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.
1455 operand bundle set ::= '[' operand bundle ']'
1456 operand bundle ::= tag '(' [ bundle operand ] (, bundle operand )* ')'
1457 bundle operand ::= SSA value
1458 tag ::= string constant
1460 Operand bundles are **not** part of a function's signature, and a
1461 given function may be called from multiple places with different kinds
1462 of operand bundles. This reflects the fact that the operand bundles
1463 are conceptually a part of the ``call`` (or ``invoke``), not the
1464 callee being dispatched to.
1466 Operand bundles are a generic mechanism intended to support
1467 runtime-introspection-like functionality for managed languages. While
1468 the exact semantics of an operand bundle depend on the bundle tag,
1469 there are certain limitations to how much the presence of an operand
1470 bundle can influence the semantics of a program. These restrictions
1471 are described as the semantics of an "unknown" operand bundle. As
1472 long as the behavior of an operand bundle is describable within these
1473 restrictions, LLVM does not need to have special knowledge of the
1474 operand bundle to not miscompile programs containing it.
1476 - The bundle operands for an unknown operand bundle escape in unknown
1477 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``).
1481 - An operand bundle at a call site cannot change the implementation
1482 of the called function. Inter-procedural optimizations work as
1483 usual as long as they take into account the first two properties.
1487 Module-Level Inline Assembly
1488 ----------------------------
1490 Modules may contain "module-level inline asm" blocks, which corresponds
1491 to the GCC "file scope inline asm" blocks. These blocks are internally
1492 concatenated by LLVM and treated as a single unit, but may be separated
1493 in the ``.ll`` file if desired. The syntax is very simple:
1495 .. code-block:: llvm
1497 module asm "inline asm code goes here"
1498 module asm "more can go here"
1500 The strings can contain any character by escaping non-printable
1501 characters. The escape sequence used is simply "\\xx" where "xx" is the
1502 two digit hex code for the number.
1504 Note that the assembly string *must* be parseable by LLVM's integrated assembler
1505 (unless it is disabled), even when emitting a ``.s`` file.
1507 .. _langref_datalayout:
1512 A module may specify a target specific data layout string that specifies
1513 how data is to be laid out in memory. The syntax for the data layout is
1516 .. code-block:: llvm
1518 target datalayout = "layout specification"
1520 The *layout specification* consists of a list of specifications
1521 separated by the minus sign character ('-'). Each specification starts
1522 with a letter and may include other information after the letter to
1523 define some aspect of the data layout. The specifications accepted are
1527 Specifies that the target lays out data in big-endian form. That is,
1528 the bits with the most significance have the lowest address
1531 Specifies that the target lays out data in little-endian form. That
1532 is, the bits with the least significance have the lowest address
1535 Specifies the natural alignment of the stack in bits. Alignment
1536 promotion of stack variables is limited to the natural stack
1537 alignment to avoid dynamic stack realignment. The stack alignment
1538 must be a multiple of 8-bits. If omitted, the natural stack
1539 alignment defaults to "unspecified", which does not prevent any
1540 alignment promotions.
1541 ``p[n]:<size>:<abi>:<pref>``
1542 This specifies the *size* of a pointer and its ``<abi>`` and
1543 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1544 bits. The address space, ``n``, is optional, and if not specified,
1545 denotes the default address space 0. The value of ``n`` must be
1546 in the range [1,2^23).
1547 ``i<size>:<abi>:<pref>``
1548 This specifies the alignment for an integer type of a given bit
1549 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1550 ``v<size>:<abi>:<pref>``
1551 This specifies the alignment for a vector type of a given bit
1553 ``f<size>:<abi>:<pref>``
1554 This specifies the alignment for a floating point type of a given bit
1555 ``<size>``. Only values of ``<size>`` that are supported by the target
1556 will work. 32 (float) and 64 (double) are supported on all targets; 80
1557 or 128 (different flavors of long double) are also supported on some
1560 This specifies the alignment for an object of aggregate type.
1562 If present, specifies that llvm names are mangled in the output. The
1565 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1566 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1567 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1568 symbols get a ``_`` prefix.
1569 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1570 functions also get a suffix based on the frame size.
