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 that 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. The ``"\01"`` prefix
83 can be used on global variables to suppress mangling.
84 #. Unnamed values are represented as an unsigned numeric value with
85 their prefix. For example, ``%12``, ``@2``, ``%44``.
86 #. Constants, which are described in the section Constants_ below.
88 LLVM requires that values start with a prefix for two reasons: Compilers
89 don't need to worry about name clashes with reserved words, and the set
90 of reserved words may be expanded in the future without penalty.
91 Additionally, unnamed identifiers allow a compiler to quickly come up
92 with a temporary variable without having to avoid symbol table
95 Reserved words in LLVM are very similar to reserved words in other
96 languages. There are keywords for different opcodes ('``add``',
97 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
98 '``i32``', etc...), and others. These reserved words cannot conflict
99 with variable names, because none of them start with a prefix character
100 (``'%'`` or ``'@'``).
102 Here is an example of LLVM code to multiply the integer variable
109 %result = mul i32 %X, 8
111 After strength reduction:
115 %result = shl i32 %X, 3
121 %0 = add i32 %X, %X ; yields i32:%0
122 %1 = add i32 %0, %0 ; yields i32:%1
123 %result = add i32 %1, %1
125 This last way of multiplying ``%X`` by 8 illustrates several important
126 lexical features of LLVM:
128 #. Comments are delimited with a '``;``' and go until the end of line.
129 #. Unnamed temporaries are created when the result of a computation is
130 not assigned to a named value.
131 #. Unnamed temporaries are numbered sequentially (using a per-function
132 incrementing counter, starting with 0). Note that basic blocks and unnamed
133 function parameters are included in this numbering. For example, if the
134 entry basic block is not given a label name and all function parameters are
135 named, then it will get number 0.
137 It also shows a convention that we follow in this document. When
138 demonstrating instructions, we will follow an instruction with a comment
139 that defines the type and name of value produced.
147 LLVM programs are composed of ``Module``'s, each of which is a
148 translation unit of the input programs. Each module consists of
149 functions, global variables, and symbol table entries. Modules may be
150 combined together with the LLVM linker, which merges function (and
151 global variable) definitions, resolves forward declarations, and merges
152 symbol table entries. Here is an example of the "hello world" module:
156 ; Declare the string constant as a global constant.
157 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
159 ; External declaration of the puts function
160 declare i32 @puts(i8* nocapture) nounwind
162 ; Definition of main function
163 define i32 @main() { ; i32()*
164 ; Convert [13 x i8]* to i8 *...
165 %cast210 = getelementptr [13 x i8], [13 x i8]* @.str, i64 0, i64 0
167 ; Call puts function to write out the string to stdout.
168 call i32 @puts(i8* %cast210)
173 !0 = !{i32 42, null, !"string"}
176 This example is made up of a :ref:`global variable <globalvars>` named
177 "``.str``", an external declaration of the "``puts``" function, a
178 :ref:`function definition <functionstructure>` for "``main``" and
179 :ref:`named metadata <namedmetadatastructure>` "``foo``".
181 In general, a module is made up of a list of global values (where both
182 functions and global variables are global values). Global values are
183 represented by a pointer to a memory location (in this case, a pointer
184 to an array of char, and a pointer to a function), and have one of the
185 following :ref:`linkage types <linkage>`.
192 All Global Variables and Functions have one of the following types of
196 Global values with "``private``" linkage are only directly
197 accessible by objects in the current module. In particular, linking
198 code into a module with an private global value may cause the
199 private to be renamed as necessary to avoid collisions. Because the
200 symbol is private to the module, all references can be updated. This
201 doesn't show up in any symbol table in the object file.
203 Similar to private, but the value shows as a local symbol
204 (``STB_LOCAL`` in the case of ELF) in the object file. This
205 corresponds to the notion of the '``static``' keyword in C.
206 ``available_externally``
207 Globals with "``available_externally``" linkage are never emitted
208 into the object file corresponding to the LLVM module. They exist to
209 allow inlining and other optimizations to take place given knowledge
210 of the definition of the global, which is known to be somewhere
211 outside the module. Globals with ``available_externally`` linkage
212 are allowed to be discarded at will, and are otherwise the same as
213 ``linkonce_odr``. This linkage type is only allowed on definitions,
216 Globals with "``linkonce``" linkage are merged with other globals of
217 the same name when linkage occurs. This can be used to implement
218 some forms of inline functions, templates, or other code which must
219 be generated in each translation unit that uses it, but where the
220 body may be overridden with a more definitive definition later.
221 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
222 that ``linkonce`` linkage does not actually allow the optimizer to
223 inline the body of this function into callers because it doesn't
224 know if this definition of the function is the definitive definition
225 within the program or whether it will be overridden by a stronger
226 definition. To enable inlining and other optimizations, use
227 "``linkonce_odr``" linkage.
229 "``weak``" linkage has the same merging semantics as ``linkonce``
230 linkage, except that unreferenced globals with ``weak`` linkage may
231 not be discarded. This is used for globals that are declared "weak"
234 "``common``" linkage is most similar to "``weak``" linkage, but they
235 are used for tentative definitions in C, such as "``int X;``" at
236 global scope. Symbols with "``common``" linkage are merged in the
237 same way as ``weak symbols``, and they may not be deleted if
238 unreferenced. ``common`` symbols may not have an explicit section,
239 must have a zero initializer, and may not be marked
240 ':ref:`constant <globalvars>`'. Functions and aliases may not have
243 .. _linkage_appending:
246 "``appending``" linkage may only be applied to global variables of
247 pointer to array type. When two global variables with appending
248 linkage are linked together, the two global arrays are appended
249 together. This is the LLVM, typesafe, equivalent of having the
250 system linker append together "sections" with identical names when
253 The semantics of this linkage follow the ELF object file model: the
254 symbol is weak until linked, if not linked, the symbol becomes null
255 instead of being an undefined reference.
256 ``linkonce_odr``, ``weak_odr``
257 Some languages allow differing globals to be merged, such as two
258 functions with different semantics. Other languages, such as
259 ``C++``, ensure that only equivalent globals are ever merged (the
260 "one definition rule" --- "ODR"). Such languages can use the
261 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
262 global will only be merged with equivalent globals. These linkage
263 types are otherwise the same as their non-``odr`` versions.
265 If none of the above identifiers are used, the global is externally
266 visible, meaning that it participates in linkage and can be used to
267 resolve external symbol references.
269 It is illegal for a function *declaration* to have any linkage type
270 other than ``external`` or ``extern_weak``.
277 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
278 :ref:`invokes <i_invoke>` can all have an optional calling convention
279 specified for the call. The calling convention of any pair of dynamic
280 caller/callee must match, or the behavior of the program is undefined.
281 The following calling conventions are supported by LLVM, and more may be
284 "``ccc``" - The C calling convention
285 This calling convention (the default if no other calling convention
286 is specified) matches the target C calling conventions. This calling
287 convention supports varargs function calls and tolerates some
288 mismatch in the declared prototype and implemented declaration of
289 the function (as does normal C).
290 "``fastcc``" - The fast calling convention
291 This calling convention attempts to make calls as fast as possible
292 (e.g. by passing things in registers). This calling convention
293 allows the target to use whatever tricks it wants to produce fast
294 code for the target, without having to conform to an externally
295 specified ABI (Application Binary Interface). `Tail calls can only
296 be optimized when this, the GHC or the HiPE convention is
297 used. <CodeGenerator.html#id80>`_ This calling convention does not
298 support varargs and requires the prototype of all callees to exactly
299 match the prototype of the function definition.
300 "``coldcc``" - The cold calling convention
301 This calling convention attempts to make code in the caller as
302 efficient as possible under the assumption that the call is not
303 commonly executed. As such, these calls often preserve all registers
304 so that the call does not break any live ranges in the caller side.
305 This calling convention does not support varargs and requires the
306 prototype of all callees to exactly match the prototype of the
307 function definition. Furthermore the inliner doesn't consider such function
309 "``cc 10``" - GHC convention
310 This calling convention has been implemented specifically for use by
311 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
312 It passes everything in registers, going to extremes to achieve this
313 by disabling callee save registers. This calling convention should
314 not be used lightly but only for specific situations such as an
315 alternative to the *register pinning* performance technique often
316 used when implementing functional programming languages. At the
317 moment only X86 supports this convention and it has the following
320 - On *X86-32* only supports up to 4 bit type parameters. No
321 floating point types are supported.
322 - On *X86-64* only supports up to 10 bit type parameters and 6
323 floating point parameters.
325 This calling convention supports `tail call
326 optimization <CodeGenerator.html#id80>`_ but requires both the
327 caller and callee are using it.
328 "``cc 11``" - The HiPE calling convention
329 This calling convention has been implemented specifically for use by
330 the `High-Performance Erlang
331 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
332 native code compiler of the `Ericsson's Open Source Erlang/OTP
333 system <http://www.erlang.org/download.shtml>`_. It uses more
334 registers for argument passing than the ordinary C calling
335 convention and defines no callee-saved registers. The calling
336 convention properly supports `tail call
337 optimization <CodeGenerator.html#id80>`_ but requires that both the
338 caller and the callee use it. It uses a *register pinning*
339 mechanism, similar to GHC's convention, for keeping frequently
340 accessed runtime components pinned to specific hardware registers.
341 At the moment only X86 supports this convention (both 32 and 64
343 "``webkit_jscc``" - WebKit's JavaScript calling convention
344 This calling convention has been implemented for `WebKit FTL JIT
345 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
346 stack right to left (as cdecl does), and returns a value in the
347 platform's customary return register.
348 "``anyregcc``" - Dynamic calling convention for code patching
349 This is a special convention that supports patching an arbitrary code
350 sequence in place of a call site. This convention forces the call
351 arguments into registers but allows them to be dynamically
352 allocated. This can currently only be used with calls to
353 llvm.experimental.patchpoint because only this intrinsic records
354 the location of its arguments in a side table. See :doc:`StackMaps`.
355 "``preserve_mostcc``" - The `PreserveMost` calling convention
356 This calling convention attempts to make the code in the caller as
357 unintrusive as possible. This convention behaves identically to the `C`
358 calling convention on how arguments and return values are passed, but it
359 uses a different set of caller/callee-saved registers. This alleviates the
360 burden of saving and recovering a large register set before and after the
361 call in the caller. If the arguments are passed in callee-saved registers,
362 then they will be preserved by the callee across the call. This doesn't
363 apply for values returned in callee-saved registers.
365 - On X86-64 the callee preserves all general purpose registers, except for
366 R11. R11 can be used as a scratch register. Floating-point registers
367 (XMMs/YMMs) are not preserved and need to be saved by the caller.
369 The idea behind this convention is to support calls to runtime functions
370 that have a hot path and a cold path. The hot path is usually a small piece
371 of code that doesn't use many registers. The cold path might need to call out to
372 another function and therefore only needs to preserve the caller-saved
373 registers, which haven't already been saved by the caller. The
374 `PreserveMost` calling convention is very similar to the `cold` calling
375 convention in terms of caller/callee-saved registers, but they are used for
376 different types of function calls. `coldcc` is for function calls that are
377 rarely executed, whereas `preserve_mostcc` function calls are intended to be
378 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
379 doesn't prevent the inliner from inlining the function call.
381 This calling convention will be used by a future version of the ObjectiveC
382 runtime and should therefore still be considered experimental at this time.
383 Although this convention was created to optimize certain runtime calls to
384 the ObjectiveC runtime, it is not limited to this runtime and might be used
385 by other runtimes in the future too. The current implementation only
386 supports X86-64, but the intention is to support more architectures in the
388 "``preserve_allcc``" - The `PreserveAll` calling convention
389 This calling convention attempts to make the code in the caller even less
390 intrusive than the `PreserveMost` calling convention. This calling
391 convention also behaves identical to the `C` calling convention on how
392 arguments and return values are passed, but it uses a different set of
393 caller/callee-saved registers. This removes the burden of saving and
394 recovering a large register set before and after the call in the caller. If
395 the arguments are passed in callee-saved registers, then they will be
396 preserved by the callee across the call. This doesn't apply for values
397 returned in callee-saved registers.
399 - On X86-64 the callee preserves all general purpose registers, except for
400 R11. R11 can be used as a scratch register. Furthermore it also preserves
401 all floating-point registers (XMMs/YMMs).
403 The idea behind this convention is to support calls to runtime functions
404 that don't need to call out to any other functions.
406 This calling convention, like the `PreserveMost` calling convention, will be
407 used by a future version of the ObjectiveC runtime and should be considered
408 experimental at this time.
409 "``cc <n>``" - Numbered convention
410 Any calling convention may be specified by number, allowing
411 target-specific calling conventions to be used. Target specific
412 calling conventions start at 64.
414 More calling conventions can be added/defined on an as-needed basis, to
415 support Pascal conventions or any other well-known target-independent
418 .. _visibilitystyles:
423 All Global Variables and Functions have one of the following visibility
426 "``default``" - Default style
427 On targets that use the ELF object file format, default visibility
428 means that the declaration is visible to other modules and, in
429 shared libraries, means that the declared entity may be overridden.
430 On Darwin, default visibility means that the declaration is visible
431 to other modules. Default visibility corresponds to "external
432 linkage" in the language.
433 "``hidden``" - Hidden style
434 Two declarations of an object with hidden visibility refer to the
435 same object if they are in the same shared object. Usually, hidden
436 visibility indicates that the symbol will not be placed into the
437 dynamic symbol table, so no other module (executable or shared
438 library) can reference it directly.
439 "``protected``" - Protected style
440 On ELF, protected visibility indicates that the symbol will be
441 placed in the dynamic symbol table, but that references within the
442 defining module will bind to the local symbol. That is, the symbol
443 cannot be overridden by another module.
445 A symbol with ``internal`` or ``private`` linkage must have ``default``
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.
471 Thread Local Storage Models
472 ---------------------------
474 A variable may be defined as ``thread_local``, which means that it will
475 not be shared by threads (each thread will have a separated copy of the
476 variable). Not all targets support thread-local variables. Optionally, a
477 TLS model may be specified:
480 For variables that are only used within the current shared library.
482 For variables in modules that will not be loaded dynamically.
484 For variables defined in the executable and only used within it.
486 If no explicit model is given, the "general dynamic" model is used.
488 The models correspond to the ELF TLS models; see `ELF Handling For
489 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
490 more information on under which circumstances the different models may
491 be used. The target may choose a different TLS model if the specified
492 model is not supported, or if a better choice of model can be made.
494 A model can also be specified in a alias, but then it only governs how
495 the alias is accessed. It will not have any effect in the aliasee.
502 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
503 types <t_struct>`. Literal types are uniqued structurally, but identified types
504 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
505 to forward declare a type that is not yet available.
507 An example of a identified structure specification is:
511 %mytype = type { %mytype*, i32 }
513 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
514 literal types are uniqued in recent versions of LLVM.
521 Global variables define regions of memory allocated at compilation time
524 Global variable definitions must be initialized.
526 Global variables in other translation units can also be declared, in which
527 case they don't have an initializer.
529 Either global variable definitions or declarations may have an explicit section
530 to be placed in and may have an optional explicit alignment specified.
532 A variable may be defined as a global ``constant``, which indicates that
533 the contents of the variable will **never** be modified (enabling better
534 optimization, allowing the global data to be placed in the read-only
535 section of an executable, etc). Note that variables that need runtime
536 initialization cannot be marked ``constant`` as there is a store to the
539 LLVM explicitly allows *declarations* of global variables to be marked
540 constant, even if the final definition of the global is not. This
541 capability can be used to enable slightly better optimization of the
542 program, but requires the language definition to guarantee that
543 optimizations based on the 'constantness' are valid for the translation
544 units that do not include the definition.
546 As SSA values, global variables define pointer values that are in scope
547 (i.e. they dominate) all basic blocks in the program. Global variables
548 always define a pointer to their "content" type because they describe a
549 region of memory, and all memory objects in LLVM are accessed through
552 Global variables can be marked with ``unnamed_addr`` which indicates
553 that the address is not significant, only the content. Constants marked
554 like this can be merged with other constants if they have the same
555 initializer. Note that a constant with significant address *can* be
556 merged with a ``unnamed_addr`` constant, the result being a constant
557 whose address is significant.
559 A global variable may be declared to reside in a target-specific
560 numbered address space. For targets that support them, address spaces
561 may affect how optimizations are performed and/or what target
562 instructions are used to access the variable. The default address space
563 is zero. The address space qualifier must precede any other attributes.
565 LLVM allows an explicit section to be specified for globals. If the
566 target supports it, it will emit globals to the section specified.
567 Additionally, the global can placed in a comdat if the target has the necessary
570 By default, global initializers are optimized by assuming that global
571 variables defined within the module are not modified from their
572 initial values before the start of the global initializer. This is
573 true even for variables potentially accessible from outside the
574 module, including those with external linkage or appearing in
575 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
576 by marking the variable with ``externally_initialized``.
578 An explicit alignment may be specified for a global, which must be a
579 power of 2. If not present, or if the alignment is set to zero, the
580 alignment of the global is set by the target to whatever it feels
581 convenient. If an explicit alignment is specified, the global is forced
582 to have exactly that alignment. Targets and optimizers are not allowed
583 to over-align the global if the global has an assigned section. In this
584 case, the extra alignment could be observable: for example, code could
585 assume that the globals are densely packed in their section and try to
586 iterate over them as an array, alignment padding would break this
587 iteration. The maximum alignment is ``1 << 29``.
589 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
591 Variables and aliases can have a
592 :ref:`Thread Local Storage Model <tls_model>`.
596 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
597 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
598 <global | constant> <Type> [<InitializerConstant>]
599 [, section "name"] [, comdat [($name)]]
600 [, align <Alignment>]
602 For example, the following defines a global in a numbered address space
603 with an initializer, section, and alignment:
607 @G = addrspace(5) constant float 1.0, section "foo", align 4
609 The following example just declares a global variable
613 @G = external global i32
615 The following example defines a thread-local global with the
616 ``initialexec`` TLS model:
620 @G = thread_local(initialexec) global i32 0, align 4
622 .. _functionstructure:
627 LLVM function definitions consist of the "``define``" keyword, an
628 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
629 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
630 an optional :ref:`calling convention <callingconv>`,
631 an optional ``unnamed_addr`` attribute, a return type, an optional
632 :ref:`parameter attribute <paramattrs>` for the return type, a function
633 name, a (possibly empty) argument list (each with optional :ref:`parameter
634 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
635 an optional section, an optional alignment,
636 an optional :ref:`comdat <langref_comdats>`,
637 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
638 an optional :ref:`prologue <prologuedata>`, an opening
639 curly brace, a list of basic blocks, and a closing curly brace.
641 LLVM function declarations consist of the "``declare``" keyword, an
642 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
643 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
644 an optional :ref:`calling convention <callingconv>`,
645 an optional ``unnamed_addr`` attribute, a return type, an optional
646 :ref:`parameter attribute <paramattrs>` for the return type, a function
647 name, a possibly empty list of arguments, an optional alignment, an optional
648 :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
649 and an optional :ref:`prologue <prologuedata>`.
651 A function definition contains a list of basic blocks, forming the CFG (Control
652 Flow Graph) for the function. Each basic block may optionally start with a label
653 (giving the basic block a symbol table entry), contains a list of instructions,
654 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
655 function return). If an explicit label is not provided, a block is assigned an
656 implicit numbered label, using the next value from the same counter as used for
657 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
658 entry block does not have an explicit label, it will be assigned label "%0",
659 then the first unnamed temporary in that block will be "%1", etc.
661 The first basic block in a function is special in two ways: it is
662 immediately executed on entrance to the function, and it is not allowed
663 to have predecessor basic blocks (i.e. there can not be any branches to
664 the entry block of a function). Because the block can have no
665 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
667 LLVM allows an explicit section to be specified for functions. If the
668 target supports it, it will emit functions to the section specified.
669 Additionally, the function can be placed in a COMDAT.
671 An explicit alignment may be specified for a function. If not present,
672 or if the alignment is set to zero, the alignment of the function is set
673 by the target to whatever it feels convenient. If an explicit alignment
674 is specified, the function is forced to have at least that much
675 alignment. All alignments must be a power of 2.
677 If the ``unnamed_addr`` attribute is given, the address is known to not
678 be significant and two identical functions can be merged.
682 define [linkage] [visibility] [DLLStorageClass]
684 <ResultType> @<FunctionName> ([argument list])
685 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
686 [align N] [gc] [prefix Constant] [prologue Constant] { ... }
688 The argument list is a comma seperated sequence of arguments where each
689 argument is of the following form
693 <type> [parameter Attrs] [name]
701 Aliases, unlike function or variables, don't create any new data. They
702 are just a new symbol and metadata for an existing position.
704 Aliases have a name and an aliasee that is either a global value or a
707 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
708 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
709 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
713 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy> @<Aliasee>
715 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
716 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
717 might not correctly handle dropping a weak symbol that is aliased.
719 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
720 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
723 Since aliases are only a second name, some restrictions apply, of which
724 some can only be checked when producing an object file:
726 * The expression defining the aliasee must be computable at assembly
727 time. Since it is just a name, no relocations can be used.
729 * No alias in the expression can be weak as the possibility of the
730 intermediate alias being overridden cannot be represented in an
733 * No global value in the expression can be a declaration, since that
734 would require a relocation, which is not possible.
741 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
743 Comdats have a name which represents the COMDAT key. All global objects that
744 specify this key will only end up in the final object file if the linker chooses
745 that key over some other key. Aliases are placed in the same COMDAT that their
746 aliasee computes to, if any.
748 Comdats have a selection kind to provide input on how the linker should
749 choose between keys in two different object files.
753 $<Name> = comdat SelectionKind
755 The selection kind must be one of the following:
758 The linker may choose any COMDAT key, the choice is arbitrary.
760 The linker may choose any COMDAT key but the sections must contain the
763 The linker will choose the section containing the largest COMDAT key.
765 The linker requires that only section with this COMDAT key exist.
767 The linker may choose any COMDAT key but the sections must contain the
770 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
771 ``any`` as a selection kind.
773 Here is an example of a COMDAT group where a function will only be selected if
774 the COMDAT key's section is the largest:
778 $foo = comdat largest
779 @foo = global i32 2, comdat($foo)
781 define void @bar() comdat($foo) {
785 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
791 @foo = global i32 2, comdat
794 In a COFF object file, this will create a COMDAT section with selection kind
795 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
796 and another COMDAT section with selection kind
797 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
798 section and contains the contents of the ``@bar`` symbol.
800 There are some restrictions on the properties of the global object.
801 It, or an alias to it, must have the same name as the COMDAT group when
803 The contents and size of this object may be used during link-time to determine
804 which COMDAT groups get selected depending on the selection kind.
805 Because the name of the object must match the name of the COMDAT group, the
806 linkage of the global object must not be local; local symbols can get renamed
807 if a collision occurs in the symbol table.
809 The combined use of COMDATS and section attributes may yield surprising results.
816 @g1 = global i32 42, section "sec", comdat($foo)
817 @g2 = global i32 42, section "sec", comdat($bar)
819 From the object file perspective, this requires the creation of two sections
820 with the same name. This is necessary because both globals belong to different
821 COMDAT groups and COMDATs, at the object file level, are represented by
824 Note that certain IR constructs like global variables and functions may create
825 COMDATs in the object file in addition to any which are specified using COMDAT
826 IR. This arises, for example, when a global variable has linkonce_odr linkage.
828 .. _namedmetadatastructure:
833 Named metadata is a collection of metadata. :ref:`Metadata
834 nodes <metadata>` (but not metadata strings) are the only valid
835 operands for a named metadata.
839 ; Some unnamed metadata nodes, which are referenced by the named metadata.
844 !name = !{!0, !1, !2}
851 The return type and each parameter of a function type may have a set of
852 *parameter attributes* associated with them. Parameter attributes are
853 used to communicate additional information about the result or
854 parameters of a function. Parameter attributes are considered to be part
855 of the function, not of the function type, so functions with different
856 parameter attributes can have the same function type.
858 Parameter attributes are simple keywords that follow the type specified.
859 If multiple parameter attributes are needed, they are space separated.
864 declare i32 @printf(i8* noalias nocapture, ...)
865 declare i32 @atoi(i8 zeroext)
866 declare signext i8 @returns_signed_char()
868 Note that any attributes for the function result (``nounwind``,
869 ``readonly``) come immediately after the argument list.
871 Currently, only the following parameter attributes are defined:
874 This indicates to the code generator that the parameter or return
875 value should be zero-extended to the extent required by the target's
876 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
877 the caller (for a parameter) or the callee (for a return value).
879 This indicates to the code generator that the parameter or return
880 value should be sign-extended to the extent required by the target's
881 ABI (which is usually 32-bits) by the caller (for a parameter) or
882 the callee (for a return value).
884 This indicates that this parameter or return value should be treated
885 in a special target-dependent fashion during while emitting code for
886 a function call or return (usually, by putting it in a register as
887 opposed to memory, though some targets use it to distinguish between
888 two different kinds of registers). Use of this attribute is
891 This indicates that the pointer parameter should really be passed by
892 value to the function. The attribute implies that a hidden copy of
893 the pointee is made between the caller and the callee, so the callee
894 is unable to modify the value in the caller. This attribute is only
895 valid on LLVM pointer arguments. It is generally used to pass
896 structs and arrays by value, but is also valid on pointers to
897 scalars. The copy is considered to belong to the caller not the
898 callee (for example, ``readonly`` functions should not write to
899 ``byval`` parameters). This is not a valid attribute for return
902 The byval attribute also supports specifying an alignment with the
903 align attribute. It indicates the alignment of the stack slot to
904 form and the known alignment of the pointer specified to the call
905 site. If the alignment is not specified, then the code generator
906 makes a target-specific assumption.
912 The ``inalloca`` argument attribute allows the caller to take the
913 address of outgoing stack arguments. An ``inalloca`` argument must
914 be a pointer to stack memory produced by an ``alloca`` instruction.
915 The alloca, or argument allocation, must also be tagged with the
916 inalloca keyword. Only the last argument may have the ``inalloca``
917 attribute, and that argument is guaranteed to be passed in memory.
919 An argument allocation may be used by a call at most once because
920 the call may deallocate it. The ``inalloca`` attribute cannot be
921 used in conjunction with other attributes that affect argument
922 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
923 ``inalloca`` attribute also disables LLVM's implicit lowering of
924 large aggregate return values, which means that frontend authors
925 must lower them with ``sret`` pointers.
927 When the call site is reached, the argument allocation must have
928 been the most recent stack allocation that is still live, or the
929 results are undefined. It is possible to allocate additional stack
930 space after an argument allocation and before its call site, but it
931 must be cleared off with :ref:`llvm.stackrestore
934 See :doc:`InAlloca` for more information on how to use this
938 This indicates that the pointer parameter specifies the address of a
939 structure that is the return value of the function in the source
940 program. This pointer must be guaranteed by the caller to be valid:
941 loads and stores to the structure may be assumed by the callee
942 not to trap and to be properly aligned. This may only be applied to
943 the first parameter. This is not a valid attribute for return
947 This indicates that the pointer value may be assumed by the optimizer to
948 have the specified alignment.
950 Note that this attribute has additional semantics when combined with the
956 This indicates that objects accessed via pointer values
957 :ref:`based <pointeraliasing>` on the argument or return value are not also
958 accessed, during the execution of the function, via pointer values not
959 *based* on the argument or return value. The attribute on a return value
960 also has additional semantics described below. The caller shares the
961 responsibility with the callee for ensuring that these requirements are met.
962 For further details, please see the discussion of the NoAlias response in
963 :ref:`alias analysis <Must, May, or No>`.
965 Note that this definition of ``noalias`` is intentionally similar
966 to the definition of ``restrict`` in C99 for function arguments.
968 For function return values, C99's ``restrict`` is not meaningful,
969 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
970 attribute on return values are stronger than the semantics of the attribute
971 when used on function arguments. On function return values, the ``noalias``
972 attribute indicates that the function acts like a system memory allocation
973 function, returning a pointer to allocated storage disjoint from the
974 storage for any other object accessible to the caller.
977 This indicates that the callee does not make any copies of the
978 pointer that outlive the callee itself. This is not a valid
979 attribute for return values.
984 This indicates that the pointer parameter can be excised using the
985 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
986 attribute for return values and can only be applied to one parameter.
989 This indicates that the function always returns the argument as its return
990 value. This is an optimization hint to the code generator when generating
991 the caller, allowing tail call optimization and omission of register saves
992 and restores in some cases; it is not checked or enforced when generating
993 the callee. The parameter and the function return type must be valid
994 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
995 valid attribute for return values and can only be applied to one parameter.
998 This indicates that the parameter or return pointer is not null. This
999 attribute may only be applied to pointer typed parameters. This is not
1000 checked or enforced by LLVM, the caller must ensure that the pointer
1001 passed in is non-null, or the callee must ensure that the returned pointer
1004 ``dereferenceable(<n>)``
1005 This indicates that the parameter or return pointer is dereferenceable. This
1006 attribute may only be applied to pointer typed parameters. A pointer that
1007 is dereferenceable can be loaded from speculatively without a risk of
1008 trapping. The number of bytes known to be dereferenceable must be provided
1009 in parentheses. It is legal for the number of bytes to be less than the
1010 size of the pointee type. The ``nonnull`` attribute does not imply
1011 dereferenceability (consider a pointer to one element past the end of an
1012 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1013 ``addrspace(0)`` (which is the default address space).
1015 ``dereferenceable_or_null(<n>)``
1016 This indicates that the parameter or return value isn't both
1017 non-null and non-dereferenceable (up to ``<n>`` bytes) at the same
1018 time. All non-null pointers tagged with
1019 ``dereferenceable_or_null(<n>)`` are ``dereferenceable(<n>)``.
1020 For address space 0 ``dereferenceable_or_null(<n>)`` implies that
1021 a pointer is exactly one of ``dereferenceable(<n>)`` or ``null``,
1022 and in other address spaces ``dereferenceable_or_null(<n>)``
1023 implies that a pointer is at least one of ``dereferenceable(<n>)``
1024 or ``null`` (i.e. it may be both ``null`` and
1025 ``dereferenceable(<n>)``). This attribute may only be applied to
1026 pointer typed parameters.
1030 Garbage Collector Strategy Names
1031 --------------------------------
1033 Each function may specify a garbage collector strategy name, which is simply a
1036 .. code-block:: llvm
1038 define void @f() gc "name" { ... }
1040 The supported values of *name* includes those :ref:`built in to LLVM
1041 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1042 strategy will cause the compiler to alter its output in order to support the
1043 named garbage collection algorithm. Note that LLVM itself does not contain a
1044 garbage collector, this functionality is restricted to generating machine code
1045 which can interoperate with a collector provided externally.
1052 Prefix data is data associated with a function which the code
1053 generator will emit immediately before the function's entrypoint.
1054 The purpose of this feature is to allow frontends to associate
1055 language-specific runtime metadata with specific functions and make it
1056 available through the function pointer while still allowing the
1057 function pointer to be called.
