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
371 - On X86-64 the callee preserves all general purpose registers, except for
372 R11. R11 can be used as a scratch register. Floating-point registers
373 (XMMs/YMMs) are not preserved and need to be saved by the caller.
375 The idea behind this convention is to support calls to runtime functions
376 that have a hot path and a cold path. The hot path is usually a small piece
377 of code that doesn't many registers. The cold path might need to call out to
378 another function and therefore only needs to preserve the caller-saved
379 registers, which haven't already been saved by the caller. The
380 `PreserveMost` calling convention is very similar to the `cold` calling
381 convention in terms of caller/callee-saved registers, but they are used for
382 different types of function calls. `coldcc` is for function calls that are
383 rarely executed, whereas `preserve_mostcc` function calls are intended to be
384 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
385 doesn't prevent the inliner from inlining the function call.
387 This calling convention will be used by a future version of the ObjectiveC
388 runtime and should therefore still be considered experimental at this time.
389 Although this convention was created to optimize certain runtime calls to
390 the ObjectiveC runtime, it is not limited to this runtime and might be used
391 by other runtimes in the future too. The current implementation only
392 supports X86-64, but the intention is to support more architectures in the
394 "``preserve_allcc``" - The `PreserveAll` calling convention
395 This calling convention attempts to make the code in the caller even less
396 intrusive than the `PreserveMost` calling convention. This calling
397 convention also behaves identical to the `C` calling convention on how
398 arguments and return values are passed, but it uses a different set of
399 caller/callee-saved registers. This removes the burden of saving and
400 recovering a large register set before and after the call in the caller.
402 - On X86-64 the callee preserves all general purpose registers, except for
403 R11. R11 can be used as a scratch register. Furthermore it also preserves
404 all floating-point registers (XMMs/YMMs).
406 The idea behind this convention is to support calls to runtime functions
407 that don't need to call out to any other functions.
409 This calling convention, like the `PreserveMost` calling convention, will be
410 used by a future version of the ObjectiveC runtime and should be considered
411 experimental at this time.
412 "``cc <n>``" - Numbered convention
413 Any calling convention may be specified by number, allowing
414 target-specific calling conventions to be used. Target specific
415 calling conventions start at 64.
417 More calling conventions can be added/defined on an as-needed basis, to
418 support Pascal conventions or any other well-known target-independent
421 .. _visibilitystyles:
426 All Global Variables and Functions have one of the following visibility
429 "``default``" - Default style
430 On targets that use the ELF object file format, default visibility
431 means that the declaration is visible to other modules and, in
432 shared libraries, means that the declared entity may be overridden.
433 On Darwin, default visibility means that the declaration is visible
434 to other modules. Default visibility corresponds to "external
435 linkage" in the language.
436 "``hidden``" - Hidden style
437 Two declarations of an object with hidden visibility refer to the
438 same object if they are in the same shared object. Usually, hidden
439 visibility indicates that the symbol will not be placed into the
440 dynamic symbol table, so no other module (executable or shared
441 library) can reference it directly.
442 "``protected``" - Protected style
443 On ELF, protected visibility indicates that the symbol will be
444 placed in the dynamic symbol table, but that references within the
445 defining module will bind to the local symbol. That is, the symbol
446 cannot be overridden by another module.
453 All Global Variables, Functions and Aliases can have one of the following
457 "``dllimport``" causes the compiler to reference a function or variable via
458 a global pointer to a pointer that is set up by the DLL exporting the
459 symbol. On Microsoft Windows targets, the pointer name is formed by
460 combining ``__imp_`` and the function or variable name.
462 "``dllexport``" causes the compiler to provide a global pointer to a pointer
463 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
464 Microsoft Windows targets, the pointer name is formed by combining
465 ``__imp_`` and the function or variable name. Since this storage class
466 exists for defining a dll interface, the compiler, assembler and linker know
467 it is externally referenced and must refrain from deleting the symbol.
472 LLVM IR allows you to specify name aliases for certain types. This can
473 make it easier to read the IR and make the IR more condensed
474 (particularly when recursive types are involved). An example of a name
479 %mytype = type { %mytype*, i32 }
481 You may give a name to any :ref:`type <typesystem>` except
482 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
483 expected with the syntax "%mytype".
485 Note that type names are aliases for the structural type that they
486 indicate, and that you can therefore specify multiple names for the same
487 type. This often leads to confusing behavior when dumping out a .ll
488 file. Since LLVM IR uses structural typing, the name is not part of the
489 type. When printing out LLVM IR, the printer will pick *one name* to
490 render all types of a particular shape. This means that if you have code
491 where two different source types end up having the same LLVM type, that
492 the dumper will sometimes print the "wrong" or unexpected type. This is
493 an important design point and isn't going to change.
500 Global variables define regions of memory allocated at compilation time
503 Global variables definitions must be initialized, may have an explicit section
504 to be placed in, and may have an optional explicit alignment specified.
506 Global variables in other translation units can also be declared, in which
507 case they don't have an initializer.
509 A variable may be defined as ``thread_local``, which means that it will
510 not be shared by threads (each thread will have a separated copy of the
511 variable). Not all targets support thread-local variables. Optionally, a
512 TLS model may be specified:
515 For variables that are only used within the current shared library.
517 For variables in modules that will not be loaded dynamically.
519 For variables defined in the executable and only used within it.
521 The models correspond to the ELF TLS models; see `ELF Handling For
522 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
523 more information on under which circumstances the different models may
524 be used. The target may choose a different TLS model if the specified
525 model is not supported, or if a better choice of model can be made.
527 A variable may be defined as a global ``constant``, which indicates that
528 the contents of the variable will **never** be modified (enabling better
529 optimization, allowing the global data to be placed in the read-only
530 section of an executable, etc). Note that variables that need runtime
531 initialization cannot be marked ``constant`` as there is a store to the
534 LLVM explicitly allows *declarations* of global variables to be marked
535 constant, even if the final definition of the global is not. This
536 capability can be used to enable slightly better optimization of the
537 program, but requires the language definition to guarantee that
538 optimizations based on the 'constantness' are valid for the translation
539 units that do not include the definition.
541 As SSA values, global variables define pointer values that are in scope
542 (i.e. they dominate) all basic blocks in the program. Global variables
543 always define a pointer to their "content" type because they describe a
544 region of memory, and all memory objects in LLVM are accessed through
547 Global variables can be marked with ``unnamed_addr`` which indicates
548 that the address is not significant, only the content. Constants marked
549 like this can be merged with other constants if they have the same
550 initializer. Note that a constant with significant address *can* be
551 merged with a ``unnamed_addr`` constant, the result being a constant
552 whose address is significant.
554 A global variable may be declared to reside in a target-specific
555 numbered address space. For targets that support them, address spaces
556 may affect how optimizations are performed and/or what target
557 instructions are used to access the variable. The default address space
558 is zero. The address space qualifier must precede any other attributes.
560 LLVM allows an explicit section to be specified for globals. If the
561 target supports it, it will emit globals to the section specified.
563 By default, global initializers are optimized by assuming that global
564 variables defined within the module are not modified from their
565 initial values before the start of the global initializer. This is
566 true even for variables potentially accessible from outside the
567 module, including those with external linkage or appearing in
568 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
569 by marking the variable with ``externally_initialized``.
571 An explicit alignment may be specified for a global, which must be a
572 power of 2. If not present, or if the alignment is set to zero, the
573 alignment of the global is set by the target to whatever it feels
574 convenient. If an explicit alignment is specified, the global is forced
575 to have exactly that alignment. Targets and optimizers are not allowed
576 to over-align the global if the global has an assigned section. In this
577 case, the extra alignment could be observable: for example, code could
578 assume that the globals are densely packed in their section and try to
579 iterate over them as an array, alignment padding would break this
582 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
586 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
587 [AddrSpace] [unnamed_addr] [ExternallyInitialized]
588 <global | constant> <Type>
589 [, section "name"] [, align <Alignment>]
591 For example, the following defines a global in a numbered address space
592 with an initializer, section, and alignment:
596 @G = addrspace(5) constant float 1.0, section "foo", align 4
598 The following example just declares a global variable
602 @G = external global i32
604 The following example defines a thread-local global with the
605 ``initialexec`` TLS model:
609 @G = thread_local(initialexec) global i32 0, align 4
611 .. _functionstructure:
616 LLVM function definitions consist of the "``define``" keyword, an
617 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
618 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
619 an optional :ref:`calling convention <callingconv>`,
620 an optional ``unnamed_addr`` attribute, a return type, an optional
621 :ref:`parameter attribute <paramattrs>` for the return type, a function
622 name, a (possibly empty) argument list (each with optional :ref:`parameter
623 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
624 an optional section, an optional alignment, an optional :ref:`garbage
625 collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
626 curly brace, a list of basic blocks, and a closing curly brace.
628 LLVM function declarations consist of the "``declare``" keyword, an
629 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
630 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
631 an optional :ref:`calling convention <callingconv>`,
632 an optional ``unnamed_addr`` attribute, a return type, an optional
633 :ref:`parameter attribute <paramattrs>` for the return type, a function
634 name, a possibly empty list of arguments, an optional alignment, an optional
635 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
637 A function definition contains a list of basic blocks, forming the CFG (Control
638 Flow Graph) for the function. Each basic block may optionally start with a label
639 (giving the basic block a symbol table entry), contains a list of instructions,
640 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
641 function return). If an explicit label is not provided, a block is assigned an
642 implicit numbered label, using the next value from the same counter as used for
643 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
644 entry block does not have an explicit label, it will be assigned label "%0",
645 then the first unnamed temporary in that block will be "%1", etc.
647 The first basic block in a function is special in two ways: it is
648 immediately executed on entrance to the function, and it is not allowed
649 to have predecessor basic blocks (i.e. there can not be any branches to
650 the entry block of a function). Because the block can have no
651 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
653 LLVM allows an explicit section to be specified for functions. If the
654 target supports it, it will emit functions to the section specified.
656 An explicit alignment may be specified for a function. If not present,
657 or if the alignment is set to zero, the alignment of the function is set
658 by the target to whatever it feels convenient. If an explicit alignment
659 is specified, the function is forced to have at least that much
660 alignment. All alignments must be a power of 2.
662 If the ``unnamed_addr`` attribute is given, the address is know to not
663 be significant and two identical functions can be merged.
667 define [linkage] [visibility] [DLLStorageClass]
669 <ResultType> @<FunctionName> ([argument list])
670 [fn Attrs] [section "name"] [align N]
671 [gc] [prefix Constant] { ... }
678 Aliases act as "second name" for the aliasee value (which can be either
679 function, global variable, another alias or bitcast of global value).
680 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
681 :ref:`visibility style <visibility>`, and an optional :ref:`DLL storage class
686 @<Name> = [Visibility] [DLLStorageClass] alias [Linkage] <AliaseeTy> @<Aliasee>
688 The linkage must be one of ``private``, ``linker_private``,
689 ``linker_private_weak``, ``internal``, ``linkonce``, ``weak``,
690 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
691 might not correctly handle dropping a weak symbol that is aliased by a non-weak
694 .. _namedmetadatastructure:
699 Named metadata is a collection of metadata. :ref:`Metadata
700 nodes <metadata>` (but not metadata strings) are the only valid
701 operands for a named metadata.
705 ; Some unnamed metadata nodes, which are referenced by the named metadata.
706 !0 = metadata !{metadata !"zero"}
707 !1 = metadata !{metadata !"one"}
708 !2 = metadata !{metadata !"two"}
710 !name = !{!0, !1, !2}
717 The return type and each parameter of a function type may have a set of
718 *parameter attributes* associated with them. Parameter attributes are
719 used to communicate additional information about the result or
720 parameters of a function. Parameter attributes are considered to be part
721 of the function, not of the function type, so functions with different
722 parameter attributes can have the same function type.
724 Parameter attributes are simple keywords that follow the type specified.
725 If multiple parameter attributes are needed, they are space separated.
730 declare i32 @printf(i8* noalias nocapture, ...)
731 declare i32 @atoi(i8 zeroext)
732 declare signext i8 @returns_signed_char()
734 Note that any attributes for the function result (``nounwind``,
735 ``readonly``) come immediately after the argument list.
737 Currently, only the following parameter attributes are defined:
740 This indicates to the code generator that the parameter or return
741 value should be zero-extended to the extent required by the target's
742 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
743 the caller (for a parameter) or the callee (for a return value).
745 This indicates to the code generator that the parameter or return
746 value should be sign-extended to the extent required by the target's
747 ABI (which is usually 32-bits) by the caller (for a parameter) or
748 the callee (for a return value).
750 This indicates that this parameter or return value should be treated
751 in a special target-dependent fashion during while emitting code for
752 a function call or return (usually, by putting it in a register as
753 opposed to memory, though some targets use it to distinguish between
754 two different kinds of registers). Use of this attribute is
757 This indicates that the pointer parameter should really be passed by
758 value to the function. The attribute implies that a hidden copy of
759 the pointee is made between the caller and the callee, so the callee
760 is unable to modify the value in the caller. This attribute is only
761 valid on LLVM pointer arguments. It is generally used to pass
762 structs and arrays by value, but is also valid on pointers to
763 scalars. The copy is considered to belong to the caller not the
764 callee (for example, ``readonly`` functions should not write to
765 ``byval`` parameters). This is not a valid attribute for return
768 The byval attribute also supports specifying an alignment with the
769 align attribute. It indicates the alignment of the stack slot to
770 form and the known alignment of the pointer specified to the call
771 site. If the alignment is not specified, then the code generator
772 makes a target-specific assumption.
778 .. Warning:: This feature is unstable and not fully implemented.
780 The ``inalloca`` argument attribute allows the caller to take the
781 address of outgoing stack arguments. An ``inalloca`` argument must
782 be a pointer to stack memory produced by an ``alloca`` instruction.
783 The alloca, or argument allocation, must also be tagged with the
784 inalloca keyword. Only the past argument may have the ``inalloca``
785 attribute, and that argument is guaranteed to be passed in memory.
787 An argument allocation may be used by a call at most once because
788 the call may deallocate it. The ``inalloca`` attribute cannot be
789 used in conjunction with other attributes that affect argument
790 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``.
792 When the call site is reached, the argument allocation must have
793 been the most recent stack allocation that is still live, or the
794 results are undefined. It is possible to allocate additional stack
795 space after an argument allocation and before its call site, but it
796 must be cleared off with :ref:`llvm.stackrestore
799 See :doc:`InAlloca` for more information on how to use this
803 This indicates that the pointer parameter specifies the address of a
804 structure that is the return value of the function in the source
805 program. This pointer must be guaranteed by the caller to be valid:
806 loads and stores to the structure may be assumed by the callee
807 not to trap and to be properly aligned. This may only be applied to
808 the first parameter. This is not a valid attribute for return
811 This indicates that pointer values :ref:`based <pointeraliasing>` on
812 the argument or return value do not alias pointer values which are
813 not *based* on it, ignoring certain "irrelevant" dependencies. For a
814 call to the parent function, dependencies between memory references
815 from before or after the call and from those during the call are
816 "irrelevant" to the ``noalias`` keyword for the arguments and return
817 value used in that call. The caller shares the responsibility with
818 the callee for ensuring that these requirements are met. For further
819 details, please see the discussion of the NoAlias response in `alias
820 analysis <AliasAnalysis.html#MustMayNo>`_.
822 Note that this definition of ``noalias`` is intentionally similar
823 to the definition of ``restrict`` in C99 for function arguments,
824 though it is slightly weaker.
826 For function return values, C99's ``restrict`` is not meaningful,
827 while LLVM's ``noalias`` is.
829 This indicates that the callee does not make any copies of the
830 pointer that outlive the callee itself. This is not a valid
831 attribute for return values.
836 This indicates that the pointer parameter can be excised using the
837 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
838 attribute for return values and can only be applied to one parameter.
841 This indicates that the function always returns the argument as its return
842 value. This is an optimization hint to the code generator when generating
843 the caller, allowing tail call optimization and omission of register saves
844 and restores in some cases; it is not checked or enforced when generating
845 the callee. The parameter and the function return type must be valid
846 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
847 valid attribute for return values and can only be applied to one parameter.
851 Garbage Collector Names
852 -----------------------
854 Each function may specify a garbage collector name, which is simply a
859 define void @f() gc "name" { ... }
861 The compiler declares the supported values of *name*. Specifying a
862 collector which will cause the compiler to alter its output in order to
863 support the named garbage collection algorithm.
870 Prefix data is data associated with a function which the code generator
871 will emit immediately before the function body. The purpose of this feature
872 is to allow frontends to associate language-specific runtime metadata with
873 specific functions and make it available through the function pointer while
874 still allowing the function pointer to be called. To access the data for a
875 given function, a program may bitcast the function pointer to a pointer to
876 the constant's type. This implies that the IR symbol points to the start
879 To maintain the semantics of ordinary function calls, the prefix data must
880 have a particular format. Specifically, it must begin with a sequence of
881 bytes which decode to a sequence of machine instructions, valid for the
882 module's target, which transfer control to the point immediately succeeding
883 the prefix data, without performing any other visible action. This allows
884 the inliner and other passes to reason about the semantics of the function
885 definition without needing to reason about the prefix data. Obviously this
886 makes the format of the prefix data highly target dependent.
888 Prefix data is laid out as if it were an initializer for a global variable
889 of the prefix data's type. No padding is automatically placed between the
890 prefix data and the function body. If padding is required, it must be part
893 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
894 which encodes the ``nop`` instruction:
898 define void @f() prefix i8 144 { ... }
900 Generally prefix data can be formed by encoding a relative branch instruction
901 which skips the metadata, as in this example of valid prefix data for the
902 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
906 %0 = type <{ i8, i8, i8* }>
908 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
910 A function may have prefix data but no body. This has similar semantics
911 to the ``available_externally`` linkage in that the data may be used by the
912 optimizers but will not be emitted in the object file.
919 Attribute groups are groups of attributes that are referenced by objects within
920 the IR. They are important for keeping ``.ll`` files readable, because a lot of
921 functions will use the same set of attributes. In the degenerative case of a
922 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
923 group will capture the important command line flags used to build that file.
925 An attribute group is a module-level object. To use an attribute group, an
926 object references the attribute group's ID (e.g. ``#37``). An object may refer
927 to more than one attribute group. In that situation, the attributes from the
928 different groups are merged.
930 Here is an example of attribute groups for a function that should always be
931 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
935 ; Target-independent attributes:
936 attributes #0 = { alwaysinline alignstack=4 }
938 ; Target-dependent attributes:
939 attributes #1 = { "no-sse" }
941 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
942 define void @f() #0 #1 { ... }
949 Function attributes are set to communicate additional information about
950 a function. Function attributes are considered to be part of the
951 function, not of the function type, so functions with different function
952 attributes can have the same function type.
954 Function attributes are simple keywords that follow the type specified.
955 If multiple attributes are needed, they are space separated. For
960 define void @f() noinline { ... }
961 define void @f() alwaysinline { ... }
962 define void @f() alwaysinline optsize { ... }
963 define void @f() optsize { ... }
966 This attribute indicates that, when emitting the prologue and
967 epilogue, the backend should forcibly align the stack pointer.
968 Specify the desired alignment, which must be a power of two, in
971 This attribute indicates that the inliner should attempt to inline
972 this function into callers whenever possible, ignoring any active
973 inlining size threshold for this caller.
975 This indicates that the callee function at a call site should be
976 recognized as a built-in function, even though the function's declaration
977 uses the ``nobuiltin`` attribute. This is only valid at call sites for
978 direct calls to functions which are declared with the ``nobuiltin``
981 This attribute indicates that this function is rarely called. When
982 computing edge weights, basic blocks post-dominated by a cold
983 function call are also considered to be cold; and, thus, given low
986 This attribute indicates that the source code contained a hint that
987 inlining this function is desirable (such as the "inline" keyword in
988 C/C++). It is just a hint; it imposes no requirements on the
991 This attribute suggests that optimization passes and code generator
992 passes make choices that keep the code size of this function as small
993 as possible and perform optimizations that may sacrifice runtime
994 performance in order to minimize the size of the generated code.
996 This attribute disables prologue / epilogue emission for the
997 function. This can have very system-specific consequences.
999 This indicates that the callee function at a call site is not recognized as
1000 a built-in function. LLVM will retain the original call and not replace it
1001 with equivalent code based on the semantics of the built-in function, unless
1002 the call site uses the ``builtin`` attribute. This is valid at call sites
1003 and on function declarations and definitions.
1005 This attribute indicates that calls to the function cannot be
1006 duplicated. A call to a ``noduplicate`` function may be moved
1007 within its parent function, but may not be duplicated within
1008 its parent function.
1010 A function containing a ``noduplicate`` call may still
1011 be an inlining candidate, provided that the call is not
1012 duplicated by inlining. That implies that the function has
1013 internal linkage and only has one call site, so the original
1014 call is dead after inlining.
1016 This attributes disables implicit floating point instructions.
1018 This attribute indicates that the inliner should never inline this
1019 function in any situation. This attribute may not be used together
1020 with the ``alwaysinline`` attribute.
1022 This attribute suppresses lazy symbol binding for the function. This
1023 may make calls to the function faster, at the cost of extra program
1024 startup time if the function is not called during program startup.
1026 This attribute indicates that the code generator should not use a
1027 red zone, even if the target-specific ABI normally permits it.
1029 This function attribute indicates that the function never returns
1030 normally. This produces undefined behavior at runtime if the
1031 function ever does dynamically return.
1033 This function attribute indicates that the function never returns
1034 with an unwind or exceptional control flow. If the function does
1035 unwind, its runtime behavior is undefined.
1037 This function attribute indicates that the function is not optimized
1038 by any optimization or code generator passes with the
1039 exception of interprocedural optimization passes.
1040 This attribute cannot be used together with the ``alwaysinline``
1041 attribute; this attribute is also incompatible
1042 with the ``minsize`` attribute and the ``optsize`` attribute.
1044 This attribute requires the ``noinline`` attribute to be specified on
1045 the function as well, so the function is never inlined into any caller.
1046 Only functions with the ``alwaysinline`` attribute are valid
1047 candidates for inlining into the body of this function.
1049 This attribute suggests that optimization passes and code generator
1050 passes make choices that keep the code size of this function low,
1051 and otherwise do optimizations specifically to reduce code size as
1052 long as they do not significantly impact runtime performance.
1054 On a function, this attribute indicates that the function computes its
1055 result (or decides to unwind an exception) based strictly on its arguments,
1056 without dereferencing any pointer arguments or otherwise accessing
1057 any mutable state (e.g. memory, control registers, etc) visible to
1058 caller functions. It does not write through any pointer arguments
1059 (including ``byval`` arguments) and never changes any state visible
1060 to callers. This means that it cannot unwind exceptions by calling
1061 the ``C++`` exception throwing methods.
1063 On an argument, this attribute indicates that the function does not
1064 dereference that pointer argument, even though it may read or write the
1065 memory that the pointer points to if accessed through other pointers.
1067 On a function, this attribute indicates that the function does not write
1068 through any pointer arguments (including ``byval`` arguments) or otherwise
1069 modify any state (e.g. memory, control registers, etc) visible to
1070 caller functions. It may dereference pointer arguments and read
1071 state that may be set in the caller. A readonly function always
1072 returns the same value (or unwinds an exception identically) when
1073 called with the same set of arguments and global state. It cannot
1074 unwind an exception by calling the ``C++`` exception throwing
1077 On an argument, this attribute indicates that the function does not write
1078 through this pointer argument, even though it may write to the memory that
1079 the pointer points to.
1081 This attribute indicates that this function can return twice. The C
1082 ``setjmp`` is an example of such a function. The compiler disables
1083 some optimizations (like tail calls) in the caller of these
1085 ``sanitize_address``
1086 This attribute indicates that AddressSanitizer checks
1087 (dynamic address safety analysis) are enabled for this function.
1089 This attribute indicates that MemorySanitizer checks (dynamic detection
1090 of accesses to uninitialized memory) are enabled for this function.
1092 This attribute indicates that ThreadSanitizer checks
1093 (dynamic thread safety analysis) are enabled for this function.
