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 an alias, but then it only governs how
495 the alias is accessed. It will not have any effect in the aliasee.
497 For platforms without linker support of ELF TLS model, the -femulated-tls
498 flag can be used to generate GCC compatible emulated TLS code.
505 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
506 types <t_struct>`. Literal types are uniqued structurally, but identified types
507 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
508 to forward declare a type that is not yet available.
510 An example of an identified structure specification is:
514 %mytype = type { %mytype*, i32 }
516 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
517 literal types are uniqued in recent versions of LLVM.
524 Global variables define regions of memory allocated at compilation time
527 Global variable definitions must be initialized.
529 Global variables in other translation units can also be declared, in which
530 case they don't have an initializer.
532 Either global variable definitions or declarations may have an explicit section
533 to be placed in and may have an optional explicit alignment specified.
535 A variable may be defined as a global ``constant``, which indicates that
536 the contents of the variable will **never** be modified (enabling better
537 optimization, allowing the global data to be placed in the read-only
538 section of an executable, etc). Note that variables that need runtime
539 initialization cannot be marked ``constant`` as there is a store to the
542 LLVM explicitly allows *declarations* of global variables to be marked
543 constant, even if the final definition of the global is not. This
544 capability can be used to enable slightly better optimization of the
545 program, but requires the language definition to guarantee that
546 optimizations based on the 'constantness' are valid for the translation
547 units that do not include the definition.
549 As SSA values, global variables define pointer values that are in scope
550 (i.e. they dominate) all basic blocks in the program. Global variables
551 always define a pointer to their "content" type because they describe a
552 region of memory, and all memory objects in LLVM are accessed through
555 Global variables can be marked with ``unnamed_addr`` which indicates
556 that the address is not significant, only the content. Constants marked
557 like this can be merged with other constants if they have the same
558 initializer. Note that a constant with significant address *can* be
559 merged with a ``unnamed_addr`` constant, the result being a constant
560 whose address is significant.
562 A global variable may be declared to reside in a target-specific
563 numbered address space. For targets that support them, address spaces
564 may affect how optimizations are performed and/or what target
565 instructions are used to access the variable. The default address space
566 is zero. The address space qualifier must precede any other attributes.
568 LLVM allows an explicit section to be specified for globals. If the
569 target supports it, it will emit globals to the section specified.
570 Additionally, the global can placed in a comdat if the target has the necessary
573 By default, global initializers are optimized by assuming that global
574 variables defined within the module are not modified from their
575 initial values before the start of the global initializer. This is
576 true even for variables potentially accessible from outside the
577 module, including those with external linkage or appearing in
578 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
579 by marking the variable with ``externally_initialized``.
581 An explicit alignment may be specified for a global, which must be a
582 power of 2. If not present, or if the alignment is set to zero, the
583 alignment of the global is set by the target to whatever it feels
584 convenient. If an explicit alignment is specified, the global is forced
585 to have exactly that alignment. Targets and optimizers are not allowed
586 to over-align the global if the global has an assigned section. In this
587 case, the extra alignment could be observable: for example, code could
588 assume that the globals are densely packed in their section and try to
589 iterate over them as an array, alignment padding would break this
590 iteration. The maximum alignment is ``1 << 29``.
592 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
594 Variables and aliases can have a
595 :ref:`Thread Local Storage Model <tls_model>`.
599 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
600 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
601 <global | constant> <Type> [<InitializerConstant>]
602 [, section "name"] [, comdat [($name)]]
603 [, align <Alignment>]
605 For example, the following defines a global in a numbered address space
606 with an initializer, section, and alignment:
610 @G = addrspace(5) constant float 1.0, section "foo", align 4
612 The following example just declares a global variable
616 @G = external global i32
618 The following example defines a thread-local global with the
619 ``initialexec`` TLS model:
623 @G = thread_local(initialexec) global i32 0, align 4
625 .. _functionstructure:
630 LLVM function definitions consist of the "``define``" keyword, an
631 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
632 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
633 an optional :ref:`calling convention <callingconv>`,
634 an optional ``unnamed_addr`` attribute, a return type, an optional
635 :ref:`parameter attribute <paramattrs>` for the return type, a function
636 name, a (possibly empty) argument list (each with optional :ref:`parameter
637 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
638 an optional section, an optional alignment,
639 an optional :ref:`comdat <langref_comdats>`,
640 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
641 an optional :ref:`prologue <prologuedata>`,
642 an optional :ref:`personality <personalityfn>`,
643 an opening curly brace, a list of basic blocks, and a closing curly brace.
645 LLVM function declarations consist of the "``declare``" keyword, an
646 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
647 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
648 an optional :ref:`calling convention <callingconv>`,
649 an optional ``unnamed_addr`` attribute, a return type, an optional
650 :ref:`parameter attribute <paramattrs>` for the return type, a function
651 name, a possibly empty list of arguments, an optional alignment, an optional
652 :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
653 and an optional :ref:`prologue <prologuedata>`.
655 A function definition contains a list of basic blocks, forming the CFG (Control
656 Flow Graph) for the function. Each basic block may optionally start with a label
657 (giving the basic block a symbol table entry), contains a list of instructions,
658 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
659 function return). If an explicit label is not provided, a block is assigned an
660 implicit numbered label, using the next value from the same counter as used for
661 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
662 entry block does not have an explicit label, it will be assigned label "%0",
663 then the first unnamed temporary in that block will be "%1", etc.
665 The first basic block in a function is special in two ways: it is
666 immediately executed on entrance to the function, and it is not allowed
667 to have predecessor basic blocks (i.e. there can not be any branches to
668 the entry block of a function). Because the block can have no
669 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
671 LLVM allows an explicit section to be specified for functions. If the
672 target supports it, it will emit functions to the section specified.
673 Additionally, the function can be placed in a COMDAT.
675 An explicit alignment may be specified for a function. If not present,
676 or if the alignment is set to zero, the alignment of the function is set
677 by the target to whatever it feels convenient. If an explicit alignment
678 is specified, the function is forced to have at least that much
679 alignment. All alignments must be a power of 2.
681 If the ``unnamed_addr`` attribute is given, the address is known to not
682 be significant and two identical functions can be merged.
686 define [linkage] [visibility] [DLLStorageClass]
688 <ResultType> @<FunctionName> ([argument list])
689 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
690 [align N] [gc] [prefix Constant] [prologue Constant]
691 [personality Constant] { ... }
693 The argument list is a comma separated sequence of arguments where each
694 argument is of the following form:
698 <type> [parameter Attrs] [name]
706 Aliases, unlike function or variables, don't create any new data. They
707 are just a new symbol and metadata for an existing position.
709 Aliases have a name and an aliasee that is either a global value or a
712 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
713 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
714 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
718 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy>, <AliaseeTy>* @<Aliasee>
720 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
721 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
722 might not correctly handle dropping a weak symbol that is aliased.
724 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
725 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
728 Since aliases are only a second name, some restrictions apply, of which
729 some can only be checked when producing an object file:
731 * The expression defining the aliasee must be computable at assembly
732 time. Since it is just a name, no relocations can be used.
734 * No alias in the expression can be weak as the possibility of the
735 intermediate alias being overridden cannot be represented in an
738 * No global value in the expression can be a declaration, since that
739 would require a relocation, which is not possible.
746 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
748 Comdats have a name which represents the COMDAT key. All global objects that
749 specify this key will only end up in the final object file if the linker chooses
750 that key over some other key. Aliases are placed in the same COMDAT that their
751 aliasee computes to, if any.
753 Comdats have a selection kind to provide input on how the linker should
754 choose between keys in two different object files.
758 $<Name> = comdat SelectionKind
760 The selection kind must be one of the following:
763 The linker may choose any COMDAT key, the choice is arbitrary.
765 The linker may choose any COMDAT key but the sections must contain the
768 The linker will choose the section containing the largest COMDAT key.
770 The linker requires that only section with this COMDAT key exist.
772 The linker may choose any COMDAT key but the sections must contain the
775 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
776 ``any`` as a selection kind.
778 Here is an example of a COMDAT group where a function will only be selected if
779 the COMDAT key's section is the largest:
783 $foo = comdat largest
784 @foo = global i32 2, comdat($foo)
786 define void @bar() comdat($foo) {
790 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
796 @foo = global i32 2, comdat
799 In a COFF object file, this will create a COMDAT section with selection kind
800 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
801 and another COMDAT section with selection kind
802 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
803 section and contains the contents of the ``@bar`` symbol.
805 There are some restrictions on the properties of the global object.
806 It, or an alias to it, must have the same name as the COMDAT group when
808 The contents and size of this object may be used during link-time to determine
809 which COMDAT groups get selected depending on the selection kind.
810 Because the name of the object must match the name of the COMDAT group, the
811 linkage of the global object must not be local; local symbols can get renamed
812 if a collision occurs in the symbol table.
814 The combined use of COMDATS and section attributes may yield surprising results.
821 @g1 = global i32 42, section "sec", comdat($foo)
822 @g2 = global i32 42, section "sec", comdat($bar)
824 From the object file perspective, this requires the creation of two sections
825 with the same name. This is necessary because both globals belong to different
826 COMDAT groups and COMDATs, at the object file level, are represented by
829 Note that certain IR constructs like global variables and functions may
830 create COMDATs in the object file in addition to any which are specified using
831 COMDAT IR. This arises when the code generator is configured to emit globals
832 in individual sections (e.g. when `-data-sections` or `-function-sections`
833 is supplied to `llc`).
835 .. _namedmetadatastructure:
840 Named metadata is a collection of metadata. :ref:`Metadata
841 nodes <metadata>` (but not metadata strings) are the only valid
842 operands for a named metadata.
844 #. Named metadata are represented as a string of characters with the
845 metadata prefix. The rules for metadata names are the same as for
846 identifiers, but quoted names are not allowed. ``"\xx"`` type escapes
847 are still valid, which allows any character to be part of a name.
851 ; Some unnamed metadata nodes, which are referenced by the named metadata.
856 !name = !{!0, !1, !2}
863 The return type and each parameter of a function type may have a set of
864 *parameter attributes* associated with them. Parameter attributes are
865 used to communicate additional information about the result or
866 parameters of a function. Parameter attributes are considered to be part
867 of the function, not of the function type, so functions with different
868 parameter attributes can have the same function type.
870 Parameter attributes are simple keywords that follow the type specified.
871 If multiple parameter attributes are needed, they are space separated.
876 declare i32 @printf(i8* noalias nocapture, ...)
877 declare i32 @atoi(i8 zeroext)
878 declare signext i8 @returns_signed_char()
880 Note that any attributes for the function result (``nounwind``,
881 ``readonly``) come immediately after the argument list.
883 Currently, only the following parameter attributes are defined:
886 This indicates to the code generator that the parameter or return
887 value should be zero-extended to the extent required by the target's
888 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
889 the caller (for a parameter) or the callee (for a return value).
891 This indicates to the code generator that the parameter or return
892 value should be sign-extended to the extent required by the target's
893 ABI (which is usually 32-bits) by the caller (for a parameter) or
894 the callee (for a return value).
896 This indicates that this parameter or return value should be treated
897 in a special target-dependent fashion while emitting code for
898 a function call or return (usually, by putting it in a register as
899 opposed to memory, though some targets use it to distinguish between
900 two different kinds of registers). Use of this attribute is
903 This indicates that the pointer parameter should really be passed by
904 value to the function. The attribute implies that a hidden copy of
905 the pointee is made between the caller and the callee, so the callee
906 is unable to modify the value in the caller. This attribute is only
907 valid on LLVM pointer arguments. It is generally used to pass
908 structs and arrays by value, but is also valid on pointers to
909 scalars. The copy is considered to belong to the caller not the
910 callee (for example, ``readonly`` functions should not write to
911 ``byval`` parameters). This is not a valid attribute for return
914 The byval attribute also supports specifying an alignment with the
915 align attribute. It indicates the alignment of the stack slot to
916 form and the known alignment of the pointer specified to the call
917 site. If the alignment is not specified, then the code generator
918 makes a target-specific assumption.
924 The ``inalloca`` argument attribute allows the caller to take the
925 address of outgoing stack arguments. An ``inalloca`` argument must
926 be a pointer to stack memory produced by an ``alloca`` instruction.
927 The alloca, or argument allocation, must also be tagged with the
928 inalloca keyword. Only the last argument may have the ``inalloca``
929 attribute, and that argument is guaranteed to be passed in memory.
931 An argument allocation may be used by a call at most once because
932 the call may deallocate it. The ``inalloca`` attribute cannot be
933 used in conjunction with other attributes that affect argument
934 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
935 ``inalloca`` attribute also disables LLVM's implicit lowering of
936 large aggregate return values, which means that frontend authors
937 must lower them with ``sret`` pointers.
939 When the call site is reached, the argument allocation must have
940 been the most recent stack allocation that is still live, or the
941 results are undefined. It is possible to allocate additional stack
942 space after an argument allocation and before its call site, but it
943 must be cleared off with :ref:`llvm.stackrestore
946 See :doc:`InAlloca` for more information on how to use this
950 This indicates that the pointer parameter specifies the address of a
951 structure that is the return value of the function in the source
952 program. This pointer must be guaranteed by the caller to be valid:
953 loads and stores to the structure may be assumed by the callee
954 not to trap and to be properly aligned. This may only be applied to
955 the first parameter. This is not a valid attribute for return
959 This indicates that the pointer value may be assumed by the optimizer to
960 have the specified alignment.
962 Note that this attribute has additional semantics when combined with the
968 This indicates that objects accessed via pointer values
969 :ref:`based <pointeraliasing>` on the argument or return value are not also
970 accessed, during the execution of the function, via pointer values not
971 *based* on the argument or return value. The attribute on a return value
972 also has additional semantics described below. The caller shares the
973 responsibility with the callee for ensuring that these requirements are met.
974 For further details, please see the discussion of the NoAlias response in
975 :ref:`alias analysis <Must, May, or No>`.
977 Note that this definition of ``noalias`` is intentionally similar
978 to the definition of ``restrict`` in C99 for function arguments.
980 For function return values, C99's ``restrict`` is not meaningful,
981 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
982 attribute on return values are stronger than the semantics of the attribute
983 when used on function arguments. On function return values, the ``noalias``
984 attribute indicates that the function acts like a system memory allocation
985 function, returning a pointer to allocated storage disjoint from the
986 storage for any other object accessible to the caller.
989 This indicates that the callee does not make any copies of the
990 pointer that outlive the callee itself. This is not a valid
991 attribute for return values.
996 This indicates that the pointer parameter can be excised using the
997 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
998 attribute for return values and can only be applied to one parameter.
1001 This indicates that the function always returns the argument as its return
1002 value. This is an optimization hint to the code generator when generating
1003 the caller, allowing tail call optimization and omission of register saves
1004 and restores in some cases; it is not checked or enforced when generating
1005 the callee. The parameter and the function return type must be valid
1006 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
1007 valid attribute for return values and can only be applied to one parameter.
1010 This indicates that the parameter or return pointer is not null. This
1011 attribute may only be applied to pointer typed parameters. This is not
1012 checked or enforced by LLVM, the caller must ensure that the pointer
1013 passed in is non-null, or the callee must ensure that the returned pointer
1016 ``dereferenceable(<n>)``
1017 This indicates that the parameter or return pointer is dereferenceable. This
1018 attribute may only be applied to pointer typed parameters. A pointer that
1019 is dereferenceable can be loaded from speculatively without a risk of
1020 trapping. The number of bytes known to be dereferenceable must be provided
1021 in parentheses. It is legal for the number of bytes to be less than the
1022 size of the pointee type. The ``nonnull`` attribute does not imply
1023 dereferenceability (consider a pointer to one element past the end of an
1024 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1025 ``addrspace(0)`` (which is the default address space).
1027 ``dereferenceable_or_null(<n>)``
1028 This indicates that the parameter or return value isn't both
1029 non-null and non-dereferenceable (up to ``<n>`` bytes) at the same
1030 time. All non-null pointers tagged with
1031 ``dereferenceable_or_null(<n>)`` are ``dereferenceable(<n>)``.
1032 For address space 0 ``dereferenceable_or_null(<n>)`` implies that
1033 a pointer is exactly one of ``dereferenceable(<n>)`` or ``null``,
1034 and in other address spaces ``dereferenceable_or_null(<n>)``
1035 implies that a pointer is at least one of ``dereferenceable(<n>)``
1036 or ``null`` (i.e. it may be both ``null`` and
1037 ``dereferenceable(<n>)``). This attribute may only be applied to
1038 pointer typed parameters.
1042 Garbage Collector Strategy Names
1043 --------------------------------
1045 Each function may specify a garbage collector strategy name, which is simply a
1048 .. code-block:: llvm
1050 define void @f() gc "name" { ... }
1052 The supported values of *name* includes those :ref:`built in to LLVM
1053 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1054 strategy will cause the compiler to alter its output in order to support the
1055 named garbage collection algorithm. Note that LLVM itself does not contain a
1056 garbage collector, this functionality is restricted to generating machine code
1057 which can interoperate with a collector provided externally.
1064 Prefix data is data associated with a function which the code
1065 generator will emit immediately before the function's entrypoint.
1066 The purpose of this feature is to allow frontends to associate
1067 language-specific runtime metadata with specific functions and make it
1068 available through the function pointer while still allowing the
1069 function pointer to be called.
1071 To access the data for a given function, a program may bitcast the
1072 function pointer to a pointer to the constant's type and dereference
1073 index -1. This implies that the IR symbol points just past the end of
1074 the prefix data. For instance, take the example of a function annotated
1075 with a single ``i32``,
1077 .. code-block:: llvm
1079 define void @f() prefix i32 123 { ... }
1081 The prefix data can be referenced as,
1083 .. code-block:: llvm
1085 %0 = bitcast void* () @f to i32*
1086 %a = getelementptr inbounds i32, i32* %0, i32 -1
1087 %b = load i32, i32* %a
1089 Prefix data is laid out as if it were an initializer for a global variable
1090 of the prefix data's type. The function will be placed such that the
1091 beginning of the prefix data is aligned. This means that if the size
1092 of the prefix data is not a multiple of the alignment size, the
1093 function's entrypoint will not be aligned. If alignment of the
1094 function's entrypoint is desired, padding must be added to the prefix
1097 A function may have prefix data but no body. This has similar semantics
1098 to the ``available_externally`` linkage in that the data may be used by the
1099 optimizers but will not be emitted in the object file.
1106 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1107 be inserted prior to the function body. This can be used for enabling
1108 function hot-patching and instrumentation.
1110 To maintain the semantics of ordinary function calls, the prologue data must
1111 have a particular format. Specifically, it must begin with a sequence of
1112 bytes which decode to a sequence of machine instructions, valid for the
1113 module's target, which transfer control to the point immediately succeeding
1114 the prologue data, without performing any other visible action. This allows
1115 the inliner and other passes to reason about the semantics of the function
1116 definition without needing to reason about the prologue data. Obviously this
1117 makes the format of the prologue data highly target dependent.
1119 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1120 which encodes the ``nop`` instruction:
1122 .. code-block:: llvm
1124 define void @f() prologue i8 144 { ... }
1126 Generally prologue data can be formed by encoding a relative branch instruction
1127 which skips the metadata, as in this example of valid prologue data for the
1128 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1130 .. code-block:: llvm
1132 %0 = type <{ i8, i8, i8* }>
1134 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1136 A function may have prologue data but no body. This has similar semantics
1137 to the ``available_externally`` linkage in that the data may be used by the
1138 optimizers but will not be emitted in the object file.
1142 Personality Function
1143 --------------------
1145 The ``personality`` attribute permits functions to specify what function
1146 to use for exception handling.
1153 Attribute groups are groups of attributes that are referenced by objects within
1154 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1155 functions will use the same set of attributes. In the degenerative case of a
1156 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1157 group will capture the important command line flags used to build that file.
1159 An attribute group is a module-level object. To use an attribute group, an
1160 object references the attribute group's ID (e.g. ``#37``). An object may refer
1161 to more than one attribute group. In that situation, the attributes from the
1162 different groups are merged.
1164 Here is an example of attribute groups for a function that should always be
1165 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1167 .. code-block:: llvm
1169 ; Target-independent attributes:
1170 attributes #0 = { alwaysinline alignstack=4 }
1172 ; Target-dependent attributes:
1173 attributes #1 = { "no-sse" }
1175 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1176 define void @f() #0 #1 { ... }
1183 Function attributes are set to communicate additional information about
1184 a function. Function attributes are considered to be part of the
1185 function, not of the function type, so functions with different function
1186 attributes can have the same function type.
1188 Function attributes are simple keywords that follow the type specified.
1189 If multiple attributes are needed, they are space separated. For
1192 .. code-block:: llvm
1194 define void @f() noinline { ... }
1195 define void @f() alwaysinline { ... }
1196 define void @f() alwaysinline optsize { ... }
1197 define void @f() optsize { ... }
1200 This attribute indicates that, when emitting the prologue and
1201 epilogue, the backend should forcibly align the stack pointer.
1202 Specify the desired alignment, which must be a power of two, in
1205 This attribute indicates that the inliner should attempt to inline
1206 this function into callers whenever possible, ignoring any active
1207 inlining size threshold for this caller.
1209 This indicates that the callee function at a call site should be
1210 recognized as a built-in function, even though the function's declaration
1211 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1212 direct calls to functions that are declared with the ``nobuiltin``
1215 This attribute indicates that this function is rarely called. When
1216 computing edge weights, basic blocks post-dominated by a cold
1217 function call are also considered to be cold; and, thus, given low
1220 This attribute indicates that the callee is dependent on a convergent
1221 thread execution pattern under certain parallel execution models.
1222 Transformations that are execution model agnostic may not make the execution
1223 of a convergent operation control dependent on any additional values.
1225 This attribute indicates that the source code contained a hint that
1226 inlining this function is desirable (such as the "inline" keyword in
1227 C/C++). It is just a hint; it imposes no requirements on the
1230 This attribute indicates that the function should be added to a
1231 jump-instruction table at code-generation time, and that all address-taken
1232 references to this function should be replaced with a reference to the
1233 appropriate jump-instruction-table function pointer. Note that this creates
1234 a new pointer for the original function, which means that code that depends
1235 on function-pointer identity can break. So, any function annotated with
1236 ``jumptable`` must also be ``unnamed_addr``.
1238 This attribute suggests that optimization passes and code generator
1239 passes make choices that keep the code size of this function as small
1240 as possible and perform optimizations that may sacrifice runtime
1241 performance in order to minimize the size of the generated code.
1243 This attribute disables prologue / epilogue emission for the
1244 function. This can have very system-specific consequences.
1246 This indicates that the callee function at a call site is not recognized as
1247 a built-in function. LLVM will retain the original call and not replace it
1248 with equivalent code based on the semantics of the built-in function, unless
1249 the call site uses the ``builtin`` attribute. This is valid at call sites
1250 and on function declarations and definitions.
1252 This attribute indicates that calls to the function cannot be
1253 duplicated. A call to a ``noduplicate`` function may be moved
1254 within its parent function, but may not be duplicated within
1255 its parent function.
1257 A function containing a ``noduplicate`` call may still
1258 be an inlining candidate, provided that the call is not
1259 duplicated by inlining. That implies that the function has
1260 internal linkage and only has one call site, so the original
1261 call is dead after inlining.
1263 This attributes disables implicit floating point instructions.
1265 This attribute indicates that the inliner should never inline this
1266 function in any situation. This attribute may not be used together
1267 with the ``alwaysinline`` attribute.
1269 This attribute suppresses lazy symbol binding for the function. This
1270 may make calls to the function faster, at the cost of extra program
1271 startup time if the function is not called during program startup.
1273 This attribute indicates that the code generator should not use a
1274 red zone, even if the target-specific ABI normally permits it.
1276 This function attribute indicates that the function never returns
1277 normally. This produces undefined behavior at runtime if the
1278 function ever does dynamically return.
1280 This function attribute indicates that the function never raises an
1281 exception. If the function does raise an exception, its runtime
1282 behavior is undefined. However, functions marked nounwind may still
1283 trap or generate asynchronous exceptions. Exception handling schemes
1284 that are recognized by LLVM to handle asynchronous exceptions, such
1285 as SEH, will still provide their implementation defined semantics.
1287 This function attribute indicates that the function is not optimized
1288 by any optimization or code generator passes with the
1289 exception of interprocedural optimization passes.
1290 This attribute cannot be used together with the ``alwaysinline``
1291 attribute; this attribute is also incompatible
1292 with the ``minsize`` attribute and the ``optsize`` attribute.
1294 This attribute requires the ``noinline`` attribute to be specified on
1295 the function as well, so the function is never inlined into any caller.
1296 Only functions with the ``alwaysinline`` attribute are valid
1297 candidates for inlining into the body of this function.
1299 This attribute suggests that optimization passes and code generator
1300 passes make choices that keep the code size of this function low,
1301 and otherwise do optimizations specifically to reduce code size as
1302 long as they do not significantly impact runtime performance.
1304 On a function, this attribute indicates that the function computes its
1305 result (or decides to unwind an exception) based strictly on its arguments,
1306 without dereferencing any pointer arguments or otherwise accessing
1307 any mutable state (e.g. memory, control registers, etc) visible to
1308 caller functions. It does not write through any pointer arguments
1309 (including ``byval`` arguments) and never changes any state visible
1310 to callers. This means that it cannot unwind exceptions by calling
1311 the ``C++`` exception throwing methods.
1313 On an argument, this attribute indicates that the function does not
1314 dereference that pointer argument, even though it may read or write the
1315 memory that the pointer points to if accessed through other pointers.
1317 On a function, this attribute indicates that the function does not write
1318 through any pointer arguments (including ``byval`` arguments) or otherwise
1319 modify any state (e.g. memory, control registers, etc) visible to
1320 caller functions. It may dereference pointer arguments and read
1321 state that may be set in the caller. A readonly function always
1322 returns the same value (or unwinds an exception identically) when
1323 called with the same set of arguments and global state. It cannot
1324 unwind an exception by calling the ``C++`` exception throwing
1327 On an argument, this attribute indicates that the function does not write
1328 through this pointer argument, even though it may write to the memory that
1329 the pointer points to.
1331 This attribute indicates that the only memory accesses inside function are
1332 loads and stores from objects pointed to by its pointer-typed arguments,
1333 with arbitrary offsets. Or in other words, all memory operations in the
1334 function can refer to memory only using pointers based on its function
1336 Note that ``argmemonly`` can be used together with ``readonly`` attribute
1337 in order to specify that function reads only from its arguments.
1339 This attribute indicates that this function can return twice. The C
1340 ``setjmp`` is an example of such a function. The compiler disables
1341 some optimizations (like tail calls) in the caller of these
1344 This attribute indicates that
1345 `SafeStack <http://clang.llvm.org/docs/SafeStack.html>`_
1346 protection is enabled for this function.
1348 If a function that has a ``safestack`` attribute is inlined into a
1349 function that doesn't have a ``safestack`` attribute or which has an
1350 ``ssp``, ``sspstrong`` or ``sspreq`` attribute, then the resulting
1351 function will have a ``safestack`` attribute.
1352 ``sanitize_address``
1353 This attribute indicates that AddressSanitizer checks
1354 (dynamic address safety analysis) are enabled for this function.
1356 This attribute indicates that MemorySanitizer checks (dynamic detection
1357 of accesses to uninitialized memory) are enabled for this function.
1359 This attribute indicates that ThreadSanitizer checks
1360 (dynamic thread safety analysis) are enabled for this function.
1362 This attribute indicates that the function should emit a stack
1363 smashing protector. It is in the form of a "canary" --- a random value
1364 placed on the stack before the local variables that's checked upon
1365 return from the function to see if it has been overwritten. A
1366 heuristic is used to determine if a function needs stack protectors
1367 or not. The heuristic used will enable protectors for functions with:
1369 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1370 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1371 - Calls to alloca() with variable sizes or constant sizes greater than
1372 ``ssp-buffer-size``.
1374 Variables that are identified as requiring a protector will be arranged
1375 on the stack such that they are adjacent to the stack protector guard.
1377 If a function that has an ``ssp`` attribute is inlined into a
1378 function that doesn't have an ``ssp`` attribute, then the resulting
1379 function will have an ``ssp`` attribute.
1381 This attribute indicates that the function should *always* emit a
1382 stack smashing protector. This overrides the ``ssp`` function
1385 Variables that are identified as requiring a protector will be arranged
1386 on the stack such that they are adjacent to the stack protector guard.
1387 The specific layout rules are:
1389 #. Large arrays and structures containing large arrays
1390 (``>= ssp-buffer-size``) are closest to the stack protector.
1391 #. Small arrays and structures containing small arrays
1392 (``< ssp-buffer-size``) are 2nd closest to the protector.
1393 #. Variables that have had their address taken are 3rd closest to the
1396 If a function that has an ``sspreq`` attribute is inlined into a
1397 function that doesn't have an ``sspreq`` attribute or which has an
1398 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1399 an ``sspreq`` attribute.
1401 This attribute indicates that the function should emit a stack smashing
1402 protector. This attribute causes a strong heuristic to be used when
1403 determining if a function needs stack protectors. The strong heuristic
1404 will enable protectors for functions with:
1406 - Arrays of any size and type
1407 - Aggregates containing an array of any size and type.
1408 - Calls to alloca().
1409 - Local variables that have had their address taken.
1411 Variables that are identified as requiring a protector will be arranged
1412 on the stack such that they are adjacent to the stack protector guard.
1413 The specific layout rules are:
1415 #. Large arrays and structures containing large arrays
1416 (``>= ssp-buffer-size``) are closest to the stack protector.
1417 #. Small arrays and structures containing small arrays
1418 (``< ssp-buffer-size``) are 2nd closest to the protector.
1419 #. Variables that have had their address taken are 3rd closest to the
1422 This overrides the ``ssp`` function attribute.
1424 If a function that has an ``sspstrong`` attribute is inlined into a
1425 function that doesn't have an ``sspstrong`` attribute, then the
1426 resulting function will have an ``sspstrong`` attribute.
1428 This attribute indicates that the function will delegate to some other
1429 function with a tail call. The prototype of a thunk should not be used for
1430 optimization purposes. The caller is expected to cast the thunk prototype to
1431 match the thunk target prototype.
1433 This attribute indicates that the ABI being targeted requires that
1434 an unwind table entry be produced for this function even if we can
1435 show that no exceptions passes by it. This is normally the case for
1436 the ELF x86-64 abi, but it can be disabled for some compilation
1445 Note: operand bundles are a work in progress, and they should be
1446 considered experimental at this time.
1448 Operand bundles are tagged sets of SSA values that can be associated
1449 with certain LLVM instructions (currently only ``call`` s and
1450 ``invoke`` s). In a way they are like metadata, but dropping them is
1451 incorrect and will change program semantics.
1454 operand bundle set ::= '[' operand bundle ']'
1455 operand bundle ::= tag '(' [ bundle operand ] (, bundle operand )* ')'
1456 bundle operand ::= SSA value
1457 tag ::= string constant
1459 Operand bundles are **not** part of a function's signature, and a
1460 given function may be called from multiple places with different kinds
1461 of operand bundles. This reflects the fact that the operand bundles
1462 are conceptually a part of the ``call`` (or ``invoke``), not the
1463 callee being dispatched to.
1465 Operand bundles are a generic mechanism intended to support
1466 runtime-introspection-like functionality for managed languages. While
1467 the exact semantics of an operand bundle depend on the bundle tag,
1468 there are certain limitations to how much the presence of an operand
1469 bundle can influence the semantics of a program. These restrictions
1470 are described as the semantics of an "unknown" operand bundle. As
1471 long as the behavior of an operand bundle is describable within these
1472 restrictions, LLVM does not need to have special knowledge of the
1473 operand bundle to not miscompile programs containing it.
1475 - The bundle operands for an unknown operand bundle escape in unknown
1476 ways before control is transferred to the callee or invokee.
1478 - Calls and invokes with operand bundles have unknown read / write
1479 effect on the heap on entry and exit (even if the call target is
1480 ``readnone`` or ``readonly``).
1482 - An operand bundle at a call site cannot change the implementation
1483 of the called function. Inter-procedural optimizations work as
1484 usual as long as they take into account the first two properties.
1488 Module-Level Inline Assembly
1489 ----------------------------
1491 Modules may contain "module-level inline asm" blocks, which corresponds
1492 to the GCC "file scope inline asm" blocks. These blocks are internally
1493 concatenated by LLVM and treated as a single unit, but may be separated
1494 in the ``.ll`` file if desired. The syntax is very simple:
1496 .. code-block:: llvm
1498 module asm "inline asm code goes here"
1499 module asm "more can go here"
1501 The strings can contain any character by escaping non-printable
1502 characters. The escape sequence used is simply "\\xx" where "xx" is the
1503 two digit hex code for the number.
1505 Note that the assembly string *must* be parseable by LLVM's integrated assembler
1506 (unless it is disabled), even when emitting a ``.s`` file.
1508 .. _langref_datalayout:
1513 A module may specify a target specific data layout string that specifies
1514 how data is to be laid out in memory. The syntax for the data layout is
1517 .. code-block:: llvm
1519 target datalayout = "layout specification"
1521 The *layout specification* consists of a list of specifications
1522 separated by the minus sign character ('-'). Each specification starts
1523 with a letter and may include other information after the letter to
1524 define some aspect of the data layout. The specifications accepted are
1528 Specifies that the target lays out data in big-endian form. That is,
1529 the bits with the most significance have the lowest address
1532 Specifies that the target lays out data in little-endian form. That
1533 is, the bits with the least significance have the lowest address
1536 Specifies the natural alignment of the stack in bits. Alignment
1537 promotion of stack variables is limited to the natural stack
1538 alignment to avoid dynamic stack realignment. The stack alignment
1539 must be a multiple of 8-bits. If omitted, the natural stack
1540 alignment defaults to "unspecified", which does not prevent any
1541 alignment promotions.
1542 ``p[n]:<size>:<abi>:<pref>``
1543 This specifies the *size* of a pointer and its ``<abi>`` and
1544 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1545 bits. The address space, ``n``, is optional, and if not specified,
1546 denotes the default address space 0. The value of ``n`` must be
1547 in the range [1,2^23).
1548 ``i<size>:<abi>:<pref>``
1549 This specifies the alignment for an integer type of a given bit
1550 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1551 ``v<size>:<abi>:<pref>``
1552 This specifies the alignment for a vector type of a given bit
1554 ``f<size>:<abi>:<pref>``
1555 This specifies the alignment for a floating point type of a given bit
1556 ``<size>``. Only values of ``<size>`` that are supported by the target
1557 will work. 32 (float) and 64 (double) are supported on all targets; 80
1558 or 128 (different flavors of long double) are also supported on some
1561 This specifies the alignment for an object of aggregate type.
1563 If present, specifies that llvm names are mangled in the output. The
1566 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1567 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1568 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1569 symbols get a ``_`` prefix.