1571 ``n<size1>:<size2>:<size3>...``
1572 This specifies a set of native integer widths for the target CPU in
1573 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1574 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1575 this set are considered to support most general arithmetic operations
1578 On every specification that takes a ``<abi>:<pref>``, specifying the
1579 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1580 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1582 When constructing the data layout for a given target, LLVM starts with a
1583 default set of specifications which are then (possibly) overridden by
1584 the specifications in the ``datalayout`` keyword. The default
1585 specifications are given in this list:
1587 - ``E`` - big endian
1588 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1589 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1590 same as the default address space.
1591 - ``S0`` - natural stack alignment is unspecified
1592 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1593 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1594 - ``i16:16:16`` - i16 is 16-bit aligned
1595 - ``i32:32:32`` - i32 is 32-bit aligned
1596 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1597 alignment of 64-bits
1598 - ``f16:16:16`` - half is 16-bit aligned
1599 - ``f32:32:32`` - float is 32-bit aligned
1600 - ``f64:64:64`` - double is 64-bit aligned
1601 - ``f128:128:128`` - quad is 128-bit aligned
1602 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1603 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1604 - ``a:0:64`` - aggregates are 64-bit aligned
1606 When LLVM is determining the alignment for a given type, it uses the
1609 #. If the type sought is an exact match for one of the specifications,
1610 that specification is used.
1611 #. If no match is found, and the type sought is an integer type, then
1612 the smallest integer type that is larger than the bitwidth of the
1613 sought type is used. If none of the specifications are larger than
1614 the bitwidth then the largest integer type is used. For example,
1615 given the default specifications above, the i7 type will use the
1616 alignment of i8 (next largest) while both i65 and i256 will use the
1617 alignment of i64 (largest specified).
1618 #. If no match is found, and the type sought is a vector type, then the
1619 largest vector type that is smaller than the sought vector type will
1620 be used as a fall back. This happens because <128 x double> can be
1621 implemented in terms of 64 <2 x double>, for example.
1623 The function of the data layout string may not be what you expect.
1624 Notably, this is not a specification from the frontend of what alignment
1625 the code generator should use.
1627 Instead, if specified, the target data layout is required to match what
1628 the ultimate *code generator* expects. This string is used by the
1629 mid-level optimizers to improve code, and this only works if it matches
1630 what the ultimate code generator uses. There is no way to generate IR
1631 that does not embed this target-specific detail into the IR. If you
1632 don't specify the string, the default specifications will be used to
1633 generate a Data Layout and the optimization phases will operate
1634 accordingly and introduce target specificity into the IR with respect to
1635 these default specifications.
1642 A module may specify a target triple string that describes the target
1643 host. The syntax for the target triple is simply:
1645 .. code-block:: llvm
1647 target triple = "x86_64-apple-macosx10.7.0"
1649 The *target triple* string consists of a series of identifiers delimited
1650 by the minus sign character ('-'). The canonical forms are:
1654 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1655 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1657 This information is passed along to the backend so that it generates
1658 code for the proper architecture. It's possible to override this on the
1659 command line with the ``-mtriple`` command line option.
1661 .. _pointeraliasing:
1663 Pointer Aliasing Rules
1664 ----------------------
1666 Any memory access must be done through a pointer value associated with
1667 an address range of the memory access, otherwise the behavior is
1668 undefined. Pointer values are associated with address ranges according
1669 to the following rules:
1671 - A pointer value is associated with the addresses associated with any
1672 value it is *based* on.
1673 - An address of a global variable is associated with the address range
1674 of the variable's storage.
1675 - The result value of an allocation instruction is associated with the
1676 address range of the allocated storage.
1677 - A null pointer in the default address-space is associated with no
1679 - An integer constant other than zero or a pointer value returned from
1680 a function not defined within LLVM may be associated with address
1681 ranges allocated through mechanisms other than those provided by
1682 LLVM. Such ranges shall not overlap with any ranges of addresses
1683 allocated by mechanisms provided by LLVM.
1685 A pointer value is *based* on another pointer value according to the
1688 - A pointer value formed from a ``getelementptr`` operation is *based*
1689 on the first value operand of the ``getelementptr``.
1690 - The result value of a ``bitcast`` is *based* on the operand of the
1692 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1693 values that contribute (directly or indirectly) to the computation of
1694 the pointer's value.