1095 This attribute indicates that the function should emit a stack
1096 smashing protector. It is in the form of a "canary" --- a random value
1097 placed on the stack before the local variables that's checked upon
1098 return from the function to see if it has been overwritten. A
1099 heuristic is used to determine if a function needs stack protectors
1100 or not. The heuristic used will enable protectors for functions with:
1102 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1103 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1104 - Calls to alloca() with variable sizes or constant sizes greater than
1105 ``ssp-buffer-size``.
1107 If a function that has an ``ssp`` attribute is inlined into a
1108 function that doesn't have an ``ssp`` attribute, then the resulting
1109 function will have an ``ssp`` attribute.
1111 This attribute indicates that the function should *always* emit a
1112 stack smashing protector. This overrides the ``ssp`` function
1115 If a function that has an ``sspreq`` attribute is inlined into a
1116 function that doesn't have an ``sspreq`` attribute or which has an
1117 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1118 an ``sspreq`` attribute.
1120 This attribute indicates that the function should emit a stack smashing
1121 protector. This attribute causes a strong heuristic to be used when
1122 determining if a function needs stack protectors. The strong heuristic
1123 will enable protectors for functions with:
1125 - Arrays of any size and type
1126 - Aggregates containing an array of any size and type.
1127 - Calls to alloca().
1128 - Local variables that have had their address taken.
1130 This overrides the ``ssp`` function attribute.
1132 If a function that has an ``sspstrong`` attribute is inlined into a
1133 function that doesn't have an ``sspstrong`` attribute, then the
1134 resulting function will have an ``sspstrong`` attribute.
1136 This attribute indicates that the ABI being targeted requires that
1137 an unwind table entry be produce for this function even if we can
1138 show that no exceptions passes by it. This is normally the case for
1139 the ELF x86-64 abi, but it can be disabled for some compilation
1144 Module-Level Inline Assembly
1145 ----------------------------
1147 Modules may contain "module-level inline asm" blocks, which corresponds
1148 to the GCC "file scope inline asm" blocks. These blocks are internally
1149 concatenated by LLVM and treated as a single unit, but may be separated
1150 in the ``.ll`` file if desired. The syntax is very simple:
1152 .. code-block:: llvm
1154 module asm "inline asm code goes here"
1155 module asm "more can go here"
1157 The strings can contain any character by escaping non-printable
1158 characters. The escape sequence used is simply "\\xx" where "xx" is the
1159 two digit hex code for the number.
1161 The inline asm code is simply printed to the machine code .s file when
1162 assembly code is generated.
1164 .. _langref_datalayout:
1169 A module may specify a target specific data layout string that specifies
1170 how data is to be laid out in memory. The syntax for the data layout is
1173 .. code-block:: llvm
1175 target datalayout = "layout specification"
1177 The *layout specification* consists of a list of specifications
1178 separated by the minus sign character ('-'). Each specification starts
1179 with a letter and may include other information after the letter to
1180 define some aspect of the data layout. The specifications accepted are
1184 Specifies that the target lays out data in big-endian form. That is,
1185 the bits with the most significance have the lowest address
1188 Specifies that the target lays out data in little-endian form. That
1189 is, the bits with the least significance have the lowest address
1192 Specifies the natural alignment of the stack in bits. Alignment
1193 promotion of stack variables is limited to the natural stack
1194 alignment to avoid dynamic stack realignment. The stack alignment
1195 must be a multiple of 8-bits. If omitted, the natural stack
1196 alignment defaults to "unspecified", which does not prevent any
1197 alignment promotions.
1198 ``p[n]:<size>:<abi>:<pref>``
1199 This specifies the *size* of a pointer and its ``<abi>`` and
1200 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1201 bits. The address space, ``n`` is optional, and if not specified,
1202 denotes the default address space 0. The value of ``n`` must be
1203 in the range [1,2^23).
1204 ``i<size>:<abi>:<pref>``
1205 This specifies the alignment for an integer type of a given bit
1206 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1207 ``v<size>:<abi>:<pref>``
1208 This specifies the alignment for a vector type of a given bit
1210 ``f<size>:<abi>:<pref>``
1211 This specifies the alignment for a floating point type of a given bit
1212 ``<size>``. Only values of ``<size>`` that are supported by the target
1213 will work. 32 (float) and 64 (double) are supported on all targets; 80
1214 or 128 (different flavors of long double) are also supported on some
1217 This specifies the alignment for an object of aggregate type.
1219 If present, specifies that llvm names are mangled in the output. The
1222 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1223 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1224 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1225 symbols get a ``_`` prefix.
1226 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1227 functions also get a suffix based on the frame size.
1228 ``n<size1>:<size2>:<size3>...``
1229 This specifies a set of native integer widths for the target CPU in
1230 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1231 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1232 this set are considered to support most general arithmetic operations
1235 On every specification that takes a ``<abi>:<pref>``, specifying the
1236 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1237 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1239 When constructing the data layout for a given target, LLVM starts with a
1240 default set of specifications which are then (possibly) overridden by
1241 the specifications in the ``datalayout`` keyword. The default
1242 specifications are given in this list:
1244 - ``E`` - big endian
1245 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1246 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1247 same as the default address space.
1248 - ``S0`` - natural stack alignment is unspecified
1249 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1250 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1251 - ``i16:16:16`` - i16 is 16-bit aligned
1252 - ``i32:32:32`` - i32 is 32-bit aligned
1253 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1254 alignment of 64-bits
1255 - ``f16:16:16`` - half is 16-bit aligned
1256 - ``f32:32:32`` - float is 32-bit aligned
1257 - ``f64:64:64`` - double is 64-bit aligned
1258 - ``f128:128:128`` - quad is 128-bit aligned
1259 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1260 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1261 - ``a:0:64`` - aggregates are 64-bit aligned
1263 When LLVM is determining the alignment for a given type, it uses the
1266 #. If the type sought is an exact match for one of the specifications,
1267 that specification is used.
1268 #. If no match is found, and the type sought is an integer type, then
1269 the smallest integer type that is larger than the bitwidth of the
1270 sought type is used. If none of the specifications are larger than
1271 the bitwidth then the largest integer type is used. For example,
1272 given the default specifications above, the i7 type will use the
1273 alignment of i8 (next largest) while both i65 and i256 will use the
1274 alignment of i64 (largest specified).
1275 #. If no match is found, and the type sought is a vector type, then the
1276 largest vector type that is smaller than the sought vector type will
1277 be used as a fall back. This happens because <128 x double> can be
1278 implemented in terms of 64 <2 x double>, for example.
1280 The function of the data layout string may not be what you expect.
1281 Notably, this is not a specification from the frontend of what alignment
1282 the code generator should use.
1284 Instead, if specified, the target data layout is required to match what
1285 the ultimate *code generator* expects. This string is used by the
1286 mid-level optimizers to improve code, and this only works if it matches
1287 what the ultimate code generator uses. If you would like to generate IR
1288 that does not embed this target-specific detail into the IR, then you
1289 don't have to specify the string. This will disable some optimizations
1290 that require precise layout information, but this also prevents those
1291 optimizations from introducing target specificity into the IR.
1298 A module may specify a target triple string that describes the target
1299 host. The syntax for the target triple is simply:
1301 .. code-block:: llvm
1303 target triple = "x86_64-apple-macosx10.7.0"
1305 The *target triple* string consists of a series of identifiers delimited
1306 by the minus sign character ('-'). The canonical forms are:
1310 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1311 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1313 This information is passed along to the backend so that it generates
1314 code for the proper architecture. It's possible to override this on the
1315 command line with the ``-mtriple`` command line option.
1317 .. _pointeraliasing:
1319 Pointer Aliasing Rules
1320 ----------------------
1322 Any memory access must be done through a pointer value associated with
1323 an address range of the memory access, otherwise the behavior is
1324 undefined. Pointer values are associated with address ranges according
1325 to the following rules:
1327 - A pointer value is associated with the addresses associated with any
1328 value it is *based* on.
1329 - An address of a global variable is associated with the address range
1330 of the variable's storage.
1331 - The result value of an allocation instruction is associated with the
1332 address range of the allocated storage.
1333 - A null pointer in the default address-space is associated with no
1335 - An integer constant other than zero or a pointer value returned from
1336 a function not defined within LLVM may be associated with address
1337 ranges allocated through mechanisms other than those provided by
1338 LLVM. Such ranges shall not overlap with any ranges of addresses
1339 allocated by mechanisms provided by LLVM.
1341 A pointer value is *based* on another pointer value according to the
1344 - A pointer value formed from a ``getelementptr`` operation is *based*
1345 on the first operand of the ``getelementptr``.
1346 - The result value of a ``bitcast`` is *based* on the operand of the
1348 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1349 values that contribute (directly or indirectly) to the computation of
1350 the pointer's value.
1351 - The "*based* on" relationship is transitive.
1353 Note that this definition of *"based"* is intentionally similar to the
1354 definition of *"based"* in C99, though it is slightly weaker.
1356 LLVM IR does not associate types with memory. The result type of a
1357 ``load`` merely indicates the size and alignment of the memory from
1358 which to load, as well as the interpretation of the value. The first
1359 operand type of a ``store`` similarly only indicates the size and
1360 alignment of the store.
1362 Consequently, type-based alias analysis, aka TBAA, aka
1363 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1364 :ref:`Metadata <metadata>` may be used to encode additional information
1365 which specialized optimization passes may use to implement type-based
1370 Volatile Memory Accesses
1371 ------------------------
1373 Certain memory accesses, such as :ref:`load <i_load>`'s,
1374 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1375 marked ``volatile``. The optimizers must not change the number of
1376 volatile operations or change their order of execution relative to other
1377 volatile operations. The optimizers *may* change the order of volatile
1378 operations relative to non-volatile operations. This is not Java's
1379 "volatile" and has no cross-thread synchronization behavior.
1381 IR-level volatile loads and stores cannot safely be optimized into
1382 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1383 flagged volatile. Likewise, the backend should never split or merge
1384 target-legal volatile load/store instructions.
1386 .. admonition:: Rationale
1388 Platforms may rely on volatile loads and stores of natively supported
1389 data width to be executed as single instruction. For example, in C
1390 this holds for an l-value of volatile primitive type with native
1391 hardware support, but not necessarily for aggregate types. The
1392 frontend upholds these expectations, which are intentionally
1393 unspecified in the IR. The rules above ensure that IR transformation
1394 do not violate the frontend's contract with the language.
1398 Memory Model for Concurrent Operations
1399 --------------------------------------
1401 The LLVM IR does not define any way to start parallel threads of
1402 execution or to register signal handlers. Nonetheless, there are
1403 platform-specific ways to create them, and we define LLVM IR's behavior
1404 in their presence. This model is inspired by the C++0x memory model.
1406 For a more informal introduction to this model, see the :doc:`Atomics`.
1408 We define a *happens-before* partial order as the least partial order
1411 - Is a superset of single-thread program order, and
1412 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1413 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1414 techniques, like pthread locks, thread creation, thread joining,
1415 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1416 Constraints <ordering>`).
1418 Note that program order does not introduce *happens-before* edges
1419 between a thread and signals executing inside that thread.
1421 Every (defined) read operation (load instructions, memcpy, atomic
1422 loads/read-modify-writes, etc.) R reads a series of bytes written by
1423 (defined) write operations (store instructions, atomic
1424 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1425 section, initialized globals are considered to have a write of the
1426 initializer which is atomic and happens before any other read or write
1427 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1428 may see any write to the same byte, except:
1430 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1431 write\ :sub:`2` happens before R\ :sub:`byte`, then
1432 R\ :sub:`byte` does not see write\ :sub:`1`.
1433 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1434 R\ :sub:`byte` does not see write\ :sub:`3`.
1436 Given that definition, R\ :sub:`byte` is defined as follows:
1438 - If R is volatile, the result is target-dependent. (Volatile is
1439 supposed to give guarantees which can support ``sig_atomic_t`` in
1440 C/C++, and may be used for accesses to addresses which do not behave
1441 like normal memory. It does not generally provide cross-thread
1443 - Otherwise, if there is no write to the same byte that happens before
1444 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1445 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1446 R\ :sub:`byte` returns the value written by that write.
1447 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1448 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1449 Memory Ordering Constraints <ordering>` section for additional
1450 constraints on how the choice is made.
1451 - Otherwise R\ :sub:`byte` returns ``undef``.
1453 R returns the value composed of the series of bytes it read. This
1454 implies that some bytes within the value may be ``undef`` **without**
1455 the entire value being ``undef``. Note that this only defines the
1456 semantics of the operation; it doesn't mean that targets will emit more
1457 than one instruction to read the series of bytes.
1459 Note that in cases where none of the atomic intrinsics are used, this
1460 model places only one restriction on IR transformations on top of what
1461 is required for single-threaded execution: introducing a store to a byte
1462 which might not otherwise be stored is not allowed in general.
1463 (Specifically, in the case where another thread might write to and read
1464 from an address, introducing a store can change a load that may see
1465 exactly one write into a load that may see multiple writes.)
1469 Atomic Memory Ordering Constraints
1470 ----------------------------------
1472 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1473 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1474 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1475 an ordering parameter that determines which other atomic instructions on
1476 the same address they *synchronize with*. These semantics are borrowed
1477 from Java and C++0x, but are somewhat more colloquial. If these
1478 descriptions aren't precise enough, check those specs (see spec
1479 references in the :doc:`atomics guide <Atomics>`).
1480 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1481 differently since they don't take an address. See that instruction's
1482 documentation for details.
1484 For a simpler introduction to the ordering constraints, see the
1488 The set of values that can be read is governed by the happens-before
1489 partial order. A value cannot be read unless some operation wrote
1490 it. This is intended to provide a guarantee strong enough to model
1491 Java's non-volatile shared variables. This ordering cannot be
1492 specified for read-modify-write operations; it is not strong enough
1493 to make them atomic in any interesting way.
1495 In addition to the guarantees of ``unordered``, there is a single
1496 total order for modifications by ``monotonic`` operations on each
1497 address. All modification orders must be compatible with the
1498 happens-before order. There is no guarantee that the modification
1499 orders can be combined to a global total order for the whole program
1500 (and this often will not be possible). The read in an atomic
1501 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1502 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1503 order immediately before the value it writes. If one atomic read
1504 happens before another atomic read of the same address, the later
1505 read must see the same value or a later value in the address's
1506 modification order. This disallows reordering of ``monotonic`` (or
1507 stronger) operations on the same address. If an address is written
1508 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1509 read that address repeatedly, the other threads must eventually see
1510 the write. This corresponds to the C++0x/C1x
1511 ``memory_order_relaxed``.
1513 In addition to the guarantees of ``monotonic``, a
1514 *synchronizes-with* edge may be formed with a ``release`` operation.
1515 This is intended to model C++'s ``memory_order_acquire``.
1517 In addition to the guarantees of ``monotonic``, if this operation
1518 writes a value which is subsequently read by an ``acquire``
1519 operation, it *synchronizes-with* that operation. (This isn't a
1520 complete description; see the C++0x definition of a release
1521 sequence.) This corresponds to the C++0x/C1x
1522 ``memory_order_release``.
1523 ``acq_rel`` (acquire+release)
1524 Acts as both an ``acquire`` and ``release`` operation on its
1525 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1526 ``seq_cst`` (sequentially consistent)
1527 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1528 operation which only reads, ``release`` for an operation which only
1529 writes), there is a global total order on all
1530 sequentially-consistent operations on all addresses, which is
1531 consistent with the *happens-before* partial order and with the
1532 modification orders of all the affected addresses. Each
1533 sequentially-consistent read sees the last preceding write to the
1534 same address in this global order. This corresponds to the C++0x/C1x
1535 ``memory_order_seq_cst`` and Java volatile.
1539 If an atomic operation is marked ``singlethread``, it only *synchronizes
1540 with* or participates in modification and seq\_cst total orderings with
1541 other operations running in the same thread (for example, in signal
1549 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1550 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1551 :ref:`frem <i_frem>`) have the following flags that can set to enable
1552 otherwise unsafe floating point operations
1555 No NaNs - Allow optimizations to assume the arguments and result are not
1556 NaN. Such optimizations are required to retain defined behavior over
1557 NaNs, but the value of the result is undefined.
1560 No Infs - Allow optimizations to assume the arguments and result are not
1561 +/-Inf. Such optimizations are required to retain defined behavior over
1562 +/-Inf, but the value of the result is undefined.
1565 No Signed Zeros - Allow optimizations to treat the sign of a zero
1566 argument or result as insignificant.
1569 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1570 argument rather than perform division.
1573 Fast - Allow algebraically equivalent transformations that may
1574 dramatically change results in floating point (e.g. reassociate). This
1575 flag implies all the others.
1582 The LLVM type system is one of the most important features of the
1583 intermediate representation. Being typed enables a number of
1584 optimizations to be performed on the intermediate representation
1585 directly, without having to do extra analyses on the side before the
1586 transformation. A strong type system makes it easier to read the
1587 generated code and enables novel analyses and transformations that are
1588 not feasible to perform on normal three address code representations.
1598 The void type does not represent any value and has no size.
1616 The function type can be thought of as a function signature. It consists of a
1617 return type and a list of formal parameter types. The return type of a function
1618 type is a void type or first class type --- except for :ref:`label <t_label>`
1619 and :ref:`metadata <t_metadata>` types.
1625 <returntype> (<parameter list>)
1627 ...where '``<parameter list>``' is a comma-separated list of type
1628 specifiers. Optionally, the parameter list may include a type ``...``, which
1629 indicates that the function takes a variable number of arguments. Variable
1630 argument functions can access their arguments with the :ref:`variable argument
1631 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1632 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1636 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1637 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1638 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1639 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1640 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1641 | ``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. |
1642 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1643 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1644 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1651 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1652 Values of these types are the only ones which can be produced by
1660 These are the types that are valid in registers from CodeGen's perspective.
1669 The integer type is a very simple type that simply specifies an
1670 arbitrary bit width for the integer type desired. Any bit width from 1
1671 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1679 The number of bits the integer will occupy is specified by the ``N``
1685 +----------------+------------------------------------------------+
1686 | ``i1`` | a single-bit integer. |
1687 +----------------+------------------------------------------------+
1688 | ``i32`` | a 32-bit integer. |
1689 +----------------+------------------------------------------------+
1690 | ``i1942652`` | a really big integer of over 1 million bits. |
1691 +----------------+------------------------------------------------+
1695 Floating Point Types
1696 """"""""""""""""""""
1705 - 16-bit floating point value
1708 - 32-bit floating point value
1711 - 64-bit floating point value
1714 - 128-bit floating point value (112-bit mantissa)
1717 - 80-bit floating point value (X87)
1720 - 128-bit floating point value (two 64-bits)
1729 The x86mmx type represents a value held in an MMX register on an x86
1730 machine. The operations allowed on it are quite limited: parameters and
1731 return values, load and store, and bitcast. User-specified MMX
1732 instructions are represented as intrinsic or asm calls with arguments
1733 and/or results of this type. There are no arrays, vectors or constants
1750 The pointer type is used to specify memory locations. Pointers are
1751 commonly used to reference objects in memory.
1753 Pointer types may have an optional address space attribute defining the
1754 numbered address space where the pointed-to object resides. The default
1755 address space is number zero. The semantics of non-zero address spaces
1756 are target-specific.
1758 Note that LLVM does not permit pointers to void (``void*``) nor does it
1759 permit pointers to labels (``label*``). Use ``i8*`` instead.
1769 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1770 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1771 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1772 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1773 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1774 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1775 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1784 A vector type is a simple derived type that represents a vector of
1785 elements. Vector types are used when multiple primitive data are
1786 operated in parallel using a single instruction (SIMD). A vector type
1787 requires a size (number of elements) and an underlying primitive data
1788 type. Vector types are considered :ref:`first class <t_firstclass>`.
1794 < <# elements> x <elementtype> >
1796 The number of elements is a constant integer value larger than 0;
1797 elementtype may be any integer or floating point type, or a pointer to
1798 these types. Vectors of size zero are not allowed.
1802 +-------------------+--------------------------------------------------+
1803 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1804 +-------------------+--------------------------------------------------+
1805 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1806 +-------------------+--------------------------------------------------+
1807 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1808 +-------------------+--------------------------------------------------+
1809 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1810 +-------------------+--------------------------------------------------+
1819 The label type represents code labels.
1834 The metadata type represents embedded metadata. No derived types may be
1835 created from metadata except for :ref:`function <t_function>` arguments.
1848 Aggregate Types are a subset of derived types that can contain multiple
1849 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1850 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1860 The array type is a very simple derived type that arranges elements
1861 sequentially in memory. The array type requires a size (number of
1862 elements) and an underlying data type.
1868 [<# elements> x <elementtype>]
1870 The number of elements is a constant integer value; ``elementtype`` may
1871 be any type with a size.
1875 +------------------+--------------------------------------+
1876 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1877 +------------------+--------------------------------------+
1878 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1879 +------------------+--------------------------------------+
1880 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1881 +------------------+--------------------------------------+
1883 Here are some examples of multidimensional arrays:
1885 +-----------------------------+----------------------------------------------------------+
1886 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1887 +-----------------------------+----------------------------------------------------------+
1888 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1889 +-----------------------------+----------------------------------------------------------+
1890 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1891 +-----------------------------+----------------------------------------------------------+
1893 There is no restriction on indexing beyond the end of the array implied
1894 by a static type (though there are restrictions on indexing beyond the
1895 bounds of an allocated object in some cases). This means that
1896 single-dimension 'variable sized array' addressing can be implemented in
1897 LLVM with a zero length array type. An implementation of 'pascal style
1898 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1908 The structure type is used to represent a collection of data members
1909 together in memory. The elements of a structure may be any type that has
1912 Structures in memory are accessed using '``load``' and '``store``' by
1913 getting a pointer to a field with the '``getelementptr``' instruction.
1914 Structures in registers are accessed using the '``extractvalue``' and
1915 '``insertvalue``' instructions.
1917 Structures may optionally be "packed" structures, which indicate that
1918 the alignment of the struct is one byte, and that there is no padding
1919 between the elements. In non-packed structs, padding between field types
1920 is inserted as defined by the DataLayout string in the module, which is
1921 required to match what the underlying code generator expects.
1923 Structures can either be "literal" or "identified". A literal structure
1924 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1925 identified types are always defined at the top level with a name.
1926 Literal types are uniqued by their contents and can never be recursive
1927 or opaque since there is no way to write one. Identified types can be
1928 recursive, can be opaqued, and are never uniqued.
1934 %T1 = type { <type list> } ; Identified normal struct type
1935 %T2 = type <{ <type list> }> ; Identified packed struct type
1939 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1940 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1941 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1942 | ``{ 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``. |
1943 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1944 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1945 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1949 Opaque Structure Types
1950 """"""""""""""""""""""
1954 Opaque structure types are used to represent named structure types that
1955 do not have a body specified. This corresponds (for example) to the C
1956 notion of a forward declared structure.
1967 +--------------+-------------------+
1968 | ``opaque`` | An opaque type. |
1969 +--------------+-------------------+
1974 LLVM has several different basic types of constants. This section
1975 describes them all and their syntax.
1980 **Boolean constants**
1981 The two strings '``true``' and '``false``' are both valid constants
1983 **Integer constants**
1984 Standard integers (such as '4') are constants of the
1985 :ref:`integer <t_integer>` type. Negative numbers may be used with
1987 **Floating point constants**
1988 Floating point constants use standard decimal notation (e.g.
1989 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1990 hexadecimal notation (see below). The assembler requires the exact
1991 decimal value of a floating-point constant. For example, the
1992 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1993 decimal in binary. Floating point constants must have a :ref:`floating
1994 point <t_floating>` type.
1995 **Null pointer constants**
1996 The identifier '``null``' is recognized as a null pointer constant
1997 and must be of :ref:`pointer type <t_pointer>`.
1999 The one non-intuitive notation for constants is the hexadecimal form of
2000 floating point constants. For example, the form
2001 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2002 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2003 constants are required (and the only time that they are generated by the
2004 disassembler) is when a floating point constant must be emitted but it
2005 cannot be represented as a decimal floating point number in a reasonable
2006 number of digits. For example, NaN's, infinities, and other special
2007 values are represented in their IEEE hexadecimal format so that assembly
2008 and disassembly do not cause any bits to change in the constants.