1570 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1571 functions also get a suffix based on the frame size.
1572 ``n<size1>:<size2>:<size3>...``
1573 This specifies a set of native integer widths for the target CPU in
1574 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1575 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1576 this set are considered to support most general arithmetic operations
1579 On every specification that takes a ``<abi>:<pref>``, specifying the
1580 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1581 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1583 When constructing the data layout for a given target, LLVM starts with a
1584 default set of specifications which are then (possibly) overridden by
1585 the specifications in the ``datalayout`` keyword. The default
1586 specifications are given in this list:
1588 - ``E`` - big endian
1589 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1590 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1591 same as the default address space.
1592 - ``S0`` - natural stack alignment is unspecified
1593 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1594 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1595 - ``i16:16:16`` - i16 is 16-bit aligned
1596 - ``i32:32:32`` - i32 is 32-bit aligned
1597 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1598 alignment of 64-bits
1599 - ``f16:16:16`` - half is 16-bit aligned
1600 - ``f32:32:32`` - float is 32-bit aligned
1601 - ``f64:64:64`` - double is 64-bit aligned
1602 - ``f128:128:128`` - quad is 128-bit aligned
1603 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1604 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1605 - ``a:0:64`` - aggregates are 64-bit aligned
1607 When LLVM is determining the alignment for a given type, it uses the
1610 #. If the type sought is an exact match for one of the specifications,
1611 that specification is used.
1612 #. If no match is found, and the type sought is an integer type, then
1613 the smallest integer type that is larger than the bitwidth of the
1614 sought type is used. If none of the specifications are larger than
1615 the bitwidth then the largest integer type is used. For example,
1616 given the default specifications above, the i7 type will use the
1617 alignment of i8 (next largest) while both i65 and i256 will use the
1618 alignment of i64 (largest specified).
1619 #. If no match is found, and the type sought is a vector type, then the
1620 largest vector type that is smaller than the sought vector type will
1621 be used as a fall back. This happens because <128 x double> can be
1622 implemented in terms of 64 <2 x double>, for example.
1624 The function of the data layout string may not be what you expect.
1625 Notably, this is not a specification from the frontend of what alignment
1626 the code generator should use.
1628 Instead, if specified, the target data layout is required to match what
1629 the ultimate *code generator* expects. This string is used by the
1630 mid-level optimizers to improve code, and this only works if it matches
1631 what the ultimate code generator uses. There is no way to generate IR
1632 that does not embed this target-specific detail into the IR. If you
1633 don't specify the string, the default specifications will be used to
1634 generate a Data Layout and the optimization phases will operate
1635 accordingly and introduce target specificity into the IR with respect to
1636 these default specifications.
1643 A module may specify a target triple string that describes the target
1644 host. The syntax for the target triple is simply:
1646 .. code-block:: llvm
1648 target triple = "x86_64-apple-macosx10.7.0"
1650 The *target triple* string consists of a series of identifiers delimited
1651 by the minus sign character ('-'). The canonical forms are:
1655 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1656 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1658 This information is passed along to the backend so that it generates
1659 code for the proper architecture. It's possible to override this on the
1660 command line with the ``-mtriple`` command line option.
1662 .. _pointeraliasing:
1664 Pointer Aliasing Rules
1665 ----------------------
1667 Any memory access must be done through a pointer value associated with
1668 an address range of the memory access, otherwise the behavior is
1669 undefined. Pointer values are associated with address ranges according
1670 to the following rules:
1672 - A pointer value is associated with the addresses associated with any
1673 value it is *based* on.
1674 - An address of a global variable is associated with the address range
1675 of the variable's storage.
1676 - The result value of an allocation instruction is associated with the
1677 address range of the allocated storage.
1678 - A null pointer in the default address-space is associated with no
1680 - An integer constant other than zero or a pointer value returned from
1681 a function not defined within LLVM may be associated with address
1682 ranges allocated through mechanisms other than those provided by
1683 LLVM. Such ranges shall not overlap with any ranges of addresses
1684 allocated by mechanisms provided by LLVM.
1686 A pointer value is *based* on another pointer value according to the
1689 - A pointer value formed from a ``getelementptr`` operation is *based*
1690 on the first value operand of the ``getelementptr``.
1691 - The result value of a ``bitcast`` is *based* on the operand of the
1693 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1694 values that contribute (directly or indirectly) to the computation of
1695 the pointer's value.
1696 - The "*based* on" relationship is transitive.
1698 Note that this definition of *"based"* is intentionally similar to the
1699 definition of *"based"* in C99, though it is slightly weaker.
1701 LLVM IR does not associate types with memory. The result type of a
1702 ``load`` merely indicates the size and alignment of the memory from
1703 which to load, as well as the interpretation of the value. The first
1704 operand type of a ``store`` similarly only indicates the size and
1705 alignment of the store.
1707 Consequently, type-based alias analysis, aka TBAA, aka
1708 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1709 :ref:`Metadata <metadata>` may be used to encode additional information
1710 which specialized optimization passes may use to implement type-based
1715 Volatile Memory Accesses
1716 ------------------------
1718 Certain memory accesses, such as :ref:`load <i_load>`'s,
1719 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1720 marked ``volatile``. The optimizers must not change the number of
1721 volatile operations or change their order of execution relative to other
1722 volatile operations. The optimizers *may* change the order of volatile
1723 operations relative to non-volatile operations. This is not Java's
1724 "volatile" and has no cross-thread synchronization behavior.
1726 IR-level volatile loads and stores cannot safely be optimized into
1727 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1728 flagged volatile. Likewise, the backend should never split or merge
1729 target-legal volatile load/store instructions.
1731 .. admonition:: Rationale
1733 Platforms may rely on volatile loads and stores of natively supported
1734 data width to be executed as single instruction. For example, in C
1735 this holds for an l-value of volatile primitive type with native
1736 hardware support, but not necessarily for aggregate types. The
1737 frontend upholds these expectations, which are intentionally
1738 unspecified in the IR. The rules above ensure that IR transformations
1739 do not violate the frontend's contract with the language.
1743 Memory Model for Concurrent Operations
1744 --------------------------------------
1746 The LLVM IR does not define any way to start parallel threads of
1747 execution or to register signal handlers. Nonetheless, there are
1748 platform-specific ways to create them, and we define LLVM IR's behavior
1749 in their presence. This model is inspired by the C++0x memory model.
1751 For a more informal introduction to this model, see the :doc:`Atomics`.
1753 We define a *happens-before* partial order as the least partial order
1756 - Is a superset of single-thread program order, and
1757 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1758 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1759 techniques, like pthread locks, thread creation, thread joining,
1760 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1761 Constraints <ordering>`).
1763 Note that program order does not introduce *happens-before* edges
1764 between a thread and signals executing inside that thread.
1766 Every (defined) read operation (load instructions, memcpy, atomic
1767 loads/read-modify-writes, etc.) R reads a series of bytes written by
1768 (defined) write operations (store instructions, atomic
1769 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1770 section, initialized globals are considered to have a write of the
1771 initializer which is atomic and happens before any other read or write
1772 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1773 may see any write to the same byte, except:
1775 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1776 write\ :sub:`2` happens before R\ :sub:`byte`, then
1777 R\ :sub:`byte` does not see write\ :sub:`1`.
1778 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1779 R\ :sub:`byte` does not see write\ :sub:`3`.
1781 Given that definition, R\ :sub:`byte` is defined as follows:
1783 - If R is volatile, the result is target-dependent. (Volatile is
1784 supposed to give guarantees which can support ``sig_atomic_t`` in
1785 C/C++, and may be used for accesses to addresses that do not behave
1786 like normal memory. It does not generally provide cross-thread
1788 - Otherwise, if there is no write to the same byte that happens before
1789 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1790 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1791 R\ :sub:`byte` returns the value written by that write.
1792 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1793 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1794 Memory Ordering Constraints <ordering>` section for additional
1795 constraints on how the choice is made.
1796 - Otherwise R\ :sub:`byte` returns ``undef``.
1798 R returns the value composed of the series of bytes it read. This
1799 implies that some bytes within the value may be ``undef`` **without**
1800 the entire value being ``undef``. Note that this only defines the
1801 semantics of the operation; it doesn't mean that targets will emit more
1802 than one instruction to read the series of bytes.
1804 Note that in cases where none of the atomic intrinsics are used, this
1805 model places only one restriction on IR transformations on top of what
1806 is required for single-threaded execution: introducing a store to a byte
1807 which might not otherwise be stored is not allowed in general.
1808 (Specifically, in the case where another thread might write to and read
1809 from an address, introducing a store can change a load that may see
1810 exactly one write into a load that may see multiple writes.)
1814 Atomic Memory Ordering Constraints
1815 ----------------------------------
1817 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1818 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1819 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1820 ordering parameters that determine which other atomic instructions on
1821 the same address they *synchronize with*. These semantics are borrowed
1822 from Java and C++0x, but are somewhat more colloquial. If these
1823 descriptions aren't precise enough, check those specs (see spec
1824 references in the :doc:`atomics guide <Atomics>`).
1825 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1826 differently since they don't take an address. See that instruction's
1827 documentation for details.
1829 For a simpler introduction to the ordering constraints, see the
1833 The set of values that can be read is governed by the happens-before
1834 partial order. A value cannot be read unless some operation wrote
1835 it. This is intended to provide a guarantee strong enough to model
1836 Java's non-volatile shared variables. This ordering cannot be
1837 specified for read-modify-write operations; it is not strong enough
1838 to make them atomic in any interesting way.
1840 In addition to the guarantees of ``unordered``, there is a single
1841 total order for modifications by ``monotonic`` operations on each
1842 address. All modification orders must be compatible with the
1843 happens-before order. There is no guarantee that the modification
1844 orders can be combined to a global total order for the whole program
1845 (and this often will not be possible). The read in an atomic
1846 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1847 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1848 order immediately before the value it writes. If one atomic read
1849 happens before another atomic read of the same address, the later
1850 read must see the same value or a later value in the address's
1851 modification order. This disallows reordering of ``monotonic`` (or
1852 stronger) operations on the same address. If an address is written
1853 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1854 read that address repeatedly, the other threads must eventually see
1855 the write. This corresponds to the C++0x/C1x
1856 ``memory_order_relaxed``.
1858 In addition to the guarantees of ``monotonic``, a
1859 *synchronizes-with* edge may be formed with a ``release`` operation.
1860 This is intended to model C++'s ``memory_order_acquire``.
1862 In addition to the guarantees of ``monotonic``, if this operation
1863 writes a value which is subsequently read by an ``acquire``
1864 operation, it *synchronizes-with* that operation. (This isn't a
1865 complete description; see the C++0x definition of a release
1866 sequence.) This corresponds to the C++0x/C1x
1867 ``memory_order_release``.
1868 ``acq_rel`` (acquire+release)
1869 Acts as both an ``acquire`` and ``release`` operation on its
1870 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1871 ``seq_cst`` (sequentially consistent)
1872 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1873 operation that only reads, ``release`` for an operation that only
1874 writes), there is a global total order on all
1875 sequentially-consistent operations on all addresses, which is
1876 consistent with the *happens-before* partial order and with the
1877 modification orders of all the affected addresses. Each
1878 sequentially-consistent read sees the last preceding write to the
1879 same address in this global order. This corresponds to the C++0x/C1x
1880 ``memory_order_seq_cst`` and Java volatile.
1884 If an atomic operation is marked ``singlethread``, it only *synchronizes
1885 with* or participates in modification and seq\_cst total orderings with
1886 other operations running in the same thread (for example, in signal
1894 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1895 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1896 :ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) have the following flags that can
1897 be set to enable otherwise unsafe floating point operations
1900 No NaNs - Allow optimizations to assume the arguments and result are not
1901 NaN. Such optimizations are required to retain defined behavior over
1902 NaNs, but the value of the result is undefined.
1905 No Infs - Allow optimizations to assume the arguments and result are not
1906 +/-Inf. Such optimizations are required to retain defined behavior over
1907 +/-Inf, but the value of the result is undefined.
1910 No Signed Zeros - Allow optimizations to treat the sign of a zero
1911 argument or result as insignificant.
1914 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1915 argument rather than perform division.
1918 Fast - Allow algebraically equivalent transformations that may
1919 dramatically change results in floating point (e.g. reassociate). This
1920 flag implies all the others.
1924 Use-list Order Directives
1925 -------------------------
1927 Use-list directives encode the in-memory order of each use-list, allowing the
1928 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1929 indexes that are assigned to the referenced value's uses. The referenced
1930 value's use-list is immediately sorted by these indexes.
1932 Use-list directives may appear at function scope or global scope. They are not
1933 instructions, and have no effect on the semantics of the IR. When they're at
1934 function scope, they must appear after the terminator of the final basic block.
1936 If basic blocks have their address taken via ``blockaddress()`` expressions,
1937 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1944 uselistorder <ty> <value>, { <order-indexes> }
1945 uselistorder_bb @function, %block { <order-indexes> }
1951 define void @foo(i32 %arg1, i32 %arg2) {
1953 ; ... instructions ...
1955 ; ... instructions ...
1957 ; At function scope.
1958 uselistorder i32 %arg1, { 1, 0, 2 }
1959 uselistorder label %bb, { 1, 0 }
1963 uselistorder i32* @global, { 1, 2, 0 }
1964 uselistorder i32 7, { 1, 0 }
1965 uselistorder i32 (i32) @bar, { 1, 0 }
1966 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1973 The LLVM type system is one of the most important features of the
1974 intermediate representation. Being typed enables a number of
1975 optimizations to be performed on the intermediate representation
1976 directly, without having to do extra analyses on the side before the
1977 transformation. A strong type system makes it easier to read the
1978 generated code and enables novel analyses and transformations that are
1979 not feasible to perform on normal three address code representations.
1989 The void type does not represent any value and has no size.
2007 The function type can be thought of as a function signature. It consists of a
2008 return type and a list of formal parameter types. The return type of a function
2009 type is a void type or first class type --- except for :ref:`label <t_label>`
2010 and :ref:`metadata <t_metadata>` types.
2016 <returntype> (<parameter list>)
2018 ...where '``<parameter list>``' is a comma-separated list of type
2019 specifiers. Optionally, the parameter list may include a type ``...``, which
2020 indicates that the function takes a variable number of arguments. Variable
2021 argument functions can access their arguments with the :ref:`variable argument
2022 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
2023 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
2027 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2028 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
2029 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2030 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
2031 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2032 | ``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. |
2033 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2034 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
2035 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2042 The :ref:`first class <t_firstclass>` types are perhaps the most important.
2043 Values of these types are the only ones which can be produced by
2051 These are the types that are valid in registers from CodeGen's perspective.
2060 The integer type is a very simple type that simply specifies an
2061 arbitrary bit width for the integer type desired. Any bit width from 1
2062 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2070 The number of bits the integer will occupy is specified by the ``N``
2076 +----------------+------------------------------------------------+
2077 | ``i1`` | a single-bit integer. |
2078 +----------------+------------------------------------------------+
2079 | ``i32`` | a 32-bit integer. |
2080 +----------------+------------------------------------------------+
2081 | ``i1942652`` | a really big integer of over 1 million bits. |
2082 +----------------+------------------------------------------------+
2086 Floating Point Types
2087 """"""""""""""""""""
2096 - 16-bit floating point value
2099 - 32-bit floating point value
2102 - 64-bit floating point value
2105 - 128-bit floating point value (112-bit mantissa)
2108 - 80-bit floating point value (X87)
2111 - 128-bit floating point value (two 64-bits)
2118 The x86_mmx type represents a value held in an MMX register on an x86
2119 machine. The operations allowed on it are quite limited: parameters and
2120 return values, load and store, and bitcast. User-specified MMX
2121 instructions are represented as intrinsic or asm calls with arguments
2122 and/or results of this type. There are no arrays, vectors or constants
2139 The pointer type is used to specify memory locations. Pointers are
2140 commonly used to reference objects in memory.
2142 Pointer types may have an optional address space attribute defining the
2143 numbered address space where the pointed-to object resides. The default
2144 address space is number zero. The semantics of non-zero address spaces
2145 are target-specific.
2147 Note that LLVM does not permit pointers to void (``void*``) nor does it
2148 permit pointers to labels (``label*``). Use ``i8*`` instead.
2158 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2159 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2160 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2161 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2162 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2163 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2164 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2173 A vector type is a simple derived type that represents a vector of
2174 elements. Vector types are used when multiple primitive data are
2175 operated in parallel using a single instruction (SIMD). A vector type
2176 requires a size (number of elements) and an underlying primitive data
2177 type. Vector types are considered :ref:`first class <t_firstclass>`.
2183 < <# elements> x <elementtype> >
2185 The number of elements is a constant integer value larger than 0;
2186 elementtype may be any integer, floating point or pointer type. Vectors
2187 of size zero are not allowed.
2191 +-------------------+--------------------------------------------------+
2192 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2193 +-------------------+--------------------------------------------------+
2194 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2195 +-------------------+--------------------------------------------------+
2196 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2197 +-------------------+--------------------------------------------------+
2198 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2199 +-------------------+--------------------------------------------------+
2208 The label type represents code labels.
2223 The token type is used when a value is associated with an instruction
2224 but all uses of the value must not attempt to introspect or obscure it.
2225 As such, it is not appropriate to have a :ref:`phi <i_phi>` or
2226 :ref:`select <i_select>` of type token.
2243 The metadata type represents embedded metadata. No derived types may be
2244 created from metadata except for :ref:`function <t_function>` arguments.
2257 Aggregate Types are a subset of derived types that can contain multiple
2258 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2259 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2269 The array type is a very simple derived type that arranges elements
2270 sequentially in memory. The array type requires a size (number of
2271 elements) and an underlying data type.
2277 [<# elements> x <elementtype>]
2279 The number of elements is a constant integer value; ``elementtype`` may
2280 be any type with a size.
2284 +------------------+--------------------------------------+
2285 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2286 +------------------+--------------------------------------+
2287 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2288 +------------------+--------------------------------------+
2289 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2290 +------------------+--------------------------------------+
2292 Here are some examples of multidimensional arrays:
2294 +-----------------------------+----------------------------------------------------------+
2295 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2296 +-----------------------------+----------------------------------------------------------+
2297 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2298 +-----------------------------+----------------------------------------------------------+
2299 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2300 +-----------------------------+----------------------------------------------------------+
2302 There is no restriction on indexing beyond the end of the array implied
2303 by a static type (though there are restrictions on indexing beyond the
2304 bounds of an allocated object in some cases). This means that
2305 single-dimension 'variable sized array' addressing can be implemented in
2306 LLVM with a zero length array type. An implementation of 'pascal style
2307 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2317 The structure type is used to represent a collection of data members
2318 together in memory. The elements of a structure may be any type that has
2321 Structures in memory are accessed using '``load``' and '``store``' by
2322 getting a pointer to a field with the '``getelementptr``' instruction.
2323 Structures in registers are accessed using the '``extractvalue``' and
2324 '``insertvalue``' instructions.
2326 Structures may optionally be "packed" structures, which indicate that
2327 the alignment of the struct is one byte, and that there is no padding
2328 between the elements. In non-packed structs, padding between field types
2329 is inserted as defined by the DataLayout string in the module, which is
2330 required to match what the underlying code generator expects.
2332 Structures can either be "literal" or "identified". A literal structure
2333 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2334 identified types are always defined at the top level with a name.
2335 Literal types are uniqued by their contents and can never be recursive
2336 or opaque since there is no way to write one. Identified types can be
2337 recursive, can be opaqued, and are never uniqued.
2343 %T1 = type { <type list> } ; Identified normal struct type
2344 %T2 = type <{ <type list> }> ; Identified packed struct type
2348 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2349 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2350 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2351 | ``{ 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``. |
2352 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2353 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2354 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2358 Opaque Structure Types
2359 """"""""""""""""""""""
2363 Opaque structure types are used to represent named structure types that
2364 do not have a body specified. This corresponds (for example) to the C
2365 notion of a forward declared structure.
2376 +--------------+-------------------+
2377 | ``opaque`` | An opaque type. |
2378 +--------------+-------------------+
2385 LLVM has several different basic types of constants. This section
2386 describes them all and their syntax.
2391 **Boolean constants**
2392 The two strings '``true``' and '``false``' are both valid constants
2394 **Integer constants**
2395 Standard integers (such as '4') are constants of the
2396 :ref:`integer <t_integer>` type. Negative numbers may be used with
2398 **Floating point constants**
2399 Floating point constants use standard decimal notation (e.g.
2400 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2401 hexadecimal notation (see below). The assembler requires the exact
2402 decimal value of a floating-point constant. For example, the
2403 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2404 decimal in binary. Floating point constants must have a :ref:`floating
2405 point <t_floating>` type.
2406 **Null pointer constants**
2407 The identifier '``null``' is recognized as a null pointer constant
2408 and must be of :ref:`pointer type <t_pointer>`.
2410 The one non-intuitive notation for constants is the hexadecimal form of
2411 floating point constants. For example, the form
2412 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2413 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2414 constants are required (and the only time that they are generated by the
2415 disassembler) is when a floating point constant must be emitted but it
2416 cannot be represented as a decimal floating point number in a reasonable
2417 number of digits. For example, NaN's, infinities, and other special
2418 values are represented in their IEEE hexadecimal format so that assembly
2419 and disassembly do not cause any bits to change in the constants.
2421 When using the hexadecimal form, constants of types half, float, and
2422 double are represented using the 16-digit form shown above (which
2423 matches the IEEE754 representation for double); half and float values
2424 must, however, be exactly representable as IEEE 754 half and single
2425 precision, respectively. Hexadecimal format is always used for long
2426 double, and there are three forms of long double. The 80-bit format used
2427 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2428 128-bit format used by PowerPC (two adjacent doubles) is represented by
2429 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2430 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2431 will only work if they match the long double format on your target.
2432 The IEEE 16-bit format (half precision) is represented by ``0xH``
2433 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2434 (sign bit at the left).
2436 There are no constants of type x86_mmx.
2438 .. _complexconstants:
2443 Complex constants are a (potentially recursive) combination of simple
2444 constants and smaller complex constants.
2446 **Structure constants**
2447 Structure constants are represented with notation similar to
2448 structure type definitions (a comma separated list of elements,
2449 surrounded by braces (``{}``)). For example:
2450 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2451 "``@G = external global i32``". Structure constants must have
2452 :ref:`structure type <t_struct>`, and the number and types of elements
2453 must match those specified by the type.
2455 Array constants are represented with notation similar to array type
2456 definitions (a comma separated list of elements, surrounded by
2457 square brackets (``[]``)). For example:
2458 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2459 :ref:`array type <t_array>`, and the number and types of elements must
2460 match those specified by the type. As a special case, character array
2461 constants may also be represented as a double-quoted string using the ``c``
2462 prefix. For example: "``c"Hello World\0A\00"``".
2463 **Vector constants**
2464 Vector constants are represented with notation similar to vector
2465 type definitions (a comma separated list of elements, surrounded by
2466 less-than/greater-than's (``<>``)). For example:
2467 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2468 must have :ref:`vector type <t_vector>`, and the number and types of
2469 elements must match those specified by the type.
2470 **Zero initialization**
2471 The string '``zeroinitializer``' can be used to zero initialize a
2472 value to zero of *any* type, including scalar and
2473 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2474 having to print large zero initializers (e.g. for large arrays) and
2475 is always exactly equivalent to using explicit zero initializers.
2477 A metadata node is a constant tuple without types. For example:
2478 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2479 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2480 Unlike other typed constants that are meant to be interpreted as part of
2481 the instruction stream, metadata is a place to attach additional
2482 information such as debug info.
2484 Global Variable and Function Addresses
2485 --------------------------------------
2487 The addresses of :ref:`global variables <globalvars>` and
2488 :ref:`functions <functionstructure>` are always implicitly valid
2489 (link-time) constants. These constants are explicitly referenced when
2490 the :ref:`identifier for the global <identifiers>` is used and always have
2491 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2494 .. code-block:: llvm
2498 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2505 The string '``undef``' can be used anywhere a constant is expected, and
2506 indicates that the user of the value may receive an unspecified
2507 bit-pattern. Undefined values may be of any type (other than '``label``'
2508 or '``void``') and be used anywhere a constant is permitted.
2510 Undefined values are useful because they indicate to the compiler that
2511 the program is well defined no matter what value is used. This gives the
2512 compiler more freedom to optimize. Here are some examples of
2513 (potentially surprising) transformations that are valid (in pseudo IR):
2515 .. code-block:: llvm
2525 This is safe because all of the output bits are affected by the undef
2526 bits. Any output bit can have a zero or one depending on the input bits.
2528 .. code-block:: llvm
2539 These logical operations have bits that are not always affected by the
2540 input. For example, if ``%X`` has a zero bit, then the output of the
2541 '``and``' operation will always be a zero for that bit, no matter what
2542 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2543 optimize or assume that the result of the '``and``' is '``undef``'.
2544 However, it is safe to assume that all bits of the '``undef``' could be
2545 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2546 all the bits of the '``undef``' operand to the '``or``' could be set,
2547 allowing the '``or``' to be folded to -1.
2549 .. code-block:: llvm
2551 %A = select undef, %X, %Y
2552 %B = select undef, 42, %Y
2553 %C = select %X, %Y, undef
2563 This set of examples shows that undefined '``select``' (and conditional
2564 branch) conditions can go *either way*, but they have to come from one
2565 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2566 both known to have a clear low bit, then ``%A`` would have to have a
2567 cleared low bit. However, in the ``%C`` example, the optimizer is
2568 allowed to assume that the '``undef``' operand could be the same as
2569 ``%Y``, allowing the whole '``select``' to be eliminated.
2571 .. code-block:: llvm
2573 %A = xor undef, undef
2590 This example points out that two '``undef``' operands are not
2591 necessarily the same. This can be surprising to people (and also matches
2592 C semantics) where they assume that "``X^X``" is always zero, even if
2593 ``X`` is undefined. This isn't true for a number of reasons, but the
2594 short answer is that an '``undef``' "variable" can arbitrarily change
2595 its value over its "live range". This is true because the variable
2596 doesn't actually *have a live range*. Instead, the value is logically
2597 read from arbitrary registers that happen to be around when needed, so
2598 the value is not necessarily consistent over time. In fact, ``%A`` and
2599 ``%C`` need to have the same semantics or the core LLVM "replace all
2600 uses with" concept would not hold.
2602 .. code-block:: llvm
2610 These examples show the crucial difference between an *undefined value*
2611 and *undefined behavior*. An undefined value (like '``undef``') is
2612 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2613 operation can be constant folded to '``undef``', because the '``undef``'
2614 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2615 However, in the second example, we can make a more aggressive
2616 assumption: because the ``undef`` is allowed to be an arbitrary value,
2617 we are allowed to assume that it could be zero. Since a divide by zero
2618 has *undefined behavior*, we are allowed to assume that the operation
2619 does not execute at all. This allows us to delete the divide and all
2620 code after it. Because the undefined operation "can't happen", the
2621 optimizer can assume that it occurs in dead code.
2623 .. code-block:: llvm
2625 a: store undef -> %X
2626 b: store %X -> undef
2631 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2632 value can be assumed to not have any effect; we can assume that the
2633 value is overwritten with bits that happen to match what was already
2634 there. However, a store *to* an undefined location could clobber
2635 arbitrary memory, therefore, it has undefined behavior.
2642 Poison values are similar to :ref:`undef values <undefvalues>`, however
2643 they also represent the fact that an instruction or constant expression
2644 that cannot evoke side effects has nevertheless detected a condition
2645 that results in undefined behavior.
2647 There is currently no way of representing a poison value in the IR; they
2648 only exist when produced by operations such as :ref:`add <i_add>` with
2651 Poison value behavior is defined in terms of value *dependence*:
2653 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2654 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2655 their dynamic predecessor basic block.
2656 - Function arguments depend on the corresponding actual argument values
2657 in the dynamic callers of their functions.
2658 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2659 instructions that dynamically transfer control back to them.
2660 - :ref:`Invoke <i_invoke>` instructions depend on the
2661 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2662 call instructions that dynamically transfer control back to them.
2663 - Non-volatile loads and stores depend on the most recent stores to all
2664 of the referenced memory addresses, following the order in the IR
2665 (including loads and stores implied by intrinsics such as
2666 :ref:`@llvm.memcpy <int_memcpy>`.)
2667 - An instruction with externally visible side effects depends on the
2668 most recent preceding instruction with externally visible side
2669 effects, following the order in the IR. (This includes :ref:`volatile
2670 operations <volatile>`.)
2671 - An instruction *control-depends* on a :ref:`terminator
2672 instruction <terminators>` if the terminator instruction has
2673 multiple successors and the instruction is always executed when
2674 control transfers to one of the successors, and may not be executed
2675 when control is transferred to another.
2676 - Additionally, an instruction also *control-depends* on a terminator
2677 instruction if the set of instructions it otherwise depends on would
2678 be different if the terminator had transferred control to a different
2680 - Dependence is transitive.
2682 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2683 with the additional effect that any instruction that has a *dependence*
2684 on a poison value has undefined behavior.
2686 Here are some examples:
2688 .. code-block:: llvm
2691 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2692 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2693 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2694 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2696 store i32 %poison, i32* @g ; Poison value stored to memory.
2697 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2699 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2701 %narrowaddr = bitcast i32* @g to i16*
2702 %wideaddr = bitcast i32* @g to i64*
2703 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2704 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2706 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2707 br i1 %cmp, label %true, label %end ; Branch to either destination.
2710 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2711 ; it has undefined behavior.
2715 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2716 ; Both edges into this PHI are
2717 ; control-dependent on %cmp, so this
2718 ; always results in a poison value.
2720 store volatile i32 0, i32* @g ; This would depend on the store in %true
2721 ; if %cmp is true, or the store in %entry
2722 ; otherwise, so this is undefined behavior.
2724 br i1 %cmp, label %second_true, label %second_end
2725 ; The same branch again, but this time the
2726 ; true block doesn't have side effects.
2733 store volatile i32 0, i32* @g ; This time, the instruction always depends
2734 ; on the store in %end. Also, it is
2735 ; control-equivalent to %end, so this is
2736 ; well-defined (ignoring earlier undefined
2737 ; behavior in this example).
2741 Addresses of Basic Blocks
2742 -------------------------
2744 ``blockaddress(@function, %block)``
2746 The '``blockaddress``' constant computes the address of the specified
2747 basic block in the specified function, and always has an ``i8*`` type.
2748 Taking the address of the entry block is illegal.
2750 This value only has defined behavior when used as an operand to the
2751 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2752 against null. Pointer equality tests between labels addresses results in
2753 undefined behavior --- though, again, comparison against null is ok, and
2754 no label is equal to the null pointer. This may be passed around as an
2755 opaque pointer sized value as long as the bits are not inspected. This
2756 allows ``ptrtoint`` and arithmetic to be performed on these values so
2757 long as the original value is reconstituted before the ``indirectbr``
2760 Finally, some targets may provide defined semantics when using the value
2761 as the operand to an inline assembly, but that is target specific.
2765 Constant Expressions
2766 --------------------
2768 Constant expressions are used to allow expressions involving other
2769 constants to be used as constants. Constant expressions may be of any
2770 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2771 that does not have side effects (e.g. load and call are not supported).
2772 The following is the syntax for constant expressions:
2774 ``trunc (CST to TYPE)``
2775 Truncate a constant to another type. The bit size of CST must be
2776 larger than the bit size of TYPE. Both types must be integers.
2777 ``zext (CST to TYPE)``
2778 Zero extend a constant to another type. The bit size of CST must be
2779 smaller than the bit size of TYPE. Both types must be integers.
2780 ``sext (CST to TYPE)``
2781 Sign extend a constant to another type. The bit size of CST must be
2782 smaller than the bit size of TYPE. Both types must be integers.
2783 ``fptrunc (CST to TYPE)``
2784 Truncate a floating point constant to another floating point type.
2785 The size of CST must be larger than the size of TYPE. Both types
2786 must be floating point.
2787 ``fpext (CST to TYPE)``
2788 Floating point extend a constant to another type. The size of CST
2789 must be smaller or equal to the size of TYPE. Both types must be
2791 ``fptoui (CST to TYPE)``
2792 Convert a floating point constant to the corresponding unsigned
2793 integer constant. TYPE must be a scalar or vector integer type. CST
2794 must be of scalar or vector floating point type. Both CST and TYPE
2795 must be scalars, or vectors of the same number of elements. If the
2796 value won't fit in the integer type, the results are undefined.
2797 ``fptosi (CST to TYPE)``
2798 Convert a floating point constant to the corresponding signed
2799 integer constant. TYPE must be a scalar or vector integer type. CST
2800 must be of scalar or vector floating point type. Both CST and TYPE
2801 must be scalars, or vectors of the same number of elements. If the
2802 value won't fit in the integer type, the results are undefined.
2803 ``uitofp (CST to TYPE)``
2804 Convert an unsigned integer constant to the corresponding floating
2805 point constant. TYPE must be a scalar or vector floating point type.
2806 CST must be of scalar or vector integer type. Both CST and TYPE must
2807 be scalars, or vectors of the same number of elements. If the value
2808 won't fit in the floating point type, the results are undefined.
2809 ``sitofp (CST to TYPE)``
2810 Convert a signed integer constant to the corresponding floating
2811 point constant. TYPE must be a scalar or vector floating point type.
2812 CST must be of scalar or vector integer type. Both CST and TYPE must
2813 be scalars, or vectors of the same number of elements. If the value
2814 won't fit in the floating point type, the results are undefined.
2815 ``ptrtoint (CST to TYPE)``
2816 Convert a pointer typed constant to the corresponding integer
2817 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2818 pointer type. The ``CST`` value is zero extended, truncated, or
2819 unchanged to make it fit in ``TYPE``.
2820 ``inttoptr (CST to TYPE)``
2821 Convert an integer constant to a pointer constant. TYPE must be a
2822 pointer type. CST must be of integer type. The CST value is zero
2823 extended, truncated, or unchanged to make it fit in a pointer size.
2824 This one is *really* dangerous!
2825 ``bitcast (CST to TYPE)``
2826 Convert a constant, CST, to another TYPE. The constraints of the
2827 operands are the same as those for the :ref:`bitcast
2828 instruction <i_bitcast>`.
2829 ``addrspacecast (CST to TYPE)``
2830 Convert a constant pointer or constant vector of pointer, CST, to another
2831 TYPE in a different address space. The constraints of the operands are the
2832 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2833 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2834 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2835 constants. As with the :ref:`getelementptr <i_getelementptr>`
2836 instruction, the index list may have zero or more indexes, which are
2837 required to make sense for the type of "pointer to TY".