1059 To access the data for a given function, a program may bitcast the
1060 function pointer to a pointer to the constant's type and dereference
1061 index -1. This implies that the IR symbol points just past the end of
1062 the prefix data. For instance, take the example of a function annotated
1063 with a single ``i32``,
1065 .. code-block:: llvm
1067 define void @f() prefix i32 123 { ... }
1069 The prefix data can be referenced as,
1071 .. code-block:: llvm
1073 %0 = bitcast void* () @f to i32*
1074 %a = getelementptr inbounds i32, i32* %0, i32 -1
1075 %b = load i32, i32* %a
1077 Prefix data is laid out as if it were an initializer for a global variable
1078 of the prefix data's type. The function will be placed such that the
1079 beginning of the prefix data is aligned. This means that if the size
1080 of the prefix data is not a multiple of the alignment size, the
1081 function's entrypoint will not be aligned. If alignment of the
1082 function's entrypoint is desired, padding must be added to the prefix
1085 A function may have prefix data but no body. This has similar semantics
1086 to the ``available_externally`` linkage in that the data may be used by the
1087 optimizers but will not be emitted in the object file.
1094 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1095 be inserted prior to the function body. This can be used for enabling
1096 function hot-patching and instrumentation.
1098 To maintain the semantics of ordinary function calls, the prologue data must
1099 have a particular format. Specifically, it must begin with a sequence of
1100 bytes which decode to a sequence of machine instructions, valid for the
1101 module's target, which transfer control to the point immediately succeeding
1102 the prologue data, without performing any other visible action. This allows
1103 the inliner and other passes to reason about the semantics of the function
1104 definition without needing to reason about the prologue data. Obviously this
1105 makes the format of the prologue data highly target dependent.
1107 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1108 which encodes the ``nop`` instruction:
1110 .. code-block:: llvm
1112 define void @f() prologue i8 144 { ... }
1114 Generally prologue data can be formed by encoding a relative branch instruction
1115 which skips the metadata, as in this example of valid prologue data for the
1116 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1118 .. code-block:: llvm
1120 %0 = type <{ i8, i8, i8* }>
1122 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1124 A function may have prologue data but no body. This has similar semantics
1125 to the ``available_externally`` linkage in that the data may be used by the
1126 optimizers but will not be emitted in the object file.
1133 Attribute groups are groups of attributes that are referenced by objects within
1134 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1135 functions will use the same set of attributes. In the degenerative case of a
1136 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1137 group will capture the important command line flags used to build that file.
1139 An attribute group is a module-level object. To use an attribute group, an
1140 object references the attribute group's ID (e.g. ``#37``). An object may refer
1141 to more than one attribute group. In that situation, the attributes from the
1142 different groups are merged.
1144 Here is an example of attribute groups for a function that should always be
1145 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1147 .. code-block:: llvm
1149 ; Target-independent attributes:
1150 attributes #0 = { alwaysinline alignstack=4 }
1152 ; Target-dependent attributes:
1153 attributes #1 = { "no-sse" }
1155 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1156 define void @f() #0 #1 { ... }
1163 Function attributes are set to communicate additional information about
1164 a function. Function attributes are considered to be part of the
1165 function, not of the function type, so functions with different function
1166 attributes can have the same function type.
1168 Function attributes are simple keywords that follow the type specified.
1169 If multiple attributes are needed, they are space separated. For
1172 .. code-block:: llvm
1174 define void @f() noinline { ... }
1175 define void @f() alwaysinline { ... }
1176 define void @f() alwaysinline optsize { ... }
1177 define void @f() optsize { ... }
1180 This attribute indicates that, when emitting the prologue and
1181 epilogue, the backend should forcibly align the stack pointer.
1182 Specify the desired alignment, which must be a power of two, in
1185 This attribute indicates that the inliner should attempt to inline
1186 this function into callers whenever possible, ignoring any active
1187 inlining size threshold for this caller.
1189 This indicates that the callee function at a call site should be
1190 recognized as a built-in function, even though the function's declaration
1191 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1192 direct calls to functions that are declared with the ``nobuiltin``
1195 This attribute indicates that this function is rarely called. When
1196 computing edge weights, basic blocks post-dominated by a cold
1197 function call are also considered to be cold; and, thus, given low
1200 This attribute indicates that the source code contained a hint that
1201 inlining this function is desirable (such as the "inline" keyword in
1202 C/C++). It is just a hint; it imposes no requirements on the
1205 This attribute indicates that the function should be added to a
1206 jump-instruction table at code-generation time, and that all address-taken
1207 references to this function should be replaced with a reference to the
1208 appropriate jump-instruction-table function pointer. Note that this creates
1209 a new pointer for the original function, which means that code that depends
1210 on function-pointer identity can break. So, any function annotated with
1211 ``jumptable`` must also be ``unnamed_addr``.
1213 This attribute suggests that optimization passes and code generator
1214 passes make choices that keep the code size of this function as small
1215 as possible and perform optimizations that may sacrifice runtime
1216 performance in order to minimize the size of the generated code.
1218 This attribute disables prologue / epilogue emission for the
1219 function. This can have very system-specific consequences.
1221 This indicates that the callee function at a call site is not recognized as
1222 a built-in function. LLVM will retain the original call and not replace it
1223 with equivalent code based on the semantics of the built-in function, unless
1224 the call site uses the ``builtin`` attribute. This is valid at call sites
1225 and on function declarations and definitions.
1227 This attribute indicates that calls to the function cannot be
1228 duplicated. A call to a ``noduplicate`` function may be moved
1229 within its parent function, but may not be duplicated within
1230 its parent function.
1232 A function containing a ``noduplicate`` call may still
1233 be an inlining candidate, provided that the call is not
1234 duplicated by inlining. That implies that the function has
1235 internal linkage and only has one call site, so the original
1236 call is dead after inlining.
1238 This attributes disables implicit floating point instructions.
1240 This attribute indicates that the inliner should never inline this
1241 function in any situation. This attribute may not be used together
1242 with the ``alwaysinline`` attribute.
1244 This attribute suppresses lazy symbol binding for the function. This
1245 may make calls to the function faster, at the cost of extra program
1246 startup time if the function is not called during program startup.
1248 This attribute indicates that the code generator should not use a
1249 red zone, even if the target-specific ABI normally permits it.
1251 This function attribute indicates that the function never returns
1252 normally. This produces undefined behavior at runtime if the
1253 function ever does dynamically return.
1255 This function attribute indicates that the function never raises an
1256 exception. If the function does raise an exception, its runtime
1257 behavior is undefined. However, functions marked nounwind may still
1258 trap or generate asynchronous exceptions. Exception handling schemes
1259 that are recognized by LLVM to handle asynchronous exceptions, such
1260 as SEH, will still provide their implementation defined semantics.
1262 This function attribute indicates that the function is not optimized
1263 by any optimization or code generator passes with the
1264 exception of interprocedural optimization passes.
1265 This attribute cannot be used together with the ``alwaysinline``
1266 attribute; this attribute is also incompatible
1267 with the ``minsize`` attribute and the ``optsize`` attribute.
1269 This attribute requires the ``noinline`` attribute to be specified on
1270 the function as well, so the function is never inlined into any caller.
1271 Only functions with the ``alwaysinline`` attribute are valid
1272 candidates for inlining into the body of this function.
1274 This attribute suggests that optimization passes and code generator
1275 passes make choices that keep the code size of this function low,
1276 and otherwise do optimizations specifically to reduce code size as
1277 long as they do not significantly impact runtime performance.
1279 On a function, this attribute indicates that the function computes its
1280 result (or decides to unwind an exception) based strictly on its arguments,
1281 without dereferencing any pointer arguments or otherwise accessing
1282 any mutable state (e.g. memory, control registers, etc) visible to
1283 caller functions. It does not write through any pointer arguments
1284 (including ``byval`` arguments) and never changes any state visible
1285 to callers. This means that it cannot unwind exceptions by calling
1286 the ``C++`` exception throwing methods.
1288 On an argument, this attribute indicates that the function does not
1289 dereference that pointer argument, even though it may read or write the
1290 memory that the pointer points to if accessed through other pointers.
1292 On a function, this attribute indicates that the function does not write
1293 through any pointer arguments (including ``byval`` arguments) or otherwise
1294 modify any state (e.g. memory, control registers, etc) visible to
1295 caller functions. It may dereference pointer arguments and read
1296 state that may be set in the caller. A readonly function always
1297 returns the same value (or unwinds an exception identically) when
1298 called with the same set of arguments and global state. It cannot
1299 unwind an exception by calling the ``C++`` exception throwing
1302 On an argument, this attribute indicates that the function does not write
1303 through this pointer argument, even though it may write to the memory that
1304 the pointer points to.
1306 This attribute indicates that this function can return twice. The C
1307 ``setjmp`` is an example of such a function. The compiler disables
1308 some optimizations (like tail calls) in the caller of these
1310 ``sanitize_address``
1311 This attribute indicates that AddressSanitizer checks
1312 (dynamic address safety analysis) are enabled for this function.
1314 This attribute indicates that MemorySanitizer checks (dynamic detection
1315 of accesses to uninitialized memory) are enabled for this function.
1317 This attribute indicates that ThreadSanitizer checks
1318 (dynamic thread safety analysis) are enabled for this function.
1320 This attribute indicates that the function should emit a stack
1321 smashing protector. It is in the form of a "canary" --- a random value
1322 placed on the stack before the local variables that's checked upon
1323 return from the function to see if it has been overwritten. A
1324 heuristic is used to determine if a function needs stack protectors
1325 or not. The heuristic used will enable protectors for functions with:
1327 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1328 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1329 - Calls to alloca() with variable sizes or constant sizes greater than
1330 ``ssp-buffer-size``.
1332 Variables that are identified as requiring a protector will be arranged
1333 on the stack such that they are adjacent to the stack protector guard.
1335 If a function that has an ``ssp`` attribute is inlined into a
1336 function that doesn't have an ``ssp`` attribute, then the resulting
1337 function will have an ``ssp`` attribute.
1339 This attribute indicates that the function should *always* emit a
1340 stack smashing protector. This overrides the ``ssp`` function
1343 Variables that are identified as requiring a protector will be arranged
1344 on the stack such that they are adjacent to the stack protector guard.
1345 The specific layout rules are:
1347 #. Large arrays and structures containing large arrays
1348 (``>= ssp-buffer-size``) are closest to the stack protector.
1349 #. Small arrays and structures containing small arrays
1350 (``< ssp-buffer-size``) are 2nd closest to the protector.
1351 #. Variables that have had their address taken are 3rd closest to the
1354 If a function that has an ``sspreq`` attribute is inlined into a
1355 function that doesn't have an ``sspreq`` attribute or which has an
1356 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1357 an ``sspreq`` attribute.
1359 This attribute indicates that the function should emit a stack smashing
1360 protector. This attribute causes a strong heuristic to be used when
1361 determining if a function needs stack protectors. The strong heuristic
1362 will enable protectors for functions with:
1364 - Arrays of any size and type
1365 - Aggregates containing an array of any size and type.
1366 - Calls to alloca().
1367 - Local variables that have had their address taken.
1369 Variables that are identified as requiring a protector will be arranged
1370 on the stack such that they are adjacent to the stack protector guard.
1371 The specific layout rules are:
1373 #. Large arrays and structures containing large arrays
1374 (``>= ssp-buffer-size``) are closest to the stack protector.
1375 #. Small arrays and structures containing small arrays
1376 (``< ssp-buffer-size``) are 2nd closest to the protector.
1377 #. Variables that have had their address taken are 3rd closest to the
1380 This overrides the ``ssp`` function attribute.
1382 If a function that has an ``sspstrong`` attribute is inlined into a
1383 function that doesn't have an ``sspstrong`` attribute, then the
1384 resulting function will have an ``sspstrong`` attribute.
1386 This attribute indicates that the function will delegate to some other
1387 function with a tail call. The prototype of a thunk should not be used for
1388 optimization purposes. The caller is expected to cast the thunk prototype to
1389 match the thunk target prototype.
1391 This attribute indicates that the ABI being targeted requires that
1392 an unwind table entry be produce for this function even if we can
1393 show that no exceptions passes by it. This is normally the case for
1394 the ELF x86-64 abi, but it can be disabled for some compilation
1399 Module-Level Inline Assembly
1400 ----------------------------
1402 Modules may contain "module-level inline asm" blocks, which corresponds
1403 to the GCC "file scope inline asm" blocks. These blocks are internally
1404 concatenated by LLVM and treated as a single unit, but may be separated
1405 in the ``.ll`` file if desired. The syntax is very simple:
1407 .. code-block:: llvm
1409 module asm "inline asm code goes here"
1410 module asm "more can go here"
1412 The strings can contain any character by escaping non-printable
1413 characters. The escape sequence used is simply "\\xx" where "xx" is the
1414 two digit hex code for the number.
1416 The inline asm code is simply printed to the machine code .s file when
1417 assembly code is generated.
1419 .. _langref_datalayout:
1424 A module may specify a target specific data layout string that specifies
1425 how data is to be laid out in memory. The syntax for the data layout is
1428 .. code-block:: llvm
1430 target datalayout = "layout specification"
1432 The *layout specification* consists of a list of specifications
1433 separated by the minus sign character ('-'). Each specification starts
1434 with a letter and may include other information after the letter to
1435 define some aspect of the data layout. The specifications accepted are
1439 Specifies that the target lays out data in big-endian form. That is,
1440 the bits with the most significance have the lowest address
1443 Specifies that the target lays out data in little-endian form. That
1444 is, the bits with the least significance have the lowest address
1447 Specifies the natural alignment of the stack in bits. Alignment
1448 promotion of stack variables is limited to the natural stack
1449 alignment to avoid dynamic stack realignment. The stack alignment
1450 must be a multiple of 8-bits. If omitted, the natural stack
1451 alignment defaults to "unspecified", which does not prevent any
1452 alignment promotions.
1453 ``p[n]:<size>:<abi>:<pref>``
1454 This specifies the *size* of a pointer and its ``<abi>`` and
1455 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1456 bits. The address space, ``n`` is optional, and if not specified,
1457 denotes the default address space 0. The value of ``n`` must be
1458 in the range [1,2^23).
1459 ``i<size>:<abi>:<pref>``
1460 This specifies the alignment for an integer type of a given bit
1461 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1462 ``v<size>:<abi>:<pref>``
1463 This specifies the alignment for a vector type of a given bit
1465 ``f<size>:<abi>:<pref>``
1466 This specifies the alignment for a floating point type of a given bit
1467 ``<size>``. Only values of ``<size>`` that are supported by the target
1468 will work. 32 (float) and 64 (double) are supported on all targets; 80
1469 or 128 (different flavors of long double) are also supported on some
1472 This specifies the alignment for an object of aggregate type.
1474 If present, specifies that llvm names are mangled in the output. The
1477 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1478 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1479 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1480 symbols get a ``_`` prefix.
1481 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1482 functions also get a suffix based on the frame size.
1483 ``n<size1>:<size2>:<size3>...``
1484 This specifies a set of native integer widths for the target CPU in
1485 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1486 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1487 this set are considered to support most general arithmetic operations
1490 On every specification that takes a ``<abi>:<pref>``, specifying the
1491 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1492 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1494 When constructing the data layout for a given target, LLVM starts with a
1495 default set of specifications which are then (possibly) overridden by
1496 the specifications in the ``datalayout`` keyword. The default
1497 specifications are given in this list:
1499 - ``E`` - big endian
1500 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1501 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1502 same as the default address space.
1503 - ``S0`` - natural stack alignment is unspecified
1504 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1505 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1506 - ``i16:16:16`` - i16 is 16-bit aligned
1507 - ``i32:32:32`` - i32 is 32-bit aligned
1508 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1509 alignment of 64-bits
1510 - ``f16:16:16`` - half is 16-bit aligned
1511 - ``f32:32:32`` - float is 32-bit aligned
1512 - ``f64:64:64`` - double is 64-bit aligned
1513 - ``f128:128:128`` - quad is 128-bit aligned
1514 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1515 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1516 - ``a:0:64`` - aggregates are 64-bit aligned
1518 When LLVM is determining the alignment for a given type, it uses the
1521 #. If the type sought is an exact match for one of the specifications,
1522 that specification is used.
1523 #. If no match is found, and the type sought is an integer type, then
1524 the smallest integer type that is larger than the bitwidth of the
1525 sought type is used. If none of the specifications are larger than
1526 the bitwidth then the largest integer type is used. For example,
1527 given the default specifications above, the i7 type will use the
1528 alignment of i8 (next largest) while both i65 and i256 will use the
1529 alignment of i64 (largest specified).
1530 #. If no match is found, and the type sought is a vector type, then the
1531 largest vector type that is smaller than the sought vector type will
1532 be used as a fall back. This happens because <128 x double> can be
1533 implemented in terms of 64 <2 x double>, for example.
1535 The function of the data layout string may not be what you expect.
1536 Notably, this is not a specification from the frontend of what alignment
1537 the code generator should use.
1539 Instead, if specified, the target data layout is required to match what
1540 the ultimate *code generator* expects. This string is used by the
1541 mid-level optimizers to improve code, and this only works if it matches
1542 what the ultimate code generator uses. There is no way to generate IR
1543 that does not embed this target-specific detail into the IR. If you
1544 don't specify the string, the default specifications will be used to
1545 generate a Data Layout and the optimization phases will operate
1546 accordingly and introduce target specificity into the IR with respect to
1547 these default specifications.
1554 A module may specify a target triple string that describes the target
1555 host. The syntax for the target triple is simply:
1557 .. code-block:: llvm
1559 target triple = "x86_64-apple-macosx10.7.0"
1561 The *target triple* string consists of a series of identifiers delimited
1562 by the minus sign character ('-'). The canonical forms are:
1566 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1567 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1569 This information is passed along to the backend so that it generates
1570 code for the proper architecture. It's possible to override this on the
1571 command line with the ``-mtriple`` command line option.
1573 .. _pointeraliasing:
1575 Pointer Aliasing Rules
1576 ----------------------
1578 Any memory access must be done through a pointer value associated with
1579 an address range of the memory access, otherwise the behavior is
1580 undefined. Pointer values are associated with address ranges according
1581 to the following rules:
1583 - A pointer value is associated with the addresses associated with any
1584 value it is *based* on.
1585 - An address of a global variable is associated with the address range
1586 of the variable's storage.
1587 - The result value of an allocation instruction is associated with the
1588 address range of the allocated storage.
1589 - A null pointer in the default address-space is associated with no
1591 - An integer constant other than zero or a pointer value returned from
1592 a function not defined within LLVM may be associated with address
1593 ranges allocated through mechanisms other than those provided by
1594 LLVM. Such ranges shall not overlap with any ranges of addresses
1595 allocated by mechanisms provided by LLVM.
1597 A pointer value is *based* on another pointer value according to the
1600 - A pointer value formed from a ``getelementptr`` operation is *based*
1601 on the first value operand of the ``getelementptr``.
1602 - The result value of a ``bitcast`` is *based* on the operand of the
1604 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1605 values that contribute (directly or indirectly) to the computation of
1606 the pointer's value.
1607 - The "*based* on" relationship is transitive.
1609 Note that this definition of *"based"* is intentionally similar to the
1610 definition of *"based"* in C99, though it is slightly weaker.
1612 LLVM IR does not associate types with memory. The result type of a
1613 ``load`` merely indicates the size and alignment of the memory from
1614 which to load, as well as the interpretation of the value. The first
1615 operand type of a ``store`` similarly only indicates the size and
1616 alignment of the store.
1618 Consequently, type-based alias analysis, aka TBAA, aka
1619 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1620 :ref:`Metadata <metadata>` may be used to encode additional information
1621 which specialized optimization passes may use to implement type-based
1626 Volatile Memory Accesses
1627 ------------------------
1629 Certain memory accesses, such as :ref:`load <i_load>`'s,
1630 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1631 marked ``volatile``. The optimizers must not change the number of
1632 volatile operations or change their order of execution relative to other
1633 volatile operations. The optimizers *may* change the order of volatile
1634 operations relative to non-volatile operations. This is not Java's
1635 "volatile" and has no cross-thread synchronization behavior.
1637 IR-level volatile loads and stores cannot safely be optimized into
1638 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1639 flagged volatile. Likewise, the backend should never split or merge
1640 target-legal volatile load/store instructions.
1642 .. admonition:: Rationale
1644 Platforms may rely on volatile loads and stores of natively supported
1645 data width to be executed as single instruction. For example, in C
1646 this holds for an l-value of volatile primitive type with native
1647 hardware support, but not necessarily for aggregate types. The
1648 frontend upholds these expectations, which are intentionally
1649 unspecified in the IR. The rules above ensure that IR transformation
1650 do not violate the frontend's contract with the language.
1654 Memory Model for Concurrent Operations
1655 --------------------------------------
1657 The LLVM IR does not define any way to start parallel threads of
1658 execution or to register signal handlers. Nonetheless, there are
1659 platform-specific ways to create them, and we define LLVM IR's behavior
1660 in their presence. This model is inspired by the C++0x memory model.
1662 For a more informal introduction to this model, see the :doc:`Atomics`.
1664 We define a *happens-before* partial order as the least partial order
1667 - Is a superset of single-thread program order, and
1668 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1669 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1670 techniques, like pthread locks, thread creation, thread joining,
1671 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1672 Constraints <ordering>`).
1674 Note that program order does not introduce *happens-before* edges
1675 between a thread and signals executing inside that thread.
1677 Every (defined) read operation (load instructions, memcpy, atomic
1678 loads/read-modify-writes, etc.) R reads a series of bytes written by
1679 (defined) write operations (store instructions, atomic
1680 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1681 section, initialized globals are considered to have a write of the
1682 initializer which is atomic and happens before any other read or write
1683 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1684 may see any write to the same byte, except:
1686 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1687 write\ :sub:`2` happens before R\ :sub:`byte`, then
1688 R\ :sub:`byte` does not see write\ :sub:`1`.
1689 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1690 R\ :sub:`byte` does not see write\ :sub:`3`.
1692 Given that definition, R\ :sub:`byte` is defined as follows:
1694 - If R is volatile, the result is target-dependent. (Volatile is
1695 supposed to give guarantees which can support ``sig_atomic_t`` in
1696 C/C++, and may be used for accesses to addresses that do not behave
1697 like normal memory. It does not generally provide cross-thread
1699 - Otherwise, if there is no write to the same byte that happens before
1700 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1701 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1702 R\ :sub:`byte` returns the value written by that write.
1703 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1704 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1705 Memory Ordering Constraints <ordering>` section for additional
1706 constraints on how the choice is made.
1707 - Otherwise R\ :sub:`byte` returns ``undef``.
1709 R returns the value composed of the series of bytes it read. This
1710 implies that some bytes within the value may be ``undef`` **without**
1711 the entire value being ``undef``. Note that this only defines the
1712 semantics of the operation; it doesn't mean that targets will emit more
1713 than one instruction to read the series of bytes.
1715 Note that in cases where none of the atomic intrinsics are used, this
1716 model places only one restriction on IR transformations on top of what
1717 is required for single-threaded execution: introducing a store to a byte
1718 which might not otherwise be stored is not allowed in general.
1719 (Specifically, in the case where another thread might write to and read
1720 from an address, introducing a store can change a load that may see
1721 exactly one write into a load that may see multiple writes.)
1725 Atomic Memory Ordering Constraints
1726 ----------------------------------
1728 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1729 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1730 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1731 ordering parameters that determine which other atomic instructions on
1732 the same address they *synchronize with*. These semantics are borrowed
1733 from Java and C++0x, but are somewhat more colloquial. If these
1734 descriptions aren't precise enough, check those specs (see spec
1735 references in the :doc:`atomics guide <Atomics>`).
1736 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1737 differently since they don't take an address. See that instruction's
1738 documentation for details.
1740 For a simpler introduction to the ordering constraints, see the
1744 The set of values that can be read is governed by the happens-before
1745 partial order. A value cannot be read unless some operation wrote
1746 it. This is intended to provide a guarantee strong enough to model
1747 Java's non-volatile shared variables. This ordering cannot be
1748 specified for read-modify-write operations; it is not strong enough
1749 to make them atomic in any interesting way.
1751 In addition to the guarantees of ``unordered``, there is a single
1752 total order for modifications by ``monotonic`` operations on each
1753 address. All modification orders must be compatible with the
1754 happens-before order. There is no guarantee that the modification
1755 orders can be combined to a global total order for the whole program
1756 (and this often will not be possible). The read in an atomic
1757 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1758 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1759 order immediately before the value it writes. If one atomic read
1760 happens before another atomic read of the same address, the later
1761 read must see the same value or a later value in the address's
1762 modification order. This disallows reordering of ``monotonic`` (or
1763 stronger) operations on the same address. If an address is written
1764 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1765 read that address repeatedly, the other threads must eventually see
1766 the write. This corresponds to the C++0x/C1x
1767 ``memory_order_relaxed``.
1769 In addition to the guarantees of ``monotonic``, a
1770 *synchronizes-with* edge may be formed with a ``release`` operation.
1771 This is intended to model C++'s ``memory_order_acquire``.
1773 In addition to the guarantees of ``monotonic``, if this operation
1774 writes a value which is subsequently read by an ``acquire``
1775 operation, it *synchronizes-with* that operation. (This isn't a
1776 complete description; see the C++0x definition of a release
1777 sequence.) This corresponds to the C++0x/C1x
1778 ``memory_order_release``.
1779 ``acq_rel`` (acquire+release)
1780 Acts as both an ``acquire`` and ``release`` operation on its
1781 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1782 ``seq_cst`` (sequentially consistent)
1783 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1784 operation that only reads, ``release`` for an operation that only
1785 writes), there is a global total order on all
1786 sequentially-consistent operations on all addresses, which is
1787 consistent with the *happens-before* partial order and with the
1788 modification orders of all the affected addresses. Each
1789 sequentially-consistent read sees the last preceding write to the
1790 same address in this global order. This corresponds to the C++0x/C1x
1791 ``memory_order_seq_cst`` and Java volatile.
1795 If an atomic operation is marked ``singlethread``, it only *synchronizes
1796 with* or participates in modification and seq\_cst total orderings with
1797 other operations running in the same thread (for example, in signal
1805 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1806 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1807 :ref:`frem <i_frem>`) have the following flags that can be set to enable
1808 otherwise unsafe floating point operations
1811 No NaNs - Allow optimizations to assume the arguments and result are not
1812 NaN. Such optimizations are required to retain defined behavior over
1813 NaNs, but the value of the result is undefined.
1816 No Infs - Allow optimizations to assume the arguments and result are not
1817 +/-Inf. Such optimizations are required to retain defined behavior over
1818 +/-Inf, but the value of the result is undefined.
1821 No Signed Zeros - Allow optimizations to treat the sign of a zero
1822 argument or result as insignificant.
1825 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1826 argument rather than perform division.
1829 Fast - Allow algebraically equivalent transformations that may
1830 dramatically change results in floating point (e.g. reassociate). This
1831 flag implies all the others.
1835 Use-list Order Directives
1836 -------------------------
1838 Use-list directives encode the in-memory order of each use-list, allowing the
1839 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1840 indexes that are assigned to the referenced value's uses. The referenced
1841 value's use-list is immediately sorted by these indexes.
1843 Use-list directives may appear at function scope or global scope. They are not
1844 instructions, and have no effect on the semantics of the IR. When they're at
1845 function scope, they must appear after the terminator of the final basic block.
1847 If basic blocks have their address taken via ``blockaddress()`` expressions,
1848 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1855 uselistorder <ty> <value>, { <order-indexes> }
1856 uselistorder_bb @function, %block { <order-indexes> }
1862 define void @foo(i32 %arg1, i32 %arg2) {
1864 ; ... instructions ...
1866 ; ... instructions ...
1868 ; At function scope.
1869 uselistorder i32 %arg1, { 1, 0, 2 }
1870 uselistorder label %bb, { 1, 0 }
1874 uselistorder i32* @global, { 1, 2, 0 }
1875 uselistorder i32 7, { 1, 0 }
1876 uselistorder i32 (i32) @bar, { 1, 0 }
1877 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1884 The LLVM type system is one of the most important features of the
1885 intermediate representation. Being typed enables a number of
1886 optimizations to be performed on the intermediate representation
1887 directly, without having to do extra analyses on the side before the
1888 transformation. A strong type system makes it easier to read the
1889 generated code and enables novel analyses and transformations that are
1890 not feasible to perform on normal three address code representations.
1900 The void type does not represent any value and has no size.
1918 The function type can be thought of as a function signature. It consists of a
1919 return type and a list of formal parameter types. The return type of a function
1920 type is a void type or first class type --- except for :ref:`label <t_label>`
1921 and :ref:`metadata <t_metadata>` types.
1927 <returntype> (<parameter list>)
1929 ...where '``<parameter list>``' is a comma-separated list of type
1930 specifiers. Optionally, the parameter list may include a type ``...``, which
1931 indicates that the function takes a variable number of arguments. Variable
1932 argument functions can access their arguments with the :ref:`variable argument
1933 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1934 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1938 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1939 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1940 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1941 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1942 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1943 | ``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. |
1944 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1945 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1946 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1953 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1954 Values of these types are the only ones which can be produced by
1962 These are the types that are valid in registers from CodeGen's perspective.
1971 The integer type is a very simple type that simply specifies an
1972 arbitrary bit width for the integer type desired. Any bit width from 1
1973 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1981 The number of bits the integer will occupy is specified by the ``N``
1987 +----------------+------------------------------------------------+
1988 | ``i1`` | a single-bit integer. |
1989 +----------------+------------------------------------------------+
1990 | ``i32`` | a 32-bit integer. |
1991 +----------------+------------------------------------------------+
1992 | ``i1942652`` | a really big integer of over 1 million bits. |
1993 +----------------+------------------------------------------------+
1997 Floating Point Types
1998 """"""""""""""""""""
2007 - 16-bit floating point value
2010 - 32-bit floating point value
2013 - 64-bit floating point value
2016 - 128-bit floating point value (112-bit mantissa)
2019 - 80-bit floating point value (X87)
2022 - 128-bit floating point value (two 64-bits)
2029 The x86_mmx type represents a value held in an MMX register on an x86
2030 machine. The operations allowed on it are quite limited: parameters and
2031 return values, load and store, and bitcast. User-specified MMX
2032 instructions are represented as intrinsic or asm calls with arguments
2033 and/or results of this type. There are no arrays, vectors or constants
2050 The pointer type is used to specify memory locations. Pointers are
2051 commonly used to reference objects in memory.
2053 Pointer types may have an optional address space attribute defining the
2054 numbered address space where the pointed-to object resides. The default
2055 address space is number zero. The semantics of non-zero address spaces
2056 are target-specific.
2058 Note that LLVM does not permit pointers to void (``void*``) nor does it
2059 permit pointers to labels (``label*``). Use ``i8*`` instead.
2069 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2070 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2071 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2072 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2073 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2074 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2075 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2084 A vector type is a simple derived type that represents a vector of
2085 elements. Vector types are used when multiple primitive data are
2086 operated in parallel using a single instruction (SIMD). A vector type
2087 requires a size (number of elements) and an underlying primitive data
2088 type. Vector types are considered :ref:`first class <t_firstclass>`.
2094 < <# elements> x <elementtype> >
2096 The number of elements is a constant integer value larger than 0;
2097 elementtype may be any integer, floating point or pointer type. Vectors
2098 of size zero are not allowed.