1695 - The "*based* on" relationship is transitive.
1697 Note that this definition of *"based"* is intentionally similar to the
1698 definition of *"based"* in C99, though it is slightly weaker.
1700 LLVM IR does not associate types with memory. The result type of a
1701 ``load`` merely indicates the size and alignment of the memory from
1702 which to load, as well as the interpretation of the value. The first
1703 operand type of a ``store`` similarly only indicates the size and
1704 alignment of the store.
1706 Consequently, type-based alias analysis, aka TBAA, aka
1707 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1708 :ref:`Metadata <metadata>` may be used to encode additional information
1709 which specialized optimization passes may use to implement type-based
1714 Volatile Memory Accesses
1715 ------------------------
1717 Certain memory accesses, such as :ref:`load <i_load>`'s,
1718 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1719 marked ``volatile``. The optimizers must not change the number of
1720 volatile operations or change their order of execution relative to other
1721 volatile operations. The optimizers *may* change the order of volatile
1722 operations relative to non-volatile operations. This is not Java's
1723 "volatile" and has no cross-thread synchronization behavior.
1725 IR-level volatile loads and stores cannot safely be optimized into
1726 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1727 flagged volatile. Likewise, the backend should never split or merge
1728 target-legal volatile load/store instructions.
1730 .. admonition:: Rationale
1732 Platforms may rely on volatile loads and stores of natively supported
1733 data width to be executed as single instruction. For example, in C
1734 this holds for an l-value of volatile primitive type with native
1735 hardware support, but not necessarily for aggregate types. The
1736 frontend upholds these expectations, which are intentionally
1737 unspecified in the IR. The rules above ensure that IR transformations
1738 do not violate the frontend's contract with the language.
1742 Memory Model for Concurrent Operations
1743 --------------------------------------
1745 The LLVM IR does not define any way to start parallel threads of
1746 execution or to register signal handlers. Nonetheless, there are
1747 platform-specific ways to create them, and we define LLVM IR's behavior
1748 in their presence. This model is inspired by the C++0x memory model.
1750 For a more informal introduction to this model, see the :doc:`Atomics`.
1752 We define a *happens-before* partial order as the least partial order
1755 - Is a superset of single-thread program order, and
1756 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1757 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1758 techniques, like pthread locks, thread creation, thread joining,
1759 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1760 Constraints <ordering>`).
1762 Note that program order does not introduce *happens-before* edges
1763 between a thread and signals executing inside that thread.
1765 Every (defined) read operation (load instructions, memcpy, atomic
1766 loads/read-modify-writes, etc.) R reads a series of bytes written by
1767 (defined) write operations (store instructions, atomic
1768 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1769 section, initialized globals are considered to have a write of the
1770 initializer which is atomic and happens before any other read or write
1771 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1772 may see any write to the same byte, except:
1774 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1775 write\ :sub:`2` happens before R\ :sub:`byte`, then
1776 R\ :sub:`byte` does not see write\ :sub:`1`.
1777 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1778 R\ :sub:`byte` does not see write\ :sub:`3`.
1780 Given that definition, R\ :sub:`byte` is defined as follows:
1782 - If R is volatile, the result is target-dependent. (Volatile is
1783 supposed to give guarantees which can support ``sig_atomic_t`` in
1784 C/C++, and may be used for accesses to addresses that do not behave
1785 like normal memory. It does not generally provide cross-thread
1787 - Otherwise, if there is no write to the same byte that happens before
1788 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1789 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1790 R\ :sub:`byte` returns the value written by that write.
1791 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1792 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1793 Memory Ordering Constraints <ordering>` section for additional
1794 constraints on how the choice is made.
1795 - Otherwise R\ :sub:`byte` returns ``undef``.
1797 R returns the value composed of the series of bytes it read. This
1798 implies that some bytes within the value may be ``undef`` **without**
1799 the entire value being ``undef``. Note that this only defines the
1800 semantics of the operation; it doesn't mean that targets will emit more
1801 than one instruction to read the series of bytes.
1803 Note that in cases where none of the atomic intrinsics are used, this
1804 model places only one restriction on IR transformations on top of what
1805 is required for single-threaded execution: introducing a store to a byte
1806 which might not otherwise be stored is not allowed in general.
