attribute defining the
1762 numbered address space where the pointed-to object resides. The default
1763 address space is number zero. The semantics of non-zero address spaces
1764 are target-specific.
1766 Note that LLVM does not permit pointers to void (``void*``) nor does it
1767 permit pointers to labels (``label*``). Use ``i8*`` instead.
1777 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1778 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1779 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1780 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1781 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1782 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1783 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1792 A vector type is a simple derived type that represents a vector of
1793 elements. Vector types are used when multiple primitive data are
1794 operated in parallel using a single instruction (SIMD). A vector type
1795 requires a size (number of elements) and an underlying primitive data
1796 type. Vector types are considered :ref:`first class <t_firstclass>`.
1802 < <# elements> x <elementtype> >
1804 The number of elements is a constant integer value larger than 0;
1805 elementtype may be any integer or floating point type, or a pointer to
1806 these types. Vectors of size zero are not allowed.
1810 +-------------------+--------------------------------------------------+
1811 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1812 +-------------------+--------------------------------------------------+
1813 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1814 +-------------------+--------------------------------------------------+
1815 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1816 +-------------------+--------------------------------------------------+
1817 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1818 +-------------------+--------------------------------------------------+
1827 The label type represents code labels.
1842 The metadata type represents embedded metadata. No derived types may be
1843 created from metadata except for :ref:`function <t_function>` arguments.
1856 Aggregate Types are a subset of derived types that can contain multiple
1857 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1858 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1868 The array type is a very simple derived type that arranges elements
1869 sequentially in memory. The array type requires a size (number of
1870 elements) and an underlying data type.
1876 [<# elements> x <elementtype>]
1878 The number of elements is a constant integer value; ``elementtype`` may
1879 be any type with a size.
1883 +------------------+--------------------------------------+
1884 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1885 +------------------+--------------------------------------+
1886 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1887 +------------------+--------------------------------------+
1888 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1889 +------------------+--------------------------------------+
1891 Here are some examples of multidimensional arrays:
1893 +-----------------------------+----------------------------------------------------------+
1894 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1895 +-----------------------------+----------------------------------------------------------+
1896 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1897 +-----------------------------+----------------------------------------------------------+
1898 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1899 +-----------------------------+----------------------------------------------------------+
1901 There is no restriction on indexing beyond the end of the array implied
1902 by a static type (though there are restrictions on indexing beyond the
1903 bounds of an allocated object in some cases). This means that
1904 single-dimension 'variable sized array' addressing can be implemented in
1905 LLVM with a zero length array type. An implementation of 'pascal style
1906 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1916 The structure type is used to represent a collection of data members
1917 together in memory. The elements of a structure may be any type that has
1920 Structures in memory are accessed using '``load``' and '``store``' by
1921 getting a pointer to a field with the '``getelementptr``' instruction.
1922 Structures in registers are accessed using the '``extractvalue``' and
1923 '``insertvalue``' instructions.
1925 Structures may optionally be "packed" structures, which indicate that
1926 the alignment of the struct is one byte, and that there is no padding
1927 between the elements. In non-packed structs, padding between field types
1928 is inserted as defined by the DataLayout string in the module, which is
1929 required to match what the underlying code generator expects.
1931 Structures can either be "literal" or "identified". A literal structure
1932 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1933 identified types are always defined at the top level with a name.
1934 Literal types are uniqued by their contents and can never be recursive
1935 or opaque since there is no way to write one. Identified types can be
1936 recursive, can be opaqued, and are never uniqued.
1942 %T1 = type { <type list> } ; Identified normal struct type
1943 %T2 = type <{ <type list> }> ; Identified packed struct type
1947 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1948 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1949 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1950 | ``{ 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``. |
1951 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1952 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1953 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1957 Opaque Structure Types
1958 """"""""""""""""""""""
1962 Opaque structure types are used to represent named structure types that
1963 do not have a body specified. This corresponds (for example) to the C
1964 notion of a forward declared structure.
1975 +--------------+-------------------+
1976 | ``opaque`` | An opaque type. |
1977 +--------------+-------------------+
1982 LLVM has several different basic types of constants. This section
1983 describes them all and their syntax.
1988 **Boolean constants**
1989 The two strings '``true``' and '``false``' are both valid constants
1991 **Integer constants**
1992 Standard integers (such as '4') are constants of the
1993 :ref:`integer <t_integer>` type. Negative numbers may be used with
1995 **Floating point constants**
1996 Floating point constants use standard decimal notation (e.g.
1997 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1998 hexadecimal notation (see below). The assembler requires the exact
1999 decimal value of a floating-point constant. For example, the
2000 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2001 decimal in binary. Floating point constants must have a :ref:`floating
2002 point <t_floating>` type.
2003 **Null pointer constants**
2004 The identifier '``null``' is recognized as a null pointer constant
2005 and must be of :ref:`pointer type <t_pointer>`.
2007 The one non-intuitive notation for constants is the hexadecimal form of
2008 floating point constants. For example, the form
2009 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2010 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2011 constants are required (and the only time that they are generated by the
2012 disassembler) is when a floating point constant must be emitted but it
2013 cannot be represented as a decimal floating point number in a reasonable
2014 number of digits. For example, NaN's, infinities, and other special
2015 values are represented in their IEEE hexadecimal format so that assembly
2016 and disassembly do not cause any bits to change in the constants.
2018 When using the hexadecimal form, constants of types half, float, and
2019 double are represented using the 16-digit form shown above (which
2020 matches the IEEE754 representation for double); half and float values
2021 must, however, be exactly representable as IEEE 754 half and single
2022 precision, respectively. Hexadecimal format is always used for long
2023 double, and there are three forms of long double. The 80-bit format used
2024 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2025 128-bit format used by PowerPC (two adjacent doubles) is represented by
2026 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2027 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2028 will only work if they match the long double format on your target.
2029 The IEEE 16-bit format (half precision) is represented by ``0xH``
2030 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2031 (sign bit at the left).
2033 There are no constants of type x86mmx.
2035 .. _complexconstants:
2040 Complex constants are a (potentially recursive) combination of simple
2041 constants and smaller complex constants.
2043 **Structure constants**
2044 Structure constants are represented with notation similar to
2045 structure type definitions (a comma separated list of elements,
2046 surrounded by braces (``{}``)). For example:
2047 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2048 "``@G = external global i32``". Structure constants must have
2049 :ref:`structure type <t_struct>`, and the number and types of elements
2050 must match those specified by the type.
2052 Array constants are represented with notation similar to array type
2053 definitions (a comma separated list of elements, surrounded by
2054 square brackets (``[]``)). For example:
2055 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2056 :ref:`array type <t_array>`, and the number and types of elements must
2057 match those specified by the type.
2058 **Vector constants**
2059 Vector constants are represented with notation similar to vector
2060 type definitions (a comma separated list of elements, surrounded by
2061 less-than/greater-than's (``<>``)). For example:
2062 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2063 must have :ref:`vector type <t_vector>`, and the number and types of
2064 elements must match those specified by the type.
2065 **Zero initialization**
2066 The string '``zeroinitializer``' can be used to zero initialize a
2067 value to zero of *any* type, including scalar and
2068 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2069 having to print large zero initializers (e.g. for large arrays) and
2070 is always exactly equivalent to using explicit zero initializers.
2072 A metadata node is a structure-like constant with :ref:`metadata
2073 type <t_metadata>`. For example:
2074 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2075 constants that are meant to be interpreted as part of the
2076 instruction stream, metadata is a place to attach additional
2077 information such as debug info.
2079 Global Variable and Function Addresses
2080 --------------------------------------
2082 The addresses of :ref:`global variables <globalvars>` and
2083 :ref:`functions <functionstructure>` are always implicitly valid
2084 (link-time) constants. These constants are explicitly referenced when
2085 the :ref:`identifier for the global <identifiers>` is used and always have
2086 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2089 .. code-block:: llvm
2093 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2100 The string '``undef``' can be used anywhere a constant is expected, and
2101 indicates that the user of the value may receive an unspecified
2102 bit-pattern. Undefined values may be of any type (other than '``label``'
2103 or '``void``') and be used anywhere a constant is permitted.
2105 Undefined values are useful because they indicate to the compiler that
2106 the program is well defined no matter what value is used. This gives the
2107 compiler more freedom to optimize. Here are some examples of
2108 (potentially surprising) transformations that are valid (in pseudo IR):
2110 .. code-block:: llvm
2120 This is safe because all of the output bits are affected by the undef
2121 bits. Any output bit can have a zero or one depending on the input bits.
2123 .. code-block:: llvm
2134 These logical operations have bits that are not always affected by the
2135 input. For example, if ``%X`` has a zero bit, then the output of the
2136 '``and``' operation will always be a zero for that bit, no matter what
2137 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2138 optimize or assume that the result of the '``and``' is '``undef``'.
2139 However, it is safe to assume that all bits of the '``undef``' could be
2140 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2141 all the bits of the '``undef``' operand to the '``or``' could be set,
2142 allowing the '``or``' to be folded to -1.
2144 .. code-block:: llvm
2146 %A = select undef, %X, %Y
2147 %B = select undef, 42, %Y
2148 %C = select %X, %Y, undef
2158 This set of examples shows that undefined '``select``' (and conditional
2159 branch) conditions can go *either way*, but they have to come from one
2160 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2161 both known to have a clear low bit, then ``%A`` would have to have a
2162 cleared low bit. However, in the ``%C`` example, the optimizer is
2163 allowed to assume that the '``undef``' operand could be the same as
2164 ``%Y``, allowing the whole '``select``' to be eliminated.
2166 .. code-block:: llvm
2168 %A = xor undef, undef
2185 This example points out that two '``undef``' operands are not
2186 necessarily the same. This can be surprising to people (and also matches
2187 C semantics) where they assume that "``X^X``" is always zero, even if
2188 ``X`` is undefined. This isn't true for a number of reasons, but the
2189 short answer is that an '``undef``' "variable" can arbitrarily change
2190 its value over its "live range". This is true because the variable
2191 doesn't actually *have a live range*. Instead, the value is logically
2192 read from arbitrary registers that happen to be around when needed, so
2193 the value is not necessarily consistent over time. In fact, ``%A`` and
2194 ``%C`` need to have the same semantics or the core LLVM "replace all
2195 uses with" concept would not hold.
2197 .. code-block:: llvm
2205 These examples show the crucial difference between an *undefined value*
2206 and *undefined behavior*. An undefined value (like '``undef``') is
2207 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2208 operation can be constant folded to '``undef``', because the '``undef``'
2209 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2210 However, in the second example, we can make a more aggressive
2211 assumption: because the ``undef`` is allowed to be an arbitrary value,
2212 we are allowed to assume that it could be zero. Since a divide by zero
2213 has *undefined behavior*, we are allowed to assume that the operation
2214 does not execute at all. This allows us to delete the divide and all
2215 code after it. Because the undefined operation "can't happen", the
2216 optimizer can assume that it occurs in dead code.
2218 .. code-block:: llvm
2220 a: store undef -> %X
2221 b: store %X -> undef
2226 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2227 value can be assumed to not have any effect; we can assume that the
2228 value is overwritten with bits that happen to match what was already
2229 there. However, a store *to* an undefined location could clobber
2230 arbitrary memory, therefore, it has undefined behavior.
2237 Poison values are similar to :ref:`undef values <undefvalues>`, however
2238 they also represent the fact that an instruction or constant expression
2239 which cannot evoke side effects has nevertheless detected a condition
2240 which results in undefined behavior.
2242 There is currently no way of representing a poison value in the IR; they
2243 only exist when produced by operations such as :ref:`add <i_add>` with
2246 Poison value behavior is defined in terms of value *dependence*:
2248 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2249 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2250 their dynamic predecessor basic block.
2251 - Function arguments depend on the corresponding actual argument values
2252 in the dynamic callers of their functions.
2253 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2254 instructions that dynamically transfer control back to them.
2255 - :ref:`Invoke <i_invoke>` instructions depend on the
2256 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2257 call instructions that dynamically transfer control back to them.
2258 - Non-volatile loads and stores depend on the most recent stores to all
2259 of the referenced memory addresses, following the order in the IR
2260 (including loads and stores implied by intrinsics such as
2261 :ref:`@llvm.memcpy <int_memcpy>`.)
2262 - An instruction with externally visible side effects depends on the
2263 most recent preceding instruction with externally visible side
2264 effects, following the order in the IR. (This includes :ref:`volatile
2265 operations <volatile>`.)
2266 - An instruction *control-depends* on a :ref:`terminator
2267 instruction <terminators>` if the terminator instruction has
2268 multiple successors and the instruction is always executed when
2269 control transfers to one of the successors, and may not be executed
2270 when control is transferred to another.
2271 - Additionally, an instruction also *control-depends* on a terminator
2272 instruction if the set of instructions it otherwise depends on would
2273 be different if the terminator had transferred control to a different
2275 - Dependence is transitive.
2277 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2278 with the additional affect that any instruction which has a *dependence*
2279 on a poison value has undefined behavior.
2281 Here are some examples:
2283 .. code-block:: llvm
2286 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2287 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2288 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2289 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2291 store i32 %poison, i32* @g ; Poison value stored to memory.
2292 %poison2 = load i32* @g ; Poison value loaded back from memory.
2294 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2296 %narrowaddr = bitcast i32* @g to i16*
2297 %wideaddr = bitcast i32* @g to i64*
2298 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2299 %poison4 = load i64* %wideaddr ; Returns a poison value.
2301 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2302 br i1 %cmp, label %true, label %end ; Branch to either destination.
2305 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2306 ; it has undefined behavior.
2310 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2311 ; Both edges into this PHI are
2312 ; control-dependent on %cmp, so this
2313 ; always results in a poison value.
2315 store volatile i32 0, i32* @g ; This would depend on the store in %true
2316 ; if %cmp is true, or the store in %entry
2317 ; otherwise, so this is undefined behavior.
2319 br i1 %cmp, label %second_true, label %second_end
2320 ; The same branch again, but this time the
2321 ; true block doesn't have side effects.
2328 store volatile i32 0, i32* @g ; This time, the instruction always depends
2329 ; on the store in %end. Also, it is
2330 ; control-equivalent to %end, so this is
2331 ; well-defined (ignoring earlier undefined
2332 ; behavior in this example).
2336 Addresses of Basic Blocks
2337 -------------------------
2339 ``blockaddress(@function, %block)``
2341 The '``blockaddress``' constant computes the address of the specified
2342 basic block in the specified function, and always has an ``i8*`` type.
2343 Taking the address of the entry block is illegal.
2345 This value only has defined behavior when used as an operand to the
2346 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2347 against null. Pointer equality tests between labels addresses results in
2348 undefined behavior --- though, again, comparison against null is ok, and
2349 no label is equal to the null pointer. This may be passed around as an
2350 opaque pointer sized value as long as the bits are not inspected. This
2351 allows ``ptrtoint`` and arithmetic to be performed on these values so
2352 long as the original value is reconstituted before the ``indirectbr``
2355 Finally, some targets may provide defined semantics when using the value
2356 as the operand to an inline assembly, but that is target specific.
2360 Constant Expressions
2361 --------------------
2363 Constant expressions are used to allow expressions involving other
2364 constants to be used as constants. Constant expressions may be of any
2365 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2366 that does not have side effects (e.g. load and call are not supported).
2367 The following is the syntax for constant expressions:
2369 ``trunc (CST to TYPE)``
2370 Truncate a constant to another type. The bit size of CST must be
2371 larger than the bit size of TYPE. Both types must be integers.
2372 ``zext (CST to TYPE)``
2373 Zero extend a constant to another type. The bit size of CST must be
2374 smaller than the bit size of TYPE. Both types must be integers.
2375 ``sext (CST to TYPE)``
2376 Sign extend a constant to another type. The bit size of CST must be
2377 smaller than the bit size of TYPE. Both types must be integers.
2378 ``fptrunc (CST to TYPE)``
2379 Truncate a floating point constant to another floating point type.
2380 The size of CST must be larger than the size of TYPE. Both types
2381 must be floating point.
2382 ``fpext (CST to TYPE)``
2383 Floating point extend a constant to another type. The size of CST
2384 must be smaller or equal to the size of TYPE. Both types must be
2386 ``fptoui (CST to TYPE)``
2387 Convert a floating point constant to the corresponding unsigned
2388 integer constant. TYPE must be a scalar or vector integer type. CST
2389 must be of scalar or vector floating point type. Both CST and TYPE
2390 must be scalars, or vectors of the same number of elements. If the
2391 value won't fit in the integer type, the results are undefined.
2392 ``fptosi (CST to TYPE)``
2393 Convert a floating point constant to the corresponding signed
2394 integer constant. TYPE must be a scalar or vector integer type. CST
2395 must be of scalar or vector floating point type. Both CST and TYPE
2396 must be scalars, or vectors of the same number of elements. If the
2397 value won't fit in the integer type, the results are undefined.
2398 ``uitofp (CST to TYPE)``
2399 Convert an unsigned integer constant to the corresponding floating
2400 point constant. TYPE must be a scalar or vector floating point type.
2401 CST must be of scalar or vector integer type. Both CST and TYPE must
2402 be scalars, or vectors of the same number of elements. If the value
2403 won't fit in the floating point type, the results are undefined.
2404 ``sitofp (CST to TYPE)``
2405 Convert a signed integer constant to the corresponding floating
2406 point constant. TYPE must be a scalar or vector floating point type.
2407 CST must be of scalar or vector integer type. Both CST and TYPE must
2408 be scalars, or vectors of the same number of elements. If the value
2409 won't fit in the floating point type, the results are undefined.
2410 ``ptrtoint (CST to TYPE)``
2411 Convert a pointer typed constant to the corresponding integer
2412 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2413 pointer type. The ``CST`` value is zero extended, truncated, or
2414 unchanged to make it fit in ``TYPE``.
2415 ``inttoptr (CST to TYPE)``
2416 Convert an integer constant to a pointer constant. TYPE must be a
2417 pointer type. CST must be of integer type. The CST value is zero
2418 extended, truncated, or unchanged to make it fit in a pointer size.
2419 This one is *really* dangerous!
2420 ``bitcast (CST to TYPE)``
2421 Convert a constant, CST, to another TYPE. The constraints of the
2422 operands are the same as those for the :ref:`bitcast
2423 instruction <i_bitcast>`.
2424 ``addrspacecast (CST to TYPE)``
2425 Convert a constant pointer or constant vector of pointer, CST, to another
2426 TYPE in a different address space. The constraints of the operands are the
2427 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2428 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2429 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2430 constants. As with the :ref:`getelementptr <i_getelementptr>`
2431 instruction, the index list may have zero or more indexes, which are
2432 required to make sense for the type of "CSTPTR".