2102 +-------------------+--------------------------------------------------+
2103 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2104 +-------------------+--------------------------------------------------+
2105 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2106 +-------------------+--------------------------------------------------+
2107 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2108 +-------------------+--------------------------------------------------+
2109 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2110 +-------------------+--------------------------------------------------+
2119 The label type represents code labels.
2134 The metadata type represents embedded metadata. No derived types may be
2135 created from metadata except for :ref:`function <t_function>` arguments.
2148 Aggregate Types are a subset of derived types that can contain multiple
2149 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2150 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2160 The array type is a very simple derived type that arranges elements
2161 sequentially in memory. The array type requires a size (number of
2162 elements) and an underlying data type.
2168 [<# elements> x <elementtype>]
2170 The number of elements is a constant integer value; ``elementtype`` may
2171 be any type with a size.
2175 +------------------+--------------------------------------+
2176 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2177 +------------------+--------------------------------------+
2178 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2179 +------------------+--------------------------------------+
2180 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2181 +------------------+--------------------------------------+
2183 Here are some examples of multidimensional arrays:
2185 +-----------------------------+----------------------------------------------------------+
2186 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2187 +-----------------------------+----------------------------------------------------------+
2188 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2189 +-----------------------------+----------------------------------------------------------+
2190 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2191 +-----------------------------+----------------------------------------------------------+
2193 There is no restriction on indexing beyond the end of the array implied
2194 by a static type (though there are restrictions on indexing beyond the
2195 bounds of an allocated object in some cases). This means that
2196 single-dimension 'variable sized array' addressing can be implemented in
2197 LLVM with a zero length array type. An implementation of 'pascal style
2198 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2208 The structure type is used to represent a collection of data members
2209 together in memory. The elements of a structure may be any type that has
2212 Structures in memory are accessed using '``load``' and '``store``' by
2213 getting a pointer to a field with the '``getelementptr``' instruction.
2214 Structures in registers are accessed using the '``extractvalue``' and
2215 '``insertvalue``' instructions.
2217 Structures may optionally be "packed" structures, which indicate that
2218 the alignment of the struct is one byte, and that there is no padding
2219 between the elements. In non-packed structs, padding between field types
2220 is inserted as defined by the DataLayout string in the module, which is
2221 required to match what the underlying code generator expects.
2223 Structures can either be "literal" or "identified". A literal structure
2224 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2225 identified types are always defined at the top level with a name.
2226 Literal types are uniqued by their contents and can never be recursive
2227 or opaque since there is no way to write one. Identified types can be
2228 recursive, can be opaqued, and are never uniqued.
2234 %T1 = type { <type list> } ; Identified normal struct type
2235 %T2 = type <{ <type list> }> ; Identified packed struct type
2239 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2240 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2241 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2242 | ``{ 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``. |
2243 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2244 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2245 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2249 Opaque Structure Types
2250 """"""""""""""""""""""
2254 Opaque structure types are used to represent named structure types that
2255 do not have a body specified. This corresponds (for example) to the C
2256 notion of a forward declared structure.
2267 +--------------+-------------------+
2268 | ``opaque`` | An opaque type. |
2269 +--------------+-------------------+
2276 LLVM has several different basic types of constants. This section
2277 describes them all and their syntax.
2282 **Boolean constants**
2283 The two strings '``true``' and '``false``' are both valid constants
2285 **Integer constants**
2286 Standard integers (such as '4') are constants of the
2287 :ref:`integer <t_integer>` type. Negative numbers may be used with
2289 **Floating point constants**
2290 Floating point constants use standard decimal notation (e.g.
2291 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2292 hexadecimal notation (see below). The assembler requires the exact
2293 decimal value of a floating-point constant. For example, the
2294 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2295 decimal in binary. Floating point constants must have a :ref:`floating
2296 point <t_floating>` type.
2297 **Null pointer constants**
2298 The identifier '``null``' is recognized as a null pointer constant
2299 and must be of :ref:`pointer type <t_pointer>`.
2301 The one non-intuitive notation for constants is the hexadecimal form of
2302 floating point constants. For example, the form
2303 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2304 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2305 constants are required (and the only time that they are generated by the
2306 disassembler) is when a floating point constant must be emitted but it
2307 cannot be represented as a decimal floating point number in a reasonable
2308 number of digits. For example, NaN's, infinities, and other special
2309 values are represented in their IEEE hexadecimal format so that assembly
2310 and disassembly do not cause any bits to change in the constants.
2312 When using the hexadecimal form, constants of types half, float, and
2313 double are represented using the 16-digit form shown above (which
2314 matches the IEEE754 representation for double); half and float values
2315 must, however, be exactly representable as IEEE 754 half and single
2316 precision, respectively. Hexadecimal format is always used for long
2317 double, and there are three forms of long double. The 80-bit format used
2318 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2319 128-bit format used by PowerPC (two adjacent doubles) is represented by
2320 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2321 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2322 will only work if they match the long double format on your target.
2323 The IEEE 16-bit format (half precision) is represented by ``0xH``
2324 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2325 (sign bit at the left).
2327 There are no constants of type x86_mmx.
2329 .. _complexconstants:
2334 Complex constants are a (potentially recursive) combination of simple
2335 constants and smaller complex constants.
2337 **Structure constants**
2338 Structure constants are represented with notation similar to
2339 structure type definitions (a comma separated list of elements,
2340 surrounded by braces (``{}``)). For example:
2341 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2342 "``@G = external global i32``". Structure constants must have
2343 :ref:`structure type <t_struct>`, and the number and types of elements
2344 must match those specified by the type.
2346 Array constants are represented with notation similar to array type
2347 definitions (a comma separated list of elements, surrounded by
2348 square brackets (``[]``)). For example:
2349 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2350 :ref:`array type <t_array>`, and the number and types of elements must
2351 match those specified by the type. As a special case, character array
2352 constants may also be represented as a double-quoted string using the ``c``
2353 prefix. For example: "``c"Hello World\0A\00"``".
2354 **Vector constants**
2355 Vector constants are represented with notation similar to vector
2356 type definitions (a comma separated list of elements, surrounded by
2357 less-than/greater-than's (``<>``)). For example:
2358 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2359 must have :ref:`vector type <t_vector>`, and the number and types of
2360 elements must match those specified by the type.
2361 **Zero initialization**
2362 The string '``zeroinitializer``' can be used to zero initialize a
2363 value to zero of *any* type, including scalar and
2364 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2365 having to print large zero initializers (e.g. for large arrays) and
2366 is always exactly equivalent to using explicit zero initializers.
2368 A metadata node is a constant tuple without types. For example:
2369 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2370 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2371 Unlike other typed constants that are meant to be interpreted as part of
2372 the instruction stream, metadata is a place to attach additional
2373 information such as debug info.
2375 Global Variable and Function Addresses
2376 --------------------------------------
2378 The addresses of :ref:`global variables <globalvars>` and
2379 :ref:`functions <functionstructure>` are always implicitly valid
2380 (link-time) constants. These constants are explicitly referenced when
2381 the :ref:`identifier for the global <identifiers>` is used and always have
2382 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2385 .. code-block:: llvm
2389 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2396 The string '``undef``' can be used anywhere a constant is expected, and
2397 indicates that the user of the value may receive an unspecified
2398 bit-pattern. Undefined values may be of any type (other than '``label``'
2399 or '``void``') and be used anywhere a constant is permitted.
2401 Undefined values are useful because they indicate to the compiler that
2402 the program is well defined no matter what value is used. This gives the
2403 compiler more freedom to optimize. Here are some examples of
2404 (potentially surprising) transformations that are valid (in pseudo IR):
2406 .. code-block:: llvm
2416 This is safe because all of the output bits are affected by the undef
2417 bits. Any output bit can have a zero or one depending on the input bits.
2419 .. code-block:: llvm
2430 These logical operations have bits that are not always affected by the
2431 input. For example, if ``%X`` has a zero bit, then the output of the
2432 '``and``' operation will always be a zero for that bit, no matter what
2433 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2434 optimize or assume that the result of the '``and``' is '``undef``'.
2435 However, it is safe to assume that all bits of the '``undef``' could be
2436 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2437 all the bits of the '``undef``' operand to the '``or``' could be set,
2438 allowing the '``or``' to be folded to -1.
2440 .. code-block:: llvm
2442 %A = select undef, %X, %Y
2443 %B = select undef, 42, %Y
2444 %C = select %X, %Y, undef
2454 This set of examples shows that undefined '``select``' (and conditional
2455 branch) conditions can go *either way*, but they have to come from one
2456 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2457 both known to have a clear low bit, then ``%A`` would have to have a
2458 cleared low bit. However, in the ``%C`` example, the optimizer is
2459 allowed to assume that the '``undef``' operand could be the same as
2460 ``%Y``, allowing the whole '``select``' to be eliminated.
2462 .. code-block:: llvm
2464 %A = xor undef, undef
2481 This example points out that two '``undef``' operands are not
2482 necessarily the same. This can be surprising to people (and also matches
2483 C semantics) where they assume that "``X^X``" is always zero, even if
2484 ``X`` is undefined. This isn't true for a number of reasons, but the
2485 short answer is that an '``undef``' "variable" can arbitrarily change
2486 its value over its "live range". This is true because the variable
2487 doesn't actually *have a live range*. Instead, the value is logically
2488 read from arbitrary registers that happen to be around when needed, so
2489 the value is not necessarily consistent over time. In fact, ``%A`` and
2490 ``%C`` need to have the same semantics or the core LLVM "replace all
2491 uses with" concept would not hold.
2493 .. code-block:: llvm
2501 These examples show the crucial difference between an *undefined value*
2502 and *undefined behavior*. An undefined value (like '``undef``') is
2503 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2504 operation can be constant folded to '``undef``', because the '``undef``'
2505 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2506 However, in the second example, we can make a more aggressive
2507 assumption: because the ``undef`` is allowed to be an arbitrary value,
2508 we are allowed to assume that it could be zero. Since a divide by zero
2509 has *undefined behavior*, we are allowed to assume that the operation
2510 does not execute at all. This allows us to delete the divide and all
2511 code after it. Because the undefined operation "can't happen", the
2512 optimizer can assume that it occurs in dead code.
2514 .. code-block:: llvm
2516 a: store undef -> %X
2517 b: store %X -> undef
2522 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2523 value can be assumed to not have any effect; we can assume that the
2524 value is overwritten with bits that happen to match what was already
2525 there. However, a store *to* an undefined location could clobber
2526 arbitrary memory, therefore, it has undefined behavior.
2533 Poison values are similar to :ref:`undef values <undefvalues>`, however
2534 they also represent the fact that an instruction or constant expression
2535 that cannot evoke side effects has nevertheless detected a condition
2536 that results in undefined behavior.
2538 There is currently no way of representing a poison value in the IR; they
2539 only exist when produced by operations such as :ref:`add <i_add>` with
2542 Poison value behavior is defined in terms of value *dependence*:
2544 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2545 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2546 their dynamic predecessor basic block.
2547 - Function arguments depend on the corresponding actual argument values
2548 in the dynamic callers of their functions.
2549 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2550 instructions that dynamically transfer control back to them.
2551 - :ref:`Invoke <i_invoke>` instructions depend on the
2552 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2553 call instructions that dynamically transfer control back to them.
2554 - Non-volatile loads and stores depend on the most recent stores to all
2555 of the referenced memory addresses, following the order in the IR
2556 (including loads and stores implied by intrinsics such as
2557 :ref:`@llvm.memcpy <int_memcpy>`.)
2558 - An instruction with externally visible side effects depends on the
2559 most recent preceding instruction with externally visible side
2560 effects, following the order in the IR. (This includes :ref:`volatile
2561 operations <volatile>`.)
2562 - An instruction *control-depends* on a :ref:`terminator
2563 instruction <terminators>` if the terminator instruction has
2564 multiple successors and the instruction is always executed when
2565 control transfers to one of the successors, and may not be executed
2566 when control is transferred to another.
2567 - Additionally, an instruction also *control-depends* on a terminator
2568 instruction if the set of instructions it otherwise depends on would
2569 be different if the terminator had transferred control to a different
2571 - Dependence is transitive.
2573 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2574 with the additional effect that any instruction that has a *dependence*
2575 on a poison value has undefined behavior.
2577 Here are some examples:
2579 .. code-block:: llvm
2582 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2583 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2584 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2585 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2587 store i32 %poison, i32* @g ; Poison value stored to memory.
2588 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2590 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2592 %narrowaddr = bitcast i32* @g to i16*
2593 %wideaddr = bitcast i32* @g to i64*
2594 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2595 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2597 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2598 br i1 %cmp, label %true, label %end ; Branch to either destination.
2601 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2602 ; it has undefined behavior.
2606 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2607 ; Both edges into this PHI are
2608 ; control-dependent on %cmp, so this
2609 ; always results in a poison value.
2611 store volatile i32 0, i32* @g ; This would depend on the store in %true
2612 ; if %cmp is true, or the store in %entry
2613 ; otherwise, so this is undefined behavior.
2615 br i1 %cmp, label %second_true, label %second_end
2616 ; The same branch again, but this time the
2617 ; true block doesn't have side effects.
2624 store volatile i32 0, i32* @g ; This time, the instruction always depends
2625 ; on the store in %end. Also, it is
2626 ; control-equivalent to %end, so this is
2627 ; well-defined (ignoring earlier undefined
2628 ; behavior in this example).
2632 Addresses of Basic Blocks
2633 -------------------------
2635 ``blockaddress(@function, %block)``
2637 The '``blockaddress``' constant computes the address of the specified
2638 basic block in the specified function, and always has an ``i8*`` type.
2639 Taking the address of the entry block is illegal.
2641 This value only has defined behavior when used as an operand to the
2642 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2643 against null. Pointer equality tests between labels addresses results in
2644 undefined behavior --- though, again, comparison against null is ok, and
2645 no label is equal to the null pointer. This may be passed around as an
2646 opaque pointer sized value as long as the bits are not inspected. This
2647 allows ``ptrtoint`` and arithmetic to be performed on these values so
2648 long as the original value is reconstituted before the ``indirectbr``
2651 Finally, some targets may provide defined semantics when using the value
2652 as the operand to an inline assembly, but that is target specific.
2656 Constant Expressions
2657 --------------------
2659 Constant expressions are used to allow expressions involving other
2660 constants to be used as constants. Constant expressions may be of any
2661 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2662 that does not have side effects (e.g. load and call are not supported).
2663 The following is the syntax for constant expressions:
2665 ``trunc (CST to TYPE)``
2666 Truncate a constant to another type. The bit size of CST must be
2667 larger than the bit size of TYPE. Both types must be integers.
2668 ``zext (CST to TYPE)``
2669 Zero extend a constant to another type. The bit size of CST must be
2670 smaller than the bit size of TYPE. Both types must be integers.
2671 ``sext (CST to TYPE)``
2672 Sign extend a constant to another type. The bit size of CST must be
2673 smaller than the bit size of TYPE. Both types must be integers.
2674 ``fptrunc (CST to TYPE)``
2675 Truncate a floating point constant to another floating point type.
2676 The size of CST must be larger than the size of TYPE. Both types
2677 must be floating point.
2678 ``fpext (CST to TYPE)``
2679 Floating point extend a constant to another type. The size of CST
2680 must be smaller or equal to the size of TYPE. Both types must be
2682 ``fptoui (CST to TYPE)``
2683 Convert a floating point constant to the corresponding unsigned
2684 integer constant. TYPE must be a scalar or vector integer type. CST
2685 must be of scalar or vector floating point type. Both CST and TYPE
2686 must be scalars, or vectors of the same number of elements. If the
2687 value won't fit in the integer type, the results are undefined.
2688 ``fptosi (CST to TYPE)``
2689 Convert a floating point constant to the corresponding signed
2690 integer constant. TYPE must be a scalar or vector integer type. CST
2691 must be of scalar or vector floating point type. Both CST and TYPE
2692 must be scalars, or vectors of the same number of elements. If the
2693 value won't fit in the integer type, the results are undefined.
2694 ``uitofp (CST to TYPE)``
2695 Convert an unsigned integer constant to the corresponding floating
2696 point constant. TYPE must be a scalar or vector floating point type.
2697 CST must be of scalar or vector integer type. Both CST and TYPE must
2698 be scalars, or vectors of the same number of elements. If the value
2699 won't fit in the floating point type, the results are undefined.
2700 ``sitofp (CST to TYPE)``
2701 Convert a signed integer constant to the corresponding floating
2702 point constant. TYPE must be a scalar or vector floating point type.
2703 CST must be of scalar or vector integer type. Both CST and TYPE must
2704 be scalars, or vectors of the same number of elements. If the value
2705 won't fit in the floating point type, the results are undefined.
2706 ``ptrtoint (CST to TYPE)``
2707 Convert a pointer typed constant to the corresponding integer
2708 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2709 pointer type. The ``CST`` value is zero extended, truncated, or
2710 unchanged to make it fit in ``TYPE``.
2711 ``inttoptr (CST to TYPE)``
2712 Convert an integer constant to a pointer constant. TYPE must be a
2713 pointer type. CST must be of integer type. The CST value is zero
2714 extended, truncated, or unchanged to make it fit in a pointer size.
2715 This one is *really* dangerous!
2716 ``bitcast (CST to TYPE)``
2717 Convert a constant, CST, to another TYPE. The constraints of the
2718 operands are the same as those for the :ref:`bitcast
2719 instruction <i_bitcast>`.
2720 ``addrspacecast (CST to TYPE)``
2721 Convert a constant pointer or constant vector of pointer, CST, to another
2722 TYPE in a different address space. The constraints of the operands are the
2723 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2724 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2725 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2726 constants. As with the :ref:`getelementptr <i_getelementptr>`
2727 instruction, the index list may have zero or more indexes, which are
2728 required to make sense for the type of "pointer to TY".
2729 ``select (COND, VAL1, VAL2)``
2730 Perform the :ref:`select operation <i_select>` on constants.
2731 ``icmp COND (VAL1, VAL2)``
2732 Performs the :ref:`icmp operation <i_icmp>` on constants.
2733 ``fcmp COND (VAL1, VAL2)``
2734 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2735 ``extractelement (VAL, IDX)``
2736 Perform the :ref:`extractelement operation <i_extractelement>` on
2738 ``insertelement (VAL, ELT, IDX)``
2739 Perform the :ref:`insertelement operation <i_insertelement>` on
2741 ``shufflevector (VEC1, VEC2, IDXMASK)``
2742 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2744 ``extractvalue (VAL, IDX0, IDX1, ...)``
2745 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2746 constants. The index list is interpreted in a similar manner as
2747 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2748 least one index value must be specified.
2749 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2750 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2751 The index list is interpreted in a similar manner as indices in a
2752 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2753 value must be specified.
2754 ``OPCODE (LHS, RHS)``
2755 Perform the specified operation of the LHS and RHS constants. OPCODE
2756 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2757 binary <bitwiseops>` operations. The constraints on operands are
2758 the same as those for the corresponding instruction (e.g. no bitwise
2759 operations on floating point values are allowed).
2766 Inline Assembler Expressions
2767 ----------------------------
2769 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2770 Inline Assembly <moduleasm>`) through the use of a special value. This
2771 value represents the inline assembler as a string (containing the
2772 instructions to emit), a list of operand constraints (stored as a
2773 string), a flag that indicates whether or not the inline asm expression
2774 has side effects, and a flag indicating whether the function containing
2775 the asm needs to align its stack conservatively. An example inline
2776 assembler expression is:
2778 .. code-block:: llvm
2780 i32 (i32) asm "bswap $0", "=r,r"
2782 Inline assembler expressions may **only** be used as the callee operand
2783 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2784 Thus, typically we have:
2786 .. code-block:: llvm
2788 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2790 Inline asms with side effects not visible in the constraint list must be
2791 marked as having side effects. This is done through the use of the
2792 '``sideeffect``' keyword, like so:
2794 .. code-block:: llvm
2796 call void asm sideeffect "eieio", ""()
2798 In some cases inline asms will contain code that will not work unless
2799 the stack is aligned in some way, such as calls or SSE instructions on
2800 x86, yet will not contain code that does that alignment within the asm.
2801 The compiler should make conservative assumptions about what the asm
2802 might contain and should generate its usual stack alignment code in the
2803 prologue if the '``alignstack``' keyword is present:
2805 .. code-block:: llvm
2807 call void asm alignstack "eieio", ""()
2809 Inline asms also support using non-standard assembly dialects. The
2810 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2811 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2812 the only supported dialects. An example is:
2814 .. code-block:: llvm
2816 call void asm inteldialect "eieio", ""()
2818 If multiple keywords appear the '``sideeffect``' keyword must come
2819 first, the '``alignstack``' keyword second and the '``inteldialect``'
2825 The call instructions that wrap inline asm nodes may have a
2826 "``!srcloc``" MDNode attached to it that contains a list of constant
2827 integers. If present, the code generator will use the integer as the
2828 location cookie value when report errors through the ``LLVMContext``
2829 error reporting mechanisms. This allows a front-end to correlate backend
2830 errors that occur with inline asm back to the source code that produced
2833 .. code-block:: llvm
2835 call void asm sideeffect "something bad", ""(), !srcloc !42
2837 !42 = !{ i32 1234567 }
2839 It is up to the front-end to make sense of the magic numbers it places
2840 in the IR. If the MDNode contains multiple constants, the code generator
2841 will use the one that corresponds to the line of the asm that the error
2849 LLVM IR allows metadata to be attached to instructions in the program
2850 that can convey extra information about the code to the optimizers and
2851 code generator. One example application of metadata is source-level
2852 debug information. There are two metadata primitives: strings and nodes.
2854 Metadata does not have a type, and is not a value. If referenced from a
2855 ``call`` instruction, it uses the ``metadata`` type.
2857 All metadata are identified in syntax by a exclamation point ('``!``').
2859 .. _metadata-string:
2861 Metadata Nodes and Metadata Strings
2862 -----------------------------------
2864 A metadata string is a string surrounded by double quotes. It can
2865 contain any character by escaping non-printable characters with
2866 "``\xx``" where "``xx``" is the two digit hex code. For example:
2869 Metadata nodes are represented with notation similar to structure
2870 constants (a comma separated list of elements, surrounded by braces and
2871 preceded by an exclamation point). Metadata nodes can have any values as
2872 their operand. For example:
2874 .. code-block:: llvm
2876 !{ !"test\00", i32 10}
2878 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
2880 .. code-block:: llvm
2882 !0 = distinct !{!"test\00", i32 10}
2884 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
2885 content. They can also occur when transformations cause uniquing collisions
2886 when metadata operands change.
2888 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2889 metadata nodes, which can be looked up in the module symbol table. For
2892 .. code-block:: llvm
2896 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2897 function is using two metadata arguments:
2899 .. code-block:: llvm
2901 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2903 Metadata can be attached with an instruction. Here metadata ``!21`` is
2904 attached to the ``add`` instruction using the ``!dbg`` identifier:
2906 .. code-block:: llvm
2908 %indvar.next = add i64 %indvar, 1, !dbg !21
2910 More information about specific metadata nodes recognized by the
2911 optimizers and code generator is found below.
2913 .. _specialized-metadata:
2915 Specialized Metadata Nodes
2916 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2918 Specialized metadata nodes are custom data structures in metadata (as opposed
2919 to generic tuples). Their fields are labelled, and can be specified in any
2922 These aren't inherently debug info centric, but currently all the specialized
2923 metadata nodes are related to debug info.
2930 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
2931 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
2932 tuples containing the debug info to be emitted along with the compile unit,
2933 regardless of code optimizations (some nodes are only emitted if there are
2934 references to them from instructions).
2936 .. code-block:: llvm
2938 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
2939 isOptimized: true, flags: "-O2", runtimeVersion: 2,
2940 splitDebugFilename: "abc.debug", emissionKind: 1,
2941 enums: !2, retainedTypes: !3, subprograms: !4,
2942 globals: !5, imports: !6)
2944 Compile unit descriptors provide the root scope for objects declared in a
2945 specific compilation unit. File descriptors are defined using this scope.
2946 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
2947 keep track of subprograms, global variables, type information, and imported
2948 entities (declarations and namespaces).
2955 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
2957 .. code-block:: llvm
2959 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
2961 Files are sometimes used in ``scope:`` fields, and are the only valid target
2962 for ``file:`` fields.
2969 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
2970 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
2972 .. code-block:: llvm
2974 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
2975 encoding: DW_ATE_unsigned_char)
2976 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
2978 The ``encoding:`` describes the details of the type. Usually it's one of the
2981 .. code-block:: llvm
2987 DW_ATE_signed_char = 6
2989 DW_ATE_unsigned_char = 8
2991 .. _DISubroutineType:
2996 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
2997 refers to a tuple; the first operand is the return type, while the rest are the
2998 types of the formal arguments in order. If the first operand is ``null``, that
2999 represents a function with no return value (such as ``void foo() {}`` in C++).
3001 .. code-block:: llvm
3003 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3004 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3005 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3012 ``DIDerivedType`` nodes represent types derived from other types, such as
3015 .. code-block:: llvm
3017 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3018 encoding: DW_ATE_unsigned_char)
3019 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3022 The following ``tag:`` values are valid:
3024 .. code-block:: llvm
3026 DW_TAG_formal_parameter = 5
3028 DW_TAG_pointer_type = 15
3029 DW_TAG_reference_type = 16
3031 DW_TAG_ptr_to_member_type = 31
3032 DW_TAG_const_type = 38
3033 DW_TAG_volatile_type = 53
3034 DW_TAG_restrict_type = 55
3036 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3037 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3038 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3039 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3040 argument of a subprogram.
3042 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3044 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3045 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3048 Note that the ``void *`` type is expressed as a type derived from NULL.
3050 .. _DICompositeType:
3055 ``DICompositeType`` nodes represent types composed of other types, like
3056 structures and unions. ``elements:`` points to a tuple of the composed types.
3058 If the source language supports ODR, the ``identifier:`` field gives the unique
3059 identifier used for type merging between modules. When specified, other types
3060 can refer to composite types indirectly via a :ref:`metadata string
3061 <metadata-string>` that matches their identifier.
3063 .. code-block:: llvm
3065 !0 = !DIEnumerator(name: "SixKind", value: 7)
3066 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3067 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3068 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3069 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3070 elements: !{!0, !1, !2})
3072 The following ``tag:`` values are valid:
3074 .. code-block:: llvm
3076 DW_TAG_array_type = 1
3077 DW_TAG_class_type = 2
3078 DW_TAG_enumeration_type = 4
3079 DW_TAG_structure_type = 19
3080 DW_TAG_union_type = 23
3081 DW_TAG_subroutine_type = 21
3082 DW_TAG_inheritance = 28
3085 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3086 descriptors <DISubrange>`, each representing the range of subscripts at that
3087 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3088 array type is a native packed vector.
3090 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3091 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3092 value for the set. All enumeration type descriptors are collected in the
3093 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3095 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3096 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3097 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3104 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3105 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3107 .. code-block:: llvm
3109 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3110 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3111 !2 = !DISubrange(count: -1) ; empty array.
3118 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3119 variants of :ref:`DICompositeType`.
3121 .. code-block:: llvm
3123 !0 = !DIEnumerator(name: "SixKind", value: 7)
3124 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3125 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3127 DITemplateTypeParameter
3128 """""""""""""""""""""""
3130 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3131 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3132 :ref:`DISubprogram` ``templateParams:`` fields.
3134 .. code-block:: llvm
3136 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3138 DITemplateValueParameter
3139 """"""""""""""""""""""""
3141 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3142 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3143 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3144 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3145 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3147 .. code-block:: llvm
3149 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3154 ``DINamespace`` nodes represent namespaces in the source language.
3156 .. code-block:: llvm
3158 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3163 ``DIGlobalVariable`` nodes represent global variables in the source language.
3165 .. code-block:: llvm
3167 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3168 file: !2, line: 7, type: !3, isLocal: true,
3169 isDefinition: false, variable: i32* @foo,
3172 All global variables should be referenced by the `globals:` field of a
3173 :ref:`compile unit <DICompileUnit>`.
3180 ``DISubprogram`` nodes represent functions from the source language. The
3181 ``variables:`` field points at :ref:`variables <DILocalVariable>` that must be
3182 retained, even if their IR counterparts are optimized out of the IR. The
3183 ``type:`` field must point at an :ref:`DISubroutineType`.
3185 .. code-block:: llvm
3187 !0 = !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3188 file: !2, line: 7, type: !3, isLocal: true,
3189 isDefinition: false, scopeLine: 8, containingType: !4,
3190 virtuality: DW_VIRTUALITY_pure_virtual, virtualIndex: 10,
3191 flags: DIFlagPrototyped, isOptimized: true,
3192 function: void ()* @_Z3foov,
3193 templateParams: !5, declaration: !6, variables: !7)
3200 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3201 <DISubprogram>`. The line number and column numbers are used to dinstinguish
3202 two lexical blocks at same depth. They are valid targets for ``scope:``
3205 .. code-block:: llvm
3207 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3209 Usually lexical blocks are ``distinct`` to prevent node merging based on
3212 .. _DILexicalBlockFile:
3217 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3218 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
3219 indicate textual inclusion, or the ``discriminator:`` field can be used to
3220 discriminate between control flow within a single block in the source language.
3222 .. code-block:: llvm
3224 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3225 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3226 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3233 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
3234 mandatory, and points at an :ref:`DILexicalBlockFile`, an
3235 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
3237 .. code-block:: llvm
3239 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3241 .. _DILocalVariable:
3246 ``DILocalVariable`` nodes represent local variables in the source language.
3247 Instead of ``DW_TAG_variable``, they use LLVM-specific fake tags to
3248 discriminate between local variables (``DW_TAG_auto_variable``) and subprogram
3249 arguments (``DW_TAG_arg_variable``). In the latter case, the ``arg:`` field
3250 specifies the argument position, and this variable will be included in the
3251 ``variables:`` field of its :ref:`DISubprogram`.
3253 .. code-block:: llvm
3255 !0 = !DILocalVariable(tag: DW_TAG_arg_variable, name: "this", arg: 0,
3256 scope: !3, file: !2, line: 7, type: !3,
3257 flags: DIFlagArtificial)
3258 !1 = !DILocalVariable(tag: DW_TAG_arg_variable, name: "x", arg: 1,
3259 scope: !4, file: !2, line: 7, type: !3)
3260 !1 = !DILocalVariable(tag: DW_TAG_auto_variable, name: "y",
3261 scope: !5, file: !2, line: 7, type: !3)
3266 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
3267 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
3268 describe how the referenced LLVM variable relates to the source language
3271 The current supported vocabulary is limited:
3273 - ``DW_OP_deref`` dereferences the working expression.
3274 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
3275 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
3276 here, respectively) of the variable piece from the working expression.
3278 .. code-block:: llvm
3280 !0 = !DIExpression(DW_OP_deref)
3281 !1 = !DIExpression(DW_OP_plus, 3)
3282 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
3283 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
3288 ``DIObjCProperty`` nodes represent Objective-C property nodes.
3290 .. code-block:: llvm
3292 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
3293 getter: "getFoo", attributes: 7, type: !2)
3298 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
3301 .. code-block:: llvm
3303 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
3304 entity: !1, line: 7)
3309 In LLVM IR, memory does not have types, so LLVM's own type system is not
3310 suitable for doing TBAA. Instead, metadata is added to the IR to
3311 describe a type system of a higher level language. This can be used to
3312 implement typical C/C++ TBAA, but it can also be used to implement
3313 custom alias analysis behavior for other languages.