2838 ``select (COND, VAL1, VAL2)``
2839 Perform the :ref:`select operation <i_select>` on constants.
2840 ``icmp COND (VAL1, VAL2)``
2841 Performs the :ref:`icmp operation <i_icmp>` on constants.
2842 ``fcmp COND (VAL1, VAL2)``
2843 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2844 ``extractelement (VAL, IDX)``
2845 Perform the :ref:`extractelement operation <i_extractelement>` on
2847 ``insertelement (VAL, ELT, IDX)``
2848 Perform the :ref:`insertelement operation <i_insertelement>` on
2850 ``shufflevector (VEC1, VEC2, IDXMASK)``
2851 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2853 ``extractvalue (VAL, IDX0, IDX1, ...)``
2854 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2855 constants. The index list is interpreted in a similar manner as
2856 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2857 least one index value must be specified.
2858 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2859 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2860 The index list is interpreted in a similar manner as indices in a
2861 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2862 value must be specified.
2863 ``OPCODE (LHS, RHS)``
2864 Perform the specified operation of the LHS and RHS constants. OPCODE
2865 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2866 binary <bitwiseops>` operations. The constraints on operands are
2867 the same as those for the corresponding instruction (e.g. no bitwise
2868 operations on floating point values are allowed).
2875 Inline Assembler Expressions
2876 ----------------------------
2878 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2879 Inline Assembly <moduleasm>`) through the use of a special value. This value
2880 represents the inline assembler as a template string (containing the
2881 instructions to emit), a list of operand constraints (stored as a string), a
2882 flag that indicates whether or not the inline asm expression has side effects,
2883 and a flag indicating whether the function containing the asm needs to align its
2884 stack conservatively.
2886 The template string supports argument substitution of the operands using "``$``"
2887 followed by a number, to indicate substitution of the given register/memory
2888 location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
2889 be used, where ``MODIFIER`` is a target-specific annotation for how to print the
2890 operand (See :ref:`inline-asm-modifiers`).
2892 A literal "``$``" may be included by using "``$$``" in the template. To include
2893 other special characters into the output, the usual "``\XX``" escapes may be
2894 used, just as in other strings. Note that after template substitution, the
2895 resulting assembly string is parsed by LLVM's integrated assembler unless it is
2896 disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
2897 syntax known to LLVM.
2899 LLVM's support for inline asm is modeled closely on the requirements of Clang's
2900 GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
2901 modifier codes listed here are similar or identical to those in GCC's inline asm
2902 support. However, to be clear, the syntax of the template and constraint strings
2903 described here is *not* the same as the syntax accepted by GCC and Clang, and,
2904 while most constraint letters are passed through as-is by Clang, some get
2905 translated to other codes when converting from the C source to the LLVM
2908 An example inline assembler expression is:
2910 .. code-block:: llvm
2912 i32 (i32) asm "bswap $0", "=r,r"
2914 Inline assembler expressions may **only** be used as the callee operand
2915 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2916 Thus, typically we have:
2918 .. code-block:: llvm
2920 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2922 Inline asms with side effects not visible in the constraint list must be
2923 marked as having side effects. This is done through the use of the
2924 '``sideeffect``' keyword, like so:
2926 .. code-block:: llvm
2928 call void asm sideeffect "eieio", ""()
2930 In some cases inline asms will contain code that will not work unless
2931 the stack is aligned in some way, such as calls or SSE instructions on
2932 x86, yet will not contain code that does that alignment within the asm.
2933 The compiler should make conservative assumptions about what the asm
2934 might contain and should generate its usual stack alignment code in the
2935 prologue if the '``alignstack``' keyword is present:
2937 .. code-block:: llvm
2939 call void asm alignstack "eieio", ""()
2941 Inline asms also support using non-standard assembly dialects. The
2942 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2943 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2944 the only supported dialects. An example is:
2946 .. code-block:: llvm
2948 call void asm inteldialect "eieio", ""()
2950 If multiple keywords appear the '``sideeffect``' keyword must come
2951 first, the '``alignstack``' keyword second and the '``inteldialect``'
2954 Inline Asm Constraint String
2955 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2957 The constraint list is a comma-separated string, each element containing one or
2958 more constraint codes.
2960 For each element in the constraint list an appropriate register or memory
2961 operand will be chosen, and it will be made available to assembly template
2962 string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
2965 There are three different types of constraints, which are distinguished by a
2966 prefix symbol in front of the constraint code: Output, Input, and Clobber. The
2967 constraints must always be given in that order: outputs first, then inputs, then
2968 clobbers. They cannot be intermingled.
2970 There are also three different categories of constraint codes:
2972 - Register constraint. This is either a register class, or a fixed physical
2973 register. This kind of constraint will allocate a register, and if necessary,
2974 bitcast the argument or result to the appropriate type.
2975 - Memory constraint. This kind of constraint is for use with an instruction
2976 taking a memory operand. Different constraints allow for different addressing
2977 modes used by the target.
2978 - Immediate value constraint. This kind of constraint is for an integer or other
2979 immediate value which can be rendered directly into an instruction. The
2980 various target-specific constraints allow the selection of a value in the
2981 proper range for the instruction you wish to use it with.
2986 Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
2987 indicates that the assembly will write to this operand, and the operand will
2988 then be made available as a return value of the ``asm`` expression. Output
2989 constraints do not consume an argument from the call instruction. (Except, see
2990 below about indirect outputs).
2992 Normally, it is expected that no output locations are written to by the assembly
2993 expression until *all* of the inputs have been read. As such, LLVM may assign
2994 the same register to an output and an input. If this is not safe (e.g. if the
2995 assembly contains two instructions, where the first writes to one output, and
2996 the second reads an input and writes to a second output), then the "``&``"
2997 modifier must be used (e.g. "``=&r``") to specify that the output is an
2998 "early-clobber" output. Marking an ouput as "early-clobber" ensures that LLVM
2999 will not use the same register for any inputs (other than an input tied to this
3005 Input constraints do not have a prefix -- just the constraint codes. Each input
3006 constraint will consume one argument from the call instruction. It is not
3007 permitted for the asm to write to any input register or memory location (unless
3008 that input is tied to an output). Note also that multiple inputs may all be
3009 assigned to the same register, if LLVM can determine that they necessarily all
3010 contain the same value.
3012 Instead of providing a Constraint Code, input constraints may also "tie"
3013 themselves to an output constraint, by providing an integer as the constraint
3014 string. Tied inputs still consume an argument from the call instruction, and
3015 take up a position in the asm template numbering as is usual -- they will simply
3016 be constrained to always use the same register as the output they've been tied
3017 to. For example, a constraint string of "``=r,0``" says to assign a register for
3018 output, and use that register as an input as well (it being the 0'th
3021 It is permitted to tie an input to an "early-clobber" output. In that case, no
3022 *other* input may share the same register as the input tied to the early-clobber
3023 (even when the other input has the same value).
3025 You may only tie an input to an output which has a register constraint, not a
3026 memory constraint. Only a single input may be tied to an output.
3028 There is also an "interesting" feature which deserves a bit of explanation: if a
3029 register class constraint allocates a register which is too small for the value
3030 type operand provided as input, the input value will be split into multiple
3031 registers, and all of them passed to the inline asm.
3033 However, this feature is often not as useful as you might think.
3035 Firstly, the registers are *not* guaranteed to be consecutive. So, on those
3036 architectures that have instructions which operate on multiple consecutive
3037 instructions, this is not an appropriate way to support them. (e.g. the 32-bit
3038 SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
3039 hardware then loads into both the named register, and the next register. This
3040 feature of inline asm would not be useful to support that.)
3042 A few of the targets provide a template string modifier allowing explicit access
3043 to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
3044 ``D``). On such an architecture, you can actually access the second allocated
3045 register (yet, still, not any subsequent ones). But, in that case, you're still
3046 probably better off simply splitting the value into two separate operands, for
3047 clarity. (e.g. see the description of the ``A`` constraint on X86, which,
3048 despite existing only for use with this feature, is not really a good idea to
3051 Indirect inputs and outputs
3052 """""""""""""""""""""""""""
3054 Indirect output or input constraints can be specified by the "``*``" modifier
3055 (which goes after the "``=``" in case of an output). This indicates that the asm
3056 will write to or read from the contents of an *address* provided as an input
3057 argument. (Note that in this way, indirect outputs act more like an *input* than
3058 an output: just like an input, they consume an argument of the call expression,
3059 rather than producing a return value. An indirect output constraint is an
3060 "output" only in that the asm is expected to write to the contents of the input
3061 memory location, instead of just read from it).
3063 This is most typically used for memory constraint, e.g. "``=*m``", to pass the
3064 address of a variable as a value.
3066 It is also possible to use an indirect *register* constraint, but only on output
3067 (e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3068 value normally, and then, separately emit a store to the address provided as
3069 input, after the provided inline asm. (It's not clear what value this
3070 functionality provides, compared to writing the store explicitly after the asm
3071 statement, and it can only produce worse code, since it bypasses many
3072 optimization passes. I would recommend not using it.)
3078 A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3079 consume an input operand, nor generate an output. Clobbers cannot use any of the
3080 general constraint code letters -- they may use only explicit register
3081 constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3082 "``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3083 memory locations -- not only the memory pointed to by a declared indirect
3089 After a potential prefix comes constraint code, or codes.
3091 A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3092 followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3095 The one and two letter constraint codes are typically chosen to be the same as
3096 GCC's constraint codes.
3098 A single constraint may include one or more than constraint code in it, leaving
3099 it up to LLVM to choose which one to use. This is included mainly for
3100 compatibility with the translation of GCC inline asm coming from clang.
3102 There are two ways to specify alternatives, and either or both may be used in an
3103 inline asm constraint list:
3105 1) Append the codes to each other, making a constraint code set. E.g. "``im``"
3106 or "``{eax}m``". This means "choose any of the options in the set". The
3107 choice of constraint is made independently for each constraint in the
3110 2) Use "``|``" between constraint code sets, creating alternatives. Every
3111 constraint in the constraint list must have the same number of alternative
3112 sets. With this syntax, the same alternative in *all* of the items in the
3113 constraint list will be chosen together.
3115 Putting those together, you might have a two operand constraint string like
3116 ``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3117 operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3118 may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3120 However, the use of either of the alternatives features is *NOT* recommended, as
3121 LLVM is not able to make an intelligent choice about which one to use. (At the
3122 point it currently needs to choose, not enough information is available to do so
3123 in a smart way.) Thus, it simply tries to make a choice that's most likely to
3124 compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3125 always choose to use memory, not registers). And, if given multiple registers,
3126 or multiple register classes, it will simply choose the first one. (In fact, it
3127 doesn't currently even ensure explicitly specified physical registers are
3128 unique, so specifying multiple physical registers as alternatives, like
3129 ``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3132 Supported Constraint Code List
3133 """"""""""""""""""""""""""""""
3135 The constraint codes are, in general, expected to behave the same way they do in
3136 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3137 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3138 and GCC likely indicates a bug in LLVM.
3140 Some constraint codes are typically supported by all targets:
3142 - ``r``: A register in the target's general purpose register class.
3143 - ``m``: A memory address operand. It is target-specific what addressing modes
3144 are supported, typical examples are register, or register + register offset,
3145 or register + immediate offset (of some target-specific size).
3146 - ``i``: An integer constant (of target-specific width). Allows either a simple
3147 immediate, or a relocatable value.
3148 - ``n``: An integer constant -- *not* including relocatable values.
3149 - ``s``: An integer constant, but allowing *only* relocatable values.
3150 - ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3151 useful to pass a label for an asm branch or call.
3153 .. FIXME: but that surely isn't actually okay to jump out of an asm
3154 block without telling llvm about the control transfer???)
3156 - ``{register-name}``: Requires exactly the named physical register.
3158 Other constraints are target-specific:
3162 - ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3163 - ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3164 i.e. 0 to 4095 with optional shift by 12.
3165 - ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3166 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3167 - ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3168 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3169 - ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3170 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3171 - ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3172 32-bit register. This is a superset of ``K``: in addition to the bitmask
3173 immediate, also allows immediate integers which can be loaded with a single
3174 ``MOVZ`` or ``MOVL`` instruction.
3175 - ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3176 64-bit register. This is a superset of ``L``.
3177 - ``Q``: Memory address operand must be in a single register (no
3178 offsets). (However, LLVM currently does this for the ``m`` constraint as
3180 - ``r``: A 32 or 64-bit integer register (W* or X*).
3181 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3182 - ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3186 - ``r``: A 32 or 64-bit integer register.
3187 - ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3188 - ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3193 - ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3194 operand. Treated the same as operand ``m``, at the moment.
3196 ARM and ARM's Thumb2 mode:
3198 - ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3199 - ``I``: An immediate integer valid for a data-processing instruction.
3200 - ``J``: An immediate integer between -4095 and 4095.
3201 - ``K``: An immediate integer whose bitwise inverse is valid for a
3202 data-processing instruction. (Can be used with template modifier "``B``" to
3203 print the inverted value).
3204 - ``L``: An immediate integer whose negation is valid for a data-processing
3205 instruction. (Can be used with template modifier "``n``" to print the negated
3207 - ``M``: A power of two or a integer between 0 and 32.
3208 - ``N``: Invalid immediate constraint.
3209 - ``O``: Invalid immediate constraint.
3210 - ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3211 - ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3213 - ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3215 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3216 ``d0-d31``, or ``q0-q15``.
3217 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3218 ``d0-d7``, or ``q0-q3``.
3219 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3224 - ``I``: An immediate integer between 0 and 255.
3225 - ``J``: An immediate integer between -255 and -1.
3226 - ``K``: An immediate integer between 0 and 255, with optional left-shift by
3228 - ``L``: An immediate integer between -7 and 7.
3229 - ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3230 - ``N``: An immediate integer between 0 and 31.
3231 - ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3232 - ``r``: A low 32-bit GPR register (``r0-r7``).
3233 - ``l``: A low 32-bit GPR register (``r0-r7``).
3234 - ``h``: A high GPR register (``r0-r7``).
3235 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3236 ``d0-d31``, or ``q0-q15``.
3237 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3238 ``d0-d7``, or ``q0-q3``.
3239 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3245 - ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3247 - ``r``: A 32 or 64-bit register.
3251 - ``r``: An 8 or 16-bit register.
3255 - ``I``: An immediate signed 16-bit integer.
3256 - ``J``: An immediate integer zero.
3257 - ``K``: An immediate unsigned 16-bit integer.
3258 - ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3259 - ``N``: An immediate integer between -65535 and -1.
3260 - ``O``: An immediate signed 15-bit integer.
3261 - ``P``: An immediate integer between 1 and 65535.
3262 - ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3263 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3264 - ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3265 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3267 - ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3268 ``sc`` instruction on the given subtarget (details vary).
3269 - ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3270 - ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
3271 (``W0-W31``). In the case of MSA registers, it is recommended to use the ``w``
3272 argument modifier for compatibility with GCC.
3273 - ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3275 - ``l``: The ``lo`` register, 32 or 64-bit.
3280 - ``b``: A 1-bit integer register.
3281 - ``c`` or ``h``: A 16-bit integer register.
3282 - ``r``: A 32-bit integer register.
3283 - ``l`` or ``N``: A 64-bit integer register.
3284 - ``f``: A 32-bit float register.
3285 - ``d``: A 64-bit float register.
3290 - ``I``: An immediate signed 16-bit integer.
3291 - ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3292 - ``K``: An immediate unsigned 16-bit integer.
3293 - ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3294 - ``M``: An immediate integer greater than 31.
3295 - ``N``: An immediate integer that is an exact power of 2.
3296 - ``O``: The immediate integer constant 0.
3297 - ``P``: An immediate integer constant whose negation is a signed 16-bit
3299 - ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3300 treated the same as ``m``.
3301 - ``r``: A 32 or 64-bit integer register.
3302 - ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3304 - ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3305 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3306 - ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3307 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3308 altivec vector register (``V0-V31``).
3310 .. FIXME: is this a bug that v accepts QPX registers? I think this
3311 is supposed to only use the altivec vector registers?
3313 - ``y``: Condition register (``CR0-CR7``).
3314 - ``wc``: An individual CR bit in a CR register.
3315 - ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3316 register set (overlapping both the floating-point and vector register files).
3317 - ``ws``: A 32 or 64-bit floating point register, from the full VSX register
3322 - ``I``: An immediate 13-bit signed integer.
3323 - ``r``: A 32-bit integer register.
3327 - ``I``: An immediate unsigned 8-bit integer.
3328 - ``J``: An immediate unsigned 12-bit integer.
3329 - ``K``: An immediate signed 16-bit integer.
3330 - ``L``: An immediate signed 20-bit integer.
3331 - ``M``: An immediate integer 0x7fffffff.
3332 - ``Q``, ``R``, ``S``, ``T``: A memory address operand, treated the same as
3333 ``m``, at the moment.
3334 - ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3335 - ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3336 address context evaluates as zero).
3337 - ``h``: A 32-bit value in the high part of a 64bit data register
3339 - ``f``: A 32, 64, or 128-bit floating point register.
3343 - ``I``: An immediate integer between 0 and 31.
3344 - ``J``: An immediate integer between 0 and 64.
3345 - ``K``: An immediate signed 8-bit integer.
3346 - ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3348 - ``M``: An immediate integer between 0 and 3.
3349 - ``N``: An immediate unsigned 8-bit integer.
3350 - ``O``: An immediate integer between 0 and 127.
3351 - ``e``: An immediate 32-bit signed integer.
3352 - ``Z``: An immediate 32-bit unsigned integer.
3353 - ``o``, ``v``: Treated the same as ``m``, at the moment.
3354 - ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3355 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3356 registers, and on X86-64, it is all of the integer registers.
3357 - ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3358 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3359 - ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3360 - ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3361 existed since i386, and can be accessed without the REX prefix.
3362 - ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3363 - ``y``: A 64-bit MMX register, if MMX is enabled.
3364 - ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3365 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3366 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3367 512-bit vector operand in an AVX512 register, Otherwise, an error.
3368 - ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3369 - ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3370 32-bit mode, a 64-bit integer operand will get split into two registers). It
3371 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3372 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3373 you're better off splitting it yourself, before passing it to the asm
3378 - ``r``: A 32-bit integer register.
3381 .. _inline-asm-modifiers:
3383 Asm template argument modifiers
3384 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3386 In the asm template string, modifiers can be used on the operand reference, like
3389 The modifiers are, in general, expected to behave the same way they do in
3390 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3391 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3392 and GCC likely indicates a bug in LLVM.
3396 - ``c``: Print an immediate integer constant unadorned, without
3397 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3398 - ``n``: Negate and print immediate integer constant unadorned, without the
3399 target-specific immediate punctuation (e.g. no ``$`` prefix).
3400 - ``l``: Print as an unadorned label, without the target-specific label
3401 punctuation (e.g. no ``$`` prefix).
3405 - ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3406 instead of ``x30``, print ``w30``.
3407 - ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3408 - ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3409 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3418 - ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3422 - ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3423 as ``d4[1]`` instead of ``s9``)
3424 - ``B``: Bitwise invert and print an immediate integer constant without ``#``
3426 - ``L``: Print the low 16-bits of an immediate integer constant.
3427 - ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3428 register operands subsequent to the specified one (!), so use carefully.
3429 - ``Q``: Print the low-order register of a register-pair, or the low-order
3430 register of a two-register operand.
3431 - ``R``: Print the high-order register of a register-pair, or the high-order
3432 register of a two-register operand.
3433 - ``H``: Print the second register of a register-pair. (On a big-endian system,
3434 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3437 .. FIXME: H doesn't currently support printing the second register
3438 of a two-register operand.
3440 - ``e``: Print the low doubleword register of a NEON quad register.
3441 - ``f``: Print the high doubleword register of a NEON quad register.
3442 - ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3447 - ``L``: Print the second register of a two-register operand. Requires that it
3448 has been allocated consecutively to the first.
3450 .. FIXME: why is it restricted to consecutive ones? And there's
3451 nothing that ensures that happens, is there?
3453 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3454 nothing. Used to print 'addi' vs 'add' instructions.
3458 No additional modifiers.
3462 - ``X``: Print an immediate integer as hexadecimal
3463 - ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3464 - ``d``: Print an immediate integer as decimal.
3465 - ``m``: Subtract one and print an immediate integer as decimal.
3466 - ``z``: Print $0 if an immediate zero, otherwise print normally.
3467 - ``L``: Print the low-order register of a two-register operand, or prints the
3468 address of the low-order word of a double-word memory operand.
3470 .. FIXME: L seems to be missing memory operand support.
3472 - ``M``: Print the high-order register of a two-register operand, or prints the
3473 address of the high-order word of a double-word memory operand.
3475 .. FIXME: M seems to be missing memory operand support.
3477 - ``D``: Print the second register of a two-register operand, or prints the
3478 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3479 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3481 - ``w``: No effect. Provided for compatibility with GCC which requires this
3482 modifier in order to print MSA registers (``W0-W31``) with the ``f``
3491 - ``L``: Print the second register of a two-register operand. Requires that it
3492 has been allocated consecutively to the first.
3494 .. FIXME: why is it restricted to consecutive ones? And there's
3495 nothing that ensures that happens, is there?
3497 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3498 nothing. Used to print 'addi' vs 'add' instructions.
3499 - ``y``: For a memory operand, prints formatter for a two-register X-form
3500 instruction. (Currently always prints ``r0,OPERAND``).
3501 - ``U``: Prints 'u' if the memory operand is an update form, and nothing
3502 otherwise. (NOTE: LLVM does not support update form, so this will currently
3503 always print nothing)
3504 - ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
3505 not support indexed form, so this will currently always print nothing)
3513 SystemZ implements only ``n``, and does *not* support any of the other
3514 target-independent modifiers.
3518 - ``c``: Print an unadorned integer or symbol name. (The latter is
3519 target-specific behavior for this typically target-independent modifier).
3520 - ``A``: Print a register name with a '``*``' before it.
3521 - ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
3523 - ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
3525 - ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
3527 - ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
3529 - ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
3530 available, otherwise the 32-bit register name; do nothing on a memory operand.
3531 - ``n``: Negate and print an unadorned integer, or, for operands other than an
3532 immediate integer (e.g. a relocatable symbol expression), print a '-' before
3533 the operand. (The behavior for relocatable symbol expressions is a
3534 target-specific behavior for this typically target-independent modifier)
3535 - ``H``: Print a memory reference with additional offset +8.
3536 - ``P``: Print a memory reference or operand for use as the argument of a call
3537 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
3541 No additional modifiers.
3547 The call instructions that wrap inline asm nodes may have a
3548 "``!srcloc``" MDNode attached to it that contains a list of constant
3549 integers. If present, the code generator will use the integer as the
3550 location cookie value when report errors through the ``LLVMContext``
3551 error reporting mechanisms. This allows a front-end to correlate backend
3552 errors that occur with inline asm back to the source code that produced
3555 .. code-block:: llvm
3557 call void asm sideeffect "something bad", ""(), !srcloc !42
3559 !42 = !{ i32 1234567 }
3561 It is up to the front-end to make sense of the magic numbers it places
3562 in the IR. If the MDNode contains multiple constants, the code generator
3563 will use the one that corresponds to the line of the asm that the error
3571 LLVM IR allows metadata to be attached to instructions in the program
3572 that can convey extra information about the code to the optimizers and
3573 code generator. One example application of metadata is source-level
3574 debug information. There are two metadata primitives: strings and nodes.
3576 Metadata does not have a type, and is not a value. If referenced from a
3577 ``call`` instruction, it uses the ``metadata`` type.
3579 All metadata are identified in syntax by a exclamation point ('``!``').
3581 .. _metadata-string:
3583 Metadata Nodes and Metadata Strings
3584 -----------------------------------
3586 A metadata string is a string surrounded by double quotes. It can
3587 contain any character by escaping non-printable characters with
3588 "``\xx``" where "``xx``" is the two digit hex code. For example:
3591 Metadata nodes are represented with notation similar to structure
3592 constants (a comma separated list of elements, surrounded by braces and
3593 preceded by an exclamation point). Metadata nodes can have any values as
3594 their operand. For example:
3596 .. code-block:: llvm
3598 !{ !"test\00", i32 10}
3600 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
3602 .. code-block:: llvm
3604 !0 = distinct !{!"test\00", i32 10}
3606 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
3607 content. They can also occur when transformations cause uniquing collisions
3608 when metadata operands change.
3610 A :ref:`named metadata <namedmetadatastructure>` is a collection of
3611 metadata nodes, which can be looked up in the module symbol table. For
3614 .. code-block:: llvm
3618 Metadata can be used as function arguments. Here ``llvm.dbg.value``
3619 function is using two metadata arguments:
3621 .. code-block:: llvm
3623 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3625 Metadata can be attached with an instruction. Here metadata ``!21`` is
3626 attached to the ``add`` instruction using the ``!dbg`` identifier:
3628 .. code-block:: llvm
3630 %indvar.next = add i64 %indvar, 1, !dbg !21
3632 More information about specific metadata nodes recognized by the
3633 optimizers and code generator is found below.
3635 .. _specialized-metadata:
3637 Specialized Metadata Nodes
3638 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3640 Specialized metadata nodes are custom data structures in metadata (as opposed
3641 to generic tuples). Their fields are labelled, and can be specified in any
3644 These aren't inherently debug info centric, but currently all the specialized
3645 metadata nodes are related to debug info.
3652 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
3653 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
3654 tuples containing the debug info to be emitted along with the compile unit,
3655 regardless of code optimizations (some nodes are only emitted if there are
3656 references to them from instructions).
3658 .. code-block:: llvm
3660 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
3661 isOptimized: true, flags: "-O2", runtimeVersion: 2,
3662 splitDebugFilename: "abc.debug", emissionKind: 1,
3663 enums: !2, retainedTypes: !3, subprograms: !4,
3664 globals: !5, imports: !6)
3666 Compile unit descriptors provide the root scope for objects declared in a
3667 specific compilation unit. File descriptors are defined using this scope.
3668 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
3669 keep track of subprograms, global variables, type information, and imported
3670 entities (declarations and namespaces).
3677 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
3679 .. code-block:: llvm
3681 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
3683 Files are sometimes used in ``scope:`` fields, and are the only valid target
3684 for ``file:`` fields.
3691 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
3692 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
3694 .. code-block:: llvm
3696 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3697 encoding: DW_ATE_unsigned_char)
3698 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
3700 The ``encoding:`` describes the details of the type. Usually it's one of the
3703 .. code-block:: llvm
3709 DW_ATE_signed_char = 6
3711 DW_ATE_unsigned_char = 8
3713 .. _DISubroutineType:
3718 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3719 refers to a tuple; the first operand is the return type, while the rest are the
3720 types of the formal arguments in order. If the first operand is ``null``, that
3721 represents a function with no return value (such as ``void foo() {}`` in C++).
3723 .. code-block:: llvm
3725 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3726 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3727 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3734 ``DIDerivedType`` nodes represent types derived from other types, such as
3737 .. code-block:: llvm
3739 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3740 encoding: DW_ATE_unsigned_char)
3741 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3744 The following ``tag:`` values are valid:
3746 .. code-block:: llvm
3748 DW_TAG_formal_parameter = 5
3750 DW_TAG_pointer_type = 15
3751 DW_TAG_reference_type = 16
3753 DW_TAG_ptr_to_member_type = 31
3754 DW_TAG_const_type = 38
3755 DW_TAG_volatile_type = 53
3756 DW_TAG_restrict_type = 55
3758 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3759 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3760 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3761 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3762 argument of a subprogram.
3764 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3766 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3767 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3770 Note that the ``void *`` type is expressed as a type derived from NULL.
3772 .. _DICompositeType:
3777 ``DICompositeType`` nodes represent types composed of other types, like
3778 structures and unions. ``elements:`` points to a tuple of the composed types.
3780 If the source language supports ODR, the ``identifier:`` field gives the unique
3781 identifier used for type merging between modules. When specified, other types
3782 can refer to composite types indirectly via a :ref:`metadata string
3783 <metadata-string>` that matches their identifier.
3785 .. code-block:: llvm
3787 !0 = !DIEnumerator(name: "SixKind", value: 7)
3788 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3789 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3790 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3791 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3792 elements: !{!0, !1, !2})
3794 The following ``tag:`` values are valid:
3796 .. code-block:: llvm
3798 DW_TAG_array_type = 1
3799 DW_TAG_class_type = 2
3800 DW_TAG_enumeration_type = 4
3801 DW_TAG_structure_type = 19
3802 DW_TAG_union_type = 23
3803 DW_TAG_subroutine_type = 21
3804 DW_TAG_inheritance = 28
3807 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3808 descriptors <DISubrange>`, each representing the range of subscripts at that
3809 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3810 array type is a native packed vector.
3812 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3813 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3814 value for the set. All enumeration type descriptors are collected in the
3815 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3817 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3818 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3819 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3826 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3827 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3829 .. code-block:: llvm
3831 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3832 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3833 !2 = !DISubrange(count: -1) ; empty array.
3840 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3841 variants of :ref:`DICompositeType`.
3843 .. code-block:: llvm
3845 !0 = !DIEnumerator(name: "SixKind", value: 7)
3846 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3847 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3849 DITemplateTypeParameter
3850 """""""""""""""""""""""
3852 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3853 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3854 :ref:`DISubprogram` ``templateParams:`` fields.
3856 .. code-block:: llvm
3858 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3860 DITemplateValueParameter
3861 """"""""""""""""""""""""
3863 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3864 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3865 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3866 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3867 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3869 .. code-block:: llvm
3871 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3876 ``DINamespace`` nodes represent namespaces in the source language.
3878 .. code-block:: llvm
3880 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3885 ``DIGlobalVariable`` nodes represent global variables in the source language.
3887 .. code-block:: llvm
3889 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3890 file: !2, line: 7, type: !3, isLocal: true,
3891 isDefinition: false, variable: i32* @foo,
3894 All global variables should be referenced by the `globals:` field of a
3895 :ref:`compile unit <DICompileUnit>`.
3902 ``DISubprogram`` nodes represent functions from the source language. The
3903 ``variables:`` field points at :ref:`variables <DILocalVariable>` that must be
3904 retained, even if their IR counterparts are optimized out of the IR. The
3905 ``type:`` field must point at an :ref:`DISubroutineType`.
3907 .. code-block:: llvm
3909 !0 = !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3910 file: !2, line: 7, type: !3, isLocal: true,
3911 isDefinition: false, scopeLine: 8, containingType: !4,
3912 virtuality: DW_VIRTUALITY_pure_virtual, virtualIndex: 10,
3913 flags: DIFlagPrototyped, isOptimized: true,
3914 function: void ()* @_Z3foov,
3915 templateParams: !5, declaration: !6, variables: !7)
3922 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3923 <DISubprogram>`. The line number and column numbers are used to distinguish
3924 two lexical blocks at same depth. They are valid targets for ``scope:``
3927 .. code-block:: llvm
3929 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3931 Usually lexical blocks are ``distinct`` to prevent node merging based on
3934 .. _DILexicalBlockFile:
3939 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3940 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
3941 indicate textual inclusion, or the ``discriminator:`` field can be used to
3942 discriminate between control flow within a single block in the source language.
3944 .. code-block:: llvm
3946 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3947 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3948 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3955 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
3956 mandatory, and points at an :ref:`DILexicalBlockFile`, an
3957 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
3959 .. code-block:: llvm
3961 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3963 .. _DILocalVariable:
3968 ``DILocalVariable`` nodes represent local variables in the source language. If
3969 the ``arg:`` field is set to non-zero, then this variable is a subprogram
3970 parameter, and it will be included in the ``variables:`` field of its
3971 :ref:`DISubprogram`.
3973 .. code-block:: llvm
3975 !0 = !DILocalVariable(name: "this", arg: 1, scope: !3, file: !2, line: 7,
3976 type: !3, flags: DIFlagArtificial)
3977 !1 = !DILocalVariable(name: "x", arg: 2, scope: !4, file: !2, line: 7,
3979 !2 = !DILocalVariable(name: "y", scope: !5, file: !2, line: 7, type: !3)
3984 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
3985 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
3986 describe how the referenced LLVM variable relates to the source language
3989 The current supported vocabulary is limited:
3991 - ``DW_OP_deref`` dereferences the working expression.
3992 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
3993 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
3994 here, respectively) of the variable piece from the working expression.
3996 .. code-block:: llvm
3998 !0 = !DIExpression(DW_OP_deref)
3999 !1 = !DIExpression(DW_OP_plus, 3)
4000 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
4001 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
4006 ``DIObjCProperty`` nodes represent Objective-C property nodes.
4008 .. code-block:: llvm
4010 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
4011 getter: "getFoo", attributes: 7, type: !2)
4016 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
4019 .. code-block:: llvm
4021 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
4022 entity: !1, line: 7)
4027 In LLVM IR, memory does not have types, so LLVM's own type system is not
4028 suitable for doing TBAA. Instead, metadata is added to the IR to
4029 describe a type system of a higher level language. This can be used to
4030 implement typical C/C++ TBAA, but it can also be used to implement
4031 custom alias analysis behavior for other languages.
4033 The current metadata format is very simple. TBAA metadata nodes have up
4034 to three fields, e.g.:
4036 .. code-block:: llvm
4038 !0 = !{ !"an example type tree" }
4039 !1 = !{ !"int", !0 }
4040 !2 = !{ !"float", !0 }
4041 !3 = !{ !"const float", !2, i64 1 }
4043 The first field is an identity field. It can be any value, usually a
4044 metadata string, which uniquely identifies the type. The most important
4045 name in the tree is the name of the root node. Two trees with different
4046 root node names are entirely disjoint, even if they have leaves with
4049 The second field identifies the type's parent node in the tree, or is
4050 null or omitted for a root node. A type is considered to alias all of
4051 its descendants and all of its ancestors in the tree. Also, a type is
4052 considered to alias all types in other trees, so that bitcode produced
4053 from multiple front-ends is handled conservatively.
4055 If the third field is present, it's an integer which if equal to 1
4056 indicates that the type is "constant" (meaning
4057 ``pointsToConstantMemory`` should return true; see `other useful
4058 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
4060 '``tbaa.struct``' Metadata
4061 ^^^^^^^^^^^^^^^^^^^^^^^^^^
4063 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
4064 aggregate assignment operations in C and similar languages, however it
4065 is defined to copy a contiguous region of memory, which is more than
4066 strictly necessary for aggregate types which contain holes due to
4067 padding. Also, it doesn't contain any TBAA information about the fields
4070 ``!tbaa.struct`` metadata can describe which memory subregions in a
4071 memcpy are padding and what the TBAA tags of the struct are.
4073 The current metadata format is very simple. ``!tbaa.struct`` metadata
4074 nodes are a list of operands which are in conceptual groups of three.
4075 For each group of three, the first operand gives the byte offset of a
4076 field in bytes, the second gives its size in bytes, and the third gives
4079 .. code-block:: llvm
4081 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
4083 This describes a struct with two fields. The first is at offset 0 bytes
4084 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4085 and has size 4 bytes and has tbaa tag !2.