1807 (Specifically, in the case where another thread might write to and read
1808 from an address, introducing a store can change a load that may see
1809 exactly one write into a load that may see multiple writes.)
1813 Atomic Memory Ordering Constraints
1814 ----------------------------------
1816 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1817 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1818 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1819 ordering parameters that determine which other atomic instructions on
1820 the same address they *synchronize with*. These semantics are borrowed
1821 from Java and C++0x, but are somewhat more colloquial. If these
1822 descriptions aren't precise enough, check those specs (see spec
1823 references in the :doc:`atomics guide <Atomics>`).
1824 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1825 differently since they don't take an address. See that instruction's
1826 documentation for details.
1828 For a simpler introduction to the ordering constraints, see the
1832 The set of values that can be read is governed by the happens-before
1833 partial order. A value cannot be read unless some operation wrote
1834 it. This is intended to provide a guarantee strong enough to model
1835 Java's non-volatile shared variables. This ordering cannot be
1836 specified for read-modify-write operations; it is not strong enough
1837 to make them atomic in any interesting way.
1839 In addition to the guarantees of ``unordered``, there is a single
1840 total order for modifications by ``monotonic`` operations on each
1841 address. All modification orders must be compatible with the
1842 happens-before order. There is no guarantee that the modification
1843 orders can be combined to a global total order for the whole program
1844 (and this often will not be possible). The read in an atomic
1845 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1846 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1847 order immediately before the value it writes. If one atomic read
1848 happens before another atomic read of the same address, the later
1849 read must see the same value or a later value in the address's
1850 modification order. This disallows reordering of ``monotonic`` (or
1851 stronger) operations on the same address. If an address is written
1852 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1853 read that address repeatedly, the other threads must eventually see
1854 the write. This corresponds to the C++0x/C1x
1855 ``memory_order_relaxed``.
1857 In addition to the guarantees of ``monotonic``, a
1858 *synchronizes-with* edge may be formed with a ``release`` operation.
1859 This is intended to model C++'s ``memory_order_acquire``.
1861 In addition to the guarantees of ``monotonic``, if this operation
1862 writes a value which is subsequently read by an ``acquire``
1863 operation, it *synchronizes-with* that operation. (This isn't a
1864 complete description; see the C++0x definition of a release
1865 sequence.) This corresponds to the C++0x/C1x
1866 ``memory_order_release``.
1867 ``acq_rel`` (acquire+release)
1868 Acts as both an ``acquire`` and ``release`` operation on its
1869 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1870 ``seq_cst`` (sequentially consistent)
1871 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1872 operation that only reads, ``release`` for an operation that only
1873 writes), there is a global total order on all
1874 sequentially-consistent operations on all addresses, which is
1875 consistent with the *happens-before* partial order and with the
1876 modification orders of all the affected addresses. Each
1877 sequentially-consistent read sees the last preceding write to the
1878 same address in this global order. This corresponds to the C++0x/C1x
1879 ``memory_order_seq_cst`` and Java volatile.
1883 If an atomic operation is marked ``singlethread``, it only *synchronizes
1884 with* or participates in modification and seq\_cst total orderings with
1885 other operations running in the same thread (for example, in signal
1893 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1894 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1895 :ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) have the following flags that can
1896 be set to enable otherwise unsafe floating point operations
1899 No NaNs - Allow optimizations to assume the arguments and result are not
1900 NaN. Such optimizations are required to retain defined behavior over
1901 NaNs, but the value of the result is undefined.
1904 No Infs - Allow optimizations to assume the arguments and result are not
1905 +/-Inf. Such optimizations are required to retain defined behavior over
1906 +/-Inf, but the value of the result is undefined.
1909 No Signed Zeros - Allow optimizations to treat the sign of a zero
1910 argument or result as insignificant.
1913 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1914 argument rather than perform division.
1917 Fast - Allow algebraically equivalent transformations that may
1918 dramatically change results in floating point (e.g. reassociate). This
1919 flag implies all the others.
1923 Use-list Order Directives
1924 -------------------------
1926 Use-list directives encode the in-memory order of each use-list, allowing the
1927 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1928 indexes that are assigned to the referenced value's uses. The referenced
1929 value's use-list is immediately sorted by these indexes.