2433 ``select (COND, VAL1, VAL2)``
2434 Perform the :ref:`select operation <i_select>` on constants.
2435 ``icmp COND (VAL1, VAL2)``
2436 Performs the :ref:`icmp operation <i_icmp>` on constants.
2437 ``fcmp COND (VAL1, VAL2)``
2438 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2439 ``extractelement (VAL, IDX)``
2440 Perform the :ref:`extractelement operation <i_extractelement>` on
2442 ``insertelement (VAL, ELT, IDX)``
2443 Perform the :ref:`insertelement operation <i_insertelement>` on
2445 ``shufflevector (VEC1, VEC2, IDXMASK)``
2446 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2448 ``extractvalue (VAL, IDX0, IDX1, ...)``
2449 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2450 constants. The index list is interpreted in a similar manner as
2451 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2452 least one index value must be specified.
2453 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2454 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2455 The index list is interpreted in a similar manner as indices in a
2456 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2457 value must be specified.
2458 ``OPCODE (LHS, RHS)``
2459 Perform the specified operation of the LHS and RHS constants. OPCODE
2460 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2461 binary <bitwiseops>` operations. The constraints on operands are
2462 the same as those for the corresponding instruction (e.g. no bitwise
2463 operations on floating point values are allowed).
2470 Inline Assembler Expressions
2471 ----------------------------
2473 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2474 Inline Assembly <moduleasm>`) through the use of a special value. This
2475 value represents the inline assembler as a string (containing the
2476 instructions to emit), a list of operand constraints (stored as a
2477 string), a flag that indicates whether or not the inline asm expression
2478 has side effects, and a flag indicating whether the function containing
2479 the asm needs to align its stack conservatively. An example inline
2480 assembler expression is:
2482 .. code-block:: llvm
2484 i32 (i32) asm "bswap $0", "=r,r"
2486 Inline assembler expressions may **only** be used as the callee operand
2487 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2488 Thus, typically we have:
2490 .. code-block:: llvm
2492 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2494 Inline asms with side effects not visible in the constraint list must be
2495 marked as having side effects. This is done through the use of the
2496 '``sideeffect``' keyword, like so:
2498 .. code-block:: llvm
2500 call void asm sideeffect "eieio", ""()
2502 In some cases inline asms will contain code that will not work unless
2503 the stack is aligned in some way, such as calls or SSE instructions on
2504 x86, yet will not contain code that does that alignment within the asm.
2505 The compiler should make conservative assumptions about what the asm
2506 might contain and should generate its usual stack alignment code in the
2507 prologue if the '``alignstack``' keyword is present:
2509 .. code-block:: llvm
2511 call void asm alignstack "eieio", ""()
2513 Inline asms also support using non-standard assembly dialects. The
2514 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2515 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2516 the only supported dialects. An example is:
2518 .. code-block:: llvm
2520 call void asm inteldialect "eieio", ""()
2522 If multiple keywords appear the '``sideeffect``' keyword must come
2523 first, the '``alignstack``' keyword second and the '``inteldialect``'
2529 The call instructions that wrap inline asm nodes may have a
2530 "``!srcloc``" MDNode attached to it that contains a list of constant
2531 integers. If present, the code generator will use the integer as the
2532 location cookie value when report errors through the ``LLVMContext``
2533 error reporting mechanisms. This allows a front-end to correlate backend
2534 errors that occur with inline asm back to the source code that produced
2537 .. code-block:: llvm
2539 call void asm sideeffect "something bad", ""(), !srcloc !42
2541 !42 = !{ i32 1234567 }
2543 It is up to the front-end to make sense of the magic numbers it places
2544 in the IR. If the MDNode contains multiple constants, the code generator
2545 will use the one that corresponds to the line of the asm that the error
2550 Metadata Nodes and Metadata Strings
2551 -----------------------------------
2553 LLVM IR allows metadata to be attached to instructions in the program
2554 that can convey extra information about the code to the optimizers and
2555 code generator. One example application of metadata is source-level
2556 debug information. There are two metadata primitives: strings and nodes.
2557 All metadata has the ``metadata`` type and is identified in syntax by a
2558 preceding exclamation point ('``!``').
2560 A metadata string is a string surrounded by double quotes. It can
2561 contain any character by escaping non-printable characters with
2562 "``\xx``" where "``xx``" is the two digit hex code. For example:
2565 Metadata nodes are represented with notation similar to structure
2566 constants (a comma separated list of elements, surrounded by braces and
2567 preceded by an exclamation point). Metadata nodes can have any values as
2568 their operand. For example:
2570 .. code-block:: llvm
2572 !{ metadata !"test\00", i32 10}
2574 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2575 metadata nodes, which can be looked up in the module symbol table. For
2578 .. code-block:: llvm
2580 !foo = metadata !{!4, !3}
2582 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2583 function is using two metadata arguments:
2585 .. code-block:: llvm
2587 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2589 Metadata can be attached with an instruction. Here metadata ``!21`` is
2590 attached to the ``add`` instruction using the ``!dbg`` identifier:
2592 .. code-block:: llvm
2594 %indvar.next = add i64 %indvar, 1, !dbg !21
2596 More information about specific metadata nodes recognized by the
2597 optimizers and code generator is found below.
2602 In LLVM IR, memory does not have types, so LLVM's own type system is not
2603 suitable for doing TBAA. Instead, metadata is added to the IR to
2604 describe a type system of a higher level language. This can be used to
2605 implement typical C/C++ TBAA, but it can also be used to implement
2606 custom alias analysis behavior for other languages.
2608 The current metadata format is very simple. TBAA metadata nodes have up
2609 to three fields, e.g.:
2611 .. code-block:: llvm
2613 !0 = metadata !{ metadata !"an example type tree" }
2614 !1 = metadata !{ metadata !"int", metadata !0 }
2615 !2 = metadata !{ metadata !"float", metadata !0 }
2616 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2618 The first field is an identity field. It can be any value, usually a
2619 metadata string, which uniquely identifies the type. The most important
2620 name in the tree is the name of the root node. Two trees with different
2621 root node names are entirely disjoint, even if they have leaves with
2624 The second field identifies the type's parent node in the tree, or is
2625 null or omitted for a root node. A type is considered to alias all of
2626 its descendants and all of its ancestors in the tree. Also, a type is
2627 considered to alias all types in other trees, so that bitcode produced
2628 from multiple front-ends is handled conservatively.
2630 If the third field is present, it's an integer which if equal to 1
2631 indicates that the type is "constant" (meaning
2632 ``pointsToConstantMemory`` should return true; see `other useful
2633 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2635 '``tbaa.struct``' Metadata
2636 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2638 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2639 aggregate assignment operations in C and similar languages, however it
2640 is defined to copy a contiguous region of memory, which is more than
2641 strictly necessary for aggregate types which contain holes due to
2642 padding. Also, it doesn't contain any TBAA information about the fields
2645 ``!tbaa.struct`` metadata can describe which memory subregions in a
2646 memcpy are padding and what the TBAA tags of the struct are.
2648 The current metadata format is very simple. ``!tbaa.struct`` metadata
2649 nodes are a list of operands which are in conceptual groups of three.
2650 For each group of three, the first operand gives the byte offset of a
2651 field in bytes, the second gives its size in bytes, and the third gives
2654 .. code-block:: llvm
2656 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2658 This describes a struct with two fields. The first is at offset 0 bytes
2659 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2660 and has size 4 bytes and has tbaa tag !2.
2662 Note that the fields need not be contiguous. In this example, there is a
2663 4 byte gap between the two fields. This gap represents padding which
2664 does not carry useful data and need not be preserved.
2666 '``fpmath``' Metadata
2667 ^^^^^^^^^^^^^^^^^^^^^
2669 ``fpmath`` metadata may be attached to any instruction of floating point
2670 type. It can be used to express the maximum acceptable error in the
2671 result of that instruction, in ULPs, thus potentially allowing the
2672 compiler to use a more efficient but less accurate method of computing
2673 it. ULP is defined as follows:
2675 If ``x`` is a real number that lies between two finite consecutive
2676 floating-point numbers ``a`` and ``b``, without being equal to one
2677 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2678 distance between the two non-equal finite floating-point numbers
2679 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2681 The metadata node shall consist of a single positive floating point
2682 number representing the maximum relative error, for example:
2684 .. code-block:: llvm
2686 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2688 '``range``' Metadata
2689 ^^^^^^^^^^^^^^^^^^^^
2691 ``range`` metadata may be attached only to loads of integer types. It
2692 expresses the possible ranges the loaded value is in. The ranges are
2693 represented with a flattened list of integers. The loaded value is known
2694 to be in the union of the ranges defined by each consecutive pair. Each
2695 pair has the following properties:
2697 - The type must match the type loaded by the instruction.
2698 - The pair ``a,b`` represents the range ``[a,b)``.
2699 - Both ``a`` and ``b`` are constants.
2700 - The range is allowed to wrap.
2701 - The range should not represent the full or empty set. That is,
2704 In addition, the pairs must be in signed order of the lower bound and
2705 they must be non-contiguous.
2709 .. code-block:: llvm
2711 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2712 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2713 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2714 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2716 !0 = metadata !{ i8 0, i8 2 }
2717 !1 = metadata !{ i8 255, i8 2 }
2718 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2719 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2724 It is sometimes useful to attach information to loop constructs. Currently,
2725 loop metadata is implemented as metadata attached to the branch instruction
2726 in the loop latch block. This type of metadata refer to a metadata node that is
2727 guaranteed to be separate for each loop. The loop identifier metadata is
2728 specified with the name ``llvm.loop``.
2730 The loop identifier metadata is implemented using a metadata that refers to
2731 itself to avoid merging it with any other identifier metadata, e.g.,
2732 during module linkage or function inlining. That is, each loop should refer
2733 to their own identification metadata even if they reside in separate functions.
2734 The following example contains loop identifier metadata for two separate loop
2737 .. code-block:: llvm
2739 !0 = metadata !{ metadata !0 }
2740 !1 = metadata !{ metadata !1 }
2742 The loop identifier metadata can be used to specify additional per-loop
2743 metadata. Any operands after the first operand can be treated as user-defined
2744 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2745 by the loop vectorizer to indicate how many times to unroll the loop:
2747 .. code-block:: llvm
2749 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2751 !0 = metadata !{ metadata !0, metadata !1 }
2752 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2757 Metadata types used to annotate memory accesses with information helpful
2758 for optimizations are prefixed with ``llvm.mem``.
2760 '``llvm.mem.parallel_loop_access``' Metadata
2761 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2763 For a loop to be parallel, in addition to using
2764 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2765 also all of the memory accessing instructions in the loop body need to be
2766 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2767 is at least one memory accessing instruction not marked with the metadata,
2768 the loop must be considered a sequential loop. This causes parallel loops to be
2769 converted to sequential loops due to optimization passes that are unaware of
2770 the parallel semantics and that insert new memory instructions to the loop
2773 Example of a loop that is considered parallel due to its correct use of
2774 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2775 metadata types that refer to the same loop identifier metadata.
2777 .. code-block:: llvm
2781 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2783 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2785 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2789 !0 = metadata !{ metadata !0 }
2791 It is also possible to have nested parallel loops. In that case the
2792 memory accesses refer to a list of loop identifier metadata nodes instead of
2793 the loop identifier metadata node directly:
2795 .. code-block:: llvm
2802 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2804 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2806 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2810 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2812 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2814 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2816 outer.for.end: ; preds = %for.body
2818 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2819 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2820 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2822 '``llvm.vectorizer``'
2823 ^^^^^^^^^^^^^^^^^^^^^
2825 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2826 vectorization parameters such as vectorization factor and unroll factor.
2828 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2829 loop identification metadata.
2831 '``llvm.vectorizer.unroll``' Metadata
2832 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2834 This metadata instructs the loop vectorizer to unroll the specified
2835 loop exactly ``N`` times.
2837 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2838 operand is an integer specifying the unroll factor. For example:
2840 .. code-block:: llvm
2842 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2844 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2847 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2848 determined automatically.
2850 '``llvm.vectorizer.width``' Metadata
2851 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2853 This metadata sets the target width of the vectorizer to ``N``. Without
2854 this metadata, the vectorizer will choose a width automatically.
2855 Regardless of this metadata, the vectorizer will only vectorize loops if
2856 it believes it is valid to do so.
2858 The first operand is the string ``llvm.vectorizer.width`` and the second
2859 operand is an integer specifying the width. For example:
2861 .. code-block:: llvm
2863 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2865 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2868 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2871 Module Flags Metadata
2872 =====================
2874 Information about the module as a whole is difficult to convey to LLVM's
2875 subsystems. The LLVM IR isn't sufficient to transmit this information.
2876 The ``llvm.module.flags`` named metadata exists in order to facilitate
2877 this. These flags are in the form of key / value pairs --- much like a
2878 dictionary --- making it easy for any subsystem who cares about a flag to
2881 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2882 Each triplet has the following form:
2884 - The first element is a *behavior* flag, which specifies the behavior
2885 when two (or more) modules are merged together, and it encounters two
2886 (or more) metadata with the same ID. The supported behaviors are
2888 - The second element is a metadata string that is a unique ID for the
2889 metadata. Each module may only have one flag entry for each unique ID (not
2890 including entries with the **Require** behavior).
2891 - The third element is the value of the flag.
2893 When two (or more) modules are merged together, the resulting
2894 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2895 each unique metadata ID string, there will be exactly one entry in the merged
2896 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2897 be determined by the merge behavior flag, as described below. The only exception
2898 is that entries with the *Require* behavior are always preserved.
2900 The following behaviors are supported:
2911 Emits an error if two values disagree, otherwise the resulting value
2912 is that of the operands.
2916 Emits a warning if two values disagree. The result value will be the
2917 operand for the flag from the first module being linked.
2921 Adds a requirement that another module flag be present and have a
2922 specified value after linking is performed. The value must be a
2923 metadata pair, where the first element of the pair is the ID of the
2924 module flag to be restricted, and the second element of the pair is
2925 the value the module flag should be restricted to. This behavior can
2926 be used to restrict the allowable results (via triggering of an
2927 error) of linking IDs with the **Override** behavior.
2931 Uses the specified value, regardless of the behavior or value of the
2932 other module. If both modules specify **Override**, but the values
2933 differ, an error will be emitted.
2937 Appends the two values, which are required to be metadata nodes.
2941 Appends the two values, which are required to be metadata
2942 nodes. However, duplicate entries in the second list are dropped
2943 during the append operation.
2945 It is an error for a particular unique flag ID to have multiple behaviors,
2946 except in the case of **Require** (which adds restrictions on another metadata
2947 value) or **Override**.
2949 An example of module flags:
2951 .. code-block:: llvm
2953 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2954 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2955 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2956 !3 = metadata !{ i32 3, metadata !"qux",
2958 metadata !"foo", i32 1
2961 !llvm.module.flags = !{ !0, !1, !2, !3 }
2963 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2964 if two or more ``!"foo"`` flags are seen is to emit an error if their
2965 values are not equal.
2967 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2968 behavior if two or more ``!"bar"`` flags are seen is to use the value
2971 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2972 behavior if two or more ``!"qux"`` flags are seen is to emit a
2973 warning if their values are not equal.
2975 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2979 metadata !{ metadata !"foo", i32 1 }
2981 The behavior is to emit an error if the ``llvm.module.flags`` does not
2982 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2985 Objective-C Garbage Collection Module Flags Metadata
2986 ----------------------------------------------------
2988 On the Mach-O platform, Objective-C stores metadata about garbage
2989 collection in a special section called "image info". The metadata
2990 consists of a version number and a bitmask specifying what types of
2991 garbage collection are supported (if any) by the file. If two or more
2992 modules are linked together their garbage collection metadata needs to
2993 be merged rather than appended together.
2995 The Objective-C garbage collection module flags metadata consists of the
2996 following key-value pairs:
3005 * - ``Objective-C Version``
3006 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3008 * - ``Objective-C Image Info Version``
3009 - **[Required]** --- The version of the image info section. Currently
3012 * - ``Objective-C Image Info Section``
3013 - **[Required]** --- The section to place the metadata. Valid values are
3014 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3015 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3016 Objective-C ABI version 2.
3018 * - ``Objective-C Garbage Collection``
3019 - **[Required]** --- Specifies whether garbage collection is supported or
3020 not. Valid values are 0, for no garbage collection, and 2, for garbage
3021 collection supported.
3023 * - ``Objective-C GC Only``
3024 - **[Optional]** --- Specifies that only garbage collection is supported.
3025 If present, its value must be 6. This flag requires that the
3026 ``Objective-C Garbage Collection`` flag have the value 2.
3028 Some important flag interactions:
3030 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3031 merged with a module with ``Objective-C Garbage Collection`` set to
3032 2, then the resulting module has the
3033 ``Objective-C Garbage Collection`` flag set to 0.
3034 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3035 merged with a module with ``Objective-C GC Only`` set to 6.
3037 Automatic Linker Flags Module Flags Metadata
3038 --------------------------------------------
3040 Some targets support embedding flags to the linker inside individual object
3041 files. Typically this is used in conjunction with language extensions which
3042 allow source files to explicitly declare the libraries they depend on, and have
3043 these automatically be transmitted to the linker via object files.
3045 These flags are encoded in the IR using metadata in the module flags section,
3046 using the ``Linker Options`` key. The merge behavior for this flag is required
3047 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3048 node which should be a list of other metadata nodes, each of which should be a
3049 list of metadata strings defining linker options.
3051 For example, the following metadata section specifies two separate sets of
3052 linker options, presumably to link against ``libz`` and the ``Cocoa``
3055 !0 = metadata !{ i32 6, metadata !"Linker Options",
3057 metadata !{ metadata !"-lz" },
3058 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3059 !llvm.module.flags = !{ !0 }
3061 The metadata encoding as lists of lists of options, as opposed to a collapsed
3062 list of options, is chosen so that the IR encoding can use multiple option
3063 strings to specify e.g., a single library, while still having that specifier be
3064 preserved as an atomic element that can be recognized by a target specific
3065 assembly writer or object file emitter.
3067 Each individual option is required to be either a valid option for the target's
3068 linker, or an option that is reserved by the target specific assembly writer or
3069 object file emitter. No other aspect of these options is defined by the IR.
3071 .. _intrinsicglobalvariables:
3073 Intrinsic Global Variables
3074 ==========================
3076 LLVM has a number of "magic" global variables that contain data that
3077 affect code generation or other IR semantics. These are documented here.
3078 All globals of this sort should have a section specified as
3079 "``llvm.metadata``". This section and all globals that start with
3080 "``llvm.``" are reserved for use by LLVM.