3315 The current metadata format is very simple. TBAA metadata nodes have up
3316 to three fields, e.g.:
3318 .. code-block:: llvm
3320 !0 = !{ !"an example type tree" }
3321 !1 = !{ !"int", !0 }
3322 !2 = !{ !"float", !0 }
3323 !3 = !{ !"const float", !2, i64 1 }
3325 The first field is an identity field. It can be any value, usually a
3326 metadata string, which uniquely identifies the type. The most important
3327 name in the tree is the name of the root node. Two trees with different
3328 root node names are entirely disjoint, even if they have leaves with
3331 The second field identifies the type's parent node in the tree, or is
3332 null or omitted for a root node. A type is considered to alias all of
3333 its descendants and all of its ancestors in the tree. Also, a type is
3334 considered to alias all types in other trees, so that bitcode produced
3335 from multiple front-ends is handled conservatively.
3337 If the third field is present, it's an integer which if equal to 1
3338 indicates that the type is "constant" (meaning
3339 ``pointsToConstantMemory`` should return true; see `other useful
3340 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
3342 '``tbaa.struct``' Metadata
3343 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3345 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
3346 aggregate assignment operations in C and similar languages, however it
3347 is defined to copy a contiguous region of memory, which is more than
3348 strictly necessary for aggregate types which contain holes due to
3349 padding. Also, it doesn't contain any TBAA information about the fields
3352 ``!tbaa.struct`` metadata can describe which memory subregions in a
3353 memcpy are padding and what the TBAA tags of the struct are.
3355 The current metadata format is very simple. ``!tbaa.struct`` metadata
3356 nodes are a list of operands which are in conceptual groups of three.
3357 For each group of three, the first operand gives the byte offset of a
3358 field in bytes, the second gives its size in bytes, and the third gives
3361 .. code-block:: llvm
3363 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
3365 This describes a struct with two fields. The first is at offset 0 bytes
3366 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
3367 and has size 4 bytes and has tbaa tag !2.
3369 Note that the fields need not be contiguous. In this example, there is a
3370 4 byte gap between the two fields. This gap represents padding which
3371 does not carry useful data and need not be preserved.
3373 '``noalias``' and '``alias.scope``' Metadata
3374 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3376 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
3377 noalias memory-access sets. This means that some collection of memory access
3378 instructions (loads, stores, memory-accessing calls, etc.) that carry
3379 ``noalias`` metadata can specifically be specified not to alias with some other
3380 collection of memory access instructions that carry ``alias.scope`` metadata.
3381 Each type of metadata specifies a list of scopes where each scope has an id and
3382 a domain. When evaluating an aliasing query, if for some domain, the set
3383 of scopes with that domain in one instruction's ``alias.scope`` list is a
3384 subset of (or equal to) the set of scopes for that domain in another
3385 instruction's ``noalias`` list, then the two memory accesses are assumed not to
3388 The metadata identifying each domain is itself a list containing one or two
3389 entries. The first entry is the name of the domain. Note that if the name is a
3390 string then it can be combined accross functions and translation units. A
3391 self-reference can be used to create globally unique domain names. A
3392 descriptive string may optionally be provided as a second list entry.
3394 The metadata identifying each scope is also itself a list containing two or
3395 three entries. The first entry is the name of the scope. Note that if the name
3396 is a string then it can be combined accross functions and translation units. A
3397 self-reference can be used to create globally unique scope names. A metadata
3398 reference to the scope's domain is the second entry. A descriptive string may
3399 optionally be provided as a third list entry.
3403 .. code-block:: llvm
3405 ; Two scope domains:
3409 ; Some scopes in these domains:
3415 !5 = !{!4} ; A list containing only scope !4
3419 ; These two instructions don't alias:
3420 %0 = load float, float* %c, align 4, !alias.scope !5
3421 store float %0, float* %arrayidx.i, align 4, !noalias !5
3423 ; These two instructions also don't alias (for domain !1, the set of scopes
3424 ; in the !alias.scope equals that in the !noalias list):
3425 %2 = load float, float* %c, align 4, !alias.scope !5
3426 store float %2, float* %arrayidx.i2, align 4, !noalias !6
3428 ; These two instructions may alias (for domain !0, the set of scopes in
3429 ; the !noalias list is not a superset of, or equal to, the scopes in the
3430 ; !alias.scope list):
3431 %2 = load float, float* %c, align 4, !alias.scope !6
3432 store float %0, float* %arrayidx.i, align 4, !noalias !7
3434 '``fpmath``' Metadata
3435 ^^^^^^^^^^^^^^^^^^^^^
3437 ``fpmath`` metadata may be attached to any instruction of floating point
3438 type. It can be used to express the maximum acceptable error in the
3439 result of that instruction, in ULPs, thus potentially allowing the
3440 compiler to use a more efficient but less accurate method of computing
3441 it. ULP is defined as follows:
3443 If ``x`` is a real number that lies between two finite consecutive
3444 floating-point numbers ``a`` and ``b``, without being equal to one
3445 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
3446 distance between the two non-equal finite floating-point numbers
3447 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
3449 The metadata node shall consist of a single positive floating point
3450 number representing the maximum relative error, for example:
3452 .. code-block:: llvm
3454 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3458 '``range``' Metadata
3459 ^^^^^^^^^^^^^^^^^^^^
3461 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
3462 integer types. It expresses the possible ranges the loaded value or the value
3463 returned by the called function at this call site is in. The ranges are
3464 represented with a flattened list of integers. The loaded value or the value
3465 returned is known to be in the union of the ranges defined by each consecutive
3466 pair. Each pair has the following properties:
3468 - The type must match the type loaded by the instruction.
3469 - The pair ``a,b`` represents the range ``[a,b)``.
3470 - Both ``a`` and ``b`` are constants.
3471 - The range is allowed to wrap.
3472 - The range should not represent the full or empty set. That is,
3475 In addition, the pairs must be in signed order of the lower bound and
3476 they must be non-contiguous.
3480 .. code-block:: llvm
3482 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
3483 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3484 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
3485 %d = invoke i8 @bar() to label %cont
3486 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
3488 !0 = !{ i8 0, i8 2 }
3489 !1 = !{ i8 255, i8 2 }
3490 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
3491 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
3496 It is sometimes useful to attach information to loop constructs. Currently,
3497 loop metadata is implemented as metadata attached to the branch instruction
3498 in the loop latch block. This type of metadata refer to a metadata node that is
3499 guaranteed to be separate for each loop. The loop identifier metadata is
3500 specified with the name ``llvm.loop``.
3502 The loop identifier metadata is implemented using a metadata that refers to
3503 itself to avoid merging it with any other identifier metadata, e.g.,
3504 during module linkage or function inlining. That is, each loop should refer
3505 to their own identification metadata even if they reside in separate functions.
3506 The following example contains loop identifier metadata for two separate loop
3509 .. code-block:: llvm
3514 The loop identifier metadata can be used to specify additional
3515 per-loop metadata. Any operands after the first operand can be treated
3516 as user-defined metadata. For example the ``llvm.loop.unroll.count``
3517 suggests an unroll factor to the loop unroller:
3519 .. code-block:: llvm
3521 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
3524 !1 = !{!"llvm.loop.unroll.count", i32 4}
3526 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
3527 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3529 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
3530 used to control per-loop vectorization and interleaving parameters such as
3531 vectorization width and interleave count. These metadata should be used in
3532 conjunction with ``llvm.loop`` loop identification metadata. The
3533 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
3534 optimization hints and the optimizer will only interleave and vectorize loops if
3535 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
3536 which contains information about loop-carried memory dependencies can be helpful
3537 in determining the safety of these transformations.
3539 '``llvm.loop.interleave.count``' Metadata
3540 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3542 This metadata suggests an interleave count to the loop interleaver.
3543 The first operand is the string ``llvm.loop.interleave.count`` and the
3544 second operand is an integer specifying the interleave count. For
3547 .. code-block:: llvm
3549 !0 = !{!"llvm.loop.interleave.count", i32 4}
3551 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
3552 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
3553 then the interleave count will be determined automatically.
3555 '``llvm.loop.vectorize.enable``' Metadata
3556 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3558 This metadata selectively enables or disables vectorization for the loop. The
3559 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
3560 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
3561 0 disables vectorization:
3563 .. code-block:: llvm
3565 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
3566 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
3568 '``llvm.loop.vectorize.width``' Metadata
3569 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3571 This metadata sets the target width of the vectorizer. The first
3572 operand is the string ``llvm.loop.vectorize.width`` and the second
3573 operand is an integer specifying the width. For example:
3575 .. code-block:: llvm
3577 !0 = !{!"llvm.loop.vectorize.width", i32 4}
3579 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
3580 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
3581 0 or if the loop does not have this metadata the width will be
3582 determined automatically.
3584 '``llvm.loop.unroll``'
3585 ^^^^^^^^^^^^^^^^^^^^^^
3587 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
3588 optimization hints such as the unroll factor. ``llvm.loop.unroll``
3589 metadata should be used in conjunction with ``llvm.loop`` loop
3590 identification metadata. The ``llvm.loop.unroll`` metadata are only
3591 optimization hints and the unrolling will only be performed if the
3592 optimizer believes it is safe to do so.
3594 '``llvm.loop.unroll.count``' Metadata
3595 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3597 This metadata suggests an unroll factor to the loop unroller. The
3598 first operand is the string ``llvm.loop.unroll.count`` and the second
3599 operand is a positive integer specifying the unroll factor. For
3602 .. code-block:: llvm
3604 !0 = !{!"llvm.loop.unroll.count", i32 4}
3606 If the trip count of the loop is less than the unroll count the loop
3607 will be partially unrolled.
3609 '``llvm.loop.unroll.disable``' Metadata
3610 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3612 This metadata either disables loop unrolling. The metadata has a single operand
3613 which is the string ``llvm.loop.unroll.disable``. For example:
3615 .. code-block:: llvm
3617 !0 = !{!"llvm.loop.unroll.disable"}
3619 '``llvm.loop.unroll.runtime.disable``' Metadata
3620 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3622 This metadata either disables runtime loop unrolling. The metadata has a single
3623 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
3625 .. code-block:: llvm
3627 !0 = !{!"llvm.loop.unroll.runtime.disable"}
3629 '``llvm.loop.unroll.full``' Metadata
3630 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3632 This metadata either suggests that the loop should be unrolled fully. The
3633 metadata has a single operand which is the string ``llvm.loop.unroll.disable``.
3636 .. code-block:: llvm
3638 !0 = !{!"llvm.loop.unroll.full"}
3643 Metadata types used to annotate memory accesses with information helpful
3644 for optimizations are prefixed with ``llvm.mem``.
3646 '``llvm.mem.parallel_loop_access``' Metadata
3647 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3649 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
3650 or metadata containing a list of loop identifiers for nested loops.
3651 The metadata is attached to memory accessing instructions and denotes that
3652 no loop carried memory dependence exist between it and other instructions denoted
3653 with the same loop identifier.
3655 Precisely, given two instructions ``m1`` and ``m2`` that both have the
3656 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
3657 set of loops associated with that metadata, respectively, then there is no loop
3658 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
3661 As a special case, if all memory accessing instructions in a loop have
3662 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
3663 loop has no loop carried memory dependences and is considered to be a parallel
3666 Note that if not all memory access instructions have such metadata referring to
3667 the loop, then the loop is considered not being trivially parallel. Additional
3668 memory dependence analysis is required to make that determination. As a fail
3669 safe mechanism, this causes loops that were originally parallel to be considered
3670 sequential (if optimization passes that are unaware of the parallel semantics
3671 insert new memory instructions into the loop body).
3673 Example of a loop that is considered parallel due to its correct use of
3674 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
3675 metadata types that refer to the same loop identifier metadata.
3677 .. code-block:: llvm
3681 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
3683 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3685 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
3691 It is also possible to have nested parallel loops. In that case the
3692 memory accesses refer to a list of loop identifier metadata nodes instead of
3693 the loop identifier metadata node directly:
3695 .. code-block:: llvm
3699 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
3701 br label %inner.for.body
3705 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3707 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
3709 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
3713 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
3715 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
3717 outer.for.end: ; preds = %for.body
3719 !0 = !{!1, !2} ; a list of loop identifiers
3720 !1 = !{!1} ; an identifier for the inner loop
3721 !2 = !{!2} ; an identifier for the outer loop
3726 The ``llvm.bitsets`` global metadata is used to implement
3727 :doc:`bitsets <BitSets>`.
3729 Module Flags Metadata
3730 =====================
3732 Information about the module as a whole is difficult to convey to LLVM's
3733 subsystems. The LLVM IR isn't sufficient to transmit this information.
3734 The ``llvm.module.flags`` named metadata exists in order to facilitate
3735 this. These flags are in the form of key / value pairs --- much like a
3736 dictionary --- making it easy for any subsystem who cares about a flag to
3739 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
3740 Each triplet has the following form:
3742 - The first element is a *behavior* flag, which specifies the behavior
3743 when two (or more) modules are merged together, and it encounters two
3744 (or more) metadata with the same ID. The supported behaviors are
3746 - The second element is a metadata string that is a unique ID for the
3747 metadata. Each module may only have one flag entry for each unique ID (not
3748 including entries with the **Require** behavior).
3749 - The third element is the value of the flag.
3751 When two (or more) modules are merged together, the resulting
3752 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
3753 each unique metadata ID string, there will be exactly one entry in the merged
3754 modules ``llvm.module.flags`` metadata table, and the value for that entry will
3755 be determined by the merge behavior flag, as described below. The only exception
3756 is that entries with the *Require* behavior are always preserved.
3758 The following behaviors are supported:
3769 Emits an error if two values disagree, otherwise the resulting value
3770 is that of the operands.
3774 Emits a warning if two values disagree. The result value will be the
3775 operand for the flag from the first module being linked.
3779 Adds a requirement that another module flag be present and have a
3780 specified value after linking is performed. The value must be a
3781 metadata pair, where the first element of the pair is the ID of the
3782 module flag to be restricted, and the second element of the pair is
3783 the value the module flag should be restricted to. This behavior can
3784 be used to restrict the allowable results (via triggering of an
3785 error) of linking IDs with the **Override** behavior.
3789 Uses the specified value, regardless of the behavior or value of the
3790 other module. If both modules specify **Override**, but the values
3791 differ, an error will be emitted.
3795 Appends the two values, which are required to be metadata nodes.
3799 Appends the two values, which are required to be metadata
3800 nodes. However, duplicate entries in the second list are dropped
3801 during the append operation.
3803 It is an error for a particular unique flag ID to have multiple behaviors,
3804 except in the case of **Require** (which adds restrictions on another metadata
3805 value) or **Override**.
3807 An example of module flags:
3809 .. code-block:: llvm
3811 !0 = !{ i32 1, !"foo", i32 1 }
3812 !1 = !{ i32 4, !"bar", i32 37 }
3813 !2 = !{ i32 2, !"qux", i32 42 }
3814 !3 = !{ i32 3, !"qux",
3819 !llvm.module.flags = !{ !0, !1, !2, !3 }
3821 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3822 if two or more ``!"foo"`` flags are seen is to emit an error if their
3823 values are not equal.
3825 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3826 behavior if two or more ``!"bar"`` flags are seen is to use the value
3829 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3830 behavior if two or more ``!"qux"`` flags are seen is to emit a
3831 warning if their values are not equal.
3833 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3839 The behavior is to emit an error if the ``llvm.module.flags`` does not
3840 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3843 Objective-C Garbage Collection Module Flags Metadata
3844 ----------------------------------------------------
3846 On the Mach-O platform, Objective-C stores metadata about garbage
3847 collection in a special section called "image info". The metadata
3848 consists of a version number and a bitmask specifying what types of
3849 garbage collection are supported (if any) by the file. If two or more
3850 modules are linked together their garbage collection metadata needs to
3851 be merged rather than appended together.
3853 The Objective-C garbage collection module flags metadata consists of the
3854 following key-value pairs:
3863 * - ``Objective-C Version``
3864 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3866 * - ``Objective-C Image Info Version``
3867 - **[Required]** --- The version of the image info section. Currently
3870 * - ``Objective-C Image Info Section``
3871 - **[Required]** --- The section to place the metadata. Valid values are
3872 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3873 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3874 Objective-C ABI version 2.
3876 * - ``Objective-C Garbage Collection``
3877 - **[Required]** --- Specifies whether garbage collection is supported or
3878 not. Valid values are 0, for no garbage collection, and 2, for garbage
3879 collection supported.
3881 * - ``Objective-C GC Only``
3882 - **[Optional]** --- Specifies that only garbage collection is supported.
3883 If present, its value must be 6. This flag requires that the
3884 ``Objective-C Garbage Collection`` flag have the value 2.
3886 Some important flag interactions:
3888 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3889 merged with a module with ``Objective-C Garbage Collection`` set to
3890 2, then the resulting module has the
3891 ``Objective-C Garbage Collection`` flag set to 0.
3892 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3893 merged with a module with ``Objective-C GC Only`` set to 6.
3895 Automatic Linker Flags Module Flags Metadata
3896 --------------------------------------------
3898 Some targets support embedding flags to the linker inside individual object
3899 files. Typically this is used in conjunction with language extensions which
3900 allow source files to explicitly declare the libraries they depend on, and have
3901 these automatically be transmitted to the linker via object files.
3903 These flags are encoded in the IR using metadata in the module flags section,
3904 using the ``Linker Options`` key. The merge behavior for this flag is required
3905 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3906 node which should be a list of other metadata nodes, each of which should be a
3907 list of metadata strings defining linker options.
3909 For example, the following metadata section specifies two separate sets of
3910 linker options, presumably to link against ``libz`` and the ``Cocoa``
3913 !0 = !{ i32 6, !"Linker Options",
3916 !{ !"-framework", !"Cocoa" } } }
3917 !llvm.module.flags = !{ !0 }
3919 The metadata encoding as lists of lists of options, as opposed to a collapsed
3920 list of options, is chosen so that the IR encoding can use multiple option
3921 strings to specify e.g., a single library, while still having that specifier be
3922 preserved as an atomic element that can be recognized by a target specific
3923 assembly writer or object file emitter.
3925 Each individual option is required to be either a valid option for the target's
3926 linker, or an option that is reserved by the target specific assembly writer or
3927 object file emitter. No other aspect of these options is defined by the IR.
3929 C type width Module Flags Metadata
3930 ----------------------------------
3932 The ARM backend emits a section into each generated object file describing the
3933 options that it was compiled with (in a compiler-independent way) to prevent
3934 linking incompatible objects, and to allow automatic library selection. Some
3935 of these options are not visible at the IR level, namely wchar_t width and enum
3938 To pass this information to the backend, these options are encoded in module
3939 flags metadata, using the following key-value pairs:
3949 - * 0 --- sizeof(wchar_t) == 4
3950 * 1 --- sizeof(wchar_t) == 2
3953 - * 0 --- Enums are at least as large as an ``int``.
3954 * 1 --- Enums are stored in the smallest integer type which can
3955 represent all of its values.
3957 For example, the following metadata section specifies that the module was
3958 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
3959 enum is the smallest type which can represent all of its values::
3961 !llvm.module.flags = !{!0, !1}
3962 !0 = !{i32 1, !"short_wchar", i32 1}
3963 !1 = !{i32 1, !"short_enum", i32 0}
3965 .. _intrinsicglobalvariables:
3967 Intrinsic Global Variables
3968 ==========================
3970 LLVM has a number of "magic" global variables that contain data that
3971 affect code generation or other IR semantics. These are documented here.
3972 All globals of this sort should have a section specified as
3973 "``llvm.metadata``". This section and all globals that start with
3974 "``llvm.``" are reserved for use by LLVM.
3978 The '``llvm.used``' Global Variable
3979 -----------------------------------
3981 The ``@llvm.used`` global is an array which has
3982 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3983 pointers to named global variables, functions and aliases which may optionally
3984 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3987 .. code-block:: llvm
3992 @llvm.used = appending global [2 x i8*] [
3994 i8* bitcast (i32* @Y to i8*)
3995 ], section "llvm.metadata"
3997 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3998 and linker are required to treat the symbol as if there is a reference to the
3999 symbol that it cannot see (which is why they have to be named). For example, if
4000 a variable has internal linkage and no references other than that from the
4001 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4002 references from inline asms and other things the compiler cannot "see", and
4003 corresponds to "``attribute((used))``" in GNU C.
4005 On some targets, the code generator must emit a directive to the
4006 assembler or object file to prevent the assembler and linker from
4007 molesting the symbol.
4009 .. _gv_llvmcompilerused:
4011 The '``llvm.compiler.used``' Global Variable
4012 --------------------------------------------
4014 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4015 directive, except that it only prevents the compiler from touching the
4016 symbol. On targets that support it, this allows an intelligent linker to
4017 optimize references to the symbol without being impeded as it would be
4020 This is a rare construct that should only be used in rare circumstances,
4021 and should not be exposed to source languages.
4023 .. _gv_llvmglobalctors:
4025 The '``llvm.global_ctors``' Global Variable
4026 -------------------------------------------
4028 .. code-block:: llvm
4030 %0 = type { i32, void ()*, i8* }
4031 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4033 The ``@llvm.global_ctors`` array contains a list of constructor
4034 functions, priorities, and an optional associated global or function.
4035 The functions referenced by this array will be called in ascending order
4036 of priority (i.e. lowest first) when the module is loaded. The order of
4037 functions with the same priority is not defined.
4039 If the third field is present, non-null, and points to a global variable
4040 or function, the initializer function will only run if the associated
4041 data from the current module is not discarded.
4043 .. _llvmglobaldtors:
4045 The '``llvm.global_dtors``' Global Variable
4046 -------------------------------------------
4048 .. code-block:: llvm
4050 %0 = type { i32, void ()*, i8* }
4051 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4053 The ``@llvm.global_dtors`` array contains a list of destructor
4054 functions, priorities, and an optional associated global or function.
4055 The functions referenced by this array will be called in descending
4056 order of priority (i.e. highest first) when the module is unloaded. The
4057 order of functions with the same priority is not defined.
4059 If the third field is present, non-null, and points to a global variable
4060 or function, the destructor function will only run if the associated
4061 data from the current module is not discarded.
4063 Instruction Reference
4064 =====================
4066 The LLVM instruction set consists of several different classifications
4067 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4068 instructions <binaryops>`, :ref:`bitwise binary
4069 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4070 :ref:`other instructions <otherops>`.
4074 Terminator Instructions
4075 -----------------------
4077 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4078 program ends with a "Terminator" instruction, which indicates which
4079 block should be executed after the current block is finished. These
4080 terminator instructions typically yield a '``void``' value: they produce
4081 control flow, not values (the one exception being the
4082 ':ref:`invoke <i_invoke>`' instruction).
4084 The terminator instructions are: ':ref:`ret <i_ret>`',
4085 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4086 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4087 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
4091 '``ret``' Instruction
4092 ^^^^^^^^^^^^^^^^^^^^^
4099 ret <type> <value> ; Return a value from a non-void function
4100 ret void ; Return from void function
4105 The '``ret``' instruction is used to return control flow (and optionally
4106 a value) from a function back to the caller.
4108 There are two forms of the '``ret``' instruction: one that returns a
4109 value and then causes control flow, and one that just causes control
4115 The '``ret``' instruction optionally accepts a single argument, the
4116 return value. The type of the return value must be a ':ref:`first
4117 class <t_firstclass>`' type.
4119 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4120 return type and contains a '``ret``' instruction with no return value or
4121 a return value with a type that does not match its type, or if it has a
4122 void return type and contains a '``ret``' instruction with a return
4128 When the '``ret``' instruction is executed, control flow returns back to
4129 the calling function's context. If the caller is a
4130 ":ref:`call <i_call>`" instruction, execution continues at the
4131 instruction after the call. If the caller was an
4132 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4133 beginning of the "normal" destination block. If the instruction returns
4134 a value, that value shall set the call or invoke instruction's return
4140 .. code-block:: llvm
4142 ret i32 5 ; Return an integer value of 5
4143 ret void ; Return from a void function
4144 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4148 '``br``' Instruction
4149 ^^^^^^^^^^^^^^^^^^^^
4156 br i1 <cond>, label <iftrue>, label <iffalse>
4157 br label <dest> ; Unconditional branch
4162 The '``br``' instruction is used to cause control flow to transfer to a
4163 different basic block in the current function. There are two forms of
4164 this instruction, corresponding to a conditional branch and an
4165 unconditional branch.
4170 The conditional branch form of the '``br``' instruction takes a single
4171 '``i1``' value and two '``label``' values. The unconditional form of the
4172 '``br``' instruction takes a single '``label``' value as a target.
4177 Upon execution of a conditional '``br``' instruction, the '``i1``'
4178 argument is evaluated. If the value is ``true``, control flows to the
4179 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4180 to the '``iffalse``' ``label`` argument.
4185 .. code-block:: llvm
4188 %cond = icmp eq i32 %a, %b
4189 br i1 %cond, label %IfEqual, label %IfUnequal
4197 '``switch``' Instruction
4198 ^^^^^^^^^^^^^^^^^^^^^^^^
4205 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
4210 The '``switch``' instruction is used to transfer control flow to one of
4211 several different places. It is a generalization of the '``br``'
4212 instruction, allowing a branch to occur to one of many possible
4218 The '``switch``' instruction uses three parameters: an integer
4219 comparison value '``value``', a default '``label``' destination, and an
4220 array of pairs of comparison value constants and '``label``'s. The table
4221 is not allowed to contain duplicate constant entries.
4226 The ``switch`` instruction specifies a table of values and destinations.
4227 When the '``switch``' instruction is executed, this table is searched
4228 for the given value. If the value is found, control flow is transferred
4229 to the corresponding destination; otherwise, control flow is transferred
4230 to the default destination.
4235 Depending on properties of the target machine and the particular
4236 ``switch`` instruction, this instruction may be code generated in
4237 different ways. For example, it could be generated as a series of
4238 chained conditional branches or with a lookup table.
4243 .. code-block:: llvm
4245 ; Emulate a conditional br instruction
4246 %Val = zext i1 %value to i32
4247 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
4249 ; Emulate an unconditional br instruction
4250 switch i32 0, label %dest [ ]
4252 ; Implement a jump table:
4253 switch i32 %val, label %otherwise [ i32 0, label %onzero
4255 i32 2, label %ontwo ]
4259 '``indirectbr``' Instruction
4260 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4267 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
4272 The '``indirectbr``' instruction implements an indirect branch to a
4273 label within the current function, whose address is specified by
4274 "``address``". Address must be derived from a
4275 :ref:`blockaddress <blockaddress>` constant.
4280 The '``address``' argument is the address of the label to jump to. The
4281 rest of the arguments indicate the full set of possible destinations
4282 that the address may point to. Blocks are allowed to occur multiple
4283 times in the destination list, though this isn't particularly useful.
4285 This destination list is required so that dataflow analysis has an
4286 accurate understanding of the CFG.
4291 Control transfers to the block specified in the address argument. All
4292 possible destination blocks must be listed in the label list, otherwise
4293 this instruction has undefined behavior. This implies that jumps to
4294 labels defined in other functions have undefined behavior as well.
4299 This is typically implemented with a jump through a register.
4304 .. code-block:: llvm
4306 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
4310 '``invoke``' Instruction
4311 ^^^^^^^^^^^^^^^^^^^^^^^^
4318 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
4319 to label <normal label> unwind label <exception label>
4324 The '``invoke``' instruction causes control to transfer to a specified
4325 function, with the possibility of control flow transfer to either the
4326 '``normal``' label or the '``exception``' label. If the callee function
4327 returns with the "``ret``" instruction, control flow will return to the
4328 "normal" label. If the callee (or any indirect callees) returns via the
4329 ":ref:`resume <i_resume>`" instruction or other exception handling
4330 mechanism, control is interrupted and continued at the dynamically
4331 nearest "exception" label.
4333 The '``exception``' label is a `landing
4334 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
4335 '``exception``' label is required to have the
4336 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
4337 information about the behavior of the program after unwinding happens,
4338 as its first non-PHI instruction. The restrictions on the
4339 "``landingpad``" instruction's tightly couples it to the "``invoke``"
4340 instruction, so that the important information contained within the
4341 "``landingpad``" instruction can't be lost through normal code motion.
4346 This instruction requires several arguments:
4348 #. The optional "cconv" marker indicates which :ref:`calling
4349 convention <callingconv>` the call should use. If none is
4350 specified, the call defaults to using C calling conventions.
4351 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
4352 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
4354 #. '``ptr to function ty``': shall be the signature of the pointer to
4355 function value being invoked. In most cases, this is a direct
4356 function invocation, but indirect ``invoke``'s are just as possible,
4357 branching off an arbitrary pointer to function value.
4358 #. '``function ptr val``': An LLVM value containing a pointer to a
4359 function to be invoked.
4360 #. '``function args``': argument list whose types match the function
4361 signature argument types and parameter attributes. All arguments must
4362 be of :ref:`first class <t_firstclass>` type. If the function signature
4363 indicates the function accepts a variable number of arguments, the
4364 extra arguments can be specified.
4365 #. '``normal label``': the label reached when the called function
4366 executes a '``ret``' instruction.
4367 #. '``exception label``': the label reached when a callee returns via
4368 the :ref:`resume <i_resume>` instruction or other exception handling
4370 #. The optional :ref:`function attributes <fnattrs>` list. Only
4371 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
4372 attributes are valid here.
4377 This instruction is designed to operate as a standard '``call``'
4378 instruction in most regards. The primary difference is that it
4379 establishes an association with a label, which is used by the runtime
4380 library to unwind the stack.
4382 This instruction is used in languages with destructors to ensure that
4383 proper cleanup is performed in the case of either a ``longjmp`` or a
4384 thrown exception. Additionally, this is important for implementation of
4385 '``catch``' clauses in high-level languages that support them.
4387 For the purposes of the SSA form, the definition of the value returned
4388 by the '``invoke``' instruction is deemed to occur on the edge from the
4389 current block to the "normal" label. If the callee unwinds then no
4390 return value is available.
4395 .. code-block:: llvm
4397 %retval = invoke i32 @Test(i32 15) to label %Continue
4398 unwind label %TestCleanup ; i32:retval set
4399 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
4400 unwind label %TestCleanup ; i32:retval set
4404 '``resume``' Instruction
4405 ^^^^^^^^^^^^^^^^^^^^^^^^
4412 resume <type> <value>
4417 The '``resume``' instruction is a terminator instruction that has no
4423 The '``resume``' instruction requires one argument, which must have the
4424 same type as the result of any '``landingpad``' instruction in the same
4430 The '``resume``' instruction resumes propagation of an existing
4431 (in-flight) exception whose unwinding was interrupted with a
4432 :ref:`landingpad <i_landingpad>` instruction.
4437 .. code-block:: llvm
4439 resume { i8*, i32 } %exn
4443 '``unreachable``' Instruction
4444 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4456 The '``unreachable``' instruction has no defined semantics. This
4457 instruction is used to inform the optimizer that a particular portion of
4458 the code is not reachable. This can be used to indicate that the code
4459 after a no-return function cannot be reached, and other facts.