2010 When using the hexadecimal form, constants of types half, float, and
2011 double are represented using the 16-digit form shown above (which
2012 matches the IEEE754 representation for double); half and float values
2013 must, however, be exactly representable as IEEE 754 half and single
2014 precision, respectively. Hexadecimal format is always used for long
2015 double, and there are three forms of long double. The 80-bit format used
2016 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2017 128-bit format used by PowerPC (two adjacent doubles) is represented by
2018 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2019 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2020 will only work if they match the long double format on your target.
2021 The IEEE 16-bit format (half precision) is represented by ``0xH``
2022 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2023 (sign bit at the left).
2025 There are no constants of type x86mmx.
2027 .. _complexconstants:
2032 Complex constants are a (potentially recursive) combination of simple
2033 constants and smaller complex constants.
2035 **Structure constants**
2036 Structure constants are represented with notation similar to
2037 structure type definitions (a comma separated list of elements,
2038 surrounded by braces (``{}``)). For example:
2039 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2040 "``@G = external global i32``". Structure constants must have
2041 :ref:`structure type <t_struct>`, and the number and types of elements
2042 must match those specified by the type.
2044 Array constants are represented with notation similar to array type
2045 definitions (a comma separated list of elements, surrounded by
2046 square brackets (``[]``)). For example:
2047 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2048 :ref:`array type <t_array>`, and the number and types of elements must
2049 match those specified by the type.
2050 **Vector constants**
2051 Vector constants are represented with notation similar to vector
2052 type definitions (a comma separated list of elements, surrounded by
2053 less-than/greater-than's (``<>``)). For example:
2054 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2055 must have :ref:`vector type <t_vector>`, and the number and types of
2056 elements must match those specified by the type.
2057 **Zero initialization**
2058 The string '``zeroinitializer``' can be used to zero initialize a
2059 value to zero of *any* type, including scalar and
2060 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2061 having to print large zero initializers (e.g. for large arrays) and
2062 is always exactly equivalent to using explicit zero initializers.
2064 A metadata node is a structure-like constant with :ref:`metadata
2065 type <t_metadata>`. For example:
2066 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2067 constants that are meant to be interpreted as part of the
2068 instruction stream, metadata is a place to attach additional
2069 information such as debug info.
2071 Global Variable and Function Addresses
2072 --------------------------------------
2074 The addresses of :ref:`global variables <globalvars>` and
2075 :ref:`functions <functionstructure>` are always implicitly valid
2076 (link-time) constants. These constants are explicitly referenced when
2077 the :ref:`identifier for the global <identifiers>` is used and always have
2078 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2081 .. code-block:: llvm
2085 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2092 The string '``undef``' can be used anywhere a constant is expected, and
2093 indicates that the user of the value may receive an unspecified
2094 bit-pattern. Undefined values may be of any type (other than '``label``'
2095 or '``void``') and be used anywhere a constant is permitted.
2097 Undefined values are useful because they indicate to the compiler that
2098 the program is well defined no matter what value is used. This gives the
2099 compiler more freedom to optimize. Here are some examples of
2100 (potentially surprising) transformations that are valid (in pseudo IR):
2102 .. code-block:: llvm
2112 This is safe because all of the output bits are affected by the undef
2113 bits. Any output bit can have a zero or one depending on the input bits.
2115 .. code-block:: llvm
2126 These logical operations have bits that are not always affected by the
2127 input. For example, if ``%X`` has a zero bit, then the output of the
2128 '``and``' operation will always be a zero for that bit, no matter what
2129 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2130 optimize or assume that the result of the '``and``' is '``undef``'.
2131 However, it is safe to assume that all bits of the '``undef``' could be
2132 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2133 all the bits of the '``undef``' operand to the '``or``' could be set,
2134 allowing the '``or``' to be folded to -1.
2136 .. code-block:: llvm
2138 %A = select undef, %X, %Y
2139 %B = select undef, 42, %Y
2140 %C = select %X, %Y, undef
2150 This set of examples shows that undefined '``select``' (and conditional
2151 branch) conditions can go *either way*, but they have to come from one
2152 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2153 both known to have a clear low bit, then ``%A`` would have to have a
2154 cleared low bit. However, in the ``%C`` example, the optimizer is
2155 allowed to assume that the '``undef``' operand could be the same as
2156 ``%Y``, allowing the whole '``select``' to be eliminated.
2158 .. code-block:: llvm
2160 %A = xor undef, undef
2177 This example points out that two '``undef``' operands are not
2178 necessarily the same. This can be surprising to people (and also matches
2179 C semantics) where they assume that "``X^X``" is always zero, even if
2180 ``X`` is undefined. This isn't true for a number of reasons, but the
2181 short answer is that an '``undef``' "variable" can arbitrarily change
2182 its value over its "live range". This is true because the variable
2183 doesn't actually *have a live range*. Instead, the value is logically
2184 read from arbitrary registers that happen to be around when needed, so
2185 the value is not necessarily consistent over time. In fact, ``%A`` and
2186 ``%C`` need to have the same semantics or the core LLVM "replace all
2187 uses with" concept would not hold.
2189 .. code-block:: llvm
2197 These examples show the crucial difference between an *undefined value*
2198 and *undefined behavior*. An undefined value (like '``undef``') is
2199 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2200 operation can be constant folded to '``undef``', because the '``undef``'
2201 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2202 However, in the second example, we can make a more aggressive
2203 assumption: because the ``undef`` is allowed to be an arbitrary value,
2204 we are allowed to assume that it could be zero. Since a divide by zero
2205 has *undefined behavior*, we are allowed to assume that the operation
2206 does not execute at all. This allows us to delete the divide and all
2207 code after it. Because the undefined operation "can't happen", the
2208 optimizer can assume that it occurs in dead code.
2210 .. code-block:: llvm
2212 a: store undef -> %X
2213 b: store %X -> undef
2218 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2219 value can be assumed to not have any effect; we can assume that the
2220 value is overwritten with bits that happen to match what was already
2221 there. However, a store *to* an undefined location could clobber
2222 arbitrary memory, therefore, it has undefined behavior.
2229 Poison values are similar to :ref:`undef values <undefvalues>`, however
2230 they also represent the fact that an instruction or constant expression
2231 which cannot evoke side effects has nevertheless detected a condition
2232 which results in undefined behavior.
2234 There is currently no way of representing a poison value in the IR; they
2235 only exist when produced by operations such as :ref:`add <i_add>` with
2238 Poison value behavior is defined in terms of value *dependence*:
2240 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2241 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2242 their dynamic predecessor basic block.
2243 - Function arguments depend on the corresponding actual argument values
2244 in the dynamic callers of their functions.
2245 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2246 instructions that dynamically transfer control back to them.
2247 - :ref:`Invoke <i_invoke>` instructions depend on the
2248 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2249 call instructions that dynamically transfer control back to them.
2250 - Non-volatile loads and stores depend on the most recent stores to all
2251 of the referenced memory addresses, following the order in the IR
2252 (including loads and stores implied by intrinsics such as
2253 :ref:`@llvm.memcpy <int_memcpy>`.)
2254 - An instruction with externally visible side effects depends on the
2255 most recent preceding instruction with externally visible side
2256 effects, following the order in the IR. (This includes :ref:`volatile
2257 operations <volatile>`.)
2258 - An instruction *control-depends* on a :ref:`terminator
2259 instruction <terminators>` if the terminator instruction has
2260 multiple successors and the instruction is always executed when
2261 control transfers to one of the successors, and may not be executed
2262 when control is transferred to another.
2263 - Additionally, an instruction also *control-depends* on a terminator
2264 instruction if the set of instructions it otherwise depends on would
2265 be different if the terminator had transferred control to a different
2267 - Dependence is transitive.
2269 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2270 with the additional affect that any instruction which has a *dependence*
2271 on a poison value has undefined behavior.
2273 Here are some examples:
2275 .. code-block:: llvm
2278 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2279 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2280 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2281 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2283 store i32 %poison, i32* @g ; Poison value stored to memory.
2284 %poison2 = load i32* @g ; Poison value loaded back from memory.
2286 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2288 %narrowaddr = bitcast i32* @g to i16*
2289 %wideaddr = bitcast i32* @g to i64*
2290 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2291 %poison4 = load i64* %wideaddr ; Returns a poison value.
2293 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2294 br i1 %cmp, label %true, label %end ; Branch to either destination.
2297 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2298 ; it has undefined behavior.
2302 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2303 ; Both edges into this PHI are
2304 ; control-dependent on %cmp, so this
2305 ; always results in a poison value.
2307 store volatile i32 0, i32* @g ; This would depend on the store in %true
2308 ; if %cmp is true, or the store in %entry
2309 ; otherwise, so this is undefined behavior.
2311 br i1 %cmp, label %second_true, label %second_end
2312 ; The same branch again, but this time the
2313 ; true block doesn't have side effects.
2320 store volatile i32 0, i32* @g ; This time, the instruction always depends
2321 ; on the store in %end. Also, it is
2322 ; control-equivalent to %end, so this is
2323 ; well-defined (ignoring earlier undefined
2324 ; behavior in this example).
2328 Addresses of Basic Blocks
2329 -------------------------
2331 ``blockaddress(@function, %block)``
2333 The '``blockaddress``' constant computes the address of the specified
2334 basic block in the specified function, and always has an ``i8*`` type.
2335 Taking the address of the entry block is illegal.
2337 This value only has defined behavior when used as an operand to the
2338 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2339 against null. Pointer equality tests between labels addresses results in
2340 undefined behavior --- though, again, comparison against null is ok, and
2341 no label is equal to the null pointer. This may be passed around as an
2342 opaque pointer sized value as long as the bits are not inspected. This
2343 allows ``ptrtoint`` and arithmetic to be performed on these values so
2344 long as the original value is reconstituted before the ``indirectbr``
2347 Finally, some targets may provide defined semantics when using the value
2348 as the operand to an inline assembly, but that is target specific.
2352 Constant Expressions
2353 --------------------
2355 Constant expressions are used to allow expressions involving other
2356 constants to be used as constants. Constant expressions may be of any
2357 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2358 that does not have side effects (e.g. load and call are not supported).
2359 The following is the syntax for constant expressions:
2361 ``trunc (CST to TYPE)``
2362 Truncate a constant to another type. The bit size of CST must be
2363 larger than the bit size of TYPE. Both types must be integers.
2364 ``zext (CST to TYPE)``
2365 Zero extend a constant to another type. The bit size of CST must be
2366 smaller than the bit size of TYPE. Both types must be integers.
2367 ``sext (CST to TYPE)``
2368 Sign extend a constant to another type. The bit size of CST must be
2369 smaller than the bit size of TYPE. Both types must be integers.
2370 ``fptrunc (CST to TYPE)``
2371 Truncate a floating point constant to another floating point type.
2372 The size of CST must be larger than the size of TYPE. Both types
2373 must be floating point.
2374 ``fpext (CST to TYPE)``
2375 Floating point extend a constant to another type. The size of CST
2376 must be smaller or equal to the size of TYPE. Both types must be
2378 ``fptoui (CST to TYPE)``
2379 Convert a floating point constant to the corresponding unsigned
2380 integer constant. TYPE must be a scalar or vector integer type. CST
2381 must be of scalar or vector floating point type. Both CST and TYPE
2382 must be scalars, or vectors of the same number of elements. If the
2383 value won't fit in the integer type, the results are undefined.
2384 ``fptosi (CST to TYPE)``
2385 Convert a floating point constant to the corresponding signed
2386 integer constant. TYPE must be a scalar or vector integer type. CST
2387 must be of scalar or vector floating point type. Both CST and TYPE
2388 must be scalars, or vectors of the same number of elements. If the
2389 value won't fit in the integer type, the results are undefined.
2390 ``uitofp (CST to TYPE)``
2391 Convert an unsigned integer constant to the corresponding floating
2392 point constant. TYPE must be a scalar or vector floating point type.
2393 CST must be of scalar or vector integer type. Both CST and TYPE must
2394 be scalars, or vectors of the same number of elements. If the value
2395 won't fit in the floating point type, the results are undefined.
2396 ``sitofp (CST to TYPE)``
2397 Convert a signed integer constant to the corresponding floating
2398 point constant. TYPE must be a scalar or vector floating point type.
2399 CST must be of scalar or vector integer type. Both CST and TYPE must
2400 be scalars, or vectors of the same number of elements. If the value
2401 won't fit in the floating point type, the results are undefined.
2402 ``ptrtoint (CST to TYPE)``
2403 Convert a pointer typed constant to the corresponding integer
2404 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2405 pointer type. The ``CST`` value is zero extended, truncated, or
2406 unchanged to make it fit in ``TYPE``.
2407 ``inttoptr (CST to TYPE)``
2408 Convert an integer constant to a pointer constant. TYPE must be a
2409 pointer type. CST must be of integer type. The CST value is zero
2410 extended, truncated, or unchanged to make it fit in a pointer size.
2411 This one is *really* dangerous!
2412 ``bitcast (CST to TYPE)``
2413 Convert a constant, CST, to another TYPE. The constraints of the
2414 operands are the same as those for the :ref:`bitcast
2415 instruction <i_bitcast>`.
2416 ``addrspacecast (CST to TYPE)``
2417 Convert a constant pointer or constant vector of pointer, CST, to another
2418 TYPE in a different address space. The constraints of the operands are the
2419 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2420 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2421 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2422 constants. As with the :ref:`getelementptr <i_getelementptr>`
2423 instruction, the index list may have zero or more indexes, which are
2424 required to make sense for the type of "CSTPTR".
2425 ``select (COND, VAL1, VAL2)``
2426 Perform the :ref:`select operation <i_select>` on constants.
2427 ``icmp COND (VAL1, VAL2)``
2428 Performs the :ref:`icmp operation <i_icmp>` on constants.
2429 ``fcmp COND (VAL1, VAL2)``
2430 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2431 ``extractelement (VAL, IDX)``
2432 Perform the :ref:`extractelement operation <i_extractelement>` on
2434 ``insertelement (VAL, ELT, IDX)``
2435 Perform the :ref:`insertelement operation <i_insertelement>` on
2437 ``shufflevector (VEC1, VEC2, IDXMASK)``
2438 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2440 ``extractvalue (VAL, IDX0, IDX1, ...)``
2441 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2442 constants. The index list is interpreted in a similar manner as
2443 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2444 least one index value must be specified.
2445 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2446 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2447 The index list is interpreted in a similar manner as indices in a
2448 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2449 value must be specified.
2450 ``OPCODE (LHS, RHS)``
2451 Perform the specified operation of the LHS and RHS constants. OPCODE
2452 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2453 binary <bitwiseops>` operations. The constraints on operands are
2454 the same as those for the corresponding instruction (e.g. no bitwise
2455 operations on floating point values are allowed).
2462 Inline Assembler Expressions
2463 ----------------------------
2465 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2466 Inline Assembly <moduleasm>`) through the use of a special value. This
2467 value represents the inline assembler as a string (containing the
2468 instructions to emit), a list of operand constraints (stored as a
2469 string), a flag that indicates whether or not the inline asm expression
2470 has side effects, and a flag indicating whether the function containing
2471 the asm needs to align its stack conservatively. An example inline
2472 assembler expression is:
2474 .. code-block:: llvm
2476 i32 (i32) asm "bswap $0", "=r,r"
2478 Inline assembler expressions may **only** be used as the callee operand
2479 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2480 Thus, typically we have:
2482 .. code-block:: llvm
2484 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2486 Inline asms with side effects not visible in the constraint list must be
2487 marked as having side effects. This is done through the use of the
2488 '``sideeffect``' keyword, like so:
2490 .. code-block:: llvm
2492 call void asm sideeffect "eieio", ""()
2494 In some cases inline asms will contain code that will not work unless
2495 the stack is aligned in some way, such as calls or SSE instructions on
2496 x86, yet will not contain code that does that alignment within the asm.
2497 The compiler should make conservative assumptions about what the asm
2498 might contain and should generate its usual stack alignment code in the
2499 prologue if the '``alignstack``' keyword is present:
2501 .. code-block:: llvm
2503 call void asm alignstack "eieio", ""()
2505 Inline asms also support using non-standard assembly dialects. The
2506 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2507 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2508 the only supported dialects. An example is:
2510 .. code-block:: llvm
2512 call void asm inteldialect "eieio", ""()
2514 If multiple keywords appear the '``sideeffect``' keyword must come
2515 first, the '``alignstack``' keyword second and the '``inteldialect``'
2521 The call instructions that wrap inline asm nodes may have a
2522 "``!srcloc``" MDNode attached to it that contains a list of constant
2523 integers. If present, the code generator will use the integer as the
2524 location cookie value when report errors through the ``LLVMContext``
2525 error reporting mechanisms. This allows a front-end to correlate backend
2526 errors that occur with inline asm back to the source code that produced
2529 .. code-block:: llvm
2531 call void asm sideeffect "something bad", ""(), !srcloc !42
2533 !42 = !{ i32 1234567 }
2535 It is up to the front-end to make sense of the magic numbers it places
2536 in the IR. If the MDNode contains multiple constants, the code generator
2537 will use the one that corresponds to the line of the asm that the error
2542 Metadata Nodes and Metadata Strings
2543 -----------------------------------
2545 LLVM IR allows metadata to be attached to instructions in the program
2546 that can convey extra information about the code to the optimizers and
2547 code generator. One example application of metadata is source-level
2548 debug information. There are two metadata primitives: strings and nodes.
2549 All metadata has the ``metadata`` type and is identified in syntax by a
2550 preceding exclamation point ('``!``').
2552 A metadata string is a string surrounded by double quotes. It can
2553 contain any character by escaping non-printable characters with
2554 "``\xx``" where "``xx``" is the two digit hex code. For example:
2557 Metadata nodes are represented with notation similar to structure
2558 constants (a comma separated list of elements, surrounded by braces and
2559 preceded by an exclamation point). Metadata nodes can have any values as
2560 their operand. For example:
2562 .. code-block:: llvm
2564 !{ metadata !"test\00", i32 10}
2566 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2567 metadata nodes, which can be looked up in the module symbol table. For
2570 .. code-block:: llvm
2572 !foo = metadata !{!4, !3}
2574 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2575 function is using two metadata arguments:
2577 .. code-block:: llvm
2579 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2581 Metadata can be attached with an instruction. Here metadata ``!21`` is
2582 attached to the ``add`` instruction using the ``!dbg`` identifier:
2584 .. code-block:: llvm
2586 %indvar.next = add i64 %indvar, 1, !dbg !21
2588 More information about specific metadata nodes recognized by the
2589 optimizers and code generator is found below.
2594 In LLVM IR, memory does not have types, so LLVM's own type system is not
2595 suitable for doing TBAA. Instead, metadata is added to the IR to
2596 describe a type system of a higher level language. This can be used to
2597 implement typical C/C++ TBAA, but it can also be used to implement
2598 custom alias analysis behavior for other languages.
2600 The current metadata format is very simple. TBAA metadata nodes have up
2601 to three fields, e.g.:
2603 .. code-block:: llvm
2605 !0 = metadata !{ metadata !"an example type tree" }
2606 !1 = metadata !{ metadata !"int", metadata !0 }
2607 !2 = metadata !{ metadata !"float", metadata !0 }
2608 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2610 The first field is an identity field. It can be any value, usually a
2611 metadata string, which uniquely identifies the type. The most important
2612 name in the tree is the name of the root node. Two trees with different
2613 root node names are entirely disjoint, even if they have leaves with
2616 The second field identifies the type's parent node in the tree, or is
2617 null or omitted for a root node. A type is considered to alias all of
2618 its descendants and all of its ancestors in the tree. Also, a type is
2619 considered to alias all types in other trees, so that bitcode produced
2620 from multiple front-ends is handled conservatively.
2622 If the third field is present, it's an integer which if equal to 1
2623 indicates that the type is "constant" (meaning
2624 ``pointsToConstantMemory`` should return true; see `other useful
2625 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2627 '``tbaa.struct``' Metadata
2628 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2630 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2631 aggregate assignment operations in C and similar languages, however it
2632 is defined to copy a contiguous region of memory, which is more than
2633 strictly necessary for aggregate types which contain holes due to
2634 padding. Also, it doesn't contain any TBAA information about the fields
2637 ``!tbaa.struct`` metadata can describe which memory subregions in a
2638 memcpy are padding and what the TBAA tags of the struct are.
2640 The current metadata format is very simple. ``!tbaa.struct`` metadata
2641 nodes are a list of operands which are in conceptual groups of three.
2642 For each group of three, the first operand gives the byte offset of a
2643 field in bytes, the second gives its size in bytes, and the third gives
2646 .. code-block:: llvm
2648 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2650 This describes a struct with two fields. The first is at offset 0 bytes
2651 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2652 and has size 4 bytes and has tbaa tag !2.
2654 Note that the fields need not be contiguous. In this example, there is a
2655 4 byte gap between the two fields. This gap represents padding which
2656 does not carry useful data and need not be preserved.
2658 '``fpmath``' Metadata
2659 ^^^^^^^^^^^^^^^^^^^^^
2661 ``fpmath`` metadata may be attached to any instruction of floating point
2662 type. It can be used to express the maximum acceptable error in the
2663 result of that instruction, in ULPs, thus potentially allowing the
2664 compiler to use a more efficient but less accurate method of computing
2665 it. ULP is defined as follows:
2667 If ``x`` is a real number that lies between two finite consecutive
2668 floating-point numbers ``a`` and ``b``, without being equal to one
2669 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2670 distance between the two non-equal finite floating-point numbers
2671 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2673 The metadata node shall consist of a single positive floating point
2674 number representing the maximum relative error, for example:
2676 .. code-block:: llvm
2678 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2680 '``range``' Metadata
2681 ^^^^^^^^^^^^^^^^^^^^
2683 ``range`` metadata may be attached only to loads of integer types. It
2684 expresses the possible ranges the loaded value is in. The ranges are
2685 represented with a flattened list of integers. The loaded value is known
2686 to be in the union of the ranges defined by each consecutive pair. Each
2687 pair has the following properties:
2689 - The type must match the type loaded by the instruction.
2690 - The pair ``a,b`` represents the range ``[a,b)``.
2691 - Both ``a`` and ``b`` are constants.
2692 - The range is allowed to wrap.
2693 - The range should not represent the full or empty set. That is,
2696 In addition, the pairs must be in signed order of the lower bound and
2697 they must be non-contiguous.
2701 .. code-block:: llvm
2703 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2704 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2705 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2706 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2708 !0 = metadata !{ i8 0, i8 2 }
2709 !1 = metadata !{ i8 255, i8 2 }
2710 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2711 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2716 It is sometimes useful to attach information to loop constructs. Currently,
2717 loop metadata is implemented as metadata attached to the branch instruction
2718 in the loop latch block. This type of metadata refer to a metadata node that is
2719 guaranteed to be separate for each loop. The loop identifier metadata is
2720 specified with the name ``llvm.loop``.
2722 The loop identifier metadata is implemented using a metadata that refers to
2723 itself to avoid merging it with any other identifier metadata, e.g.,
2724 during module linkage or function inlining. That is, each loop should refer
2725 to their own identification metadata even if they reside in separate functions.
2726 The following example contains loop identifier metadata for two separate loop
2729 .. code-block:: llvm
2731 !0 = metadata !{ metadata !0 }
2732 !1 = metadata !{ metadata !1 }
2734 The loop identifier metadata can be used to specify additional per-loop
2735 metadata. Any operands after the first operand can be treated as user-defined
2736 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2737 by the loop vectorizer to indicate how many times to unroll the loop:
2739 .. code-block:: llvm
2741 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2743 !0 = metadata !{ metadata !0, metadata !1 }
2744 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2749 Metadata types used to annotate memory accesses with information helpful
2750 for optimizations are prefixed with ``llvm.mem``.