3084 The '``llvm.used``' Global Variable
3085 -----------------------------------
3087 The ``@llvm.used`` global is an array which has
3088 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3089 pointers to named global variables, functions and aliases which may optionally
3090 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3093 .. code-block:: llvm
3098 @llvm.used = appending global [2 x i8*] [
3100 i8* bitcast (i32* @Y to i8*)
3101 ], section "llvm.metadata"
3103 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3104 and linker are required to treat the symbol as if there is a reference to the
3105 symbol that it cannot see (which is why they have to be named). For example, if
3106 a variable has internal linkage and no references other than that from the
3107 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3108 references from inline asms and other things the compiler cannot "see", and
3109 corresponds to "``attribute((used))``" in GNU C.
3111 On some targets, the code generator must emit a directive to the
3112 assembler or object file to prevent the assembler and linker from
3113 molesting the symbol.
3115 .. _gv_llvmcompilerused:
3117 The '``llvm.compiler.used``' Global Variable
3118 --------------------------------------------
3120 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3121 directive, except that it only prevents the compiler from touching the
3122 symbol. On targets that support it, this allows an intelligent linker to
3123 optimize references to the symbol without being impeded as it would be
3126 This is a rare construct that should only be used in rare circumstances,
3127 and should not be exposed to source languages.
3129 .. _gv_llvmglobalctors:
3131 The '``llvm.global_ctors``' Global Variable
3132 -------------------------------------------
3134 .. code-block:: llvm
3136 %0 = type { i32, void ()* }
3137 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3139 The ``@llvm.global_ctors`` array contains a list of constructor
3140 functions and associated priorities. The functions referenced by this
3141 array will be called in ascending order of priority (i.e. lowest first)
3142 when the module is loaded. The order of functions with the same priority
3145 .. _llvmglobaldtors:
3147 The '``llvm.global_dtors``' Global Variable
3148 -------------------------------------------
3150 .. code-block:: llvm
3152 %0 = type { i32, void ()* }
3153 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3155 The ``@llvm.global_dtors`` array contains a list of destructor functions
3156 and associated priorities. The functions referenced by this array will
3157 be called in descending order of priority (i.e. highest first) when the
3158 module is loaded. The order of functions with the same priority is not
3161 Instruction Reference
3162 =====================
3164 The LLVM instruction set consists of several different classifications
3165 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3166 instructions <binaryops>`, :ref:`bitwise binary
3167 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3168 :ref:`other instructions <otherops>`.
3172 Terminator Instructions
3173 -----------------------
3175 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3176 program ends with a "Terminator" instruction, which indicates which
3177 block should be executed after the current block is finished. These
3178 terminator instructions typically yield a '``void``' value: they produce
3179 control flow, not values (the one exception being the
3180 ':ref:`invoke <i_invoke>`' instruction).
3182 The terminator instructions are: ':ref:`ret <i_ret>`',
3183 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3184 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3185 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3189 '``ret``' Instruction
3190 ^^^^^^^^^^^^^^^^^^^^^
3197 ret <type> <value> ; Return a value from a non-void function
3198 ret void ; Return from void function
3203 The '``ret``' instruction is used to return control flow (and optionally
3204 a value) from a function back to the caller.
3206 There are two forms of the '``ret``' instruction: one that returns a
3207 value and then causes control flow, and one that just causes control
3213 The '``ret``' instruction optionally accepts a single argument, the
3214 return value. The type of the return value must be a ':ref:`first
3215 class <t_firstclass>`' type.
3217 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3218 return type and contains a '``ret``' instruction with no return value or
3219 a return value with a type that does not match its type, or if it has a
3220 void return type and contains a '``ret``' instruction with a return
3226 When the '``ret``' instruction is executed, control flow returns back to
3227 the calling function's context. If the caller is a
3228 ":ref:`call <i_call>`" instruction, execution continues at the
3229 instruction after the call. If the caller was an
3230 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3231 beginning of the "normal" destination block. If the instruction returns
3232 a value, that value shall set the call or invoke instruction's return
3238 .. code-block:: llvm
3240 ret i32 5 ; Return an integer value of 5
3241 ret void ; Return from a void function
3242 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3246 '``br``' Instruction
3247 ^^^^^^^^^^^^^^^^^^^^
3254 br i1 <cond>, label <iftrue>, label <iffalse>
3255 br label <dest> ; Unconditional branch
3260 The '``br``' instruction is used to cause control flow to transfer to a
3261 different basic block in the current function. There are two forms of
3262 this instruction, corresponding to a conditional branch and an
3263 unconditional branch.
3268 The conditional branch form of the '``br``' instruction takes a single
3269 '``i1``' value and two '``label``' values. The unconditional form of the
3270 '``br``' instruction takes a single '``label``' value as a target.
3275 Upon execution of a conditional '``br``' instruction, the '``i1``'
3276 argument is evaluated. If the value is ``true``, control flows to the
3277 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3278 to the '``iffalse``' ``label`` argument.
3283 .. code-block:: llvm
3286 %cond = icmp eq i32 %a, %b
3287 br i1 %cond, label %IfEqual, label %IfUnequal
3295 '``switch``' Instruction
3296 ^^^^^^^^^^^^^^^^^^^^^^^^
3303 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3308 The '``switch``' instruction is used to transfer control flow to one of
3309 several different places. It is a generalization of the '``br``'
3310 instruction, allowing a branch to occur to one of many possible
3316 The '``switch``' instruction uses three parameters: an integer
3317 comparison value '``value``', a default '``label``' destination, and an
3318 array of pairs of comparison value constants and '``label``'s. The table
3319 is not allowed to contain duplicate constant entries.
3324 The ``switch`` instruction specifies a table of values and destinations.
3325 When the '``switch``' instruction is executed, this table is searched
3326 for the given value. If the value is found, control flow is transferred
3327 to the corresponding destination; otherwise, control flow is transferred
3328 to the default destination.
3333 Depending on properties of the target machine and the particular
3334 ``switch`` instruction, this instruction may be code generated in
3335 different ways. For example, it could be generated as a series of
3336 chained conditional branches or with a lookup table.
3341 .. code-block:: llvm
3343 ; Emulate a conditional br instruction
3344 %Val = zext i1 %value to i32
3345 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3347 ; Emulate an unconditional br instruction
3348 switch i32 0, label %dest [ ]
3350 ; Implement a jump table:
3351 switch i32 %val, label %otherwise [ i32 0, label %onzero
3353 i32 2, label %ontwo ]
3357 '``indirectbr``' Instruction
3358 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3365 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3370 The '``indirectbr``' instruction implements an indirect branch to a
3371 label within the current function, whose address is specified by
3372 "``address``". Address must be derived from a
3373 :ref:`blockaddress <blockaddress>` constant.
3378 The '``address``' argument is the address of the label to jump to. The
3379 rest of the arguments indicate the full set of possible destinations
3380 that the address may point to. Blocks are allowed to occur multiple
3381 times in the destination list, though this isn't particularly useful.
3383 This destination list is required so that dataflow analysis has an
3384 accurate understanding of the CFG.
3389 Control transfers to the block specified in the address argument. All
3390 possible destination blocks must be listed in the label list, otherwise
3391 this instruction has undefined behavior. This implies that jumps to
3392 labels defined in other functions have undefined behavior as well.
3397 This is typically implemented with a jump through a register.
3402 .. code-block:: llvm
3404 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3408 '``invoke``' Instruction
3409 ^^^^^^^^^^^^^^^^^^^^^^^^
3416 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3417 to label <normal label> unwind label <exception label>
3422 The '``invoke``' instruction causes control to transfer to a specified
3423 function, with the possibility of control flow transfer to either the
3424 '``normal``' label or the '``exception``' label. If the callee function
3425 returns with the "``ret``" instruction, control flow will return to the
3426 "normal" label. If the callee (or any indirect callees) returns via the
3427 ":ref:`resume <i_resume>`" instruction or other exception handling
3428 mechanism, control is interrupted and continued at the dynamically
3429 nearest "exception" label.
3431 The '``exception``' label is a `landing
3432 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3433 '``exception``' label is required to have the
3434 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3435 information about the behavior of the program after unwinding happens,
3436 as its first non-PHI instruction. The restrictions on the
3437 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3438 instruction, so that the important information contained within the
3439 "``landingpad``" instruction can't be lost through normal code motion.
3444 This instruction requires several arguments:
3446 #. The optional "cconv" marker indicates which :ref:`calling
3447 convention <callingconv>` the call should use. If none is
3448 specified, the call defaults to using C calling conventions.
3449 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3450 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3452 #. '``ptr to function ty``': shall be the signature of the pointer to
3453 function value being invoked. In most cases, this is a direct
3454 function invocation, but indirect ``invoke``'s are just as possible,
3455 branching off an arbitrary pointer to function value.
3456 #. '``function ptr val``': An LLVM value containing a pointer to a
3457 function to be invoked.
3458 #. '``function args``': argument list whose types match the function
3459 signature argument types and parameter attributes. All arguments must
3460 be of :ref:`first class <t_firstclass>` type. If the function signature
3461 indicates the function accepts a variable number of arguments, the
3462 extra arguments can be specified.
3463 #. '``normal label``': the label reached when the called function
3464 executes a '``ret``' instruction.
3465 #. '``exception label``': the label reached when a callee returns via
3466 the :ref:`resume <i_resume>` instruction or other exception handling
3468 #. The optional :ref:`function attributes <fnattrs>` list. Only
3469 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3470 attributes are valid here.
3475 This instruction is designed to operate as a standard '``call``'
3476 instruction in most regards. The primary difference is that it
3477 establishes an association with a label, which is used by the runtime
3478 library to unwind the stack.
3480 This instruction is used in languages with destructors to ensure that
3481 proper cleanup is performed in the case of either a ``longjmp`` or a
3482 thrown exception. Additionally, this is important for implementation of
3483 '``catch``' clauses in high-level languages that support them.
3485 For the purposes of the SSA form, the definition of the value returned
3486 by the '``invoke``' instruction is deemed to occur on the edge from the
3487 current block to the "normal" label. If the callee unwinds then no
3488 return value is available.
3493 .. code-block:: llvm
3495 %retval = invoke i32 @Test(i32 15) to label %Continue
3496 unwind label %TestCleanup ; {i32}:retval set
3497 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3498 unwind label %TestCleanup ; {i32}:retval set
3502 '``resume``' Instruction
3503 ^^^^^^^^^^^^^^^^^^^^^^^^
3510 resume <type> <value>
3515 The '``resume``' instruction is a terminator instruction that has no
3521 The '``resume``' instruction requires one argument, which must have the
3522 same type as the result of any '``landingpad``' instruction in the same
3528 The '``resume``' instruction resumes propagation of an existing
3529 (in-flight) exception whose unwinding was interrupted with a
3530 :ref:`landingpad <i_landingpad>` instruction.
3535 .. code-block:: llvm
3537 resume { i8*, i32 } %exn
3541 '``unreachable``' Instruction
3542 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3554 The '``unreachable``' instruction has no defined semantics. This
3555 instruction is used to inform the optimizer that a particular portion of
3556 the code is not reachable. This can be used to indicate that the code
3557 after a no-return function cannot be reached, and other facts.
3562 The '``unreachable``' instruction has no defined semantics.
3569 Binary operators are used to do most of the computation in a program.
3570 They require two operands of the same type, execute an operation on
3571 them, and produce a single value. The operands might represent multiple
3572 data, as is the case with the :ref:`vector <t_vector>` data type. The
3573 result value has the same type as its operands.
3575 There are several different binary operators:
3579 '``add``' Instruction
3580 ^^^^^^^^^^^^^^^^^^^^^
3587 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3588 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3589 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3590 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3595 The '``add``' instruction returns the sum of its two operands.
3600 The two arguments to the '``add``' instruction must be
3601 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3602 arguments must have identical types.
3607 The value produced is the integer sum of the two operands.
3609 If the sum has unsigned overflow, the result returned is the
3610 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3613 Because LLVM integers use a two's complement representation, this
3614 instruction is appropriate for both signed and unsigned integers.
3616 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3617 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3618 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3619 unsigned and/or signed overflow, respectively, occurs.
3624 .. code-block:: llvm
3626 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3630 '``fadd``' Instruction
3631 ^^^^^^^^^^^^^^^^^^^^^^
3638 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3643 The '``fadd``' instruction returns the sum of its two operands.
3648 The two arguments to the '``fadd``' instruction must be :ref:`floating
3649 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3650 Both arguments must have identical types.
3655 The value produced is the floating point sum of the two operands. This
3656 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3657 which are optimization hints to enable otherwise unsafe floating point
3663 .. code-block:: llvm
3665 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3667 '``sub``' Instruction
3668 ^^^^^^^^^^^^^^^^^^^^^
3675 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3676 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3677 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3678 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3683 The '``sub``' instruction returns the difference of its two operands.
3685 Note that the '``sub``' instruction is used to represent the '``neg``'
3686 instruction present in most other intermediate representations.
3691 The two arguments to the '``sub``' instruction must be
3692 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3693 arguments must have identical types.
3698 The value produced is the integer difference of the two operands.
3700 If the difference has unsigned overflow, the result returned is the
3701 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3704 Because LLVM integers use a two's complement representation, this
3705 instruction is appropriate for both signed and unsigned integers.
3707 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3708 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3709 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3710 unsigned and/or signed overflow, respectively, occurs.
3715 .. code-block:: llvm
3717 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3718 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3722 '``fsub``' Instruction
3723 ^^^^^^^^^^^^^^^^^^^^^^
3730 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3735 The '``fsub``' instruction returns the difference of its two operands.
3737 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3738 instruction present in most other intermediate representations.
3743 The two arguments to the '``fsub``' instruction must be :ref:`floating
3744 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3745 Both arguments must have identical types.
3750 The value produced is the floating point difference of the two operands.
3751 This instruction can also take any number of :ref:`fast-math
3752 flags <fastmath>`, which are optimization hints to enable otherwise
3753 unsafe floating point optimizations:
3758 .. code-block:: llvm
3760 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3761 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3763 '``mul``' Instruction
3764 ^^^^^^^^^^^^^^^^^^^^^
3771 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3772 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3773 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3774 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3779 The '``mul``' instruction returns the product of its two operands.
3784 The two arguments to the '``mul``' instruction must be
3785 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3786 arguments must have identical types.
3791 The value produced is the integer product of the two operands.
3793 If the result of the multiplication has unsigned overflow, the result
3794 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3795 bit width of the result.
3797 Because LLVM integers use a two's complement representation, and the
3798 result is the same width as the operands, this instruction returns the
3799 correct result for both signed and unsigned integers. If a full product
3800 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3801 sign-extended or zero-extended as appropriate to the width of the full
3804 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3805 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3806 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3807 unsigned and/or signed overflow, respectively, occurs.
3812 .. code-block:: llvm
3814 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3818 '``fmul``' Instruction
3819 ^^^^^^^^^^^^^^^^^^^^^^
3826 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3831 The '``fmul``' instruction returns the product of its two operands.
3836 The two arguments to the '``fmul``' instruction must be :ref:`floating
3837 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3838 Both arguments must have identical types.
3843 The value produced is the floating point product of the two operands.
3844 This instruction can also take any number of :ref:`fast-math
3845 flags <fastmath>`, which are optimization hints to enable otherwise
3846 unsafe floating point optimizations:
3851 .. code-block:: llvm
3853 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3855 '``udiv``' Instruction
3856 ^^^^^^^^^^^^^^^^^^^^^^
3863 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3864 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3869 The '``udiv``' instruction returns the quotient of its two operands.
3874 The two arguments to the '``udiv``' instruction must be
3875 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3876 arguments must have identical types.
3881 The value produced is the unsigned integer quotient of the two operands.
3883 Note that unsigned integer division and signed integer division are
3884 distinct operations; for signed integer division, use '``sdiv``'.
3886 Division by zero leads to undefined behavior.
3888 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3889 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3890 such, "((a udiv exact b) mul b) == a").
3895 .. code-block:: llvm
3897 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3899 '``sdiv``' Instruction
3900 ^^^^^^^^^^^^^^^^^^^^^^
3907 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3908 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3913 The '``sdiv``' instruction returns the quotient of its two operands.
3918 The two arguments to the '``sdiv``' instruction must be
3919 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3920 arguments must have identical types.
3925 The value produced is the signed integer quotient of the two operands
3926 rounded towards zero.
3928 Note that signed integer division and unsigned integer division are
3929 distinct operations; for unsigned integer division, use '``udiv``'.
3931 Division by zero leads to undefined behavior. Overflow also leads to
3932 undefined behavior; this is a rare case, but can occur, for example, by
3933 doing a 32-bit division of -2147483648 by -1.
3935 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3936 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3941 .. code-block:: llvm
3943 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3947 '``fdiv``' Instruction
3948 ^^^^^^^^^^^^^^^^^^^^^^
3955 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3960 The '``fdiv``' instruction returns the quotient of its two operands.
3965 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3966 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3967 Both arguments must have identical types.
3972 The value produced is the floating point quotient of the two operands.
3973 This instruction can also take any number of :ref:`fast-math
3974 flags <fastmath>`, which are optimization hints to enable otherwise
3975 unsafe floating point optimizations:
3980 .. code-block:: llvm
3982 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3984 '``urem``' Instruction
3985 ^^^^^^^^^^^^^^^^^^^^^^
3992 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3997 The '``urem``' instruction returns the remainder from the unsigned
3998 division of its two arguments.
4003 The two arguments to the '``urem``' instruction must be
4004 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4005 arguments must have identical types.
4010 This instruction returns the unsigned integer *remainder* of a division.
4011 This instruction always performs an unsigned division to get the
4014 Note that unsigned integer remainder and signed integer remainder are
4015 distinct operations; for signed integer remainder, use '``srem``'.
4017 Taking the remainder of a division by zero leads to undefined behavior.
4022 .. code-block:: llvm
4024 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
4026 '``srem``' Instruction
4027 ^^^^^^^^^^^^^^^^^^^^^^
4034 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
4039 The '``srem``' instruction returns the remainder from the signed
4040 division of its two operands. This instruction can also take
4041 :ref:`vector <t_vector>` versions of the values in which case the elements
4047 The two arguments to the '``srem``' instruction must be
4048 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4049 arguments must have identical types.
4054 This instruction returns the *remainder* of a division (where the result
4055 is either zero or has the same sign as the dividend, ``op1``), not the
4056 *modulo* operator (where the result is either zero or has the same sign
4057 as the divisor, ``op2``) of a value. For more information about the
4058 difference, see `The Math
4059 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4060 table of how this is implemented in various languages, please see
4062 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4064 Note that signed integer remainder and unsigned integer remainder are
4065 distinct operations; for unsigned integer remainder, use '``urem``'.
4067 Taking the remainder of a division by zero leads to undefined behavior.
4068 Overflow also leads to undefined behavior; this is a rare case, but can
4069 occur, for example, by taking the remainder of a 32-bit division of
4070 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4071 rule lets srem be implemented using instructions that return both the
4072 result of the division and the remainder.)