4464 The '``unreachable``' instruction has no defined semantics.
4471 Binary operators are used to do most of the computation in a program.
4472 They require two operands of the same type, execute an operation on
4473 them, and produce a single value. The operands might represent multiple
4474 data, as is the case with the :ref:`vector <t_vector>` data type. The
4475 result value has the same type as its operands.
4477 There are several different binary operators:
4481 '``add``' Instruction
4482 ^^^^^^^^^^^^^^^^^^^^^
4489 <result> = add <ty> <op1>, <op2> ; yields ty:result
4490 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
4491 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
4492 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
4497 The '``add``' instruction returns the sum of its two operands.
4502 The two arguments to the '``add``' instruction must be
4503 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4504 arguments must have identical types.
4509 The value produced is the integer sum of the two operands.
4511 If the sum has unsigned overflow, the result returned is the
4512 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4515 Because LLVM integers use a two's complement representation, this
4516 instruction is appropriate for both signed and unsigned integers.
4518 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4519 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4520 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
4521 unsigned and/or signed overflow, respectively, occurs.
4526 .. code-block:: llvm
4528 <result> = add i32 4, %var ; yields i32:result = 4 + %var
4532 '``fadd``' Instruction
4533 ^^^^^^^^^^^^^^^^^^^^^^
4540 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4545 The '``fadd``' instruction returns the sum of its two operands.
4550 The two arguments to the '``fadd``' instruction must be :ref:`floating
4551 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4552 Both arguments must have identical types.
4557 The value produced is the floating point sum of the two operands. This
4558 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
4559 which are optimization hints to enable otherwise unsafe floating point
4565 .. code-block:: llvm
4567 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
4569 '``sub``' Instruction
4570 ^^^^^^^^^^^^^^^^^^^^^
4577 <result> = sub <ty> <op1>, <op2> ; yields ty:result
4578 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
4579 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
4580 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
4585 The '``sub``' instruction returns the difference of its two operands.
4587 Note that the '``sub``' instruction is used to represent the '``neg``'
4588 instruction present in most other intermediate representations.
4593 The two arguments to the '``sub``' instruction must be
4594 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4595 arguments must have identical types.
4600 The value produced is the integer difference of the two operands.
4602 If the difference has unsigned overflow, the result returned is the
4603 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4606 Because LLVM integers use a two's complement representation, this
4607 instruction is appropriate for both signed and unsigned integers.
4609 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4610 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4611 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
4612 unsigned and/or signed overflow, respectively, occurs.
4617 .. code-block:: llvm
4619 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
4620 <result> = sub i32 0, %val ; yields i32:result = -%var
4624 '``fsub``' Instruction
4625 ^^^^^^^^^^^^^^^^^^^^^^
4632 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4637 The '``fsub``' instruction returns the difference of its two operands.
4639 Note that the '``fsub``' instruction is used to represent the '``fneg``'
4640 instruction present in most other intermediate representations.
4645 The two arguments to the '``fsub``' instruction must be :ref:`floating
4646 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4647 Both arguments must have identical types.
4652 The value produced is the floating point difference of the two operands.
4653 This instruction can also take any number of :ref:`fast-math
4654 flags <fastmath>`, which are optimization hints to enable otherwise
4655 unsafe floating point optimizations:
4660 .. code-block:: llvm
4662 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
4663 <result> = fsub float -0.0, %val ; yields float:result = -%var
4665 '``mul``' Instruction
4666 ^^^^^^^^^^^^^^^^^^^^^
4673 <result> = mul <ty> <op1>, <op2> ; yields ty:result
4674 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
4675 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
4676 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
4681 The '``mul``' instruction returns the product of its two operands.
4686 The two arguments to the '``mul``' instruction must be
4687 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4688 arguments must have identical types.
4693 The value produced is the integer product of the two operands.
4695 If the result of the multiplication has unsigned overflow, the result
4696 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
4697 bit width of the result.
4699 Because LLVM integers use a two's complement representation, and the
4700 result is the same width as the operands, this instruction returns the
4701 correct result for both signed and unsigned integers. If a full product
4702 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
4703 sign-extended or zero-extended as appropriate to the width of the full
4706 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4707 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4708 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
4709 unsigned and/or signed overflow, respectively, occurs.
4714 .. code-block:: llvm
4716 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
4720 '``fmul``' Instruction
4721 ^^^^^^^^^^^^^^^^^^^^^^
4728 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4733 The '``fmul``' instruction returns the product of its two operands.
4738 The two arguments to the '``fmul``' instruction must be :ref:`floating
4739 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4740 Both arguments must have identical types.
4745 The value produced is the floating point product of the two operands.
4746 This instruction can also take any number of :ref:`fast-math
4747 flags <fastmath>`, which are optimization hints to enable otherwise
4748 unsafe floating point optimizations:
4753 .. code-block:: llvm
4755 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
4757 '``udiv``' Instruction
4758 ^^^^^^^^^^^^^^^^^^^^^^
4765 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
4766 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
4771 The '``udiv``' instruction returns the quotient of its two operands.
4776 The two arguments to the '``udiv``' instruction must be
4777 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4778 arguments must have identical types.
4783 The value produced is the unsigned integer quotient of the two operands.
4785 Note that unsigned integer division and signed integer division are
4786 distinct operations; for signed integer division, use '``sdiv``'.
4788 Division by zero leads to undefined behavior.
4790 If the ``exact`` keyword is present, the result value of the ``udiv`` is
4791 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
4792 such, "((a udiv exact b) mul b) == a").
4797 .. code-block:: llvm
4799 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
4801 '``sdiv``' Instruction
4802 ^^^^^^^^^^^^^^^^^^^^^^
4809 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
4810 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
4815 The '``sdiv``' instruction returns the quotient of its two operands.
4820 The two arguments to the '``sdiv``' instruction must be
4821 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4822 arguments must have identical types.
4827 The value produced is the signed integer quotient of the two operands
4828 rounded towards zero.
4830 Note that signed integer division and unsigned integer division are
4831 distinct operations; for unsigned integer division, use '``udiv``'.
4833 Division by zero leads to undefined behavior. Overflow also leads to
4834 undefined behavior; this is a rare case, but can occur, for example, by
4835 doing a 32-bit division of -2147483648 by -1.
4837 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4838 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4843 .. code-block:: llvm
4845 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
4849 '``fdiv``' Instruction
4850 ^^^^^^^^^^^^^^^^^^^^^^
4857 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4862 The '``fdiv``' instruction returns the quotient of its two operands.
4867 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4868 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4869 Both arguments must have identical types.
4874 The value produced is the floating point quotient of the two operands.
4875 This instruction can also take any number of :ref:`fast-math
4876 flags <fastmath>`, which are optimization hints to enable otherwise
4877 unsafe floating point optimizations:
4882 .. code-block:: llvm
4884 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
4886 '``urem``' Instruction
4887 ^^^^^^^^^^^^^^^^^^^^^^
4894 <result> = urem <ty> <op1>, <op2> ; yields ty:result
4899 The '``urem``' instruction returns the remainder from the unsigned
4900 division of its two arguments.
4905 The two arguments to the '``urem``' instruction must be
4906 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4907 arguments must have identical types.
4912 This instruction returns the unsigned integer *remainder* of a division.
4913 This instruction always performs an unsigned division to get the
4916 Note that unsigned integer remainder and signed integer remainder are
4917 distinct operations; for signed integer remainder, use '``srem``'.
4919 Taking the remainder of a division by zero leads to undefined behavior.
4924 .. code-block:: llvm
4926 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
4928 '``srem``' Instruction
4929 ^^^^^^^^^^^^^^^^^^^^^^
4936 <result> = srem <ty> <op1>, <op2> ; yields ty:result
4941 The '``srem``' instruction returns the remainder from the signed
4942 division of its two operands. This instruction can also take
4943 :ref:`vector <t_vector>` versions of the values in which case the elements
4949 The two arguments to the '``srem``' instruction must be
4950 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4951 arguments must have identical types.
4956 This instruction returns the *remainder* of a division (where the result
4957 is either zero or has the same sign as the dividend, ``op1``), not the
4958 *modulo* operator (where the result is either zero or has the same sign
4959 as the divisor, ``op2``) of a value. For more information about the
4960 difference, see `The Math
4961 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4962 table of how this is implemented in various languages, please see
4964 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4966 Note that signed integer remainder and unsigned integer remainder are
4967 distinct operations; for unsigned integer remainder, use '``urem``'.
4969 Taking the remainder of a division by zero leads to undefined behavior.
4970 Overflow also leads to undefined behavior; this is a rare case, but can
4971 occur, for example, by taking the remainder of a 32-bit division of
4972 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4973 rule lets srem be implemented using instructions that return both the
4974 result of the division and the remainder.)
4979 .. code-block:: llvm
4981 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
4985 '``frem``' Instruction
4986 ^^^^^^^^^^^^^^^^^^^^^^
4993 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4998 The '``frem``' instruction returns the remainder from the division of
5004 The two arguments to the '``frem``' instruction must be :ref:`floating
5005 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5006 Both arguments must have identical types.
5011 This instruction returns the *remainder* of a division. The remainder
5012 has the same sign as the dividend. This instruction can also take any
5013 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
5014 to enable otherwise unsafe floating point optimizations:
5019 .. code-block:: llvm
5021 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
5025 Bitwise Binary Operations
5026 -------------------------
5028 Bitwise binary operators are used to do various forms of bit-twiddling
5029 in a program. They are generally very efficient instructions and can
5030 commonly be strength reduced from other instructions. They require two
5031 operands of the same type, execute an operation on them, and produce a
5032 single value. The resulting value is the same type as its operands.
5034 '``shl``' Instruction
5035 ^^^^^^^^^^^^^^^^^^^^^
5042 <result> = shl <ty> <op1>, <op2> ; yields ty:result
5043 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
5044 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
5045 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
5050 The '``shl``' instruction returns the first operand shifted to the left
5051 a specified number of bits.
5056 Both arguments to the '``shl``' instruction must be the same
5057 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5058 '``op2``' is treated as an unsigned value.
5063 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
5064 where ``n`` is the width of the result. If ``op2`` is (statically or
5065 dynamically) equal to or larger than the number of bits in
5066 ``op1``, the result is undefined. If the arguments are vectors, each
5067 vector element of ``op1`` is shifted by the corresponding shift amount
5070 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
5071 value <poisonvalues>` if it shifts out any non-zero bits. If the
5072 ``nsw`` keyword is present, then the shift produces a :ref:`poison
5073 value <poisonvalues>` if it shifts out any bits that disagree with the
5074 resultant sign bit. As such, NUW/NSW have the same semantics as they
5075 would if the shift were expressed as a mul instruction with the same
5076 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
5081 .. code-block:: llvm
5083 <result> = shl i32 4, %var ; yields i32: 4 << %var
5084 <result> = shl i32 4, 2 ; yields i32: 16
5085 <result> = shl i32 1, 10 ; yields i32: 1024
5086 <result> = shl i32 1, 32 ; undefined
5087 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
5089 '``lshr``' Instruction
5090 ^^^^^^^^^^^^^^^^^^^^^^
5097 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
5098 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
5103 The '``lshr``' instruction (logical shift right) returns the first
5104 operand shifted to the right a specified number of bits with zero fill.
5109 Both arguments to the '``lshr``' instruction must be the same
5110 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5111 '``op2``' is treated as an unsigned value.
5116 This instruction always performs a logical shift right operation. The
5117 most significant bits of the result will be filled with zero bits after
5118 the shift. If ``op2`` is (statically or dynamically) equal to or larger
5119 than the number of bits in ``op1``, the result is undefined. If the
5120 arguments are vectors, each vector element of ``op1`` is shifted by the
5121 corresponding shift amount in ``op2``.
5123 If the ``exact`` keyword is present, the result value of the ``lshr`` is
5124 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5130 .. code-block:: llvm
5132 <result> = lshr i32 4, 1 ; yields i32:result = 2
5133 <result> = lshr i32 4, 2 ; yields i32:result = 1
5134 <result> = lshr i8 4, 3 ; yields i8:result = 0
5135 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
5136 <result> = lshr i32 1, 32 ; undefined
5137 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
5139 '``ashr``' Instruction
5140 ^^^^^^^^^^^^^^^^^^^^^^
5147 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
5148 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
5153 The '``ashr``' instruction (arithmetic shift right) returns the first
5154 operand shifted to the right a specified number of bits with sign
5160 Both arguments to the '``ashr``' instruction must be the same
5161 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5162 '``op2``' is treated as an unsigned value.
5167 This instruction always performs an arithmetic shift right operation,
5168 The most significant bits of the result will be filled with the sign bit
5169 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
5170 than the number of bits in ``op1``, the result is undefined. If the
5171 arguments are vectors, each vector element of ``op1`` is shifted by the
5172 corresponding shift amount in ``op2``.
5174 If the ``exact`` keyword is present, the result value of the ``ashr`` is
5175 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5181 .. code-block:: llvm
5183 <result> = ashr i32 4, 1 ; yields i32:result = 2
5184 <result> = ashr i32 4, 2 ; yields i32:result = 1
5185 <result> = ashr i8 4, 3 ; yields i8:result = 0
5186 <result> = ashr i8 -2, 1 ; yields i8:result = -1
5187 <result> = ashr i32 1, 32 ; undefined
5188 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
5190 '``and``' Instruction
5191 ^^^^^^^^^^^^^^^^^^^^^
5198 <result> = and <ty> <op1>, <op2> ; yields ty:result
5203 The '``and``' instruction returns the bitwise logical and of its two
5209 The two arguments to the '``and``' instruction must be
5210 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5211 arguments must have identical types.
5216 The truth table used for the '``and``' instruction is:
5233 .. code-block:: llvm
5235 <result> = and i32 4, %var ; yields i32:result = 4 & %var
5236 <result> = and i32 15, 40 ; yields i32:result = 8
5237 <result> = and i32 4, 8 ; yields i32:result = 0
5239 '``or``' Instruction
5240 ^^^^^^^^^^^^^^^^^^^^
5247 <result> = or <ty> <op1>, <op2> ; yields ty:result
5252 The '``or``' instruction returns the bitwise logical inclusive or of its
5258 The two arguments to the '``or``' instruction must be
5259 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5260 arguments must have identical types.
5265 The truth table used for the '``or``' instruction is:
5284 <result> = or i32 4, %var ; yields i32:result = 4 | %var
5285 <result> = or i32 15, 40 ; yields i32:result = 47
5286 <result> = or i32 4, 8 ; yields i32:result = 12
5288 '``xor``' Instruction
5289 ^^^^^^^^^^^^^^^^^^^^^
5296 <result> = xor <ty> <op1>, <op2> ; yields ty:result
5301 The '``xor``' instruction returns the bitwise logical exclusive or of
5302 its two operands. The ``xor`` is used to implement the "one's
5303 complement" operation, which is the "~" operator in C.
5308 The two arguments to the '``xor``' instruction must be
5309 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5310 arguments must have identical types.
5315 The truth table used for the '``xor``' instruction is:
5332 .. code-block:: llvm
5334 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
5335 <result> = xor i32 15, 40 ; yields i32:result = 39
5336 <result> = xor i32 4, 8 ; yields i32:result = 12
5337 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
5342 LLVM supports several instructions to represent vector operations in a
5343 target-independent manner. These instructions cover the element-access
5344 and vector-specific operations needed to process vectors effectively.
5345 While LLVM does directly support these vector operations, many
5346 sophisticated algorithms will want to use target-specific intrinsics to
5347 take full advantage of a specific target.
5349 .. _i_extractelement:
5351 '``extractelement``' Instruction
5352 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5359 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
5364 The '``extractelement``' instruction extracts a single scalar element
5365 from a vector at a specified index.
5370 The first operand of an '``extractelement``' instruction is a value of
5371 :ref:`vector <t_vector>` type. The second operand is an index indicating
5372 the position from which to extract the element. The index may be a
5373 variable of any integer type.
5378 The result is a scalar of the same type as the element type of ``val``.
5379 Its value is the value at position ``idx`` of ``val``. If ``idx``
5380 exceeds the length of ``val``, the results are undefined.
5385 .. code-block:: llvm
5387 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
5389 .. _i_insertelement:
5391 '``insertelement``' Instruction
5392 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5399 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
5404 The '``insertelement``' instruction inserts a scalar element into a
5405 vector at a specified index.
5410 The first operand of an '``insertelement``' instruction is a value of
5411 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
5412 type must equal the element type of the first operand. The third operand
5413 is an index indicating the position at which to insert the value. The
5414 index may be a variable of any integer type.
5419 The result is a vector of the same type as ``val``. Its element values
5420 are those of ``val`` except at position ``idx``, where it gets the value
5421 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
5427 .. code-block:: llvm
5429 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
5431 .. _i_shufflevector:
5433 '``shufflevector``' Instruction
5434 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5441 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
5446 The '``shufflevector``' instruction constructs a permutation of elements
5447 from two input vectors, returning a vector with the same element type as
5448 the input and length that is the same as the shuffle mask.
5453 The first two operands of a '``shufflevector``' instruction are vectors
5454 with the same type. The third argument is a shuffle mask whose element
5455 type is always 'i32'. The result of the instruction is a vector whose
5456 length is the same as the shuffle mask and whose element type is the
5457 same as the element type of the first two operands.
5459 The shuffle mask operand is required to be a constant vector with either
5460 constant integer or undef values.
5465 The elements of the two input vectors are numbered from left to right
5466 across both of the vectors. The shuffle mask operand specifies, for each
5467 element of the result vector, which element of the two input vectors the
5468 result element gets. The element selector may be undef (meaning "don't
5469 care") and the second operand may be undef if performing a shuffle from
5475 .. code-block:: llvm
5477 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5478 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
5479 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
5480 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
5481 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
5482 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
5483 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5484 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
5486 Aggregate Operations
5487 --------------------
5489 LLVM supports several instructions for working with
5490 :ref:`aggregate <t_aggregate>` values.
5494 '``extractvalue``' Instruction
5495 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5502 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
5507 The '``extractvalue``' instruction extracts the value of a member field
5508 from an :ref:`aggregate <t_aggregate>` value.
5513 The first operand of an '``extractvalue``' instruction is a value of
5514 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
5515 constant indices to specify which value to extract in a similar manner
5516 as indices in a '``getelementptr``' instruction.
5518 The major differences to ``getelementptr`` indexing are:
5520 - Since the value being indexed is not a pointer, the first index is
5521 omitted and assumed to be zero.
5522 - At least one index must be specified.
5523 - Not only struct indices but also array indices must be in bounds.
5528 The result is the value at the position in the aggregate specified by
5534 .. code-block:: llvm
5536 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
5540 '``insertvalue``' Instruction
5541 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5548 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
5553 The '``insertvalue``' instruction inserts a value into a member field in
5554 an :ref:`aggregate <t_aggregate>` value.
5559 The first operand of an '``insertvalue``' instruction is a value of
5560 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
5561 a first-class value to insert. The following operands are constant
5562 indices indicating the position at which to insert the value in a
5563 similar manner as indices in a '``extractvalue``' instruction. The value
5564 to insert must have the same type as the value identified by the
5570 The result is an aggregate of the same type as ``val``. Its value is
5571 that of ``val`` except that the value at the position specified by the
5572 indices is that of ``elt``.
5577 .. code-block:: llvm
5579 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
5580 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
5581 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
5585 Memory Access and Addressing Operations
5586 ---------------------------------------
5588 A key design point of an SSA-based representation is how it represents
5589 memory. In LLVM, no memory locations are in SSA form, which makes things
5590 very simple. This section describes how to read, write, and allocate
5595 '``alloca``' Instruction
5596 ^^^^^^^^^^^^^^^^^^^^^^^^
5603 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
5608 The '``alloca``' instruction allocates memory on the stack frame of the
5609 currently executing function, to be automatically released when this
5610 function returns to its caller. The object is always allocated in the
5611 generic address space (address space zero).
5616 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
5617 bytes of memory on the runtime stack, returning a pointer of the
5618 appropriate type to the program. If "NumElements" is specified, it is
5619 the number of elements allocated, otherwise "NumElements" is defaulted
5620 to be one. If a constant alignment is specified, the value result of the
5621 allocation is guaranteed to be aligned to at least that boundary. The
5622 alignment may not be greater than ``1 << 29``. If not specified, or if
5623 zero, the target can choose to align the allocation on any convenient
5624 boundary compatible with the type.
5626 '``type``' may be any sized type.
5631 Memory is allocated; a pointer is returned. The operation is undefined
5632 if there is insufficient stack space for the allocation. '``alloca``'d
5633 memory is automatically released when the function returns. The
5634 '``alloca``' instruction is commonly used to represent automatic
5635 variables that must have an address available. When the function returns
5636 (either with the ``ret`` or ``resume`` instructions), the memory is
5637 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5638 The order in which memory is allocated (ie., which way the stack grows)
5644 .. code-block:: llvm
5646 %ptr = alloca i32 ; yields i32*:ptr
5647 %ptr = alloca i32, i32 4 ; yields i32*:ptr
5648 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
5649 %ptr = alloca i32, align 1024 ; yields i32*:ptr
5653 '``load``' Instruction
5654 ^^^^^^^^^^^^^^^^^^^^^^
5661 <result> = load [volatile] <ty>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !nonnull !<index>][, !dereferenceable !<index>][, !dereferenceable_or_null !<index>]
5662 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5663 !<index> = !{ i32 1 }
5668 The '``load``' instruction is used to read from memory.
5673 The argument to the ``load`` instruction specifies the memory address
5674 from which to load. The type specified must be a :ref:`first
5675 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
5676 then the optimizer is not allowed to modify the number or order of
5677 execution of this ``load`` with other :ref:`volatile
5678 operations <volatile>`.
5680 If the ``load`` is marked as ``atomic``, it takes an extra
5681 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5682 ``release`` and ``acq_rel`` orderings are not valid on ``load``
5683 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5684 when they may see multiple atomic stores. The type of the pointee must
5685 be an integer type whose bit width is a power of two greater than or
5686 equal to eight and less than or equal to a target-specific size limit.
5687 ``align`` must be explicitly specified on atomic loads, and the load has
5688 undefined behavior if the alignment is not set to a value which is at
5689 least the size in bytes of the pointee. ``!nontemporal`` does not have
5690 any defined semantics for atomic loads.
5692 The optional constant ``align`` argument specifies the alignment of the
5693 operation (that is, the alignment of the memory address). A value of 0
5694 or an omitted ``align`` argument means that the operation has the ABI
5695 alignment for the target. It is the responsibility of the code emitter
5696 to ensure that the alignment information is correct. Overestimating the
5697 alignment results in undefined behavior. Underestimating the alignment
5698 may produce less efficient code. An alignment of 1 is always safe. The
5699 maximum possible alignment is ``1 << 29``.
5701 The optional ``!nontemporal`` metadata must reference a single
5702 metadata name ``<index>`` corresponding to a metadata node with one
5703 ``i32`` entry of value 1. The existence of the ``!nontemporal``
5704 metadata on the instruction tells the optimizer and code generator
5705 that this load is not expected to be reused in the cache. The code
5706 generator may select special instructions to save cache bandwidth, such
5707 as the ``MOVNT`` instruction on x86.
5709 The optional ``!invariant.load`` metadata must reference a single
5710 metadata name ``<index>`` corresponding to a metadata node with no
5711 entries. The existence of the ``!invariant.load`` metadata on the
5712 instruction tells the optimizer and code generator that the address
5713 operand to this load points to memory which can be assumed unchanged.
5714 Being invariant does not imply that a location is dereferenceable,
5715 but it does imply that once the location is known dereferenceable
5716 its value is henceforth unchanging.
5718 The optional ``!nonnull`` metadata must reference a single
5719 metadata name ``<index>`` corresponding to a metadata node with no
5720 entries. The existence of the ``!nonnull`` metadata on the
5721 instruction tells the optimizer that the value loaded is known to
5722 never be null. This is analogous to the ''nonnull'' attribute
5723 on parameters and return values. This metadata can only be applied
5724 to loads of a pointer type.
5726 The optional ``!dereferenceable`` metadata must reference a single
5727 metadata name ``<index>`` corresponding to a metadata node with one ``i64``
5728 entry. The existence of the ``!dereferenceable`` metadata on the instruction
5729 tells the optimizer that the value loaded is known to be dereferenceable.
5730 The number of bytes known to be dereferenceable is specified by the integer
5731 value in the metadata node. This is analogous to the ''dereferenceable''
5732 attribute on parameters and return values. This metadata can only be applied
5733 to loads of a pointer type.
5735 The optional ``!dereferenceable_or_null`` metadata must reference a single
5736 metadata name ``<index>`` corresponding to a metadata node with one ``i64``
5737 entry. The existence of the ``!dereferenceable_or_null`` metadata on the
5738 instruction tells the optimizer that the value loaded is known to be either
5739 dereferenceable or null.
5740 The number of bytes known to be dereferenceable is specified by the integer
5741 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
5742 attribute on parameters and return values. This metadata can only be applied
5743 to loads of a pointer type.
5748 The location of memory pointed to is loaded. If the value being loaded
5749 is of scalar type then the number of bytes read does not exceed the
5750 minimum number of bytes needed to hold all bits of the type. For
5751 example, loading an ``i24`` reads at most three bytes. When loading a
5752 value of a type like ``i20`` with a size that is not an integral number
5753 of bytes, the result is undefined if the value was not originally
5754 written using a store of the same type.
5759 .. code-block:: llvm
5761 %ptr = alloca i32 ; yields i32*:ptr
5762 store i32 3, i32* %ptr ; yields void
5763 %val = load i32, i32* %ptr ; yields i32:val = i32 3
5767 '``store``' Instruction
5768 ^^^^^^^^^^^^^^^^^^^^^^^
5775 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
5776 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
5781 The '``store``' instruction is used to write to memory.
5786 There are two arguments to the ``store`` instruction: a value to store
5787 and an address at which to store it. The type of the ``<pointer>``
5788 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
5789 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
5790 then the optimizer is not allowed to modify the number or order of
5791 execution of this ``store`` with other :ref:`volatile
5792 operations <volatile>`.
5794 If the ``store`` is marked as ``atomic``, it takes an extra
5795 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5796 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
5797 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5798 when they may see multiple atomic stores. The type of the pointee must
5799 be an integer type whose bit width is a power of two greater than or
5800 equal to eight and less than or equal to a target-specific size limit.
5801 ``align`` must be explicitly specified on atomic stores, and the store
5802 has undefined behavior if the alignment is not set to a value which is
5803 at least the size in bytes of the pointee. ``!nontemporal`` does not
5804 have any defined semantics for atomic stores.
5806 The optional constant ``align`` argument specifies the alignment of the
5807 operation (that is, the alignment of the memory address). A value of 0
5808 or an omitted ``align`` argument means that the operation has the ABI
5809 alignment for the target. It is the responsibility of the code emitter
5810 to ensure that the alignment information is correct. Overestimating the
5811 alignment results in undefined behavior. Underestimating the
5812 alignment may produce less efficient code. An alignment of 1 is always
5813 safe. The maximum possible alignment is ``1 << 29``.
5815 The optional ``!nontemporal`` metadata must reference a single metadata
5816 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
5817 value 1. The existence of the ``!nontemporal`` metadata on the instruction
5818 tells the optimizer and code generator that this load is not expected to
5819 be reused in the cache. The code generator may select special
5820 instructions to save cache bandwidth, such as the MOVNT instruction on
5826 The contents of memory are updated to contain ``<value>`` at the
5827 location specified by the ``<pointer>`` operand. If ``<value>`` is
5828 of scalar type then the number of bytes written does not exceed the
5829 minimum number of bytes needed to hold all bits of the type. For
5830 example, storing an ``i24`` writes at most three bytes. When writing a
5831 value of a type like ``i20`` with a size that is not an integral number
5832 of bytes, it is unspecified what happens to the extra bits that do not
5833 belong to the type, but they will typically be overwritten.
5838 .. code-block:: llvm
5840 %ptr = alloca i32 ; yields i32*:ptr
5841 store i32 3, i32* %ptr ; yields void
5842 %val = load i32* %ptr ; yields i32:val = i32 3
5846 '``fence``' Instruction
5847 ^^^^^^^^^^^^^^^^^^^^^^^
5854 fence [singlethread] <ordering> ; yields void
5859 The '``fence``' instruction is used to introduce happens-before edges
5865 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5866 defines what *synchronizes-with* edges they add. They can only be given
5867 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5872 A fence A which has (at least) ``release`` ordering semantics
5873 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5874 semantics if and only if there exist atomic operations X and Y, both
5875 operating on some atomic object M, such that A is sequenced before X, X
5876 modifies M (either directly or through some side effect of a sequence
5877 headed by X), Y is sequenced before B, and Y observes M. This provides a
5878 *happens-before* dependency between A and B. Rather than an explicit
5879 ``fence``, one (but not both) of the atomic operations X or Y might
5880 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5881 still *synchronize-with* the explicit ``fence`` and establish the
5882 *happens-before* edge.
5884 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5885 ``acquire`` and ``release`` semantics specified above, participates in
5886 the global program order of other ``seq_cst`` operations and/or fences.
5888 The optional ":ref:`singlethread <singlethread>`" argument specifies
5889 that the fence only synchronizes with other fences in the same thread.
5890 (This is useful for interacting with signal handlers.)
5895 .. code-block:: llvm
5897 fence acquire ; yields void
5898 fence singlethread seq_cst ; yields void
5902 '``cmpxchg``' Instruction
5903 ^^^^^^^^^^^^^^^^^^^^^^^^^
5910 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
5915 The '``cmpxchg``' instruction is used to atomically modify memory. It
5916 loads a value in memory and compares it to a given value. If they are
5917 equal, it tries to store a new value into the memory.
5922 There are three arguments to the '``cmpxchg``' instruction: an address
5923 to operate on, a value to compare to the value currently be at that
5924 address, and a new value to place at that address if the compared values
5925 are equal. The type of '<cmp>' must be an integer type whose bit width
5926 is a power of two greater than or equal to eight and less than or equal
5927 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5928 type, and the type of '<pointer>' must be a pointer to that type. If the
5929 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5930 to modify the number or order of execution of this ``cmpxchg`` with
5931 other :ref:`volatile operations <volatile>`.
5933 The success and failure :ref:`ordering <ordering>` arguments specify how this
5934 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
5935 must be at least ``monotonic``, the ordering constraint on failure must be no
5936 stronger than that on success, and the failure ordering cannot be either
5937 ``release`` or ``acq_rel``.
5939 The optional "``singlethread``" argument declares that the ``cmpxchg``
5940 is only atomic with respect to code (usually signal handlers) running in
5941 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5942 respect to all other code in the system.
5944 The pointer passed into cmpxchg must have alignment greater than or
5945 equal to the size in memory of the operand.
5950 The contents of memory at the location specified by the '``<pointer>``' operand
5951 is read and compared to '``<cmp>``'; if the read value is the equal, the
5952 '``<new>``' is written. The original value at the location is returned, together
5953 with a flag indicating success (true) or failure (false).
5955 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
5956 permitted: the operation may not write ``<new>`` even if the comparison
5959 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
5960 if the value loaded equals ``cmp``.
5962 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5963 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5964 load with an ordering parameter determined the second ordering parameter.