4087 Note that the fields need not be contiguous. In this example, there is a
4088 4 byte gap between the two fields. This gap represents padding which
4089 does not carry useful data and need not be preserved.
4091 '``noalias``' and '``alias.scope``' Metadata
4092 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4094 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4095 noalias memory-access sets. This means that some collection of memory access
4096 instructions (loads, stores, memory-accessing calls, etc.) that carry
4097 ``noalias`` metadata can specifically be specified not to alias with some other
4098 collection of memory access instructions that carry ``alias.scope`` metadata.
4099 Each type of metadata specifies a list of scopes where each scope has an id and
4100 a domain. When evaluating an aliasing query, if for some domain, the set
4101 of scopes with that domain in one instruction's ``alias.scope`` list is a
4102 subset of (or equal to) the set of scopes for that domain in another
4103 instruction's ``noalias`` list, then the two memory accesses are assumed not to
4106 The metadata identifying each domain is itself a list containing one or two
4107 entries. The first entry is the name of the domain. Note that if the name is a
4108 string then it can be combined across functions and translation units. A
4109 self-reference can be used to create globally unique domain names. A
4110 descriptive string may optionally be provided as a second list entry.
4112 The metadata identifying each scope is also itself a list containing two or
4113 three entries. The first entry is the name of the scope. Note that if the name
4114 is a string then it can be combined across functions and translation units. A
4115 self-reference can be used to create globally unique scope names. A metadata
4116 reference to the scope's domain is the second entry. A descriptive string may
4117 optionally be provided as a third list entry.
4121 .. code-block:: llvm
4123 ; Two scope domains:
4127 ; Some scopes in these domains:
4133 !5 = !{!4} ; A list containing only scope !4
4137 ; These two instructions don't alias:
4138 %0 = load float, float* %c, align 4, !alias.scope !5
4139 store float %0, float* %arrayidx.i, align 4, !noalias !5
4141 ; These two instructions also don't alias (for domain !1, the set of scopes
4142 ; in the !alias.scope equals that in the !noalias list):
4143 %2 = load float, float* %c, align 4, !alias.scope !5
4144 store float %2, float* %arrayidx.i2, align 4, !noalias !6
4146 ; These two instructions may alias (for domain !0, the set of scopes in
4147 ; the !noalias list is not a superset of, or equal to, the scopes in the
4148 ; !alias.scope list):
4149 %2 = load float, float* %c, align 4, !alias.scope !6
4150 store float %0, float* %arrayidx.i, align 4, !noalias !7
4152 '``fpmath``' Metadata
4153 ^^^^^^^^^^^^^^^^^^^^^
4155 ``fpmath`` metadata may be attached to any instruction of floating point
4156 type. It can be used to express the maximum acceptable error in the
4157 result of that instruction, in ULPs, thus potentially allowing the
4158 compiler to use a more efficient but less accurate method of computing
4159 it. ULP is defined as follows:
4161 If ``x`` is a real number that lies between two finite consecutive
4162 floating-point numbers ``a`` and ``b``, without being equal to one
4163 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4164 distance between the two non-equal finite floating-point numbers
4165 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4167 The metadata node shall consist of a single positive floating point
4168 number representing the maximum relative error, for example:
4170 .. code-block:: llvm
4172 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
4176 '``range``' Metadata
4177 ^^^^^^^^^^^^^^^^^^^^
4179 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4180 integer types. It expresses the possible ranges the loaded value or the value
4181 returned by the called function at this call site is in. The ranges are
4182 represented with a flattened list of integers. The loaded value or the value
4183 returned is known to be in the union of the ranges defined by each consecutive
4184 pair. Each pair has the following properties:
4186 - The type must match the type loaded by the instruction.
4187 - The pair ``a,b`` represents the range ``[a,b)``.
4188 - Both ``a`` and ``b`` are constants.
4189 - The range is allowed to wrap.
4190 - The range should not represent the full or empty set. That is,
4193 In addition, the pairs must be in signed order of the lower bound and
4194 they must be non-contiguous.
4198 .. code-block:: llvm
4200 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4201 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
4202 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4203 %d = invoke i8 @bar() to label %cont
4204 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
4206 !0 = !{ i8 0, i8 2 }
4207 !1 = !{ i8 255, i8 2 }
4208 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4209 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
4211 '``unpredictable``' Metadata
4212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4214 ``unpredictable`` metadata may be attached to any branch or switch
4215 instruction. It can be used to express the unpredictability of control
4216 flow. Similar to the llvm.expect intrinsic, it may be used to alter
4217 optimizations related to compare and branch instructions. The metadata
4218 is treated as a boolean value; if it exists, it signals that the branch
4219 or switch that it is attached to is completely unpredictable.
4224 It is sometimes useful to attach information to loop constructs. Currently,
4225 loop metadata is implemented as metadata attached to the branch instruction
4226 in the loop latch block. This type of metadata refer to a metadata node that is
4227 guaranteed to be separate for each loop. The loop identifier metadata is
4228 specified with the name ``llvm.loop``.
4230 The loop identifier metadata is implemented using a metadata that refers to
4231 itself to avoid merging it with any other identifier metadata, e.g.,
4232 during module linkage or function inlining. That is, each loop should refer
4233 to their own identification metadata even if they reside in separate functions.
4234 The following example contains loop identifier metadata for two separate loop
4237 .. code-block:: llvm
4242 The loop identifier metadata can be used to specify additional
4243 per-loop metadata. Any operands after the first operand can be treated
4244 as user-defined metadata. For example the ``llvm.loop.unroll.count``
4245 suggests an unroll factor to the loop unroller:
4247 .. code-block:: llvm
4249 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
4252 !1 = !{!"llvm.loop.unroll.count", i32 4}
4254 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
4255 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4257 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
4258 used to control per-loop vectorization and interleaving parameters such as
4259 vectorization width and interleave count. These metadata should be used in
4260 conjunction with ``llvm.loop`` loop identification metadata. The
4261 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
4262 optimization hints and the optimizer will only interleave and vectorize loops if
4263 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
4264 which contains information about loop-carried memory dependencies can be helpful
4265 in determining the safety of these transformations.
4267 '``llvm.loop.interleave.count``' Metadata
4268 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4270 This metadata suggests an interleave count to the loop interleaver.
4271 The first operand is the string ``llvm.loop.interleave.count`` and the
4272 second operand is an integer specifying the interleave count. For
4275 .. code-block:: llvm
4277 !0 = !{!"llvm.loop.interleave.count", i32 4}
4279 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
4280 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
4281 then the interleave count will be determined automatically.
4283 '``llvm.loop.vectorize.enable``' Metadata
4284 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4286 This metadata selectively enables or disables vectorization for the loop. The
4287 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
4288 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
4289 0 disables vectorization:
4291 .. code-block:: llvm
4293 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
4294 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
4296 '``llvm.loop.vectorize.width``' Metadata
4297 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4299 This metadata sets the target width of the vectorizer. The first
4300 operand is the string ``llvm.loop.vectorize.width`` and the second
4301 operand is an integer specifying the width. For example:
4303 .. code-block:: llvm
4305 !0 = !{!"llvm.loop.vectorize.width", i32 4}
4307 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
4308 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
4309 0 or if the loop does not have this metadata the width will be
4310 determined automatically.
4312 '``llvm.loop.unroll``'
4313 ^^^^^^^^^^^^^^^^^^^^^^
4315 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
4316 optimization hints such as the unroll factor. ``llvm.loop.unroll``
4317 metadata should be used in conjunction with ``llvm.loop`` loop
4318 identification metadata. The ``llvm.loop.unroll`` metadata are only
4319 optimization hints and the unrolling will only be performed if the
4320 optimizer believes it is safe to do so.
4322 '``llvm.loop.unroll.count``' Metadata
4323 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4325 This metadata suggests an unroll factor to the loop unroller. The
4326 first operand is the string ``llvm.loop.unroll.count`` and the second
4327 operand is a positive integer specifying the unroll factor. For
4330 .. code-block:: llvm
4332 !0 = !{!"llvm.loop.unroll.count", i32 4}
4334 If the trip count of the loop is less than the unroll count the loop
4335 will be partially unrolled.
4337 '``llvm.loop.unroll.disable``' Metadata
4338 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4340 This metadata disables loop unrolling. The metadata has a single operand
4341 which is the string ``llvm.loop.unroll.disable``. For example:
4343 .. code-block:: llvm
4345 !0 = !{!"llvm.loop.unroll.disable"}
4347 '``llvm.loop.unroll.runtime.disable``' Metadata
4348 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4350 This metadata disables runtime loop unrolling. The metadata has a single
4351 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
4353 .. code-block:: llvm
4355 !0 = !{!"llvm.loop.unroll.runtime.disable"}
4357 '``llvm.loop.unroll.enable``' Metadata
4358 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4360 This metadata suggests that the loop should be fully unrolled if the trip count
4361 is known at compile time and partially unrolled if the trip count is not known
4362 at compile time. The metadata has a single operand which is the string
4363 ``llvm.loop.unroll.enable``. For example:
4365 .. code-block:: llvm
4367 !0 = !{!"llvm.loop.unroll.enable"}
4369 '``llvm.loop.unroll.full``' Metadata
4370 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4372 This metadata suggests that the loop should be unrolled fully. The
4373 metadata has a single operand which is the string ``llvm.loop.unroll.full``.
4376 .. code-block:: llvm
4378 !0 = !{!"llvm.loop.unroll.full"}
4383 Metadata types used to annotate memory accesses with information helpful
4384 for optimizations are prefixed with ``llvm.mem``.
4386 '``llvm.mem.parallel_loop_access``' Metadata
4387 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4389 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
4390 or metadata containing a list of loop identifiers for nested loops.
4391 The metadata is attached to memory accessing instructions and denotes that
4392 no loop carried memory dependence exist between it and other instructions denoted
4393 with the same loop identifier.
4395 Precisely, given two instructions ``m1`` and ``m2`` that both have the
4396 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
4397 set of loops associated with that metadata, respectively, then there is no loop
4398 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
4401 As a special case, if all memory accessing instructions in a loop have
4402 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
4403 loop has no loop carried memory dependences and is considered to be a parallel
4406 Note that if not all memory access instructions have such metadata referring to
4407 the loop, then the loop is considered not being trivially parallel. Additional
4408 memory dependence analysis is required to make that determination. As a fail
4409 safe mechanism, this causes loops that were originally parallel to be considered
4410 sequential (if optimization passes that are unaware of the parallel semantics
4411 insert new memory instructions into the loop body).
4413 Example of a loop that is considered parallel due to its correct use of
4414 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
4415 metadata types that refer to the same loop identifier metadata.
4417 .. code-block:: llvm
4421 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
4423 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4425 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
4431 It is also possible to have nested parallel loops. In that case the
4432 memory accesses refer to a list of loop identifier metadata nodes instead of
4433 the loop identifier metadata node directly:
4435 .. code-block:: llvm
4439 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
4441 br label %inner.for.body
4445 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4447 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
4449 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
4453 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
4455 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
4457 outer.for.end: ; preds = %for.body
4459 !0 = !{!1, !2} ; a list of loop identifiers
4460 !1 = !{!1} ; an identifier for the inner loop
4461 !2 = !{!2} ; an identifier for the outer loop
4466 The ``llvm.bitsets`` global metadata is used to implement
4467 :doc:`bitsets <BitSets>`.
4469 '``invariant.group``' Metadata
4470 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4472 The ``invariant.group`` metadata may be attached to ``load``/``store`` instructions.
4473 The existence of the ``invariant.group`` metadata on the instruction tells
4474 the optimizer that every ``load`` and ``store`` to the same pointer operand
4475 within the same invariant group can be assumed to load or store the same
4476 value (but see the ``llvm.invariant.group.barrier`` intrinsic which affects
4477 when two pointers are considered the same).
4481 .. code-block:: llvm
4483 @unknownPtr = external global i8
4486 store i8 42, i8* %ptr, !invariant.group !0
4487 call void @foo(i8* %ptr)
4489 %a = load i8, i8* %ptr, !invariant.group !0 ; Can assume that value under %ptr didn't change
4490 call void @foo(i8* %ptr)
4491 %b = load i8, i8* %ptr, !invariant.group !1 ; Can't assume anything, because group changed
4493 %newPtr = call i8* @getPointer(i8* %ptr)
4494 %c = load i8, i8* %newPtr, !invariant.group !0 ; Can't assume anything, because we only have information about %ptr
4496 %unknownValue = load i8, i8* @unknownPtr
4497 store i8 %unknownValue, i8* %ptr, !invariant.group !0 ; Can assume that %unknownValue == 42
4499 call void @foo(i8* %ptr)
4500 %newPtr2 = call i8* @llvm.invariant.group.barrier(i8* %ptr)
4501 %d = load i8, i8* %newPtr2, !invariant.group !0 ; Can't step through invariant.group.barrier to get value of %ptr
4504 declare void @foo(i8*)
4505 declare i8* @getPointer(i8*)
4506 declare i8* @llvm.invariant.group.barrier(i8*)
4508 !0 = !{!"magic ptr"}
4509 !1 = !{!"other ptr"}
4513 Module Flags Metadata
4514 =====================
4516 Information about the module as a whole is difficult to convey to LLVM's
4517 subsystems. The LLVM IR isn't sufficient to transmit this information.
4518 The ``llvm.module.flags`` named metadata exists in order to facilitate
4519 this. These flags are in the form of key / value pairs --- much like a
4520 dictionary --- making it easy for any subsystem who cares about a flag to
4523 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
4524 Each triplet has the following form:
4526 - The first element is a *behavior* flag, which specifies the behavior
4527 when two (or more) modules are merged together, and it encounters two
4528 (or more) metadata with the same ID. The supported behaviors are
4530 - The second element is a metadata string that is a unique ID for the
4531 metadata. Each module may only have one flag entry for each unique ID (not
4532 including entries with the **Require** behavior).
4533 - The third element is the value of the flag.
4535 When two (or more) modules are merged together, the resulting
4536 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
4537 each unique metadata ID string, there will be exactly one entry in the merged
4538 modules ``llvm.module.flags`` metadata table, and the value for that entry will
4539 be determined by the merge behavior flag, as described below. The only exception
4540 is that entries with the *Require* behavior are always preserved.
4542 The following behaviors are supported:
4553 Emits an error if two values disagree, otherwise the resulting value
4554 is that of the operands.
4558 Emits a warning if two values disagree. The result value will be the
4559 operand for the flag from the first module being linked.
4563 Adds a requirement that another module flag be present and have a
4564 specified value after linking is performed. The value must be a
4565 metadata pair, where the first element of the pair is the ID of the
4566 module flag to be restricted, and the second element of the pair is
4567 the value the module flag should be restricted to. This behavior can
4568 be used to restrict the allowable results (via triggering of an
4569 error) of linking IDs with the **Override** behavior.
4573 Uses the specified value, regardless of the behavior or value of the
4574 other module. If both modules specify **Override**, but the values
4575 differ, an error will be emitted.
4579 Appends the two values, which are required to be metadata nodes.
4583 Appends the two values, which are required to be metadata
4584 nodes. However, duplicate entries in the second list are dropped
4585 during the append operation.
4587 It is an error for a particular unique flag ID to have multiple behaviors,
4588 except in the case of **Require** (which adds restrictions on another metadata
4589 value) or **Override**.
4591 An example of module flags:
4593 .. code-block:: llvm
4595 !0 = !{ i32 1, !"foo", i32 1 }
4596 !1 = !{ i32 4, !"bar", i32 37 }
4597 !2 = !{ i32 2, !"qux", i32 42 }
4598 !3 = !{ i32 3, !"qux",
4603 !llvm.module.flags = !{ !0, !1, !2, !3 }
4605 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
4606 if two or more ``!"foo"`` flags are seen is to emit an error if their
4607 values are not equal.
4609 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
4610 behavior if two or more ``!"bar"`` flags are seen is to use the value
4613 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
4614 behavior if two or more ``!"qux"`` flags are seen is to emit a
4615 warning if their values are not equal.
4617 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
4623 The behavior is to emit an error if the ``llvm.module.flags`` does not
4624 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
4627 Objective-C Garbage Collection Module Flags Metadata
4628 ----------------------------------------------------
4630 On the Mach-O platform, Objective-C stores metadata about garbage
4631 collection in a special section called "image info". The metadata
4632 consists of a version number and a bitmask specifying what types of
4633 garbage collection are supported (if any) by the file. If two or more
4634 modules are linked together their garbage collection metadata needs to
4635 be merged rather than appended together.
4637 The Objective-C garbage collection module flags metadata consists of the
4638 following key-value pairs:
4647 * - ``Objective-C Version``
4648 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
4650 * - ``Objective-C Image Info Version``
4651 - **[Required]** --- The version of the image info section. Currently
4654 * - ``Objective-C Image Info Section``
4655 - **[Required]** --- The section to place the metadata. Valid values are
4656 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
4657 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
4658 Objective-C ABI version 2.
4660 * - ``Objective-C Garbage Collection``
4661 - **[Required]** --- Specifies whether garbage collection is supported or
4662 not. Valid values are 0, for no garbage collection, and 2, for garbage
4663 collection supported.
4665 * - ``Objective-C GC Only``
4666 - **[Optional]** --- Specifies that only garbage collection is supported.
4667 If present, its value must be 6. This flag requires that the
4668 ``Objective-C Garbage Collection`` flag have the value 2.
4670 Some important flag interactions:
4672 - If a module with ``Objective-C Garbage Collection`` set to 0 is
4673 merged with a module with ``Objective-C Garbage Collection`` set to
4674 2, then the resulting module has the
4675 ``Objective-C Garbage Collection`` flag set to 0.
4676 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
4677 merged with a module with ``Objective-C GC Only`` set to 6.
4679 Automatic Linker Flags Module Flags Metadata
4680 --------------------------------------------
4682 Some targets support embedding flags to the linker inside individual object
4683 files. Typically this is used in conjunction with language extensions which
4684 allow source files to explicitly declare the libraries they depend on, and have
4685 these automatically be transmitted to the linker via object files.
4687 These flags are encoded in the IR using metadata in the module flags section,
4688 using the ``Linker Options`` key. The merge behavior for this flag is required
4689 to be ``AppendUnique``, and the value for the key is expected to be a metadata
4690 node which should be a list of other metadata nodes, each of which should be a
4691 list of metadata strings defining linker options.
4693 For example, the following metadata section specifies two separate sets of
4694 linker options, presumably to link against ``libz`` and the ``Cocoa``
4697 !0 = !{ i32 6, !"Linker Options",
4700 !{ !"-framework", !"Cocoa" } } }
4701 !llvm.module.flags = !{ !0 }
4703 The metadata encoding as lists of lists of options, as opposed to a collapsed
4704 list of options, is chosen so that the IR encoding can use multiple option
4705 strings to specify e.g., a single library, while still having that specifier be
4706 preserved as an atomic element that can be recognized by a target specific
4707 assembly writer or object file emitter.
4709 Each individual option is required to be either a valid option for the target's
4710 linker, or an option that is reserved by the target specific assembly writer or
4711 object file emitter. No other aspect of these options is defined by the IR.
4713 C type width Module Flags Metadata
4714 ----------------------------------
4716 The ARM backend emits a section into each generated object file describing the
4717 options that it was compiled with (in a compiler-independent way) to prevent
4718 linking incompatible objects, and to allow automatic library selection. Some
4719 of these options are not visible at the IR level, namely wchar_t width and enum
4722 To pass this information to the backend, these options are encoded in module
4723 flags metadata, using the following key-value pairs:
4733 - * 0 --- sizeof(wchar_t) == 4
4734 * 1 --- sizeof(wchar_t) == 2
4737 - * 0 --- Enums are at least as large as an ``int``.
4738 * 1 --- Enums are stored in the smallest integer type which can
4739 represent all of its values.
4741 For example, the following metadata section specifies that the module was
4742 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
4743 enum is the smallest type which can represent all of its values::
4745 !llvm.module.flags = !{!0, !1}
4746 !0 = !{i32 1, !"short_wchar", i32 1}
4747 !1 = !{i32 1, !"short_enum", i32 0}
4749 .. _intrinsicglobalvariables:
4751 Intrinsic Global Variables
4752 ==========================
4754 LLVM has a number of "magic" global variables that contain data that
4755 affect code generation or other IR semantics. These are documented here.
4756 All globals of this sort should have a section specified as
4757 "``llvm.metadata``". This section and all globals that start with
4758 "``llvm.``" are reserved for use by LLVM.
4762 The '``llvm.used``' Global Variable
4763 -----------------------------------
4765 The ``@llvm.used`` global is an array which has
4766 :ref:`appending linkage <linkage_appending>`. This array contains a list of
4767 pointers to named global variables, functions and aliases which may optionally
4768 have a pointer cast formed of bitcast or getelementptr. For example, a legal
4771 .. code-block:: llvm
4776 @llvm.used = appending global [2 x i8*] [
4778 i8* bitcast (i32* @Y to i8*)
4779 ], section "llvm.metadata"
4781 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4782 and linker are required to treat the symbol as if there is a reference to the
4783 symbol that it cannot see (which is why they have to be named). For example, if
4784 a variable has internal linkage and no references other than that from the
4785 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4786 references from inline asms and other things the compiler cannot "see", and
4787 corresponds to "``attribute((used))``" in GNU C.
4789 On some targets, the code generator must emit a directive to the
4790 assembler or object file to prevent the assembler and linker from
4791 molesting the symbol.
4793 .. _gv_llvmcompilerused:
4795 The '``llvm.compiler.used``' Global Variable
4796 --------------------------------------------
4798 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4799 directive, except that it only prevents the compiler from touching the
4800 symbol. On targets that support it, this allows an intelligent linker to
4801 optimize references to the symbol without being impeded as it would be
4804 This is a rare construct that should only be used in rare circumstances,
4805 and should not be exposed to source languages.
4807 .. _gv_llvmglobalctors:
4809 The '``llvm.global_ctors``' Global Variable
4810 -------------------------------------------
4812 .. code-block:: llvm
4814 %0 = type { i32, void ()*, i8* }
4815 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4817 The ``@llvm.global_ctors`` array contains a list of constructor
4818 functions, priorities, and an optional associated global or function.
4819 The functions referenced by this array will be called in ascending order
4820 of priority (i.e. lowest first) when the module is loaded. The order of
4821 functions with the same priority is not defined.
4823 If the third field is present, non-null, and points to a global variable
4824 or function, the initializer function will only run if the associated
4825 data from the current module is not discarded.
4827 .. _llvmglobaldtors:
4829 The '``llvm.global_dtors``' Global Variable
4830 -------------------------------------------
4832 .. code-block:: llvm
4834 %0 = type { i32, void ()*, i8* }
4835 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4837 The ``@llvm.global_dtors`` array contains a list of destructor
4838 functions, priorities, and an optional associated global or function.
4839 The functions referenced by this array will be called in descending
4840 order of priority (i.e. highest first) when the module is unloaded. The
4841 order of functions with the same priority is not defined.
4843 If the third field is present, non-null, and points to a global variable
4844 or function, the destructor function will only run if the associated
4845 data from the current module is not discarded.
4847 Instruction Reference
4848 =====================
4850 The LLVM instruction set consists of several different classifications
4851 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4852 instructions <binaryops>`, :ref:`bitwise binary
4853 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4854 :ref:`other instructions <otherops>`.
4858 Terminator Instructions
4859 -----------------------
4861 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4862 program ends with a "Terminator" instruction, which indicates which
4863 block should be executed after the current block is finished. These
4864 terminator instructions typically yield a '``void``' value: they produce
4865 control flow, not values (the one exception being the
4866 ':ref:`invoke <i_invoke>`' instruction).
4868 The terminator instructions are: ':ref:`ret <i_ret>`',
4869 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4870 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4871 ':ref:`resume <i_resume>`', ':ref:`catchpad <i_catchpad>`',
4872 ':ref:`catchendpad <i_catchendpad>`',
4873 ':ref:`catchret <i_catchret>`',
4874 ':ref:`cleanupendpad <i_cleanupendpad>`',
4875 ':ref:`cleanupret <i_cleanupret>`',
4876 ':ref:`terminatepad <i_terminatepad>`',
4877 and ':ref:`unreachable <i_unreachable>`'.
4881 '``ret``' Instruction
4882 ^^^^^^^^^^^^^^^^^^^^^
4889 ret <type> <value> ; Return a value from a non-void function
4890 ret void ; Return from void function
4895 The '``ret``' instruction is used to return control flow (and optionally
4896 a value) from a function back to the caller.
4898 There are two forms of the '``ret``' instruction: one that returns a
4899 value and then causes control flow, and one that just causes control
4905 The '``ret``' instruction optionally accepts a single argument, the
4906 return value. The type of the return value must be a ':ref:`first
4907 class <t_firstclass>`' type.
4909 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4910 return type and contains a '``ret``' instruction with no return value or
4911 a return value with a type that does not match its type, or if it has a
4912 void return type and contains a '``ret``' instruction with a return
4918 When the '``ret``' instruction is executed, control flow returns back to
4919 the calling function's context. If the caller is a
4920 ":ref:`call <i_call>`" instruction, execution continues at the
4921 instruction after the call. If the caller was an
4922 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4923 beginning of the "normal" destination block. If the instruction returns
4924 a value, that value shall set the call or invoke instruction's return
4930 .. code-block:: llvm
4932 ret i32 5 ; Return an integer value of 5
4933 ret void ; Return from a void function
4934 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4938 '``br``' Instruction
4939 ^^^^^^^^^^^^^^^^^^^^
4946 br i1 <cond>, label <iftrue>, label <iffalse>
4947 br label <dest> ; Unconditional branch
4952 The '``br``' instruction is used to cause control flow to transfer to a
4953 different basic block in the current function. There are two forms of
4954 this instruction, corresponding to a conditional branch and an
4955 unconditional branch.
4960 The conditional branch form of the '``br``' instruction takes a single
4961 '``i1``' value and two '``label``' values. The unconditional form of the
4962 '``br``' instruction takes a single '``label``' value as a target.
4967 Upon execution of a conditional '``br``' instruction, the '``i1``'
4968 argument is evaluated. If the value is ``true``, control flows to the
4969 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4970 to the '``iffalse``' ``label`` argument.
4975 .. code-block:: llvm
4978 %cond = icmp eq i32 %a, %b
4979 br i1 %cond, label %IfEqual, label %IfUnequal
4987 '``switch``' Instruction
4988 ^^^^^^^^^^^^^^^^^^^^^^^^
4995 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
5000 The '``switch``' instruction is used to transfer control flow to one of
5001 several different places. It is a generalization of the '``br``'
5002 instruction, allowing a branch to occur to one of many possible
5008 The '``switch``' instruction uses three parameters: an integer
5009 comparison value '``value``', a default '``label``' destination, and an
5010 array of pairs of comparison value constants and '``label``'s. The table
5011 is not allowed to contain duplicate constant entries.
5016 The ``switch`` instruction specifies a table of values and destinations.
5017 When the '``switch``' instruction is executed, this table is searched
5018 for the given value. If the value is found, control flow is transferred
5019 to the corresponding destination; otherwise, control flow is transferred
5020 to the default destination.
5025 Depending on properties of the target machine and the particular
5026 ``switch`` instruction, this instruction may be code generated in
5027 different ways. For example, it could be generated as a series of
5028 chained conditional branches or with a lookup table.
5033 .. code-block:: llvm
5035 ; Emulate a conditional br instruction
5036 %Val = zext i1 %value to i32
5037 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
5039 ; Emulate an unconditional br instruction
5040 switch i32 0, label %dest [ ]
5042 ; Implement a jump table:
5043 switch i32 %val, label %otherwise [ i32 0, label %onzero
5045 i32 2, label %ontwo ]
5049 '``indirectbr``' Instruction
5050 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5057 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
5062 The '``indirectbr``' instruction implements an indirect branch to a
5063 label within the current function, whose address is specified by
5064 "``address``". Address must be derived from a
5065 :ref:`blockaddress <blockaddress>` constant.
5070 The '``address``' argument is the address of the label to jump to. The
5071 rest of the arguments indicate the full set of possible destinations
5072 that the address may point to. Blocks are allowed to occur multiple
5073 times in the destination list, though this isn't particularly useful.
5075 This destination list is required so that dataflow analysis has an
5076 accurate understanding of the CFG.
5081 Control transfers to the block specified in the address argument. All
5082 possible destination blocks must be listed in the label list, otherwise
5083 this instruction has undefined behavior. This implies that jumps to
5084 labels defined in other functions have undefined behavior as well.
5089 This is typically implemented with a jump through a register.
5094 .. code-block:: llvm
5096 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
5100 '``invoke``' Instruction
5101 ^^^^^^^^^^^^^^^^^^^^^^^^
5108 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
5109 [operand bundles] to label <normal label> unwind label <exception label>
5114 The '``invoke``' instruction causes control to transfer to a specified
5115 function, with the possibility of control flow transfer to either the
5116 '``normal``' label or the '``exception``' label. If the callee function
5117 returns with the "``ret``" instruction, control flow will return to the
5118 "normal" label. If the callee (or any indirect callees) returns via the
5119 ":ref:`resume <i_resume>`" instruction or other exception handling
5120 mechanism, control is interrupted and continued at the dynamically
5121 nearest "exception" label.
5123 The '``exception``' label is a `landing
5124 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
5125 '``exception``' label is required to have the
5126 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
5127 information about the behavior of the program after unwinding happens,
5128 as its first non-PHI instruction. The restrictions on the
5129 "``landingpad``" instruction's tightly couples it to the "``invoke``"
5130 instruction, so that the important information contained within the
5131 "``landingpad``" instruction can't be lost through normal code motion.
5136 This instruction requires several arguments:
5138 #. The optional "cconv" marker indicates which :ref:`calling
5139 convention <callingconv>` the call should use. If none is
5140 specified, the call defaults to using C calling conventions.
5141 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5142 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5144 #. '``ptr to function ty``': shall be the signature of the pointer to
5145 function value being invoked. In most cases, this is a direct
5146 function invocation, but indirect ``invoke``'s are just as possible,
5147 branching off an arbitrary pointer to function value.
5148 #. '``function ptr val``': An LLVM value containing a pointer to a
5149 function to be invoked.
5150 #. '``function args``': argument list whose types match the function
5151 signature argument types and parameter attributes. All arguments must
5152 be of :ref:`first class <t_firstclass>` type. If the function signature
5153 indicates the function accepts a variable number of arguments, the
5154 extra arguments can be specified.
5155 #. '``normal label``': the label reached when the called function
5156 executes a '``ret``' instruction.
5157 #. '``exception label``': the label reached when a callee returns via
5158 the :ref:`resume <i_resume>` instruction or other exception handling
5160 #. The optional :ref:`function attributes <fnattrs>` list. Only
5161 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5162 attributes are valid here.
5163 #. The optional :ref:`operand bundles <opbundles>` list.
5168 This instruction is designed to operate as a standard '``call``'
5169 instruction in most regards. The primary difference is that it
5170 establishes an association with a label, which is used by the runtime
5171 library to unwind the stack.
5173 This instruction is used in languages with destructors to ensure that
5174 proper cleanup is performed in the case of either a ``longjmp`` or a
5175 thrown exception. Additionally, this is important for implementation of
5176 '``catch``' clauses in high-level languages that support them.
5178 For the purposes of the SSA form, the definition of the value returned
5179 by the '``invoke``' instruction is deemed to occur on the edge from the
5180 current block to the "normal" label. If the callee unwinds then no
5181 return value is available.
5186 .. code-block:: llvm
5188 %retval = invoke i32 @Test(i32 15) to label %Continue
5189 unwind label %TestCleanup ; i32:retval set
5190 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
5191 unwind label %TestCleanup ; i32:retval set
5195 '``resume``' Instruction
5196 ^^^^^^^^^^^^^^^^^^^^^^^^
5203 resume <type> <value>
5208 The '``resume``' instruction is a terminator instruction that has no
5214 The '``resume``' instruction requires one argument, which must have the
5215 same type as the result of any '``landingpad``' instruction in the same
5221 The '``resume``' instruction resumes propagation of an existing
5222 (in-flight) exception whose unwinding was interrupted with a
5223 :ref:`landingpad <i_landingpad>` instruction.
5228 .. code-block:: llvm
5230 resume { i8*, i32 } %exn
5234 '``catchpad``' Instruction
5235 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5242 <resultval> = catchpad [<args>*]
5243 to label <normal label> unwind label <exception label>
5248 The '``catchpad``' instruction is used by `LLVM's exception handling
5249 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5250 is a catch block --- one where a personality routine attempts to transfer
5251 control to catch an exception.
5252 The ``args`` correspond to whatever information the personality
5253 routine requires to know if this is an appropriate place to catch the
5254 exception. Control is transfered to the ``exception`` label if the
5255 ``catchpad`` is not an appropriate handler for the in-flight exception.
5256 The ``normal`` label should contain the code found in the ``catch``
5257 portion of a ``try``/``catch`` sequence. The ``resultval`` has the type
5258 :ref:`token <t_token>` and is used to match the ``catchpad`` to
5259 corresponding :ref:`catchrets <i_catchret>`.
5264 The instruction takes a list of arbitrary values which are interpreted
5265 by the :ref:`personality function <personalityfn>`.
5267 The ``catchpad`` must be provided a ``normal`` label to transfer control
5268 to if the ``catchpad`` matches the exception and an ``exception``
5269 label to transfer control to if it doesn't.
5274 When the call stack is being unwound due to an exception being thrown,
5275 the exception is compared against the ``args``. If it doesn't match,
5276 then control is transfered to the ``exception`` basic block.
5277 As with calling conventions, how the personality function results are
5278 represented in LLVM IR is target specific.
5280 The ``catchpad`` instruction has several restrictions:
5282 - A catch block is a basic block which is the unwind destination of
5283 an exceptional instruction.
5284 - A catch block must have a '``catchpad``' instruction as its
5285 first non-PHI instruction.
5286 - A catch block's ``exception`` edge must refer to a catch block or a
5288 - There can be only one '``catchpad``' instruction within the
5290 - A basic block that is not a catch block may not include a
5291 '``catchpad``' instruction.
5292 - A catch block which has another catch block as a predecessor may not have
5293 any other predecessors.
5294 - It is undefined behavior for control to transfer from a ``catchpad`` to a
5295 ``ret`` without first executing a ``catchret`` that consumes the
5296 ``catchpad`` or unwinding through its ``catchendpad``.
5297 - It is undefined behavior for control to transfer from a ``catchpad`` to
5298 itself without first executing a ``catchret`` that consumes the
5299 ``catchpad`` or unwinding through its ``catchendpad``.
5304 .. code-block:: llvm
5306 ;; A catch block which can catch an integer.