4077 .. code-block:: llvm
4079 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4083 '``frem``' Instruction
4084 ^^^^^^^^^^^^^^^^^^^^^^
4091 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4096 The '``frem``' instruction returns the remainder from the division of
4102 The two arguments to the '``frem``' instruction must be :ref:`floating
4103 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4104 Both arguments must have identical types.
4109 This instruction returns the *remainder* of a division. The remainder
4110 has the same sign as the dividend. This instruction can also take any
4111 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4112 to enable otherwise unsafe floating point optimizations:
4117 .. code-block:: llvm
4119 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4123 Bitwise Binary Operations
4124 -------------------------
4126 Bitwise binary operators are used to do various forms of bit-twiddling
4127 in a program. They are generally very efficient instructions and can
4128 commonly be strength reduced from other instructions. They require two
4129 operands of the same type, execute an operation on them, and produce a
4130 single value. The resulting value is the same type as its operands.
4132 '``shl``' Instruction
4133 ^^^^^^^^^^^^^^^^^^^^^
4140 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4141 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4142 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4143 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4148 The '``shl``' instruction returns the first operand shifted to the left
4149 a specified number of bits.
4154 Both arguments to the '``shl``' instruction must be the same
4155 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4156 '``op2``' is treated as an unsigned value.
4161 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4162 where ``n`` is the width of the result. If ``op2`` is (statically or
4163 dynamically) negative or equal to or larger than the number of bits in
4164 ``op1``, the result is undefined. If the arguments are vectors, each
4165 vector element of ``op1`` is shifted by the corresponding shift amount
4168 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4169 value <poisonvalues>` if it shifts out any non-zero bits. If the
4170 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4171 value <poisonvalues>` if it shifts out any bits that disagree with the
4172 resultant sign bit. As such, NUW/NSW have the same semantics as they
4173 would if the shift were expressed as a mul instruction with the same
4174 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4179 .. code-block:: llvm
4181 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4182 <result> = shl i32 4, 2 ; yields {i32}: 16
4183 <result> = shl i32 1, 10 ; yields {i32}: 1024
4184 <result> = shl i32 1, 32 ; undefined
4185 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4187 '``lshr``' Instruction
4188 ^^^^^^^^^^^^^^^^^^^^^^
4195 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4196 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4201 The '``lshr``' instruction (logical shift right) returns the first
4202 operand shifted to the right a specified number of bits with zero fill.
4207 Both arguments to the '``lshr``' instruction must be the same
4208 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4209 '``op2``' is treated as an unsigned value.
4214 This instruction always performs a logical shift right operation. The
4215 most significant bits of the result will be filled with zero bits after
4216 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4217 than the number of bits in ``op1``, the result is undefined. If the
4218 arguments are vectors, each vector element of ``op1`` is shifted by the
4219 corresponding shift amount in ``op2``.
4221 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4222 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4228 .. code-block:: llvm
4230 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4231 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4232 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4233 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4234 <result> = lshr i32 1, 32 ; undefined
4235 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4237 '``ashr``' Instruction
4238 ^^^^^^^^^^^^^^^^^^^^^^
4245 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4246 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4251 The '``ashr``' instruction (arithmetic shift right) returns the first
4252 operand shifted to the right a specified number of bits with sign
4258 Both arguments to the '``ashr``' instruction must be the same
4259 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4260 '``op2``' is treated as an unsigned value.
4265 This instruction always performs an arithmetic shift right operation,
4266 The most significant bits of the result will be filled with the sign bit
4267 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4268 than the number of bits in ``op1``, the result is undefined. If the
4269 arguments are vectors, each vector element of ``op1`` is shifted by the
4270 corresponding shift amount in ``op2``.
4272 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4273 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4279 .. code-block:: llvm
4281 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4282 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4283 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4284 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4285 <result> = ashr i32 1, 32 ; undefined
4286 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4288 '``and``' Instruction
4289 ^^^^^^^^^^^^^^^^^^^^^
4296 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4301 The '``and``' instruction returns the bitwise logical and of its two
4307 The two arguments to the '``and``' instruction must be
4308 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4309 arguments must have identical types.
4314 The truth table used for the '``and``' instruction is:
4331 .. code-block:: llvm
4333 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4334 <result> = and i32 15, 40 ; yields {i32}:result = 8
4335 <result> = and i32 4, 8 ; yields {i32}:result = 0
4337 '``or``' Instruction
4338 ^^^^^^^^^^^^^^^^^^^^
4345 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4350 The '``or``' instruction returns the bitwise logical inclusive or of its
4356 The two arguments to the '``or``' instruction must be
4357 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4358 arguments must have identical types.
4363 The truth table used for the '``or``' instruction is:
4382 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4383 <result> = or i32 15, 40 ; yields {i32}:result = 47
4384 <result> = or i32 4, 8 ; yields {i32}:result = 12
4386 '``xor``' Instruction
4387 ^^^^^^^^^^^^^^^^^^^^^
4394 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4399 The '``xor``' instruction returns the bitwise logical exclusive or of
4400 its two operands. The ``xor`` is used to implement the "one's
4401 complement" operation, which is the "~" operator in C.
4406 The two arguments to the '``xor``' instruction must be
4407 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4408 arguments must have identical types.
4413 The truth table used for the '``xor``' instruction is:
4430 .. code-block:: llvm
4432 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4433 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4434 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4435 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4440 LLVM supports several instructions to represent vector operations in a
4441 target-independent manner. These instructions cover the element-access
4442 and vector-specific operations needed to process vectors effectively.
4443 While LLVM does directly support these vector operations, many
4444 sophisticated algorithms will want to use target-specific intrinsics to
4445 take full advantage of a specific target.
4447 .. _i_extractelement:
4449 '``extractelement``' Instruction
4450 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4457 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4462 The '``extractelement``' instruction extracts a single scalar element
4463 from a vector at a specified index.
4468 The first operand of an '``extractelement``' instruction is a value of
4469 :ref:`vector <t_vector>` type. The second operand is an index indicating
4470 the position from which to extract the element. The index may be a
4476 The result is a scalar of the same type as the element type of ``val``.
4477 Its value is the value at position ``idx`` of ``val``. If ``idx``
4478 exceeds the length of ``val``, the results are undefined.
4483 .. code-block:: llvm
4485 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4487 .. _i_insertelement:
4489 '``insertelement``' Instruction
4490 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4497 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4502 The '``insertelement``' instruction inserts a scalar element into a
4503 vector at a specified index.
4508 The first operand of an '``insertelement``' instruction is a value of
4509 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4510 type must equal the element type of the first operand. The third operand
4511 is an index indicating the position at which to insert the value. The
4512 index may be a variable.
4517 The result is a vector of the same type as ``val``. Its element values
4518 are those of ``val`` except at position ``idx``, where it gets the value
4519 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4525 .. code-block:: llvm
4527 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4529 .. _i_shufflevector:
4531 '``shufflevector``' Instruction
4532 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4539 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4544 The '``shufflevector``' instruction constructs a permutation of elements
4545 from two input vectors, returning a vector with the same element type as
4546 the input and length that is the same as the shuffle mask.
4551 The first two operands of a '``shufflevector``' instruction are vectors
4552 with the same type. The third argument is a shuffle mask whose element
4553 type is always 'i32'. The result of the instruction is a vector whose
4554 length is the same as the shuffle mask and whose element type is the
4555 same as the element type of the first two operands.
4557 The shuffle mask operand is required to be a constant vector with either
4558 constant integer or undef values.
4563 The elements of the two input vectors are numbered from left to right
4564 across both of the vectors. The shuffle mask operand specifies, for each
4565 element of the result vector, which element of the two input vectors the
4566 result element gets. The element selector may be undef (meaning "don't
4567 care") and the second operand may be undef if performing a shuffle from
4573 .. code-block:: llvm
4575 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4576 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4577 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4578 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4579 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4580 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4581 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4582 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4584 Aggregate Operations
4585 --------------------
4587 LLVM supports several instructions for working with
4588 :ref:`aggregate <t_aggregate>` values.
4592 '``extractvalue``' Instruction
4593 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4600 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4605 The '``extractvalue``' instruction extracts the value of a member field
4606 from an :ref:`aggregate <t_aggregate>` value.
4611 The first operand of an '``extractvalue``' instruction is a value of
4612 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4613 constant indices to specify which value to extract in a similar manner
4614 as indices in a '``getelementptr``' instruction.
4616 The major differences to ``getelementptr`` indexing are:
4618 - Since the value being indexed is not a pointer, the first index is
4619 omitted and assumed to be zero.
4620 - At least one index must be specified.
4621 - Not only struct indices but also array indices must be in bounds.
4626 The result is the value at the position in the aggregate specified by
4632 .. code-block:: llvm
4634 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4638 '``insertvalue``' Instruction
4639 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4646 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4651 The '``insertvalue``' instruction inserts a value into a member field in
4652 an :ref:`aggregate <t_aggregate>` value.
4657 The first operand of an '``insertvalue``' instruction is a value of
4658 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4659 a first-class value to insert. The following operands are constant
4660 indices indicating the position at which to insert the value in a
4661 similar manner as indices in a '``extractvalue``' instruction. The value
4662 to insert must have the same type as the value identified by the
4668 The result is an aggregate of the same type as ``val``. Its value is
4669 that of ``val`` except that the value at the position specified by the
4670 indices is that of ``elt``.
4675 .. code-block:: llvm
4677 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4678 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4679 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4683 Memory Access and Addressing Operations
4684 ---------------------------------------
4686 A key design point of an SSA-based representation is how it represents
4687 memory. In LLVM, no memory locations are in SSA form, which makes things
4688 very simple. This section describes how to read, write, and allocate
4693 '``alloca``' Instruction
4694 ^^^^^^^^^^^^^^^^^^^^^^^^
4701 <result> = alloca <type>[, inalloca][, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4706 The '``alloca``' instruction allocates memory on the stack frame of the
4707 currently executing function, to be automatically released when this
4708 function returns to its caller. The object is always allocated in the
4709 generic address space (address space zero).
4714 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4715 bytes of memory on the runtime stack, returning a pointer of the
4716 appropriate type to the program. If "NumElements" is specified, it is
4717 the number of elements allocated, otherwise "NumElements" is defaulted
4718 to be one. If a constant alignment is specified, the value result of the
4719 allocation is guaranteed to be aligned to at least that boundary. If not
4720 specified, or if zero, the target can choose to align the allocation on
4721 any convenient boundary compatible with the type.
4723 '``type``' may be any sized type.
4728 Memory is allocated; a pointer is returned. The operation is undefined
4729 if there is insufficient stack space for the allocation. '``alloca``'d
4730 memory is automatically released when the function returns. The
4731 '``alloca``' instruction is commonly used to represent automatic
4732 variables that must have an address available. When the function returns
4733 (either with the ``ret`` or ``resume`` instructions), the memory is
4734 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4735 The order in which memory is allocated (ie., which way the stack grows)
4741 .. code-block:: llvm
4743 %ptr = alloca i32 ; yields {i32*}:ptr
4744 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4745 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4746 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4750 '``load``' Instruction
4751 ^^^^^^^^^^^^^^^^^^^^^^
4758 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4759 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4760 !<index> = !{ i32 1 }
4765 The '``load``' instruction is used to read from memory.
4770 The argument to the ``load`` instruction specifies the memory address
4771 from which to load. The pointer must point to a :ref:`first
4772 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4773 then the optimizer is not allowed to modify the number or order of
4774 execution of this ``load`` with other :ref:`volatile
4775 operations <volatile>`.
4777 If the ``load`` is marked as ``atomic``, it takes an extra
4778 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4779 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4780 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4781 when they may see multiple atomic stores. The type of the pointee must
4782 be an integer type whose bit width is a power of two greater than or
4783 equal to eight and less than or equal to a target-specific size limit.
4784 ``align`` must be explicitly specified on atomic loads, and the load has
4785 undefined behavior if the alignment is not set to a value which is at
4786 least the size in bytes of the pointee. ``!nontemporal`` does not have
4787 any defined semantics for atomic loads.
4789 The optional constant ``align`` argument specifies the alignment of the
4790 operation (that is, the alignment of the memory address). A value of 0
4791 or an omitted ``align`` argument means that the operation has the ABI
4792 alignment for the target. It is the responsibility of the code emitter
4793 to ensure that the alignment information is correct. Overestimating the
4794 alignment results in undefined behavior. Underestimating the alignment
4795 may produce less efficient code. An alignment of 1 is always safe.
4797 The optional ``!nontemporal`` metadata must reference a single
4798 metadata name ``<index>`` corresponding to a metadata node with one
4799 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4800 metadata on the instruction tells the optimizer and code generator
4801 that this load is not expected to be reused in the cache. The code
4802 generator may select special instructions to save cache bandwidth, such
4803 as the ``MOVNT`` instruction on x86.
4805 The optional ``!invariant.load`` metadata must reference a single
4806 metadata name ``<index>`` corresponding to a metadata node with no
4807 entries. The existence of the ``!invariant.load`` metadata on the
4808 instruction tells the optimizer and code generator that this load
4809 address points to memory which does not change value during program
4810 execution. The optimizer may then move this load around, for example, by
4811 hoisting it out of loops using loop invariant code motion.
4816 The location of memory pointed to is loaded. If the value being loaded
4817 is of scalar type then the number of bytes read does not exceed the
4818 minimum number of bytes needed to hold all bits of the type. For
4819 example, loading an ``i24`` reads at most three bytes. When loading a
4820 value of a type like ``i20`` with a size that is not an integral number
4821 of bytes, the result is undefined if the value was not originally
4822 written using a store of the same type.
4827 .. code-block:: llvm
4829 %ptr = alloca i32 ; yields {i32*}:ptr
4830 store i32 3, i32* %ptr ; yields {void}
4831 %val = load i32* %ptr ; yields {i32}:val = i32 3
4835 '``store``' Instruction
4836 ^^^^^^^^^^^^^^^^^^^^^^^
4843 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4844 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4849 The '``store``' instruction is used to write to memory.
4854 There are two arguments to the ``store`` instruction: a value to store
4855 and an address at which to store it. The type of the ``<pointer>``
4856 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4857 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4858 then the optimizer is not allowed to modify the number or order of
4859 execution of this ``store`` with other :ref:`volatile
4860 operations <volatile>`.
4862 If the ``store`` is marked as ``atomic``, it takes an extra
4863 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4864 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4865 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4866 when they may see multiple atomic stores. The type of the pointee must
4867 be an integer type whose bit width is a power of two greater than or
4868 equal to eight and less than or equal to a target-specific size limit.
4869 ``align`` must be explicitly specified on atomic stores, and the store
4870 has undefined behavior if the alignment is not set to a value which is
4871 at least the size in bytes of the pointee. ``!nontemporal`` does not
4872 have any defined semantics for atomic stores.
4874 The optional constant ``align`` argument specifies the alignment of the
4875 operation (that is, the alignment of the memory address). A value of 0
4876 or an omitted ``align`` argument means that the operation has the ABI
4877 alignment for the target. It is the responsibility of the code emitter
4878 to ensure that the alignment information is correct. Overestimating the
4879 alignment results in undefined behavior. Underestimating the
4880 alignment may produce less efficient code. An alignment of 1 is always
4883 The optional ``!nontemporal`` metadata must reference a single metadata
4884 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4885 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4886 tells the optimizer and code generator that this load is not expected to
4887 be reused in the cache. The code generator may select special
4888 instructions to save cache bandwidth, such as the MOVNT instruction on
4894 The contents of memory are updated to contain ``<value>`` at the
4895 location specified by the ``<pointer>`` operand. If ``<value>`` is
4896 of scalar type then the number of bytes written does not exceed the
4897 minimum number of bytes needed to hold all bits of the type. For
4898 example, storing an ``i24`` writes at most three bytes. When writing a
4899 value of a type like ``i20`` with a size that is not an integral number
4900 of bytes, it is unspecified what happens to the extra bits that do not
4901 belong to the type, but they will typically be overwritten.
4906 .. code-block:: llvm
4908 %ptr = alloca i32 ; yields {i32*}:ptr
4909 store i32 3, i32* %ptr ; yields {void}
4910 %val = load i32* %ptr ; yields {i32}:val = i32 3
4914 '``fence``' Instruction
4915 ^^^^^^^^^^^^^^^^^^^^^^^
4922 fence [singlethread] <ordering> ; yields {void}
4927 The '``fence``' instruction is used to introduce happens-before edges
4933 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4934 defines what *synchronizes-with* edges they add. They can only be given
4935 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4940 A fence A which has (at least) ``release`` ordering semantics
4941 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4942 semantics if and only if there exist atomic operations X and Y, both
4943 operating on some atomic object M, such that A is sequenced before X, X
4944 modifies M (either directly or through some side effect of a sequence
4945 headed by X), Y is sequenced before B, and Y observes M. This provides a
4946 *happens-before* dependency between A and B. Rather than an explicit
4947 ``fence``, one (but not both) of the atomic operations X or Y might
4948 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4949 still *synchronize-with* the explicit ``fence`` and establish the
4950 *happens-before* edge.
4952 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4953 ``acquire`` and ``release`` semantics specified above, participates in
4954 the global program order of other ``seq_cst`` operations and/or fences.
4956 The optional ":ref:`singlethread <singlethread>`" argument specifies
4957 that the fence only synchronizes with other fences in the same thread.
4958 (This is useful for interacting with signal handlers.)
4963 .. code-block:: llvm
4965 fence acquire ; yields {void}
4966 fence singlethread seq_cst ; yields {void}
4970 '``cmpxchg``' Instruction
4971 ^^^^^^^^^^^^^^^^^^^^^^^^^
4978 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4983 The '``cmpxchg``' instruction is used to atomically modify memory. It
4984 loads a value in memory and compares it to a given value. If they are
4985 equal, it stores a new value into the memory.
4990 There are three arguments to the '``cmpxchg``' instruction: an address
4991 to operate on, a value to compare to the value currently be at that
4992 address, and a new value to place at that address if the compared values
4993 are equal. The type of '<cmp>' must be an integer type whose bit width
4994 is a power of two greater than or equal to eight and less than or equal
4995 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4996 type, and the type of '<pointer>' must be a pointer to that type. If the
4997 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4998 to modify the number or order of execution of this ``cmpxchg`` with
4999 other :ref:`volatile operations <volatile>`.
5001 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
5002 synchronizes with other atomic operations.
5004 The optional "``singlethread``" argument declares that the ``cmpxchg``
5005 is only atomic with respect to code (usually signal handlers) running in
5006 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5007 respect to all other code in the system.
5009 The pointer passed into cmpxchg must have alignment greater than or
5010 equal to the size in memory of the operand.