5969 .. code-block:: llvm
5972 %orig = atomic load i32, i32* %ptr unordered ; yields i32
5976 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5977 %squared = mul i32 %cmp, %cmp
5978 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
5979 %value_loaded = extractvalue { i32, i1 } %val_success, 0
5980 %success = extractvalue { i32, i1 } %val_success, 1
5981 br i1 %success, label %done, label %loop
5988 '``atomicrmw``' Instruction
5989 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5996 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
6001 The '``atomicrmw``' instruction is used to atomically modify memory.
6006 There are three arguments to the '``atomicrmw``' instruction: an
6007 operation to apply, an address whose value to modify, an argument to the
6008 operation. The operation must be one of the following keywords:
6022 The type of '<value>' must be an integer type whose bit width is a power
6023 of two greater than or equal to eight and less than or equal to a
6024 target-specific size limit. The type of the '``<pointer>``' operand must
6025 be a pointer to that type. If the ``atomicrmw`` is marked as
6026 ``volatile``, then the optimizer is not allowed to modify the number or
6027 order of execution of this ``atomicrmw`` with other :ref:`volatile
6028 operations <volatile>`.
6033 The contents of memory at the location specified by the '``<pointer>``'
6034 operand are atomically read, modified, and written back. The original
6035 value at the location is returned. The modification is specified by the
6038 - xchg: ``*ptr = val``
6039 - add: ``*ptr = *ptr + val``
6040 - sub: ``*ptr = *ptr - val``
6041 - and: ``*ptr = *ptr & val``
6042 - nand: ``*ptr = ~(*ptr & val)``
6043 - or: ``*ptr = *ptr | val``
6044 - xor: ``*ptr = *ptr ^ val``
6045 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
6046 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
6047 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
6049 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
6055 .. code-block:: llvm
6057 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
6059 .. _i_getelementptr:
6061 '``getelementptr``' Instruction
6062 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6069 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
6070 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
6071 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
6076 The '``getelementptr``' instruction is used to get the address of a
6077 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
6078 address calculation only and does not access memory.
6083 The first argument is always a type used as the basis for the calculations.
6084 The second argument is always a pointer or a vector of pointers, and is the
6085 base address to start from. The remaining arguments are indices
6086 that indicate which of the elements of the aggregate object are indexed.
6087 The interpretation of each index is dependent on the type being indexed
6088 into. The first index always indexes the pointer value given as the
6089 first argument, the second index indexes a value of the type pointed to
6090 (not necessarily the value directly pointed to, since the first index
6091 can be non-zero), etc. The first type indexed into must be a pointer
6092 value, subsequent types can be arrays, vectors, and structs. Note that
6093 subsequent types being indexed into can never be pointers, since that
6094 would require loading the pointer before continuing calculation.
6096 The type of each index argument depends on the type it is indexing into.
6097 When indexing into a (optionally packed) structure, only ``i32`` integer
6098 **constants** are allowed (when using a vector of indices they must all
6099 be the **same** ``i32`` integer constant). When indexing into an array,
6100 pointer or vector, integers of any width are allowed, and they are not
6101 required to be constant. These integers are treated as signed values
6104 For example, let's consider a C code fragment and how it gets compiled
6120 int *foo(struct ST *s) {
6121 return &s[1].Z.B[5][13];
6124 The LLVM code generated by Clang is:
6126 .. code-block:: llvm
6128 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
6129 %struct.ST = type { i32, double, %struct.RT }
6131 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
6133 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
6140 In the example above, the first index is indexing into the
6141 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
6142 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
6143 indexes into the third element of the structure, yielding a
6144 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
6145 structure. The third index indexes into the second element of the
6146 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
6147 dimensions of the array are subscripted into, yielding an '``i32``'
6148 type. The '``getelementptr``' instruction returns a pointer to this
6149 element, thus computing a value of '``i32*``' type.
6151 Note that it is perfectly legal to index partially through a structure,
6152 returning a pointer to an inner element. Because of this, the LLVM code
6153 for the given testcase is equivalent to:
6155 .. code-block:: llvm
6157 define i32* @foo(%struct.ST* %s) {
6158 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
6159 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
6160 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
6161 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
6162 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
6166 If the ``inbounds`` keyword is present, the result value of the
6167 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
6168 pointer is not an *in bounds* address of an allocated object, or if any
6169 of the addresses that would be formed by successive addition of the
6170 offsets implied by the indices to the base address with infinitely
6171 precise signed arithmetic are not an *in bounds* address of that
6172 allocated object. The *in bounds* addresses for an allocated object are
6173 all the addresses that point into the object, plus the address one byte
6174 past the end. In cases where the base is a vector of pointers the
6175 ``inbounds`` keyword applies to each of the computations element-wise.
6177 If the ``inbounds`` keyword is not present, the offsets are added to the
6178 base address with silently-wrapping two's complement arithmetic. If the
6179 offsets have a different width from the pointer, they are sign-extended
6180 or truncated to the width of the pointer. The result value of the
6181 ``getelementptr`` may be outside the object pointed to by the base
6182 pointer. The result value may not necessarily be used to access memory
6183 though, even if it happens to point into allocated storage. See the
6184 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
6187 The getelementptr instruction is often confusing. For some more insight
6188 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
6193 .. code-block:: llvm
6195 ; yields [12 x i8]*:aptr
6196 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
6198 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
6200 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
6202 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
6204 In cases where the pointer argument is a vector of pointers, each index
6205 must be a vector with the same number of elements. For example:
6207 .. code-block:: llvm
6209 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets,
6211 Conversion Operations
6212 ---------------------
6214 The instructions in this category are the conversion instructions
6215 (casting) which all take a single operand and a type. They perform
6216 various bit conversions on the operand.
6218 '``trunc .. to``' Instruction
6219 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6226 <result> = trunc <ty> <value> to <ty2> ; yields ty2
6231 The '``trunc``' instruction truncates its operand to the type ``ty2``.
6236 The '``trunc``' instruction takes a value to trunc, and a type to trunc
6237 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
6238 of the same number of integers. The bit size of the ``value`` must be
6239 larger than the bit size of the destination type, ``ty2``. Equal sized
6240 types are not allowed.
6245 The '``trunc``' instruction truncates the high order bits in ``value``
6246 and converts the remaining bits to ``ty2``. Since the source size must
6247 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
6248 It will always truncate bits.
6253 .. code-block:: llvm
6255 %X = trunc i32 257 to i8 ; yields i8:1
6256 %Y = trunc i32 123 to i1 ; yields i1:true
6257 %Z = trunc i32 122 to i1 ; yields i1:false
6258 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
6260 '``zext .. to``' Instruction
6261 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6268 <result> = zext <ty> <value> to <ty2> ; yields ty2
6273 The '``zext``' instruction zero extends its operand to type ``ty2``.
6278 The '``zext``' instruction takes a value to cast, and a type to cast it
6279 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
6280 the same number of integers. The bit size of the ``value`` must be
6281 smaller than the bit size of the destination type, ``ty2``.
6286 The ``zext`` fills the high order bits of the ``value`` with zero bits
6287 until it reaches the size of the destination type, ``ty2``.
6289 When zero extending from i1, the result will always be either 0 or 1.
6294 .. code-block:: llvm
6296 %X = zext i32 257 to i64 ; yields i64:257
6297 %Y = zext i1 true to i32 ; yields i32:1
6298 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
6300 '``sext .. to``' Instruction
6301 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6308 <result> = sext <ty> <value> to <ty2> ; yields ty2
6313 The '``sext``' sign extends ``value`` to the type ``ty2``.
6318 The '``sext``' instruction takes a value to cast, and a type to cast it
6319 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
6320 the same number of integers. The bit size of the ``value`` must be
6321 smaller than the bit size of the destination type, ``ty2``.
6326 The '``sext``' instruction performs a sign extension by copying the sign
6327 bit (highest order bit) of the ``value`` until it reaches the bit size
6328 of the type ``ty2``.
6330 When sign extending from i1, the extension always results in -1 or 0.
6335 .. code-block:: llvm
6337 %X = sext i8 -1 to i16 ; yields i16 :65535
6338 %Y = sext i1 true to i32 ; yields i32:-1
6339 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
6341 '``fptrunc .. to``' Instruction
6342 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6349 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
6354 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
6359 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
6360 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
6361 The size of ``value`` must be larger than the size of ``ty2``. This
6362 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
6367 The '``fptrunc``' instruction truncates a ``value`` from a larger
6368 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
6369 point <t_floating>` type. If the value cannot fit within the
6370 destination type, ``ty2``, then the results are undefined.
6375 .. code-block:: llvm
6377 %X = fptrunc double 123.0 to float ; yields float:123.0
6378 %Y = fptrunc double 1.0E+300 to float ; yields undefined
6380 '``fpext .. to``' Instruction
6381 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6388 <result> = fpext <ty> <value> to <ty2> ; yields ty2
6393 The '``fpext``' extends a floating point ``value`` to a larger floating
6399 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
6400 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
6401 to. The source type must be smaller than the destination type.
6406 The '``fpext``' instruction extends the ``value`` from a smaller
6407 :ref:`floating point <t_floating>` type to a larger :ref:`floating
6408 point <t_floating>` type. The ``fpext`` cannot be used to make a
6409 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
6410 *no-op cast* for a floating point cast.
6415 .. code-block:: llvm
6417 %X = fpext float 3.125 to double ; yields double:3.125000e+00
6418 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
6420 '``fptoui .. to``' Instruction
6421 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6428 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
6433 The '``fptoui``' converts a floating point ``value`` to its unsigned
6434 integer equivalent of type ``ty2``.
6439 The '``fptoui``' instruction takes a value to cast, which must be a
6440 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6441 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6442 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6443 type with the same number of elements as ``ty``
6448 The '``fptoui``' instruction converts its :ref:`floating
6449 point <t_floating>` operand into the nearest (rounding towards zero)
6450 unsigned integer value. If the value cannot fit in ``ty2``, the results
6456 .. code-block:: llvm
6458 %X = fptoui double 123.0 to i32 ; yields i32:123
6459 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
6460 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
6462 '``fptosi .. to``' Instruction
6463 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6470 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
6475 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
6476 ``value`` to type ``ty2``.
6481 The '``fptosi``' instruction takes a value to cast, which must be a
6482 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6483 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6484 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6485 type with the same number of elements as ``ty``
6490 The '``fptosi``' instruction converts its :ref:`floating
6491 point <t_floating>` operand into the nearest (rounding towards zero)
6492 signed integer value. If the value cannot fit in ``ty2``, the results
6498 .. code-block:: llvm
6500 %X = fptosi double -123.0 to i32 ; yields i32:-123
6501 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
6502 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
6504 '``uitofp .. to``' Instruction
6505 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6512 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
6517 The '``uitofp``' instruction regards ``value`` as an unsigned integer
6518 and converts that value to the ``ty2`` type.
6523 The '``uitofp``' instruction takes a value to cast, which must be a
6524 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6525 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6526 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6527 type with the same number of elements as ``ty``
6532 The '``uitofp``' instruction interprets its operand as an unsigned
6533 integer quantity and converts it to the corresponding floating point
6534 value. If the value cannot fit in the floating point value, the results
6540 .. code-block:: llvm
6542 %X = uitofp i32 257 to float ; yields float:257.0
6543 %Y = uitofp i8 -1 to double ; yields double:255.0
6545 '``sitofp .. to``' Instruction
6546 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6553 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
6558 The '``sitofp``' instruction regards ``value`` as a signed integer and
6559 converts that value to the ``ty2`` type.
6564 The '``sitofp``' instruction takes a value to cast, which must be a
6565 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6566 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6567 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6568 type with the same number of elements as ``ty``
6573 The '``sitofp``' instruction interprets its operand as a signed integer
6574 quantity and converts it to the corresponding floating point value. If
6575 the value cannot fit in the floating point value, the results are
6581 .. code-block:: llvm
6583 %X = sitofp i32 257 to float ; yields float:257.0
6584 %Y = sitofp i8 -1 to double ; yields double:-1.0
6588 '``ptrtoint .. to``' Instruction
6589 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6596 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
6601 The '``ptrtoint``' instruction converts the pointer or a vector of
6602 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
6607 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
6608 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
6609 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
6610 a vector of integers type.
6615 The '``ptrtoint``' instruction converts ``value`` to integer type
6616 ``ty2`` by interpreting the pointer value as an integer and either
6617 truncating or zero extending that value to the size of the integer type.
6618 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
6619 ``value`` is larger than ``ty2`` then a truncation is done. If they are
6620 the same size, then nothing is done (*no-op cast*) other than a type
6626 .. code-block:: llvm
6628 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
6629 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
6630 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
6634 '``inttoptr .. to``' Instruction
6635 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6642 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
6647 The '``inttoptr``' instruction converts an integer ``value`` to a
6648 pointer type, ``ty2``.
6653 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
6654 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
6660 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
6661 applying either a zero extension or a truncation depending on the size
6662 of the integer ``value``. If ``value`` is larger than the size of a
6663 pointer then a truncation is done. If ``value`` is smaller than the size
6664 of a pointer then a zero extension is done. If they are the same size,
6665 nothing is done (*no-op cast*).
6670 .. code-block:: llvm
6672 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
6673 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
6674 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
6675 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
6679 '``bitcast .. to``' Instruction
6680 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6687 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
6692 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
6698 The '``bitcast``' instruction takes a value to cast, which must be a
6699 non-aggregate first class value, and a type to cast it to, which must
6700 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
6701 bit sizes of ``value`` and the destination type, ``ty2``, must be
6702 identical. If the source type is a pointer, the destination type must
6703 also be a pointer of the same size. This instruction supports bitwise
6704 conversion of vectors to integers and to vectors of other types (as
6705 long as they have the same size).
6710 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
6711 is always a *no-op cast* because no bits change with this
6712 conversion. The conversion is done as if the ``value`` had been stored
6713 to memory and read back as type ``ty2``. Pointer (or vector of
6714 pointers) types may only be converted to other pointer (or vector of
6715 pointers) types with the same address space through this instruction.
6716 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
6717 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
6722 .. code-block:: llvm
6724 %X = bitcast i8 255 to i8 ; yields i8 :-1
6725 %Y = bitcast i32* %x to sint* ; yields sint*:%x
6726 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
6727 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
6729 .. _i_addrspacecast:
6731 '``addrspacecast .. to``' Instruction
6732 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6739 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
6744 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
6745 address space ``n`` to type ``pty2`` in address space ``m``.
6750 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
6751 to cast and a pointer type to cast it to, which must have a different
6757 The '``addrspacecast``' instruction converts the pointer value
6758 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
6759 value modification, depending on the target and the address space
6760 pair. Pointer conversions within the same address space must be
6761 performed with the ``bitcast`` instruction. Note that if the address space
6762 conversion is legal then both result and operand refer to the same memory
6768 .. code-block:: llvm
6770 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
6771 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
6772 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
6779 The instructions in this category are the "miscellaneous" instructions,
6780 which defy better classification.
6784 '``icmp``' Instruction
6785 ^^^^^^^^^^^^^^^^^^^^^^
6792 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6797 The '``icmp``' instruction returns a boolean value or a vector of
6798 boolean values based on comparison of its two integer, integer vector,
6799 pointer, or pointer vector operands.
6804 The '``icmp``' instruction takes three operands. The first operand is
6805 the condition code indicating the kind of comparison to perform. It is
6806 not a value, just a keyword. The possible condition code are:
6809 #. ``ne``: not equal
6810 #. ``ugt``: unsigned greater than
6811 #. ``uge``: unsigned greater or equal
6812 #. ``ult``: unsigned less than
6813 #. ``ule``: unsigned less or equal
6814 #. ``sgt``: signed greater than
6815 #. ``sge``: signed greater or equal
6816 #. ``slt``: signed less than
6817 #. ``sle``: signed less or equal
6819 The remaining two arguments must be :ref:`integer <t_integer>` or
6820 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
6821 must also be identical types.
6826 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
6827 code given as ``cond``. The comparison performed always yields either an
6828 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
6830 #. ``eq``: yields ``true`` if the operands are equal, ``false``
6831 otherwise. No sign interpretation is necessary or performed.
6832 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
6833 otherwise. No sign interpretation is necessary or performed.
6834 #. ``ugt``: interprets the operands as unsigned values and yields
6835 ``true`` if ``op1`` is greater than ``op2``.
6836 #. ``uge``: interprets the operands as unsigned values and yields
6837 ``true`` if ``op1`` is greater than or equal to ``op2``.
6838 #. ``ult``: interprets the operands as unsigned values and yields
6839 ``true`` if ``op1`` is less than ``op2``.
6840 #. ``ule``: interprets the operands as unsigned values and yields
6841 ``true`` if ``op1`` is less than or equal to ``op2``.
6842 #. ``sgt``: interprets the operands as signed values and yields ``true``
6843 if ``op1`` is greater than ``op2``.
6844 #. ``sge``: interprets the operands as signed values and yields ``true``
6845 if ``op1`` is greater than or equal to ``op2``.
6846 #. ``slt``: interprets the operands as signed values and yields ``true``
6847 if ``op1`` is less than ``op2``.
6848 #. ``sle``: interprets the operands as signed values and yields ``true``
6849 if ``op1`` is less than or equal to ``op2``.
6851 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
6852 are compared as if they were integers.
6854 If the operands are integer vectors, then they are compared element by
6855 element. The result is an ``i1`` vector with the same number of elements
6856 as the values being compared. Otherwise, the result is an ``i1``.
6861 .. code-block:: llvm
6863 <result> = icmp eq i32 4, 5 ; yields: result=false
6864 <result> = icmp ne float* %X, %X ; yields: result=false
6865 <result> = icmp ult i16 4, 5 ; yields: result=true
6866 <result> = icmp sgt i16 4, 5 ; yields: result=false
6867 <result> = icmp ule i16 -4, 5 ; yields: result=false
6868 <result> = icmp sge i16 4, 5 ; yields: result=false
6870 Note that the code generator does not yet support vector types with the
6871 ``icmp`` instruction.
6875 '``fcmp``' Instruction
6876 ^^^^^^^^^^^^^^^^^^^^^^
6883 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6888 The '``fcmp``' instruction returns a boolean value or vector of boolean
6889 values based on comparison of its operands.
6891 If the operands are floating point scalars, then the result type is a
6892 boolean (:ref:`i1 <t_integer>`).
6894 If the operands are floating point vectors, then the result type is a
6895 vector of boolean with the same number of elements as the operands being
6901 The '``fcmp``' instruction takes three operands. The first operand is
6902 the condition code indicating the kind of comparison to perform. It is
6903 not a value, just a keyword. The possible condition code are:
6905 #. ``false``: no comparison, always returns false
6906 #. ``oeq``: ordered and equal
6907 #. ``ogt``: ordered and greater than
6908 #. ``oge``: ordered and greater than or equal
6909 #. ``olt``: ordered and less than
6910 #. ``ole``: ordered and less than or equal
6911 #. ``one``: ordered and not equal
6912 #. ``ord``: ordered (no nans)
6913 #. ``ueq``: unordered or equal
6914 #. ``ugt``: unordered or greater than
6915 #. ``uge``: unordered or greater than or equal
6916 #. ``ult``: unordered or less than
6917 #. ``ule``: unordered or less than or equal
6918 #. ``une``: unordered or not equal
6919 #. ``uno``: unordered (either nans)
6920 #. ``true``: no comparison, always returns true
6922 *Ordered* means that neither operand is a QNAN while *unordered* means
6923 that either operand may be a QNAN.
6925 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6926 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6927 type. They must have identical types.
6932 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6933 condition code given as ``cond``. If the operands are vectors, then the
6934 vectors are compared element by element. Each comparison performed
6935 always yields an :ref:`i1 <t_integer>` result, as follows:
6937 #. ``false``: always yields ``false``, regardless of operands.
6938 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6939 is equal to ``op2``.
6940 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6941 is greater than ``op2``.
6942 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6943 is greater than or equal to ``op2``.
6944 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6945 is less than ``op2``.
6946 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6947 is less than or equal to ``op2``.
6948 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6949 is not equal to ``op2``.
6950 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6951 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6953 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6954 greater than ``op2``.
6955 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6956 greater than or equal to ``op2``.
6957 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6959 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6960 less than or equal to ``op2``.
6961 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6962 not equal to ``op2``.
6963 #. ``uno``: yields ``true`` if either operand is a QNAN.
6964 #. ``true``: always yields ``true``, regardless of operands.
6969 .. code-block:: llvm
6971 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6972 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6973 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6974 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6976 Note that the code generator does not yet support vector types with the
6977 ``fcmp`` instruction.
6981 '``phi``' Instruction
6982 ^^^^^^^^^^^^^^^^^^^^^
6989 <result> = phi <ty> [ <val0>, <label0>], ...
6994 The '``phi``' instruction is used to implement the φ node in the SSA
6995 graph representing the function.
7000 The type of the incoming values is specified with the first type field.
7001 After this, the '``phi``' instruction takes a list of pairs as
7002 arguments, with one pair for each predecessor basic block of the current
7003 block. Only values of :ref:`first class <t_firstclass>` type may be used as
7004 the value arguments to the PHI node. Only labels may be used as the
7007 There must be no non-phi instructions between the start of a basic block
7008 and the PHI instructions: i.e. PHI instructions must be first in a basic
7011 For the purposes of the SSA form, the use of each incoming value is
7012 deemed to occur on the edge from the corresponding predecessor block to
7013 the current block (but after any definition of an '``invoke``'
7014 instruction's return value on the same edge).
7019 At runtime, the '``phi``' instruction logically takes on the value
7020 specified by the pair corresponding to the predecessor basic block that
7021 executed just prior to the current block.
7026 .. code-block:: llvm
7028 Loop: ; Infinite loop that counts from 0 on up...
7029 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
7030 %nextindvar = add i32 %indvar, 1
7035 '``select``' Instruction
7036 ^^^^^^^^^^^^^^^^^^^^^^^^
7043 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
7045 selty is either i1 or {<N x i1>}
7050 The '``select``' instruction is used to choose one value based on a
7051 condition, without IR-level branching.
7056 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
7057 values indicating the condition, and two values of the same :ref:`first
7058 class <t_firstclass>` type.
7063 If the condition is an i1 and it evaluates to 1, the instruction returns
7064 the first value argument; otherwise, it returns the second value
7067 If the condition is a vector of i1, then the value arguments must be
7068 vectors of the same size, and the selection is done element by element.
7070 If the condition is an i1 and the value arguments are vectors of the
7071 same size, then an entire vector is selected.
7076 .. code-block:: llvm
7078 %X = select i1 true, i8 17, i8 42 ; yields i8:17
7082 '``call``' Instruction
7083 ^^^^^^^^^^^^^^^^^^^^^^
7090 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
7095 The '``call``' instruction represents a simple function call.
7100 This instruction requires several arguments:
7102 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
7103 should perform tail call optimization. The ``tail`` marker is a hint that
7104 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
7105 means that the call must be tail call optimized in order for the program to
7106 be correct. The ``musttail`` marker provides these guarantees:
7108 #. The call will not cause unbounded stack growth if it is part of a
7109 recursive cycle in the call graph.
7110 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
7113 Both markers imply that the callee does not access allocas or varargs from
7114 the caller. Calls marked ``musttail`` must obey the following additional
7117 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
7118 or a pointer bitcast followed by a ret instruction.
7119 - The ret instruction must return the (possibly bitcasted) value
7120 produced by the call or void.
7121 - The caller and callee prototypes must match. Pointer types of
7122 parameters or return types may differ in pointee type, but not
7124 - The calling conventions of the caller and callee must match.
7125 - All ABI-impacting function attributes, such as sret, byval, inreg,
7126 returned, and inalloca, must match.
7127 - The callee must be varargs iff the caller is varargs. Bitcasting a
7128 non-varargs function to the appropriate varargs type is legal so
7129 long as the non-varargs prefixes obey the other rules.
7131 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
7132 the following conditions are met:
7134 - Caller and callee both have the calling convention ``fastcc``.
7135 - The call is in tail position (ret immediately follows call and ret
7136 uses value of call or is void).
7137 - Option ``-tailcallopt`` is enabled, or
7138 ``llvm::GuaranteedTailCallOpt`` is ``true``.
7139 - `Platform-specific constraints are
7140 met. <CodeGenerator.html#tailcallopt>`_
7142 #. The optional "cconv" marker indicates which :ref:`calling
7143 convention <callingconv>` the call should use. If none is
7144 specified, the call defaults to using C calling conventions. The
7145 calling convention of the call must match the calling convention of
7146 the target function, or else the behavior is undefined.
7147 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
7148 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
7150 #. '``ty``': the type of the call instruction itself which is also the
7151 type of the return value. Functions that return no value are marked
7153 #. '``fnty``': shall be the signature of the pointer to function value
7154 being invoked. The argument types must match the types implied by
7155 this signature. This type can be omitted if the function is not
7156 varargs and if the function type does not return a pointer to a
7158 #. '``fnptrval``': An LLVM value containing a pointer to a function to
7159 be invoked. In most cases, this is a direct function invocation, but
7160 indirect ``call``'s are just as possible, calling an arbitrary pointer
7162 #. '``function args``': argument list whose types match the function
7163 signature argument types and parameter attributes. All arguments must
7164 be of :ref:`first class <t_firstclass>` type. If the function signature
7165 indicates the function accepts a variable number of arguments, the
7166 extra arguments can be specified.
7167 #. The optional :ref:`function attributes <fnattrs>` list. Only
7168 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
7169 attributes are valid here.
7174 The '``call``' instruction is used to cause control flow to transfer to
7175 a specified function, with its incoming arguments bound to the specified
7176 values. Upon a '``ret``' instruction in the called function, control
7177 flow continues with the instruction after the function call, and the
7178 return value of the function is bound to the result argument.
7183 .. code-block:: llvm
7185 %retval = call i32 @test(i32 %argc)
7186 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
7187 %X = tail call i32 @foo() ; yields i32
7188 %Y = tail call fastcc i32 @foo() ; yields i32
7189 call void %foo(i8 97 signext)
7191 %struct.A = type { i32, i8 }
7192 %r = call %struct.A @foo() ; yields { i32, i8 }
7193 %gr = extractvalue %struct.A %r, 0 ; yields i32
7194 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
7195 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
7196 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
7198 llvm treats calls to some functions with names and arguments that match
7199 the standard C99 library as being the C99 library functions, and may
7200 perform optimizations or generate code for them under that assumption.
7201 This is something we'd like to change in the future to provide better
7202 support for freestanding environments and non-C-based languages.
7206 '``va_arg``' Instruction
7207 ^^^^^^^^^^^^^^^^^^^^^^^^
7214 <resultval> = va_arg <va_list*> <arglist>, <argty>
7219 The '``va_arg``' instruction is used to access arguments passed through
7220 the "variable argument" area of a function call. It is used to implement
7221 the ``va_arg`` macro in C.
7226 This instruction takes a ``va_list*`` value and the type of the
7227 argument. It returns a value of the specified argument type and
7228 increments the ``va_list`` to point to the next argument. The actual
7229 type of ``va_list`` is target specific.
7234 The '``va_arg``' instruction loads an argument of the specified type
7235 from the specified ``va_list`` and causes the ``va_list`` to point to
7236 the next argument. For more information, see the variable argument
7237 handling :ref:`Intrinsic Functions <int_varargs>`.
7239 It is legal for this instruction to be called in a function which does
7240 not take a variable number of arguments, for example, the ``vfprintf``
7243 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
7244 function <intrinsics>` because it takes a type as an argument.
7249 See the :ref:`variable argument processing <int_varargs>` section.
7251 Note that the code generator does not yet fully support va\_arg on many
7252 targets. Also, it does not currently support va\_arg with aggregate
7253 types on any target.
7257 '``landingpad``' Instruction
7258 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7265 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
7266 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
7268 <clause> := catch <type> <value>
7269 <clause> := filter <array constant type> <array constant>
7274 The '``landingpad``' instruction is used by `LLVM's exception handling
7275 system <ExceptionHandling.html#overview>`_ to specify that a basic block
7276 is a landing pad --- one where the exception lands, and corresponds to the
7277 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
7278 defines values supplied by the personality function (``pers_fn``) upon
7279 re-entry to the function. The ``resultval`` has the type ``resultty``.
7284 This instruction takes a ``pers_fn`` value. This is the personality
7285 function associated with the unwinding mechanism. The optional
7286 ``cleanup`` flag indicates that the landing pad block is a cleanup.
7288 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
7289 contains the global variable representing the "type" that may be caught
7290 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
7291 clause takes an array constant as its argument. Use
7292 "``[0 x i8**] undef``" for a filter which cannot throw. The
7293 '``landingpad``' instruction must contain *at least* one ``clause`` or
7294 the ``cleanup`` flag.
7299 The '``landingpad``' instruction defines the values which are set by the
7300 personality function (``pers_fn``) upon re-entry to the function, and
7301 therefore the "result type" of the ``landingpad`` instruction. As with
7302 calling conventions, how the personality function results are
7303 represented in LLVM IR is target specific.
7305 The clauses are applied in order from top to bottom. If two
7306 ``landingpad`` instructions are merged together through inlining, the
7307 clauses from the calling function are appended to the list of clauses.
7308 When the call stack is being unwound due to an exception being thrown,
7309 the exception is compared against each ``clause`` in turn. If it doesn't
7310 match any of the clauses, and the ``cleanup`` flag is not set, then
7311 unwinding continues further up the call stack.
7313 The ``landingpad`` instruction has several restrictions:
7315 - A landing pad block is a basic block which is the unwind destination
7316 of an '``invoke``' instruction.
7317 - A landing pad block must have a '``landingpad``' instruction as its
7318 first non-PHI instruction.
7319 - There can be only one '``landingpad``' instruction within the landing
7321 - A basic block that is not a landing pad block may not include a
7322 '``landingpad``' instruction.
7323 - All '``landingpad``' instructions in a function must have the same
7324 personality function.
7329 .. code-block:: llvm
7331 ;; A landing pad which can catch an integer.
7332 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7334 ;; A landing pad that is a cleanup.
7335 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7337 ;; A landing pad which can catch an integer and can only throw a double.
7338 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7340 filter [1 x i8**] [@_ZTId]
7347 LLVM supports the notion of an "intrinsic function". These functions
7348 have well known names and semantics and are required to follow certain
7349 restrictions. Overall, these intrinsics represent an extension mechanism
7350 for the LLVM language that does not require changing all of the
7351 transformations in LLVM when adding to the language (or the bitcode
7352 reader/writer, the parser, etc...).
7354 Intrinsic function names must all start with an "``llvm.``" prefix. This
7355 prefix is reserved in LLVM for intrinsic names; thus, function names may
7356 not begin with this prefix. Intrinsic functions must always be external
7357 functions: you cannot define the body of intrinsic functions. Intrinsic
7358 functions may only be used in call or invoke instructions: it is illegal
7359 to take the address of an intrinsic function. Additionally, because
7360 intrinsic functions are part of the LLVM language, it is required if any
7361 are added that they be documented here.
7363 Some intrinsic functions can be overloaded, i.e., the intrinsic
7364 represents a family of functions that perform the same operation but on
7365 different data types. Because LLVM can represent over 8 million
7366 different integer types, overloading is used commonly to allow an
7367 intrinsic function to operate on any integer type. One or more of the
7368 argument types or the result type can be overloaded to accept any
7369 integer type. Argument types may also be defined as exactly matching a
7370 previous argument's type or the result type. This allows an intrinsic
7371 function which accepts multiple arguments, but needs all of them to be
7372 of the same type, to only be overloaded with respect to a single
7373 argument or the result.