2752 '``llvm.mem.parallel_loop_access``' Metadata
2753 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2755 For a loop to be parallel, in addition to using
2756 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2757 also all of the memory accessing instructions in the loop body need to be
2758 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2759 is at least one memory accessing instruction not marked with the metadata,
2760 the loop must be considered a sequential loop. This causes parallel loops to be
2761 converted to sequential loops due to optimization passes that are unaware of
2762 the parallel semantics and that insert new memory instructions to the loop
2765 Example of a loop that is considered parallel due to its correct use of
2766 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2767 metadata types that refer to the same loop identifier metadata.
2769 .. code-block:: llvm
2773 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2775 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2777 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2781 !0 = metadata !{ metadata !0 }
2783 It is also possible to have nested parallel loops. In that case the
2784 memory accesses refer to a list of loop identifier metadata nodes instead of
2785 the loop identifier metadata node directly:
2787 .. code-block:: llvm
2794 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2796 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2798 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
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 %outer.for.end, label %outer.for.body, !llvm.loop !2
2808 outer.for.end: ; preds = %for.body
2810 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2811 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2812 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2814 '``llvm.vectorizer``'
2815 ^^^^^^^^^^^^^^^^^^^^^
2817 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2818 vectorization parameters such as vectorization factor and unroll factor.
2820 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2821 loop identification metadata.
2823 '``llvm.vectorizer.unroll``' Metadata
2824 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2826 This metadata instructs the loop vectorizer to unroll the specified
2827 loop exactly ``N`` times.
2829 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2830 operand is an integer specifying the unroll factor. For example:
2832 .. code-block:: llvm
2834 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2836 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2839 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2840 determined automatically.
2842 '``llvm.vectorizer.width``' Metadata
2843 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2845 This metadata sets the target width of the vectorizer to ``N``. Without
2846 this metadata, the vectorizer will choose a width automatically.
2847 Regardless of this metadata, the vectorizer will only vectorize loops if
2848 it believes it is valid to do so.
2850 The first operand is the string ``llvm.vectorizer.width`` and the second
2851 operand is an integer specifying the width. For example:
2853 .. code-block:: llvm
2855 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2857 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2860 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2863 Module Flags Metadata
2864 =====================
2866 Information about the module as a whole is difficult to convey to LLVM's
2867 subsystems. The LLVM IR isn't sufficient to transmit this information.
2868 The ``llvm.module.flags`` named metadata exists in order to facilitate
2869 this. These flags are in the form of key / value pairs --- much like a
2870 dictionary --- making it easy for any subsystem who cares about a flag to
2873 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2874 Each triplet has the following form:
2876 - The first element is a *behavior* flag, which specifies the behavior
2877 when two (or more) modules are merged together, and it encounters two
2878 (or more) metadata with the same ID. The supported behaviors are
2880 - The second element is a metadata string that is a unique ID for the
2881 metadata. Each module may only have one flag entry for each unique ID (not
2882 including entries with the **Require** behavior).
2883 - The third element is the value of the flag.
2885 When two (or more) modules are merged together, the resulting
2886 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2887 each unique metadata ID string, there will be exactly one entry in the merged
2888 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2889 be determined by the merge behavior flag, as described below. The only exception
2890 is that entries with the *Require* behavior are always preserved.
2892 The following behaviors are supported:
2903 Emits an error if two values disagree, otherwise the resulting value
2904 is that of the operands.
2908 Emits a warning if two values disagree. The result value will be the
2909 operand for the flag from the first module being linked.
2913 Adds a requirement that another module flag be present and have a
2914 specified value after linking is performed. The value must be a
2915 metadata pair, where the first element of the pair is the ID of the
2916 module flag to be restricted, and the second element of the pair is
2917 the value the module flag should be restricted to. This behavior can
2918 be used to restrict the allowable results (via triggering of an
2919 error) of linking IDs with the **Override** behavior.
2923 Uses the specified value, regardless of the behavior or value of the
2924 other module. If both modules specify **Override**, but the values
2925 differ, an error will be emitted.
2929 Appends the two values, which are required to be metadata nodes.
2933 Appends the two values, which are required to be metadata
2934 nodes. However, duplicate entries in the second list are dropped
2935 during the append operation.
2937 It is an error for a particular unique flag ID to have multiple behaviors,
2938 except in the case of **Require** (which adds restrictions on another metadata
2939 value) or **Override**.
2941 An example of module flags:
2943 .. code-block:: llvm
2945 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2946 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2947 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2948 !3 = metadata !{ i32 3, metadata !"qux",
2950 metadata !"foo", i32 1
2953 !llvm.module.flags = !{ !0, !1, !2, !3 }
2955 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2956 if two or more ``!"foo"`` flags are seen is to emit an error if their
2957 values are not equal.
2959 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2960 behavior if two or more ``!"bar"`` flags are seen is to use the value
2963 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2964 behavior if two or more ``!"qux"`` flags are seen is to emit a
2965 warning if their values are not equal.
2967 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2971 metadata !{ metadata !"foo", i32 1 }
2973 The behavior is to emit an error if the ``llvm.module.flags`` does not
2974 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2977 Objective-C Garbage Collection Module Flags Metadata
2978 ----------------------------------------------------
2980 On the Mach-O platform, Objective-C stores metadata about garbage
2981 collection in a special section called "image info". The metadata
2982 consists of a version number and a bitmask specifying what types of
2983 garbage collection are supported (if any) by the file. If two or more
2984 modules are linked together their garbage collection metadata needs to
2985 be merged rather than appended together.
2987 The Objective-C garbage collection module flags metadata consists of the
2988 following key-value pairs:
2997 * - ``Objective-C Version``
2998 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3000 * - ``Objective-C Image Info Version``
3001 - **[Required]** --- The version of the image info section. Currently
3004 * - ``Objective-C Image Info Section``
3005 - **[Required]** --- The section to place the metadata. Valid values are
3006 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3007 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3008 Objective-C ABI version 2.
3010 * - ``Objective-C Garbage Collection``
3011 - **[Required]** --- Specifies whether garbage collection is supported or
3012 not. Valid values are 0, for no garbage collection, and 2, for garbage
3013 collection supported.
3015 * - ``Objective-C GC Only``
3016 - **[Optional]** --- Specifies that only garbage collection is supported.
3017 If present, its value must be 6. This flag requires that the
3018 ``Objective-C Garbage Collection`` flag have the value 2.
3020 Some important flag interactions:
3022 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3023 merged with a module with ``Objective-C Garbage Collection`` set to
3024 2, then the resulting module has the
3025 ``Objective-C Garbage Collection`` flag set to 0.
3026 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3027 merged with a module with ``Objective-C GC Only`` set to 6.
3029 Automatic Linker Flags Module Flags Metadata
3030 --------------------------------------------
3032 Some targets support embedding flags to the linker inside individual object
3033 files. Typically this is used in conjunction with language extensions which
3034 allow source files to explicitly declare the libraries they depend on, and have
3035 these automatically be transmitted to the linker via object files.
3037 These flags are encoded in the IR using metadata in the module flags section,
3038 using the ``Linker Options`` key. The merge behavior for this flag is required
3039 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3040 node which should be a list of other metadata nodes, each of which should be a
3041 list of metadata strings defining linker options.
3043 For example, the following metadata section specifies two separate sets of
3044 linker options, presumably to link against ``libz`` and the ``Cocoa``
3047 !0 = metadata !{ i32 6, metadata !"Linker Options",
3049 metadata !{ metadata !"-lz" },
3050 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3051 !llvm.module.flags = !{ !0 }
3053 The metadata encoding as lists of lists of options, as opposed to a collapsed
3054 list of options, is chosen so that the IR encoding can use multiple option
3055 strings to specify e.g., a single library, while still having that specifier be
3056 preserved as an atomic element that can be recognized by a target specific
3057 assembly writer or object file emitter.
3059 Each individual option is required to be either a valid option for the target's
3060 linker, or an option that is reserved by the target specific assembly writer or
3061 object file emitter. No other aspect of these options is defined by the IR.
3063 .. _intrinsicglobalvariables:
3065 Intrinsic Global Variables
3066 ==========================
3068 LLVM has a number of "magic" global variables that contain data that
3069 affect code generation or other IR semantics. These are documented here.
3070 All globals of this sort should have a section specified as
3071 "``llvm.metadata``". This section and all globals that start with
3072 "``llvm.``" are reserved for use by LLVM.
3076 The '``llvm.used``' Global Variable
3077 -----------------------------------
3079 The ``@llvm.used`` global is an array which has
3080 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3081 pointers to named global variables, functions and aliases which may optionally
3082 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3085 .. code-block:: llvm
3090 @llvm.used = appending global [2 x i8*] [
3092 i8* bitcast (i32* @Y to i8*)
3093 ], section "llvm.metadata"
3095 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3096 and linker are required to treat the symbol as if there is a reference to the
3097 symbol that it cannot see (which is why they have to be named). For example, if
3098 a variable has internal linkage and no references other than that from the
3099 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3100 references from inline asms and other things the compiler cannot "see", and
3101 corresponds to "``attribute((used))``" in GNU C.
3103 On some targets, the code generator must emit a directive to the
3104 assembler or object file to prevent the assembler and linker from
3105 molesting the symbol.
3107 .. _gv_llvmcompilerused:
3109 The '``llvm.compiler.used``' Global Variable
3110 --------------------------------------------
3112 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3113 directive, except that it only prevents the compiler from touching the
3114 symbol. On targets that support it, this allows an intelligent linker to
3115 optimize references to the symbol without being impeded as it would be
3118 This is a rare construct that should only be used in rare circumstances,
3119 and should not be exposed to source languages.
3121 .. _gv_llvmglobalctors:
3123 The '``llvm.global_ctors``' Global Variable
3124 -------------------------------------------
3126 .. code-block:: llvm
3128 %0 = type { i32, void ()* }
3129 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3131 The ``@llvm.global_ctors`` array contains a list of constructor
3132 functions and associated priorities. The functions referenced by this
3133 array will be called in ascending order of priority (i.e. lowest first)
3134 when the module is loaded. The order of functions with the same priority
3137 .. _llvmglobaldtors:
3139 The '``llvm.global_dtors``' Global Variable
3140 -------------------------------------------
3142 .. code-block:: llvm
3144 %0 = type { i32, void ()* }
3145 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3147 The ``@llvm.global_dtors`` array contains a list of destructor functions
3148 and associated priorities. The functions referenced by this array will
3149 be called in descending order of priority (i.e. highest first) when the
3150 module is loaded. The order of functions with the same priority is not
3153 Instruction Reference
3154 =====================
3156 The LLVM instruction set consists of several different classifications
3157 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3158 instructions <binaryops>`, :ref:`bitwise binary
3159 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3160 :ref:`other instructions <otherops>`.
3164 Terminator Instructions
3165 -----------------------
3167 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3168 program ends with a "Terminator" instruction, which indicates which
3169 block should be executed after the current block is finished. These
3170 terminator instructions typically yield a '``void``' value: they produce
3171 control flow, not values (the one exception being the
3172 ':ref:`invoke <i_invoke>`' instruction).
3174 The terminator instructions are: ':ref:`ret <i_ret>`',
3175 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3176 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3177 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3181 '``ret``' Instruction
3182 ^^^^^^^^^^^^^^^^^^^^^
3189 ret <type> <value> ; Return a value from a non-void function
3190 ret void ; Return from void function
3195 The '``ret``' instruction is used to return control flow (and optionally
3196 a value) from a function back to the caller.
3198 There are two forms of the '``ret``' instruction: one that returns a
3199 value and then causes control flow, and one that just causes control
3205 The '``ret``' instruction optionally accepts a single argument, the
3206 return value. The type of the return value must be a ':ref:`first
3207 class <t_firstclass>`' type.
3209 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3210 return type and contains a '``ret``' instruction with no return value or
3211 a return value with a type that does not match its type, or if it has a
3212 void return type and contains a '``ret``' instruction with a return
3218 When the '``ret``' instruction is executed, control flow returns back to
3219 the calling function's context. If the caller is a
3220 ":ref:`call <i_call>`" instruction, execution continues at the
3221 instruction after the call. If the caller was an
3222 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3223 beginning of the "normal" destination block. If the instruction returns
3224 a value, that value shall set the call or invoke instruction's return
3230 .. code-block:: llvm
3232 ret i32 5 ; Return an integer value of 5
3233 ret void ; Return from a void function
3234 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3238 '``br``' Instruction
3239 ^^^^^^^^^^^^^^^^^^^^
3246 br i1 <cond>, label <iftrue>, label <iffalse>
3247 br label <dest> ; Unconditional branch
3252 The '``br``' instruction is used to cause control flow to transfer to a
3253 different basic block in the current function. There are two forms of
3254 this instruction, corresponding to a conditional branch and an
3255 unconditional branch.
3260 The conditional branch form of the '``br``' instruction takes a single
3261 '``i1``' value and two '``label``' values. The unconditional form of the
3262 '``br``' instruction takes a single '``label``' value as a target.
3267 Upon execution of a conditional '``br``' instruction, the '``i1``'
3268 argument is evaluated. If the value is ``true``, control flows to the
3269 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3270 to the '``iffalse``' ``label`` argument.
3275 .. code-block:: llvm
3278 %cond = icmp eq i32 %a, %b
3279 br i1 %cond, label %IfEqual, label %IfUnequal
3287 '``switch``' Instruction
3288 ^^^^^^^^^^^^^^^^^^^^^^^^
3295 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3300 The '``switch``' instruction is used to transfer control flow to one of
3301 several different places. It is a generalization of the '``br``'
3302 instruction, allowing a branch to occur to one of many possible
3308 The '``switch``' instruction uses three parameters: an integer
3309 comparison value '``value``', a default '``label``' destination, and an
3310 array of pairs of comparison value constants and '``label``'s. The table
3311 is not allowed to contain duplicate constant entries.
3316 The ``switch`` instruction specifies a table of values and destinations.
3317 When the '``switch``' instruction is executed, this table is searched
3318 for the given value. If the value is found, control flow is transferred
3319 to the corresponding destination; otherwise, control flow is transferred
3320 to the default destination.
3325 Depending on properties of the target machine and the particular
3326 ``switch`` instruction, this instruction may be code generated in
3327 different ways. For example, it could be generated as a series of
3328 chained conditional branches or with a lookup table.
3333 .. code-block:: llvm
3335 ; Emulate a conditional br instruction
3336 %Val = zext i1 %value to i32
3337 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3339 ; Emulate an unconditional br instruction
3340 switch i32 0, label %dest [ ]
3342 ; Implement a jump table:
3343 switch i32 %val, label %otherwise [ i32 0, label %onzero
3345 i32 2, label %ontwo ]
3349 '``indirectbr``' Instruction
3350 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3357 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3362 The '``indirectbr``' instruction implements an indirect branch to a
3363 label within the current function, whose address is specified by
3364 "``address``". Address must be derived from a
3365 :ref:`blockaddress <blockaddress>` constant.
3370 The '``address``' argument is the address of the label to jump to. The
3371 rest of the arguments indicate the full set of possible destinations
3372 that the address may point to. Blocks are allowed to occur multiple
3373 times in the destination list, though this isn't particularly useful.
3375 This destination list is required so that dataflow analysis has an
3376 accurate understanding of the CFG.
3381 Control transfers to the block specified in the address argument. All
3382 possible destination blocks must be listed in the label list, otherwise
3383 this instruction has undefined behavior. This implies that jumps to
3384 labels defined in other functions have undefined behavior as well.
3389 This is typically implemented with a jump through a register.
3394 .. code-block:: llvm
3396 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3400 '``invoke``' Instruction
3401 ^^^^^^^^^^^^^^^^^^^^^^^^
3408 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3409 to label <normal label> unwind label <exception label>
3414 The '``invoke``' instruction causes control to transfer to a specified
3415 function, with the possibility of control flow transfer to either the
3416 '``normal``' label or the '``exception``' label. If the callee function
3417 returns with the "``ret``" instruction, control flow will return to the
3418 "normal" label. If the callee (or any indirect callees) returns via the
3419 ":ref:`resume <i_resume>`" instruction or other exception handling
3420 mechanism, control is interrupted and continued at the dynamically
3421 nearest "exception" label.
3423 The '``exception``' label is a `landing
3424 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3425 '``exception``' label is required to have the
3426 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3427 information about the behavior of the program after unwinding happens,
3428 as its first non-PHI instruction. The restrictions on the
3429 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3430 instruction, so that the important information contained within the
3431 "``landingpad``" instruction can't be lost through normal code motion.
3436 This instruction requires several arguments:
3438 #. The optional "cconv" marker indicates which :ref:`calling
3439 convention <callingconv>` the call should use. If none is
3440 specified, the call defaults to using C calling conventions.
3441 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3442 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3444 #. '``ptr to function ty``': shall be the signature of the pointer to
3445 function value being invoked. In most cases, this is a direct
3446 function invocation, but indirect ``invoke``'s are just as possible,
3447 branching off an arbitrary pointer to function value.
3448 #. '``function ptr val``': An LLVM value containing a pointer to a
3449 function to be invoked.
3450 #. '``function args``': argument list whose types match the function
3451 signature argument types and parameter attributes. All arguments must
3452 be of :ref:`first class <t_firstclass>` type. If the function signature
3453 indicates the function accepts a variable number of arguments, the
3454 extra arguments can be specified.
3455 #. '``normal label``': the label reached when the called function
3456 executes a '``ret``' instruction.
3457 #. '``exception label``': the label reached when a callee returns via
3458 the :ref:`resume <i_resume>` instruction or other exception handling
3460 #. The optional :ref:`function attributes <fnattrs>` list. Only
3461 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3462 attributes are valid here.
3467 This instruction is designed to operate as a standard '``call``'
3468 instruction in most regards. The primary difference is that it
3469 establishes an association with a label, which is used by the runtime
3470 library to unwind the stack.
3472 This instruction is used in languages with destructors to ensure that
3473 proper cleanup is performed in the case of either a ``longjmp`` or a
3474 thrown exception. Additionally, this is important for implementation of
3475 '``catch``' clauses in high-level languages that support them.
3477 For the purposes of the SSA form, the definition of the value returned
3478 by the '``invoke``' instruction is deemed to occur on the edge from the
3479 current block to the "normal" label. If the callee unwinds then no
3480 return value is available.
3485 .. code-block:: llvm
3487 %retval = invoke i32 @Test(i32 15) to label %Continue
3488 unwind label %TestCleanup ; {i32}:retval set
3489 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3490 unwind label %TestCleanup ; {i32}:retval set
3494 '``resume``' Instruction
3495 ^^^^^^^^^^^^^^^^^^^^^^^^
3502 resume <type> <value>
3507 The '``resume``' instruction is a terminator instruction that has no
3513 The '``resume``' instruction requires one argument, which must have the
3514 same type as the result of any '``landingpad``' instruction in the same
3520 The '``resume``' instruction resumes propagation of an existing
3521 (in-flight) exception whose unwinding was interrupted with a
3522 :ref:`landingpad <i_landingpad>` instruction.
3527 .. code-block:: llvm
3529 resume { i8*, i32 } %exn
3533 '``unreachable``' Instruction
3534 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3546 The '``unreachable``' instruction has no defined semantics. This
3547 instruction is used to inform the optimizer that a particular portion of
3548 the code is not reachable. This can be used to indicate that the code
3549 after a no-return function cannot be reached, and other facts.
3554 The '``unreachable``' instruction has no defined semantics.
3561 Binary operators are used to do most of the computation in a program.
3562 They require two operands of the same type, execute an operation on
3563 them, and produce a single value. The operands might represent multiple
3564 data, as is the case with the :ref:`vector <t_vector>` data type. The
3565 result value has the same type as its operands.
3567 There are several different binary operators:
3571 '``add``' Instruction
3572 ^^^^^^^^^^^^^^^^^^^^^
3579 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3580 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3581 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3582 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3587 The '``add``' instruction returns the sum of its two operands.
3592 The two arguments to the '``add``' instruction must be
3593 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3594 arguments must have identical types.
3599 The value produced is the integer sum of the two operands.
3601 If the sum has unsigned overflow, the result returned is the
3602 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3605 Because LLVM integers use a two's complement representation, this
3606 instruction is appropriate for both signed and unsigned integers.
3608 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3609 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3610 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3611 unsigned and/or signed overflow, respectively, occurs.
3616 .. code-block:: llvm
3618 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3622 '``fadd``' Instruction
3623 ^^^^^^^^^^^^^^^^^^^^^^
3630 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3635 The '``fadd``' instruction returns the sum of its two operands.
3640 The two arguments to the '``fadd``' instruction must be :ref:`floating
3641 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3642 Both arguments must have identical types.
3647 The value produced is the floating point sum of the two operands. This
3648 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3649 which are optimization hints to enable otherwise unsafe floating point
3655 .. code-block:: llvm
3657 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3659 '``sub``' Instruction
3660 ^^^^^^^^^^^^^^^^^^^^^
3667 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3668 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3669 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3670 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3675 The '``sub``' instruction returns the difference of its two operands.
3677 Note that the '``sub``' instruction is used to represent the '``neg``'
3678 instruction present in most other intermediate representations.
3683 The two arguments to the '``sub``' instruction must be
3684 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3685 arguments must have identical types.
3690 The value produced is the integer difference of the two operands.
3692 If the difference has unsigned overflow, the result returned is the
3693 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3696 Because LLVM integers use a two's complement representation, this
3697 instruction is appropriate for both signed and unsigned integers.
3699 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3700 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3701 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3702 unsigned and/or signed overflow, respectively, occurs.
3707 .. code-block:: llvm
3709 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3710 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3714 '``fsub``' Instruction
3715 ^^^^^^^^^^^^^^^^^^^^^^
3722 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3727 The '``fsub``' instruction returns the difference of its two operands.
3729 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3730 instruction present in most other intermediate representations.
3735 The two arguments to the '``fsub``' instruction must be :ref:`floating
3736 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3737 Both arguments must have identical types.
3742 The value produced is the floating point difference of the two operands.
3743 This instruction can also take any number of :ref:`fast-math
3744 flags <fastmath>`, which are optimization hints to enable otherwise
3745 unsafe floating point optimizations:
3750 .. code-block:: llvm
3752 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3753 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3755 '``mul``' Instruction
3756 ^^^^^^^^^^^^^^^^^^^^^
3763 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3764 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3765 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3766 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3771 The '``mul``' instruction returns the product of its two operands.
3776 The two arguments to the '``mul``' instruction must be
3777 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3778 arguments must have identical types.
3783 The value produced is the integer product of the two operands.
3785 If the result of the multiplication has unsigned overflow, the result
3786 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3787 bit width of the result.
3789 Because LLVM integers use a two's complement representation, and the
3790 result is the same width as the operands, this instruction returns the
3791 correct result for both signed and unsigned integers. If a full product
3792 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3793 sign-extended or zero-extended as appropriate to the width of the full
3796 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3797 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3798 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3799 unsigned and/or signed overflow, respectively, occurs.
3804 .. code-block:: llvm
3806 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3810 '``fmul``' Instruction
3811 ^^^^^^^^^^^^^^^^^^^^^^
3818 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3823 The '``fmul``' instruction returns the product of its two operands.
3828 The two arguments to the '``fmul``' instruction must be :ref:`floating
3829 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3830 Both arguments must have identical types.
3835 The value produced is the floating point product of the two operands.
3836 This instruction can also take any number of :ref:`fast-math
3837 flags <fastmath>`, which are optimization hints to enable otherwise
3838 unsafe floating point optimizations:
3843 .. code-block:: llvm
3845 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3847 '``udiv``' Instruction
3848 ^^^^^^^^^^^^^^^^^^^^^^
3855 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3856 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3861 The '``udiv``' instruction returns the quotient of its two operands.
3866 The two arguments to the '``udiv``' instruction must be
3867 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3868 arguments must have identical types.
3873 The value produced is the unsigned integer quotient of the two operands.
3875 Note that unsigned integer division and signed integer division are
3876 distinct operations; for signed integer division, use '``sdiv``'.
3878 Division by zero leads to undefined behavior.
3880 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3881 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3882 such, "((a udiv exact b) mul b) == a").
3887 .. code-block:: llvm
3889 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3891 '``sdiv``' Instruction
3892 ^^^^^^^^^^^^^^^^^^^^^^
3899 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3900 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3905 The '``sdiv``' instruction returns the quotient of its two operands.