1931 Use-list directives may appear at function scope or global scope. They are not
1932 instructions, and have no effect on the semantics of the IR. When they're at
1933 function scope, they must appear after the terminator of the final basic block.
1935 If basic blocks have their address taken via ``blockaddress()`` expressions,
1936 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1943 uselistorder <ty> <value>, { <order-indexes> }
1944 uselistorder_bb @function, %block { <order-indexes> }
1950 define void @foo(i32 %arg1, i32 %arg2) {
1952 ; ... instructions ...
1954 ; ... instructions ...
1956 ; At function scope.
1957 uselistorder i32 %arg1, { 1, 0, 2 }
1958 uselistorder label %bb, { 1, 0 }
1962 uselistorder i32* @global, { 1, 2, 0 }
1963 uselistorder i32 7, { 1, 0 }
1964 uselistorder i32 (i32) @bar, { 1, 0 }
1965 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1972 The LLVM type system is one of the most important features of the
1973 intermediate representation. Being typed enables a number of
1974 optimizations to be performed on the intermediate representation
1975 directly, without having to do extra analyses on the side before the
1976 transformation. A strong type system makes it easier to read the
1977 generated code and enables novel analyses and transformations that are
1978 not feasible to perform on normal three address code representations.
1988 The void type does not represent any value and has no size.
2006 The function type can be thought of as a function signature. It consists of a
2007 return type and a list of formal parameter types. The return type of a function
2008 type is a void type or first class type --- except for :ref:`label <t_label>`
2009 and :ref:`metadata <t_metadata>` types.
2015 <returntype> (<parameter list>)
2017 ...where '``<parameter list>``' is a comma-separated list of type
2018 specifiers. Optionally, the parameter list may include a type ``...``, which
2019 indicates that the function takes a variable number of arguments. Variable
2020 argument functions can access their arguments with the :ref:`variable argument
2021 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
2022 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
2026 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2027 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
2028 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2029 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
2030 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2031 | ``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. |
2032 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2033 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
2034 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2041 The :ref:`first class <t_firstclass>` types are perhaps the most important.
2042 Values of these types are the only ones which can be produced by
2050 These are the types that are valid in registers from CodeGen's perspective.
2059 The integer type is a very simple type that simply specifies an
2060 arbitrary bit width for the integer type desired. Any bit width from 1
2061 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2069 The number of bits the integer will occupy is specified by the ``N``
2075 +----------------+------------------------------------------------+
2076 | ``i1`` | a single-bit integer. |
2077 +----------------+------------------------------------------------+
2078 | ``i32`` | a 32-bit integer. |
2079 +----------------+------------------------------------------------+
2080 | ``i1942652`` | a really big integer of over 1 million bits. |
2081 +----------------+------------------------------------------------+
2085 Floating Point Types
2086 """"""""""""""""""""
2095 - 16-bit floating point value
2098 - 32-bit floating point value
2101 - 64-bit floating point value
2104 - 128-bit floating point value (112-bit mantissa)
2107 - 80-bit floating point value (X87)
2110 - 128-bit floating point value (two 64-bits)
2117 The x86_mmx type represents a value held in an MMX register on an x86
2118 machine. The operations allowed on it are quite limited: parameters and
2119 return values, load and store, and bitcast. User-specified MMX
2120 instructions are represented as intrinsic or asm calls with arguments
2121 and/or results of this type. There are no arrays, vectors or constants
2138 The pointer type is used to specify memory locations. Pointers are
2139 commonly used to reference objects in memory.
2141 Pointer types may have an optional address space attribute defining the
2142 numbered address space where the pointed-to object resides. The default
2143 address space is number zero. The semantics of non-zero address spaces
2144 are target-specific.
2146 Note that LLVM does not permit pointers to void (``void*``) nor does it
2147 permit pointers to labels (``label*``). Use ``i8*`` instead.
2157 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2158 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2159 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2160 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2161 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2162 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2163 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2172 A vector type is a simple derived type that represents a vector of
2173 elements. Vector types are used when multiple primitive data are
2174 operated in parallel using a single instruction (SIMD). A vector type
2175 requires a size (number of elements) and an underlying primitive data
2176 type. Vector types are considered :ref:`first class <t_firstclass>`.