5307 %tok = catchpad [i8** @_ZTIi]
5308 to label %int.handler unwind label %terminate
5312 '``catchendpad``' Instruction
5313 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5320 catchendpad unwind label <nextaction>
5321 catchendpad unwind to caller
5326 The '``catchendpad``' instruction is used by `LLVM's exception handling
5327 system <ExceptionHandling.html#overview>`_ to communicate to the
5328 :ref:`personality function <personalityfn>` which invokes are associated
5329 with a chain of :ref:`catchpad <i_catchpad>` instructions; propagating an
5330 exception out of a catch handler is represented by unwinding through its
5331 ``catchendpad``. Unwinding to the outer scope when a chain of catch handlers
5332 do not handle an exception is also represented by unwinding through their
5335 The ``nextaction`` label indicates where control should transfer to if
5336 none of the ``catchpad`` instructions are suitable for catching the
5337 in-flight exception.
5339 If a ``nextaction`` label is not present, the instruction unwinds out of
5340 its parent function. The
5341 :ref:`personality function <personalityfn>` will continue processing
5342 exception handling actions in the caller.
5347 The instruction optionally takes a label, ``nextaction``, indicating
5348 where control should transfer to if none of the preceding
5349 ``catchpad`` instructions are suitable for the in-flight exception.
5354 When the call stack is being unwound due to an exception being thrown
5355 and none of the constituent ``catchpad`` instructions match, then
5356 control is transfered to ``nextaction`` if it is present. If it is not
5357 present, control is transfered to the caller.
5359 The ``catchendpad`` instruction has several restrictions:
5361 - A catch-end block is a basic block which is the unwind destination of
5362 an exceptional instruction.
5363 - A catch-end block must have a '``catchendpad``' instruction as its
5364 first non-PHI instruction.
5365 - There can be only one '``catchendpad``' instruction within the
5367 - A basic block that is not a catch-end block may not include a
5368 '``catchendpad``' instruction.
5369 - Exactly one catch block may unwind to a ``catchendpad``.
5370 - It is undefined behavior to execute a ``catchendpad`` if none of the
5371 '``catchpad``'s chained to it have been executed.
5372 - It is undefined behavior to execute a ``catchendpad`` twice without an
5373 intervening execution of one or more of the '``catchpad``'s chained to it.
5374 - It is undefined behavior to execute a ``catchendpad`` if, after the most
5375 recent execution of the normal successor edge of any ``catchpad`` chained
5376 to it, some ``catchret`` consuming that ``catchpad`` has already been
5378 - It is undefined behavior to execute a ``catchendpad`` if, after the most
5379 recent execution of the normal successor edge of any ``catchpad`` chained
5380 to it, any other ``catchpad`` or ``cleanuppad`` has been executed but has
5381 not had a corresponding
5382 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5387 .. code-block:: llvm
5389 catchendpad unwind label %terminate
5390 catchendpad unwind to caller
5394 '``catchret``' Instruction
5395 ^^^^^^^^^^^^^^^^^^^^^^^^^^
5402 catchret <value> to label <normal>
5407 The '``catchret``' instruction is a terminator instruction that has a
5414 The first argument to a '``catchret``' indicates which ``catchpad`` it
5415 exits. It must be a :ref:`catchpad <i_catchpad>`.
5416 The second argument to a '``catchret``' specifies where control will
5422 The '``catchret``' instruction ends the existing (in-flight) exception
5423 whose unwinding was interrupted with a
5424 :ref:`catchpad <i_catchpad>` instruction.
5425 The :ref:`personality function <personalityfn>` gets a chance to execute
5426 arbitrary code to, for example, run a C++ destructor.
5427 Control then transfers to ``normal``.
5428 It may be passed an optional, personality specific, value.
5430 It is undefined behavior to execute a ``catchret`` whose ``catchpad`` has
5433 It is undefined behavior to execute a ``catchret`` if, after the most recent
5434 execution of its ``catchpad``, some ``catchret`` or ``catchendpad`` linked
5435 to the same ``catchpad`` has already been executed.
5437 It is undefined behavior to execute a ``catchret`` if, after the most recent
5438 execution of its ``catchpad``, any other ``catchpad`` or ``cleanuppad`` has
5439 been executed but has not had a corresponding
5440 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5445 .. code-block:: llvm
5447 catchret %catch label %continue
5449 .. _i_cleanupendpad:
5451 '``cleanupendpad``' Instruction
5452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5459 cleanupendpad <value> unwind label <nextaction>
5460 cleanupendpad <value> unwind to caller
5465 The '``cleanupendpad``' instruction is used by `LLVM's exception handling
5466 system <ExceptionHandling.html#overview>`_ to communicate to the
5467 :ref:`personality function <personalityfn>` which invokes are associated
5468 with a :ref:`cleanuppad <i_cleanuppad>` instructions; propagating an exception
5469 out of a cleanup is represented by unwinding through its ``cleanupendpad``.
5471 The ``nextaction`` label indicates where control should unwind to next, in the
5472 event that a cleanup is exited by means of an(other) exception being raised.
5474 If a ``nextaction`` label is not present, the instruction unwinds out of
5475 its parent function. The
5476 :ref:`personality function <personalityfn>` will continue processing
5477 exception handling actions in the caller.
5482 The '``cleanupendpad``' instruction requires one argument, which indicates
5483 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5484 It also has an optional successor, ``nextaction``, indicating where control
5490 When and exception propagates to a ``cleanupendpad``, control is transfered to
5491 ``nextaction`` if it is present. If it is not present, control is transfered to
5494 The ``cleanupendpad`` instruction has several restrictions:
5496 - A cleanup-end block is a basic block which is the unwind destination of
5497 an exceptional instruction.
5498 - A cleanup-end block must have a '``cleanupendpad``' instruction as its
5499 first non-PHI instruction.
5500 - There can be only one '``cleanupendpad``' instruction within the
5502 - A basic block that is not a cleanup-end block may not include a
5503 '``cleanupendpad``' instruction.
5504 - It is undefined behavior to execute a ``cleanupendpad`` whose ``cleanuppad``
5505 has not been executed.
5506 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5507 recent execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5508 consuming the same ``cleanuppad`` has already been executed.
5509 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5510 recent execution of its ``cleanuppad``, any other ``cleanuppad`` or
5511 ``catchpad`` has been executed but has not had a corresponding
5512 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5517 .. code-block:: llvm
5519 cleanupendpad %cleanup unwind label %terminate
5520 cleanupendpad %cleanup unwind to caller
5524 '``cleanupret``' Instruction
5525 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5532 cleanupret <value> unwind label <continue>
5533 cleanupret <value> unwind to caller
5538 The '``cleanupret``' instruction is a terminator instruction that has
5539 an optional successor.
5545 The '``cleanupret``' instruction requires one argument, which indicates
5546 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5547 It also has an optional successor, ``continue``.
5552 The '``cleanupret``' instruction indicates to the
5553 :ref:`personality function <personalityfn>` that one
5554 :ref:`cleanuppad <i_cleanuppad>` it transferred control to has ended.
5555 It transfers control to ``continue`` or unwinds out of the function.
5557 It is undefined behavior to execute a ``cleanupret`` whose ``cleanuppad`` has
5560 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5561 execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5562 consuming the same ``cleanuppad`` has already been executed.
5564 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5565 execution of its ``cleanuppad``, any other ``cleanuppad`` or ``catchpad`` has
5566 been executed but has not had a corresponding
5567 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5572 .. code-block:: llvm
5574 cleanupret %cleanup unwind to caller
5575 cleanupret %cleanup unwind label %continue
5579 '``terminatepad``' Instruction
5580 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5587 terminatepad [<args>*] unwind label <exception label>
5588 terminatepad [<args>*] unwind to caller
5593 The '``terminatepad``' instruction is used by `LLVM's exception handling
5594 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5595 is a terminate block --- one where a personality routine may decide to
5596 terminate the program.
5597 The ``args`` correspond to whatever information the personality
5598 routine requires to know if this is an appropriate place to terminate the
5599 program. Control is transferred to the ``exception`` label if the
5600 personality routine decides not to terminate the program for the
5601 in-flight exception.
5606 The instruction takes a list of arbitrary values which are interpreted
5607 by the :ref:`personality function <personalityfn>`.
5609 The ``terminatepad`` may be given an ``exception`` label to
5610 transfer control to if the in-flight exception matches the ``args``.
5615 When the call stack is being unwound due to an exception being thrown,
5616 the exception is compared against the ``args``. If it matches,
5617 then control is transfered to the ``exception`` basic block. Otherwise,
5618 the program is terminated via personality-specific means. Typically,
5619 the first argument to ``terminatepad`` specifies what function the
5620 personality should defer to in order to terminate the program.
5622 The ``terminatepad`` instruction has several restrictions:
5624 - A terminate block is a basic block which is the unwind destination of
5625 an exceptional instruction.
5626 - A terminate block must have a '``terminatepad``' instruction as its
5627 first non-PHI instruction.
5628 - There can be only one '``terminatepad``' instruction within the
5630 - A basic block that is not a terminate block may not include a
5631 '``terminatepad``' instruction.
5636 .. code-block:: llvm
5638 ;; A terminate block which only permits integers.
5639 terminatepad [i8** @_ZTIi] unwind label %continue
5643 '``unreachable``' Instruction
5644 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5656 The '``unreachable``' instruction has no defined semantics. This
5657 instruction is used to inform the optimizer that a particular portion of
5658 the code is not reachable. This can be used to indicate that the code
5659 after a no-return function cannot be reached, and other facts.
5664 The '``unreachable``' instruction has no defined semantics.
5671 Binary operators are used to do most of the computation in a program.
5672 They require two operands of the same type, execute an operation on
5673 them, and produce a single value. The operands might represent multiple
5674 data, as is the case with the :ref:`vector <t_vector>` data type. The
5675 result value has the same type as its operands.
5677 There are several different binary operators:
5681 '``add``' Instruction
5682 ^^^^^^^^^^^^^^^^^^^^^
5689 <result> = add <ty> <op1>, <op2> ; yields ty:result
5690 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
5691 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
5692 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
5697 The '``add``' instruction returns the sum of its two operands.
5702 The two arguments to the '``add``' instruction must be
5703 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5704 arguments must have identical types.
5709 The value produced is the integer sum of the two operands.
5711 If the sum has unsigned overflow, the result returned is the
5712 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5715 Because LLVM integers use a two's complement representation, this
5716 instruction is appropriate for both signed and unsigned integers.
5718 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5719 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5720 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
5721 unsigned and/or signed overflow, respectively, occurs.
5726 .. code-block:: llvm
5728 <result> = add i32 4, %var ; yields i32:result = 4 + %var
5732 '``fadd``' Instruction
5733 ^^^^^^^^^^^^^^^^^^^^^^
5740 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5745 The '``fadd``' instruction returns the sum of its two operands.
5750 The two arguments to the '``fadd``' instruction must be :ref:`floating
5751 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5752 Both arguments must have identical types.
5757 The value produced is the floating point sum of the two operands. This
5758 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
5759 which are optimization hints to enable otherwise unsafe floating point
5765 .. code-block:: llvm
5767 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
5769 '``sub``' Instruction
5770 ^^^^^^^^^^^^^^^^^^^^^
5777 <result> = sub <ty> <op1>, <op2> ; yields ty:result
5778 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
5779 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
5780 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
5785 The '``sub``' instruction returns the difference of its two operands.
5787 Note that the '``sub``' instruction is used to represent the '``neg``'
5788 instruction present in most other intermediate representations.
5793 The two arguments to the '``sub``' instruction must be
5794 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5795 arguments must have identical types.
5800 The value produced is the integer difference of the two operands.
5802 If the difference has unsigned overflow, the result returned is the
5803 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5806 Because LLVM integers use a two's complement representation, this
5807 instruction is appropriate for both signed and unsigned integers.
5809 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5810 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5811 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
5812 unsigned and/or signed overflow, respectively, occurs.
5817 .. code-block:: llvm
5819 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
5820 <result> = sub i32 0, %val ; yields i32:result = -%var
5824 '``fsub``' Instruction
5825 ^^^^^^^^^^^^^^^^^^^^^^
5832 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5837 The '``fsub``' instruction returns the difference of its two operands.
5839 Note that the '``fsub``' instruction is used to represent the '``fneg``'
5840 instruction present in most other intermediate representations.
5845 The two arguments to the '``fsub``' instruction must be :ref:`floating
5846 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5847 Both arguments must have identical types.
5852 The value produced is the floating point difference of the two operands.
5853 This instruction can also take any number of :ref:`fast-math
5854 flags <fastmath>`, which are optimization hints to enable otherwise
5855 unsafe floating point optimizations:
5860 .. code-block:: llvm
5862 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
5863 <result> = fsub float -0.0, %val ; yields float:result = -%var
5865 '``mul``' Instruction
5866 ^^^^^^^^^^^^^^^^^^^^^
5873 <result> = mul <ty> <op1>, <op2> ; yields ty:result
5874 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
5875 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
5876 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
5881 The '``mul``' instruction returns the product of its two operands.
5886 The two arguments to the '``mul``' instruction must be
5887 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5888 arguments must have identical types.
5893 The value produced is the integer product of the two operands.
5895 If the result of the multiplication has unsigned overflow, the result
5896 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
5897 bit width of the result.
5899 Because LLVM integers use a two's complement representation, and the
5900 result is the same width as the operands, this instruction returns the
5901 correct result for both signed and unsigned integers. If a full product
5902 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
5903 sign-extended or zero-extended as appropriate to the width of the full
5906 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5907 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5908 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
5909 unsigned and/or signed overflow, respectively, occurs.
5914 .. code-block:: llvm
5916 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
5920 '``fmul``' Instruction
5921 ^^^^^^^^^^^^^^^^^^^^^^
5928 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5933 The '``fmul``' instruction returns the product of its two operands.
5938 The two arguments to the '``fmul``' instruction must be :ref:`floating
5939 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5940 Both arguments must have identical types.
5945 The value produced is the floating point product of the two operands.
5946 This instruction can also take any number of :ref:`fast-math
5947 flags <fastmath>`, which are optimization hints to enable otherwise
5948 unsafe floating point optimizations:
5953 .. code-block:: llvm
5955 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
5957 '``udiv``' Instruction
5958 ^^^^^^^^^^^^^^^^^^^^^^
5965 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
5966 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
5971 The '``udiv``' instruction returns the quotient of its two operands.
5976 The two arguments to the '``udiv``' instruction must be
5977 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5978 arguments must have identical types.
5983 The value produced is the unsigned integer quotient of the two operands.
5985 Note that unsigned integer division and signed integer division are
5986 distinct operations; for signed integer division, use '``sdiv``'.
5988 Division by zero leads to undefined behavior.
5990 If the ``exact`` keyword is present, the result value of the ``udiv`` is
5991 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
5992 such, "((a udiv exact b) mul b) == a").
5997 .. code-block:: llvm
5999 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
6001 '``sdiv``' Instruction
6002 ^^^^^^^^^^^^^^^^^^^^^^
6009 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
6010 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
6015 The '``sdiv``' instruction returns the quotient of its two operands.
6020 The two arguments to the '``sdiv``' instruction must be
6021 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6022 arguments must have identical types.
6027 The value produced is the signed integer quotient of the two operands
6028 rounded towards zero.
6030 Note that signed integer division and unsigned integer division are
6031 distinct operations; for unsigned integer division, use '``udiv``'.
6033 Division by zero leads to undefined behavior. Overflow also leads to
6034 undefined behavior; this is a rare case, but can occur, for example, by
6035 doing a 32-bit division of -2147483648 by -1.
6037 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
6038 a :ref:`poison value <poisonvalues>` if the result would be rounded.
6043 .. code-block:: llvm
6045 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
6049 '``fdiv``' Instruction
6050 ^^^^^^^^^^^^^^^^^^^^^^
6057 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6062 The '``fdiv``' instruction returns the quotient of its two operands.
6067 The two arguments to the '``fdiv``' instruction must be :ref:`floating
6068 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6069 Both arguments must have identical types.
6074 The value produced is the floating point quotient of the two operands.
6075 This instruction can also take any number of :ref:`fast-math
6076 flags <fastmath>`, which are optimization hints to enable otherwise
6077 unsafe floating point optimizations:
6082 .. code-block:: llvm
6084 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
6086 '``urem``' Instruction
6087 ^^^^^^^^^^^^^^^^^^^^^^
6094 <result> = urem <ty> <op1>, <op2> ; yields ty:result
6099 The '``urem``' instruction returns the remainder from the unsigned
6100 division of its two arguments.
6105 The two arguments to the '``urem``' instruction must be
6106 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6107 arguments must have identical types.
6112 This instruction returns the unsigned integer *remainder* of a division.
6113 This instruction always performs an unsigned division to get the
6116 Note that unsigned integer remainder and signed integer remainder are
6117 distinct operations; for signed integer remainder, use '``srem``'.
6119 Taking the remainder of a division by zero leads to undefined behavior.
6124 .. code-block:: llvm
6126 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
6128 '``srem``' Instruction
6129 ^^^^^^^^^^^^^^^^^^^^^^
6136 <result> = srem <ty> <op1>, <op2> ; yields ty:result
6141 The '``srem``' instruction returns the remainder from the signed
6142 division of its two operands. This instruction can also take
6143 :ref:`vector <t_vector>` versions of the values in which case the elements
6149 The two arguments to the '``srem``' instruction must be
6150 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6151 arguments must have identical types.
6156 This instruction returns the *remainder* of a division (where the result
6157 is either zero or has the same sign as the dividend, ``op1``), not the
6158 *modulo* operator (where the result is either zero or has the same sign
6159 as the divisor, ``op2``) of a value. For more information about the
6160 difference, see `The Math
6161 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
6162 table of how this is implemented in various languages, please see
6164 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
6166 Note that signed integer remainder and unsigned integer remainder are
6167 distinct operations; for unsigned integer remainder, use '``urem``'.
6169 Taking the remainder of a division by zero leads to undefined behavior.
6170 Overflow also leads to undefined behavior; this is a rare case, but can
6171 occur, for example, by taking the remainder of a 32-bit division of
6172 -2147483648 by -1. (The remainder doesn't actually overflow, but this
6173 rule lets srem be implemented using instructions that return both the
6174 result of the division and the remainder.)
6179 .. code-block:: llvm
6181 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
6185 '``frem``' Instruction
6186 ^^^^^^^^^^^^^^^^^^^^^^
6193 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6198 The '``frem``' instruction returns the remainder from the division of
6204 The two arguments to the '``frem``' instruction must be :ref:`floating
6205 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6206 Both arguments must have identical types.
6211 This instruction returns the *remainder* of a division. The remainder
6212 has the same sign as the dividend. This instruction can also take any
6213 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
6214 to enable otherwise unsafe floating point optimizations:
6219 .. code-block:: llvm
6221 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
6225 Bitwise Binary Operations
6226 -------------------------
6228 Bitwise binary operators are used to do various forms of bit-twiddling
6229 in a program. They are generally very efficient instructions and can
6230 commonly be strength reduced from other instructions. They require two
6231 operands of the same type, execute an operation on them, and produce a
6232 single value. The resulting value is the same type as its operands.
6234 '``shl``' Instruction
6235 ^^^^^^^^^^^^^^^^^^^^^
6242 <result> = shl <ty> <op1>, <op2> ; yields ty:result
6243 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
6244 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
6245 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
6250 The '``shl``' instruction returns the first operand shifted to the left
6251 a specified number of bits.
6256 Both arguments to the '``shl``' instruction must be the same
6257 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6258 '``op2``' is treated as an unsigned value.
6263 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
6264 where ``n`` is the width of the result. If ``op2`` is (statically or
6265 dynamically) equal to or larger than the number of bits in
6266 ``op1``, the result is undefined. If the arguments are vectors, each
6267 vector element of ``op1`` is shifted by the corresponding shift amount
6270 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
6271 value <poisonvalues>` if it shifts out any non-zero bits. If the
6272 ``nsw`` keyword is present, then the shift produces a :ref:`poison
6273 value <poisonvalues>` if it shifts out any bits that disagree with the
6274 resultant sign bit. As such, NUW/NSW have the same semantics as they
6275 would if the shift were expressed as a mul instruction with the same
6276 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
6281 .. code-block:: llvm
6283 <result> = shl i32 4, %var ; yields i32: 4 << %var
6284 <result> = shl i32 4, 2 ; yields i32: 16
6285 <result> = shl i32 1, 10 ; yields i32: 1024
6286 <result> = shl i32 1, 32 ; undefined
6287 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
6289 '``lshr``' Instruction
6290 ^^^^^^^^^^^^^^^^^^^^^^
6297 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
6298 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
6303 The '``lshr``' instruction (logical shift right) returns the first
6304 operand shifted to the right a specified number of bits with zero fill.
6309 Both arguments to the '``lshr``' instruction must be the same
6310 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6311 '``op2``' is treated as an unsigned value.
6316 This instruction always performs a logical shift right operation. The
6317 most significant bits of the result will be filled with zero bits after
6318 the shift. If ``op2`` is (statically or dynamically) equal to or larger
6319 than the number of bits in ``op1``, the result is undefined. If the
6320 arguments are vectors, each vector element of ``op1`` is shifted by the
6321 corresponding shift amount in ``op2``.
6323 If the ``exact`` keyword is present, the result value of the ``lshr`` is
6324 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6330 .. code-block:: llvm
6332 <result> = lshr i32 4, 1 ; yields i32:result = 2
6333 <result> = lshr i32 4, 2 ; yields i32:result = 1
6334 <result> = lshr i8 4, 3 ; yields i8:result = 0
6335 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
6336 <result> = lshr i32 1, 32 ; undefined
6337 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
6339 '``ashr``' Instruction
6340 ^^^^^^^^^^^^^^^^^^^^^^
6347 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
6348 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
6353 The '``ashr``' instruction (arithmetic shift right) returns the first
6354 operand shifted to the right a specified number of bits with sign
6360 Both arguments to the '``ashr``' instruction must be the same
6361 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6362 '``op2``' is treated as an unsigned value.
6367 This instruction always performs an arithmetic shift right operation,
6368 The most significant bits of the result will be filled with the sign bit
6369 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
6370 than the number of bits in ``op1``, the result is undefined. If the
6371 arguments are vectors, each vector element of ``op1`` is shifted by the
6372 corresponding shift amount in ``op2``.
6374 If the ``exact`` keyword is present, the result value of the ``ashr`` is
6375 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6381 .. code-block:: llvm
6383 <result> = ashr i32 4, 1 ; yields i32:result = 2
6384 <result> = ashr i32 4, 2 ; yields i32:result = 1
6385 <result> = ashr i8 4, 3 ; yields i8:result = 0
6386 <result> = ashr i8 -2, 1 ; yields i8:result = -1
6387 <result> = ashr i32 1, 32 ; undefined
6388 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
6390 '``and``' Instruction
6391 ^^^^^^^^^^^^^^^^^^^^^
6398 <result> = and <ty> <op1>, <op2> ; yields ty:result
6403 The '``and``' instruction returns the bitwise logical and of its two
6409 The two arguments to the '``and``' instruction must be
6410 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6411 arguments must have identical types.
6416 The truth table used for the '``and``' instruction is:
6433 .. code-block:: llvm
6435 <result> = and i32 4, %var ; yields i32:result = 4 & %var
6436 <result> = and i32 15, 40 ; yields i32:result = 8
6437 <result> = and i32 4, 8 ; yields i32:result = 0
6439 '``or``' Instruction
6440 ^^^^^^^^^^^^^^^^^^^^
6447 <result> = or <ty> <op1>, <op2> ; yields ty:result
6452 The '``or``' instruction returns the bitwise logical inclusive or of its
6458 The two arguments to the '``or``' instruction must be
6459 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6460 arguments must have identical types.
6465 The truth table used for the '``or``' instruction is:
6484 <result> = or i32 4, %var ; yields i32:result = 4 | %var
6485 <result> = or i32 15, 40 ; yields i32:result = 47
6486 <result> = or i32 4, 8 ; yields i32:result = 12
6488 '``xor``' Instruction
6489 ^^^^^^^^^^^^^^^^^^^^^
6496 <result> = xor <ty> <op1>, <op2> ; yields ty:result
6501 The '``xor``' instruction returns the bitwise logical exclusive or of
6502 its two operands. The ``xor`` is used to implement the "one's
6503 complement" operation, which is the "~" operator in C.
6508 The two arguments to the '``xor``' instruction must be
6509 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6510 arguments must have identical types.
6515 The truth table used for the '``xor``' instruction is:
6532 .. code-block:: llvm
6534 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
6535 <result> = xor i32 15, 40 ; yields i32:result = 39
6536 <result> = xor i32 4, 8 ; yields i32:result = 12
6537 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
6542 LLVM supports several instructions to represent vector operations in a
6543 target-independent manner. These instructions cover the element-access
6544 and vector-specific operations needed to process vectors effectively.
6545 While LLVM does directly support these vector operations, many
6546 sophisticated algorithms will want to use target-specific intrinsics to
6547 take full advantage of a specific target.
6549 .. _i_extractelement:
6551 '``extractelement``' Instruction
6552 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6559 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
6564 The '``extractelement``' instruction extracts a single scalar element
6565 from a vector at a specified index.
6570 The first operand of an '``extractelement``' instruction is a value of
6571 :ref:`vector <t_vector>` type. The second operand is an index indicating
6572 the position from which to extract the element. The index may be a
6573 variable of any integer type.
6578 The result is a scalar of the same type as the element type of ``val``.
6579 Its value is the value at position ``idx`` of ``val``. If ``idx``
6580 exceeds the length of ``val``, the results are undefined.
6585 .. code-block:: llvm
6587 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
6589 .. _i_insertelement:
6591 '``insertelement``' Instruction
6592 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6599 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
6604 The '``insertelement``' instruction inserts a scalar element into a
6605 vector at a specified index.
6610 The first operand of an '``insertelement``' instruction is a value of
6611 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
6612 type must equal the element type of the first operand. The third operand
6613 is an index indicating the position at which to insert the value. The
6614 index may be a variable of any integer type.
6619 The result is a vector of the same type as ``val``. Its element values
6620 are those of ``val`` except at position ``idx``, where it gets the value
6621 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
6627 .. code-block:: llvm
6629 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
6631 .. _i_shufflevector:
6633 '``shufflevector``' Instruction
6634 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6641 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
6646 The '``shufflevector``' instruction constructs a permutation of elements
6647 from two input vectors, returning a vector with the same element type as
6648 the input and length that is the same as the shuffle mask.
6653 The first two operands of a '``shufflevector``' instruction are vectors
6654 with the same type. The third argument is a shuffle mask whose element
6655 type is always 'i32'. The result of the instruction is a vector whose
6656 length is the same as the shuffle mask and whose element type is the
6657 same as the element type of the first two operands.
6659 The shuffle mask operand is required to be a constant vector with either
6660 constant integer or undef values.
6665 The elements of the two input vectors are numbered from left to right
6666 across both of the vectors. The shuffle mask operand specifies, for each
6667 element of the result vector, which element of the two input vectors the
6668 result element gets. The element selector may be undef (meaning "don't
6669 care") and the second operand may be undef if performing a shuffle from
6675 .. code-block:: llvm
6677 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6678 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
6679 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
6680 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
6681 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
6682 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
6683 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6684 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
6686 Aggregate Operations
6687 --------------------
6689 LLVM supports several instructions for working with
6690 :ref:`aggregate <t_aggregate>` values.
6694 '``extractvalue``' Instruction
6695 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6702 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
6707 The '``extractvalue``' instruction extracts the value of a member field
6708 from an :ref:`aggregate <t_aggregate>` value.
6713 The first operand of an '``extractvalue``' instruction is a value of
6714 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The other operands are
6715 constant indices to specify which value to extract in a similar manner
6716 as indices in a '``getelementptr``' instruction.
6718 The major differences to ``getelementptr`` indexing are:
6720 - Since the value being indexed is not a pointer, the first index is
6721 omitted and assumed to be zero.
6722 - At least one index must be specified.
6723 - Not only struct indices but also array indices must be in bounds.
6728 The result is the value at the position in the aggregate specified by
6734 .. code-block:: llvm
6736 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
6740 '``insertvalue``' Instruction
6741 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6748 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
6753 The '``insertvalue``' instruction inserts a value into a member field in
6754 an :ref:`aggregate <t_aggregate>` value.
6759 The first operand of an '``insertvalue``' instruction is a value of
6760 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
6761 a first-class value to insert. The following operands are constant
6762 indices indicating the position at which to insert the value in a
6763 similar manner as indices in a '``extractvalue``' instruction. The value
6764 to insert must have the same type as the value identified by the
6770 The result is an aggregate of the same type as ``val``. Its value is
6771 that of ``val`` except that the value at the position specified by the
6772 indices is that of ``elt``.
6777 .. code-block:: llvm
6779 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
6780 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
6781 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
6785 Memory Access and Addressing Operations
6786 ---------------------------------------
6788 A key design point of an SSA-based representation is how it represents
6789 memory. In LLVM, no memory locations are in SSA form, which makes things
6790 very simple. This section describes how to read, write, and allocate
6795 '``alloca``' Instruction
6796 ^^^^^^^^^^^^^^^^^^^^^^^^
6803 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
6808 The '``alloca``' instruction allocates memory on the stack frame of the
6809 currently executing function, to be automatically released when this
6810 function returns to its caller. The object is always allocated in the
6811 generic address space (address space zero).
6816 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
6817 bytes of memory on the runtime stack, returning a pointer of the
6818 appropriate type to the program. If "NumElements" is specified, it is
6819 the number of elements allocated, otherwise "NumElements" is defaulted
6820 to be one. If a constant alignment is specified, the value result of the
6821 allocation is guaranteed to be aligned to at least that boundary. The
6822 alignment may not be greater than ``1 << 29``. If not specified, or if
6823 zero, the target can choose to align the allocation on any convenient
6824 boundary compatible with the type.
6826 '``type``' may be any sized type.
6831 Memory is allocated; a pointer is returned. The operation is undefined
6832 if there is insufficient stack space for the allocation. '``alloca``'d
6833 memory is automatically released when the function returns. The
6834 '``alloca``' instruction is commonly used to represent automatic
6835 variables that must have an address available. When the function returns
6836 (either with the ``ret`` or ``resume`` instructions), the memory is
6837 reclaimed. Allocating zero bytes is legal, but the result is undefined.
6838 The order in which memory is allocated (ie., which way the stack grows)
6844 .. code-block:: llvm
6846 %ptr = alloca i32 ; yields i32*:ptr
6847 %ptr = alloca i32, i32 4 ; yields i32*:ptr
6848 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
6849 %ptr = alloca i32, align 1024 ; yields i32*:ptr
6853 '``load``' Instruction
6854 ^^^^^^^^^^^^^^^^^^^^^^
6861 <result> = load [volatile] <ty>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !invariant.group !<index>][, !nonnull !<index>][, !dereferenceable !<deref_bytes_node>][, !dereferenceable_or_null !<deref_bytes_node>][, !align !<align_node>]
6862 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>]
6863 !<index> = !{ i32 1 }
6864 !<deref_bytes_node> = !{i64 <dereferenceable_bytes>}
6865 !<align_node> = !{ i64 <value_alignment> }
6870 The '``load``' instruction is used to read from memory.
6875 The argument to the ``load`` instruction specifies the memory address
6876 from which to load. The type specified must be a :ref:`first
6877 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
6878 then the optimizer is not allowed to modify the number or order of
6879 execution of this ``load`` with other :ref:`volatile
6880 operations <volatile>`.
6882 If the ``load`` is marked as ``atomic``, it takes an extra
6883 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6884 ``release`` and ``acq_rel`` orderings are not valid on ``load``
6885 instructions. Atomic loads produce :ref:`defined <memmodel>` results
6886 when they may see multiple atomic stores. The type of the pointee must
6887 be an integer type whose bit width is a power of two greater than or
6888 equal to eight and less than or equal to a target-specific size limit.
6889 ``align`` must be explicitly specified on atomic loads, and the load has
6890 undefined behavior if the alignment is not set to a value which is at
6891 least the size in bytes of the pointee. ``!nontemporal`` does not have
6892 any defined semantics for atomic loads.
6894 The optional constant ``align`` argument specifies the alignment of the
6895 operation (that is, the alignment of the memory address). A value of 0
6896 or an omitted ``align`` argument means that the operation has the ABI
6897 alignment for the target. It is the responsibility of the code emitter
6898 to ensure that the alignment information is correct. Overestimating the
6899 alignment results in undefined behavior. Underestimating the alignment
6900 may produce less efficient code. An alignment of 1 is always safe. The
6901 maximum possible alignment is ``1 << 29``.
6903 The optional ``!nontemporal`` metadata must reference a single
6904 metadata name ``<index>`` corresponding to a metadata node with one
6905 ``i32`` entry of value 1. The existence of the ``!nontemporal``
6906 metadata on the instruction tells the optimizer and code generator
6907 that this load is not expected to be reused in the cache. The code
6908 generator may select special instructions to save cache bandwidth, such
6909 as the ``MOVNT`` instruction on x86.
6911 The optional ``!invariant.load`` metadata must reference a single
6912 metadata name ``<index>`` corresponding to a metadata node with no
6913 entries. The existence of the ``!invariant.load`` metadata on the
6914 instruction tells the optimizer and code generator that the address
6915 operand to this load points to memory which can be assumed unchanged.
6916 Being invariant does not imply that a location is dereferenceable,
6917 but it does imply that once the location is known dereferenceable
6918 its value is henceforth unchanging.
6920 The optional ``!invariant.group`` metadata must reference a single metadata name
6921 ``<index>`` corresponding to a metadata node. See ``invariant.group`` metadata.
6923 The optional ``!nonnull`` metadata must reference a single
6924 metadata name ``<index>`` corresponding to a metadata node with no
6925 entries. The existence of the ``!nonnull`` metadata on the
6926 instruction tells the optimizer that the value loaded is known to
6927 never be null. This is analogous to the ``nonnull`` attribute
6928 on parameters and return values. This metadata can only be applied
6929 to loads of a pointer type.
6931 The optional ``!dereferenceable`` metadata must reference a single metadata
6932 name ``<deref_bytes_node>`` corresponding to a metadata node with one ``i64``
6933 entry. The existence of the ``!dereferenceable`` metadata on the instruction
6934 tells the optimizer that the value loaded is known to be dereferenceable.
6935 The number of bytes known to be dereferenceable is specified by the integer
6936 value in the metadata node. This is analogous to the ''dereferenceable''
6937 attribute on parameters and return values. This metadata can only be applied
6938 to loads of a pointer type.
6940 The optional ``!dereferenceable_or_null`` metadata must reference a single
6941 metadata name ``<deref_bytes_node>`` corresponding to a metadata node with one
6942 ``i64`` entry. The existence of the ``!dereferenceable_or_null`` metadata on the
6943 instruction tells the optimizer that the value loaded is known to be either
6944 dereferenceable or null.
6945 The number of bytes known to be dereferenceable is specified by the integer
6946 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
6947 attribute on parameters and return values. This metadata can only be applied
6948 to loads of a pointer type.