3910 The two arguments to the '``sdiv``' instruction must be
3911 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3912 arguments must have identical types.
3917 The value produced is the signed integer quotient of the two operands
3918 rounded towards zero.
3920 Note that signed integer division and unsigned integer division are
3921 distinct operations; for unsigned integer division, use '``udiv``'.
3923 Division by zero leads to undefined behavior. Overflow also leads to
3924 undefined behavior; this is a rare case, but can occur, for example, by
3925 doing a 32-bit division of -2147483648 by -1.
3927 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3928 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3933 .. code-block:: llvm
3935 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3939 '``fdiv``' Instruction
3940 ^^^^^^^^^^^^^^^^^^^^^^
3947 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3952 The '``fdiv``' instruction returns the quotient of its two operands.
3957 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3958 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3959 Both arguments must have identical types.
3964 The value produced is the floating point quotient of the two operands.
3965 This instruction can also take any number of :ref:`fast-math
3966 flags <fastmath>`, which are optimization hints to enable otherwise
3967 unsafe floating point optimizations:
3972 .. code-block:: llvm
3974 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3976 '``urem``' Instruction
3977 ^^^^^^^^^^^^^^^^^^^^^^
3984 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3989 The '``urem``' instruction returns the remainder from the unsigned
3990 division of its two arguments.
3995 The two arguments to the '``urem``' instruction must be
3996 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3997 arguments must have identical types.
4002 This instruction returns the unsigned integer *remainder* of a division.
4003 This instruction always performs an unsigned division to get the
4006 Note that unsigned integer remainder and signed integer remainder are
4007 distinct operations; for signed integer remainder, use '``srem``'.
4009 Taking the remainder of a division by zero leads to undefined behavior.
4014 .. code-block:: llvm
4016 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
4018 '``srem``' Instruction
4019 ^^^^^^^^^^^^^^^^^^^^^^
4026 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
4031 The '``srem``' instruction returns the remainder from the signed
4032 division of its two operands. This instruction can also take
4033 :ref:`vector <t_vector>` versions of the values in which case the elements
4039 The two arguments to the '``srem``' instruction must be
4040 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4041 arguments must have identical types.
4046 This instruction returns the *remainder* of a division (where the result
4047 is either zero or has the same sign as the dividend, ``op1``), not the
4048 *modulo* operator (where the result is either zero or has the same sign
4049 as the divisor, ``op2``) of a value. For more information about the
4050 difference, see `The Math
4051 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4052 table of how this is implemented in various languages, please see
4054 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4056 Note that signed integer remainder and unsigned integer remainder are
4057 distinct operations; for unsigned integer remainder, use '``urem``'.
4059 Taking the remainder of a division by zero leads to undefined behavior.
4060 Overflow also leads to undefined behavior; this is a rare case, but can
4061 occur, for example, by taking the remainder of a 32-bit division of
4062 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4063 rule lets srem be implemented using instructions that return both the
4064 result of the division and the remainder.)
4069 .. code-block:: llvm
4071 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4075 '``frem``' Instruction
4076 ^^^^^^^^^^^^^^^^^^^^^^
4083 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4088 The '``frem``' instruction returns the remainder from the division of
4094 The two arguments to the '``frem``' instruction must be :ref:`floating
4095 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4096 Both arguments must have identical types.
4101 This instruction returns the *remainder* of a division. The remainder
4102 has the same sign as the dividend. This instruction can also take any
4103 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4104 to enable otherwise unsafe floating point optimizations:
4109 .. code-block:: llvm
4111 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4115 Bitwise Binary Operations
4116 -------------------------
4118 Bitwise binary operators are used to do various forms of bit-twiddling
4119 in a program. They are generally very efficient instructions and can
4120 commonly be strength reduced from other instructions. They require two
4121 operands of the same type, execute an operation on them, and produce a
4122 single value. The resulting value is the same type as its operands.
4124 '``shl``' Instruction
4125 ^^^^^^^^^^^^^^^^^^^^^
4132 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4133 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4134 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4135 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4140 The '``shl``' instruction returns the first operand shifted to the left
4141 a specified number of bits.
4146 Both arguments to the '``shl``' instruction must be the same
4147 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4148 '``op2``' is treated as an unsigned value.
4153 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4154 where ``n`` is the width of the result. If ``op2`` is (statically or
4155 dynamically) negative or equal to or larger than the number of bits in
4156 ``op1``, the result is undefined. If the arguments are vectors, each
4157 vector element of ``op1`` is shifted by the corresponding shift amount
4160 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4161 value <poisonvalues>` if it shifts out any non-zero bits. If the
4162 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4163 value <poisonvalues>` if it shifts out any bits that disagree with the
4164 resultant sign bit. As such, NUW/NSW have the same semantics as they
4165 would if the shift were expressed as a mul instruction with the same
4166 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4171 .. code-block:: llvm
4173 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4174 <result> = shl i32 4, 2 ; yields {i32}: 16
4175 <result> = shl i32 1, 10 ; yields {i32}: 1024
4176 <result> = shl i32 1, 32 ; undefined
4177 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4179 '``lshr``' Instruction
4180 ^^^^^^^^^^^^^^^^^^^^^^
4187 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4188 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4193 The '``lshr``' instruction (logical shift right) returns the first
4194 operand shifted to the right a specified number of bits with zero fill.
4199 Both arguments to the '``lshr``' instruction must be the same
4200 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4201 '``op2``' is treated as an unsigned value.
4206 This instruction always performs a logical shift right operation. The
4207 most significant bits of the result will be filled with zero bits after
4208 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4209 than the number of bits in ``op1``, the result is undefined. If the
4210 arguments are vectors, each vector element of ``op1`` is shifted by the
4211 corresponding shift amount in ``op2``.
4213 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4214 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4220 .. code-block:: llvm
4222 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4223 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4224 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4225 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4226 <result> = lshr i32 1, 32 ; undefined
4227 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4229 '``ashr``' Instruction
4230 ^^^^^^^^^^^^^^^^^^^^^^
4237 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4238 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4243 The '``ashr``' instruction (arithmetic shift right) returns the first
4244 operand shifted to the right a specified number of bits with sign
4250 Both arguments to the '``ashr``' instruction must be the same
4251 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4252 '``op2``' is treated as an unsigned value.
4257 This instruction always performs an arithmetic shift right operation,
4258 The most significant bits of the result will be filled with the sign bit
4259 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4260 than the number of bits in ``op1``, the result is undefined. If the
4261 arguments are vectors, each vector element of ``op1`` is shifted by the
4262 corresponding shift amount in ``op2``.
4264 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4265 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4271 .. code-block:: llvm
4273 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4274 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4275 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4276 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4277 <result> = ashr i32 1, 32 ; undefined
4278 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4280 '``and``' Instruction
4281 ^^^^^^^^^^^^^^^^^^^^^
4288 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4293 The '``and``' instruction returns the bitwise logical and of its two
4299 The two arguments to the '``and``' instruction must be
4300 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4301 arguments must have identical types.
4306 The truth table used for the '``and``' instruction is:
4323 .. code-block:: llvm
4325 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4326 <result> = and i32 15, 40 ; yields {i32}:result = 8
4327 <result> = and i32 4, 8 ; yields {i32}:result = 0
4329 '``or``' Instruction
4330 ^^^^^^^^^^^^^^^^^^^^
4337 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4342 The '``or``' instruction returns the bitwise logical inclusive or of its
4348 The two arguments to the '``or``' instruction must be
4349 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4350 arguments must have identical types.
4355 The truth table used for the '``or``' instruction is:
4374 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4375 <result> = or i32 15, 40 ; yields {i32}:result = 47
4376 <result> = or i32 4, 8 ; yields {i32}:result = 12
4378 '``xor``' Instruction
4379 ^^^^^^^^^^^^^^^^^^^^^
4386 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4391 The '``xor``' instruction returns the bitwise logical exclusive or of
4392 its two operands. The ``xor`` is used to implement the "one's
4393 complement" operation, which is the "~" operator in C.
4398 The two arguments to the '``xor``' instruction must be
4399 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4400 arguments must have identical types.
4405 The truth table used for the '``xor``' instruction is:
4422 .. code-block:: llvm
4424 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4425 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4426 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4427 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4432 LLVM supports several instructions to represent vector operations in a
4433 target-independent manner. These instructions cover the element-access
4434 and vector-specific operations needed to process vectors effectively.
4435 While LLVM does directly support these vector operations, many
4436 sophisticated algorithms will want to use target-specific intrinsics to
4437 take full advantage of a specific target.
4439 .. _i_extractelement:
4441 '``extractelement``' Instruction
4442 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4449 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4454 The '``extractelement``' instruction extracts a single scalar element
4455 from a vector at a specified index.
4460 The first operand of an '``extractelement``' instruction is a value of
4461 :ref:`vector <t_vector>` type. The second operand is an index indicating
4462 the position from which to extract the element. The index may be a
4468 The result is a scalar of the same type as the element type of ``val``.
4469 Its value is the value at position ``idx`` of ``val``. If ``idx``
4470 exceeds the length of ``val``, the results are undefined.
4475 .. code-block:: llvm
4477 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4479 .. _i_insertelement:
4481 '``insertelement``' Instruction
4482 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4489 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4494 The '``insertelement``' instruction inserts a scalar element into a
4495 vector at a specified index.
4500 The first operand of an '``insertelement``' instruction is a value of
4501 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4502 type must equal the element type of the first operand. The third operand
4503 is an index indicating the position at which to insert the value. The
4504 index may be a variable.
4509 The result is a vector of the same type as ``val``. Its element values
4510 are those of ``val`` except at position ``idx``, where it gets the value
4511 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4517 .. code-block:: llvm
4519 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4521 .. _i_shufflevector:
4523 '``shufflevector``' Instruction
4524 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4531 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4536 The '``shufflevector``' instruction constructs a permutation of elements
4537 from two input vectors, returning a vector with the same element type as
4538 the input and length that is the same as the shuffle mask.
4543 The first two operands of a '``shufflevector``' instruction are vectors
4544 with the same type. The third argument is a shuffle mask whose element
4545 type is always 'i32'. The result of the instruction is a vector whose
4546 length is the same as the shuffle mask and whose element type is the
4547 same as the element type of the first two operands.
4549 The shuffle mask operand is required to be a constant vector with either
4550 constant integer or undef values.
4555 The elements of the two input vectors are numbered from left to right
4556 across both of the vectors. The shuffle mask operand specifies, for each
4557 element of the result vector, which element of the two input vectors the
4558 result element gets. The element selector may be undef (meaning "don't
4559 care") and the second operand may be undef if performing a shuffle from
4565 .. code-block:: llvm
4567 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4568 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4569 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4570 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4571 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4572 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4573 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4574 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4576 Aggregate Operations
4577 --------------------
4579 LLVM supports several instructions for working with
4580 :ref:`aggregate <t_aggregate>` values.
4584 '``extractvalue``' Instruction
4585 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4592 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4597 The '``extractvalue``' instruction extracts the value of a member field
4598 from an :ref:`aggregate <t_aggregate>` value.
4603 The first operand of an '``extractvalue``' instruction is a value of
4604 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4605 constant indices to specify which value to extract in a similar manner
4606 as indices in a '``getelementptr``' instruction.
4608 The major differences to ``getelementptr`` indexing are:
4610 - Since the value being indexed is not a pointer, the first index is
4611 omitted and assumed to be zero.
4612 - At least one index must be specified.
4613 - Not only struct indices but also array indices must be in bounds.
4618 The result is the value at the position in the aggregate specified by
4624 .. code-block:: llvm
4626 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4630 '``insertvalue``' Instruction
4631 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4638 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4643 The '``insertvalue``' instruction inserts a value into a member field in
4644 an :ref:`aggregate <t_aggregate>` value.
4649 The first operand of an '``insertvalue``' instruction is a value of
4650 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4651 a first-class value to insert. The following operands are constant
4652 indices indicating the position at which to insert the value in a
4653 similar manner as indices in a '``extractvalue``' instruction. The value
4654 to insert must have the same type as the value identified by the
4660 The result is an aggregate of the same type as ``val``. Its value is
4661 that of ``val`` except that the value at the position specified by the
4662 indices is that of ``elt``.
4667 .. code-block:: llvm
4669 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4670 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4671 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4675 Memory Access and Addressing Operations
4676 ---------------------------------------
4678 A key design point of an SSA-based representation is how it represents
4679 memory. In LLVM, no memory locations are in SSA form, which makes things
4680 very simple. This section describes how to read, write, and allocate
4685 '``alloca``' Instruction
4686 ^^^^^^^^^^^^^^^^^^^^^^^^
4693 <result> = alloca <type>[, inalloca][, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4698 The '``alloca``' instruction allocates memory on the stack frame of the
4699 currently executing function, to be automatically released when this
4700 function returns to its caller. The object is always allocated in the
4701 generic address space (address space zero).
4706 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4707 bytes of memory on the runtime stack, returning a pointer of the
4708 appropriate type to the program. If "NumElements" is specified, it is
4709 the number of elements allocated, otherwise "NumElements" is defaulted
4710 to be one. If a constant alignment is specified, the value result of the
4711 allocation is guaranteed to be aligned to at least that boundary. If not
4712 specified, or if zero, the target can choose to align the allocation on
4713 any convenient boundary compatible with the type.
4715 '``type``' may be any sized type.
4720 Memory is allocated; a pointer is returned. The operation is undefined
4721 if there is insufficient stack space for the allocation. '``alloca``'d
4722 memory is automatically released when the function returns. The
4723 '``alloca``' instruction is commonly used to represent automatic
4724 variables that must have an address available. When the function returns
4725 (either with the ``ret`` or ``resume`` instructions), the memory is
4726 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4727 The order in which memory is allocated (ie., which way the stack grows)
4733 .. code-block:: llvm
4735 %ptr = alloca i32 ; yields {i32*}:ptr
4736 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4737 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4738 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4742 '``load``' Instruction
4743 ^^^^^^^^^^^^^^^^^^^^^^
4750 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4751 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4752 !<index> = !{ i32 1 }
4757 The '``load``' instruction is used to read from memory.
4762 The argument to the ``load`` instruction specifies the memory address
4763 from which to load. The pointer must point to a :ref:`first
4764 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4765 then the optimizer is not allowed to modify the number or order of
4766 execution of this ``load`` with other :ref:`volatile
4767 operations <volatile>`.
4769 If the ``load`` is marked as ``atomic``, it takes an extra
4770 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4771 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4772 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4773 when they may see multiple atomic stores. The type of the pointee must
4774 be an integer type whose bit width is a power of two greater than or
4775 equal to eight and less than or equal to a target-specific size limit.
4776 ``align`` must be explicitly specified on atomic loads, and the load has
4777 undefined behavior if the alignment is not set to a value which is at
4778 least the size in bytes of the pointee. ``!nontemporal`` does not have
4779 any defined semantics for atomic loads.
4781 The optional constant ``align`` argument specifies the alignment of the
4782 operation (that is, the alignment of the memory address). A value of 0
4783 or an omitted ``align`` argument means that the operation has the ABI
4784 alignment for the target. It is the responsibility of the code emitter
4785 to ensure that the alignment information is correct. Overestimating the
4786 alignment results in undefined behavior. Underestimating the alignment
4787 may produce less efficient code. An alignment of 1 is always safe.
4789 The optional ``!nontemporal`` metadata must reference a single
4790 metadata name ``<index>`` corresponding to a metadata node with one
4791 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4792 metadata on the instruction tells the optimizer and code generator
4793 that this load is not expected to be reused in the cache. The code
4794 generator may select special instructions to save cache bandwidth, such
4795 as the ``MOVNT`` instruction on x86.
4797 The optional ``!invariant.load`` metadata must reference a single
4798 metadata name ``<index>`` corresponding to a metadata node with no
4799 entries. The existence of the ``!invariant.load`` metadata on the
4800 instruction tells the optimizer and code generator that this load
4801 address points to memory which does not change value during program
4802 execution. The optimizer may then move this load around, for example, by
4803 hoisting it out of loops using loop invariant code motion.
4808 The location of memory pointed to is loaded. If the value being loaded
4809 is of scalar type then the number of bytes read does not exceed the
4810 minimum number of bytes needed to hold all bits of the type. For
4811 example, loading an ``i24`` reads at most three bytes. When loading a
4812 value of a type like ``i20`` with a size that is not an integral number
4813 of bytes, the result is undefined if the value was not originally
4814 written using a store of the same type.
4819 .. code-block:: llvm
4821 %ptr = alloca i32 ; yields {i32*}:ptr
4822 store i32 3, i32* %ptr ; yields {void}
4823 %val = load i32* %ptr ; yields {i32}:val = i32 3
4827 '``store``' Instruction
4828 ^^^^^^^^^^^^^^^^^^^^^^^
4835 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4836 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4841 The '``store``' instruction is used to write to memory.
4846 There are two arguments to the ``store`` instruction: a value to store
4847 and an address at which to store it. The type of the ``<pointer>``
4848 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4849 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4850 then the optimizer is not allowed to modify the number or order of
4851 execution of this ``store`` with other :ref:`volatile
4852 operations <volatile>`.
4854 If the ``store`` is marked as ``atomic``, it takes an extra
4855 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4856 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4857 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4858 when they may see multiple atomic stores. The type of the pointee must
4859 be an integer type whose bit width is a power of two greater than or
4860 equal to eight and less than or equal to a target-specific size limit.
4861 ``align`` must be explicitly specified on atomic stores, and the store
4862 has undefined behavior if the alignment is not set to a value which is
4863 at least the size in bytes of the pointee. ``!nontemporal`` does not
4864 have any defined semantics for atomic stores.
4866 The optional constant ``align`` argument specifies the alignment of the
4867 operation (that is, the alignment of the memory address). A value of 0
4868 or an omitted ``align`` argument means that the operation has the ABI
4869 alignment for the target. It is the responsibility of the code emitter
4870 to ensure that the alignment information is correct. Overestimating the
4871 alignment results in undefined behavior. Underestimating the
4872 alignment may produce less efficient code. An alignment of 1 is always
4875 The optional ``!nontemporal`` metadata must reference a single metadata
4876 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4877 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4878 tells the optimizer and code generator that this load is not expected to
4879 be reused in the cache. The code generator may select special
4880 instructions to save cache bandwidth, such as the MOVNT instruction on
4886 The contents of memory are updated to contain ``<value>`` at the
4887 location specified by the ``<pointer>`` operand. If ``<value>`` is
4888 of scalar type then the number of bytes written does not exceed the
4889 minimum number of bytes needed to hold all bits of the type. For
4890 example, storing an ``i24`` writes at most three bytes. When writing a
4891 value of a type like ``i20`` with a size that is not an integral number
4892 of bytes, it is unspecified what happens to the extra bits that do not
4893 belong to the type, but they will typically be overwritten.
4898 .. code-block:: llvm
4900 %ptr = alloca i32 ; yields {i32*}:ptr
4901 store i32 3, i32* %ptr ; yields {void}
4902 %val = load i32* %ptr ; yields {i32}:val = i32 3
4906 '``fence``' Instruction
4907 ^^^^^^^^^^^^^^^^^^^^^^^
4914 fence [singlethread] <ordering> ; yields {void}
4919 The '``fence``' instruction is used to introduce happens-before edges
4925 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4926 defines what *synchronizes-with* edges they add. They can only be given
4927 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4932 A fence A which has (at least) ``release`` ordering semantics
4933 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4934 semantics if and only if there exist atomic operations X and Y, both
4935 operating on some atomic object M, such that A is sequenced before X, X
4936 modifies M (either directly or through some side effect of a sequence
4937 headed by X), Y is sequenced before B, and Y observes M. This provides a
4938 *happens-before* dependency between A and B. Rather than an explicit
4939 ``fence``, one (but not both) of the atomic operations X or Y might
4940 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4941 still *synchronize-with* the explicit ``fence`` and establish the
4942 *happens-before* edge.
4944 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4945 ``acquire`` and ``release`` semantics specified above, participates in
4946 the global program order of other ``seq_cst`` operations and/or fences.
4948 The optional ":ref:`singlethread <singlethread>`" argument specifies
4949 that the fence only synchronizes with other fences in the same thread.
4950 (This is useful for interacting with signal handlers.)
4955 .. code-block:: llvm
4957 fence acquire ; yields {void}
4958 fence singlethread seq_cst ; yields {void}
4962 '``cmpxchg``' Instruction
4963 ^^^^^^^^^^^^^^^^^^^^^^^^^
4970 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4975 The '``cmpxchg``' instruction is used to atomically modify memory. It
4976 loads a value in memory and compares it to a given value. If they are
4977 equal, it stores a new value into the memory.
4982 There are three arguments to the '``cmpxchg``' instruction: an address
4983 to operate on, a value to compare to the value currently be at that
4984 address, and a new value to place at that address if the compared values
4985 are equal. The type of '<cmp>' must be an integer type whose bit width
4986 is a power of two greater than or equal to eight and less than or equal
4987 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4988 type, and the type of '<pointer>' must be a pointer to that type. If the
4989 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4990 to modify the number or order of execution of this ``cmpxchg`` with
4991 other :ref:`volatile operations <volatile>`.
4993 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4994 synchronizes with other atomic operations.
4996 The optional "``singlethread``" argument declares that the ``cmpxchg``
4997 is only atomic with respect to code (usually signal handlers) running in
4998 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4999 respect to all other code in the system.
5001 The pointer passed into cmpxchg must have alignment greater than or
5002 equal to the size in memory of the operand.
5007 The contents of memory at the location specified by the '``<pointer>``'
5008 operand is read and compared to '``<cmp>``'; if the read value is the
5009 equal, '``<new>``' is written. The original value at the location is
5012 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
5013 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
5014 atomic load with an ordering parameter determined by dropping any
5015 ``release`` part of the ``cmpxchg``'s ordering.
5020 .. code-block:: llvm
5023 %orig = atomic load i32* %ptr unordered ; yields {i32}
5027 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5028 %squared = mul i32 %cmp, %cmp
5029 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
5030 %success = icmp eq i32 %cmp, %old
5031 br i1 %success, label %done, label %loop
5038 '``atomicrmw``' Instruction
5039 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5046 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
5051 The '``atomicrmw``' instruction is used to atomically modify memory.
5056 There are three arguments to the '``atomicrmw``' instruction: an
5057 operation to apply, an address whose value to modify, an argument to the
5058 operation. The operation must be one of the following keywords:
5072 The type of '<value>' must be an integer type whose bit width is a power
5073 of two greater than or equal to eight and less than or equal to a
5074 target-specific size limit. The type of the '``<pointer>``' operand must
5075 be a pointer to that type. If the ``atomicrmw`` is marked as
5076 ``volatile``, then the optimizer is not allowed to modify the number or
5077 order of execution of this ``atomicrmw`` with other :ref:`volatile
5078 operations <volatile>`.
5083 The contents of memory at the location specified by the '``<pointer>``'
5084 operand are atomically read, modified, and written back. The original
5085 value at the location is returned. The modification is specified by the
5088 - xchg: ``*ptr = val``
5089 - add: ``*ptr = *ptr + val``
5090 - sub: ``*ptr = *ptr - val``
5091 - and: ``*ptr = *ptr & val``
5092 - nand: ``*ptr = ~(*ptr & val)``
5093 - or: ``*ptr = *ptr | val``
5094 - xor: ``*ptr = *ptr ^ val``
5095 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5096 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5097 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5099 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5105 .. code-block:: llvm
5107 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5109 .. _i_getelementptr:
5111 '``getelementptr``' Instruction
5112 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5119 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5120 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5121 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5126 The '``getelementptr``' instruction is used to get the address of a
5127 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5128 address calculation only and does not access memory.
5133 The first argument is always a pointer or a vector of pointers, and
5134 forms the basis of the calculation. The remaining arguments are indices
5135 that indicate which of the elements of the aggregate object are indexed.