2182 < <# elements> x <elementtype> >
2184 The number of elements is a constant integer value larger than 0;
2185 elementtype may be any integer, floating point or pointer type. Vectors
2186 of size zero are not allowed.
2190 +-------------------+--------------------------------------------------+
2191 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2192 +-------------------+--------------------------------------------------+
2193 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2194 +-------------------+--------------------------------------------------+
2195 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2196 +-------------------+--------------------------------------------------+
2197 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2198 +-------------------+--------------------------------------------------+
2207 The label type represents code labels.
2222 The token type is used when a value is associated with an instruction
2223 but all uses of the value must not attempt to introspect or obscure it.
2224 As such, it is not appropriate to have a :ref:`phi <i_phi>` or
2225 :ref:`select <i_select>` of type token.
2242 The metadata type represents embedded metadata. No derived types may be
2243 created from metadata except for :ref:`function <t_function>` arguments.
2256 Aggregate Types are a subset of derived types that can contain multiple
2257 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2258 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2268 The array type is a very simple derived type that arranges elements
2269 sequentially in memory. The array type requires a size (number of
2270 elements) and an underlying data type.
2276 [<# elements> x <elementtype>]
2278 The number of elements is a constant integer value; ``elementtype`` may
2279 be any type with a size.
2283 +------------------+--------------------------------------+
2284 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2285 +------------------+--------------------------------------+
2286 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2287 +------------------+--------------------------------------+
2288 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2289 +------------------+--------------------------------------+
2291 Here are some examples of multidimensional arrays:
2293 +-----------------------------+----------------------------------------------------------+
2294 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2295 +-----------------------------+----------------------------------------------------------+
2296 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2297 +-----------------------------+----------------------------------------------------------+
2298 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2299 +-----------------------------+----------------------------------------------------------+
2301 There is no restriction on indexing beyond the end of the array implied
2302 by a static type (though there are restrictions on indexing beyond the
2303 bounds of an allocated object in some cases). This means that
2304 single-dimension 'variable sized array' addressing can be implemented in
2305 LLVM with a zero length array type. An implementation of 'pascal style
2306 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2316 The structure type is used to represent a collection of data members
2317 together in memory. The elements of a structure may be any type that has
2320 Structures in memory are accessed using '``load``' and '``store``' by
2321 getting a pointer to a field with the '``getelementptr``' instruction.
2322 Structures in registers are accessed using the '``extractvalue``' and
2323 '``insertvalue``' instructions.
2325 Structures may optionally be "packed" structures, which indicate that
2326 the alignment of the struct is one byte, and that there is no padding
2327 between the elements. In non-packed structs, padding between field types
2328 is inserted as defined by the DataLayout string in the module, which is
2329 required to match what the underlying code generator expects.
2331 Structures can either be "literal" or "identified". A literal structure
2332 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2333 identified types are always defined at the top level with a name.
2334 Literal types are uniqued by their contents and can never be recursive
2335 or opaque since there is no way to write one. Identified types can be
2336 recursive, can be opaqued, and are never uniqued.
2342 %T1 = type { <type list> } ; Identified normal struct type
2343 %T2 = type <{ <type list> }> ; Identified packed struct type
2347 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2348 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2349 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2350 | ``{ 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``. |
2351 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2352 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2353 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2357 Opaque Structure Types
2358 """"""""""""""""""""""
2362 Opaque structure types are used to represent named structure types that
2363 do not have a body specified. This corresponds (for example) to the C
2364 notion of a forward declared structure.
2375 +--------------+-------------------+
2376 | ``opaque`` | An opaque type. |
2377 +--------------+-------------------+
2384 LLVM has several different basic types of constants. This section
2385 describes them all and their syntax.
2390 **Boolean constants**
2391 The two strings '``true``' and '``false``' are both valid constants
2393 **Integer constants**
2394 Standard integers (such as '4') are constants of the
2395 :ref:`integer <t_integer>` type. Negative numbers may be used with
2397 **Floating point constants**
2398 Floating point constants use standard decimal notation (e.g.
2399 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2400 hexadecimal notation (see below). The assembler requires the exact
2401 decimal value of a floating-point constant. For example, the
2402 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2403 decimal in binary. Floating point constants must have a :ref:`floating
2404 point <t_floating>` type.