5015 The contents of memory at the location specified by the '``<pointer>``'
5016 operand is read and compared to '``<cmp>``'; if the read value is the
5017 equal, '``<new>``' is written. The original value at the location is
5020 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
5021 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
5022 atomic load with an ordering parameter determined by dropping any
5023 ``release`` part of the ``cmpxchg``'s ordering.
5028 .. code-block:: llvm
5031 %orig = atomic load i32* %ptr unordered ; yields {i32}
5035 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5036 %squared = mul i32 %cmp, %cmp
5037 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
5038 %success = icmp eq i32 %cmp, %old
5039 br i1 %success, label %done, label %loop
5046 '``atomicrmw``' Instruction
5047 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5054 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
5059 The '``atomicrmw``' instruction is used to atomically modify memory.
5064 There are three arguments to the '``atomicrmw``' instruction: an
5065 operation to apply, an address whose value to modify, an argument to the
5066 operation. The operation must be one of the following keywords:
5080 The type of '<value>' must be an integer type whose bit width is a power
5081 of two greater than or equal to eight and less than or equal to a
5082 target-specific size limit. The type of the '``<pointer>``' operand must
5083 be a pointer to that type. If the ``atomicrmw`` is marked as
5084 ``volatile``, then the optimizer is not allowed to modify the number or
5085 order of execution of this ``atomicrmw`` with other :ref:`volatile
5086 operations <volatile>`.
5091 The contents of memory at the location specified by the '``<pointer>``'
5092 operand are atomically read, modified, and written back. The original
5093 value at the location is returned. The modification is specified by the
5096 - xchg: ``*ptr = val``
5097 - add: ``*ptr = *ptr + val``
5098 - sub: ``*ptr = *ptr - val``
5099 - and: ``*ptr = *ptr & val``
5100 - nand: ``*ptr = ~(*ptr & val)``
5101 - or: ``*ptr = *ptr | val``
5102 - xor: ``*ptr = *ptr ^ val``
5103 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5104 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5105 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5107 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5113 .. code-block:: llvm
5115 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5117 .. _i_getelementptr:
5119 '``getelementptr``' Instruction
5120 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5127 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5128 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5129 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5134 The '``getelementptr``' instruction is used to get the address of a
5135 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5136 address calculation only and does not access memory.
5141 The first argument is always a pointer or a vector of pointers, and
5142 forms the basis of the calculation. The remaining arguments are indices
5143 that indicate which of the elements of the aggregate object are indexed.
5144 The interpretation of each index is dependent on the type being indexed
5145 into. The first index always indexes the pointer value given as the
5146 first argument, the second index indexes a value of the type pointed to
5147 (not necessarily the value directly pointed to, since the first index
5148 can be non-zero), etc. The first type indexed into must be a pointer
5149 value, subsequent types can be arrays, vectors, and structs. Note that
5150 subsequent types being indexed into can never be pointers, since that
5151 would require loading the pointer before continuing calculation.
5153 The type of each index argument depends on the type it is indexing into.
5154 When indexing into a (optionally packed) structure, only ``i32`` integer
5155 **constants** are allowed (when using a vector of indices they must all
5156 be the **same** ``i32`` integer constant). When indexing into an array,
5157 pointer or vector, integers of any width are allowed, and they are not
5158 required to be constant. These integers are treated as signed values
5161 For example, let's consider a C code fragment and how it gets compiled
5177 int *foo(struct ST *s) {
5178 return &s[1].Z.B[5][13];
5181 The LLVM code generated by Clang is:
5183 .. code-block:: llvm
5185 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5186 %struct.ST = type { i32, double, %struct.RT }
5188 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5190 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5197 In the example above, the first index is indexing into the
5198 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5199 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5200 indexes into the third element of the structure, yielding a
5201 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5202 structure. The third index indexes into the second element of the
5203 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5204 dimensions of the array are subscripted into, yielding an '``i32``'
5205 type. The '``getelementptr``' instruction returns a pointer to this
5206 element, thus computing a value of '``i32*``' type.
5208 Note that it is perfectly legal to index partially through a structure,
5209 returning a pointer to an inner element. Because of this, the LLVM code
5210 for the given testcase is equivalent to:
5212 .. code-block:: llvm
5214 define i32* @foo(%struct.ST* %s) {
5215 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5216 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5217 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5218 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5219 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5223 If the ``inbounds`` keyword is present, the result value of the
5224 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5225 pointer is not an *in bounds* address of an allocated object, or if any
5226 of the addresses that would be formed by successive addition of the
5227 offsets implied by the indices to the base address with infinitely
5228 precise signed arithmetic are not an *in bounds* address of that
5229 allocated object. The *in bounds* addresses for an allocated object are
5230 all the addresses that point into the object, plus the address one byte
5231 past the end. In cases where the base is a vector of pointers the
5232 ``inbounds`` keyword applies to each of the computations element-wise.
5234 If the ``inbounds`` keyword is not present, the offsets are added to the
5235 base address with silently-wrapping two's complement arithmetic. If the
5236 offsets have a different width from the pointer, they are sign-extended
5237 or truncated to the width of the pointer. The result value of the
5238 ``getelementptr`` may be outside the object pointed to by the base
5239 pointer. The result value may not necessarily be used to access memory
5240 though, even if it happens to point into allocated storage. See the
5241 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5244 The getelementptr instruction is often confusing. For some more insight
5245 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5250 .. code-block:: llvm
5252 ; yields [12 x i8]*:aptr
5253 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5255 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5257 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5259 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5261 In cases where the pointer argument is a vector of pointers, each index
5262 must be a vector with the same number of elements. For example:
5264 .. code-block:: llvm
5266 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5268 Conversion Operations
5269 ---------------------
5271 The instructions in this category are the conversion instructions
5272 (casting) which all take a single operand and a type. They perform
5273 various bit conversions on the operand.
5275 '``trunc .. to``' Instruction
5276 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5283 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5288 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5293 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5294 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5295 of the same number of integers. The bit size of the ``value`` must be
5296 larger than the bit size of the destination type, ``ty2``. Equal sized
5297 types are not allowed.
5302 The '``trunc``' instruction truncates the high order bits in ``value``
5303 and converts the remaining bits to ``ty2``. Since the source size must
5304 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5305 It will always truncate bits.
5310 .. code-block:: llvm
5312 %X = trunc i32 257 to i8 ; yields i8:1
5313 %Y = trunc i32 123 to i1 ; yields i1:true
5314 %Z = trunc i32 122 to i1 ; yields i1:false
5315 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5317 '``zext .. to``' Instruction
5318 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5325 <result> = zext <ty> <value> to <ty2> ; yields ty2
5330 The '``zext``' instruction zero extends its operand to type ``ty2``.
5335 The '``zext``' instruction takes a value to cast, and a type to cast it
5336 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5337 the same number of integers. The bit size of the ``value`` must be
5338 smaller than the bit size of the destination type, ``ty2``.
5343 The ``zext`` fills the high order bits of the ``value`` with zero bits
5344 until it reaches the size of the destination type, ``ty2``.
5346 When zero extending from i1, the result will always be either 0 or 1.
5351 .. code-block:: llvm
5353 %X = zext i32 257 to i64 ; yields i64:257
5354 %Y = zext i1 true to i32 ; yields i32:1
5355 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5357 '``sext .. to``' Instruction
5358 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5365 <result> = sext <ty> <value> to <ty2> ; yields ty2
5370 The '``sext``' sign extends ``value`` to the type ``ty2``.
5375 The '``sext``' instruction takes a value to cast, and a type to cast it
5376 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5377 the same number of integers. The bit size of the ``value`` must be
5378 smaller than the bit size of the destination type, ``ty2``.
5383 The '``sext``' instruction performs a sign extension by copying the sign
5384 bit (highest order bit) of the ``value`` until it reaches the bit size
5385 of the type ``ty2``.
5387 When sign extending from i1, the extension always results in -1 or 0.
5392 .. code-block:: llvm
5394 %X = sext i8 -1 to i16 ; yields i16 :65535
5395 %Y = sext i1 true to i32 ; yields i32:-1
5396 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5398 '``fptrunc .. to``' Instruction
5399 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5406 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5411 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5416 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5417 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5418 The size of ``value`` must be larger than the size of ``ty2``. This
5419 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5424 The '``fptrunc``' instruction truncates a ``value`` from a larger
5425 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5426 point <t_floating>` type. If the value cannot fit within the
5427 destination type, ``ty2``, then the results are undefined.
5432 .. code-block:: llvm
5434 %X = fptrunc double 123.0 to float ; yields float:123.0
5435 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5437 '``fpext .. to``' Instruction
5438 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5445 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5450 The '``fpext``' extends a floating point ``value`` to a larger floating
5456 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5457 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5458 to. The source type must be smaller than the destination type.
5463 The '``fpext``' instruction extends the ``value`` from a smaller
5464 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5465 point <t_floating>` type. The ``fpext`` cannot be used to make a
5466 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5467 *no-op cast* for a floating point cast.
5472 .. code-block:: llvm
5474 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5475 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5477 '``fptoui .. to``' Instruction
5478 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5485 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5490 The '``fptoui``' converts a floating point ``value`` to its unsigned
5491 integer equivalent of type ``ty2``.
5496 The '``fptoui``' instruction takes a value to cast, which must be a
5497 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5498 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5499 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5500 type with the same number of elements as ``ty``
5505 The '``fptoui``' instruction converts its :ref:`floating
5506 point <t_floating>` operand into the nearest (rounding towards zero)
5507 unsigned integer value. If the value cannot fit in ``ty2``, the results
5513 .. code-block:: llvm
5515 %X = fptoui double 123.0 to i32 ; yields i32:123
5516 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5517 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5519 '``fptosi .. to``' Instruction
5520 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5527 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5532 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5533 ``value`` to type ``ty2``.
5538 The '``fptosi``' instruction takes a value to cast, which must be a
5539 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5540 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5541 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5542 type with the same number of elements as ``ty``
5547 The '``fptosi``' instruction converts its :ref:`floating
5548 point <t_floating>` operand into the nearest (rounding towards zero)
5549 signed integer value. If the value cannot fit in ``ty2``, the results
5555 .. code-block:: llvm
5557 %X = fptosi double -123.0 to i32 ; yields i32:-123
5558 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5559 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5561 '``uitofp .. to``' Instruction
5562 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5569 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5574 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5575 and converts that value to the ``ty2`` type.
5580 The '``uitofp``' instruction takes a value to cast, which must be a
5581 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5582 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5583 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5584 type with the same number of elements as ``ty``
5589 The '``uitofp``' instruction interprets its operand as an unsigned
5590 integer quantity and converts it to the corresponding floating point
5591 value. If the value cannot fit in the floating point value, the results
5597 .. code-block:: llvm
5599 %X = uitofp i32 257 to float ; yields float:257.0
5600 %Y = uitofp i8 -1 to double ; yields double:255.0
5602 '``sitofp .. to``' Instruction
5603 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5610 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5615 The '``sitofp``' instruction regards ``value`` as a signed integer and
5616 converts that value to the ``ty2`` type.
5621 The '``sitofp``' instruction takes a value to cast, which must be a
5622 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5623 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5624 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5625 type with the same number of elements as ``ty``
5630 The '``sitofp``' instruction interprets its operand as a signed integer
5631 quantity and converts it to the corresponding floating point value. If
5632 the value cannot fit in the floating point value, the results are
5638 .. code-block:: llvm
5640 %X = sitofp i32 257 to float ; yields float:257.0
5641 %Y = sitofp i8 -1 to double ; yields double:-1.0
5645 '``ptrtoint .. to``' Instruction
5646 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5653 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5658 The '``ptrtoint``' instruction converts the pointer or a vector of
5659 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5664 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5665 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5666 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5667 a vector of integers type.
5672 The '``ptrtoint``' instruction converts ``value`` to integer type
5673 ``ty2`` by interpreting the pointer value as an integer and either
5674 truncating or zero extending that value to the size of the integer type.
5675 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5676 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5677 the same size, then nothing is done (*no-op cast*) other than a type
5683 .. code-block:: llvm
5685 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5686 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5687 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5691 '``inttoptr .. to``' Instruction
5692 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5699 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5704 The '``inttoptr``' instruction converts an integer ``value`` to a
5705 pointer type, ``ty2``.
5710 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5711 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5717 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5718 applying either a zero extension or a truncation depending on the size
5719 of the integer ``value``. If ``value`` is larger than the size of a
5720 pointer then a truncation is done. If ``value`` is smaller than the size
5721 of a pointer then a zero extension is done. If they are the same size,
5722 nothing is done (*no-op cast*).
5727 .. code-block:: llvm
5729 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5730 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5731 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5732 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5736 '``bitcast .. to``' Instruction
5737 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5744 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5749 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5755 The '``bitcast``' instruction takes a value to cast, which must be a
5756 non-aggregate first class value, and a type to cast it to, which must
5757 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5758 bit sizes of ``value`` and the destination type, ``ty2``, must be
5759 identical. If the source type is a pointer, the destination type must
5760 also be a pointer of the same size. This instruction supports bitwise
5761 conversion of vectors to integers and to vectors of other types (as
5762 long as they have the same size).
5767 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5768 is always a *no-op cast* because no bits change with this
5769 conversion. The conversion is done as if the ``value`` had been stored
5770 to memory and read back as type ``ty2``. Pointer (or vector of
5771 pointers) types may only be converted to other pointer (or vector of
5772 pointers) types with the same address space through this instruction.
5773 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5774 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5779 .. code-block:: llvm
5781 %X = bitcast i8 255 to i8 ; yields i8 :-1
5782 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5783 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5784 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5786 .. _i_addrspacecast:
5788 '``addrspacecast .. to``' Instruction
5789 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5796 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5801 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5802 address space ``n`` to type ``pty2`` in address space ``m``.
5807 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5808 to cast and a pointer type to cast it to, which must have a different
5814 The '``addrspacecast``' instruction converts the pointer value
5815 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5816 value modification, depending on the target and the address space
5817 pair. Pointer conversions within the same address space must be
5818 performed with the ``bitcast`` instruction. Note that if the address space
5819 conversion is legal then both result and operand refer to the same memory
5825 .. code-block:: llvm
5827 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5828 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5829 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5836 The instructions in this category are the "miscellaneous" instructions,
5837 which defy better classification.
5841 '``icmp``' Instruction
5842 ^^^^^^^^^^^^^^^^^^^^^^
5849 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5854 The '``icmp``' instruction returns a boolean value or a vector of
5855 boolean values based on comparison of its two integer, integer vector,
5856 pointer, or pointer vector operands.
5861 The '``icmp``' instruction takes three operands. The first operand is
5862 the condition code indicating the kind of comparison to perform. It is
5863 not a value, just a keyword. The possible condition code are:
5866 #. ``ne``: not equal
5867 #. ``ugt``: unsigned greater than
5868 #. ``uge``: unsigned greater or equal
5869 #. ``ult``: unsigned less than
5870 #. ``ule``: unsigned less or equal
5871 #. ``sgt``: signed greater than
5872 #. ``sge``: signed greater or equal
5873 #. ``slt``: signed less than
5874 #. ``sle``: signed less or equal
5876 The remaining two arguments must be :ref:`integer <t_integer>` or
5877 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5878 must also be identical types.
5883 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5884 code given as ``cond``. The comparison performed always yields either an
5885 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5887 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5888 otherwise. No sign interpretation is necessary or performed.
5889 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5890 otherwise. No sign interpretation is necessary or performed.
5891 #. ``ugt``: interprets the operands as unsigned values and yields
5892 ``true`` if ``op1`` is greater than ``op2``.
5893 #. ``uge``: interprets the operands as unsigned values and yields
5894 ``true`` if ``op1`` is greater than or equal to ``op2``.
5895 #. ``ult``: interprets the operands as unsigned values and yields
5896 ``true`` if ``op1`` is less than ``op2``.
5897 #. ``ule``: interprets the operands as unsigned values and yields
5898 ``true`` if ``op1`` is less than or equal to ``op2``.
5899 #. ``sgt``: interprets the operands as signed values and yields ``true``
5900 if ``op1`` is greater than ``op2``.
5901 #. ``sge``: interprets the operands as signed values and yields ``true``
5902 if ``op1`` is greater than or equal to ``op2``.
5903 #. ``slt``: interprets the operands as signed values and yields ``true``
5904 if ``op1`` is less than ``op2``.
5905 #. ``sle``: interprets the operands as signed values and yields ``true``
5906 if ``op1`` is less than or equal to ``op2``.
5908 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5909 are compared as if they were integers.
5911 If the operands are integer vectors, then they are compared element by
5912 element. The result is an ``i1`` vector with the same number of elements
5913 as the values being compared. Otherwise, the result is an ``i1``.
5918 .. code-block:: llvm
5920 <result> = icmp eq i32 4, 5 ; yields: result=false
5921 <result> = icmp ne float* %X, %X ; yields: result=false
5922 <result> = icmp ult i16 4, 5 ; yields: result=true
5923 <result> = icmp sgt i16 4, 5 ; yields: result=false
5924 <result> = icmp ule i16 -4, 5 ; yields: result=false
5925 <result> = icmp sge i16 4, 5 ; yields: result=false
5927 Note that the code generator does not yet support vector types with the
5928 ``icmp`` instruction.
5932 '``fcmp``' Instruction
5933 ^^^^^^^^^^^^^^^^^^^^^^
5940 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5945 The '``fcmp``' instruction returns a boolean value or vector of boolean
5946 values based on comparison of its operands.
5948 If the operands are floating point scalars, then the result type is a
5949 boolean (:ref:`i1 <t_integer>`).
5951 If the operands are floating point vectors, then the result type is a
5952 vector of boolean with the same number of elements as the operands being
5958 The '``fcmp``' instruction takes three operands. The first operand is
5959 the condition code indicating the kind of comparison to perform. It is
5960 not a value, just a keyword. The possible condition code are:
5962 #. ``false``: no comparison, always returns false
5963 #. ``oeq``: ordered and equal
5964 #. ``ogt``: ordered and greater than
5965 #. ``oge``: ordered and greater than or equal
5966 #. ``olt``: ordered and less than
5967 #. ``ole``: ordered and less than or equal
5968 #. ``one``: ordered and not equal
5969 #. ``ord``: ordered (no nans)
5970 #. ``ueq``: unordered or equal
5971 #. ``ugt``: unordered or greater than
5972 #. ``uge``: unordered or greater than or equal
5973 #. ``ult``: unordered or less than
5974 #. ``ule``: unordered or less than or equal
5975 #. ``une``: unordered or not equal
5976 #. ``uno``: unordered (either nans)
5977 #. ``true``: no comparison, always returns true
5979 *Ordered* means that neither operand is a QNAN while *unordered* means
5980 that either operand may be a QNAN.
5982 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5983 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5984 type. They must have identical types.
5989 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5990 condition code given as ``cond``. If the operands are vectors, then the
5991 vectors are compared element by element. Each comparison performed
5992 always yields an :ref:`i1 <t_integer>` result, as follows:
5994 #. ``false``: always yields ``false``, regardless of operands.