5136 The interpretation of each index is dependent on the type being indexed
5137 into. The first index always indexes the pointer value given as the
5138 first argument, the second index indexes a value of the type pointed to
5139 (not necessarily the value directly pointed to, since the first index
5140 can be non-zero), etc. The first type indexed into must be a pointer
5141 value, subsequent types can be arrays, vectors, and structs. Note that
5142 subsequent types being indexed into can never be pointers, since that
5143 would require loading the pointer before continuing calculation.
5145 The type of each index argument depends on the type it is indexing into.
5146 When indexing into a (optionally packed) structure, only ``i32`` integer
5147 **constants** are allowed (when using a vector of indices they must all
5148 be the **same** ``i32`` integer constant). When indexing into an array,
5149 pointer or vector, integers of any width are allowed, and they are not
5150 required to be constant. These integers are treated as signed values
5153 For example, let's consider a C code fragment and how it gets compiled
5169 int *foo(struct ST *s) {
5170 return &s[1].Z.B[5][13];
5173 The LLVM code generated by Clang is:
5175 .. code-block:: llvm
5177 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5178 %struct.ST = type { i32, double, %struct.RT }
5180 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5182 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5189 In the example above, the first index is indexing into the
5190 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5191 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5192 indexes into the third element of the structure, yielding a
5193 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5194 structure. The third index indexes into the second element of the
5195 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5196 dimensions of the array are subscripted into, yielding an '``i32``'
5197 type. The '``getelementptr``' instruction returns a pointer to this
5198 element, thus computing a value of '``i32*``' type.
5200 Note that it is perfectly legal to index partially through a structure,
5201 returning a pointer to an inner element. Because of this, the LLVM code
5202 for the given testcase is equivalent to:
5204 .. code-block:: llvm
5206 define i32* @foo(%struct.ST* %s) {
5207 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5208 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5209 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5210 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5211 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5215 If the ``inbounds`` keyword is present, the result value of the
5216 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5217 pointer is not an *in bounds* address of an allocated object, or if any
5218 of the addresses that would be formed by successive addition of the
5219 offsets implied by the indices to the base address with infinitely
5220 precise signed arithmetic are not an *in bounds* address of that
5221 allocated object. The *in bounds* addresses for an allocated object are
5222 all the addresses that point into the object, plus the address one byte
5223 past the end. In cases where the base is a vector of pointers the
5224 ``inbounds`` keyword applies to each of the computations element-wise.
5226 If the ``inbounds`` keyword is not present, the offsets are added to the
5227 base address with silently-wrapping two's complement arithmetic. If the
5228 offsets have a different width from the pointer, they are sign-extended
5229 or truncated to the width of the pointer. The result value of the
5230 ``getelementptr`` may be outside the object pointed to by the base
5231 pointer. The result value may not necessarily be used to access memory
5232 though, even if it happens to point into allocated storage. See the
5233 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5236 The getelementptr instruction is often confusing. For some more insight
5237 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5242 .. code-block:: llvm
5244 ; yields [12 x i8]*:aptr
5245 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5247 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5249 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5251 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5253 In cases where the pointer argument is a vector of pointers, each index
5254 must be a vector with the same number of elements. For example:
5256 .. code-block:: llvm
5258 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5260 Conversion Operations
5261 ---------------------
5263 The instructions in this category are the conversion instructions
5264 (casting) which all take a single operand and a type. They perform
5265 various bit conversions on the operand.
5267 '``trunc .. to``' Instruction
5268 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5275 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5280 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5285 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5286 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5287 of the same number of integers. The bit size of the ``value`` must be
5288 larger than the bit size of the destination type, ``ty2``. Equal sized
5289 types are not allowed.
5294 The '``trunc``' instruction truncates the high order bits in ``value``
5295 and converts the remaining bits to ``ty2``. Since the source size must
5296 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5297 It will always truncate bits.
5302 .. code-block:: llvm
5304 %X = trunc i32 257 to i8 ; yields i8:1
5305 %Y = trunc i32 123 to i1 ; yields i1:true
5306 %Z = trunc i32 122 to i1 ; yields i1:false
5307 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5309 '``zext .. to``' Instruction
5310 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5317 <result> = zext <ty> <value> to <ty2> ; yields ty2
5322 The '``zext``' instruction zero extends its operand to type ``ty2``.
5327 The '``zext``' instruction takes a value to cast, and a type to cast it
5328 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5329 the same number of integers. The bit size of the ``value`` must be
5330 smaller than the bit size of the destination type, ``ty2``.
5335 The ``zext`` fills the high order bits of the ``value`` with zero bits
5336 until it reaches the size of the destination type, ``ty2``.
5338 When zero extending from i1, the result will always be either 0 or 1.
5343 .. code-block:: llvm
5345 %X = zext i32 257 to i64 ; yields i64:257
5346 %Y = zext i1 true to i32 ; yields i32:1
5347 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5349 '``sext .. to``' Instruction
5350 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5357 <result> = sext <ty> <value> to <ty2> ; yields ty2
5362 The '``sext``' sign extends ``value`` to the type ``ty2``.
5367 The '``sext``' instruction takes a value to cast, and a type to cast it
5368 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5369 the same number of integers. The bit size of the ``value`` must be
5370 smaller than the bit size of the destination type, ``ty2``.
5375 The '``sext``' instruction performs a sign extension by copying the sign
5376 bit (highest order bit) of the ``value`` until it reaches the bit size
5377 of the type ``ty2``.
5379 When sign extending from i1, the extension always results in -1 or 0.
5384 .. code-block:: llvm
5386 %X = sext i8 -1 to i16 ; yields i16 :65535
5387 %Y = sext i1 true to i32 ; yields i32:-1
5388 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5390 '``fptrunc .. to``' Instruction
5391 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5398 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5403 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5408 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5409 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5410 The size of ``value`` must be larger than the size of ``ty2``. This
5411 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5416 The '``fptrunc``' instruction truncates a ``value`` from a larger
5417 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5418 point <t_floating>` type. If the value cannot fit within the
5419 destination type, ``ty2``, then the results are undefined.
5424 .. code-block:: llvm
5426 %X = fptrunc double 123.0 to float ; yields float:123.0
5427 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5429 '``fpext .. to``' Instruction
5430 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5437 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5442 The '``fpext``' extends a floating point ``value`` to a larger floating
5448 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5449 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5450 to. The source type must be smaller than the destination type.
5455 The '``fpext``' instruction extends the ``value`` from a smaller
5456 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5457 point <t_floating>` type. The ``fpext`` cannot be used to make a
5458 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5459 *no-op cast* for a floating point cast.
5464 .. code-block:: llvm
5466 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5467 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5469 '``fptoui .. to``' Instruction
5470 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5477 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5482 The '``fptoui``' converts a floating point ``value`` to its unsigned
5483 integer equivalent of type ``ty2``.
5488 The '``fptoui``' instruction takes a value to cast, which must be a
5489 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5490 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5491 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5492 type with the same number of elements as ``ty``
5497 The '``fptoui``' instruction converts its :ref:`floating
5498 point <t_floating>` operand into the nearest (rounding towards zero)
5499 unsigned integer value. If the value cannot fit in ``ty2``, the results
5505 .. code-block:: llvm
5507 %X = fptoui double 123.0 to i32 ; yields i32:123
5508 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5509 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5511 '``fptosi .. to``' Instruction
5512 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5519 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5524 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5525 ``value`` to type ``ty2``.
5530 The '``fptosi``' instruction takes a value to cast, which must be a
5531 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5532 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5533 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5534 type with the same number of elements as ``ty``
5539 The '``fptosi``' instruction converts its :ref:`floating
5540 point <t_floating>` operand into the nearest (rounding towards zero)
5541 signed integer value. If the value cannot fit in ``ty2``, the results
5547 .. code-block:: llvm
5549 %X = fptosi double -123.0 to i32 ; yields i32:-123
5550 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5551 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5553 '``uitofp .. to``' Instruction
5554 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5561 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5566 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5567 and converts that value to the ``ty2`` type.
5572 The '``uitofp``' instruction takes a value to cast, which must be a
5573 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5574 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5575 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5576 type with the same number of elements as ``ty``
5581 The '``uitofp``' instruction interprets its operand as an unsigned
5582 integer quantity and converts it to the corresponding floating point
5583 value. If the value cannot fit in the floating point value, the results
5589 .. code-block:: llvm
5591 %X = uitofp i32 257 to float ; yields float:257.0
5592 %Y = uitofp i8 -1 to double ; yields double:255.0
5594 '``sitofp .. to``' Instruction
5595 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5602 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5607 The '``sitofp``' instruction regards ``value`` as a signed integer and
5608 converts that value to the ``ty2`` type.
5613 The '``sitofp``' instruction takes a value to cast, which must be a
5614 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5615 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5616 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5617 type with the same number of elements as ``ty``
5622 The '``sitofp``' instruction interprets its operand as a signed integer
5623 quantity and converts it to the corresponding floating point value. If
5624 the value cannot fit in the floating point value, the results are
5630 .. code-block:: llvm
5632 %X = sitofp i32 257 to float ; yields float:257.0
5633 %Y = sitofp i8 -1 to double ; yields double:-1.0
5637 '``ptrtoint .. to``' Instruction
5638 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5645 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5650 The '``ptrtoint``' instruction converts the pointer or a vector of
5651 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5656 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5657 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5658 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5659 a vector of integers type.
5664 The '``ptrtoint``' instruction converts ``value`` to integer type
5665 ``ty2`` by interpreting the pointer value as an integer and either
5666 truncating or zero extending that value to the size of the integer type.
5667 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5668 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5669 the same size, then nothing is done (*no-op cast*) other than a type
5675 .. code-block:: llvm
5677 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5678 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5679 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5683 '``inttoptr .. to``' Instruction
5684 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5691 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5696 The '``inttoptr``' instruction converts an integer ``value`` to a
5697 pointer type, ``ty2``.
5702 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5703 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5709 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5710 applying either a zero extension or a truncation depending on the size
5711 of the integer ``value``. If ``value`` is larger than the size of a
5712 pointer then a truncation is done. If ``value`` is smaller than the size
5713 of a pointer then a zero extension is done. If they are the same size,
5714 nothing is done (*no-op cast*).
5719 .. code-block:: llvm
5721 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5722 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5723 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5724 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5728 '``bitcast .. to``' Instruction
5729 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5736 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5741 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5747 The '``bitcast``' instruction takes a value to cast, which must be a
5748 non-aggregate first class value, and a type to cast it to, which must
5749 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5750 bit sizes of ``value`` and the destination type, ``ty2``, must be
5751 identical. If the source type is a pointer, the destination type must
5752 also be a pointer of the same size. This instruction supports bitwise
5753 conversion of vectors to integers and to vectors of other types (as
5754 long as they have the same size).
5759 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5760 is always a *no-op cast* because no bits change with this
5761 conversion. The conversion is done as if the ``value`` had been stored
5762 to memory and read back as type ``ty2``. Pointer (or vector of
5763 pointers) types may only be converted to other pointer (or vector of
5764 pointers) types with the same address space through this instruction.
5765 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5766 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5771 .. code-block:: llvm
5773 %X = bitcast i8 255 to i8 ; yields i8 :-1
5774 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5775 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5776 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5778 .. _i_addrspacecast:
5780 '``addrspacecast .. to``' Instruction
5781 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5788 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5793 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5794 address space ``n`` to type ``pty2`` in address space ``m``.
5799 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5800 to cast and a pointer type to cast it to, which must have a different
5806 The '``addrspacecast``' instruction converts the pointer value
5807 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5808 value modification, depending on the target and the address space
5809 pair. Pointer conversions within the same address space must be
5810 performed with the ``bitcast`` instruction. Note that if the address space
5811 conversion is legal then both result and operand refer to the same memory
5817 .. code-block:: llvm
5819 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5820 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5821 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5828 The instructions in this category are the "miscellaneous" instructions,
5829 which defy better classification.
5833 '``icmp``' Instruction
5834 ^^^^^^^^^^^^^^^^^^^^^^
5841 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5846 The '``icmp``' instruction returns a boolean value or a vector of
5847 boolean values based on comparison of its two integer, integer vector,
5848 pointer, or pointer vector operands.
5853 The '``icmp``' instruction takes three operands. The first operand is
5854 the condition code indicating the kind of comparison to perform. It is
5855 not a value, just a keyword. The possible condition code are:
5858 #. ``ne``: not equal
5859 #. ``ugt``: unsigned greater than
5860 #. ``uge``: unsigned greater or equal
5861 #. ``ult``: unsigned less than
5862 #. ``ule``: unsigned less or equal
5863 #. ``sgt``: signed greater than
5864 #. ``sge``: signed greater or equal
5865 #. ``slt``: signed less than
5866 #. ``sle``: signed less or equal
5868 The remaining two arguments must be :ref:`integer <t_integer>` or
5869 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5870 must also be identical types.
5875 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5876 code given as ``cond``. The comparison performed always yields either an
5877 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5879 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5880 otherwise. No sign interpretation is necessary or performed.
5881 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5882 otherwise. No sign interpretation is necessary or performed.
5883 #. ``ugt``: interprets the operands as unsigned values and yields
5884 ``true`` if ``op1`` is greater than ``op2``.
5885 #. ``uge``: interprets the operands as unsigned values and yields
5886 ``true`` if ``op1`` is greater than or equal to ``op2``.
5887 #. ``ult``: interprets the operands as unsigned values and yields
5888 ``true`` if ``op1`` is less than ``op2``.
5889 #. ``ule``: interprets the operands as unsigned values and yields
5890 ``true`` if ``op1`` is less than or equal to ``op2``.
5891 #. ``sgt``: interprets the operands as signed values and yields ``true``
5892 if ``op1`` is greater than ``op2``.
5893 #. ``sge``: interprets the operands as signed values and yields ``true``
5894 if ``op1`` is greater than or equal to ``op2``.
5895 #. ``slt``: interprets the operands as signed values and yields ``true``
5896 if ``op1`` is less than ``op2``.
5897 #. ``sle``: interprets the operands as signed values and yields ``true``
5898 if ``op1`` is less than or equal to ``op2``.
5900 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5901 are compared as if they were integers.
5903 If the operands are integer vectors, then they are compared element by
5904 element. The result is an ``i1`` vector with the same number of elements
5905 as the values being compared. Otherwise, the result is an ``i1``.
5910 .. code-block:: llvm
5912 <result> = icmp eq i32 4, 5 ; yields: result=false
5913 <result> = icmp ne float* %X, %X ; yields: result=false
5914 <result> = icmp ult i16 4, 5 ; yields: result=true
5915 <result> = icmp sgt i16 4, 5 ; yields: result=false
5916 <result> = icmp ule i16 -4, 5 ; yields: result=false
5917 <result> = icmp sge i16 4, 5 ; yields: result=false
5919 Note that the code generator does not yet support vector types with the
5920 ``icmp`` instruction.
5924 '``fcmp``' Instruction
5925 ^^^^^^^^^^^^^^^^^^^^^^
5932 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5937 The '``fcmp``' instruction returns a boolean value or vector of boolean
5938 values based on comparison of its operands.
5940 If the operands are floating point scalars, then the result type is a
5941 boolean (:ref:`i1 <t_integer>`).
5943 If the operands are floating point vectors, then the result type is a
5944 vector of boolean with the same number of elements as the operands being
5950 The '``fcmp``' instruction takes three operands. The first operand is
5951 the condition code indicating the kind of comparison to perform. It is
5952 not a value, just a keyword. The possible condition code are:
5954 #. ``false``: no comparison, always returns false
5955 #. ``oeq``: ordered and equal
5956 #. ``ogt``: ordered and greater than
5957 #. ``oge``: ordered and greater than or equal
5958 #. ``olt``: ordered and less than
5959 #. ``ole``: ordered and less than or equal
5960 #. ``one``: ordered and not equal
5961 #. ``ord``: ordered (no nans)
5962 #. ``ueq``: unordered or equal
5963 #. ``ugt``: unordered or greater than
5964 #. ``uge``: unordered or greater than or equal
5965 #. ``ult``: unordered or less than
5966 #. ``ule``: unordered or less than or equal
5967 #. ``une``: unordered or not equal
5968 #. ``uno``: unordered (either nans)
5969 #. ``true``: no comparison, always returns true
5971 *Ordered* means that neither operand is a QNAN while *unordered* means
5972 that either operand may be a QNAN.
5974 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5975 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5976 type. They must have identical types.
5981 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5982 condition code given as ``cond``. If the operands are vectors, then the
5983 vectors are compared element by element. Each comparison performed
5984 always yields an :ref:`i1 <t_integer>` result, as follows:
5986 #. ``false``: always yields ``false``, regardless of operands.
5987 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5988 is equal to ``op2``.
5989 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5990 is greater than ``op2``.
5991 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5992 is greater than or equal to ``op2``.
5993 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5994 is less than ``op2``.
5995 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5996 is less than or equal to ``op2``.
5997 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5998 is not equal to ``op2``.
5999 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6000 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6002 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6003 greater than ``op2``.
6004 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6005 greater than or equal to ``op2``.
6006 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6008 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6009 less than or equal to ``op2``.
6010 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6011 not equal to ``op2``.
6012 #. ``uno``: yields ``true`` if either operand is a QNAN.
6013 #. ``true``: always yields ``true``, regardless of operands.
6018 .. code-block:: llvm
6020 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6021 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6022 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6023 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6025 Note that the code generator does not yet support vector types with the
6026 ``fcmp`` instruction.
6030 '``phi``' Instruction
6031 ^^^^^^^^^^^^^^^^^^^^^
6038 <result> = phi <ty> [ <val0>, <label0>], ...
6043 The '``phi``' instruction is used to implement the φ node in the SSA
6044 graph representing the function.
6049 The type of the incoming values is specified with the first type field.
6050 After this, the '``phi``' instruction takes a list of pairs as
6051 arguments, with one pair for each predecessor basic block of the current
6052 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6053 the value arguments to the PHI node. Only labels may be used as the
6056 There must be no non-phi instructions between the start of a basic block
6057 and the PHI instructions: i.e. PHI instructions must be first in a basic
6060 For the purposes of the SSA form, the use of each incoming value is
6061 deemed to occur on the edge from the corresponding predecessor block to
6062 the current block (but after any definition of an '``invoke``'
6063 instruction's return value on the same edge).
6068 At runtime, the '``phi``' instruction logically takes on the value
6069 specified by the pair corresponding to the predecessor basic block that
6070 executed just prior to the current block.
6075 .. code-block:: llvm
6077 Loop: ; Infinite loop that counts from 0 on up...
6078 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6079 %nextindvar = add i32 %indvar, 1
6084 '``select``' Instruction
6085 ^^^^^^^^^^^^^^^^^^^^^^^^
6092 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6094 selty is either i1 or {<N x i1>}
6099 The '``select``' instruction is used to choose one value based on a
6100 condition, without branching.
6105 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6106 values indicating the condition, and two values of the same :ref:`first
6107 class <t_firstclass>` type. If the val1/val2 are vectors and the
6108 condition is a scalar, then entire vectors are selected, not individual
6114 If the condition is an i1 and it evaluates to 1, the instruction returns
6115 the first value argument; otherwise, it returns the second value
6118 If the condition is a vector of i1, then the value arguments must be
6119 vectors of the same size, and the selection is done element by element.
6124 .. code-block:: llvm
6126 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6130 '``call``' Instruction
6131 ^^^^^^^^^^^^^^^^^^^^^^
6138 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6143 The '``call``' instruction represents a simple function call.
6148 This instruction requires several arguments:
6150 #. The optional "tail" marker indicates that the callee function does
6151 not access any allocas or varargs in the caller. Note that calls may
6152 be marked "tail" even if they do not occur before a
6153 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6154 function call is eligible for tail call optimization, but `might not
6155 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6156 The code generator may optimize calls marked "tail" with either 1)
6157 automatic `sibling call
6158 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6159 callee have matching signatures, or 2) forced tail call optimization
6160 when the following extra requirements are met:
6162 - Caller and callee both have the calling convention ``fastcc``.
6163 - The call is in tail position (ret immediately follows call and ret
6164 uses value of call or is void).
6165 - Option ``-tailcallopt`` is enabled, or
6166 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6167 - `Platform specific constraints are
6168 met. <CodeGenerator.html#tailcallopt>`_
6170 #. The optional "cconv" marker indicates which :ref:`calling
6171 convention <callingconv>` the call should use. If none is
6172 specified, the call defaults to using C calling conventions. The
6173 calling convention of the call must match the calling convention of
6174 the target function, or else the behavior is undefined.
6175 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6176 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6178 #. '``ty``': the type of the call instruction itself which is also the
6179 type of the return value. Functions that return no value are marked
6181 #. '``fnty``': shall be the signature of the pointer to function value
6182 being invoked. The argument types must match the types implied by
6183 this signature. This type can be omitted if the function is not
6184 varargs and if the function type does not return a pointer to a
6186 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6187 be invoked. In most cases, this is a direct function invocation, but
6188 indirect ``call``'s are just as possible, calling an arbitrary pointer
6190 #. '``function args``': argument list whose types match the function
6191 signature argument types and parameter attributes. All arguments must
6192 be of :ref:`first class <t_firstclass>` type. If the function signature
6193 indicates the function accepts a variable number of arguments, the
6194 extra arguments can be specified.
6195 #. The optional :ref:`function attributes <fnattrs>` list. Only
6196 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6197 attributes are valid here.
6202 The '``call``' instruction is used to cause control flow to transfer to
6203 a specified function, with its incoming arguments bound to the specified
6204 values. Upon a '``ret``' instruction in the called function, control
6205 flow continues with the instruction after the function call, and the
6206 return value of the function is bound to the result argument.
6211 .. code-block:: llvm
6213 %retval = call i32 @test(i32 %argc)
6214 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6215 %X = tail call i32 @foo() ; yields i32
6216 %Y = tail call fastcc i32 @foo() ; yields i32
6217 call void %foo(i8 97 signext)
6219 %struct.A = type { i32, i8 }
6220 %r = call %struct.A @foo() ; yields { 32, i8 }
6221 %gr = extractvalue %struct.A %r, 0 ; yields i32
6222 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6223 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6224 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6226 llvm treats calls to some functions with names and arguments that match
6227 the standard C99 library as being the C99 library functions, and may
6228 perform optimizations or generate code for them under that assumption.
6229 This is something we'd like to change in the future to provide better
6230 support for freestanding environments and non-C-based languages.
6234 '``va_arg``' Instruction
6235 ^^^^^^^^^^^^^^^^^^^^^^^^
6242 <resultval> = va_arg <va_list*> <arglist>, <argty>
6247 The '``va_arg``' instruction is used to access arguments passed through
6248 the "variable argument" area of a function call. It is used to implement
6249 the ``va_arg`` macro in C.
6254 This instruction takes a ``va_list*`` value and the type of the
6255 argument. It returns a value of the specified argument type and
6256 increments the ``va_list`` to point to the next argument. The actual
6257 type of ``va_list`` is target specific.
6262 The '``va_arg``' instruction loads an argument of the specified type
6263 from the specified ``va_list`` and causes the ``va_list`` to point to
6264 the next argument. For more information, see the variable argument
6265 handling :ref:`Intrinsic Functions <int_varargs>`.
6267 It is legal for this instruction to be called in a function which does
6268 not take a variable number of arguments, for example, the ``vfprintf``
6271 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6272 function <intrinsics>` because it takes a type as an argument.
6277 See the :ref:`variable argument processing <int_varargs>` section.
6279 Note that the code generator does not yet fully support va\_arg on many
6280 targets. Also, it does not currently support va\_arg with aggregate
6281 types on any target.
6285 '``landingpad``' Instruction
6286 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6293 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6294 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6296 <clause> := catch <type> <value>
6297 <clause> := filter <array constant type> <array constant>
6302 The '``landingpad``' instruction is used by `LLVM's exception handling
6303 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6304 is a landing pad --- one where the exception lands, and corresponds to the
6305 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6306 defines values supplied by the personality function (``pers_fn``) upon
6307 re-entry to the function. The ``resultval`` has the type ``resultty``.
6312 This instruction takes a ``pers_fn`` value. This is the personality
6313 function associated with the unwinding mechanism. The optional
6314 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6316 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6317 contains the global variable representing the "type" that may be caught
6318 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6319 clause takes an array constant as its argument. Use
6320 "``[0 x i8**] undef``" for a filter which cannot throw. The
6321 '``landingpad``' instruction must contain *at least* one ``clause`` or
6322 the ``cleanup`` flag.