6950 The optional ``!align`` metadata must reference a single metadata name
6951 ``<align_node>`` corresponding to a metadata node with one ``i64`` entry.
6952 The existence of the ``!align`` metadata on the instruction tells the
6953 optimizer that the value loaded is known to be aligned to a boundary specified
6954 by the integer value in the metadata node. The alignment must be a power of 2.
6955 This is analogous to the ''align'' attribute on parameters and return values.
6956 This metadata can only be applied to loads of a pointer type.
6961 The location of memory pointed to is loaded. If the value being loaded
6962 is of scalar type then the number of bytes read does not exceed the
6963 minimum number of bytes needed to hold all bits of the type. For
6964 example, loading an ``i24`` reads at most three bytes. When loading a
6965 value of a type like ``i20`` with a size that is not an integral number
6966 of bytes, the result is undefined if the value was not originally
6967 written using a store of the same type.
6972 .. code-block:: llvm
6974 %ptr = alloca i32 ; yields i32*:ptr
6975 store i32 3, i32* %ptr ; yields void
6976 %val = load i32, i32* %ptr ; yields i32:val = i32 3
6980 '``store``' Instruction
6981 ^^^^^^^^^^^^^^^^^^^^^^^
6988 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.group !<index>] ; yields void
6989 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>] ; yields void
6994 The '``store``' instruction is used to write to memory.
6999 There are two arguments to the ``store`` instruction: a value to store
7000 and an address at which to store it. The type of the ``<pointer>``
7001 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
7002 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
7003 then the optimizer is not allowed to modify the number or order of
7004 execution of this ``store`` with other :ref:`volatile
7005 operations <volatile>`.
7007 If the ``store`` is marked as ``atomic``, it takes an extra
7008 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
7009 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
7010 instructions. Atomic loads produce :ref:`defined <memmodel>` results
7011 when they may see multiple atomic stores. The type of the pointee must
7012 be an integer type whose bit width is a power of two greater than or
7013 equal to eight and less than or equal to a target-specific size limit.
7014 ``align`` must be explicitly specified on atomic stores, and the store
7015 has undefined behavior if the alignment is not set to a value which is
7016 at least the size in bytes of the pointee. ``!nontemporal`` does not
7017 have any defined semantics for atomic stores.
7019 The optional constant ``align`` argument specifies the alignment of the
7020 operation (that is, the alignment of the memory address). A value of 0
7021 or an omitted ``align`` argument means that the operation has the ABI
7022 alignment for the target. It is the responsibility of the code emitter
7023 to ensure that the alignment information is correct. Overestimating the
7024 alignment results in undefined behavior. Underestimating the
7025 alignment may produce less efficient code. An alignment of 1 is always
7026 safe. The maximum possible alignment is ``1 << 29``.
7028 The optional ``!nontemporal`` metadata must reference a single metadata
7029 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
7030 value 1. The existence of the ``!nontemporal`` metadata on the instruction
7031 tells the optimizer and code generator that this load is not expected to
7032 be reused in the cache. The code generator may select special
7033 instructions to save cache bandwidth, such as the MOVNT instruction on
7036 The optional ``!invariant.group`` metadata must reference a
7037 single metadata name ``<index>``. See ``invariant.group`` metadata.
7042 The contents of memory are updated to contain ``<value>`` at the
7043 location specified by the ``<pointer>`` operand. If ``<value>`` is
7044 of scalar type then the number of bytes written does not exceed the
7045 minimum number of bytes needed to hold all bits of the type. For
7046 example, storing an ``i24`` writes at most three bytes. When writing a
7047 value of a type like ``i20`` with a size that is not an integral number
7048 of bytes, it is unspecified what happens to the extra bits that do not
7049 belong to the type, but they will typically be overwritten.
7054 .. code-block:: llvm
7056 %ptr = alloca i32 ; yields i32*:ptr
7057 store i32 3, i32* %ptr ; yields void
7058 %val = load i32, i32* %ptr ; yields i32:val = i32 3
7062 '``fence``' Instruction
7063 ^^^^^^^^^^^^^^^^^^^^^^^
7070 fence [singlethread] <ordering> ; yields void
7075 The '``fence``' instruction is used to introduce happens-before edges
7081 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
7082 defines what *synchronizes-with* edges they add. They can only be given
7083 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
7088 A fence A which has (at least) ``release`` ordering semantics
7089 *synchronizes with* a fence B with (at least) ``acquire`` ordering
7090 semantics if and only if there exist atomic operations X and Y, both
7091 operating on some atomic object M, such that A is sequenced before X, X
7092 modifies M (either directly or through some side effect of a sequence
7093 headed by X), Y is sequenced before B, and Y observes M. This provides a
7094 *happens-before* dependency between A and B. Rather than an explicit
7095 ``fence``, one (but not both) of the atomic operations X or Y might
7096 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
7097 still *synchronize-with* the explicit ``fence`` and establish the
7098 *happens-before* edge.
7100 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
7101 ``acquire`` and ``release`` semantics specified above, participates in
7102 the global program order of other ``seq_cst`` operations and/or fences.
7104 The optional ":ref:`singlethread <singlethread>`" argument specifies
7105 that the fence only synchronizes with other fences in the same thread.
7106 (This is useful for interacting with signal handlers.)
7111 .. code-block:: llvm
7113 fence acquire ; yields void
7114 fence singlethread seq_cst ; yields void
7118 '``cmpxchg``' Instruction
7119 ^^^^^^^^^^^^^^^^^^^^^^^^^
7126 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
7131 The '``cmpxchg``' instruction is used to atomically modify memory. It
7132 loads a value in memory and compares it to a given value. If they are
7133 equal, it tries to store a new value into the memory.
7138 There are three arguments to the '``cmpxchg``' instruction: an address
7139 to operate on, a value to compare to the value currently be at that
7140 address, and a new value to place at that address if the compared values
7141 are equal. The type of '<cmp>' must be an integer type whose bit width
7142 is a power of two greater than or equal to eight and less than or equal
7143 to a target-specific size limit. '<cmp>' and '<new>' must have the same
7144 type, and the type of '<pointer>' must be a pointer to that type. If the
7145 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
7146 to modify the number or order of execution of this ``cmpxchg`` with
7147 other :ref:`volatile operations <volatile>`.
7149 The success and failure :ref:`ordering <ordering>` arguments specify how this
7150 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
7151 must be at least ``monotonic``, the ordering constraint on failure must be no
7152 stronger than that on success, and the failure ordering cannot be either
7153 ``release`` or ``acq_rel``.
7155 The optional "``singlethread``" argument declares that the ``cmpxchg``
7156 is only atomic with respect to code (usually signal handlers) running in
7157 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
7158 respect to all other code in the system.
7160 The pointer passed into cmpxchg must have alignment greater than or
7161 equal to the size in memory of the operand.
7166 The contents of memory at the location specified by the '``<pointer>``' operand
7167 is read and compared to '``<cmp>``'; if the read value is the equal, the
7168 '``<new>``' is written. The original value at the location is returned, together
7169 with a flag indicating success (true) or failure (false).
7171 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
7172 permitted: the operation may not write ``<new>`` even if the comparison
7175 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
7176 if the value loaded equals ``cmp``.
7178 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
7179 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
7180 load with an ordering parameter determined the second ordering parameter.
7185 .. code-block:: llvm
7188 %orig = atomic load i32, i32* %ptr unordered ; yields i32
7192 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
7193 %squared = mul i32 %cmp, %cmp
7194 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
7195 %value_loaded = extractvalue { i32, i1 } %val_success, 0
7196 %success = extractvalue { i32, i1 } %val_success, 1
7197 br i1 %success, label %done, label %loop
7204 '``atomicrmw``' Instruction
7205 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7212 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
7217 The '``atomicrmw``' instruction is used to atomically modify memory.
7222 There are three arguments to the '``atomicrmw``' instruction: an
7223 operation to apply, an address whose value to modify, an argument to the
7224 operation. The operation must be one of the following keywords:
7238 The type of '<value>' must be an integer type whose bit width is a power
7239 of two greater than or equal to eight and less than or equal to a
7240 target-specific size limit. The type of the '``<pointer>``' operand must
7241 be a pointer to that type. If the ``atomicrmw`` is marked as
7242 ``volatile``, then the optimizer is not allowed to modify the number or
7243 order of execution of this ``atomicrmw`` with other :ref:`volatile
7244 operations <volatile>`.
7249 The contents of memory at the location specified by the '``<pointer>``'
7250 operand are atomically read, modified, and written back. The original
7251 value at the location is returned. The modification is specified by the
7254 - xchg: ``*ptr = val``
7255 - add: ``*ptr = *ptr + val``
7256 - sub: ``*ptr = *ptr - val``
7257 - and: ``*ptr = *ptr & val``
7258 - nand: ``*ptr = ~(*ptr & val)``
7259 - or: ``*ptr = *ptr | val``
7260 - xor: ``*ptr = *ptr ^ val``
7261 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
7262 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
7263 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
7265 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
7271 .. code-block:: llvm
7273 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
7275 .. _i_getelementptr:
7277 '``getelementptr``' Instruction
7278 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7285 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7286 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7287 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
7292 The '``getelementptr``' instruction is used to get the address of a
7293 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
7294 address calculation only and does not access memory. The instruction can also
7295 be used to calculate a vector of such addresses.
7300 The first argument is always a type used as the basis for the calculations.
7301 The second argument is always a pointer or a vector of pointers, and is the
7302 base address to start from. The remaining arguments are indices
7303 that indicate which of the elements of the aggregate object are indexed.
7304 The interpretation of each index is dependent on the type being indexed
7305 into. The first index always indexes the pointer value given as the
7306 first argument, the second index indexes a value of the type pointed to
7307 (not necessarily the value directly pointed to, since the first index
7308 can be non-zero), etc. The first type indexed into must be a pointer
7309 value, subsequent types can be arrays, vectors, and structs. Note that
7310 subsequent types being indexed into can never be pointers, since that
7311 would require loading the pointer before continuing calculation.
7313 The type of each index argument depends on the type it is indexing into.
7314 When indexing into a (optionally packed) structure, only ``i32`` integer
7315 **constants** are allowed (when using a vector of indices they must all
7316 be the **same** ``i32`` integer constant). When indexing into an array,
7317 pointer or vector, integers of any width are allowed, and they are not
7318 required to be constant. These integers are treated as signed values
7321 For example, let's consider a C code fragment and how it gets compiled
7337 int *foo(struct ST *s) {
7338 return &s[1].Z.B[5][13];
7341 The LLVM code generated by Clang is:
7343 .. code-block:: llvm
7345 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
7346 %struct.ST = type { i32, double, %struct.RT }
7348 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
7350 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
7357 In the example above, the first index is indexing into the
7358 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
7359 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
7360 indexes into the third element of the structure, yielding a
7361 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
7362 structure. The third index indexes into the second element of the
7363 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
7364 dimensions of the array are subscripted into, yielding an '``i32``'
7365 type. The '``getelementptr``' instruction returns a pointer to this
7366 element, thus computing a value of '``i32*``' type.
7368 Note that it is perfectly legal to index partially through a structure,
7369 returning a pointer to an inner element. Because of this, the LLVM code
7370 for the given testcase is equivalent to:
7372 .. code-block:: llvm
7374 define i32* @foo(%struct.ST* %s) {
7375 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
7376 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
7377 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
7378 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
7379 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
7383 If the ``inbounds`` keyword is present, the result value of the
7384 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
7385 pointer is not an *in bounds* address of an allocated object, or if any
7386 of the addresses that would be formed by successive addition of the
7387 offsets implied by the indices to the base address with infinitely
7388 precise signed arithmetic are not an *in bounds* address of that
7389 allocated object. The *in bounds* addresses for an allocated object are
7390 all the addresses that point into the object, plus the address one byte
7391 past the end. In cases where the base is a vector of pointers the
7392 ``inbounds`` keyword applies to each of the computations element-wise.
7394 If the ``inbounds`` keyword is not present, the offsets are added to the
7395 base address with silently-wrapping two's complement arithmetic. If the
7396 offsets have a different width from the pointer, they are sign-extended
7397 or truncated to the width of the pointer. The result value of the
7398 ``getelementptr`` may be outside the object pointed to by the base
7399 pointer. The result value may not necessarily be used to access memory
7400 though, even if it happens to point into allocated storage. See the
7401 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
7404 The getelementptr instruction is often confusing. For some more insight
7405 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
7410 .. code-block:: llvm
7412 ; yields [12 x i8]*:aptr
7413 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
7415 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
7417 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
7419 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
7424 The ``getelementptr`` returns a vector of pointers, instead of a single address,
7425 when one or more of its arguments is a vector. In such cases, all vector
7426 arguments should have the same number of elements, and every scalar argument
7427 will be effectively broadcast into a vector during address calculation.
7429 .. code-block:: llvm
7431 ; All arguments are vectors:
7432 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
7433 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
7435 ; Add the same scalar offset to each pointer of a vector:
7436 ; A[i] = ptrs[i] + offset*sizeof(i8)
7437 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
7439 ; Add distinct offsets to the same pointer:
7440 ; A[i] = ptr + offsets[i]*sizeof(i8)
7441 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
7443 ; In all cases described above the type of the result is <4 x i8*>
7445 The two following instructions are equivalent:
7447 .. code-block:: llvm
7449 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7450 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
7451 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
7453 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
7455 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7456 i32 2, i32 1, <4 x i32> %ind4, i64 13
7458 Let's look at the C code, where the vector version of ``getelementptr``
7463 // Let's assume that we vectorize the following loop:
7464 double *A, B; int *C;
7465 for (int i = 0; i < size; ++i) {
7469 .. code-block:: llvm
7471 ; get pointers for 8 elements from array B
7472 %ptrs = getelementptr double, double* %B, <8 x i32> %C
7473 ; load 8 elements from array B into A
7474 %A = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs,
7475 i32 8, <8 x i1> %mask, <8 x double> %passthru)
7477 Conversion Operations
7478 ---------------------
7480 The instructions in this category are the conversion instructions
7481 (casting) which all take a single operand and a type. They perform
7482 various bit conversions on the operand.
7484 '``trunc .. to``' Instruction
7485 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7492 <result> = trunc <ty> <value> to <ty2> ; yields ty2
7497 The '``trunc``' instruction truncates its operand to the type ``ty2``.
7502 The '``trunc``' instruction takes a value to trunc, and a type to trunc
7503 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
7504 of the same number of integers. The bit size of the ``value`` must be
7505 larger than the bit size of the destination type, ``ty2``. Equal sized
7506 types are not allowed.
7511 The '``trunc``' instruction truncates the high order bits in ``value``
7512 and converts the remaining bits to ``ty2``. Since the source size must
7513 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
7514 It will always truncate bits.
7519 .. code-block:: llvm
7521 %X = trunc i32 257 to i8 ; yields i8:1
7522 %Y = trunc i32 123 to i1 ; yields i1:true
7523 %Z = trunc i32 122 to i1 ; yields i1:false
7524 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
7526 '``zext .. to``' Instruction
7527 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7534 <result> = zext <ty> <value> to <ty2> ; yields ty2
7539 The '``zext``' instruction zero extends its operand to type ``ty2``.
7544 The '``zext``' instruction takes a value to cast, and a type to cast it
7545 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7546 the same number of integers. The bit size of the ``value`` must be
7547 smaller than the bit size of the destination type, ``ty2``.
7552 The ``zext`` fills the high order bits of the ``value`` with zero bits
7553 until it reaches the size of the destination type, ``ty2``.
7555 When zero extending from i1, the result will always be either 0 or 1.
7560 .. code-block:: llvm
7562 %X = zext i32 257 to i64 ; yields i64:257
7563 %Y = zext i1 true to i32 ; yields i32:1
7564 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7566 '``sext .. to``' Instruction
7567 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7574 <result> = sext <ty> <value> to <ty2> ; yields ty2
7579 The '``sext``' sign extends ``value`` to the type ``ty2``.
7584 The '``sext``' instruction takes a value to cast, and a type to cast it
7585 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7586 the same number of integers. The bit size of the ``value`` must be
7587 smaller than the bit size of the destination type, ``ty2``.
7592 The '``sext``' instruction performs a sign extension by copying the sign
7593 bit (highest order bit) of the ``value`` until it reaches the bit size
7594 of the type ``ty2``.
7596 When sign extending from i1, the extension always results in -1 or 0.
7601 .. code-block:: llvm
7603 %X = sext i8 -1 to i16 ; yields i16 :65535
7604 %Y = sext i1 true to i32 ; yields i32:-1
7605 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7607 '``fptrunc .. to``' Instruction
7608 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7615 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
7620 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
7625 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
7626 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
7627 The size of ``value`` must be larger than the size of ``ty2``. This
7628 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
7633 The '``fptrunc``' instruction casts a ``value`` from a larger
7634 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
7635 point <t_floating>` type. If the value cannot fit (i.e. overflows) within the
7636 destination type, ``ty2``, then the results are undefined. If the cast produces
7637 an inexact result, how rounding is performed (e.g. truncation, also known as
7638 round to zero) is undefined.
7643 .. code-block:: llvm
7645 %X = fptrunc double 123.0 to float ; yields float:123.0
7646 %Y = fptrunc double 1.0E+300 to float ; yields undefined
7648 '``fpext .. to``' Instruction
7649 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7656 <result> = fpext <ty> <value> to <ty2> ; yields ty2
7661 The '``fpext``' extends a floating point ``value`` to a larger floating
7667 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
7668 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
7669 to. The source type must be smaller than the destination type.
7674 The '``fpext``' instruction extends the ``value`` from a smaller
7675 :ref:`floating point <t_floating>` type to a larger :ref:`floating
7676 point <t_floating>` type. The ``fpext`` cannot be used to make a
7677 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
7678 *no-op cast* for a floating point cast.
7683 .. code-block:: llvm
7685 %X = fpext float 3.125 to double ; yields double:3.125000e+00
7686 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
7688 '``fptoui .. to``' Instruction
7689 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7696 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
7701 The '``fptoui``' converts a floating point ``value`` to its unsigned
7702 integer equivalent of type ``ty2``.
7707 The '``fptoui``' instruction takes a value to cast, which must be a
7708 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7709 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7710 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7711 type with the same number of elements as ``ty``
7716 The '``fptoui``' instruction converts its :ref:`floating
7717 point <t_floating>` operand into the nearest (rounding towards zero)
7718 unsigned integer value. If the value cannot fit in ``ty2``, the results
7724 .. code-block:: llvm
7726 %X = fptoui double 123.0 to i32 ; yields i32:123
7727 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
7728 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
7730 '``fptosi .. to``' Instruction
7731 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7738 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
7743 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
7744 ``value`` to type ``ty2``.
7749 The '``fptosi``' instruction takes a value to cast, which must be a
7750 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7751 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7752 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7753 type with the same number of elements as ``ty``
7758 The '``fptosi``' instruction converts its :ref:`floating
7759 point <t_floating>` operand into the nearest (rounding towards zero)
7760 signed integer value. If the value cannot fit in ``ty2``, the results
7766 .. code-block:: llvm
7768 %X = fptosi double -123.0 to i32 ; yields i32:-123
7769 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
7770 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
7772 '``uitofp .. to``' Instruction
7773 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7780 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
7785 The '``uitofp``' instruction regards ``value`` as an unsigned integer
7786 and converts that value to the ``ty2`` type.
7791 The '``uitofp``' instruction takes a value to cast, which must be a
7792 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7793 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7794 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7795 type with the same number of elements as ``ty``
7800 The '``uitofp``' instruction interprets its operand as an unsigned
7801 integer quantity and converts it to the corresponding floating point
7802 value. If the value cannot fit in the floating point value, the results
7808 .. code-block:: llvm
7810 %X = uitofp i32 257 to float ; yields float:257.0
7811 %Y = uitofp i8 -1 to double ; yields double:255.0
7813 '``sitofp .. to``' Instruction
7814 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7821 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
7826 The '``sitofp``' instruction regards ``value`` as a signed integer and
7827 converts that value to the ``ty2`` type.
7832 The '``sitofp``' instruction takes a value to cast, which must be a
7833 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7834 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7835 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7836 type with the same number of elements as ``ty``
7841 The '``sitofp``' instruction interprets its operand as a signed integer
7842 quantity and converts it to the corresponding floating point value. If
7843 the value cannot fit in the floating point value, the results are
7849 .. code-block:: llvm
7851 %X = sitofp i32 257 to float ; yields float:257.0
7852 %Y = sitofp i8 -1 to double ; yields double:-1.0
7856 '``ptrtoint .. to``' Instruction
7857 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7864 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
7869 The '``ptrtoint``' instruction converts the pointer or a vector of
7870 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
7875 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
7876 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
7877 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
7878 a vector of integers type.
7883 The '``ptrtoint``' instruction converts ``value`` to integer type
7884 ``ty2`` by interpreting the pointer value as an integer and either
7885 truncating or zero extending that value to the size of the integer type.
7886 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
7887 ``value`` is larger than ``ty2`` then a truncation is done. If they are
7888 the same size, then nothing is done (*no-op cast*) other than a type
7894 .. code-block:: llvm
7896 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
7897 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
7898 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
7902 '``inttoptr .. to``' Instruction
7903 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7910 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
7915 The '``inttoptr``' instruction converts an integer ``value`` to a
7916 pointer type, ``ty2``.
7921 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
7922 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
7928 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
7929 applying either a zero extension or a truncation depending on the size
7930 of the integer ``value``. If ``value`` is larger than the size of a
7931 pointer then a truncation is done. If ``value`` is smaller than the size
7932 of a pointer then a zero extension is done. If they are the same size,
7933 nothing is done (*no-op cast*).
7938 .. code-block:: llvm
7940 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
7941 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
7942 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
7943 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
7947 '``bitcast .. to``' Instruction
7948 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7955 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
7960 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
7966 The '``bitcast``' instruction takes a value to cast, which must be a
7967 non-aggregate first class value, and a type to cast it to, which must
7968 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
7969 bit sizes of ``value`` and the destination type, ``ty2``, must be
7970 identical. If the source type is a pointer, the destination type must
7971 also be a pointer of the same size. This instruction supports bitwise
7972 conversion of vectors to integers and to vectors of other types (as
7973 long as they have the same size).
7978 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
7979 is always a *no-op cast* because no bits change with this
7980 conversion. The conversion is done as if the ``value`` had been stored
7981 to memory and read back as type ``ty2``. Pointer (or vector of
7982 pointers) types may only be converted to other pointer (or vector of
7983 pointers) types with the same address space through this instruction.
7984 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
7985 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
7990 .. code-block:: llvm
7992 %X = bitcast i8 255 to i8 ; yields i8 :-1
7993 %Y = bitcast i32* %x to sint* ; yields sint*:%x
7994 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
7995 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
7997 .. _i_addrspacecast:
7999 '``addrspacecast .. to``' Instruction
8000 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8007 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
8012 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
8013 address space ``n`` to type ``pty2`` in address space ``m``.
8018 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
8019 to cast and a pointer type to cast it to, which must have a different
8025 The '``addrspacecast``' instruction converts the pointer value
8026 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
8027 value modification, depending on the target and the address space
8028 pair. Pointer conversions within the same address space must be
8029 performed with the ``bitcast`` instruction. Note that if the address space
8030 conversion is legal then both result and operand refer to the same memory
8036 .. code-block:: llvm
8038 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
8039 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
8040 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
8047 The instructions in this category are the "miscellaneous" instructions,
8048 which defy better classification.
8052 '``icmp``' Instruction
8053 ^^^^^^^^^^^^^^^^^^^^^^
8060 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8065 The '``icmp``' instruction returns a boolean value or a vector of
8066 boolean values based on comparison of its two integer, integer vector,
8067 pointer, or pointer vector operands.
8072 The '``icmp``' instruction takes three operands. The first operand is
8073 the condition code indicating the kind of comparison to perform. It is
8074 not a value, just a keyword. The possible condition code are:
8077 #. ``ne``: not equal
8078 #. ``ugt``: unsigned greater than
8079 #. ``uge``: unsigned greater or equal
8080 #. ``ult``: unsigned less than
8081 #. ``ule``: unsigned less or equal
8082 #. ``sgt``: signed greater than
8083 #. ``sge``: signed greater or equal
8084 #. ``slt``: signed less than
8085 #. ``sle``: signed less or equal
8087 The remaining two arguments must be :ref:`integer <t_integer>` or
8088 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
8089 must also be identical types.
8094 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
8095 code given as ``cond``. The comparison performed always yields either an
8096 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
8098 #. ``eq``: yields ``true`` if the operands are equal, ``false``
8099 otherwise. No sign interpretation is necessary or performed.
8100 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
8101 otherwise. No sign interpretation is necessary or performed.
8102 #. ``ugt``: interprets the operands as unsigned values and yields
8103 ``true`` if ``op1`` is greater than ``op2``.
8104 #. ``uge``: interprets the operands as unsigned values and yields
8105 ``true`` if ``op1`` is greater than or equal to ``op2``.
8106 #. ``ult``: interprets the operands as unsigned values and yields
8107 ``true`` if ``op1`` is less than ``op2``.
8108 #. ``ule``: interprets the operands as unsigned values and yields
8109 ``true`` if ``op1`` is less than or equal to ``op2``.
8110 #. ``sgt``: interprets the operands as signed values and yields ``true``
8111 if ``op1`` is greater than ``op2``.
8112 #. ``sge``: interprets the operands as signed values and yields ``true``
8113 if ``op1`` is greater than or equal to ``op2``.
8114 #. ``slt``: interprets the operands as signed values and yields ``true``
8115 if ``op1`` is less than ``op2``.
8116 #. ``sle``: interprets the operands as signed values and yields ``true``
8117 if ``op1`` is less than or equal to ``op2``.
8119 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
8120 are compared as if they were integers.
8122 If the operands are integer vectors, then they are compared element by
8123 element. The result is an ``i1`` vector with the same number of elements
8124 as the values being compared. Otherwise, the result is an ``i1``.
8129 .. code-block:: llvm
8131 <result> = icmp eq i32 4, 5 ; yields: result=false
8132 <result> = icmp ne float* %X, %X ; yields: result=false
8133 <result> = icmp ult i16 4, 5 ; yields: result=true
8134 <result> = icmp sgt i16 4, 5 ; yields: result=false
8135 <result> = icmp ule i16 -4, 5 ; yields: result=false
8136 <result> = icmp sge i16 4, 5 ; yields: result=false
8138 Note that the code generator does not yet support vector types with the
8139 ``icmp`` instruction.
8143 '``fcmp``' Instruction
8144 ^^^^^^^^^^^^^^^^^^^^^^
8151 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8156 The '``fcmp``' instruction returns a boolean value or vector of boolean
8157 values based on comparison of its operands.
8159 If the operands are floating point scalars, then the result type is a
8160 boolean (:ref:`i1 <t_integer>`).
8162 If the operands are floating point vectors, then the result type is a
8163 vector of boolean with the same number of elements as the operands being
8169 The '``fcmp``' instruction takes three operands. The first operand is
8170 the condition code indicating the kind of comparison to perform. It is
8171 not a value, just a keyword. The possible condition code are:
8173 #. ``false``: no comparison, always returns false
8174 #. ``oeq``: ordered and equal
8175 #. ``ogt``: ordered and greater than
8176 #. ``oge``: ordered and greater than or equal
8177 #. ``olt``: ordered and less than
8178 #. ``ole``: ordered and less than or equal
8179 #. ``one``: ordered and not equal
8180 #. ``ord``: ordered (no nans)
8181 #. ``ueq``: unordered or equal
8182 #. ``ugt``: unordered or greater than
8183 #. ``uge``: unordered or greater than or equal
8184 #. ``ult``: unordered or less than
8185 #. ``ule``: unordered or less than or equal
8186 #. ``une``: unordered or not equal
8187 #. ``uno``: unordered (either nans)
8188 #. ``true``: no comparison, always returns true
8190 *Ordered* means that neither operand is a QNAN while *unordered* means
8191 that either operand may be a QNAN.
8193 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
8194 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
8195 type. They must have identical types.
8200 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
8201 condition code given as ``cond``. If the operands are vectors, then the
8202 vectors are compared element by element. Each comparison performed
8203 always yields an :ref:`i1 <t_integer>` result, as follows:
8205 #. ``false``: always yields ``false``, regardless of operands.
8206 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
8207 is equal to ``op2``.
8208 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
8209 is greater than ``op2``.
8210 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
8211 is greater than or equal to ``op2``.
8212 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
8213 is less than ``op2``.
8214 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
8215 is less than or equal to ``op2``.
8216 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
8217 is not equal to ``op2``.
8218 #. ``ord``: yields ``true`` if both operands are not a QNAN.
8219 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
8221 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
8222 greater than ``op2``.
8223 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
8224 greater than or equal to ``op2``.
8225 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
8227 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
8228 less than or equal to ``op2``.
8229 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
8230 not equal to ``op2``.
8231 #. ``uno``: yields ``true`` if either operand is a QNAN.
8232 #. ``true``: always yields ``true``, regardless of operands.
8234 The ``fcmp`` instruction can also optionally take any number of
8235 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8236 otherwise unsafe floating point optimizations.
8238 Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
8239 only flags that have any effect on its semantics are those that allow
8240 assumptions to be made about the values of input arguments; namely
8241 ``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
8246 .. code-block:: llvm
8248 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
8249 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
8250 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
8251 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
8253 Note that the code generator does not yet support vector types with the
8254 ``fcmp`` instruction.
8258 '``phi``' Instruction
8259 ^^^^^^^^^^^^^^^^^^^^^
8266 <result> = phi <ty> [ <val0>, <label0>], ...
8271 The '``phi``' instruction is used to implement the φ node in the SSA
8272 graph representing the function.
8277 The type of the incoming values is specified with the first type field.
8278 After this, the '``phi``' instruction takes a list of pairs as
8279 arguments, with one pair for each predecessor basic block of the current
8280 block. Only values of :ref:`first class <t_firstclass>` type may be used as
8281 the value arguments to the PHI node. Only labels may be used as the
8284 There must be no non-phi instructions between the start of a basic block
8285 and the PHI instructions: i.e. PHI instructions must be first in a basic
8288 For the purposes of the SSA form, the use of each incoming value is
8289 deemed to occur on the edge from the corresponding predecessor block to
8290 the current block (but after any definition of an '``invoke``'
8291 instruction's return value on the same edge).
8296 At runtime, the '``phi``' instruction logically takes on the value
8297 specified by the pair corresponding to the predecessor basic block that
8298 executed just prior to the current block.
8303 .. code-block:: llvm
8305 Loop: ; Infinite loop that counts from 0 on up...
8306 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
8307 %nextindvar = add i32 %indvar, 1
8312 '``select``' Instruction
8313 ^^^^^^^^^^^^^^^^^^^^^^^^
8320 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
8322 selty is either i1 or {<N x i1>}
8327 The '``select``' instruction is used to choose one value based on a
8328 condition, without IR-level branching.
8333 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
8334 values indicating the condition, and two values of the same :ref:`first
8335 class <t_firstclass>` type.
8340 If the condition is an i1 and it evaluates to 1, the instruction returns
8341 the first value argument; otherwise, it returns the second value
8344 If the condition is a vector of i1, then the value arguments must be
8345 vectors of the same size, and the selection is done element by element.
8347 If the condition is an i1 and the value arguments are vectors of the
8348 same size, then an entire vector is selected.
8353 .. code-block:: llvm
8355 %X = select i1 true, i8 17, i8 42 ; yields i8:17
8359 '``call``' Instruction
8360 ^^^^^^^^^^^^^^^^^^^^^^
8367 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
8373 The '``call``' instruction represents a simple function call.
8378 This instruction requires several arguments:
8380 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
8381 should perform tail call optimization. The ``tail`` marker is a hint that
8382 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
8383 means that the call must be tail call optimized in order for the program to
8384 be correct. The ``musttail`` marker provides these guarantees:
8386 #. The call will not cause unbounded stack growth if it is part of a
8387 recursive cycle in the call graph.
8388 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
8391 Both markers imply that the callee does not access allocas or varargs from
8392 the caller. Calls marked ``musttail`` must obey the following additional
8395 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
8396 or a pointer bitcast followed by a ret instruction.
8397 - The ret instruction must return the (possibly bitcasted) value
8398 produced by the call or void.
8399 - The caller and callee prototypes must match. Pointer types of
8400 parameters or return types may differ in pointee type, but not
8402 - The calling conventions of the caller and callee must match.
8403 - All ABI-impacting function attributes, such as sret, byval, inreg,
8404 returned, and inalloca, must match.
8405 - The callee must be varargs iff the caller is varargs. Bitcasting a
8406 non-varargs function to the appropriate varargs type is legal so
8407 long as the non-varargs prefixes obey the other rules.
8409 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
8410 the following conditions are met:
8412 - Caller and callee both have the calling convention ``fastcc``.
8413 - The call is in tail position (ret immediately follows call and ret
8414 uses value of call or is void).
8415 - Option ``-tailcallopt`` is enabled, or
8416 ``llvm::GuaranteedTailCallOpt`` is ``true``.
8417 - `Platform-specific constraints are
8418 met. <CodeGenerator.html#tailcallopt>`_
8420 #. The optional "cconv" marker indicates which :ref:`calling
8421 convention <callingconv>` the call should use. If none is
8422 specified, the call defaults to using C calling conventions. The
8423 calling convention of the call must match the calling convention of
8424 the target function, or else the behavior is undefined.
8425 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
8426 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
8428 #. '``ty``': the type of the call instruction itself which is also the
8429 type of the return value. Functions that return no value are marked
8431 #. '``fnty``': shall be the signature of the pointer to function value
8432 being invoked. The argument types must match the types implied by
8433 this signature. This type can be omitted if the function is not
8434 varargs and if the function type does not return a pointer to a
8436 #. '``fnptrval``': An LLVM value containing a pointer to a function to
8437 be invoked. In most cases, this is a direct function invocation, but
8438 indirect ``call``'s are just as possible, calling an arbitrary pointer
8440 #. '``function args``': argument list whose types match the function
8441 signature argument types and parameter attributes. All arguments must
8442 be of :ref:`first class <t_firstclass>` type. If the function signature
8443 indicates the function accepts a variable number of arguments, the
8444 extra arguments can be specified.
8445 #. The optional :ref:`function attributes <fnattrs>` list. Only
8446 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
8447 attributes are valid here.
8448 #. The optional :ref:`operand bundles <opbundles>` list.
8453 The '``call``' instruction is used to cause control flow to transfer to
8454 a specified function, with its incoming arguments bound to the specified
8455 values. Upon a '``ret``' instruction in the called function, control
8456 flow continues with the instruction after the function call, and the
8457 return value of the function is bound to the result argument.