5995 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5996 is equal to ``op2``.
5997 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5998 is greater than ``op2``.
5999 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6000 is greater than or equal to ``op2``.
6001 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6002 is less than ``op2``.
6003 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6004 is less than or equal to ``op2``.
6005 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6006 is not equal to ``op2``.
6007 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6008 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6010 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6011 greater than ``op2``.
6012 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6013 greater than or equal to ``op2``.
6014 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6016 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6017 less than or equal to ``op2``.
6018 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6019 not equal to ``op2``.
6020 #. ``uno``: yields ``true`` if either operand is a QNAN.
6021 #. ``true``: always yields ``true``, regardless of operands.
6026 .. code-block:: llvm
6028 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6029 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6030 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6031 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6033 Note that the code generator does not yet support vector types with the
6034 ``fcmp`` instruction.
6038 '``phi``' Instruction
6039 ^^^^^^^^^^^^^^^^^^^^^
6046 <result> = phi <ty> [ <val0>, <label0>], ...
6051 The '``phi``' instruction is used to implement the φ node in the SSA
6052 graph representing the function.
6057 The type of the incoming values is specified with the first type field.
6058 After this, the '``phi``' instruction takes a list of pairs as
6059 arguments, with one pair for each predecessor basic block of the current
6060 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6061 the value arguments to the PHI node. Only labels may be used as the
6064 There must be no non-phi instructions between the start of a basic block
6065 and the PHI instructions: i.e. PHI instructions must be first in a basic
6068 For the purposes of the SSA form, the use of each incoming value is
6069 deemed to occur on the edge from the corresponding predecessor block to
6070 the current block (but after any definition of an '``invoke``'
6071 instruction's return value on the same edge).
6076 At runtime, the '``phi``' instruction logically takes on the value
6077 specified by the pair corresponding to the predecessor basic block that
6078 executed just prior to the current block.
6083 .. code-block:: llvm
6085 Loop: ; Infinite loop that counts from 0 on up...
6086 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6087 %nextindvar = add i32 %indvar, 1
6092 '``select``' Instruction
6093 ^^^^^^^^^^^^^^^^^^^^^^^^
6100 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6102 selty is either i1 or {<N x i1>}
6107 The '``select``' instruction is used to choose one value based on a
6108 condition, without branching.
6113 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6114 values indicating the condition, and two values of the same :ref:`first
6115 class <t_firstclass>` type. If the val1/val2 are vectors and the
6116 condition is a scalar, then entire vectors are selected, not individual
6122 If the condition is an i1 and it evaluates to 1, the instruction returns
6123 the first value argument; otherwise, it returns the second value
6126 If the condition is a vector of i1, then the value arguments must be
6127 vectors of the same size, and the selection is done element by element.
6132 .. code-block:: llvm
6134 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6138 '``call``' Instruction
6139 ^^^^^^^^^^^^^^^^^^^^^^
6146 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6151 The '``call``' instruction represents a simple function call.
6156 This instruction requires several arguments:
6158 #. The optional "tail" marker indicates that the callee function does
6159 not access any allocas or varargs in the caller. Note that calls may
6160 be marked "tail" even if they do not occur before a
6161 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6162 function call is eligible for tail call optimization, but `might not
6163 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6164 The code generator may optimize calls marked "tail" with either 1)
6165 automatic `sibling call
6166 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6167 callee have matching signatures, or 2) forced tail call optimization
6168 when the following extra requirements are met:
6170 - Caller and callee both have the calling convention ``fastcc``.
6171 - The call is in tail position (ret immediately follows call and ret
6172 uses value of call or is void).
6173 - Option ``-tailcallopt`` is enabled, or
6174 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6175 - `Platform specific constraints are
6176 met. <CodeGenerator.html#tailcallopt>`_
6178 #. The optional "cconv" marker indicates which :ref:`calling
6179 convention <callingconv>` the call should use. If none is
6180 specified, the call defaults to using C calling conventions. The
6181 calling convention of the call must match the calling convention of
6182 the target function, or else the behavior is undefined.
6183 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6184 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6186 #. '``ty``': the type of the call instruction itself which is also the
6187 type of the return value. Functions that return no value are marked
6189 #. '``fnty``': shall be the signature of the pointer to function value
6190 being invoked. The argument types must match the types implied by
6191 this signature. This type can be omitted if the function is not
6192 varargs and if the function type does not return a pointer to a
6194 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6195 be invoked. In most cases, this is a direct function invocation, but
6196 indirect ``call``'s are just as possible, calling an arbitrary pointer
6198 #. '``function args``': argument list whose types match the function
6199 signature argument types and parameter attributes. All arguments must
6200 be of :ref:`first class <t_firstclass>` type. If the function signature
6201 indicates the function accepts a variable number of arguments, the
6202 extra arguments can be specified.
6203 #. The optional :ref:`function attributes <fnattrs>` list. Only
6204 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6205 attributes are valid here.
6210 The '``call``' instruction is used to cause control flow to transfer to
6211 a specified function, with its incoming arguments bound to the specified
6212 values. Upon a '``ret``' instruction in the called function, control
6213 flow continues with the instruction after the function call, and the
6214 return value of the function is bound to the result argument.
6219 .. code-block:: llvm
6221 %retval = call i32 @test(i32 %argc)
6222 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6223 %X = tail call i32 @foo() ; yields i32
6224 %Y = tail call fastcc i32 @foo() ; yields i32
6225 call void %foo(i8 97 signext)
6227 %struct.A = type { i32, i8 }
6228 %r = call %struct.A @foo() ; yields { 32, i8 }
6229 %gr = extractvalue %struct.A %r, 0 ; yields i32
6230 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6231 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6232 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6234 llvm treats calls to some functions with names and arguments that match
6235 the standard C99 library as being the C99 library functions, and may
6236 perform optimizations or generate code for them under that assumption.
6237 This is something we'd like to change in the future to provide better
6238 support for freestanding environments and non-C-based languages.
6242 '``va_arg``' Instruction
6243 ^^^^^^^^^^^^^^^^^^^^^^^^
6250 <resultval> = va_arg <va_list*> <arglist>, <argty>
6255 The '``va_arg``' instruction is used to access arguments passed through
6256 the "variable argument" area of a function call. It is used to implement
6257 the ``va_arg`` macro in C.
6262 This instruction takes a ``va_list*`` value and the type of the
6263 argument. It returns a value of the specified argument type and
6264 increments the ``va_list`` to point to the next argument. The actual
6265 type of ``va_list`` is target specific.
6270 The '``va_arg``' instruction loads an argument of the specified type
6271 from the specified ``va_list`` and causes the ``va_list`` to point to
6272 the next argument. For more information, see the variable argument
6273 handling :ref:`Intrinsic Functions <int_varargs>`.
6275 It is legal for this instruction to be called in a function which does
6276 not take a variable number of arguments, for example, the ``vfprintf``
6279 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6280 function <intrinsics>` because it takes a type as an argument.
6285 See the :ref:`variable argument processing <int_varargs>` section.
6287 Note that the code generator does not yet fully support va\_arg on many
6288 targets. Also, it does not currently support va\_arg with aggregate
6289 types on any target.
6293 '``landingpad``' Instruction
6294 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6301 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6302 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6304 <clause> := catch <type> <value>
6305 <clause> := filter <array constant type> <array constant>
6310 The '``landingpad``' instruction is used by `LLVM's exception handling
6311 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6312 is a landing pad --- one where the exception lands, and corresponds to the
6313 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6314 defines values supplied by the personality function (``pers_fn``) upon
6315 re-entry to the function. The ``resultval`` has the type ``resultty``.
6320 This instruction takes a ``pers_fn`` value. This is the personality
6321 function associated with the unwinding mechanism. The optional
6322 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6324 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6325 contains the global variable representing the "type" that may be caught
6326 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6327 clause takes an array constant as its argument. Use
6328 "``[0 x i8**] undef``" for a filter which cannot throw. The
6329 '``landingpad``' instruction must contain *at least* one ``clause`` or
6330 the ``cleanup`` flag.
6335 The '``landingpad``' instruction defines the values which are set by the
6336 personality function (``pers_fn``) upon re-entry to the function, and
6337 therefore the "result type" of the ``landingpad`` instruction. As with
6338 calling conventions, how the personality function results are
6339 represented in LLVM IR is target specific.
6341 The clauses are applied in order from top to bottom. If two
6342 ``landingpad`` instructions are merged together through inlining, the
6343 clauses from the calling function are appended to the list of clauses.
6344 When the call stack is being unwound due to an exception being thrown,
6345 the exception is compared against each ``clause`` in turn. If it doesn't
6346 match any of the clauses, and the ``cleanup`` flag is not set, then
6347 unwinding continues further up the call stack.
6349 The ``landingpad`` instruction has several restrictions:
6351 - A landing pad block is a basic block which is the unwind destination
6352 of an '``invoke``' instruction.
6353 - A landing pad block must have a '``landingpad``' instruction as its
6354 first non-PHI instruction.
6355 - There can be only one '``landingpad``' instruction within the landing
6357 - A basic block that is not a landing pad block may not include a
6358 '``landingpad``' instruction.
6359 - All '``landingpad``' instructions in a function must have the same
6360 personality function.
6365 .. code-block:: llvm
6367 ;; A landing pad which can catch an integer.
6368 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6370 ;; A landing pad that is a cleanup.
6371 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6373 ;; A landing pad which can catch an integer and can only throw a double.
6374 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6376 filter [1 x i8**] [@_ZTId]
6383 LLVM supports the notion of an "intrinsic function". These functions
6384 have well known names and semantics and are required to follow certain
6385 restrictions. Overall, these intrinsics represent an extension mechanism
6386 for the LLVM language that does not require changing all of the
6387 transformations in LLVM when adding to the language (or the bitcode
6388 reader/writer, the parser, etc...).
6390 Intrinsic function names must all start with an "``llvm.``" prefix. This
6391 prefix is reserved in LLVM for intrinsic names; thus, function names may
6392 not begin with this prefix. Intrinsic functions must always be external
6393 functions: you cannot define the body of intrinsic functions. Intrinsic
6394 functions may only be used in call or invoke instructions: it is illegal
6395 to take the address of an intrinsic function. Additionally, because
6396 intrinsic functions are part of the LLVM language, it is required if any
6397 are added that they be documented here.
6399 Some intrinsic functions can be overloaded, i.e., the intrinsic
6400 represents a family of functions that perform the same operation but on
6401 different data types. Because LLVM can represent over 8 million
6402 different integer types, overloading is used commonly to allow an
6403 intrinsic function to operate on any integer type. One or more of the
6404 argument types or the result type can be overloaded to accept any
6405 integer type. Argument types may also be defined as exactly matching a
6406 previous argument's type or the result type. This allows an intrinsic
6407 function which accepts multiple arguments, but needs all of them to be
6408 of the same type, to only be overloaded with respect to a single
6409 argument or the result.
6411 Overloaded intrinsics will have the names of its overloaded argument
6412 types encoded into its function name, each preceded by a period. Only
6413 those types which are overloaded result in a name suffix. Arguments
6414 whose type is matched against another type do not. For example, the
6415 ``llvm.ctpop`` function can take an integer of any width and returns an
6416 integer of exactly the same integer width. This leads to a family of
6417 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6418 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6419 overloaded, and only one type suffix is required. Because the argument's
6420 type is matched against the return type, it does not require its own
6423 To learn how to add an intrinsic function, please see the `Extending
6424 LLVM Guide <ExtendingLLVM.html>`_.
6428 Variable Argument Handling Intrinsics
6429 -------------------------------------
6431 Variable argument support is defined in LLVM with the
6432 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6433 functions. These functions are related to the similarly named macros
6434 defined in the ``<stdarg.h>`` header file.
6436 All of these functions operate on arguments that use a target-specific
6437 value type "``va_list``". The LLVM assembly language reference manual
6438 does not define what this type is, so all transformations should be
6439 prepared to handle these functions regardless of the type used.
6441 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6442 variable argument handling intrinsic functions are used.
6444 .. code-block:: llvm
6446 define i32 @test(i32 %X, ...) {
6447 ; Initialize variable argument processing
6449 %ap2 = bitcast i8** %ap to i8*
6450 call void @llvm.va_start(i8* %ap2)
6452 ; Read a single integer argument
6453 %tmp = va_arg i8** %ap, i32
6455 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6457 %aq2 = bitcast i8** %aq to i8*
6458 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6459 call void @llvm.va_end(i8* %aq2)
6461 ; Stop processing of arguments.
6462 call void @llvm.va_end(i8* %ap2)
6466 declare void @llvm.va_start(i8*)
6467 declare void @llvm.va_copy(i8*, i8*)
6468 declare void @llvm.va_end(i8*)
6472 '``llvm.va_start``' Intrinsic
6473 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6480 declare void @llvm.va_start(i8* <arglist>)
6485 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6486 subsequent use by ``va_arg``.
6491 The argument is a pointer to a ``va_list`` element to initialize.
6496 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6497 available in C. In a target-dependent way, it initializes the
6498 ``va_list`` element to which the argument points, so that the next call
6499 to ``va_arg`` will produce the first variable argument passed to the
6500 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6501 to know the last argument of the function as the compiler can figure
6504 '``llvm.va_end``' Intrinsic
6505 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6512 declare void @llvm.va_end(i8* <arglist>)
6517 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6518 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6523 The argument is a pointer to a ``va_list`` to destroy.
6528 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6529 available in C. In a target-dependent way, it destroys the ``va_list``
6530 element to which the argument points. Calls to
6531 :ref:`llvm.va_start <int_va_start>` and
6532 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6537 '``llvm.va_copy``' Intrinsic
6538 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6545 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6550 The '``llvm.va_copy``' intrinsic copies the current argument position
6551 from the source argument list to the destination argument list.
6556 The first argument is a pointer to a ``va_list`` element to initialize.
6557 The second argument is a pointer to a ``va_list`` element to copy from.
6562 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6563 available in C. In a target-dependent way, it copies the source
6564 ``va_list`` element into the destination ``va_list`` element. This
6565 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6566 arbitrarily complex and require, for example, memory allocation.
6568 Accurate Garbage Collection Intrinsics
6569 --------------------------------------
6571 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6572 (GC) requires the implementation and generation of these intrinsics.
6573 These intrinsics allow identification of :ref:`GC roots on the
6574 stack <int_gcroot>`, as well as garbage collector implementations that
6575 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6576 Front-ends for type-safe garbage collected languages should generate
6577 these intrinsics to make use of the LLVM garbage collectors. For more
6578 details, see `Accurate Garbage Collection with
6579 LLVM <GarbageCollection.html>`_.
6581 The garbage collection intrinsics only operate on objects in the generic
6582 address space (address space zero).
6586 '``llvm.gcroot``' Intrinsic
6587 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6594 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6599 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6600 the code generator, and allows some metadata to be associated with it.
6605 The first argument specifies the address of a stack object that contains
6606 the root pointer. The second pointer (which must be either a constant or
6607 a global value address) contains the meta-data to be associated with the
6613 At runtime, a call to this intrinsic stores a null pointer into the
6614 "ptrloc" location. At compile-time, the code generator generates
6615 information to allow the runtime to find the pointer at GC safe points.
6616 The '``llvm.gcroot``' intrinsic may only be used in a function which
6617 :ref:`specifies a GC algorithm <gc>`.
6621 '``llvm.gcread``' Intrinsic
6622 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6629 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6634 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6635 locations, allowing garbage collector implementations that require read
6641 The second argument is the address to read from, which should be an
6642 address allocated from the garbage collector. The first object is a
6643 pointer to the start of the referenced object, if needed by the language
6644 runtime (otherwise null).
6649 The '``llvm.gcread``' intrinsic has the same semantics as a load
6650 instruction, but may be replaced with substantially more complex code by
6651 the garbage collector runtime, as needed. The '``llvm.gcread``'
6652 intrinsic may only be used in a function which :ref:`specifies a GC
6657 '``llvm.gcwrite``' Intrinsic
6658 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6665 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6670 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6671 locations, allowing garbage collector implementations that require write
6672 barriers (such as generational or reference counting collectors).
6677 The first argument is the reference to store, the second is the start of
6678 the object to store it to, and the third is the address of the field of
6679 Obj to store to. If the runtime does not require a pointer to the
6680 object, Obj may be null.
6685 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6686 instruction, but may be replaced with substantially more complex code by
6687 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6688 intrinsic may only be used in a function which :ref:`specifies a GC
6691 Code Generator Intrinsics
6692 -------------------------
6694 These intrinsics are provided by LLVM to expose special features that
6695 may only be implemented with code generator support.
6697 '``llvm.returnaddress``' Intrinsic
6698 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6705 declare i8 *@llvm.returnaddress(i32 <level>)
6710 The '``llvm.returnaddress``' intrinsic attempts to compute a
6711 target-specific value indicating the return address of the current
6712 function or one of its callers.
6717 The argument to this intrinsic indicates which function to return the
6718 address for. Zero indicates the calling function, one indicates its
6719 caller, etc. The argument is **required** to be a constant integer
6725 The '``llvm.returnaddress``' intrinsic either returns a pointer
6726 indicating the return address of the specified call frame, or zero if it
6727 cannot be identified. The value returned by this intrinsic is likely to
6728 be incorrect or 0 for arguments other than zero, so it should only be
6729 used for debugging purposes.
6731 Note that calling this intrinsic does not prevent function inlining or
6732 other aggressive transformations, so the value returned may not be that
6733 of the obvious source-language caller.
6735 '``llvm.frameaddress``' Intrinsic
6736 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6743 declare i8* @llvm.frameaddress(i32 <level>)
6748 The '``llvm.frameaddress``' intrinsic attempts to return the
6749 target-specific frame pointer value for the specified stack frame.
6754 The argument to this intrinsic indicates which function to return the
6755 frame pointer for. Zero indicates the calling function, one indicates
6756 its caller, etc. The argument is **required** to be a constant integer
6762 The '``llvm.frameaddress``' intrinsic either returns a pointer
6763 indicating the frame address of the specified call frame, or zero if it
6764 cannot be identified. The value returned by this intrinsic is likely to
6765 be incorrect or 0 for arguments other than zero, so it should only be
6766 used for debugging purposes.
6768 Note that calling this intrinsic does not prevent function inlining or
6769 other aggressive transformations, so the value returned may not be that
6770 of the obvious source-language caller.
6774 '``llvm.stacksave``' Intrinsic
6775 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6782 declare i8* @llvm.stacksave()
6787 The '``llvm.stacksave``' intrinsic is used to remember the current state
6788 of the function stack, for use with
6789 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6790 implementing language features like scoped automatic variable sized
6796 This intrinsic returns a opaque pointer value that can be passed to
6797 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6798 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6799 ``llvm.stacksave``, it effectively restores the state of the stack to
6800 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6801 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6802 were allocated after the ``llvm.stacksave`` was executed.