6327 The '``landingpad``' instruction defines the values which are set by the
6328 personality function (``pers_fn``) upon re-entry to the function, and
6329 therefore the "result type" of the ``landingpad`` instruction. As with
6330 calling conventions, how the personality function results are
6331 represented in LLVM IR is target specific.
6333 The clauses are applied in order from top to bottom. If two
6334 ``landingpad`` instructions are merged together through inlining, the
6335 clauses from the calling function are appended to the list of clauses.
6336 When the call stack is being unwound due to an exception being thrown,
6337 the exception is compared against each ``clause`` in turn. If it doesn't
6338 match any of the clauses, and the ``cleanup`` flag is not set, then
6339 unwinding continues further up the call stack.
6341 The ``landingpad`` instruction has several restrictions:
6343 - A landing pad block is a basic block which is the unwind destination
6344 of an '``invoke``' instruction.
6345 - A landing pad block must have a '``landingpad``' instruction as its
6346 first non-PHI instruction.
6347 - There can be only one '``landingpad``' instruction within the landing
6349 - A basic block that is not a landing pad block may not include a
6350 '``landingpad``' instruction.
6351 - All '``landingpad``' instructions in a function must have the same
6352 personality function.
6357 .. code-block:: llvm
6359 ;; A landing pad which can catch an integer.
6360 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6362 ;; A landing pad that is a cleanup.
6363 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6365 ;; A landing pad which can catch an integer and can only throw a double.
6366 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6368 filter [1 x i8**] [@_ZTId]
6375 LLVM supports the notion of an "intrinsic function". These functions
6376 have well known names and semantics and are required to follow certain
6377 restrictions. Overall, these intrinsics represent an extension mechanism
6378 for the LLVM language that does not require changing all of the
6379 transformations in LLVM when adding to the language (or the bitcode
6380 reader/writer, the parser, etc...).
6382 Intrinsic function names must all start with an "``llvm.``" prefix. This
6383 prefix is reserved in LLVM for intrinsic names; thus, function names may
6384 not begin with this prefix. Intrinsic functions must always be external
6385 functions: you cannot define the body of intrinsic functions. Intrinsic
6386 functions may only be used in call or invoke instructions: it is illegal
6387 to take the address of an intrinsic function. Additionally, because
6388 intrinsic functions are part of the LLVM language, it is required if any
6389 are added that they be documented here.
6391 Some intrinsic functions can be overloaded, i.e., the intrinsic
6392 represents a family of functions that perform the same operation but on
6393 different data types. Because LLVM can represent over 8 million
6394 different integer types, overloading is used commonly to allow an
6395 intrinsic function to operate on any integer type. One or more of the
6396 argument types or the result type can be overloaded to accept any
6397 integer type. Argument types may also be defined as exactly matching a
6398 previous argument's type or the result type. This allows an intrinsic
6399 function which accepts multiple arguments, but needs all of them to be
6400 of the same type, to only be overloaded with respect to a single
6401 argument or the result.
6403 Overloaded intrinsics will have the names of its overloaded argument
6404 types encoded into its function name, each preceded by a period. Only
6405 those types which are overloaded result in a name suffix. Arguments
6406 whose type is matched against another type do not. For example, the
6407 ``llvm.ctpop`` function can take an integer of any width and returns an
6408 integer of exactly the same integer width. This leads to a family of
6409 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6410 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6411 overloaded, and only one type suffix is required. Because the argument's
6412 type is matched against the return type, it does not require its own
6415 To learn how to add an intrinsic function, please see the `Extending
6416 LLVM Guide <ExtendingLLVM.html>`_.
6420 Variable Argument Handling Intrinsics
6421 -------------------------------------
6423 Variable argument support is defined in LLVM with the
6424 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6425 functions. These functions are related to the similarly named macros
6426 defined in the ``<stdarg.h>`` header file.
6428 All of these functions operate on arguments that use a target-specific
6429 value type "``va_list``". The LLVM assembly language reference manual
6430 does not define what this type is, so all transformations should be
6431 prepared to handle these functions regardless of the type used.
6433 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6434 variable argument handling intrinsic functions are used.
6436 .. code-block:: llvm
6438 define i32 @test(i32 %X, ...) {
6439 ; Initialize variable argument processing
6441 %ap2 = bitcast i8** %ap to i8*
6442 call void @llvm.va_start(i8* %ap2)
6444 ; Read a single integer argument
6445 %tmp = va_arg i8** %ap, i32
6447 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6449 %aq2 = bitcast i8** %aq to i8*
6450 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6451 call void @llvm.va_end(i8* %aq2)
6453 ; Stop processing of arguments.
6454 call void @llvm.va_end(i8* %ap2)
6458 declare void @llvm.va_start(i8*)
6459 declare void @llvm.va_copy(i8*, i8*)
6460 declare void @llvm.va_end(i8*)
6464 '``llvm.va_start``' Intrinsic
6465 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6472 declare void @llvm.va_start(i8* <arglist>)
6477 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6478 subsequent use by ``va_arg``.
6483 The argument is a pointer to a ``va_list`` element to initialize.
6488 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6489 available in C. In a target-dependent way, it initializes the
6490 ``va_list`` element to which the argument points, so that the next call
6491 to ``va_arg`` will produce the first variable argument passed to the
6492 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6493 to know the last argument of the function as the compiler can figure
6496 '``llvm.va_end``' Intrinsic
6497 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6504 declare void @llvm.va_end(i8* <arglist>)
6509 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6510 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6515 The argument is a pointer to a ``va_list`` to destroy.
6520 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6521 available in C. In a target-dependent way, it destroys the ``va_list``
6522 element to which the argument points. Calls to
6523 :ref:`llvm.va_start <int_va_start>` and
6524 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6529 '``llvm.va_copy``' Intrinsic
6530 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6537 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6542 The '``llvm.va_copy``' intrinsic copies the current argument position
6543 from the source argument list to the destination argument list.
6548 The first argument is a pointer to a ``va_list`` element to initialize.
6549 The second argument is a pointer to a ``va_list`` element to copy from.
6554 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6555 available in C. In a target-dependent way, it copies the source
6556 ``va_list`` element into the destination ``va_list`` element. This
6557 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6558 arbitrarily complex and require, for example, memory allocation.
6560 Accurate Garbage Collection Intrinsics
6561 --------------------------------------
6563 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6564 (GC) requires the implementation and generation of these intrinsics.
6565 These intrinsics allow identification of :ref:`GC roots on the
6566 stack <int_gcroot>`, as well as garbage collector implementations that
6567 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6568 Front-ends for type-safe garbage collected languages should generate
6569 these intrinsics to make use of the LLVM garbage collectors. For more
6570 details, see `Accurate Garbage Collection with
6571 LLVM <GarbageCollection.html>`_.
6573 The garbage collection intrinsics only operate on objects in the generic
6574 address space (address space zero).
6578 '``llvm.gcroot``' Intrinsic
6579 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6586 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6591 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6592 the code generator, and allows some metadata to be associated with it.
6597 The first argument specifies the address of a stack object that contains
6598 the root pointer. The second pointer (which must be either a constant or
6599 a global value address) contains the meta-data to be associated with the
6605 At runtime, a call to this intrinsic stores a null pointer into the
6606 "ptrloc" location. At compile-time, the code generator generates
6607 information to allow the runtime to find the pointer at GC safe points.
6608 The '``llvm.gcroot``' intrinsic may only be used in a function which
6609 :ref:`specifies a GC algorithm <gc>`.
6613 '``llvm.gcread``' Intrinsic
6614 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6621 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6626 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6627 locations, allowing garbage collector implementations that require read
6633 The second argument is the address to read from, which should be an
6634 address allocated from the garbage collector. The first object is a
6635 pointer to the start of the referenced object, if needed by the language
6636 runtime (otherwise null).
6641 The '``llvm.gcread``' intrinsic has the same semantics as a load
6642 instruction, but may be replaced with substantially more complex code by
6643 the garbage collector runtime, as needed. The '``llvm.gcread``'
6644 intrinsic may only be used in a function which :ref:`specifies a GC
6649 '``llvm.gcwrite``' Intrinsic
6650 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6657 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6662 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6663 locations, allowing garbage collector implementations that require write
6664 barriers (such as generational or reference counting collectors).
6669 The first argument is the reference to store, the second is the start of
6670 the object to store it to, and the third is the address of the field of
6671 Obj to store to. If the runtime does not require a pointer to the
6672 object, Obj may be null.
6677 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6678 instruction, but may be replaced with substantially more complex code by
6679 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6680 intrinsic may only be used in a function which :ref:`specifies a GC
6683 Code Generator Intrinsics
6684 -------------------------
6686 These intrinsics are provided by LLVM to expose special features that
6687 may only be implemented with code generator support.
6689 '``llvm.returnaddress``' Intrinsic
6690 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6697 declare i8 *@llvm.returnaddress(i32 <level>)
6702 The '``llvm.returnaddress``' intrinsic attempts to compute a
6703 target-specific value indicating the return address of the current
6704 function or one of its callers.
6709 The argument to this intrinsic indicates which function to return the
6710 address for. Zero indicates the calling function, one indicates its
6711 caller, etc. The argument is **required** to be a constant integer
6717 The '``llvm.returnaddress``' intrinsic either returns a pointer
6718 indicating the return address of the specified call frame, or zero if it
6719 cannot be identified. The value returned by this intrinsic is likely to
6720 be incorrect or 0 for arguments other than zero, so it should only be
6721 used for debugging purposes.
6723 Note that calling this intrinsic does not prevent function inlining or
6724 other aggressive transformations, so the value returned may not be that
6725 of the obvious source-language caller.
6727 '``llvm.frameaddress``' Intrinsic
6728 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6735 declare i8* @llvm.frameaddress(i32 <level>)
6740 The '``llvm.frameaddress``' intrinsic attempts to return the
6741 target-specific frame pointer value for the specified stack frame.
6746 The argument to this intrinsic indicates which function to return the
6747 frame pointer for. Zero indicates the calling function, one indicates
6748 its caller, etc. The argument is **required** to be a constant integer
6754 The '``llvm.frameaddress``' intrinsic either returns a pointer
6755 indicating the frame address of the specified call frame, or zero if it
6756 cannot be identified. The value returned by this intrinsic is likely to
6757 be incorrect or 0 for arguments other than zero, so it should only be
6758 used for debugging purposes.
6760 Note that calling this intrinsic does not prevent function inlining or
6761 other aggressive transformations, so the value returned may not be that
6762 of the obvious source-language caller.
6766 '``llvm.stacksave``' Intrinsic
6767 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6774 declare i8* @llvm.stacksave()
6779 The '``llvm.stacksave``' intrinsic is used to remember the current state
6780 of the function stack, for use with
6781 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6782 implementing language features like scoped automatic variable sized
6788 This intrinsic returns a opaque pointer value that can be passed to
6789 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6790 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6791 ``llvm.stacksave``, it effectively restores the state of the stack to
6792 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6793 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6794 were allocated after the ``llvm.stacksave`` was executed.
6796 .. _int_stackrestore:
6798 '``llvm.stackrestore``' Intrinsic
6799 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6806 declare void @llvm.stackrestore(i8* %ptr)
6811 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6812 the function stack to the state it was in when the corresponding
6813 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6814 useful for implementing language features like scoped automatic variable
6815 sized arrays in C99.
6820 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6822 '``llvm.prefetch``' Intrinsic
6823 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6830 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6835 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6836 insert a prefetch instruction if supported; otherwise, it is a noop.
6837 Prefetches have no effect on the behavior of the program but can change
6838 its performance characteristics.
6843 ``address`` is the address to be prefetched, ``rw`` is the specifier
6844 determining if the fetch should be for a read (0) or write (1), and
6845 ``locality`` is a temporal locality specifier ranging from (0) - no
6846 locality, to (3) - extremely local keep in cache. The ``cache type``
6847 specifies whether the prefetch is performed on the data (1) or
6848 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6849 arguments must be constant integers.
6854 This intrinsic does not modify the behavior of the program. In
6855 particular, prefetches cannot trap and do not produce a value. On
6856 targets that support this intrinsic, the prefetch can provide hints to
6857 the processor cache for better performance.
6859 '``llvm.pcmarker``' Intrinsic
6860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6867 declare void @llvm.pcmarker(i32 <id>)
6872 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6873 Counter (PC) in a region of code to simulators and other tools. The
6874 method is target specific, but it is expected that the marker will use
6875 exported symbols to transmit the PC of the marker. The marker makes no
6876 guarantees that it will remain with any specific instruction after
6877 optimizations. It is possible that the presence of a marker will inhibit
6878 optimizations. The intended use is to be inserted after optimizations to
6879 allow correlations of simulation runs.
6884 ``id`` is a numerical id identifying the marker.
6889 This intrinsic does not modify the behavior of the program. Backends
6890 that do not support this intrinsic may ignore it.
6892 '``llvm.readcyclecounter``' Intrinsic
6893 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6900 declare i64 @llvm.readcyclecounter()
6905 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6906 counter register (or similar low latency, high accuracy clocks) on those
6907 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6908 should map to RPCC. As the backing counters overflow quickly (on the
6909 order of 9 seconds on alpha), this should only be used for small
6915 When directly supported, reading the cycle counter should not modify any
6916 memory. Implementations are allowed to either return a application
6917 specific value or a system wide value. On backends without support, this
6918 is lowered to a constant 0.
6920 Note that runtime support may be conditional on the privilege-level code is
6921 running at and the host platform.
6923 Standard C Library Intrinsics
6924 -----------------------------
6926 LLVM provides intrinsics for a few important standard C library
6927 functions. These intrinsics allow source-language front-ends to pass
6928 information about the alignment of the pointer arguments to the code
6929 generator, providing opportunity for more efficient code generation.
6933 '``llvm.memcpy``' Intrinsic
6934 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6939 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6940 integer bit width and for different address spaces. Not all targets
6941 support all bit widths however.
6945 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6946 i32 <len>, i32 <align>, i1 <isvolatile>)
6947 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6948 i64 <len>, i32 <align>, i1 <isvolatile>)
6953 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6954 source location to the destination location.
6956 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6957 intrinsics do not return a value, takes extra alignment/isvolatile
6958 arguments and the pointers can be in specified address spaces.
6963 The first argument is a pointer to the destination, the second is a
6964 pointer to the source. The third argument is an integer argument
6965 specifying the number of bytes to copy, the fourth argument is the
6966 alignment of the source and destination locations, and the fifth is a
6967 boolean indicating a volatile access.
6969 If the call to this intrinsic has an alignment value that is not 0 or 1,
6970 then the caller guarantees that both the source and destination pointers
6971 are aligned to that boundary.
6973 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6974 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6975 very cleanly specified and it is unwise to depend on it.
6980 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6981 source location to the destination location, which are not allowed to
6982 overlap. It copies "len" bytes of memory over. If the argument is known
6983 to be aligned to some boundary, this can be specified as the fourth
6984 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
6986 '``llvm.memmove``' Intrinsic
6987 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6992 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6993 bit width and for different address space. Not all targets support all
6998 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6999 i32 <len>, i32 <align>, i1 <isvolatile>)
7000 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7001 i64 <len>, i32 <align>, i1 <isvolatile>)
7006 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7007 source location to the destination location. It is similar to the
7008 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7011 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7012 intrinsics do not return a value, takes extra alignment/isvolatile
7013 arguments and the pointers can be in specified address spaces.
7018 The first argument is a pointer to the destination, the second is a
7019 pointer to the source. The third argument is an integer argument
7020 specifying the number of bytes to copy, the fourth argument is the
7021 alignment of the source and destination locations, and the fifth is a
7022 boolean indicating a volatile access.
7024 If the call to this intrinsic has an alignment value that is not 0 or 1,
7025 then the caller guarantees that the source and destination pointers are
7026 aligned to that boundary.
7028 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7029 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7030 not very cleanly specified and it is unwise to depend on it.
7035 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7036 source location to the destination location, which may overlap. It
7037 copies "len" bytes of memory over. If the argument is known to be
7038 aligned to some boundary, this can be specified as the fourth argument,
7039 otherwise it should be set to 0 or 1 (both meaning no alignment).
7041 '``llvm.memset.*``' Intrinsics
7042 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7047 This is an overloaded intrinsic. You can use llvm.memset on any integer
7048 bit width and for different address spaces. However, not all targets
7049 support all bit widths.
7053 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7054 i32 <len>, i32 <align>, i1 <isvolatile>)
7055 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7056 i64 <len>, i32 <align>, i1 <isvolatile>)
7061 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7062 particular byte value.
7064 Note that, unlike the standard libc function, the ``llvm.memset``
7065 intrinsic does not return a value and takes extra alignment/volatile
7066 arguments. Also, the destination can be in an arbitrary address space.
7071 The first argument is a pointer to the destination to fill, the second
7072 is the byte value with which to fill it, the third argument is an
7073 integer argument specifying the number of bytes to fill, and the fourth
7074 argument is the known alignment of the destination location.
7076 If the call to this intrinsic has an alignment value that is not 0 or 1,
7077 then the caller guarantees that the destination pointer is aligned to
7080 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7081 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7082 very cleanly specified and it is unwise to depend on it.
7087 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7088 at the destination location. If the argument is known to be aligned to
7089 some boundary, this can be specified as the fourth argument, otherwise
7090 it should be set to 0 or 1 (both meaning no alignment).
7092 '``llvm.sqrt.*``' Intrinsic
7093 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7098 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7099 floating point or vector of floating point type. Not all targets support
7104 declare float @llvm.sqrt.f32(float %Val)
7105 declare double @llvm.sqrt.f64(double %Val)
7106 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7107 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7108 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7113 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7114 returning the same value as the libm '``sqrt``' functions would. Unlike
7115 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7116 negative numbers other than -0.0 (which allows for better optimization,
7117 because there is no need to worry about errno being set).
7118 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7123 The argument and return value are floating point numbers of the same
7129 This function returns the sqrt of the specified operand if it is a
7130 nonnegative floating point number.
7132 '``llvm.powi.*``' Intrinsic
7133 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7138 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7139 floating point or vector of floating point type. Not all targets support
7144 declare float @llvm.powi.f32(float %Val, i32 %power)
7145 declare double @llvm.powi.f64(double %Val, i32 %power)
7146 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7147 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7148 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7153 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7154 specified (positive or negative) power. The order of evaluation of
7155 multiplications is not defined. When a vector of floating point type is
7156 used, the second argument remains a scalar integer value.
7161 The second argument is an integer power, and the first is a value to
7162 raise to that power.
7167 This function returns the first value raised to the second power with an
7168 unspecified sequence of rounding operations.
7170 '``llvm.sin.*``' Intrinsic
7171 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7176 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7177 floating point or vector of floating point type. Not all targets support
7182 declare float @llvm.sin.f32(float %Val)
7183 declare double @llvm.sin.f64(double %Val)
7184 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7185 declare fp128 @llvm.sin.f128(fp128 %Val)
7186 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7191 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7196 The argument and return value are floating point numbers of the same
7202 This function returns the sine of the specified operand, returning the
7203 same values as the libm ``sin`` functions would, and handles error
7204 conditions in the same way.
7206 '``llvm.cos.*``' Intrinsic
7207 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7212 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7213 floating point or vector of floating point type. Not all targets support
7218 declare float @llvm.cos.f32(float %Val)
7219 declare double @llvm.cos.f64(double %Val)
7220 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7221 declare fp128 @llvm.cos.f128(fp128 %Val)
7222 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7227 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7232 The argument and return value are floating point numbers of the same
7238 This function returns the cosine of the specified operand, returning the
7239 same values as the libm ``cos`` functions would, and handles error
7240 conditions in the same way.
7242 '``llvm.pow.*``' Intrinsic
7243 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7248 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7249 floating point or vector of floating point type. Not all targets support
7254 declare float @llvm.pow.f32(float %Val, float %Power)
7255 declare double @llvm.pow.f64(double %Val, double %Power)
7256 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7257 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7258 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7263 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7264 specified (positive or negative) power.
7269 The second argument is a floating point power, and the first is a value
7270 to raise to that power.
7275 This function returns the first value raised to the second power,
7276 returning the same values as the libm ``pow`` functions would, and
7277 handles error conditions in the same way.
7279 '``llvm.exp.*``' Intrinsic
7280 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7285 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7286 floating point or vector of floating point type. Not all targets support
7291 declare float @llvm.exp.f32(float %Val)
7292 declare double @llvm.exp.f64(double %Val)
7293 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7294 declare fp128 @llvm.exp.f128(fp128 %Val)
7295 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7300 The '``llvm.exp.*``' intrinsics perform the exp function.
7305 The argument and return value are floating point numbers of the same
7311 This function returns the same values as the libm ``exp`` functions
7312 would, and handles error conditions in the same way.
7314 '``llvm.exp2.*``' Intrinsic
7315 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7320 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7321 floating point or vector of floating point type. Not all targets support
7326 declare float @llvm.exp2.f32(float %Val)
7327 declare double @llvm.exp2.f64(double %Val)
7328 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7329 declare fp128 @llvm.exp2.f128(fp128 %Val)
7330 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7335 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7340 The argument and return value are floating point numbers of the same
7346 This function returns the same values as the libm ``exp2`` functions
7347 would, and handles error conditions in the same way.
7349 '``llvm.log.*``' Intrinsic
7350 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7355 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7356 floating point or vector of floating point type. Not all targets support
7361 declare float @llvm.log.f32(float %Val)
7362 declare double @llvm.log.f64(double %Val)
7363 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7364 declare fp128 @llvm.log.f128(fp128 %Val)
7365 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7370 The '``llvm.log.*``' intrinsics perform the log function.
7375 The argument and return value are floating point numbers of the same
7381 This function returns the same values as the libm ``log`` functions
7382 would, and handles error conditions in the same way.
7384 '``llvm.log10.*``' Intrinsic
7385 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7390 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7391 floating point or vector of floating point type. Not all targets support
7396 declare float @llvm.log10.f32(float %Val)
7397 declare double @llvm.log10.f64(double %Val)
7398 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7399 declare fp128 @llvm.log10.f128(fp128 %Val)
7400 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7405 The '``llvm.log10.*``' intrinsics perform the log10 function.
7410 The argument and return value are floating point numbers of the same
7416 This function returns the same values as the libm ``log10`` functions
7417 would, and handles error conditions in the same way.
7419 '``llvm.log2.*``' Intrinsic
7420 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7425 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7426 floating point or vector of floating point type. Not all targets support
7431 declare float @llvm.log2.f32(float %Val)
7432 declare double @llvm.log2.f64(double %Val)
7433 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7434 declare fp128 @llvm.log2.f128(fp128 %Val)
7435 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7440 The '``llvm.log2.*``' intrinsics perform the log2 function.
7445 The argument and return value are floating point numbers of the same
7451 This function returns the same values as the libm ``log2`` functions
7452 would, and handles error conditions in the same way.
7454 '``llvm.fma.*``' Intrinsic
7455 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7460 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7461 floating point or vector of floating point type. Not all targets support
7466 declare float @llvm.fma.f32(float %a, float %b, float %c)
7467 declare double @llvm.fma.f64(double %a, double %b, double %c)
7468 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7469 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7470 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7475 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7481 The argument and return value are floating point numbers of the same
7487 This function returns the same values as the libm ``fma`` functions
7490 '``llvm.fabs.*``' Intrinsic
7491 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7496 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7497 floating point or vector of floating point type. Not all targets support
7502 declare float @llvm.fabs.f32(float %Val)
7503 declare double @llvm.fabs.f64(double %Val)
7504 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7505 declare fp128 @llvm.fabs.f128(fp128 %Val)
7506 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7511 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7517 The argument and return value are floating point numbers of the same
7523 This function returns the same values as the libm ``fabs`` functions
7524 would, and handles error conditions in the same way.
7526 '``llvm.copysign.*``' Intrinsic
7527 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7532 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7533 floating point or vector of floating point type. Not all targets support
7538 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7539 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7540 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7541 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7542 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7547 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7548 first operand and the sign of the second operand.