8462 .. code-block:: llvm
8464 %retval = call i32 @test(i32 %argc)
8465 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
8466 %X = tail call i32 @foo() ; yields i32
8467 %Y = tail call fastcc i32 @foo() ; yields i32
8468 call void %foo(i8 97 signext)
8470 %struct.A = type { i32, i8 }
8471 %r = call %struct.A @foo() ; yields { i32, i8 }
8472 %gr = extractvalue %struct.A %r, 0 ; yields i32
8473 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
8474 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
8475 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
8477 llvm treats calls to some functions with names and arguments that match
8478 the standard C99 library as being the C99 library functions, and may
8479 perform optimizations or generate code for them under that assumption.
8480 This is something we'd like to change in the future to provide better
8481 support for freestanding environments and non-C-based languages.
8485 '``va_arg``' Instruction
8486 ^^^^^^^^^^^^^^^^^^^^^^^^
8493 <resultval> = va_arg <va_list*> <arglist>, <argty>
8498 The '``va_arg``' instruction is used to access arguments passed through
8499 the "variable argument" area of a function call. It is used to implement
8500 the ``va_arg`` macro in C.
8505 This instruction takes a ``va_list*`` value and the type of the
8506 argument. It returns a value of the specified argument type and
8507 increments the ``va_list`` to point to the next argument. The actual
8508 type of ``va_list`` is target specific.
8513 The '``va_arg``' instruction loads an argument of the specified type
8514 from the specified ``va_list`` and causes the ``va_list`` to point to
8515 the next argument. For more information, see the variable argument
8516 handling :ref:`Intrinsic Functions <int_varargs>`.
8518 It is legal for this instruction to be called in a function which does
8519 not take a variable number of arguments, for example, the ``vfprintf``
8522 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
8523 function <intrinsics>` because it takes a type as an argument.
8528 See the :ref:`variable argument processing <int_varargs>` section.
8530 Note that the code generator does not yet fully support va\_arg on many
8531 targets. Also, it does not currently support va\_arg with aggregate
8532 types on any target.
8536 '``landingpad``' Instruction
8537 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8544 <resultval> = landingpad <resultty> <clause>+
8545 <resultval> = landingpad <resultty> cleanup <clause>*
8547 <clause> := catch <type> <value>
8548 <clause> := filter <array constant type> <array constant>
8553 The '``landingpad``' instruction is used by `LLVM's exception handling
8554 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8555 is a landing pad --- one where the exception lands, and corresponds to the
8556 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
8557 defines values supplied by the :ref:`personality function <personalityfn>` upon
8558 re-entry to the function. The ``resultval`` has the type ``resultty``.
8564 ``cleanup`` flag indicates that the landing pad block is a cleanup.
8566 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
8567 contains the global variable representing the "type" that may be caught
8568 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
8569 clause takes an array constant as its argument. Use
8570 "``[0 x i8**] undef``" for a filter which cannot throw. The
8571 '``landingpad``' instruction must contain *at least* one ``clause`` or
8572 the ``cleanup`` flag.
8577 The '``landingpad``' instruction defines the values which are set by the
8578 :ref:`personality function <personalityfn>` upon re-entry to the function, and
8579 therefore the "result type" of the ``landingpad`` instruction. As with
8580 calling conventions, how the personality function results are
8581 represented in LLVM IR is target specific.
8583 The clauses are applied in order from top to bottom. If two
8584 ``landingpad`` instructions are merged together through inlining, the
8585 clauses from the calling function are appended to the list of clauses.
8586 When the call stack is being unwound due to an exception being thrown,
8587 the exception is compared against each ``clause`` in turn. If it doesn't
8588 match any of the clauses, and the ``cleanup`` flag is not set, then
8589 unwinding continues further up the call stack.
8591 The ``landingpad`` instruction has several restrictions:
8593 - A landing pad block is a basic block which is the unwind destination
8594 of an '``invoke``' instruction.
8595 - A landing pad block must have a '``landingpad``' instruction as its
8596 first non-PHI instruction.
8597 - There can be only one '``landingpad``' instruction within the landing
8599 - A basic block that is not a landing pad block may not include a
8600 '``landingpad``' instruction.
8605 .. code-block:: llvm
8607 ;; A landing pad which can catch an integer.
8608 %res = landingpad { i8*, i32 }
8610 ;; A landing pad that is a cleanup.
8611 %res = landingpad { i8*, i32 }
8613 ;; A landing pad which can catch an integer and can only throw a double.
8614 %res = landingpad { i8*, i32 }
8616 filter [1 x i8**] [@_ZTId]
8620 '``cleanuppad``' Instruction
8621 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8628 <resultval> = cleanuppad [<args>*]
8633 The '``cleanuppad``' instruction is used by `LLVM's exception handling
8634 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8635 is a cleanup block --- one where a personality routine attempts to
8636 transfer control to run cleanup actions.
8637 The ``args`` correspond to whatever additional
8638 information the :ref:`personality function <personalityfn>` requires to
8639 execute the cleanup.
8640 The ``resultval`` has the type :ref:`token <t_token>` and is used to
8641 match the ``cleanuppad`` to corresponding :ref:`cleanuprets <i_cleanupret>`
8642 and :ref:`cleanupendpads <i_cleanupendpad>`.
8647 The instruction takes a list of arbitrary values which are interpreted
8648 by the :ref:`personality function <personalityfn>`.
8653 When the call stack is being unwound due to an exception being thrown,
8654 the :ref:`personality function <personalityfn>` transfers control to the
8655 ``cleanuppad`` with the aid of the personality-specific arguments.
8656 As with calling conventions, how the personality function results are
8657 represented in LLVM IR is target specific.
8659 The ``cleanuppad`` instruction has several restrictions:
8661 - A cleanup block is a basic block which is the unwind destination of
8662 an exceptional instruction.
8663 - A cleanup block must have a '``cleanuppad``' instruction as its
8664 first non-PHI instruction.
8665 - There can be only one '``cleanuppad``' instruction within the
8667 - A basic block that is not a cleanup block may not include a
8668 '``cleanuppad``' instruction.
8669 - All '``cleanupret``'s and '``cleanupendpad``'s which consume a ``cleanuppad``
8670 must have the same exceptional successor.
8671 - It is undefined behavior for control to transfer from a ``cleanuppad`` to a
8672 ``ret`` without first executing a ``cleanupret`` or ``cleanupendpad`` that
8673 consumes the ``cleanuppad``.
8674 - It is undefined behavior for control to transfer from a ``cleanuppad`` to
8675 itself without first executing a ``cleanupret`` or ``cleanupendpad`` that
8676 consumes the ``cleanuppad``.
8681 .. code-block:: llvm
8683 %tok = cleanuppad []
8690 LLVM supports the notion of an "intrinsic function". These functions
8691 have well known names and semantics and are required to follow certain
8692 restrictions. Overall, these intrinsics represent an extension mechanism
8693 for the LLVM language that does not require changing all of the
8694 transformations in LLVM when adding to the language (or the bitcode
8695 reader/writer, the parser, etc...).
8697 Intrinsic function names must all start with an "``llvm.``" prefix. This
8698 prefix is reserved in LLVM for intrinsic names; thus, function names may
8699 not begin with this prefix. Intrinsic functions must always be external
8700 functions: you cannot define the body of intrinsic functions. Intrinsic
8701 functions may only be used in call or invoke instructions: it is illegal
8702 to take the address of an intrinsic function. Additionally, because
8703 intrinsic functions are part of the LLVM language, it is required if any
8704 are added that they be documented here.
8706 Some intrinsic functions can be overloaded, i.e., the intrinsic
8707 represents a family of functions that perform the same operation but on
8708 different data types. Because LLVM can represent over 8 million
8709 different integer types, overloading is used commonly to allow an
8710 intrinsic function to operate on any integer type. One or more of the
8711 argument types or the result type can be overloaded to accept any
8712 integer type. Argument types may also be defined as exactly matching a
8713 previous argument's type or the result type. This allows an intrinsic
8714 function which accepts multiple arguments, but needs all of them to be
8715 of the same type, to only be overloaded with respect to a single
8716 argument or the result.
8718 Overloaded intrinsics will have the names of its overloaded argument
8719 types encoded into its function name, each preceded by a period. Only
8720 those types which are overloaded result in a name suffix. Arguments
8721 whose type is matched against another type do not. For example, the
8722 ``llvm.ctpop`` function can take an integer of any width and returns an
8723 integer of exactly the same integer width. This leads to a family of
8724 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
8725 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
8726 overloaded, and only one type suffix is required. Because the argument's
8727 type is matched against the return type, it does not require its own
8730 To learn how to add an intrinsic function, please see the `Extending
8731 LLVM Guide <ExtendingLLVM.html>`_.
8735 Variable Argument Handling Intrinsics
8736 -------------------------------------
8738 Variable argument support is defined in LLVM with the
8739 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
8740 functions. These functions are related to the similarly named macros
8741 defined in the ``<stdarg.h>`` header file.
8743 All of these functions operate on arguments that use a target-specific
8744 value type "``va_list``". The LLVM assembly language reference manual
8745 does not define what this type is, so all transformations should be
8746 prepared to handle these functions regardless of the type used.
8748 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
8749 variable argument handling intrinsic functions are used.
8751 .. code-block:: llvm
8753 ; This struct is different for every platform. For most platforms,
8754 ; it is merely an i8*.
8755 %struct.va_list = type { i8* }
8757 ; For Unix x86_64 platforms, va_list is the following struct:
8758 ; %struct.va_list = type { i32, i32, i8*, i8* }
8760 define i32 @test(i32 %X, ...) {
8761 ; Initialize variable argument processing
8762 %ap = alloca %struct.va_list
8763 %ap2 = bitcast %struct.va_list* %ap to i8*
8764 call void @llvm.va_start(i8* %ap2)
8766 ; Read a single integer argument
8767 %tmp = va_arg i8* %ap2, i32
8769 ; Demonstrate usage of llvm.va_copy and llvm.va_end
8771 %aq2 = bitcast i8** %aq to i8*
8772 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
8773 call void @llvm.va_end(i8* %aq2)
8775 ; Stop processing of arguments.
8776 call void @llvm.va_end(i8* %ap2)
8780 declare void @llvm.va_start(i8*)
8781 declare void @llvm.va_copy(i8*, i8*)
8782 declare void @llvm.va_end(i8*)
8786 '``llvm.va_start``' Intrinsic
8787 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8794 declare void @llvm.va_start(i8* <arglist>)
8799 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
8800 subsequent use by ``va_arg``.
8805 The argument is a pointer to a ``va_list`` element to initialize.
8810 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
8811 available in C. In a target-dependent way, it initializes the
8812 ``va_list`` element to which the argument points, so that the next call
8813 to ``va_arg`` will produce the first variable argument passed to the
8814 function. Unlike the C ``va_start`` macro, this intrinsic does not need
8815 to know the last argument of the function as the compiler can figure
8818 '``llvm.va_end``' Intrinsic
8819 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8826 declare void @llvm.va_end(i8* <arglist>)
8831 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
8832 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
8837 The argument is a pointer to a ``va_list`` to destroy.
8842 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
8843 available in C. In a target-dependent way, it destroys the ``va_list``
8844 element to which the argument points. Calls to
8845 :ref:`llvm.va_start <int_va_start>` and
8846 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
8851 '``llvm.va_copy``' Intrinsic
8852 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8859 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
8864 The '``llvm.va_copy``' intrinsic copies the current argument position
8865 from the source argument list to the destination argument list.
8870 The first argument is a pointer to a ``va_list`` element to initialize.
8871 The second argument is a pointer to a ``va_list`` element to copy from.
8876 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
8877 available in C. In a target-dependent way, it copies the source
8878 ``va_list`` element into the destination ``va_list`` element. This
8879 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
8880 arbitrarily complex and require, for example, memory allocation.
8882 Accurate Garbage Collection Intrinsics
8883 --------------------------------------
8885 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
8886 (GC) requires the frontend to generate code containing appropriate intrinsic
8887 calls and select an appropriate GC strategy which knows how to lower these
8888 intrinsics in a manner which is appropriate for the target collector.
8890 These intrinsics allow identification of :ref:`GC roots on the
8891 stack <int_gcroot>`, as well as garbage collector implementations that
8892 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
8893 Frontends for type-safe garbage collected languages should generate
8894 these intrinsics to make use of the LLVM garbage collectors. For more
8895 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
8897 Experimental Statepoint Intrinsics
8898 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8900 LLVM provides an second experimental set of intrinsics for describing garbage
8901 collection safepoints in compiled code. These intrinsics are an alternative
8902 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
8903 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
8904 differences in approach are covered in the `Garbage Collection with LLVM
8905 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
8906 described in :doc:`Statepoints`.
8910 '``llvm.gcroot``' Intrinsic
8911 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8918 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
8923 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
8924 the code generator, and allows some metadata to be associated with it.
8929 The first argument specifies the address of a stack object that contains
8930 the root pointer. The second pointer (which must be either a constant or
8931 a global value address) contains the meta-data to be associated with the
8937 At runtime, a call to this intrinsic stores a null pointer into the
8938 "ptrloc" location. At compile-time, the code generator generates
8939 information to allow the runtime to find the pointer at GC safe points.
8940 The '``llvm.gcroot``' intrinsic may only be used in a function which
8941 :ref:`specifies a GC algorithm <gc>`.
8945 '``llvm.gcread``' Intrinsic
8946 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8953 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
8958 The '``llvm.gcread``' intrinsic identifies reads of references from heap
8959 locations, allowing garbage collector implementations that require read
8965 The second argument is the address to read from, which should be an
8966 address allocated from the garbage collector. The first object is a
8967 pointer to the start of the referenced object, if needed by the language
8968 runtime (otherwise null).
8973 The '``llvm.gcread``' intrinsic has the same semantics as a load
8974 instruction, but may be replaced with substantially more complex code by
8975 the garbage collector runtime, as needed. The '``llvm.gcread``'
8976 intrinsic may only be used in a function which :ref:`specifies a GC
8981 '``llvm.gcwrite``' Intrinsic
8982 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8989 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
8994 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
8995 locations, allowing garbage collector implementations that require write
8996 barriers (such as generational or reference counting collectors).
9001 The first argument is the reference to store, the second is the start of
9002 the object to store it to, and the third is the address of the field of
9003 Obj to store to. If the runtime does not require a pointer to the
9004 object, Obj may be null.
9009 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
9010 instruction, but may be replaced with substantially more complex code by
9011 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
9012 intrinsic may only be used in a function which :ref:`specifies a GC
9015 Code Generator Intrinsics
9016 -------------------------
9018 These intrinsics are provided by LLVM to expose special features that
9019 may only be implemented with code generator support.
9021 '``llvm.returnaddress``' Intrinsic
9022 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9029 declare i8 *@llvm.returnaddress(i32 <level>)
9034 The '``llvm.returnaddress``' intrinsic attempts to compute a
9035 target-specific value indicating the return address of the current
9036 function or one of its callers.
9041 The argument to this intrinsic indicates which function to return the
9042 address for. Zero indicates the calling function, one indicates its
9043 caller, etc. The argument is **required** to be a constant integer
9049 The '``llvm.returnaddress``' intrinsic either returns a pointer
9050 indicating the return address of the specified call frame, or zero if it
9051 cannot be identified. The value returned by this intrinsic is likely to
9052 be incorrect or 0 for arguments other than zero, so it should only be
9053 used for debugging purposes.
9055 Note that calling this intrinsic does not prevent function inlining or
9056 other aggressive transformations, so the value returned may not be that
9057 of the obvious source-language caller.
9059 '``llvm.frameaddress``' Intrinsic
9060 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9067 declare i8* @llvm.frameaddress(i32 <level>)
9072 The '``llvm.frameaddress``' intrinsic attempts to return the
9073 target-specific frame pointer value for the specified stack frame.
9078 The argument to this intrinsic indicates which function to return the
9079 frame pointer for. Zero indicates the calling function, one indicates
9080 its caller, etc. The argument is **required** to be a constant integer
9086 The '``llvm.frameaddress``' intrinsic either returns a pointer
9087 indicating the frame address of the specified call frame, or zero if it
9088 cannot be identified. The value returned by this intrinsic is likely to
9089 be incorrect or 0 for arguments other than zero, so it should only be
9090 used for debugging purposes.
9092 Note that calling this intrinsic does not prevent function inlining or
9093 other aggressive transformations, so the value returned may not be that
9094 of the obvious source-language caller.
9096 '``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
9097 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9104 declare void @llvm.localescape(...)
9105 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
9110 The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
9111 allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
9112 live frame pointer to recover the address of the allocation. The offset is
9113 computed during frame layout of the caller of ``llvm.localescape``.
9118 All arguments to '``llvm.localescape``' must be pointers to static allocas or
9119 casts of static allocas. Each function can only call '``llvm.localescape``'
9120 once, and it can only do so from the entry block.
9122 The ``func`` argument to '``llvm.localrecover``' must be a constant
9123 bitcasted pointer to a function defined in the current module. The code
9124 generator cannot determine the frame allocation offset of functions defined in
9127 The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
9128 call frame that is currently live. The return value of '``llvm.localaddress``'
9129 is one way to produce such a value, but various runtimes also expose a suitable
9130 pointer in platform-specific ways.
9132 The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
9133 '``llvm.localescape``' to recover. It is zero-indexed.
9138 These intrinsics allow a group of functions to share access to a set of local
9139 stack allocations of a one parent function. The parent function may call the
9140 '``llvm.localescape``' intrinsic once from the function entry block, and the
9141 child functions can use '``llvm.localrecover``' to access the escaped allocas.
9142 The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
9143 the escaped allocas are allocated, which would break attempts to use
9144 '``llvm.localrecover``'.
9146 .. _int_read_register:
9147 .. _int_write_register:
9149 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
9150 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9157 declare i32 @llvm.read_register.i32(metadata)
9158 declare i64 @llvm.read_register.i64(metadata)
9159 declare void @llvm.write_register.i32(metadata, i32 @value)
9160 declare void @llvm.write_register.i64(metadata, i64 @value)
9166 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
9167 provides access to the named register. The register must be valid on
9168 the architecture being compiled to. The type needs to be compatible
9169 with the register being read.
9174 The '``llvm.read_register``' intrinsic returns the current value of the
9175 register, where possible. The '``llvm.write_register``' intrinsic sets
9176 the current value of the register, where possible.
9178 This is useful to implement named register global variables that need
9179 to always be mapped to a specific register, as is common practice on
9180 bare-metal programs including OS kernels.
9182 The compiler doesn't check for register availability or use of the used
9183 register in surrounding code, including inline assembly. Because of that,
9184 allocatable registers are not supported.
9186 Warning: So far it only works with the stack pointer on selected
9187 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
9188 work is needed to support other registers and even more so, allocatable
9193 '``llvm.stacksave``' Intrinsic
9194 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9201 declare i8* @llvm.stacksave()
9206 The '``llvm.stacksave``' intrinsic is used to remember the current state
9207 of the function stack, for use with
9208 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
9209 implementing language features like scoped automatic variable sized
9215 This intrinsic returns a opaque pointer value that can be passed to
9216 :ref:`llvm.stackrestore <int_stackrestore>`. When an
9217 ``llvm.stackrestore`` intrinsic is executed with a value saved from
9218 ``llvm.stacksave``, it effectively restores the state of the stack to
9219 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
9220 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
9221 were allocated after the ``llvm.stacksave`` was executed.
9223 .. _int_stackrestore:
9225 '``llvm.stackrestore``' Intrinsic
9226 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9233 declare void @llvm.stackrestore(i8* %ptr)
9238 The '``llvm.stackrestore``' intrinsic is used to restore the state of
9239 the function stack to the state it was in when the corresponding
9240 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
9241 useful for implementing language features like scoped automatic variable
9242 sized arrays in C99.
9247 See the description for :ref:`llvm.stacksave <int_stacksave>`.
9249 '``llvm.prefetch``' Intrinsic
9250 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9257 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
9262 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
9263 insert a prefetch instruction if supported; otherwise, it is a noop.
9264 Prefetches have no effect on the behavior of the program but can change
9265 its performance characteristics.
9270 ``address`` is the address to be prefetched, ``rw`` is the specifier
9271 determining if the fetch should be for a read (0) or write (1), and
9272 ``locality`` is a temporal locality specifier ranging from (0) - no
9273 locality, to (3) - extremely local keep in cache. The ``cache type``
9274 specifies whether the prefetch is performed on the data (1) or
9275 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
9276 arguments must be constant integers.
9281 This intrinsic does not modify the behavior of the program. In
9282 particular, prefetches cannot trap and do not produce a value. On
9283 targets that support this intrinsic, the prefetch can provide hints to
9284 the processor cache for better performance.
9286 '``llvm.pcmarker``' Intrinsic
9287 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9294 declare void @llvm.pcmarker(i32 <id>)
9299 The '``llvm.pcmarker``' intrinsic is a method to export a Program
9300 Counter (PC) in a region of code to simulators and other tools. The
9301 method is target specific, but it is expected that the marker will use
9302 exported symbols to transmit the PC of the marker. The marker makes no
9303 guarantees that it will remain with any specific instruction after
9304 optimizations. It is possible that the presence of a marker will inhibit
9305 optimizations. The intended use is to be inserted after optimizations to
9306 allow correlations of simulation runs.
9311 ``id`` is a numerical id identifying the marker.
9316 This intrinsic does not modify the behavior of the program. Backends
9317 that do not support this intrinsic may ignore it.
9319 '``llvm.readcyclecounter``' Intrinsic
9320 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9327 declare i64 @llvm.readcyclecounter()
9332 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
9333 counter register (or similar low latency, high accuracy clocks) on those
9334 targets that support it. On X86, it should map to RDTSC. On Alpha, it
9335 should map to RPCC. As the backing counters overflow quickly (on the
9336 order of 9 seconds on alpha), this should only be used for small
9342 When directly supported, reading the cycle counter should not modify any
9343 memory. Implementations are allowed to either return a application
9344 specific value or a system wide value. On backends without support, this
9345 is lowered to a constant 0.
9347 Note that runtime support may be conditional on the privilege-level code is
9348 running at and the host platform.
9350 '``llvm.clear_cache``' Intrinsic
9351 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9358 declare void @llvm.clear_cache(i8*, i8*)
9363 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
9364 in the specified range to the execution unit of the processor. On
9365 targets with non-unified instruction and data cache, the implementation
9366 flushes the instruction cache.
9371 On platforms with coherent instruction and data caches (e.g. x86), this
9372 intrinsic is a nop. On platforms with non-coherent instruction and data
9373 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
9374 instructions or a system call, if cache flushing requires special
9377 The default behavior is to emit a call to ``__clear_cache`` from the run
9380 This instrinsic does *not* empty the instruction pipeline. Modifications
9381 of the current function are outside the scope of the intrinsic.
9383 '``llvm.instrprof_increment``' Intrinsic
9384 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9391 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
9392 i32 <num-counters>, i32 <index>)
9397 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
9398 frontend for use with instrumentation based profiling. These will be
9399 lowered by the ``-instrprof`` pass to generate execution counts of a
9405 The first argument is a pointer to a global variable containing the
9406 name of the entity being instrumented. This should generally be the
9407 (mangled) function name for a set of counters.
9409 The second argument is a hash value that can be used by the consumer
9410 of the profile data to detect changes to the instrumented source, and
9411 the third is the number of counters associated with ``name``. It is an
9412 error if ``hash`` or ``num-counters`` differ between two instances of
9413 ``instrprof_increment`` that refer to the same name.
9415 The last argument refers to which of the counters for ``name`` should
9416 be incremented. It should be a value between 0 and ``num-counters``.
9421 This intrinsic represents an increment of a profiling counter. It will
9422 cause the ``-instrprof`` pass to generate the appropriate data
9423 structures and the code to increment the appropriate value, in a
9424 format that can be written out by a compiler runtime and consumed via
9425 the ``llvm-profdata`` tool.
9427 Standard C Library Intrinsics
9428 -----------------------------
9430 LLVM provides intrinsics for a few important standard C library
9431 functions. These intrinsics allow source-language front-ends to pass
9432 information about the alignment of the pointer arguments to the code
9433 generator, providing opportunity for more efficient code generation.
9437 '``llvm.memcpy``' Intrinsic
9438 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9443 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
9444 integer bit width and for different address spaces. Not all targets
9445 support all bit widths however.
9449 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9450 i32 <len>, i32 <align>, i1 <isvolatile>)
9451 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9452 i64 <len>, i32 <align>, i1 <isvolatile>)
9457 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9458 source location to the destination location.
9460 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
9461 intrinsics do not return a value, takes extra alignment/isvolatile
9462 arguments and the pointers can be in specified address spaces.
9467 The first argument is a pointer to the destination, the second is a
9468 pointer to the source. The third argument is an integer argument
9469 specifying the number of bytes to copy, the fourth argument is the
9470 alignment of the source and destination locations, and the fifth is a
9471 boolean indicating a volatile access.
9473 If the call to this intrinsic has an alignment value that is not 0 or 1,
9474 then the caller guarantees that both the source and destination pointers
9475 are aligned to that boundary.
9477 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
9478 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9479 very cleanly specified and it is unwise to depend on it.
9484 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9485 source location to the destination location, which are not allowed to
9486 overlap. It copies "len" bytes of memory over. If the argument is known
9487 to be aligned to some boundary, this can be specified as the fourth
9488 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
9490 '``llvm.memmove``' Intrinsic
9491 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9496 This is an overloaded intrinsic. You can use llvm.memmove on any integer
9497 bit width and for different address space. Not all targets support all
9502 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9503 i32 <len>, i32 <align>, i1 <isvolatile>)
9504 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9505 i64 <len>, i32 <align>, i1 <isvolatile>)
9510 The '``llvm.memmove.*``' intrinsics move a block of memory from the
9511 source location to the destination location. It is similar to the
9512 '``llvm.memcpy``' intrinsic but allows the two memory locations to
9515 Note that, unlike the standard libc function, the ``llvm.memmove.*``
9516 intrinsics do not return a value, takes extra alignment/isvolatile
9517 arguments and the pointers can be in specified address spaces.
9522 The first argument is a pointer to the destination, the second is a
9523 pointer to the source. The third argument is an integer argument
9524 specifying the number of bytes to copy, the fourth argument is the
9525 alignment of the source and destination locations, and the fifth is a
9526 boolean indicating a volatile access.
9528 If the call to this intrinsic has an alignment value that is not 0 or 1,
9529 then the caller guarantees that the source and destination pointers are
9530 aligned to that boundary.
9532 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
9533 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
9534 not very cleanly specified and it is unwise to depend on it.
9539 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
9540 source location to the destination location, which may overlap. It
9541 copies "len" bytes of memory over. If the argument is known to be
9542 aligned to some boundary, this can be specified as the fourth argument,
9543 otherwise it should be set to 0 or 1 (both meaning no alignment).
9545 '``llvm.memset.*``' Intrinsics
9546 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9551 This is an overloaded intrinsic. You can use llvm.memset on any integer
9552 bit width and for different address spaces. However, not all targets
9553 support all bit widths.
9557 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
9558 i32 <len>, i32 <align>, i1 <isvolatile>)
9559 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
9560 i64 <len>, i32 <align>, i1 <isvolatile>)
9565 The '``llvm.memset.*``' intrinsics fill a block of memory with a
9566 particular byte value.
9568 Note that, unlike the standard libc function, the ``llvm.memset``
9569 intrinsic does not return a value and takes extra alignment/volatile
9570 arguments. Also, the destination can be in an arbitrary address space.
9575 The first argument is a pointer to the destination to fill, the second
9576 is the byte value with which to fill it, the third argument is an
9577 integer argument specifying the number of bytes to fill, and the fourth
9578 argument is the known alignment of the destination location.
9580 If the call to this intrinsic has an alignment value that is not 0 or 1,
9581 then the caller guarantees that the destination pointer is aligned to
9584 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
9585 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9586 very cleanly specified and it is unwise to depend on it.
9591 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
9592 at the destination location. If the argument is known to be aligned to
9593 some boundary, this can be specified as the fourth argument, otherwise
9594 it should be set to 0 or 1 (both meaning no alignment).
9596 '``llvm.sqrt.*``' Intrinsic
9597 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9602 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
9603 floating point or vector of floating point type. Not all targets support
9608 declare float @llvm.sqrt.f32(float %Val)
9609 declare double @llvm.sqrt.f64(double %Val)
9610 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
9611 declare fp128 @llvm.sqrt.f128(fp128 %Val)
9612 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
9617 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
9618 returning the same value as the libm '``sqrt``' functions would. Unlike
9619 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
9620 negative numbers other than -0.0 (which allows for better optimization,
9621 because there is no need to worry about errno being set).
9622 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
9627 The argument and return value are floating point numbers of the same
9633 This function returns the sqrt of the specified operand if it is a
9634 nonnegative floating point number.
9636 '``llvm.powi.*``' Intrinsic
9637 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9642 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
9643 floating point or vector of floating point type. Not all targets support
9648 declare float @llvm.powi.f32(float %Val, i32 %power)
9649 declare double @llvm.powi.f64(double %Val, i32 %power)
9650 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
9651 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
9652 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
9657 The '``llvm.powi.*``' intrinsics return the first operand raised to the
9658 specified (positive or negative) power. The order of evaluation of
9659 multiplications is not defined. When a vector of floating point type is
9660 used, the second argument remains a scalar integer value.
9665 The second argument is an integer power, and the first is a value to
9666 raise to that power.
9671 This function returns the first value raised to the second power with an
9672 unspecified sequence of rounding operations.
9674 '``llvm.sin.*``' Intrinsic
9675 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9680 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
9681 floating point or vector of floating point type. Not all targets support
9686 declare float @llvm.sin.f32(float %Val)
9687 declare double @llvm.sin.f64(double %Val)
9688 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
9689 declare fp128 @llvm.sin.f128(fp128 %Val)
9690 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
9695 The '``llvm.sin.*``' intrinsics return the sine of the operand.
9700 The argument and return value are floating point numbers of the same
9706 This function returns the sine of the specified operand, returning the
9707 same values as the libm ``sin`` functions would, and handles error
9708 conditions in the same way.
9710 '``llvm.cos.*``' Intrinsic
9711 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9716 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
9717 floating point or vector of floating point type. Not all targets support
9722 declare float @llvm.cos.f32(float %Val)
9723 declare double @llvm.cos.f64(double %Val)
9724 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
9725 declare fp128 @llvm.cos.f128(fp128 %Val)
9726 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
9731 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
9736 The argument and return value are floating point numbers of the same
9742 This function returns the cosine of the specified operand, returning the
9743 same values as the libm ``cos`` functions would, and handles error
9744 conditions in the same way.
9746 '``llvm.pow.*``' Intrinsic
9747 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9752 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
9753 floating point or vector of floating point type. Not all targets support
9758 declare float @llvm.pow.f32(float %Val, float %Power)
9759 declare double @llvm.pow.f64(double %Val, double %Power)
9760 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
9761 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
9762 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
9767 The '``llvm.pow.*``' intrinsics return the first operand raised to the
9768 specified (positive or negative) power.
9773 The second argument is a floating point power, and the first is a value
9774 to raise to that power.
9779 This function returns the first value raised to the second power,
9780 returning the same values as the libm ``pow`` functions would, and
9781 handles error conditions in the same way.
9783 '``llvm.exp.*``' Intrinsic
9784 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9789 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
9790 floating point or vector of floating point type. Not all targets support
9795 declare float @llvm.exp.f32(float %Val)
9796 declare double @llvm.exp.f64(double %Val)
9797 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
9798 declare fp128 @llvm.exp.f128(fp128 %Val)
9799 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
9804 The '``llvm.exp.*``' intrinsics perform the exp function.
9809 The argument and return value are floating point numbers of the same
9815 This function returns the same values as the libm ``exp`` functions
9816 would, and handles error conditions in the same way.
9818 '``llvm.exp2.*``' Intrinsic
9819 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9824 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
9825 floating point or vector of floating point type. Not all targets support
9830 declare float @llvm.exp2.f32(float %Val)
9831 declare double @llvm.exp2.f64(double %Val)
9832 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
9833 declare fp128 @llvm.exp2.f128(fp128 %Val)
9834 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
9839 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
9844 The argument and return value are floating point numbers of the same
9850 This function returns the same values as the libm ``exp2`` functions
9851 would, and handles error conditions in the same way.
9853 '``llvm.log.*``' Intrinsic
9854 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9859 This is an overloaded intrinsic. You can use ``llvm.log`` on any
9860 floating point or vector of floating point type. Not all targets support
9865 declare float @llvm.log.f32(float %Val)
9866 declare double @llvm.log.f64(double %Val)
9867 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
9868 declare fp128 @llvm.log.f128(fp128 %Val)
9869 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
9874 The '``llvm.log.*``' intrinsics perform the log function.
9879 The argument and return value are floating point numbers of the same
9885 This function returns the same values as the libm ``log`` functions
9886 would, and handles error conditions in the same way.
9888 '``llvm.log10.*``' Intrinsic
9889 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9894 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
9895 floating point or vector of floating point type. Not all targets support
9900 declare float @llvm.log10.f32(float %Val)
9901 declare double @llvm.log10.f64(double %Val)
9902 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
9903 declare fp128 @llvm.log10.f128(fp128 %Val)
9904 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
9909 The '``llvm.log10.*``' intrinsics perform the log10 function.
9914 The argument and return value are floating point numbers of the same
9920 This function returns the same values as the libm ``log10`` functions
9921 would, and handles error conditions in the same way.
9923 '``llvm.log2.*``' Intrinsic
9924 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9929 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
9930 floating point or vector of floating point type. Not all targets support
9935 declare float @llvm.log2.f32(float %Val)
9936 declare double @llvm.log2.f64(double %Val)
9937 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
9938 declare fp128 @llvm.log2.f128(fp128 %Val)
9939 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
9944 The '``llvm.log2.*``' intrinsics perform the log2 function.
9949 The argument and return value are floating point numbers of the same
9955 This function returns the same values as the libm ``log2`` functions
9956 would, and handles error conditions in the same way.
9958 '``llvm.fma.*``' Intrinsic
9959 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9964 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
9965 floating point or vector of floating point type. Not all targets support
9970 declare float @llvm.fma.f32(float %a, float %b, float %c)
9971 declare double @llvm.fma.f64(double %a, double %b, double %c)
9972 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
9973 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
9974 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
9979 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
9985 The argument and return value are floating point numbers of the same
9991 This function returns the same values as the libm ``fma`` functions
9992 would, and does not set errno.
9994 '``llvm.fabs.*``' Intrinsic
9995 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10000 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
10001 floating point or vector of floating point type. Not all targets support
10006 declare float @llvm.fabs.f32(float %Val)
10007 declare double @llvm.fabs.f64(double %Val)
10008 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
10009 declare fp128 @llvm.fabs.f128(fp128 %Val)
10010 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
10015 The '``llvm.fabs.*``' intrinsics return the absolute value of the
10021 The argument and return value are floating point numbers of the same
10027 This function returns the same values as the libm ``fabs`` functions
10028 would, and handles error conditions in the same way.