6804 .. _int_stackrestore:
6806 '``llvm.stackrestore``' Intrinsic
6807 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6814 declare void @llvm.stackrestore(i8* %ptr)
6819 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6820 the function stack to the state it was in when the corresponding
6821 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6822 useful for implementing language features like scoped automatic variable
6823 sized arrays in C99.
6828 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6830 '``llvm.prefetch``' Intrinsic
6831 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6838 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6843 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6844 insert a prefetch instruction if supported; otherwise, it is a noop.
6845 Prefetches have no effect on the behavior of the program but can change
6846 its performance characteristics.
6851 ``address`` is the address to be prefetched, ``rw`` is the specifier
6852 determining if the fetch should be for a read (0) or write (1), and
6853 ``locality`` is a temporal locality specifier ranging from (0) - no
6854 locality, to (3) - extremely local keep in cache. The ``cache type``
6855 specifies whether the prefetch is performed on the data (1) or
6856 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6857 arguments must be constant integers.
6862 This intrinsic does not modify the behavior of the program. In
6863 particular, prefetches cannot trap and do not produce a value. On
6864 targets that support this intrinsic, the prefetch can provide hints to
6865 the processor cache for better performance.
6867 '``llvm.pcmarker``' Intrinsic
6868 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6875 declare void @llvm.pcmarker(i32 <id>)
6880 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6881 Counter (PC) in a region of code to simulators and other tools. The
6882 method is target specific, but it is expected that the marker will use
6883 exported symbols to transmit the PC of the marker. The marker makes no
6884 guarantees that it will remain with any specific instruction after
6885 optimizations. It is possible that the presence of a marker will inhibit
6886 optimizations. The intended use is to be inserted after optimizations to
6887 allow correlations of simulation runs.
6892 ``id`` is a numerical id identifying the marker.
6897 This intrinsic does not modify the behavior of the program. Backends
6898 that do not support this intrinsic may ignore it.
6900 '``llvm.readcyclecounter``' Intrinsic
6901 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6908 declare i64 @llvm.readcyclecounter()
6913 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6914 counter register (or similar low latency, high accuracy clocks) on those
6915 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6916 should map to RPCC. As the backing counters overflow quickly (on the
6917 order of 9 seconds on alpha), this should only be used for small
6923 When directly supported, reading the cycle counter should not modify any
6924 memory. Implementations are allowed to either return a application
6925 specific value or a system wide value. On backends without support, this
6926 is lowered to a constant 0.
6928 Note that runtime support may be conditional on the privilege-level code is
6929 running at and the host platform.
6931 Standard C Library Intrinsics
6932 -----------------------------
6934 LLVM provides intrinsics for a few important standard C library
6935 functions. These intrinsics allow source-language front-ends to pass
6936 information about the alignment of the pointer arguments to the code
6937 generator, providing opportunity for more efficient code generation.
6941 '``llvm.memcpy``' Intrinsic
6942 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6947 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6948 integer bit width and for different address spaces. Not all targets
6949 support all bit widths however.
6953 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6954 i32 <len>, i32 <align>, i1 <isvolatile>)
6955 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6956 i64 <len>, i32 <align>, i1 <isvolatile>)
6961 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6962 source location to the destination location.
6964 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6965 intrinsics do not return a value, takes extra alignment/isvolatile
6966 arguments and the pointers can be in specified address spaces.
6971 The first argument is a pointer to the destination, the second is a
6972 pointer to the source. The third argument is an integer argument
6973 specifying the number of bytes to copy, the fourth argument is the
6974 alignment of the source and destination locations, and the fifth is a
6975 boolean indicating a volatile access.
6977 If the call to this intrinsic has an alignment value that is not 0 or 1,
6978 then the caller guarantees that both the source and destination pointers
6979 are aligned to that boundary.
6981 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6982 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6983 very cleanly specified and it is unwise to depend on it.
6988 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6989 source location to the destination location, which are not allowed to
6990 overlap. It copies "len" bytes of memory over. If the argument is known
6991 to be aligned to some boundary, this can be specified as the fourth
6992 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
6994 '``llvm.memmove``' Intrinsic
6995 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7000 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7001 bit width and for different address space. Not all targets support all
7006 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7007 i32 <len>, i32 <align>, i1 <isvolatile>)
7008 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7009 i64 <len>, i32 <align>, i1 <isvolatile>)
7014 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7015 source location to the destination location. It is similar to the
7016 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7019 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7020 intrinsics do not return a value, takes extra alignment/isvolatile
7021 arguments and the pointers can be in specified address spaces.
7026 The first argument is a pointer to the destination, the second is a
7027 pointer to the source. The third argument is an integer argument
7028 specifying the number of bytes to copy, the fourth argument is the
7029 alignment of the source and destination locations, and the fifth is a
7030 boolean indicating a volatile access.
7032 If the call to this intrinsic has an alignment value that is not 0 or 1,
7033 then the caller guarantees that the source and destination pointers are
7034 aligned to that boundary.
7036 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7037 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7038 not very cleanly specified and it is unwise to depend on it.
7043 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7044 source location to the destination location, which may overlap. It
7045 copies "len" bytes of memory over. If the argument is known to be
7046 aligned to some boundary, this can be specified as the fourth argument,
7047 otherwise it should be set to 0 or 1 (both meaning no alignment).
7049 '``llvm.memset.*``' Intrinsics
7050 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7055 This is an overloaded intrinsic. You can use llvm.memset on any integer
7056 bit width and for different address spaces. However, not all targets
7057 support all bit widths.
7061 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7062 i32 <len>, i32 <align>, i1 <isvolatile>)
7063 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7064 i64 <len>, i32 <align>, i1 <isvolatile>)
7069 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7070 particular byte value.
7072 Note that, unlike the standard libc function, the ``llvm.memset``
7073 intrinsic does not return a value and takes extra alignment/volatile
7074 arguments. Also, the destination can be in an arbitrary address space.
7079 The first argument is a pointer to the destination to fill, the second
7080 is the byte value with which to fill it, the third argument is an
7081 integer argument specifying the number of bytes to fill, and the fourth
7082 argument is the known alignment of the destination location.
7084 If the call to this intrinsic has an alignment value that is not 0 or 1,
7085 then the caller guarantees that the destination pointer is aligned to
7088 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7089 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7090 very cleanly specified and it is unwise to depend on it.
7095 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7096 at the destination location. If the argument is known to be aligned to
7097 some boundary, this can be specified as the fourth argument, otherwise
7098 it should be set to 0 or 1 (both meaning no alignment).
7100 '``llvm.sqrt.*``' Intrinsic
7101 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7106 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7107 floating point or vector of floating point type. Not all targets support
7112 declare float @llvm.sqrt.f32(float %Val)
7113 declare double @llvm.sqrt.f64(double %Val)
7114 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7115 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7116 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7121 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7122 returning the same value as the libm '``sqrt``' functions would. Unlike
7123 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7124 negative numbers other than -0.0 (which allows for better optimization,
7125 because there is no need to worry about errno being set).
7126 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7131 The argument and return value are floating point numbers of the same
7137 This function returns the sqrt of the specified operand if it is a
7138 nonnegative floating point number.
7140 '``llvm.powi.*``' Intrinsic
7141 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7146 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7147 floating point or vector of floating point type. Not all targets support
7152 declare float @llvm.powi.f32(float %Val, i32 %power)
7153 declare double @llvm.powi.f64(double %Val, i32 %power)
7154 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7155 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7156 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7161 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7162 specified (positive or negative) power. The order of evaluation of
7163 multiplications is not defined. When a vector of floating point type is
7164 used, the second argument remains a scalar integer value.
7169 The second argument is an integer power, and the first is a value to
7170 raise to that power.
7175 This function returns the first value raised to the second power with an
7176 unspecified sequence of rounding operations.
7178 '``llvm.sin.*``' Intrinsic
7179 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7184 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7185 floating point or vector of floating point type. Not all targets support
7190 declare float @llvm.sin.f32(float %Val)
7191 declare double @llvm.sin.f64(double %Val)
7192 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7193 declare fp128 @llvm.sin.f128(fp128 %Val)
7194 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7199 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7204 The argument and return value are floating point numbers of the same
7210 This function returns the sine of the specified operand, returning the
7211 same values as the libm ``sin`` functions would, and handles error
7212 conditions in the same way.
7214 '``llvm.cos.*``' Intrinsic
7215 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7220 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7221 floating point or vector of floating point type. Not all targets support
7226 declare float @llvm.cos.f32(float %Val)
7227 declare double @llvm.cos.f64(double %Val)
7228 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7229 declare fp128 @llvm.cos.f128(fp128 %Val)
7230 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7235 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7240 The argument and return value are floating point numbers of the same
7246 This function returns the cosine of the specified operand, returning the
7247 same values as the libm ``cos`` functions would, and handles error
7248 conditions in the same way.
7250 '``llvm.pow.*``' Intrinsic
7251 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7256 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7257 floating point or vector of floating point type. Not all targets support
7262 declare float @llvm.pow.f32(float %Val, float %Power)
7263 declare double @llvm.pow.f64(double %Val, double %Power)
7264 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7265 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7266 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7271 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7272 specified (positive or negative) power.
7277 The second argument is a floating point power, and the first is a value
7278 to raise to that power.
7283 This function returns the first value raised to the second power,
7284 returning the same values as the libm ``pow`` functions would, and
7285 handles error conditions in the same way.
7287 '``llvm.exp.*``' Intrinsic
7288 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7293 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7294 floating point or vector of floating point type. Not all targets support
7299 declare float @llvm.exp.f32(float %Val)
7300 declare double @llvm.exp.f64(double %Val)
7301 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7302 declare fp128 @llvm.exp.f128(fp128 %Val)
7303 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7308 The '``llvm.exp.*``' intrinsics perform the exp function.
7313 The argument and return value are floating point numbers of the same
7319 This function returns the same values as the libm ``exp`` functions
7320 would, and handles error conditions in the same way.
7322 '``llvm.exp2.*``' Intrinsic
7323 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7328 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7329 floating point or vector of floating point type. Not all targets support
7334 declare float @llvm.exp2.f32(float %Val)
7335 declare double @llvm.exp2.f64(double %Val)
7336 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7337 declare fp128 @llvm.exp2.f128(fp128 %Val)
7338 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7343 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7348 The argument and return value are floating point numbers of the same
7354 This function returns the same values as the libm ``exp2`` functions
7355 would, and handles error conditions in the same way.
7357 '``llvm.log.*``' Intrinsic
7358 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7363 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7364 floating point or vector of floating point type. Not all targets support
7369 declare float @llvm.log.f32(float %Val)
7370 declare double @llvm.log.f64(double %Val)
7371 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7372 declare fp128 @llvm.log.f128(fp128 %Val)
7373 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7378 The '``llvm.log.*``' intrinsics perform the log function.
7383 The argument and return value are floating point numbers of the same
7389 This function returns the same values as the libm ``log`` functions
7390 would, and handles error conditions in the same way.
7392 '``llvm.log10.*``' Intrinsic
7393 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7398 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7399 floating point or vector of floating point type. Not all targets support
7404 declare float @llvm.log10.f32(float %Val)
7405 declare double @llvm.log10.f64(double %Val)
7406 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7407 declare fp128 @llvm.log10.f128(fp128 %Val)
7408 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7413 The '``llvm.log10.*``' intrinsics perform the log10 function.
7418 The argument and return value are floating point numbers of the same
7424 This function returns the same values as the libm ``log10`` functions
7425 would, and handles error conditions in the same way.
7427 '``llvm.log2.*``' Intrinsic
7428 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7433 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7434 floating point or vector of floating point type. Not all targets support
7439 declare float @llvm.log2.f32(float %Val)
7440 declare double @llvm.log2.f64(double %Val)
7441 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7442 declare fp128 @llvm.log2.f128(fp128 %Val)
7443 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7448 The '``llvm.log2.*``' intrinsics perform the log2 function.
7453 The argument and return value are floating point numbers of the same
7459 This function returns the same values as the libm ``log2`` functions
7460 would, and handles error conditions in the same way.
7462 '``llvm.fma.*``' Intrinsic
7463 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7468 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7469 floating point or vector of floating point type. Not all targets support
7474 declare float @llvm.fma.f32(float %a, float %b, float %c)
7475 declare double @llvm.fma.f64(double %a, double %b, double %c)
7476 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7477 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7478 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7483 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7489 The argument and return value are floating point numbers of the same
7495 This function returns the same values as the libm ``fma`` functions
7496 would, and does not set errno.
7498 '``llvm.fabs.*``' Intrinsic
7499 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7504 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7505 floating point or vector of floating point type. Not all targets support
7510 declare float @llvm.fabs.f32(float %Val)
7511 declare double @llvm.fabs.f64(double %Val)
7512 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7513 declare fp128 @llvm.fabs.f128(fp128 %Val)
7514 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7519 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7525 The argument and return value are floating point numbers of the same
7531 This function returns the same values as the libm ``fabs`` functions
7532 would, and handles error conditions in the same way.
7534 '``llvm.copysign.*``' Intrinsic
7535 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7540 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7541 floating point or vector of floating point type. Not all targets support
7546 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7547 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7548 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7549 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7550 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7555 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7556 first operand and the sign of the second operand.
7561 The arguments and return value are floating point numbers of the same
7567 This function returns the same values as the libm ``copysign``
7568 functions would, and handles error conditions in the same way.
7570 '``llvm.floor.*``' Intrinsic
7571 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7576 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7577 floating point or vector of floating point type. Not all targets support
7582 declare float @llvm.floor.f32(float %Val)
7583 declare double @llvm.floor.f64(double %Val)
7584 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7585 declare fp128 @llvm.floor.f128(fp128 %Val)
7586 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7591 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7596 The argument and return value are floating point numbers of the same
7602 This function returns the same values as the libm ``floor`` functions
7603 would, and handles error conditions in the same way.
7605 '``llvm.ceil.*``' Intrinsic
7606 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7611 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7612 floating point or vector of floating point type. Not all targets support
7617 declare float @llvm.ceil.f32(float %Val)
7618 declare double @llvm.ceil.f64(double %Val)
7619 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7620 declare fp128 @llvm.ceil.f128(fp128 %Val)
7621 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7626 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7631 The argument and return value are floating point numbers of the same
7637 This function returns the same values as the libm ``ceil`` functions
7638 would, and handles error conditions in the same way.
7640 '``llvm.trunc.*``' Intrinsic
7641 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7646 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7647 floating point or vector of floating point type. Not all targets support
7652 declare float @llvm.trunc.f32(float %Val)
7653 declare double @llvm.trunc.f64(double %Val)
7654 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7655 declare fp128 @llvm.trunc.f128(fp128 %Val)
7656 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7661 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7662 nearest integer not larger in magnitude than the operand.
7667 The argument and return value are floating point numbers of the same
7673 This function returns the same values as the libm ``trunc`` functions
7674 would, and handles error conditions in the same way.
7676 '``llvm.rint.*``' Intrinsic
7677 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7682 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7683 floating point or vector of floating point type. Not all targets support
7688 declare float @llvm.rint.f32(float %Val)
7689 declare double @llvm.rint.f64(double %Val)
7690 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7691 declare fp128 @llvm.rint.f128(fp128 %Val)
7692 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7697 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7698 nearest integer. It may raise an inexact floating-point exception if the
7699 operand isn't an integer.
7704 The argument and return value are floating point numbers of the same
7710 This function returns the same values as the libm ``rint`` functions
7711 would, and handles error conditions in the same way.
7713 '``llvm.nearbyint.*``' Intrinsic
7714 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7719 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7720 floating point or vector of floating point type. Not all targets support
7725 declare float @llvm.nearbyint.f32(float %Val)
7726 declare double @llvm.nearbyint.f64(double %Val)
7727 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7728 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7729 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7734 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7740 The argument and return value are floating point numbers of the same
7746 This function returns the same values as the libm ``nearbyint``
7747 functions would, and handles error conditions in the same way.
7749 '``llvm.round.*``' Intrinsic
7750 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7755 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7756 floating point or vector of floating point type. Not all targets support
7761 declare float @llvm.round.f32(float %Val)
7762 declare double @llvm.round.f64(double %Val)
7763 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7764 declare fp128 @llvm.round.f128(fp128 %Val)
7765 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7770 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7776 The argument and return value are floating point numbers of the same
7782 This function returns the same values as the libm ``round``
7783 functions would, and handles error conditions in the same way.
7785 Bit Manipulation Intrinsics
7786 ---------------------------
7788 LLVM provides intrinsics for a few important bit manipulation
7789 operations. These allow efficient code generation for some algorithms.
7791 '``llvm.bswap.*``' Intrinsics
7792 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7797 This is an overloaded intrinsic function. You can use bswap on any
7798 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7802 declare i16 @llvm.bswap.i16(i16 <id>)
7803 declare i32 @llvm.bswap.i32(i32 <id>)
7804 declare i64 @llvm.bswap.i64(i64 <id>)
7809 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7810 values with an even number of bytes (positive multiple of 16 bits).
7811 These are useful for performing operations on data that is not in the
7812 target's native byte order.
7817 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7818 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7819 intrinsic returns an i32 value that has the four bytes of the input i32
7820 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7821 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7822 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7823 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7826 '``llvm.ctpop.*``' Intrinsic
7827 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7832 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7833 bit width, or on any vector with integer elements. Not all targets
7834 support all bit widths or vector types, however.
7838 declare i8 @llvm.ctpop.i8(i8 <src>)
7839 declare i16 @llvm.ctpop.i16(i16 <src>)
7840 declare i32 @llvm.ctpop.i32(i32 <src>)
7841 declare i64 @llvm.ctpop.i64(i64 <src>)
7842 declare i256 @llvm.ctpop.i256(i256 <src>)
7843 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7848 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7854 The only argument is the value to be counted. The argument may be of any
7855 integer type, or a vector with integer elements. The return type must
7856 match the argument type.
7861 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7862 each element of a vector.
7864 '``llvm.ctlz.*``' Intrinsic
7865 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7870 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7871 integer bit width, or any vector whose elements are integers. Not all
7872 targets support all bit widths or vector types, however.
7876 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7877 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7878 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7879 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7880 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7881 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7886 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7887 leading zeros in a variable.
7892 The first argument is the value to be counted. This argument may be of
7893 any integer type, or a vectory with integer element type. The return
7894 type must match the first argument type.
7896 The second argument must be a constant and is a flag to indicate whether
7897 the intrinsic should ensure that a zero as the first argument produces a
7898 defined result. Historically some architectures did not provide a
7899 defined result for zero values as efficiently, and many algorithms are
7900 now predicated on avoiding zero-value inputs.