7553 The arguments and return value are floating point numbers of the same
7559 This function returns the same values as the libm ``copysign``
7560 functions would, and handles error conditions in the same way.
7562 '``llvm.floor.*``' Intrinsic
7563 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7568 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7569 floating point or vector of floating point type. Not all targets support
7574 declare float @llvm.floor.f32(float %Val)
7575 declare double @llvm.floor.f64(double %Val)
7576 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7577 declare fp128 @llvm.floor.f128(fp128 %Val)
7578 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7583 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7588 The argument and return value are floating point numbers of the same
7594 This function returns the same values as the libm ``floor`` functions
7595 would, and handles error conditions in the same way.
7597 '``llvm.ceil.*``' Intrinsic
7598 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7603 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7604 floating point or vector of floating point type. Not all targets support
7609 declare float @llvm.ceil.f32(float %Val)
7610 declare double @llvm.ceil.f64(double %Val)
7611 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7612 declare fp128 @llvm.ceil.f128(fp128 %Val)
7613 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7618 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7623 The argument and return value are floating point numbers of the same
7629 This function returns the same values as the libm ``ceil`` functions
7630 would, and handles error conditions in the same way.
7632 '``llvm.trunc.*``' Intrinsic
7633 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7638 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7639 floating point or vector of floating point type. Not all targets support
7644 declare float @llvm.trunc.f32(float %Val)
7645 declare double @llvm.trunc.f64(double %Val)
7646 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7647 declare fp128 @llvm.trunc.f128(fp128 %Val)
7648 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7653 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7654 nearest integer not larger in magnitude than the operand.
7659 The argument and return value are floating point numbers of the same
7665 This function returns the same values as the libm ``trunc`` functions
7666 would, and handles error conditions in the same way.
7668 '``llvm.rint.*``' Intrinsic
7669 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7674 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7675 floating point or vector of floating point type. Not all targets support
7680 declare float @llvm.rint.f32(float %Val)
7681 declare double @llvm.rint.f64(double %Val)
7682 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7683 declare fp128 @llvm.rint.f128(fp128 %Val)
7684 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7689 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7690 nearest integer. It may raise an inexact floating-point exception if the
7691 operand isn't an integer.
7696 The argument and return value are floating point numbers of the same
7702 This function returns the same values as the libm ``rint`` functions
7703 would, and handles error conditions in the same way.
7705 '``llvm.nearbyint.*``' Intrinsic
7706 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7711 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7712 floating point or vector of floating point type. Not all targets support
7717 declare float @llvm.nearbyint.f32(float %Val)
7718 declare double @llvm.nearbyint.f64(double %Val)
7719 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7720 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7721 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7726 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7732 The argument and return value are floating point numbers of the same
7738 This function returns the same values as the libm ``nearbyint``
7739 functions would, and handles error conditions in the same way.
7741 '``llvm.round.*``' Intrinsic
7742 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7747 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7748 floating point or vector of floating point type. Not all targets support
7753 declare float @llvm.round.f32(float %Val)
7754 declare double @llvm.round.f64(double %Val)
7755 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7756 declare fp128 @llvm.round.f128(fp128 %Val)
7757 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7762 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7768 The argument and return value are floating point numbers of the same
7774 This function returns the same values as the libm ``round``
7775 functions would, and handles error conditions in the same way.
7777 Bit Manipulation Intrinsics
7778 ---------------------------
7780 LLVM provides intrinsics for a few important bit manipulation
7781 operations. These allow efficient code generation for some algorithms.
7783 '``llvm.bswap.*``' Intrinsics
7784 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7789 This is an overloaded intrinsic function. You can use bswap on any
7790 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7794 declare i16 @llvm.bswap.i16(i16 <id>)
7795 declare i32 @llvm.bswap.i32(i32 <id>)
7796 declare i64 @llvm.bswap.i64(i64 <id>)
7801 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7802 values with an even number of bytes (positive multiple of 16 bits).
7803 These are useful for performing operations on data that is not in the
7804 target's native byte order.
7809 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7810 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7811 intrinsic returns an i32 value that has the four bytes of the input i32
7812 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7813 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7814 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7815 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7818 '``llvm.ctpop.*``' Intrinsic
7819 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7824 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7825 bit width, or on any vector with integer elements. Not all targets
7826 support all bit widths or vector types, however.
7830 declare i8 @llvm.ctpop.i8(i8 <src>)
7831 declare i16 @llvm.ctpop.i16(i16 <src>)
7832 declare i32 @llvm.ctpop.i32(i32 <src>)
7833 declare i64 @llvm.ctpop.i64(i64 <src>)
7834 declare i256 @llvm.ctpop.i256(i256 <src>)
7835 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7840 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7846 The only argument is the value to be counted. The argument may be of any
7847 integer type, or a vector with integer elements. The return type must
7848 match the argument type.
7853 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7854 each element of a vector.
7856 '``llvm.ctlz.*``' Intrinsic
7857 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7862 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7863 integer bit width, or any vector whose elements are integers. Not all
7864 targets support all bit widths or vector types, however.
7868 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7869 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7870 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7871 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7872 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7873 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7878 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7879 leading zeros in a variable.
7884 The first argument is the value to be counted. This argument may be of
7885 any integer type, or a vectory with integer element type. The return
7886 type must match the first argument type.
7888 The second argument must be a constant and is a flag to indicate whether
7889 the intrinsic should ensure that a zero as the first argument produces a
7890 defined result. Historically some architectures did not provide a
7891 defined result for zero values as efficiently, and many algorithms are
7892 now predicated on avoiding zero-value inputs.
7897 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7898 zeros in a variable, or within each element of the vector. If
7899 ``src == 0`` then the result is the size in bits of the type of ``src``
7900 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7901 ``llvm.ctlz(i32 2) = 30``.
7903 '``llvm.cttz.*``' Intrinsic
7904 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7909 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7910 integer bit width, or any vector of integer elements. Not all targets
7911 support all bit widths or vector types, however.
7915 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7916 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7917 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7918 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7919 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7920 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7925 The '``llvm.cttz``' family of intrinsic functions counts the number of
7931 The first argument is the value to be counted. This argument may be of
7932 any integer type, or a vectory with integer element type. The return
7933 type must match the first argument type.
7935 The second argument must be a constant and is a flag to indicate whether
7936 the intrinsic should ensure that a zero as the first argument produces a
7937 defined result. Historically some architectures did not provide a
7938 defined result for zero values as efficiently, and many algorithms are
7939 now predicated on avoiding zero-value inputs.
7944 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7945 zeros in a variable, or within each element of a vector. If ``src == 0``
7946 then the result is the size in bits of the type of ``src`` if
7947 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7948 ``llvm.cttz(2) = 1``.
7950 Arithmetic with Overflow Intrinsics
7951 -----------------------------------
7953 LLVM provides intrinsics for some arithmetic with overflow operations.
7955 '``llvm.sadd.with.overflow.*``' Intrinsics
7956 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7961 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7962 on any integer bit width.
7966 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7967 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7968 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7973 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7974 a signed addition of the two arguments, and indicate whether an overflow
7975 occurred during the signed summation.
7980 The arguments (%a and %b) and the first element of the result structure
7981 may be of integer types of any bit width, but they must have the same
7982 bit width. The second element of the result structure must be of type
7983 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7989 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7990 a signed addition of the two variables. They return a structure --- the
7991 first element of which is the signed summation, and the second element
7992 of which is a bit specifying if the signed summation resulted in an
7998 .. code-block:: llvm
8000 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8001 %sum = extractvalue {i32, i1} %res, 0
8002 %obit = extractvalue {i32, i1} %res, 1
8003 br i1 %obit, label %overflow, label %normal
8005 '``llvm.uadd.with.overflow.*``' Intrinsics
8006 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8011 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8012 on any integer bit width.
8016 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8017 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8018 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8023 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8024 an unsigned addition of the two arguments, and indicate whether a carry
8025 occurred during the unsigned summation.
8030 The arguments (%a and %b) and the first element of the result structure
8031 may be of integer types of any bit width, but they must have the same
8032 bit width. The second element of the result structure must be of type
8033 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8039 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8040 an unsigned addition of the two arguments. They return a structure --- the
8041 first element of which is the sum, and the second element of which is a
8042 bit specifying if the unsigned summation resulted in a carry.
8047 .. code-block:: llvm
8049 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8050 %sum = extractvalue {i32, i1} %res, 0
8051 %obit = extractvalue {i32, i1} %res, 1
8052 br i1 %obit, label %carry, label %normal
8054 '``llvm.ssub.with.overflow.*``' Intrinsics
8055 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8060 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8061 on any integer bit width.
8065 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8066 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8067 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8072 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8073 a signed subtraction of the two arguments, and indicate whether an
8074 overflow occurred during the signed subtraction.
8079 The arguments (%a and %b) and the first element of the result structure
8080 may be of integer types of any bit width, but they must have the same
8081 bit width. The second element of the result structure must be of type
8082 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8088 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8089 a signed subtraction of the two arguments. They return a structure --- the
8090 first element of which is the subtraction, and the second element of
8091 which is a bit specifying if the signed subtraction resulted in an
8097 .. code-block:: llvm
8099 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8100 %sum = extractvalue {i32, i1} %res, 0
8101 %obit = extractvalue {i32, i1} %res, 1
8102 br i1 %obit, label %overflow, label %normal
8104 '``llvm.usub.with.overflow.*``' Intrinsics
8105 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8110 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8111 on any integer bit width.
8115 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8116 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8117 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8122 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8123 an unsigned subtraction of the two arguments, and indicate whether an
8124 overflow occurred during the unsigned subtraction.
8129 The arguments (%a and %b) and the first element of the result structure
8130 may be of integer types of any bit width, but they must have the same
8131 bit width. The second element of the result structure must be of type
8132 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8138 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8139 an unsigned subtraction of the two arguments. They return a structure ---
8140 the first element of which is the subtraction, and the second element of
8141 which is a bit specifying if the unsigned subtraction resulted in an
8147 .. code-block:: llvm
8149 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8150 %sum = extractvalue {i32, i1} %res, 0
8151 %obit = extractvalue {i32, i1} %res, 1
8152 br i1 %obit, label %overflow, label %normal
8154 '``llvm.smul.with.overflow.*``' Intrinsics
8155 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8160 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8161 on any integer bit width.
8165 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8166 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8167 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8172 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8173 a signed multiplication of the two arguments, and indicate whether an
8174 overflow occurred during the signed multiplication.
8179 The arguments (%a and %b) and the first element of the result structure
8180 may be of integer types of any bit width, but they must have the same
8181 bit width. The second element of the result structure must be of type
8182 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8188 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8189 a signed multiplication of the two arguments. They return a structure ---
8190 the first element of which is the multiplication, and the second element
8191 of which is a bit specifying if the signed multiplication resulted in an
8197 .. code-block:: llvm
8199 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8200 %sum = extractvalue {i32, i1} %res, 0
8201 %obit = extractvalue {i32, i1} %res, 1
8202 br i1 %obit, label %overflow, label %normal
8204 '``llvm.umul.with.overflow.*``' Intrinsics
8205 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8210 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8211 on any integer bit width.
8215 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8216 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8217 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8222 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8223 a unsigned multiplication of the two arguments, and indicate whether an
8224 overflow occurred during the unsigned multiplication.
8229 The arguments (%a and %b) and the first element of the result structure
8230 may be of integer types of any bit width, but they must have the same
8231 bit width. The second element of the result structure must be of type
8232 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8238 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8239 an unsigned multiplication of the two arguments. They return a structure ---
8240 the first element of which is the multiplication, and the second
8241 element of which is a bit specifying if the unsigned multiplication
8242 resulted in an overflow.
8247 .. code-block:: llvm
8249 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8250 %sum = extractvalue {i32, i1} %res, 0
8251 %obit = extractvalue {i32, i1} %res, 1
8252 br i1 %obit, label %overflow, label %normal
8254 Specialised Arithmetic Intrinsics
8255 ---------------------------------
8257 '``llvm.fmuladd.*``' Intrinsic
8258 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8265 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8266 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8271 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8272 expressions that can be fused if the code generator determines that (a) the
8273 target instruction set has support for a fused operation, and (b) that the
8274 fused operation is more efficient than the equivalent, separate pair of mul
8275 and add instructions.
8280 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8281 multiplicands, a and b, and an addend c.
8290 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8292 is equivalent to the expression a \* b + c, except that rounding will
8293 not be performed between the multiplication and addition steps if the
8294 code generator fuses the operations. Fusion is not guaranteed, even if
8295 the target platform supports it. If a fused multiply-add is required the
8296 corresponding llvm.fma.\* intrinsic function should be used instead.
8301 .. code-block:: llvm
8303 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8305 Half Precision Floating Point Intrinsics
8306 ----------------------------------------
8308 For most target platforms, half precision floating point is a
8309 storage-only format. This means that it is a dense encoding (in memory)
8310 but does not support computation in the format.
8312 This means that code must first load the half-precision floating point
8313 value as an i16, then convert it to float with
8314 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8315 then be performed on the float value (including extending to double
8316 etc). To store the value back to memory, it is first converted to float
8317 if needed, then converted to i16 with
8318 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8321 .. _int_convert_to_fp16:
8323 '``llvm.convert.to.fp16``' Intrinsic
8324 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8331 declare i16 @llvm.convert.to.fp16(f32 %a)
8336 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8337 from single precision floating point format to half precision floating
8343 The intrinsic function contains single argument - the value to be
8349 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8350 from single precision floating point format to half precision floating
8351 point format. The return value is an ``i16`` which contains the
8357 .. code-block:: llvm
8359 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8360 store i16 %res, i16* @x, align 2
8362 .. _int_convert_from_fp16:
8364 '``llvm.convert.from.fp16``' Intrinsic
8365 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8372 declare f32 @llvm.convert.from.fp16(i16 %a)
8377 The '``llvm.convert.from.fp16``' intrinsic function performs a
8378 conversion from half precision floating point format to single precision
8379 floating point format.
8384 The intrinsic function contains single argument - the value to be
8390 The '``llvm.convert.from.fp16``' intrinsic function performs a
8391 conversion from half single precision floating point format to single
8392 precision floating point format. The input half-float value is
8393 represented by an ``i16`` value.
8398 .. code-block:: llvm
8400 %a = load i16* @x, align 2
8401 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8406 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8407 prefix), are described in the `LLVM Source Level
8408 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8411 Exception Handling Intrinsics
8412 -----------------------------
8414 The LLVM exception handling intrinsics (which all start with
8415 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8416 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8420 Trampoline Intrinsics
8421 ---------------------
8423 These intrinsics make it possible to excise one parameter, marked with
8424 the :ref:`nest <nest>` attribute, from a function. The result is a
8425 callable function pointer lacking the nest parameter - the caller does
8426 not need to provide a value for it. Instead, the value to use is stored
8427 in advance in a "trampoline", a block of memory usually allocated on the
8428 stack, which also contains code to splice the nest value into the
8429 argument list. This is used to implement the GCC nested function address
8432 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8433 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8434 It can be created as follows:
8436 .. code-block:: llvm
8438 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8439 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8440 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8441 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8442 %fp = bitcast i8* %p to i32 (i32, i32)*
8444 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8445 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8449 '``llvm.init.trampoline``' Intrinsic
8450 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8457 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8462 This fills the memory pointed to by ``tramp`` with executable code,
8463 turning it into a trampoline.
8468 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8469 pointers. The ``tramp`` argument must point to a sufficiently large and
8470 sufficiently aligned block of memory; this memory is written to by the
8471 intrinsic. Note that the size and the alignment are target-specific -
8472 LLVM currently provides no portable way of determining them, so a
8473 front-end that generates this intrinsic needs to have some
8474 target-specific knowledge. The ``func`` argument must hold a function
8475 bitcast to an ``i8*``.
8480 The block of memory pointed to by ``tramp`` is filled with target
8481 dependent code, turning it into a function. Then ``tramp`` needs to be
8482 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8483 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8484 function's signature is the same as that of ``func`` with any arguments
8485 marked with the ``nest`` attribute removed. At most one such ``nest``
8486 argument is allowed, and it must be of pointer type. Calling the new
8487 function is equivalent to calling ``func`` with the same argument list,
8488 but with ``nval`` used for the missing ``nest`` argument. If, after
8489 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8490 modified, then the effect of any later call to the returned function
8491 pointer is undefined.
8495 '``llvm.adjust.trampoline``' Intrinsic
8496 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8503 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8508 This performs any required machine-specific adjustment to the address of
8509 a trampoline (passed as ``tramp``).
8514 ``tramp`` must point to a block of memory which already has trampoline
8515 code filled in by a previous call to
8516 :ref:`llvm.init.trampoline <int_it>`.
8521 On some architectures the address of the code to be executed needs to be
8522 different to the address where the trampoline is actually stored. This
8523 intrinsic returns the executable address corresponding to ``tramp``
8524 after performing the required machine specific adjustments. The pointer
8525 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8530 This class of intrinsics exists to information about the lifetime of
8531 memory objects and ranges where variables are immutable.
8535 '``llvm.lifetime.start``' Intrinsic
8536 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8543 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8548 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8554 The first argument is a constant integer representing the size of the
8555 object, or -1 if it is variable sized. The second argument is a pointer
8561 This intrinsic indicates that before this point in the code, the value
8562 of the memory pointed to by ``ptr`` is dead. This means that it is known
8563 to never be used and has an undefined value. A load from the pointer
8564 that precedes this intrinsic can be replaced with ``'undef'``.
8568 '``llvm.lifetime.end``' Intrinsic
8569 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8576 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8581 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8587 The first argument is a constant integer representing the size of the
8588 object, or -1 if it is variable sized. The second argument is a pointer
8594 This intrinsic indicates that after this point in the code, the value of
8595 the memory pointed to by ``ptr`` is dead. This means that it is known to
8596 never be used and has an undefined value. Any stores into the memory
8597 object following this intrinsic may be removed as dead.
8599 '``llvm.invariant.start``' Intrinsic
8600 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8607 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8612 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8613 a memory object will not change.
8618 The first argument is a constant integer representing the size of the
8619 object, or -1 if it is variable sized. The second argument is a pointer
8625 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8626 the return value, the referenced memory location is constant and
8629 '``llvm.invariant.end``' Intrinsic
8630 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8637 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8642 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8643 memory object are mutable.
8648 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8649 The second argument is a constant integer representing the size of the
8650 object, or -1 if it is variable sized and the third argument is a
8651 pointer to the object.
8656 This intrinsic indicates that the memory is mutable again.
8661 This class of intrinsics is designed to be generic and has no specific
8664 '``llvm.var.annotation``' Intrinsic
8665 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8672 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8677 The '``llvm.var.annotation``' intrinsic.
8682 The first argument is a pointer to a value, the second is a pointer to a
8683 global string, the third is a pointer to a global string which is the
8684 source file name, and the last argument is the line number.
8689 This intrinsic allows annotation of local variables with arbitrary
8690 strings. This can be useful for special purpose optimizations that want
8691 to look for these annotations. These have no other defined use; they are
8692 ignored by code generation and optimization.
8694 '``llvm.ptr.annotation.*``' Intrinsic
8695 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8700 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8701 pointer to an integer of any width. *NOTE* you must specify an address space for
8702 the pointer. The identifier for the default address space is the integer
8707 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8708 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8709 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8710 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8711 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8716 The '``llvm.ptr.annotation``' intrinsic.
8721 The first argument is a pointer to an integer value of arbitrary bitwidth
8722 (result of some expression), the second is a pointer to a global string, the
8723 third is a pointer to a global string which is the source file name, and the
8724 last argument is the line number. It returns the value of the first argument.
8729 This intrinsic allows annotation of a pointer to an integer with arbitrary
8730 strings. This can be useful for special purpose optimizations that want to look
8731 for these annotations. These have no other defined use; they are ignored by code
8732 generation and optimization.
8734 '``llvm.annotation.*``' Intrinsic
8735 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8740 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8741 any integer bit width.
8745 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8746 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8747 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8748 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8749 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8754 The '``llvm.annotation``' intrinsic.
8759 The first argument is an integer value (result of some expression), the
8760 second is a pointer to a global string, the third is a pointer to a
8761 global string which is the source file name, and the last argument is
8762 the line number. It returns the value of the first argument.
8767 This intrinsic allows annotations to be put on arbitrary expressions
8768 with arbitrary strings. This can be useful for special purpose
8769 optimizations that want to look for these annotations. These have no
8770 other defined use; they are ignored by code generation and optimization.
8772 '``llvm.trap``' Intrinsic
8773 ^^^^^^^^^^^^^^^^^^^^^^^^^
8780 declare void @llvm.trap() noreturn nounwind
8785 The '``llvm.trap``' intrinsic.
8795 This intrinsic is lowered to the target dependent trap instruction. If
8796 the target does not have a trap instruction, this intrinsic will be
8797 lowered to a call of the ``abort()`` function.
8799 '``llvm.debugtrap``' Intrinsic
8800 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8807 declare void @llvm.debugtrap() nounwind
8812 The '``llvm.debugtrap``' intrinsic.
8822 This intrinsic is lowered to code which is intended to cause an
8823 execution trap with the intention of requesting the attention of a
8826 '``llvm.stackprotector``' Intrinsic
8827 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8834 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8839 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8840 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8841 is placed on the stack before local variables.
8846 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8847 The first argument is the value loaded from the stack guard
8848 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8849 enough space to hold the value of the guard.
8854 This intrinsic causes the prologue/epilogue inserter to force the position of
8855 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8856 to ensure that if a local variable on the stack is overwritten, it will destroy
8857 the value of the guard. When the function exits, the guard on the stack is
8858 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8859 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8860 calling the ``__stack_chk_fail()`` function.
8862 '``llvm.stackprotectorcheck``' Intrinsic
8863 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8870 declare void @llvm.stackprotectorcheck(i8** <guard>)
8875 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8876 created stack protector and if they are not equal calls the
8877 ``__stack_chk_fail()`` function.
8882 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8883 the variable ``@__stack_chk_guard``.
8888 This intrinsic is provided to perform the stack protector check by comparing
8889 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8890 values do not match call the ``__stack_chk_fail()`` function.
8892 The reason to provide this as an IR level intrinsic instead of implementing it
8893 via other IR operations is that in order to perform this operation at the IR
8894 level without an intrinsic, one would need to create additional basic blocks to
8895 handle the success/failure cases. This makes it difficult to stop the stack
8896 protector check from disrupting sibling tail calls in Codegen. With this
8897 intrinsic, we are able to generate the stack protector basic blocks late in
8898 codegen after the tail call decision has occurred.
8900 '``llvm.objectsize``' Intrinsic
8901 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8908 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8909 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8914 The ``llvm.objectsize`` intrinsic is designed to provide information to
8915 the optimizers to determine at compile time whether a) an operation
8916 (like memcpy) will overflow a buffer that corresponds to an object, or
8917 b) that a runtime check for overflow isn't necessary. An object in this
8918 context means an allocation of a specific class, structure, array, or
8924 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8925 argument is a pointer to or into the ``object``. The second argument is
8926 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8927 or -1 (if false) when the object size is unknown. The second argument
8928 only accepts constants.
8933 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8934 the size of the object concerned. If the size cannot be determined at
8935 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8936 on the ``min`` argument).
8938 '``llvm.expect``' Intrinsic
8939 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8946 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8947 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8952 The ``llvm.expect`` intrinsic provides information about expected (the
8953 most probable) value of ``val``, which can be used by optimizers.
8958 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8959 a value. The second argument is an expected value, this needs to be a
8960 constant value, variables are not allowed.
8965 This intrinsic is lowered to the ``val``.
8967 '``llvm.donothing``' Intrinsic
8968 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8975 declare void @llvm.donothing() nounwind readnone
8980 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8981 only intrinsic that can be called with an invoke instruction.
8991 This intrinsic does nothing, and it's removed by optimizers and ignored
8994 Stack Map Intrinsics
8995 --------------------
8997 LLVM provides experimental intrinsics to support runtime patching
8998 mechanisms commonly desired in dynamic language JITs. These intrinsics
8999 are described in :doc:`StackMaps`.