7375 Overloaded intrinsics will have the names of its overloaded argument
7376 types encoded into its function name, each preceded by a period. Only
7377 those types which are overloaded result in a name suffix. Arguments
7378 whose type is matched against another type do not. For example, the
7379 ``llvm.ctpop`` function can take an integer of any width and returns an
7380 integer of exactly the same integer width. This leads to a family of
7381 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
7382 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
7383 overloaded, and only one type suffix is required. Because the argument's
7384 type is matched against the return type, it does not require its own
7387 To learn how to add an intrinsic function, please see the `Extending
7388 LLVM Guide <ExtendingLLVM.html>`_.
7392 Variable Argument Handling Intrinsics
7393 -------------------------------------
7395 Variable argument support is defined in LLVM with the
7396 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
7397 functions. These functions are related to the similarly named macros
7398 defined in the ``<stdarg.h>`` header file.
7400 All of these functions operate on arguments that use a target-specific
7401 value type "``va_list``". The LLVM assembly language reference manual
7402 does not define what this type is, so all transformations should be
7403 prepared to handle these functions regardless of the type used.
7405 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
7406 variable argument handling intrinsic functions are used.
7408 .. code-block:: llvm
7410 ; This struct is different for every platform. For most platforms,
7411 ; it is merely an i8*.
7412 %struct.va_list = type { i8* }
7414 ; For Unix x86_64 platforms, va_list is the following struct:
7415 ; %struct.va_list = type { i32, i32, i8*, i8* }
7417 define i32 @test(i32 %X, ...) {
7418 ; Initialize variable argument processing
7419 %ap = alloca %struct.va_list
7420 %ap2 = bitcast %struct.va_list* %ap to i8*
7421 call void @llvm.va_start(i8* %ap2)
7423 ; Read a single integer argument
7424 %tmp = va_arg i8* %ap2, i32
7426 ; Demonstrate usage of llvm.va_copy and llvm.va_end
7428 %aq2 = bitcast i8** %aq to i8*
7429 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
7430 call void @llvm.va_end(i8* %aq2)
7432 ; Stop processing of arguments.
7433 call void @llvm.va_end(i8* %ap2)
7437 declare void @llvm.va_start(i8*)
7438 declare void @llvm.va_copy(i8*, i8*)
7439 declare void @llvm.va_end(i8*)
7443 '``llvm.va_start``' Intrinsic
7444 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7451 declare void @llvm.va_start(i8* <arglist>)
7456 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
7457 subsequent use by ``va_arg``.
7462 The argument is a pointer to a ``va_list`` element to initialize.
7467 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
7468 available in C. In a target-dependent way, it initializes the
7469 ``va_list`` element to which the argument points, so that the next call
7470 to ``va_arg`` will produce the first variable argument passed to the
7471 function. Unlike the C ``va_start`` macro, this intrinsic does not need
7472 to know the last argument of the function as the compiler can figure
7475 '``llvm.va_end``' Intrinsic
7476 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7483 declare void @llvm.va_end(i8* <arglist>)
7488 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
7489 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
7494 The argument is a pointer to a ``va_list`` to destroy.
7499 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
7500 available in C. In a target-dependent way, it destroys the ``va_list``
7501 element to which the argument points. Calls to
7502 :ref:`llvm.va_start <int_va_start>` and
7503 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
7508 '``llvm.va_copy``' Intrinsic
7509 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7516 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
7521 The '``llvm.va_copy``' intrinsic copies the current argument position
7522 from the source argument list to the destination argument list.
7527 The first argument is a pointer to a ``va_list`` element to initialize.
7528 The second argument is a pointer to a ``va_list`` element to copy from.
7533 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
7534 available in C. In a target-dependent way, it copies the source
7535 ``va_list`` element into the destination ``va_list`` element. This
7536 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
7537 arbitrarily complex and require, for example, memory allocation.
7539 Accurate Garbage Collection Intrinsics
7540 --------------------------------------
7542 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
7543 (GC) requires the frontend to generate code containing appropriate intrinsic
7544 calls and select an appropriate GC strategy which knows how to lower these
7545 intrinsics in a manner which is appropriate for the target collector.
7547 These intrinsics allow identification of :ref:`GC roots on the
7548 stack <int_gcroot>`, as well as garbage collector implementations that
7549 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
7550 Frontends for type-safe garbage collected languages should generate
7551 these intrinsics to make use of the LLVM garbage collectors. For more
7552 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
7554 Experimental Statepoint Intrinsics
7555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7557 LLVM provides an second experimental set of intrinsics for describing garbage
7558 collection safepoints in compiled code. These intrinsics are an alternative
7559 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
7560 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
7561 differences in approach are covered in the `Garbage Collection with LLVM
7562 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
7563 described in :doc:`Statepoints`.
7567 '``llvm.gcroot``' Intrinsic
7568 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7575 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
7580 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
7581 the code generator, and allows some metadata to be associated with it.
7586 The first argument specifies the address of a stack object that contains
7587 the root pointer. The second pointer (which must be either a constant or
7588 a global value address) contains the meta-data to be associated with the
7594 At runtime, a call to this intrinsic stores a null pointer into the
7595 "ptrloc" location. At compile-time, the code generator generates
7596 information to allow the runtime to find the pointer at GC safe points.
7597 The '``llvm.gcroot``' intrinsic may only be used in a function which
7598 :ref:`specifies a GC algorithm <gc>`.
7602 '``llvm.gcread``' Intrinsic
7603 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7610 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
7615 The '``llvm.gcread``' intrinsic identifies reads of references from heap
7616 locations, allowing garbage collector implementations that require read
7622 The second argument is the address to read from, which should be an
7623 address allocated from the garbage collector. The first object is a
7624 pointer to the start of the referenced object, if needed by the language
7625 runtime (otherwise null).
7630 The '``llvm.gcread``' intrinsic has the same semantics as a load
7631 instruction, but may be replaced with substantially more complex code by
7632 the garbage collector runtime, as needed. The '``llvm.gcread``'
7633 intrinsic may only be used in a function which :ref:`specifies a GC
7638 '``llvm.gcwrite``' Intrinsic
7639 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7646 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
7651 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
7652 locations, allowing garbage collector implementations that require write
7653 barriers (such as generational or reference counting collectors).
7658 The first argument is the reference to store, the second is the start of
7659 the object to store it to, and the third is the address of the field of
7660 Obj to store to. If the runtime does not require a pointer to the
7661 object, Obj may be null.
7666 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
7667 instruction, but may be replaced with substantially more complex code by
7668 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
7669 intrinsic may only be used in a function which :ref:`specifies a GC
7672 Code Generator Intrinsics
7673 -------------------------
7675 These intrinsics are provided by LLVM to expose special features that
7676 may only be implemented with code generator support.
7678 '``llvm.returnaddress``' Intrinsic
7679 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7686 declare i8 *@llvm.returnaddress(i32 <level>)
7691 The '``llvm.returnaddress``' intrinsic attempts to compute a
7692 target-specific value indicating the return address of the current
7693 function or one of its callers.
7698 The argument to this intrinsic indicates which function to return the
7699 address for. Zero indicates the calling function, one indicates its
7700 caller, etc. The argument is **required** to be a constant integer
7706 The '``llvm.returnaddress``' intrinsic either returns a pointer
7707 indicating the return address of the specified call frame, or zero if it
7708 cannot be identified. The value returned by this intrinsic is likely to
7709 be incorrect or 0 for arguments other than zero, so it should only be
7710 used for debugging purposes.
7712 Note that calling this intrinsic does not prevent function inlining or
7713 other aggressive transformations, so the value returned may not be that
7714 of the obvious source-language caller.
7716 '``llvm.frameaddress``' Intrinsic
7717 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7724 declare i8* @llvm.frameaddress(i32 <level>)
7729 The '``llvm.frameaddress``' intrinsic attempts to return the
7730 target-specific frame pointer value for the specified stack frame.
7735 The argument to this intrinsic indicates which function to return the
7736 frame pointer for. Zero indicates the calling function, one indicates
7737 its caller, etc. The argument is **required** to be a constant integer
7743 The '``llvm.frameaddress``' intrinsic either returns a pointer
7744 indicating the frame address of the specified call frame, or zero if it
7745 cannot be identified. The value returned by this intrinsic is likely to
7746 be incorrect or 0 for arguments other than zero, so it should only be
7747 used for debugging purposes.
7749 Note that calling this intrinsic does not prevent function inlining or
7750 other aggressive transformations, so the value returned may not be that
7751 of the obvious source-language caller.
7753 '``llvm.frameescape``' and '``llvm.framerecover``' Intrinsics
7754 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7761 declare void @llvm.frameescape(...)
7762 declare i8* @llvm.framerecover(i8* %func, i8* %fp, i32 %idx)
7767 The '``llvm.frameescape``' intrinsic escapes offsets of a collection of static
7768 allocas, and the '``llvm.framerecover``' intrinsic applies those offsets to a
7769 live frame pointer to recover the address of the allocation. The offset is
7770 computed during frame layout of the caller of ``llvm.frameescape``.
7775 All arguments to '``llvm.frameescape``' must be pointers to static allocas or
7776 casts of static allocas. Each function can only call '``llvm.frameescape``'
7777 once, and it can only do so from the entry block.
7779 The ``func`` argument to '``llvm.framerecover``' must be a constant
7780 bitcasted pointer to a function defined in the current module. The code
7781 generator cannot determine the frame allocation offset of functions defined in
7784 The ``fp`` argument to '``llvm.framerecover``' must be a frame
7785 pointer of a call frame that is currently live. The return value of
7786 '``llvm.frameaddress``' is one way to produce such a value, but most platforms
7787 also expose the frame pointer through stack unwinding mechanisms.
7789 The ``idx`` argument to '``llvm.framerecover``' indicates which alloca passed to
7790 '``llvm.frameescape``' to recover. It is zero-indexed.
7795 These intrinsics allow a group of functions to access one stack memory
7796 allocation in an ancestor stack frame. The memory returned from
7797 '``llvm.frameallocate``' may be allocated prior to stack realignment, so the
7798 memory is only aligned to the ABI-required stack alignment. Each function may
7799 only call '``llvm.frameallocate``' one or zero times from the function entry
7800 block. The frame allocation intrinsic inhibits inlining, as any frame
7801 allocations in the inlined function frame are likely to be at a different
7802 offset from the one used by '``llvm.framerecover``' called with the
7805 .. _int_read_register:
7806 .. _int_write_register:
7808 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
7809 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7816 declare i32 @llvm.read_register.i32(metadata)
7817 declare i64 @llvm.read_register.i64(metadata)
7818 declare void @llvm.write_register.i32(metadata, i32 @value)
7819 declare void @llvm.write_register.i64(metadata, i64 @value)
7825 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
7826 provides access to the named register. The register must be valid on
7827 the architecture being compiled to. The type needs to be compatible
7828 with the register being read.
7833 The '``llvm.read_register``' intrinsic returns the current value of the
7834 register, where possible. The '``llvm.write_register``' intrinsic sets
7835 the current value of the register, where possible.
7837 This is useful to implement named register global variables that need
7838 to always be mapped to a specific register, as is common practice on
7839 bare-metal programs including OS kernels.
7841 The compiler doesn't check for register availability or use of the used
7842 register in surrounding code, including inline assembly. Because of that,
7843 allocatable registers are not supported.
7845 Warning: So far it only works with the stack pointer on selected
7846 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
7847 work is needed to support other registers and even more so, allocatable
7852 '``llvm.stacksave``' Intrinsic
7853 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7860 declare i8* @llvm.stacksave()
7865 The '``llvm.stacksave``' intrinsic is used to remember the current state
7866 of the function stack, for use with
7867 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
7868 implementing language features like scoped automatic variable sized
7874 This intrinsic returns a opaque pointer value that can be passed to
7875 :ref:`llvm.stackrestore <int_stackrestore>`. When an
7876 ``llvm.stackrestore`` intrinsic is executed with a value saved from
7877 ``llvm.stacksave``, it effectively restores the state of the stack to
7878 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
7879 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
7880 were allocated after the ``llvm.stacksave`` was executed.
7882 .. _int_stackrestore:
7884 '``llvm.stackrestore``' Intrinsic
7885 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7892 declare void @llvm.stackrestore(i8* %ptr)
7897 The '``llvm.stackrestore``' intrinsic is used to restore the state of
7898 the function stack to the state it was in when the corresponding
7899 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
7900 useful for implementing language features like scoped automatic variable
7901 sized arrays in C99.
7906 See the description for :ref:`llvm.stacksave <int_stacksave>`.
7908 '``llvm.prefetch``' Intrinsic
7909 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7916 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7921 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
7922 insert a prefetch instruction if supported; otherwise, it is a noop.
7923 Prefetches have no effect on the behavior of the program but can change
7924 its performance characteristics.
7929 ``address`` is the address to be prefetched, ``rw`` is the specifier
7930 determining if the fetch should be for a read (0) or write (1), and
7931 ``locality`` is a temporal locality specifier ranging from (0) - no
7932 locality, to (3) - extremely local keep in cache. The ``cache type``
7933 specifies whether the prefetch is performed on the data (1) or
7934 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
7935 arguments must be constant integers.
7940 This intrinsic does not modify the behavior of the program. In
7941 particular, prefetches cannot trap and do not produce a value. On
7942 targets that support this intrinsic, the prefetch can provide hints to
7943 the processor cache for better performance.
7945 '``llvm.pcmarker``' Intrinsic
7946 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7953 declare void @llvm.pcmarker(i32 <id>)
7958 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7959 Counter (PC) in a region of code to simulators and other tools. The
7960 method is target specific, but it is expected that the marker will use
7961 exported symbols to transmit the PC of the marker. The marker makes no
7962 guarantees that it will remain with any specific instruction after
7963 optimizations. It is possible that the presence of a marker will inhibit
7964 optimizations. The intended use is to be inserted after optimizations to
7965 allow correlations of simulation runs.
7970 ``id`` is a numerical id identifying the marker.
7975 This intrinsic does not modify the behavior of the program. Backends
7976 that do not support this intrinsic may ignore it.
7978 '``llvm.readcyclecounter``' Intrinsic
7979 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7986 declare i64 @llvm.readcyclecounter()
7991 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
7992 counter register (or similar low latency, high accuracy clocks) on those
7993 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7994 should map to RPCC. As the backing counters overflow quickly (on the
7995 order of 9 seconds on alpha), this should only be used for small
8001 When directly supported, reading the cycle counter should not modify any
8002 memory. Implementations are allowed to either return a application
8003 specific value or a system wide value. On backends without support, this
8004 is lowered to a constant 0.
8006 Note that runtime support may be conditional on the privilege-level code is
8007 running at and the host platform.
8009 '``llvm.clear_cache``' Intrinsic
8010 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8017 declare void @llvm.clear_cache(i8*, i8*)
8022 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
8023 in the specified range to the execution unit of the processor. On
8024 targets with non-unified instruction and data cache, the implementation
8025 flushes the instruction cache.
8030 On platforms with coherent instruction and data caches (e.g. x86), this
8031 intrinsic is a nop. On platforms with non-coherent instruction and data
8032 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
8033 instructions or a system call, if cache flushing requires special
8036 The default behavior is to emit a call to ``__clear_cache`` from the run
8039 This instrinsic does *not* empty the instruction pipeline. Modifications
8040 of the current function are outside the scope of the intrinsic.
8042 '``llvm.instrprof_increment``' Intrinsic
8043 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8050 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
8051 i32 <num-counters>, i32 <index>)
8056 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
8057 frontend for use with instrumentation based profiling. These will be
8058 lowered by the ``-instrprof`` pass to generate execution counts of a
8064 The first argument is a pointer to a global variable containing the
8065 name of the entity being instrumented. This should generally be the
8066 (mangled) function name for a set of counters.
8068 The second argument is a hash value that can be used by the consumer
8069 of the profile data to detect changes to the instrumented source, and
8070 the third is the number of counters associated with ``name``. It is an
8071 error if ``hash`` or ``num-counters`` differ between two instances of
8072 ``instrprof_increment`` that refer to the same name.
8074 The last argument refers to which of the counters for ``name`` should
8075 be incremented. It should be a value between 0 and ``num-counters``.
8080 This intrinsic represents an increment of a profiling counter. It will
8081 cause the ``-instrprof`` pass to generate the appropriate data
8082 structures and the code to increment the appropriate value, in a
8083 format that can be written out by a compiler runtime and consumed via
8084 the ``llvm-profdata`` tool.
8086 Standard C Library Intrinsics
8087 -----------------------------
8089 LLVM provides intrinsics for a few important standard C library
8090 functions. These intrinsics allow source-language front-ends to pass
8091 information about the alignment of the pointer arguments to the code
8092 generator, providing opportunity for more efficient code generation.
8096 '``llvm.memcpy``' Intrinsic
8097 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8102 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
8103 integer bit width and for different address spaces. Not all targets
8104 support all bit widths however.
8108 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
8109 i32 <len>, i32 <align>, i1 <isvolatile>)
8110 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
8111 i64 <len>, i32 <align>, i1 <isvolatile>)
8116 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
8117 source location to the destination location.
8119 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
8120 intrinsics do not return a value, takes extra alignment/isvolatile
8121 arguments and the pointers can be in specified address spaces.
8126 The first argument is a pointer to the destination, the second is a
8127 pointer to the source. The third argument is an integer argument
8128 specifying the number of bytes to copy, the fourth argument is the
8129 alignment of the source and destination locations, and the fifth is a
8130 boolean indicating a volatile access.
8132 If the call to this intrinsic has an alignment value that is not 0 or 1,
8133 then the caller guarantees that both the source and destination pointers
8134 are aligned to that boundary.
8136 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
8137 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
8138 very cleanly specified and it is unwise to depend on it.
8143 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
8144 source location to the destination location, which are not allowed to
8145 overlap. It copies "len" bytes of memory over. If the argument is known
8146 to be aligned to some boundary, this can be specified as the fourth
8147 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
8149 '``llvm.memmove``' Intrinsic
8150 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8155 This is an overloaded intrinsic. You can use llvm.memmove on any integer
8156 bit width and for different address space. Not all targets support all
8161 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
8162 i32 <len>, i32 <align>, i1 <isvolatile>)
8163 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
8164 i64 <len>, i32 <align>, i1 <isvolatile>)
8169 The '``llvm.memmove.*``' intrinsics move a block of memory from the
8170 source location to the destination location. It is similar to the
8171 '``llvm.memcpy``' intrinsic but allows the two memory locations to
8174 Note that, unlike the standard libc function, the ``llvm.memmove.*``
8175 intrinsics do not return a value, takes extra alignment/isvolatile
8176 arguments and the pointers can be in specified address spaces.
8181 The first argument is a pointer to the destination, the second is a
8182 pointer to the source. The third argument is an integer argument
8183 specifying the number of bytes to copy, the fourth argument is the
8184 alignment of the source and destination locations, and the fifth is a
8185 boolean indicating a volatile access.
8187 If the call to this intrinsic has an alignment value that is not 0 or 1,
8188 then the caller guarantees that the source and destination pointers are
8189 aligned to that boundary.
8191 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
8192 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
8193 not very cleanly specified and it is unwise to depend on it.
8198 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
8199 source location to the destination location, which may overlap. It
8200 copies "len" bytes of memory over. If the argument is known to be
8201 aligned to some boundary, this can be specified as the fourth argument,
8202 otherwise it should be set to 0 or 1 (both meaning no alignment).
8204 '``llvm.memset.*``' Intrinsics
8205 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8210 This is an overloaded intrinsic. You can use llvm.memset on any integer
8211 bit width and for different address spaces. However, not all targets
8212 support all bit widths.
8216 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
8217 i32 <len>, i32 <align>, i1 <isvolatile>)
8218 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
8219 i64 <len>, i32 <align>, i1 <isvolatile>)
8224 The '``llvm.memset.*``' intrinsics fill a block of memory with a
8225 particular byte value.
8227 Note that, unlike the standard libc function, the ``llvm.memset``
8228 intrinsic does not return a value and takes extra alignment/volatile
8229 arguments. Also, the destination can be in an arbitrary address space.
8234 The first argument is a pointer to the destination to fill, the second
8235 is the byte value with which to fill it, the third argument is an
8236 integer argument specifying the number of bytes to fill, and the fourth
8237 argument is the known alignment of the destination location.
8239 If the call to this intrinsic has an alignment value that is not 0 or 1,
8240 then the caller guarantees that the destination pointer is aligned to
8243 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
8244 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
8245 very cleanly specified and it is unwise to depend on it.
8250 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
8251 at the destination location. If the argument is known to be aligned to
8252 some boundary, this can be specified as the fourth argument, otherwise
8253 it should be set to 0 or 1 (both meaning no alignment).
8255 '``llvm.sqrt.*``' Intrinsic
8256 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8261 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
8262 floating point or vector of floating point type. Not all targets support
8267 declare float @llvm.sqrt.f32(float %Val)
8268 declare double @llvm.sqrt.f64(double %Val)
8269 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
8270 declare fp128 @llvm.sqrt.f128(fp128 %Val)
8271 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
8276 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
8277 returning the same value as the libm '``sqrt``' functions would. Unlike
8278 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
8279 negative numbers other than -0.0 (which allows for better optimization,
8280 because there is no need to worry about errno being set).
8281 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
8286 The argument and return value are floating point numbers of the same
8292 This function returns the sqrt of the specified operand if it is a
8293 nonnegative floating point number.
8295 '``llvm.powi.*``' Intrinsic
8296 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8301 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
8302 floating point or vector of floating point type. Not all targets support
8307 declare float @llvm.powi.f32(float %Val, i32 %power)
8308 declare double @llvm.powi.f64(double %Val, i32 %power)
8309 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
8310 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
8311 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
8316 The '``llvm.powi.*``' intrinsics return the first operand raised to the
8317 specified (positive or negative) power. The order of evaluation of
8318 multiplications is not defined. When a vector of floating point type is
8319 used, the second argument remains a scalar integer value.
8324 The second argument is an integer power, and the first is a value to
8325 raise to that power.
8330 This function returns the first value raised to the second power with an
8331 unspecified sequence of rounding operations.
8333 '``llvm.sin.*``' Intrinsic
8334 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8339 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
8340 floating point or vector of floating point type. Not all targets support
8345 declare float @llvm.sin.f32(float %Val)
8346 declare double @llvm.sin.f64(double %Val)
8347 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
8348 declare fp128 @llvm.sin.f128(fp128 %Val)
8349 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
8354 The '``llvm.sin.*``' intrinsics return the sine of the operand.
8359 The argument and return value are floating point numbers of the same
8365 This function returns the sine of the specified operand, returning the
8366 same values as the libm ``sin`` functions would, and handles error
8367 conditions in the same way.
8369 '``llvm.cos.*``' Intrinsic
8370 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8375 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
8376 floating point or vector of floating point type. Not all targets support
8381 declare float @llvm.cos.f32(float %Val)
8382 declare double @llvm.cos.f64(double %Val)
8383 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
8384 declare fp128 @llvm.cos.f128(fp128 %Val)
8385 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
8390 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
8395 The argument and return value are floating point numbers of the same
8401 This function returns the cosine of the specified operand, returning the
8402 same values as the libm ``cos`` functions would, and handles error
8403 conditions in the same way.
8405 '``llvm.pow.*``' Intrinsic
8406 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8411 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
8412 floating point or vector of floating point type. Not all targets support
8417 declare float @llvm.pow.f32(float %Val, float %Power)
8418 declare double @llvm.pow.f64(double %Val, double %Power)
8419 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
8420 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
8421 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
8426 The '``llvm.pow.*``' intrinsics return the first operand raised to the
8427 specified (positive or negative) power.
8432 The second argument is a floating point power, and the first is a value
8433 to raise to that power.
8438 This function returns the first value raised to the second power,
8439 returning the same values as the libm ``pow`` functions would, and
8440 handles error conditions in the same way.
8442 '``llvm.exp.*``' Intrinsic
8443 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8448 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
8449 floating point or vector of floating point type. Not all targets support
8454 declare float @llvm.exp.f32(float %Val)
8455 declare double @llvm.exp.f64(double %Val)
8456 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
8457 declare fp128 @llvm.exp.f128(fp128 %Val)
8458 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
8463 The '``llvm.exp.*``' intrinsics perform the exp function.
8468 The argument and return value are floating point numbers of the same
8474 This function returns the same values as the libm ``exp`` functions
8475 would, and handles error conditions in the same way.
8477 '``llvm.exp2.*``' Intrinsic
8478 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8483 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
8484 floating point or vector of floating point type. Not all targets support
8489 declare float @llvm.exp2.f32(float %Val)
8490 declare double @llvm.exp2.f64(double %Val)
8491 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
8492 declare fp128 @llvm.exp2.f128(fp128 %Val)
8493 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
8498 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
8503 The argument and return value are floating point numbers of the same
8509 This function returns the same values as the libm ``exp2`` functions
8510 would, and handles error conditions in the same way.
8512 '``llvm.log.*``' Intrinsic
8513 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8518 This is an overloaded intrinsic. You can use ``llvm.log`` on any
8519 floating point or vector of floating point type. Not all targets support
8524 declare float @llvm.log.f32(float %Val)
8525 declare double @llvm.log.f64(double %Val)
8526 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
8527 declare fp128 @llvm.log.f128(fp128 %Val)
8528 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
8533 The '``llvm.log.*``' intrinsics perform the log function.
8538 The argument and return value are floating point numbers of the same
8544 This function returns the same values as the libm ``log`` functions
8545 would, and handles error conditions in the same way.
8547 '``llvm.log10.*``' Intrinsic
8548 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8553 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
8554 floating point or vector of floating point type. Not all targets support
8559 declare float @llvm.log10.f32(float %Val)
8560 declare double @llvm.log10.f64(double %Val)
8561 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
8562 declare fp128 @llvm.log10.f128(fp128 %Val)
8563 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
8568 The '``llvm.log10.*``' intrinsics perform the log10 function.
8573 The argument and return value are floating point numbers of the same
8579 This function returns the same values as the libm ``log10`` functions
8580 would, and handles error conditions in the same way.
8582 '``llvm.log2.*``' Intrinsic
8583 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8588 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
8589 floating point or vector of floating point type. Not all targets support
8594 declare float @llvm.log2.f32(float %Val)
8595 declare double @llvm.log2.f64(double %Val)
8596 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
8597 declare fp128 @llvm.log2.f128(fp128 %Val)
8598 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
8603 The '``llvm.log2.*``' intrinsics perform the log2 function.
8608 The argument and return value are floating point numbers of the same
8614 This function returns the same values as the libm ``log2`` functions
8615 would, and handles error conditions in the same way.
8617 '``llvm.fma.*``' Intrinsic
8618 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8623 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
8624 floating point or vector of floating point type. Not all targets support
8629 declare float @llvm.fma.f32(float %a, float %b, float %c)
8630 declare double @llvm.fma.f64(double %a, double %b, double %c)
8631 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
8632 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
8633 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
8638 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
8644 The argument and return value are floating point numbers of the same
8650 This function returns the same values as the libm ``fma`` functions
8651 would, and does not set errno.
8653 '``llvm.fabs.*``' Intrinsic
8654 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8659 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
8660 floating point or vector of floating point type. Not all targets support
8665 declare float @llvm.fabs.f32(float %Val)
8666 declare double @llvm.fabs.f64(double %Val)
8667 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
8668 declare fp128 @llvm.fabs.f128(fp128 %Val)
8669 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
8674 The '``llvm.fabs.*``' intrinsics return the absolute value of the
8680 The argument and return value are floating point numbers of the same
8686 This function returns the same values as the libm ``fabs`` functions
8687 would, and handles error conditions in the same way.
8689 '``llvm.minnum.*``' Intrinsic
8690 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8695 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
8696 floating point or vector of floating point type. Not all targets support
8701 declare float @llvm.minnum.f32(float %Val0, float %Val1)
8702 declare double @llvm.minnum.f64(double %Val0, double %Val1)
8703 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8704 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
8705 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8710 The '``llvm.minnum.*``' intrinsics return the minimum of the two
8717 The arguments and return value are floating point numbers of the same
8723 Follows the IEEE-754 semantics for minNum, which also match for libm's
8726 If either operand is a NaN, returns the other non-NaN operand. Returns
8727 NaN only if both operands are NaN. If the operands compare equal,
8728 returns a value that compares equal to both operands. This means that
8729 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8731 '``llvm.maxnum.*``' Intrinsic
8732 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8737 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
8738 floating point or vector of floating point type. Not all targets support
8743 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
8744 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
8745 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8746 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
8747 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8752 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
8759 The arguments and return value are floating point numbers of the same
8764 Follows the IEEE-754 semantics for maxNum, which also match for libm's
8767 If either operand is a NaN, returns the other non-NaN operand. Returns
8768 NaN only if both operands are NaN. If the operands compare equal,
8769 returns a value that compares equal to both operands. This means that
8770 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8772 '``llvm.copysign.*``' Intrinsic
8773 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8778 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
8779 floating point or vector of floating point type. Not all targets support
8784 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
8785 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
8786 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
8787 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
8788 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
8793 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
8794 first operand and the sign of the second operand.
8799 The arguments and return value are floating point numbers of the same
8805 This function returns the same values as the libm ``copysign``
8806 functions would, and handles error conditions in the same way.
8808 '``llvm.floor.*``' Intrinsic
8809 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8814 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
8815 floating point or vector of floating point type. Not all targets support
8820 declare float @llvm.floor.f32(float %Val)
8821 declare double @llvm.floor.f64(double %Val)
8822 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
8823 declare fp128 @llvm.floor.f128(fp128 %Val)
8824 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
8829 The '``llvm.floor.*``' intrinsics return the floor of the operand.
8834 The argument and return value are floating point numbers of the same
8840 This function returns the same values as the libm ``floor`` functions
8841 would, and handles error conditions in the same way.
8843 '``llvm.ceil.*``' Intrinsic
8844 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8849 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
8850 floating point or vector of floating point type. Not all targets support
8855 declare float @llvm.ceil.f32(float %Val)
8856 declare double @llvm.ceil.f64(double %Val)
8857 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
8858 declare fp128 @llvm.ceil.f128(fp128 %Val)
8859 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
8864 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
8869 The argument and return value are floating point numbers of the same
8875 This function returns the same values as the libm ``ceil`` functions
8876 would, and handles error conditions in the same way.
8878 '``llvm.trunc.*``' Intrinsic
8879 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8884 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
8885 floating point or vector of floating point type. Not all targets support
8890 declare float @llvm.trunc.f32(float %Val)
8891 declare double @llvm.trunc.f64(double %Val)
8892 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
8893 declare fp128 @llvm.trunc.f128(fp128 %Val)
8894 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
8899 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
8900 nearest integer not larger in magnitude than the operand.
8905 The argument and return value are floating point numbers of the same
8911 This function returns the same values as the libm ``trunc`` functions
8912 would, and handles error conditions in the same way.