2405 **Null pointer constants**
2406 The identifier '``null``' is recognized as a null pointer constant
2407 and must be of :ref:`pointer type <t_pointer>`.
2409 The one non-intuitive notation for constants is the hexadecimal form of
2410 floating point constants. For example, the form
2411 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2412 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2413 constants are required (and the only time that they are generated by the
2414 disassembler) is when a floating point constant must be emitted but it
2415 cannot be represented as a decimal floating point number in a reasonable
2416 number of digits. For example, NaN's, infinities, and other special
2417 values are represented in their IEEE hexadecimal format so that assembly
2418 and disassembly do not cause any bits to change in the constants.
2420 When using the hexadecimal form, constants of types half, float, and
2421 double are represented using the 16-digit form shown above (which
2422 matches the IEEE754 representation for double); half and float values
2423 must, however, be exactly representable as IEEE 754 half and single
2424 precision, respectively. Hexadecimal format is always used for long
2425 double, and there are three forms of long double. The 80-bit format used
2426 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2427 128-bit format used by PowerPC (two adjacent doubles) is represented by
2428 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2429 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2430 will only work if they match the long double format on your target.
2431 The IEEE 16-bit format (half precision) is represented by ``0xH``
2432 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2433 (sign bit at the left).
2435 There are no constants of type x86_mmx.
2437 .. _complexconstants:
2442 Complex constants are a (potentially recursive) combination of simple
2443 constants and smaller complex constants.
2445 **Structure constants**
2446 Structure constants are represented with notation similar to
2447 structure type definitions (a comma separated list of elements,
2448 surrounded by braces (``{}``)). For example:
2449 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2450 "``@G = external global i32``". Structure constants must have
2451 :ref:`structure type <t_struct>`, and the number and types of elements
2452 must match those specified by the type.
2454 Array constants are represented with notation similar to array type
2455 definitions (a comma separated list of elements, surrounded by
2456 square brackets (``[]``)). For example:
2457 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2458 :ref:`array type <t_array>`, and the number and types of elements must
2459 match those specified by the type. As a special case, character array
2460 constants may also be represented as a double-quoted string using the ``c``
2461 prefix. For example: "``c"Hello World\0A\00"``".
2462 **Vector constants**
2463 Vector constants are represented with notation similar to vector
2464 type definitions (a comma separated list of elements, surrounded by
2465 less-than/greater-than's (``<>``)). For example:
2466 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2467 must have :ref:`vector type <t_vector>`, and the number and types of
2468 elements must match those specified by the type.
2469 **Zero initialization**
2470 The string '``zeroinitializer``' can be used to zero initialize a
2471 value to zero of *any* type, including scalar and
2472 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2473 having to print large zero initializers (e.g. for large arrays) and
2474 is always exactly equivalent to using explicit zero initializers.
2476 A metadata node is a constant tuple without types. For example:
2477 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2478 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2479 Unlike other typed constants that are meant to be interpreted as part of
2480 the instruction stream, metadata is a place to attach additional
2481 information such as debug info.
2483 Global Variable and Function Addresses
2484 --------------------------------------
2486 The addresses of :ref:`global variables <globalvars>` and
2487 :ref:`functions <functionstructure>` are always implicitly valid
2488 (link-time) constants. These constants are explicitly referenced when
2489 the :ref:`identifier for the global <identifiers>` is used and always have
2490 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2493 .. code-block:: llvm
2497 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2504 The string '``undef``' can be used anywhere a constant is expected, and
2505 indicates that the user of the value may receive an unspecified
2506 bit-pattern. Undefined values may be of any type (other than '``label``'
2507 or '``void``') and be used anywhere a constant is permitted.
2509 Undefined values are useful because they indicate to the compiler that
2510 the program is well defined no matter what value is used. This gives the
2511 compiler more freedom to optimize. Here are some examples of
2512 (potentially surprising) transformations that are valid (in pseudo IR):
2514 .. code-block:: llvm
2524 This is safe because all of the output bits are affected by the undef
2525 bits. Any output bit can have a zero or one depending on the input bits.