7905 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7906 zeros in a variable, or within each element of the vector. If
7907 ``src == 0`` then the result is the size in bits of the type of ``src``
7908 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7909 ``llvm.ctlz(i32 2) = 30``.
7911 '``llvm.cttz.*``' Intrinsic
7912 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7917 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7918 integer bit width, or any vector of integer elements. Not all targets
7919 support all bit widths or vector types, however.
7923 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7924 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7925 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7926 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7927 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7928 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7933 The '``llvm.cttz``' family of intrinsic functions counts the number of
7939 The first argument is the value to be counted. This argument may be of
7940 any integer type, or a vectory with integer element type. The return
7941 type must match the first argument type.
7943 The second argument must be a constant and is a flag to indicate whether
7944 the intrinsic should ensure that a zero as the first argument produces a
7945 defined result. Historically some architectures did not provide a
7946 defined result for zero values as efficiently, and many algorithms are
7947 now predicated on avoiding zero-value inputs.
7952 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7953 zeros in a variable, or within each element of a vector. If ``src == 0``
7954 then the result is the size in bits of the type of ``src`` if
7955 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7956 ``llvm.cttz(2) = 1``.
7958 Arithmetic with Overflow Intrinsics
7959 -----------------------------------
7961 LLVM provides intrinsics for some arithmetic with overflow operations.
7963 '``llvm.sadd.with.overflow.*``' Intrinsics
7964 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7969 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7970 on any integer bit width.
7974 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7975 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7976 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7981 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7982 a signed addition of the two arguments, and indicate whether an overflow
7983 occurred during the signed summation.
7988 The arguments (%a and %b) and the first element of the result structure
7989 may be of integer types of any bit width, but they must have the same
7990 bit width. The second element of the result structure must be of type
7991 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7997 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7998 a signed addition of the two variables. They return a structure --- the
7999 first element of which is the signed summation, and the second element
8000 of which is a bit specifying if the signed summation resulted in an
8006 .. code-block:: llvm
8008 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8009 %sum = extractvalue {i32, i1} %res, 0
8010 %obit = extractvalue {i32, i1} %res, 1
8011 br i1 %obit, label %overflow, label %normal
8013 '``llvm.uadd.with.overflow.*``' Intrinsics
8014 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8019 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8020 on any integer bit width.
8024 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8025 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8026 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8031 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8032 an unsigned addition of the two arguments, and indicate whether a carry
8033 occurred during the unsigned summation.
8038 The arguments (%a and %b) and the first element of the result structure
8039 may be of integer types of any bit width, but they must have the same
8040 bit width. The second element of the result structure must be of type
8041 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8047 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8048 an unsigned addition of the two arguments. They return a structure --- the
8049 first element of which is the sum, and the second element of which is a
8050 bit specifying if the unsigned summation resulted in a carry.
8055 .. code-block:: llvm
8057 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8058 %sum = extractvalue {i32, i1} %res, 0
8059 %obit = extractvalue {i32, i1} %res, 1
8060 br i1 %obit, label %carry, label %normal
8062 '``llvm.ssub.with.overflow.*``' Intrinsics
8063 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8068 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8069 on any integer bit width.
8073 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8074 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8075 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8080 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8081 a signed subtraction of the two arguments, and indicate whether an
8082 overflow occurred during the signed subtraction.
8087 The arguments (%a and %b) and the first element of the result structure
8088 may be of integer types of any bit width, but they must have the same
8089 bit width. The second element of the result structure must be of type
8090 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8096 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8097 a signed subtraction of the two arguments. They return a structure --- the
8098 first element of which is the subtraction, and the second element of
8099 which is a bit specifying if the signed subtraction resulted in an
8105 .. code-block:: llvm
8107 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8108 %sum = extractvalue {i32, i1} %res, 0
8109 %obit = extractvalue {i32, i1} %res, 1
8110 br i1 %obit, label %overflow, label %normal
8112 '``llvm.usub.with.overflow.*``' Intrinsics
8113 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8118 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8119 on any integer bit width.
8123 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8124 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8125 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8130 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8131 an unsigned subtraction of the two arguments, and indicate whether an
8132 overflow occurred during the unsigned subtraction.
8137 The arguments (%a and %b) and the first element of the result structure
8138 may be of integer types of any bit width, but they must have the same
8139 bit width. The second element of the result structure must be of type
8140 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8146 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8147 an unsigned subtraction of the two arguments. They return a structure ---
8148 the first element of which is the subtraction, and the second element of
8149 which is a bit specifying if the unsigned subtraction resulted in an
8155 .. code-block:: llvm
8157 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8158 %sum = extractvalue {i32, i1} %res, 0
8159 %obit = extractvalue {i32, i1} %res, 1
8160 br i1 %obit, label %overflow, label %normal
8162 '``llvm.smul.with.overflow.*``' Intrinsics
8163 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8168 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8169 on any integer bit width.
8173 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8174 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8175 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8180 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8181 a signed multiplication of the two arguments, and indicate whether an
8182 overflow occurred during the signed multiplication.
8187 The arguments (%a and %b) and the first element of the result structure
8188 may be of integer types of any bit width, but they must have the same
8189 bit width. The second element of the result structure must be of type
8190 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8196 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8197 a signed multiplication of the two arguments. They return a structure ---
8198 the first element of which is the multiplication, and the second element
8199 of which is a bit specifying if the signed multiplication resulted in an
8205 .. code-block:: llvm
8207 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8208 %sum = extractvalue {i32, i1} %res, 0
8209 %obit = extractvalue {i32, i1} %res, 1
8210 br i1 %obit, label %overflow, label %normal
8212 '``llvm.umul.with.overflow.*``' Intrinsics
8213 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8218 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8219 on any integer bit width.
8223 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8224 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8225 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8230 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8231 a unsigned multiplication of the two arguments, and indicate whether an
8232 overflow occurred during the unsigned multiplication.
8237 The arguments (%a and %b) and the first element of the result structure
8238 may be of integer types of any bit width, but they must have the same
8239 bit width. The second element of the result structure must be of type
8240 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8246 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8247 an unsigned multiplication of the two arguments. They return a structure ---
8248 the first element of which is the multiplication, and the second
8249 element of which is a bit specifying if the unsigned multiplication
8250 resulted in an overflow.
8255 .. code-block:: llvm
8257 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8258 %sum = extractvalue {i32, i1} %res, 0
8259 %obit = extractvalue {i32, i1} %res, 1
8260 br i1 %obit, label %overflow, label %normal
8262 Specialised Arithmetic Intrinsics
8263 ---------------------------------
8265 '``llvm.fmuladd.*``' Intrinsic
8266 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8273 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8274 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8279 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8280 expressions that can be fused if the code generator determines that (a) the
8281 target instruction set has support for a fused operation, and (b) that the
8282 fused operation is more efficient than the equivalent, separate pair of mul
8283 and add instructions.
8288 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8289 multiplicands, a and b, and an addend c.
8298 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8300 is equivalent to the expression a \* b + c, except that rounding will
8301 not be performed between the multiplication and addition steps if the
8302 code generator fuses the operations. Fusion is not guaranteed, even if
8303 the target platform supports it. If a fused multiply-add is required the
8304 corresponding llvm.fma.\* intrinsic function should be used
8305 instead. This never sets errno, just as '``llvm.fma.*``'.
8310 .. code-block:: llvm
8312 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8314 Half Precision Floating Point Intrinsics
8315 ----------------------------------------
8317 For most target platforms, half precision floating point is a
8318 storage-only format. This means that it is a dense encoding (in memory)
8319 but does not support computation in the format.
8321 This means that code must first load the half-precision floating point
8322 value as an i16, then convert it to float with
8323 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8324 then be performed on the float value (including extending to double
8325 etc). To store the value back to memory, it is first converted to float
8326 if needed, then converted to i16 with
8327 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8330 .. _int_convert_to_fp16:
8332 '``llvm.convert.to.fp16``' Intrinsic
8333 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8340 declare i16 @llvm.convert.to.fp16(f32 %a)
8345 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8346 from single precision floating point format to half precision floating
8352 The intrinsic function contains single argument - the value to be
8358 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8359 from single precision floating point format to half precision floating
8360 point format. The return value is an ``i16`` which contains the
8366 .. code-block:: llvm
8368 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8369 store i16 %res, i16* @x, align 2
8371 .. _int_convert_from_fp16:
8373 '``llvm.convert.from.fp16``' Intrinsic
8374 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8381 declare f32 @llvm.convert.from.fp16(i16 %a)
8386 The '``llvm.convert.from.fp16``' intrinsic function performs a
8387 conversion from half precision floating point format to single precision
8388 floating point format.
8393 The intrinsic function contains single argument - the value to be
8399 The '``llvm.convert.from.fp16``' intrinsic function performs a
8400 conversion from half single precision floating point format to single
8401 precision floating point format. The input half-float value is
8402 represented by an ``i16`` value.
8407 .. code-block:: llvm
8409 %a = load i16* @x, align 2
8410 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8415 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8416 prefix), are described in the `LLVM Source Level
8417 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8420 Exception Handling Intrinsics
8421 -----------------------------
8423 The LLVM exception handling intrinsics (which all start with
8424 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8425 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8429 Trampoline Intrinsics
8430 ---------------------
8432 These intrinsics make it possible to excise one parameter, marked with
8433 the :ref:`nest <nest>` attribute, from a function. The result is a
8434 callable function pointer lacking the nest parameter - the caller does
8435 not need to provide a value for it. Instead, the value to use is stored
8436 in advance in a "trampoline", a block of memory usually allocated on the
8437 stack, which also contains code to splice the nest value into the
8438 argument list. This is used to implement the GCC nested function address
8441 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8442 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8443 It can be created as follows:
8445 .. code-block:: llvm
8447 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8448 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8449 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8450 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8451 %fp = bitcast i8* %p to i32 (i32, i32)*
8453 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8454 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8458 '``llvm.init.trampoline``' Intrinsic
8459 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8466 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8471 This fills the memory pointed to by ``tramp`` with executable code,
8472 turning it into a trampoline.
8477 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8478 pointers. The ``tramp`` argument must point to a sufficiently large and
8479 sufficiently aligned block of memory; this memory is written to by the
8480 intrinsic. Note that the size and the alignment are target-specific -
8481 LLVM currently provides no portable way of determining them, so a
8482 front-end that generates this intrinsic needs to have some
8483 target-specific knowledge. The ``func`` argument must hold a function
8484 bitcast to an ``i8*``.
8489 The block of memory pointed to by ``tramp`` is filled with target
8490 dependent code, turning it into a function. Then ``tramp`` needs to be
8491 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8492 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8493 function's signature is the same as that of ``func`` with any arguments
8494 marked with the ``nest`` attribute removed. At most one such ``nest``
8495 argument is allowed, and it must be of pointer type. Calling the new
8496 function is equivalent to calling ``func`` with the same argument list,
8497 but with ``nval`` used for the missing ``nest`` argument. If, after
8498 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8499 modified, then the effect of any later call to the returned function
8500 pointer is undefined.
8504 '``llvm.adjust.trampoline``' Intrinsic
8505 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8512 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8517 This performs any required machine-specific adjustment to the address of
8518 a trampoline (passed as ``tramp``).
8523 ``tramp`` must point to a block of memory which already has trampoline
8524 code filled in by a previous call to
8525 :ref:`llvm.init.trampoline <int_it>`.
8530 On some architectures the address of the code to be executed needs to be
8531 different to the address where the trampoline is actually stored. This
8532 intrinsic returns the executable address corresponding to ``tramp``
8533 after performing the required machine specific adjustments. The pointer
8534 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8539 This class of intrinsics exists to information about the lifetime of
8540 memory objects and ranges where variables are immutable.
8544 '``llvm.lifetime.start``' Intrinsic
8545 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8552 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8557 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8563 The first argument is a constant integer representing the size of the
8564 object, or -1 if it is variable sized. The second argument is a pointer
8570 This intrinsic indicates that before this point in the code, the value
8571 of the memory pointed to by ``ptr`` is dead. This means that it is known
8572 to never be used and has an undefined value. A load from the pointer
8573 that precedes this intrinsic can be replaced with ``'undef'``.
8577 '``llvm.lifetime.end``' Intrinsic
8578 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8585 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8590 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8596 The first argument is a constant integer representing the size of the
8597 object, or -1 if it is variable sized. The second argument is a pointer
8603 This intrinsic indicates that after this point in the code, the value of
8604 the memory pointed to by ``ptr`` is dead. This means that it is known to
8605 never be used and has an undefined value. Any stores into the memory
8606 object following this intrinsic may be removed as dead.
8608 '``llvm.invariant.start``' Intrinsic
8609 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8616 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8621 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8622 a memory object will not change.
8627 The first argument is a constant integer representing the size of the
8628 object, or -1 if it is variable sized. The second argument is a pointer
8634 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8635 the return value, the referenced memory location is constant and
8638 '``llvm.invariant.end``' Intrinsic
8639 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8646 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8651 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8652 memory object are mutable.
8657 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8658 The second argument is a constant integer representing the size of the
8659 object, or -1 if it is variable sized and the third argument is a
8660 pointer to the object.
8665 This intrinsic indicates that the memory is mutable again.
8670 This class of intrinsics is designed to be generic and has no specific
8673 '``llvm.var.annotation``' Intrinsic
8674 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8681 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8686 The '``llvm.var.annotation``' intrinsic.
8691 The first argument is a pointer to a value, the second is a pointer to a
8692 global string, the third is a pointer to a global string which is the
8693 source file name, and the last argument is the line number.
8698 This intrinsic allows annotation of local variables with arbitrary
8699 strings. This can be useful for special purpose optimizations that want
8700 to look for these annotations. These have no other defined use; they are
8701 ignored by code generation and optimization.
8703 '``llvm.ptr.annotation.*``' Intrinsic
8704 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8709 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8710 pointer to an integer of any width. *NOTE* you must specify an address space for
8711 the pointer. The identifier for the default address space is the integer
8716 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8717 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8718 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8719 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8720 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8725 The '``llvm.ptr.annotation``' intrinsic.
8730 The first argument is a pointer to an integer value of arbitrary bitwidth
8731 (result of some expression), the second is a pointer to a global string, the
8732 third is a pointer to a global string which is the source file name, and the
8733 last argument is the line number. It returns the value of the first argument.
8738 This intrinsic allows annotation of a pointer to an integer with arbitrary
8739 strings. This can be useful for special purpose optimizations that want to look
8740 for these annotations. These have no other defined use; they are ignored by code
8741 generation and optimization.
8743 '``llvm.annotation.*``' Intrinsic
8744 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8749 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8750 any integer bit width.
8754 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8755 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8756 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8757 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8758 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8763 The '``llvm.annotation``' intrinsic.
8768 The first argument is an integer value (result of some expression), the
8769 second is a pointer to a global string, the third is a pointer to a
8770 global string which is the source file name, and the last argument is
8771 the line number. It returns the value of the first argument.
8776 This intrinsic allows annotations to be put on arbitrary expressions
8777 with arbitrary strings. This can be useful for special purpose
8778 optimizations that want to look for these annotations. These have no
8779 other defined use; they are ignored by code generation and optimization.
8781 '``llvm.trap``' Intrinsic
8782 ^^^^^^^^^^^^^^^^^^^^^^^^^
8789 declare void @llvm.trap() noreturn nounwind
8794 The '``llvm.trap``' intrinsic.
8804 This intrinsic is lowered to the target dependent trap instruction. If
8805 the target does not have a trap instruction, this intrinsic will be
8806 lowered to a call of the ``abort()`` function.
8808 '``llvm.debugtrap``' Intrinsic
8809 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8816 declare void @llvm.debugtrap() nounwind
8821 The '``llvm.debugtrap``' intrinsic.
8831 This intrinsic is lowered to code which is intended to cause an
8832 execution trap with the intention of requesting the attention of a
8835 '``llvm.stackprotector``' Intrinsic
8836 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8843 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8848 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8849 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8850 is placed on the stack before local variables.
8855 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8856 The first argument is the value loaded from the stack guard
8857 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8858 enough space to hold the value of the guard.
8863 This intrinsic causes the prologue/epilogue inserter to force the position of
8864 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8865 to ensure that if a local variable on the stack is overwritten, it will destroy
8866 the value of the guard. When the function exits, the guard on the stack is
8867 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8868 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8869 calling the ``__stack_chk_fail()`` function.
8871 '``llvm.stackprotectorcheck``' Intrinsic
8872 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8879 declare void @llvm.stackprotectorcheck(i8** <guard>)
8884 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8885 created stack protector and if they are not equal calls the
8886 ``__stack_chk_fail()`` function.
8891 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8892 the variable ``@__stack_chk_guard``.
8897 This intrinsic is provided to perform the stack protector check by comparing
8898 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8899 values do not match call the ``__stack_chk_fail()`` function.
8901 The reason to provide this as an IR level intrinsic instead of implementing it
8902 via other IR operations is that in order to perform this operation at the IR
8903 level without an intrinsic, one would need to create additional basic blocks to
8904 handle the success/failure cases. This makes it difficult to stop the stack
8905 protector check from disrupting sibling tail calls in Codegen. With this
8906 intrinsic, we are able to generate the stack protector basic blocks late in
8907 codegen after the tail call decision has occurred.
8909 '``llvm.objectsize``' Intrinsic
8910 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8917 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8918 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8923 The ``llvm.objectsize`` intrinsic is designed to provide information to
8924 the optimizers to determine at compile time whether a) an operation
8925 (like memcpy) will overflow a buffer that corresponds to an object, or
8926 b) that a runtime check for overflow isn't necessary. An object in this
8927 context means an allocation of a specific class, structure, array, or
8933 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8934 argument is a pointer to or into the ``object``. The second argument is
8935 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8936 or -1 (if false) when the object size is unknown. The second argument
8937 only accepts constants.
8942 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8943 the size of the object concerned. If the size cannot be determined at
8944 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8945 on the ``min`` argument).
8947 '``llvm.expect``' Intrinsic
8948 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8955 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8956 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8961 The ``llvm.expect`` intrinsic provides information about expected (the
8962 most probable) value of ``val``, which can be used by optimizers.
8967 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8968 a value. The second argument is an expected value, this needs to be a
8969 constant value, variables are not allowed.
8974 This intrinsic is lowered to the ``val``.
8976 '``llvm.donothing``' Intrinsic
8977 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8984 declare void @llvm.donothing() nounwind readnone
8989 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8990 only intrinsic that can be called with an invoke instruction.
9000 This intrinsic does nothing, and it's removed by optimizers and ignored
9003 Stack Map Intrinsics
9004 --------------------
9006 LLVM provides experimental intrinsics to support runtime patching
9007 mechanisms commonly desired in dynamic language JITs. These intrinsics
9008 are described in :doc:`StackMaps`.