8914 '``llvm.rint.*``' Intrinsic
8915 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8920 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
8921 floating point or vector of floating point type. Not all targets support
8926 declare float @llvm.rint.f32(float %Val)
8927 declare double @llvm.rint.f64(double %Val)
8928 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
8929 declare fp128 @llvm.rint.f128(fp128 %Val)
8930 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
8935 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
8936 nearest integer. It may raise an inexact floating-point exception if the
8937 operand isn't an integer.
8942 The argument and return value are floating point numbers of the same
8948 This function returns the same values as the libm ``rint`` functions
8949 would, and handles error conditions in the same way.
8951 '``llvm.nearbyint.*``' Intrinsic
8952 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8957 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
8958 floating point or vector of floating point type. Not all targets support
8963 declare float @llvm.nearbyint.f32(float %Val)
8964 declare double @llvm.nearbyint.f64(double %Val)
8965 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
8966 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
8967 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
8972 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
8978 The argument and return value are floating point numbers of the same
8984 This function returns the same values as the libm ``nearbyint``
8985 functions would, and handles error conditions in the same way.
8987 '``llvm.round.*``' Intrinsic
8988 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8993 This is an overloaded intrinsic. You can use ``llvm.round`` on any
8994 floating point or vector of floating point type. Not all targets support
8999 declare float @llvm.round.f32(float %Val)
9000 declare double @llvm.round.f64(double %Val)
9001 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
9002 declare fp128 @llvm.round.f128(fp128 %Val)
9003 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
9008 The '``llvm.round.*``' intrinsics returns the operand rounded to the
9014 The argument and return value are floating point numbers of the same
9020 This function returns the same values as the libm ``round``
9021 functions would, and handles error conditions in the same way.
9023 Bit Manipulation Intrinsics
9024 ---------------------------
9026 LLVM provides intrinsics for a few important bit manipulation
9027 operations. These allow efficient code generation for some algorithms.
9029 '``llvm.bswap.*``' Intrinsics
9030 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9035 This is an overloaded intrinsic function. You can use bswap on any
9036 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
9040 declare i16 @llvm.bswap.i16(i16 <id>)
9041 declare i32 @llvm.bswap.i32(i32 <id>)
9042 declare i64 @llvm.bswap.i64(i64 <id>)
9047 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
9048 values with an even number of bytes (positive multiple of 16 bits).
9049 These are useful for performing operations on data that is not in the
9050 target's native byte order.
9055 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
9056 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
9057 intrinsic returns an i32 value that has the four bytes of the input i32
9058 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
9059 returned i32 will have its bytes in 3, 2, 1, 0 order. The
9060 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
9061 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
9064 '``llvm.ctpop.*``' Intrinsic
9065 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9070 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
9071 bit width, or on any vector with integer elements. Not all targets
9072 support all bit widths or vector types, however.
9076 declare i8 @llvm.ctpop.i8(i8 <src>)
9077 declare i16 @llvm.ctpop.i16(i16 <src>)
9078 declare i32 @llvm.ctpop.i32(i32 <src>)
9079 declare i64 @llvm.ctpop.i64(i64 <src>)
9080 declare i256 @llvm.ctpop.i256(i256 <src>)
9081 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
9086 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
9092 The only argument is the value to be counted. The argument may be of any
9093 integer type, or a vector with integer elements. The return type must
9094 match the argument type.
9099 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
9100 each element of a vector.
9102 '``llvm.ctlz.*``' Intrinsic
9103 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9108 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
9109 integer bit width, or any vector whose elements are integers. Not all
9110 targets support all bit widths or vector types, however.
9114 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
9115 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
9116 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
9117 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
9118 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
9119 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
9124 The '``llvm.ctlz``' family of intrinsic functions counts the number of
9125 leading zeros in a variable.
9130 The first argument is the value to be counted. This argument may be of
9131 any integer type, or a vector with integer element type. The return
9132 type must match the first argument type.
9134 The second argument must be a constant and is a flag to indicate whether
9135 the intrinsic should ensure that a zero as the first argument produces a
9136 defined result. Historically some architectures did not provide a
9137 defined result for zero values as efficiently, and many algorithms are
9138 now predicated on avoiding zero-value inputs.
9143 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
9144 zeros in a variable, or within each element of the vector. If
9145 ``src == 0`` then the result is the size in bits of the type of ``src``
9146 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9147 ``llvm.ctlz(i32 2) = 30``.
9149 '``llvm.cttz.*``' Intrinsic
9150 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9155 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
9156 integer bit width, or any vector of integer elements. Not all targets
9157 support all bit widths or vector types, however.
9161 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
9162 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
9163 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
9164 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
9165 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
9166 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
9171 The '``llvm.cttz``' family of intrinsic functions counts the number of
9177 The first argument is the value to be counted. This argument may be of
9178 any integer type, or a vector with integer element type. The return
9179 type must match the first argument type.
9181 The second argument must be a constant and is a flag to indicate whether
9182 the intrinsic should ensure that a zero as the first argument produces a
9183 defined result. Historically some architectures did not provide a
9184 defined result for zero values as efficiently, and many algorithms are
9185 now predicated on avoiding zero-value inputs.
9190 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
9191 zeros in a variable, or within each element of a vector. If ``src == 0``
9192 then the result is the size in bits of the type of ``src`` if
9193 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9194 ``llvm.cttz(2) = 1``.
9198 Arithmetic with Overflow Intrinsics
9199 -----------------------------------
9201 LLVM provides intrinsics for some arithmetic with overflow operations.
9203 '``llvm.sadd.with.overflow.*``' Intrinsics
9204 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9209 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
9210 on any integer bit width.
9214 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
9215 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9216 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
9221 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9222 a signed addition of the two arguments, and indicate whether an overflow
9223 occurred during the signed summation.
9228 The arguments (%a and %b) and the first element of the result structure
9229 may be of integer types of any bit width, but they must have the same
9230 bit width. The second element of the result structure must be of type
9231 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9237 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9238 a signed addition of the two variables. They return a structure --- the
9239 first element of which is the signed summation, and the second element
9240 of which is a bit specifying if the signed summation resulted in an
9246 .. code-block:: llvm
9248 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9249 %sum = extractvalue {i32, i1} %res, 0
9250 %obit = extractvalue {i32, i1} %res, 1
9251 br i1 %obit, label %overflow, label %normal
9253 '``llvm.uadd.with.overflow.*``' Intrinsics
9254 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9259 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
9260 on any integer bit width.
9264 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
9265 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
9266 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
9271 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9272 an unsigned addition of the two arguments, and indicate whether a carry
9273 occurred during the unsigned summation.
9278 The arguments (%a and %b) and the first element of the result structure
9279 may be of integer types of any bit width, but they must have the same
9280 bit width. The second element of the result structure must be of type
9281 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9287 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9288 an unsigned addition of the two arguments. They return a structure --- the
9289 first element of which is the sum, and the second element of which is a
9290 bit specifying if the unsigned summation resulted in a carry.
9295 .. code-block:: llvm
9297 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
9298 %sum = extractvalue {i32, i1} %res, 0
9299 %obit = extractvalue {i32, i1} %res, 1
9300 br i1 %obit, label %carry, label %normal
9302 '``llvm.ssub.with.overflow.*``' Intrinsics
9303 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9308 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
9309 on any integer bit width.
9313 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
9314 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
9315 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
9320 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
9321 a signed subtraction of the two arguments, and indicate whether an
9322 overflow occurred during the signed subtraction.
9327 The arguments (%a and %b) and the first element of the result structure
9328 may be of integer types of any bit width, but they must have the same
9329 bit width. The second element of the result structure must be of type
9330 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9336 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
9337 a signed subtraction of the two arguments. They return a structure --- the
9338 first element of which is the subtraction, and the second element of
9339 which is a bit specifying if the signed subtraction resulted in an
9345 .. code-block:: llvm
9347 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
9348 %sum = extractvalue {i32, i1} %res, 0
9349 %obit = extractvalue {i32, i1} %res, 1
9350 br i1 %obit, label %overflow, label %normal
9352 '``llvm.usub.with.overflow.*``' Intrinsics
9353 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9358 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
9359 on any integer bit width.
9363 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
9364 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
9365 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
9370 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
9371 an unsigned subtraction of the two arguments, and indicate whether an
9372 overflow occurred during the unsigned subtraction.
9377 The arguments (%a and %b) and the first element of the result structure
9378 may be of integer types of any bit width, but they must have the same
9379 bit width. The second element of the result structure must be of type
9380 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9386 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
9387 an unsigned subtraction of the two arguments. They return a structure ---
9388 the first element of which is the subtraction, and the second element of
9389 which is a bit specifying if the unsigned subtraction resulted in an
9395 .. code-block:: llvm
9397 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
9398 %sum = extractvalue {i32, i1} %res, 0
9399 %obit = extractvalue {i32, i1} %res, 1
9400 br i1 %obit, label %overflow, label %normal
9402 '``llvm.smul.with.overflow.*``' Intrinsics
9403 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9408 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
9409 on any integer bit width.
9413 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
9414 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
9415 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
9420 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
9421 a signed multiplication of the two arguments, and indicate whether an
9422 overflow occurred during the signed multiplication.
9427 The arguments (%a and %b) and the first element of the result structure
9428 may be of integer types of any bit width, but they must have the same
9429 bit width. The second element of the result structure must be of type
9430 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9436 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
9437 a signed multiplication of the two arguments. They return a structure ---
9438 the first element of which is the multiplication, and the second element
9439 of which is a bit specifying if the signed multiplication resulted in an
9445 .. code-block:: llvm
9447 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
9448 %sum = extractvalue {i32, i1} %res, 0
9449 %obit = extractvalue {i32, i1} %res, 1
9450 br i1 %obit, label %overflow, label %normal
9452 '``llvm.umul.with.overflow.*``' Intrinsics
9453 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9458 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
9459 on any integer bit width.
9463 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
9464 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9465 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
9470 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9471 a unsigned multiplication of the two arguments, and indicate whether an
9472 overflow occurred during the unsigned multiplication.
9477 The arguments (%a and %b) and the first element of the result structure
9478 may be of integer types of any bit width, but they must have the same
9479 bit width. The second element of the result structure must be of type
9480 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9486 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9487 an unsigned multiplication of the two arguments. They return a structure ---
9488 the first element of which is the multiplication, and the second
9489 element of which is a bit specifying if the unsigned multiplication
9490 resulted in an overflow.
9495 .. code-block:: llvm
9497 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9498 %sum = extractvalue {i32, i1} %res, 0
9499 %obit = extractvalue {i32, i1} %res, 1
9500 br i1 %obit, label %overflow, label %normal
9502 Specialised Arithmetic Intrinsics
9503 ---------------------------------
9505 '``llvm.fmuladd.*``' Intrinsic
9506 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9513 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
9514 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
9519 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
9520 expressions that can be fused if the code generator determines that (a) the
9521 target instruction set has support for a fused operation, and (b) that the
9522 fused operation is more efficient than the equivalent, separate pair of mul
9523 and add instructions.
9528 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
9529 multiplicands, a and b, and an addend c.
9538 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
9540 is equivalent to the expression a \* b + c, except that rounding will
9541 not be performed between the multiplication and addition steps if the
9542 code generator fuses the operations. Fusion is not guaranteed, even if
9543 the target platform supports it. If a fused multiply-add is required the
9544 corresponding llvm.fma.\* intrinsic function should be used
9545 instead. This never sets errno, just as '``llvm.fma.*``'.
9550 .. code-block:: llvm
9552 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
9554 Half Precision Floating Point Intrinsics
9555 ----------------------------------------
9557 For most target platforms, half precision floating point is a
9558 storage-only format. This means that it is a dense encoding (in memory)
9559 but does not support computation in the format.
9561 This means that code must first load the half-precision floating point
9562 value as an i16, then convert it to float with
9563 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
9564 then be performed on the float value (including extending to double
9565 etc). To store the value back to memory, it is first converted to float
9566 if needed, then converted to i16 with
9567 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
9570 .. _int_convert_to_fp16:
9572 '``llvm.convert.to.fp16``' Intrinsic
9573 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9580 declare i16 @llvm.convert.to.fp16.f32(float %a)
9581 declare i16 @llvm.convert.to.fp16.f64(double %a)
9586 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9587 conventional floating point type to half precision floating point format.
9592 The intrinsic function contains single argument - the value to be
9598 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9599 conventional floating point format to half precision floating point format. The
9600 return value is an ``i16`` which contains the converted number.
9605 .. code-block:: llvm
9607 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
9608 store i16 %res, i16* @x, align 2
9610 .. _int_convert_from_fp16:
9612 '``llvm.convert.from.fp16``' Intrinsic
9613 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9620 declare float @llvm.convert.from.fp16.f32(i16 %a)
9621 declare double @llvm.convert.from.fp16.f64(i16 %a)
9626 The '``llvm.convert.from.fp16``' intrinsic function performs a
9627 conversion from half precision floating point format to single precision
9628 floating point format.
9633 The intrinsic function contains single argument - the value to be
9639 The '``llvm.convert.from.fp16``' intrinsic function performs a
9640 conversion from half single precision floating point format to single
9641 precision floating point format. The input half-float value is
9642 represented by an ``i16`` value.
9647 .. code-block:: llvm
9649 %a = load i16, i16* @x, align 2
9650 %res = call float @llvm.convert.from.fp16(i16 %a)
9657 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
9658 prefix), are described in the `LLVM Source Level
9659 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
9662 Exception Handling Intrinsics
9663 -----------------------------
9665 The LLVM exception handling intrinsics (which all start with
9666 ``llvm.eh.`` prefix), are described in the `LLVM Exception
9667 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
9671 Trampoline Intrinsics
9672 ---------------------
9674 These intrinsics make it possible to excise one parameter, marked with
9675 the :ref:`nest <nest>` attribute, from a function. The result is a
9676 callable function pointer lacking the nest parameter - the caller does
9677 not need to provide a value for it. Instead, the value to use is stored
9678 in advance in a "trampoline", a block of memory usually allocated on the
9679 stack, which also contains code to splice the nest value into the
9680 argument list. This is used to implement the GCC nested function address
9683 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
9684 then the resulting function pointer has signature ``i32 (i32, i32)*``.
9685 It can be created as follows:
9687 .. code-block:: llvm
9689 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
9690 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
9691 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
9692 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
9693 %fp = bitcast i8* %p to i32 (i32, i32)*
9695 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
9696 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
9700 '``llvm.init.trampoline``' Intrinsic
9701 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9708 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
9713 This fills the memory pointed to by ``tramp`` with executable code,
9714 turning it into a trampoline.
9719 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
9720 pointers. The ``tramp`` argument must point to a sufficiently large and
9721 sufficiently aligned block of memory; this memory is written to by the
9722 intrinsic. Note that the size and the alignment are target-specific -
9723 LLVM currently provides no portable way of determining them, so a
9724 front-end that generates this intrinsic needs to have some
9725 target-specific knowledge. The ``func`` argument must hold a function
9726 bitcast to an ``i8*``.
9731 The block of memory pointed to by ``tramp`` is filled with target
9732 dependent code, turning it into a function. Then ``tramp`` needs to be
9733 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
9734 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
9735 function's signature is the same as that of ``func`` with any arguments
9736 marked with the ``nest`` attribute removed. At most one such ``nest``
9737 argument is allowed, and it must be of pointer type. Calling the new
9738 function is equivalent to calling ``func`` with the same argument list,
9739 but with ``nval`` used for the missing ``nest`` argument. If, after
9740 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
9741 modified, then the effect of any later call to the returned function
9742 pointer is undefined.
9746 '``llvm.adjust.trampoline``' Intrinsic
9747 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9754 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
9759 This performs any required machine-specific adjustment to the address of
9760 a trampoline (passed as ``tramp``).
9765 ``tramp`` must point to a block of memory which already has trampoline
9766 code filled in by a previous call to
9767 :ref:`llvm.init.trampoline <int_it>`.
9772 On some architectures the address of the code to be executed needs to be
9773 different than the address where the trampoline is actually stored. This
9774 intrinsic returns the executable address corresponding to ``tramp``
9775 after performing the required machine specific adjustments. The pointer
9776 returned can then be :ref:`bitcast and executed <int_trampoline>`.
9778 .. _int_mload_mstore:
9780 Masked Vector Load and Store Intrinsics
9781 ---------------------------------------
9783 LLVM provides intrinsics for predicated vector load and store operations. The predicate is specified by a mask operand, which holds one bit per vector element, switching the associated vector lane on or off. The memory addresses corresponding to the "off" lanes are not accessed. When all bits of the mask are on, the intrinsic is identical to a regular vector load or store. When all bits are off, no memory is accessed.
9787 '``llvm.masked.load.*``' Intrinsics
9788 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9792 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
9796 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
9797 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
9802 Reads a vector from memory according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes. The masked-off lanes in the result vector are taken from the corresponding lanes of the '``passthru``' operand.
9808 The first operand is the base pointer for the load. The second operand is the alignment of the source location. It must be a constant integer value. The third operand, mask, is a vector of boolean values with the same number of elements as the return type. The fourth is a pass-through value that is used to fill the masked-off lanes of the result. The return type, underlying type of the base pointer and the type of the '``passthru``' operand are the same vector types.
9814 The '``llvm.masked.load``' intrinsic is designed for conditional reading of selected vector elements in a single IR operation. It is useful for targets that support vector masked loads and allows vectorizing predicated basic blocks on these targets. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar load operations.
9815 The result of this operation is equivalent to a regular vector load instruction followed by a 'select' between the loaded and the passthru values, predicated on the same mask. However, using this intrinsic prevents exceptions on memory access to masked-off lanes.
9820 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
9822 ;; The result of the two following instructions is identical aside from potential memory access exception
9823 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
9824 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
9828 '``llvm.masked.store.*``' Intrinsics
9829 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9833 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
9837 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
9838 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
9843 Writes a vector to memory according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes.
9848 The first operand is the vector value to be written to memory. The second operand is the base pointer for the store, it has the same underlying type as the value operand. The third operand is the alignment of the destination location. The fourth operand, mask, is a vector of boolean values. The types of the mask and the value operand must have the same number of vector elements.
9854 The '``llvm.masked.store``' intrinsics is designed for conditional writing of selected vector elements in a single IR operation. It is useful for targets that support vector masked store and allows vectorizing predicated basic blocks on these targets. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar store operations.
9855 The result of this operation is equivalent to a load-modify-store sequence. However, using this intrinsic prevents exceptions and data races on memory access to masked-off lanes.
9859 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
9861 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
9862 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
9863 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
9864 store <16 x float> %res, <16 x float>* %ptr, align 4
9867 Masked Vector Gather and Scatter Intrinsics
9868 -------------------------------------------
9870 LLVM provides intrinsics for vector gather and scatter operations. They are similar to :ref:`Masked Vector Load and Store <int_mload_mstore>`, except they are designed for arbitrary memory accesses, rather than sequential memory accesses. Gather and scatter also employ a mask operand, which holds one bit per vector element, switching the associated vector lane on or off. The memory addresses corresponding to the "off" lanes are not accessed. When all bits are off, no memory is accessed.
9874 '``llvm.masked.gather.*``' Intrinsics
9875 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9879 This is an overloaded intrinsic. The loaded data are multiple scalar values of any integer or floating point data type gathered together into one vector.
9883 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
9884 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
9889 Reads scalar values from arbitrary memory locations and gathers them into one vector. The memory locations are provided in the vector of pointers '``ptrs``'. The memory is accessed according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes. The masked-off lanes in the result vector are taken from the corresponding lanes of the '``passthru``' operand.
9895 The first operand is a vector of pointers which holds all memory addresses to read. The second operand is an alignment of the source addresses. It must be a constant integer value. The third operand, mask, is a vector of boolean values with the same number of elements as the return type. The fourth is a pass-through value that is used to fill the masked-off lanes of the result. The return type, underlying type of the vector of pointers and the type of the '``passthru``' operand are the same vector types.
9901 The '``llvm.masked.gather``' intrinsic is designed for conditional reading of multiple scalar values from arbitrary memory locations in a single IR operation. It is useful for targets that support vector masked gathers and allows vectorizing basic blocks with data and control divergence. Other targets may support this intrinsic differently, for example by lowering it into a sequence of scalar load operations.
9902 The semantics of this operation are equivalent to a sequence of conditional scalar loads with subsequent gathering all loaded values into a single vector. The mask restricts memory access to certain lanes and facilitates vectorization of predicated basic blocks.
9907 %res = call <4 x double> @llvm.masked.gather.v4f64 (<4 x double*> %ptrs, i32 8, <4 x i1>%mask, <4 x double> <true, true, true, true>)
9909 ;; The gather with all-true mask is equivalent to the following instruction sequence
9910 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
9911 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
9912 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
9913 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
9915 %val0 = load double, double* %ptr0, align 8
9916 %val1 = load double, double* %ptr1, align 8
9917 %val2 = load double, double* %ptr2, align 8
9918 %val3 = load double, double* %ptr3, align 8
9920 %vec0 = insertelement <4 x double>undef, %val0, 0
9921 %vec01 = insertelement <4 x double>%vec0, %val1, 1
9922 %vec012 = insertelement <4 x double>%vec01, %val2, 2
9923 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
9927 '``llvm.masked.scatter.*``' Intrinsics
9928 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9932 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type. Each vector element is stored in an arbitrary memory addresses. Scatter with overlapping addresses is guaranteed to be ordered from least-significant to most-significant element.
9936 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
9937 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
9942 Writes each element from the value vector to the corresponding memory address. The memory addresses are represented as a vector of pointers. Writing is done according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes.
9947 The first operand is a vector value to be written to memory. The second operand is a vector of pointers, pointing to where the value elements should be stored. It has the same underlying type as the value operand. The third operand is an alignment of the destination addresses. The fourth operand, mask, is a vector of boolean values. The types of the mask and the value operand must have the same number of vector elements.
9953 The '``llvm.masked.scatter``' intrinsics is designed for writing selected vector elements to arbitrary memory addresses in a single IR operation. The operation may be conditional, when not all bits in the mask are switched on. It is useful for targets that support vector masked scatter and allows vectorizing basic blocks with data and control divergency. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar store operations.
9957 ;; This instruction unconditionaly stores data vector in multiple addresses
9958 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
9960 ;; It is equivalent to a list of scalar stores
9961 %val0 = extractelement <8 x i32> %value, i32 0
9962 %val1 = extractelement <8 x i32> %value, i32 1
9964 %val7 = extractelement <8 x i32> %value, i32 7
9965 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
9966 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
9968 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
9969 ;; Note: the order of the following stores is important when they overlap:
9970 store i32 %val0, i32* %ptr0, align 4
9971 store i32 %val1, i32* %ptr1, align 4
9973 store i32 %val7, i32* %ptr7, align 4
9979 This class of intrinsics provides information about the lifetime of
9980 memory objects and ranges where variables are immutable.
9984 '``llvm.lifetime.start``' Intrinsic
9985 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9992 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
9997 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
10003 The first argument is a constant integer representing the size of the
10004 object, or -1 if it is variable sized. The second argument is a pointer
10010 This intrinsic indicates that before this point in the code, the value
10011 of the memory pointed to by ``ptr`` is dead. This means that it is known
10012 to never be used and has an undefined value. A load from the pointer
10013 that precedes this intrinsic can be replaced with ``'undef'``.
10017 '``llvm.lifetime.end``' Intrinsic
10018 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10025 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
10030 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
10036 The first argument is a constant integer representing the size of the
10037 object, or -1 if it is variable sized. The second argument is a pointer
10043 This intrinsic indicates that after this point in the code, the value of
10044 the memory pointed to by ``ptr`` is dead. This means that it is known to
10045 never be used and has an undefined value. Any stores into the memory
10046 object following this intrinsic may be removed as dead.
10048 '``llvm.invariant.start``' Intrinsic
10049 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10056 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
10061 The '``llvm.invariant.start``' intrinsic specifies that the contents of
10062 a memory object will not change.
10067 The first argument is a constant integer representing the size of the
10068 object, or -1 if it is variable sized. The second argument is a pointer
10074 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
10075 the return value, the referenced memory location is constant and
10078 '``llvm.invariant.end``' Intrinsic
10079 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10086 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
10091 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
10092 memory object are mutable.
10097 The first argument is the matching ``llvm.invariant.start`` intrinsic.
10098 The second argument is a constant integer representing the size of the
10099 object, or -1 if it is variable sized and the third argument is a
10100 pointer to the object.
10105 This intrinsic indicates that the memory is mutable again.
10110 This class of intrinsics is designed to be generic and has no specific
10113 '``llvm.var.annotation``' Intrinsic
10114 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10121 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
10126 The '``llvm.var.annotation``' intrinsic.
10131 The first argument is a pointer to a value, the second is a pointer to a
10132 global string, the third is a pointer to a global string which is the
10133 source file name, and the last argument is the line number.
10138 This intrinsic allows annotation of local variables with arbitrary
10139 strings. This can be useful for special purpose optimizations that want
10140 to look for these annotations. These have no other defined use; they are
10141 ignored by code generation and optimization.
10143 '``llvm.ptr.annotation.*``' Intrinsic
10144 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10149 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
10150 pointer to an integer of any width. *NOTE* you must specify an address space for
10151 the pointer. The identifier for the default address space is the integer
10156 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
10157 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
10158 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
10159 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
10160 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
10165 The '``llvm.ptr.annotation``' intrinsic.
10170 The first argument is a pointer to an integer value of arbitrary bitwidth
10171 (result of some expression), the second is a pointer to a global string, the
10172 third is a pointer to a global string which is the source file name, and the
10173 last argument is the line number. It returns the value of the first argument.
10178 This intrinsic allows annotation of a pointer to an integer with arbitrary
10179 strings. This can be useful for special purpose optimizations that want to look
10180 for these annotations. These have no other defined use; they are ignored by code
10181 generation and optimization.
10183 '``llvm.annotation.*``' Intrinsic
10184 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10189 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
10190 any integer bit width.
10194 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
10195 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
10196 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
10197 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
10198 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
10203 The '``llvm.annotation``' intrinsic.
10208 The first argument is an integer value (result of some expression), the
10209 second is a pointer to a global string, the third is a pointer to a
10210 global string which is the source file name, and the last argument is
10211 the line number. It returns the value of the first argument.
10216 This intrinsic allows annotations to be put on arbitrary expressions
10217 with arbitrary strings. This can be useful for special purpose
10218 optimizations that want to look for these annotations. These have no
10219 other defined use; they are ignored by code generation and optimization.
10221 '``llvm.trap``' Intrinsic
10222 ^^^^^^^^^^^^^^^^^^^^^^^^^
10229 declare void @llvm.trap() noreturn nounwind
10234 The '``llvm.trap``' intrinsic.
10244 This intrinsic is lowered to the target dependent trap instruction. If
10245 the target does not have a trap instruction, this intrinsic will be
10246 lowered to a call of the ``abort()`` function.
10248 '``llvm.debugtrap``' Intrinsic
10249 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10256 declare void @llvm.debugtrap() nounwind
10261 The '``llvm.debugtrap``' intrinsic.
10271 This intrinsic is lowered to code which is intended to cause an
10272 execution trap with the intention of requesting the attention of a
10275 '``llvm.stackprotector``' Intrinsic
10276 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10283 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
10288 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
10289 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
10290 is placed on the stack before local variables.
10295 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
10296 The first argument is the value loaded from the stack guard
10297 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
10298 enough space to hold the value of the guard.
10303 This intrinsic causes the prologue/epilogue inserter to force the position of
10304 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
10305 to ensure that if a local variable on the stack is overwritten, it will destroy
10306 the value of the guard. When the function exits, the guard on the stack is
10307 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
10308 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
10309 calling the ``__stack_chk_fail()`` function.
10311 '``llvm.stackprotectorcheck``' Intrinsic
10312 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10319 declare void @llvm.stackprotectorcheck(i8** <guard>)
10324 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
10325 created stack protector and if they are not equal calls the
10326 ``__stack_chk_fail()`` function.
10331 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
10332 the variable ``@__stack_chk_guard``.
10337 This intrinsic is provided to perform the stack protector check by comparing
10338 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
10339 values do not match call the ``__stack_chk_fail()`` function.
10341 The reason to provide this as an IR level intrinsic instead of implementing it
10342 via other IR operations is that in order to perform this operation at the IR
10343 level without an intrinsic, one would need to create additional basic blocks to
10344 handle the success/failure cases. This makes it difficult to stop the stack
10345 protector check from disrupting sibling tail calls in Codegen. With this
10346 intrinsic, we are able to generate the stack protector basic blocks late in
10347 codegen after the tail call decision has occurred.
10349 '``llvm.objectsize``' Intrinsic
10350 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10357 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
10358 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
10363 The ``llvm.objectsize`` intrinsic is designed to provide information to
10364 the optimizers to determine at compile time whether a) an operation
10365 (like memcpy) will overflow a buffer that corresponds to an object, or
10366 b) that a runtime check for overflow isn't necessary. An object in this
10367 context means an allocation of a specific class, structure, array, or
10373 The ``llvm.objectsize`` intrinsic takes two arguments. The first
10374 argument is a pointer to or into the ``object``. The second argument is
10375 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
10376 or -1 (if false) when the object size is unknown. The second argument
10377 only accepts constants.
10382 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
10383 the size of the object concerned. If the size cannot be determined at
10384 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
10385 on the ``min`` argument).
10387 '``llvm.expect``' Intrinsic
10388 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10393 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
10398 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
10399 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
10400 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
10405 The ``llvm.expect`` intrinsic provides information about expected (the
10406 most probable) value of ``val``, which can be used by optimizers.
10411 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
10412 a value. The second argument is an expected value, this needs to be a
10413 constant value, variables are not allowed.
10418 This intrinsic is lowered to the ``val``.
10422 '``llvm.assume``' Intrinsic
10423 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10430 declare void @llvm.assume(i1 %cond)
10435 The ``llvm.assume`` allows the optimizer to assume that the provided
10436 condition is true. This information can then be used in simplifying other parts
10442 The condition which the optimizer may assume is always true.
10447 The intrinsic allows the optimizer to assume that the provided condition is
10448 always true whenever the control flow reaches the intrinsic call. No code is
10449 generated for this intrinsic, and instructions that contribute only to the
10450 provided condition are not used for code generation. If the condition is
10451 violated during execution, the behavior is undefined.
10453 Note that the optimizer might limit the transformations performed on values
10454 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
10455 only used to form the intrinsic's input argument. This might prove undesirable
10456 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
10457 sufficient overall improvement in code quality. For this reason,
10458 ``llvm.assume`` should not be used to document basic mathematical invariants
10459 that the optimizer can otherwise deduce or facts that are of little use to the
10464 '``llvm.bitset.test``' Intrinsic
10465 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10472 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
10478 The first argument is a pointer to be tested. The second argument is a
10479 metadata string containing the name of a :doc:`bitset <BitSets>`.
10484 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
10485 member of the given bitset.
10487 '``llvm.donothing``' Intrinsic
10488 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10495 declare void @llvm.donothing() nounwind readnone
10500 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
10501 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
10502 with an invoke instruction.
10512 This intrinsic does nothing, and it's removed by optimizers and ignored
10515 Stack Map Intrinsics
10516 --------------------
10518 LLVM provides experimental intrinsics to support runtime patching
10519 mechanisms commonly desired in dynamic language JITs. These intrinsics
10520 are described in :doc:`StackMaps`.