10030 '``llvm.minnum.*``' Intrinsic
10031 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10036 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
10037 floating point or vector of floating point type. Not all targets support
10042 declare float @llvm.minnum.f32(float %Val0, float %Val1)
10043 declare double @llvm.minnum.f64(double %Val0, double %Val1)
10044 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10045 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
10046 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10051 The '``llvm.minnum.*``' intrinsics return the minimum of the two
10058 The arguments and return value are floating point numbers of the same
10064 Follows the IEEE-754 semantics for minNum, which also match for libm's
10067 If either operand is a NaN, returns the other non-NaN operand. Returns
10068 NaN only if both operands are NaN. If the operands compare equal,
10069 returns a value that compares equal to both operands. This means that
10070 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10072 '``llvm.maxnum.*``' Intrinsic
10073 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10078 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
10079 floating point or vector of floating point type. Not all targets support
10084 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
10085 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
10086 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10087 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
10088 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10093 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
10100 The arguments and return value are floating point numbers of the same
10105 Follows the IEEE-754 semantics for maxNum, which also match for libm's
10108 If either operand is a NaN, returns the other non-NaN operand. Returns
10109 NaN only if both operands are NaN. If the operands compare equal,
10110 returns a value that compares equal to both operands. This means that
10111 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10113 '``llvm.copysign.*``' Intrinsic
10114 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10119 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
10120 floating point or vector of floating point type. Not all targets support
10125 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
10126 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
10127 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
10128 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
10129 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
10134 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
10135 first operand and the sign of the second operand.
10140 The arguments and return value are floating point numbers of the same
10146 This function returns the same values as the libm ``copysign``
10147 functions would, and handles error conditions in the same way.
10149 '``llvm.floor.*``' Intrinsic
10150 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10155 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
10156 floating point or vector of floating point type. Not all targets support
10161 declare float @llvm.floor.f32(float %Val)
10162 declare double @llvm.floor.f64(double %Val)
10163 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
10164 declare fp128 @llvm.floor.f128(fp128 %Val)
10165 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
10170 The '``llvm.floor.*``' intrinsics return the floor of the operand.
10175 The argument and return value are floating point numbers of the same
10181 This function returns the same values as the libm ``floor`` functions
10182 would, and handles error conditions in the same way.
10184 '``llvm.ceil.*``' Intrinsic
10185 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10190 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
10191 floating point or vector of floating point type. Not all targets support
10196 declare float @llvm.ceil.f32(float %Val)
10197 declare double @llvm.ceil.f64(double %Val)
10198 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
10199 declare fp128 @llvm.ceil.f128(fp128 %Val)
10200 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
10205 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
10210 The argument and return value are floating point numbers of the same
10216 This function returns the same values as the libm ``ceil`` functions
10217 would, and handles error conditions in the same way.
10219 '``llvm.trunc.*``' Intrinsic
10220 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10225 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
10226 floating point or vector of floating point type. Not all targets support
10231 declare float @llvm.trunc.f32(float %Val)
10232 declare double @llvm.trunc.f64(double %Val)
10233 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
10234 declare fp128 @llvm.trunc.f128(fp128 %Val)
10235 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
10240 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
10241 nearest integer not larger in magnitude than the operand.
10246 The argument and return value are floating point numbers of the same
10252 This function returns the same values as the libm ``trunc`` functions
10253 would, and handles error conditions in the same way.
10255 '``llvm.rint.*``' Intrinsic
10256 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10261 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
10262 floating point or vector of floating point type. Not all targets support
10267 declare float @llvm.rint.f32(float %Val)
10268 declare double @llvm.rint.f64(double %Val)
10269 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
10270 declare fp128 @llvm.rint.f128(fp128 %Val)
10271 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
10276 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
10277 nearest integer. It may raise an inexact floating-point exception if the
10278 operand isn't an integer.
10283 The argument and return value are floating point numbers of the same
10289 This function returns the same values as the libm ``rint`` functions
10290 would, and handles error conditions in the same way.
10292 '``llvm.nearbyint.*``' Intrinsic
10293 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10298 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
10299 floating point or vector of floating point type. Not all targets support
10304 declare float @llvm.nearbyint.f32(float %Val)
10305 declare double @llvm.nearbyint.f64(double %Val)
10306 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
10307 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
10308 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
10313 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
10319 The argument and return value are floating point numbers of the same
10325 This function returns the same values as the libm ``nearbyint``
10326 functions would, and handles error conditions in the same way.
10328 '``llvm.round.*``' Intrinsic
10329 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10334 This is an overloaded intrinsic. You can use ``llvm.round`` on any
10335 floating point or vector of floating point type. Not all targets support
10340 declare float @llvm.round.f32(float %Val)
10341 declare double @llvm.round.f64(double %Val)
10342 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
10343 declare fp128 @llvm.round.f128(fp128 %Val)
10344 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
10349 The '``llvm.round.*``' intrinsics returns the operand rounded to the
10355 The argument and return value are floating point numbers of the same
10361 This function returns the same values as the libm ``round``
10362 functions would, and handles error conditions in the same way.
10364 Bit Manipulation Intrinsics
10365 ---------------------------
10367 LLVM provides intrinsics for a few important bit manipulation
10368 operations. These allow efficient code generation for some algorithms.
10370 '``llvm.bswap.*``' Intrinsics
10371 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10376 This is an overloaded intrinsic function. You can use bswap on any
10377 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
10381 declare i16 @llvm.bswap.i16(i16 <id>)
10382 declare i32 @llvm.bswap.i32(i32 <id>)
10383 declare i64 @llvm.bswap.i64(i64 <id>)
10388 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
10389 values with an even number of bytes (positive multiple of 16 bits).
10390 These are useful for performing operations on data that is not in the
10391 target's native byte order.
10396 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
10397 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
10398 intrinsic returns an i32 value that has the four bytes of the input i32
10399 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
10400 returned i32 will have its bytes in 3, 2, 1, 0 order. The
10401 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
10402 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
10405 '``llvm.ctpop.*``' Intrinsic
10406 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10411 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
10412 bit width, or on any vector with integer elements. Not all targets
10413 support all bit widths or vector types, however.
10417 declare i8 @llvm.ctpop.i8(i8 <src>)
10418 declare i16 @llvm.ctpop.i16(i16 <src>)
10419 declare i32 @llvm.ctpop.i32(i32 <src>)
10420 declare i64 @llvm.ctpop.i64(i64 <src>)
10421 declare i256 @llvm.ctpop.i256(i256 <src>)
10422 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
10427 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
10433 The only argument is the value to be counted. The argument may be of any
10434 integer type, or a vector with integer elements. The return type must
10435 match the argument type.
10440 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
10441 each element of a vector.
10443 '``llvm.ctlz.*``' Intrinsic
10444 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10449 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
10450 integer bit width, or any vector whose elements are integers. Not all
10451 targets support all bit widths or vector types, however.
10455 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
10456 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
10457 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
10458 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
10459 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
10460 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10465 The '``llvm.ctlz``' family of intrinsic functions counts the number of
10466 leading zeros in a variable.
10471 The first argument is the value to be counted. This argument may be of
10472 any integer type, or a vector with integer element type. The return
10473 type must match the first argument type.
10475 The second argument must be a constant and is a flag to indicate whether
10476 the intrinsic should ensure that a zero as the first argument produces a
10477 defined result. Historically some architectures did not provide a
10478 defined result for zero values as efficiently, and many algorithms are
10479 now predicated on avoiding zero-value inputs.
10484 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
10485 zeros in a variable, or within each element of the vector. If
10486 ``src == 0`` then the result is the size in bits of the type of ``src``
10487 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10488 ``llvm.ctlz(i32 2) = 30``.
10490 '``llvm.cttz.*``' Intrinsic
10491 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10496 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
10497 integer bit width, or any vector of integer elements. Not all targets
10498 support all bit widths or vector types, however.
10502 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
10503 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
10504 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
10505 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
10506 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
10507 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10512 The '``llvm.cttz``' family of intrinsic functions counts the number of
10518 The first argument is the value to be counted. This argument may be of
10519 any integer type, or a vector with integer element type. The return
10520 type must match the first argument type.
10522 The second argument must be a constant and is a flag to indicate whether
10523 the intrinsic should ensure that a zero as the first argument produces a
10524 defined result. Historically some architectures did not provide a
10525 defined result for zero values as efficiently, and many algorithms are
10526 now predicated on avoiding zero-value inputs.
10531 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
10532 zeros in a variable, or within each element of a vector. If ``src == 0``
10533 then the result is the size in bits of the type of ``src`` if
10534 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10535 ``llvm.cttz(2) = 1``.
10539 Arithmetic with Overflow Intrinsics
10540 -----------------------------------
10542 LLVM provides intrinsics for some arithmetic with overflow operations.
10544 '``llvm.sadd.with.overflow.*``' Intrinsics
10545 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10550 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
10551 on any integer bit width.
10555 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
10556 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10557 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
10562 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10563 a signed addition of the two arguments, and indicate whether an overflow
10564 occurred during the signed summation.
10569 The arguments (%a and %b) and the first element of the result structure
10570 may be of integer types of any bit width, but they must have the same
10571 bit width. The second element of the result structure must be of type
10572 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10578 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10579 a signed addition of the two variables. They return a structure --- the
10580 first element of which is the signed summation, and the second element
10581 of which is a bit specifying if the signed summation resulted in an
10587 .. code-block:: llvm
10589 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10590 %sum = extractvalue {i32, i1} %res, 0
10591 %obit = extractvalue {i32, i1} %res, 1
10592 br i1 %obit, label %overflow, label %normal
10594 '``llvm.uadd.with.overflow.*``' Intrinsics
10595 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10600 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
10601 on any integer bit width.
10605 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
10606 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10607 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
10612 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10613 an unsigned addition of the two arguments, and indicate whether a carry
10614 occurred during the unsigned summation.
10619 The arguments (%a and %b) and the first element of the result structure
10620 may be of integer types of any bit width, but they must have the same
10621 bit width. The second element of the result structure must be of type
10622 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10628 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10629 an unsigned addition of the two arguments. They return a structure --- the
10630 first element of which is the sum, and the second element of which is a
10631 bit specifying if the unsigned summation resulted in a carry.
10636 .. code-block:: llvm
10638 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10639 %sum = extractvalue {i32, i1} %res, 0
10640 %obit = extractvalue {i32, i1} %res, 1
10641 br i1 %obit, label %carry, label %normal
10643 '``llvm.ssub.with.overflow.*``' Intrinsics
10644 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10649 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
10650 on any integer bit width.
10654 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
10655 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10656 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
10661 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10662 a signed subtraction of the two arguments, and indicate whether an
10663 overflow occurred during the signed subtraction.
10668 The arguments (%a and %b) and the first element of the result structure
10669 may be of integer types of any bit width, but they must have the same
10670 bit width. The second element of the result structure must be of type
10671 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10677 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10678 a signed subtraction of the two arguments. They return a structure --- the
10679 first element of which is the subtraction, and the second element of
10680 which is a bit specifying if the signed subtraction resulted in an
10686 .. code-block:: llvm
10688 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10689 %sum = extractvalue {i32, i1} %res, 0
10690 %obit = extractvalue {i32, i1} %res, 1
10691 br i1 %obit, label %overflow, label %normal
10693 '``llvm.usub.with.overflow.*``' Intrinsics
10694 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10699 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
10700 on any integer bit width.
10704 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
10705 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10706 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
10711 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10712 an unsigned subtraction of the two arguments, and indicate whether an
10713 overflow occurred during the unsigned subtraction.
10718 The arguments (%a and %b) and the first element of the result structure
10719 may be of integer types of any bit width, but they must have the same
10720 bit width. The second element of the result structure must be of type
10721 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10727 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10728 an unsigned subtraction of the two arguments. They return a structure ---
10729 the first element of which is the subtraction, and the second element of
10730 which is a bit specifying if the unsigned subtraction resulted in an
10736 .. code-block:: llvm
10738 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10739 %sum = extractvalue {i32, i1} %res, 0
10740 %obit = extractvalue {i32, i1} %res, 1
10741 br i1 %obit, label %overflow, label %normal
10743 '``llvm.smul.with.overflow.*``' Intrinsics
10744 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10749 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
10750 on any integer bit width.
10754 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
10755 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10756 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
10761 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10762 a signed multiplication of the two arguments, and indicate whether an
10763 overflow occurred during the signed multiplication.
10768 The arguments (%a and %b) and the first element of the result structure
10769 may be of integer types of any bit width, but they must have the same
10770 bit width. The second element of the result structure must be of type
10771 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10777 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10778 a signed multiplication of the two arguments. They return a structure ---
10779 the first element of which is the multiplication, and the second element
10780 of which is a bit specifying if the signed multiplication resulted in an
10786 .. code-block:: llvm
10788 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10789 %sum = extractvalue {i32, i1} %res, 0
10790 %obit = extractvalue {i32, i1} %res, 1
10791 br i1 %obit, label %overflow, label %normal
10793 '``llvm.umul.with.overflow.*``' Intrinsics
10794 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10799 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
10800 on any integer bit width.
10804 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
10805 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10806 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
10811 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10812 a unsigned multiplication of the two arguments, and indicate whether an
10813 overflow occurred during the unsigned multiplication.
10818 The arguments (%a and %b) and the first element of the result structure
10819 may be of integer types of any bit width, but they must have the same
10820 bit width. The second element of the result structure must be of type
10821 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10827 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10828 an unsigned multiplication of the two arguments. They return a structure ---
10829 the first element of which is the multiplication, and the second
10830 element of which is a bit specifying if the unsigned multiplication
10831 resulted in an overflow.
10836 .. code-block:: llvm
10838 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10839 %sum = extractvalue {i32, i1} %res, 0
10840 %obit = extractvalue {i32, i1} %res, 1
10841 br i1 %obit, label %overflow, label %normal
10843 Specialised Arithmetic Intrinsics
10844 ---------------------------------
10846 '``llvm.canonicalize.*``' Intrinsic
10847 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10854 declare float @llvm.canonicalize.f32(float %a)
10855 declare double @llvm.canonicalize.f64(double %b)
10860 The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
10861 encoding of a floating point number. This canonicalization is useful for
10862 implementing certain numeric primitives such as frexp. The canonical encoding is
10863 defined by IEEE-754-2008 to be:
10867 2.1.8 canonical encoding: The preferred encoding of a floating-point
10868 representation in a format. Applied to declets, significands of finite
10869 numbers, infinities, and NaNs, especially in decimal formats.
10871 This operation can also be considered equivalent to the IEEE-754-2008
10872 conversion of a floating-point value to the same format. NaNs are handled
10873 according to section 6.2.
10875 Examples of non-canonical encodings:
10877 - x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
10878 converted to a canonical representation per hardware-specific protocol.
10879 - Many normal decimal floating point numbers have non-canonical alternative
10881 - Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
10882 These are treated as non-canonical encodings of zero and with be flushed to
10883 a zero of the same sign by this operation.
10885 Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
10886 default exception handling must signal an invalid exception, and produce a
10889 This function should always be implementable as multiplication by 1.0, provided
10890 that the compiler does not constant fold the operation. Likewise, division by
10891 1.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
10892 -0.0 is also sufficient provided that the rounding mode is not -Infinity.
10894 ``@llvm.canonicalize`` must preserve the equality relation. That is:
10896 - ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
10897 - ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
10900 Additionally, the sign of zero must be conserved:
10901 ``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
10903 The payload bits of a NaN must be conserved, with two exceptions.
10904 First, environments which use only a single canonical representation of NaN
10905 must perform said canonicalization. Second, SNaNs must be quieted per the
10908 The canonicalization operation may be optimized away if:
10910 - The input is known to be canonical. For example, it was produced by a
10911 floating-point operation that is required by the standard to be canonical.
10912 - The result is consumed only by (or fused with) other floating-point
10913 operations. That is, the bits of the floating point value are not examined.
10915 '``llvm.fmuladd.*``' Intrinsic
10916 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10923 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
10924 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
10929 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
10930 expressions that can be fused if the code generator determines that (a) the
10931 target instruction set has support for a fused operation, and (b) that the
10932 fused operation is more efficient than the equivalent, separate pair of mul
10933 and add instructions.
10938 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
10939 multiplicands, a and b, and an addend c.
10948 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
10950 is equivalent to the expression a \* b + c, except that rounding will
10951 not be performed between the multiplication and addition steps if the
10952 code generator fuses the operations. Fusion is not guaranteed, even if
10953 the target platform supports it. If a fused multiply-add is required the
10954 corresponding llvm.fma.\* intrinsic function should be used
10955 instead. This never sets errno, just as '``llvm.fma.*``'.
10960 .. code-block:: llvm
10962 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
10965 '``llvm.uabsdiff.*``' and '``llvm.sabsdiff.*``' Intrinsics
10966 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10970 This is an overloaded intrinsic. The loaded data is a vector of any integer bit width.
10972 .. code-block:: llvm
10974 declare <4 x integer> @llvm.uabsdiff.v4i32(<4 x integer> %a, <4 x integer> %b)
10980 The ``llvm.uabsdiff`` intrinsic returns a vector result of the absolute difference
10981 of the two operands, treating them both as unsigned integers. The intermediate
10982 calculations are computed using infinitely precise unsigned arithmetic. The final
10983 result will be truncated to the given type.
10985 The ``llvm.sabsdiff`` intrinsic returns a vector result of the absolute difference of
10986 the two operands, treating them both as signed integers. If the result overflows, the
10987 behavior is undefined.
10991 These intrinsics are primarily used during the code generation stage of compilation.
10992 They are generated by compiler passes such as the Loop and SLP vectorizers. It is not
10993 recommended for users to create them manually.
10998 Both intrinsics take two integer of the same bitwidth.
11005 call <4 x i32> @llvm.uabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11009 %1 = zext <4 x i32> %a to <4 x i64>
11010 %2 = zext <4 x i32> %b to <4 x i64>
11011 %sub = sub <4 x i64> %1, %2
11012 %trunc = trunc <4 x i64> to <4 x i32>
11014 and the expression::
11016 call <4 x i32> @llvm.sabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11020 %sub = sub nsw <4 x i32> %a, %b
11021 %ispos = icmp sge <4 x i32> %sub, zeroinitializer
11022 %neg = sub nsw <4 x i32> zeroinitializer, %sub
11023 %1 = select <4 x i1> %ispos, <4 x i32> %sub, <4 x i32> %neg
11026 Half Precision Floating Point Intrinsics
11027 ----------------------------------------
11029 For most target platforms, half precision floating point is a
11030 storage-only format. This means that it is a dense encoding (in memory)
11031 but does not support computation in the format.
11033 This means that code must first load the half-precision floating point
11034 value as an i16, then convert it to float with
11035 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
11036 then be performed on the float value (including extending to double
11037 etc). To store the value back to memory, it is first converted to float
11038 if needed, then converted to i16 with
11039 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
11042 .. _int_convert_to_fp16:
11044 '``llvm.convert.to.fp16``' Intrinsic
11045 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11052 declare i16 @llvm.convert.to.fp16.f32(float %a)
11053 declare i16 @llvm.convert.to.fp16.f64(double %a)
11058 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11059 conventional floating point type to half precision floating point format.
11064 The intrinsic function contains single argument - the value to be
11070 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11071 conventional floating point format to half precision floating point format. The
11072 return value is an ``i16`` which contains the converted number.
11077 .. code-block:: llvm
11079 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
11080 store i16 %res, i16* @x, align 2
11082 .. _int_convert_from_fp16:
11084 '``llvm.convert.from.fp16``' Intrinsic
11085 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11092 declare float @llvm.convert.from.fp16.f32(i16 %a)
11093 declare double @llvm.convert.from.fp16.f64(i16 %a)
11098 The '``llvm.convert.from.fp16``' intrinsic function performs a
11099 conversion from half precision floating point format to single precision
11100 floating point format.
11105 The intrinsic function contains single argument - the value to be
11111 The '``llvm.convert.from.fp16``' intrinsic function performs a
11112 conversion from half single precision floating point format to single
11113 precision floating point format. The input half-float value is
11114 represented by an ``i16`` value.
11119 .. code-block:: llvm
11121 %a = load i16, i16* @x, align 2
11122 %res = call float @llvm.convert.from.fp16(i16 %a)
11124 .. _dbg_intrinsics:
11126 Debugger Intrinsics
11127 -------------------
11129 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
11130 prefix), are described in the `LLVM Source Level
11131 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
11134 Exception Handling Intrinsics
11135 -----------------------------
11137 The LLVM exception handling intrinsics (which all start with
11138 ``llvm.eh.`` prefix), are described in the `LLVM Exception
11139 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
11141 .. _int_trampoline:
11143 Trampoline Intrinsics
11144 ---------------------
11146 These intrinsics make it possible to excise one parameter, marked with
11147 the :ref:`nest <nest>` attribute, from a function. The result is a
11148 callable function pointer lacking the nest parameter - the caller does
11149 not need to provide a value for it. Instead, the value to use is stored
11150 in advance in a "trampoline", a block of memory usually allocated on the
11151 stack, which also contains code to splice the nest value into the
11152 argument list. This is used to implement the GCC nested function address
11155 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
11156 then the resulting function pointer has signature ``i32 (i32, i32)*``.
11157 It can be created as follows:
11159 .. code-block:: llvm
11161 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
11162 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
11163 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
11164 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
11165 %fp = bitcast i8* %p to i32 (i32, i32)*
11167 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
11168 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
11172 '``llvm.init.trampoline``' Intrinsic
11173 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11180 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
11185 This fills the memory pointed to by ``tramp`` with executable code,
11186 turning it into a trampoline.
11191 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
11192 pointers. The ``tramp`` argument must point to a sufficiently large and
11193 sufficiently aligned block of memory; this memory is written to by the
11194 intrinsic. Note that the size and the alignment are target-specific -
11195 LLVM currently provides no portable way of determining them, so a
11196 front-end that generates this intrinsic needs to have some
11197 target-specific knowledge. The ``func`` argument must hold a function
11198 bitcast to an ``i8*``.
11203 The block of memory pointed to by ``tramp`` is filled with target
11204 dependent code, turning it into a function. Then ``tramp`` needs to be
11205 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
11206 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
11207 function's signature is the same as that of ``func`` with any arguments
11208 marked with the ``nest`` attribute removed. At most one such ``nest``
11209 argument is allowed, and it must be of pointer type. Calling the new
11210 function is equivalent to calling ``func`` with the same argument list,
11211 but with ``nval`` used for the missing ``nest`` argument. If, after
11212 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
11213 modified, then the effect of any later call to the returned function
11214 pointer is undefined.
11218 '``llvm.adjust.trampoline``' Intrinsic
11219 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11226 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
11231 This performs any required machine-specific adjustment to the address of
11232 a trampoline (passed as ``tramp``).
11237 ``tramp`` must point to a block of memory which already has trampoline
11238 code filled in by a previous call to
11239 :ref:`llvm.init.trampoline <int_it>`.
11244 On some architectures the address of the code to be executed needs to be
11245 different than the address where the trampoline is actually stored. This
11246 intrinsic returns the executable address corresponding to ``tramp``
11247 after performing the required machine specific adjustments. The pointer
11248 returned can then be :ref:`bitcast and executed <int_trampoline>`.
11250 .. _int_mload_mstore:
11252 Masked Vector Load and Store Intrinsics
11253 ---------------------------------------
11255 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.
11259 '``llvm.masked.load.*``' Intrinsics
11260 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11264 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
11268 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11269 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11274 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.
11280 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.
11286 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.
11287 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.
11292 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
11294 ;; The result of the two following instructions is identical aside from potential memory access exception
11295 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
11296 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
11300 '``llvm.masked.store.*``' Intrinsics
11301 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11305 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
11309 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
11310 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
11315 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.
11320 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.
11326 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.
11327 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.
11331 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
11333 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
11334 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
11335 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
11336 store <16 x float> %res, <16 x float>* %ptr, align 4
11339 Masked Vector Gather and Scatter Intrinsics
11340 -------------------------------------------
11342 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.
11346 '``llvm.masked.gather.*``' Intrinsics
11347 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11351 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.
11355 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11356 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11361 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.
11367 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.
11373 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.
11374 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.
11379 %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>)
11381 ;; The gather with all-true mask is equivalent to the following instruction sequence
11382 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
11383 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
11384 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
11385 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
11387 %val0 = load double, double* %ptr0, align 8
11388 %val1 = load double, double* %ptr1, align 8
11389 %val2 = load double, double* %ptr2, align 8
11390 %val3 = load double, double* %ptr3, align 8
11392 %vec0 = insertelement <4 x double>undef, %val0, 0
11393 %vec01 = insertelement <4 x double>%vec0, %val1, 1
11394 %vec012 = insertelement <4 x double>%vec01, %val2, 2
11395 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
11399 '``llvm.masked.scatter.*``' Intrinsics
11400 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11404 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.
11408 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
11409 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
11414 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.
11419 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.
11425 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 divergence. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar store operations.
11429 ;; This instruction unconditionaly stores data vector in multiple addresses
11430 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
11432 ;; It is equivalent to a list of scalar stores
11433 %val0 = extractelement <8 x i32> %value, i32 0
11434 %val1 = extractelement <8 x i32> %value, i32 1
11436 %val7 = extractelement <8 x i32> %value, i32 7
11437 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
11438 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
11440 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
11441 ;; Note: the order of the following stores is important when they overlap:
11442 store i32 %val0, i32* %ptr0, align 4
11443 store i32 %val1, i32* %ptr1, align 4
11445 store i32 %val7, i32* %ptr7, align 4
11451 This class of intrinsics provides information about the lifetime of
11452 memory objects and ranges where variables are immutable.
11456 '``llvm.lifetime.start``' Intrinsic
11457 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11464 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
11469 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
11475 The first argument is a constant integer representing the size of the
11476 object, or -1 if it is variable sized. The second argument is a pointer
11482 This intrinsic indicates that before this point in the code, the value
11483 of the memory pointed to by ``ptr`` is dead. This means that it is known
11484 to never be used and has an undefined value. A load from the pointer
11485 that precedes this intrinsic can be replaced with ``'undef'``.
11489 '``llvm.lifetime.end``' Intrinsic
11490 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11497 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
11502 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
11508 The first argument is a constant integer representing the size of the
11509 object, or -1 if it is variable sized. The second argument is a pointer
11515 This intrinsic indicates that after this point in the code, the value of
11516 the memory pointed to by ``ptr`` is dead. This means that it is known to
11517 never be used and has an undefined value. Any stores into the memory
11518 object following this intrinsic may be removed as dead.
11520 '``llvm.invariant.start``' Intrinsic
11521 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11528 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
11533 The '``llvm.invariant.start``' intrinsic specifies that the contents of
11534 a memory object will not change.
11539 The first argument is a constant integer representing the size of the
11540 object, or -1 if it is variable sized. The second argument is a pointer
11546 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
11547 the return value, the referenced memory location is constant and
11550 '``llvm.invariant.end``' Intrinsic
11551 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11558 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
11563 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
11564 memory object are mutable.
11569 The first argument is the matching ``llvm.invariant.start`` intrinsic.
11570 The second argument is a constant integer representing the size of the
11571 object, or -1 if it is variable sized and the third argument is a
11572 pointer to the object.
11577 This intrinsic indicates that the memory is mutable again.
11579 '``llvm.invariant.group.barrier``' Intrinsic
11580 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11587 declare i8* @llvm.invariant.group.barrier(i8* <ptr>)
11592 The '``llvm.invariant.group.barrier``' intrinsic can be used when an invariant
11593 established by invariant.group metadata no longer holds, to obtain a new pointer
11594 value that does not carry the invariant information.
11600 The ``llvm.invariant.group.barrier`` takes only one argument, which is
11601 the pointer to the memory for which the ``invariant.group`` no longer holds.
11606 Returns another pointer that aliases its argument but which is considered different
11607 for the purposes of ``load``/``store`` ``invariant.group`` metadata.
11612 This class of intrinsics is designed to be generic and has no specific
11615 '``llvm.var.annotation``' Intrinsic
11616 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11623 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11628 The '``llvm.var.annotation``' intrinsic.
11633 The first argument is a pointer to a value, the second is a pointer to a
11634 global string, the third is a pointer to a global string which is the
11635 source file name, and the last argument is the line number.
11640 This intrinsic allows annotation of local variables with arbitrary
11641 strings. This can be useful for special purpose optimizations that want
11642 to look for these annotations. These have no other defined use; they are
11643 ignored by code generation and optimization.
11645 '``llvm.ptr.annotation.*``' Intrinsic
11646 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11651 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
11652 pointer to an integer of any width. *NOTE* you must specify an address space for
11653 the pointer. The identifier for the default address space is the integer
11658 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11659 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
11660 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
11661 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
11662 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
11667 The '``llvm.ptr.annotation``' intrinsic.
11672 The first argument is a pointer to an integer value of arbitrary bitwidth
11673 (result of some expression), the second is a pointer to a global string, the
11674 third is a pointer to a global string which is the source file name, and the
11675 last argument is the line number. It returns the value of the first argument.
11680 This intrinsic allows annotation of a pointer to an integer with arbitrary
11681 strings. This can be useful for special purpose optimizations that want to look
11682 for these annotations. These have no other defined use; they are ignored by code
11683 generation and optimization.
11685 '``llvm.annotation.*``' Intrinsic
11686 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11691 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
11692 any integer bit width.
11696 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
11697 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
11698 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
11699 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
11700 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
11705 The '``llvm.annotation``' intrinsic.
11710 The first argument is an integer value (result of some expression), the
11711 second is a pointer to a global string, the third is a pointer to a
11712 global string which is the source file name, and the last argument is
11713 the line number. It returns the value of the first argument.
11718 This intrinsic allows annotations to be put on arbitrary expressions
11719 with arbitrary strings. This can be useful for special purpose
11720 optimizations that want to look for these annotations. These have no
11721 other defined use; they are ignored by code generation and optimization.
11723 '``llvm.trap``' Intrinsic
11724 ^^^^^^^^^^^^^^^^^^^^^^^^^
11731 declare void @llvm.trap() noreturn nounwind
11736 The '``llvm.trap``' intrinsic.
11746 This intrinsic is lowered to the target dependent trap instruction. If
11747 the target does not have a trap instruction, this intrinsic will be
11748 lowered to a call of the ``abort()`` function.
11750 '``llvm.debugtrap``' Intrinsic
11751 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11758 declare void @llvm.debugtrap() nounwind
11763 The '``llvm.debugtrap``' intrinsic.
11773 This intrinsic is lowered to code which is intended to cause an
11774 execution trap with the intention of requesting the attention of a
11777 '``llvm.stackprotector``' Intrinsic
11778 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11785 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
11790 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
11791 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
11792 is placed on the stack before local variables.
11797 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
11798 The first argument is the value loaded from the stack guard
11799 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
11800 enough space to hold the value of the guard.
11805 This intrinsic causes the prologue/epilogue inserter to force the position of
11806 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
11807 to ensure that if a local variable on the stack is overwritten, it will destroy
11808 the value of the guard. When the function exits, the guard on the stack is
11809 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
11810 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
11811 calling the ``__stack_chk_fail()`` function.
11813 '``llvm.stackprotectorcheck``' Intrinsic
11814 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11821 declare void @llvm.stackprotectorcheck(i8** <guard>)
11826 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
11827 created stack protector and if they are not equal calls the
11828 ``__stack_chk_fail()`` function.
11833 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
11834 the variable ``@__stack_chk_guard``.
11839 This intrinsic is provided to perform the stack protector check by comparing
11840 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
11841 values do not match call the ``__stack_chk_fail()`` function.
11843 The reason to provide this as an IR level intrinsic instead of implementing it
11844 via other IR operations is that in order to perform this operation at the IR
11845 level without an intrinsic, one would need to create additional basic blocks to
11846 handle the success/failure cases. This makes it difficult to stop the stack
11847 protector check from disrupting sibling tail calls in Codegen. With this
11848 intrinsic, we are able to generate the stack protector basic blocks late in
11849 codegen after the tail call decision has occurred.
11851 '``llvm.objectsize``' Intrinsic
11852 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11859 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
11860 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
11865 The ``llvm.objectsize`` intrinsic is designed to provide information to
11866 the optimizers to determine at compile time whether a) an operation
11867 (like memcpy) will overflow a buffer that corresponds to an object, or
11868 b) that a runtime check for overflow isn't necessary. An object in this
11869 context means an allocation of a specific class, structure, array, or
11875 The ``llvm.objectsize`` intrinsic takes two arguments. The first
11876 argument is a pointer to or into the ``object``. The second argument is
11877 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
11878 or -1 (if false) when the object size is unknown. The second argument
11879 only accepts constants.
11884 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
11885 the size of the object concerned. If the size cannot be determined at
11886 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
11887 on the ``min`` argument).
11889 '``llvm.expect``' Intrinsic
11890 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11895 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
11900 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
11901 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
11902 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
11907 The ``llvm.expect`` intrinsic provides information about expected (the
11908 most probable) value of ``val``, which can be used by optimizers.
11913 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
11914 a value. The second argument is an expected value, this needs to be a
11915 constant value, variables are not allowed.
11920 This intrinsic is lowered to the ``val``.
11924 '``llvm.assume``' Intrinsic
11925 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11932 declare void @llvm.assume(i1 %cond)
11937 The ``llvm.assume`` allows the optimizer to assume that the provided
11938 condition is true. This information can then be used in simplifying other parts
11944 The condition which the optimizer may assume is always true.
11949 The intrinsic allows the optimizer to assume that the provided condition is
11950 always true whenever the control flow reaches the intrinsic call. No code is
11951 generated for this intrinsic, and instructions that contribute only to the
11952 provided condition are not used for code generation. If the condition is
11953 violated during execution, the behavior is undefined.
11955 Note that the optimizer might limit the transformations performed on values
11956 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
11957 only used to form the intrinsic's input argument. This might prove undesirable
11958 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
11959 sufficient overall improvement in code quality. For this reason,
11960 ``llvm.assume`` should not be used to document basic mathematical invariants
11961 that the optimizer can otherwise deduce or facts that are of little use to the
11966 '``llvm.bitset.test``' Intrinsic
11967 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11974 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
11980 The first argument is a pointer to be tested. The second argument is a
11981 metadata object representing an identifier for a :doc:`bitset <BitSets>`.
11986 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
11987 member of the given bitset.
11989 '``llvm.donothing``' Intrinsic
11990 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11997 declare void @llvm.donothing() nounwind readnone
12002 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
12003 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
12004 with an invoke instruction.
12014 This intrinsic does nothing, and it's removed by optimizers and ignored
12017 Stack Map Intrinsics
12018 --------------------
12020 LLVM provides experimental intrinsics to support runtime patching
12021 mechanisms commonly desired in dynamic language JITs. These intrinsics
12022 are described in :doc:`StackMaps`.