2527 .. code-block:: llvm
2538 These logical operations have bits that are not always affected by the
2539 input. For example, if ``%X`` has a zero bit, then the output of the
2540 '``and``' operation will always be a zero for that bit, no matter what
2541 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2542 optimize or assume that the result of the '``and``' is '``undef``'.
2543 However, it is safe to assume that all bits of the '``undef``' could be
2544 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2545 all the bits of the '``undef``' operand to the '``or``' could be set,
2546 allowing the '``or``' to be folded to -1.
2548 .. code-block:: llvm
2550 %A = select undef, %X, %Y
2551 %B = select undef, 42, %Y
2552 %C = select %X, %Y, undef
2562 This set of examples shows that undefined '``select``' (and conditional
2563 branch) conditions can go *either way*, but they have to come from one
2564 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2565 both known to have a clear low bit, then ``%A`` would have to have a
2566 cleared low bit. However, in the ``%C`` example, the optimizer is
2567 allowed to assume that the '``undef``' operand could be the same as
2568 ``%Y``, allowing the whole '``select``' to be eliminated.
2570 .. code-block:: llvm
2572 %A = xor undef, undef
2589 This example points out that two '``undef``' operands are not
2590 necessarily the same. This can be surprising to people (and also matches
2591 C semantics) where they assume that "``X^X``" is always zero, even if
2592 ``X`` is undefined. This isn't true for a number of reasons, but the
2593 short answer is that an '``undef``' "variable" can arbitrarily change
2594 its value over its "live range". This is true because the variable
2595 doesn't actually *have a live range*. Instead, the value is logically
2596 read from arbitrary registers that happen to be around when needed, so
2597 the value is not necessarily consistent over time. In fact, ``%A`` and
2598 ``%C`` need to have the same semantics or the core LLVM "replace all
2599 uses with" concept would not hold.
2601 .. code-block:: llvm
2609 These examples show the crucial difference between an *undefined value*
2610 and *undefined behavior*. An undefined value (like '``undef``') is
2611 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2612 operation can be constant folded to '``undef``', because the '``undef``'
2613 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2614 However, in the second example, we can make a more aggressive
2615 assumption: because the ``undef`` is allowed to be an arbitrary value,
2616 we are allowed to assume that it could be zero. Since a divide by zero
2617 has *undefined behavior*, we are allowed to assume that the operation
2618 does not execute at all. This allows us to delete the divide and all
2619 code after it. Because the undefined operation "can't happen", the
2620 optimizer can assume that it occurs in dead code.
2622 .. code-block:: llvm
2624 a: store undef -> %X
2625 b: store %X -> undef
2630 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2631 value can be assumed to not have any effect; we can assume that the
2632 value is overwritten with bits that happen to match what was already
2633 there. However, a store *to* an undefined location could clobber
2634 arbitrary memory, therefore, it has undefined behavior.
2641 Poison values are similar to :ref:`undef values <undefvalues>`, however
2642 they also represent the fact that an instruction or constant expression
2643 that cannot evoke side effects has nevertheless detected a condition
2644 that results in undefined behavior.
2646 There is currently no way of representing a poison value in the IR; they
2647 only exist when produced by operations such as :ref:`add <i_add>` with
2650 Poison value behavior is defined in terms of value *dependence*:
2652 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2653 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2654 their dynamic predecessor basic block.
2655 - Function arguments depend on the corresponding actual argument values
2656 in the dynamic callers of their functions.
2657 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2658 instructions that dynamically transfer control back to them.
2659 - :ref:`Invoke <i_invoke>` instructions depend on the
2660 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2661 call instructions that dynamically transfer control back to them.
2662 - Non-volatile loads and stores depend on the most recent stores to all
2663 of the referenced memory addresses, following the order in the IR
2664 (including loads and stores implied by intrinsics such as
2665 :ref:`@llvm.memcpy <int_memcpy>`.)
2666 - An instruction with externally visible side effects depends on the
2667 most recent preceding instruction with externally visible side
2668 effects, following the order in the IR. (This includes :ref:`volatile
2669 operations <volatile>`.)
2670 - An instruction *control-depends* on a :ref:`terminator
2671 instruction <terminators>` if the terminator instruction has
2672 multiple successors and the instruction is always executed when
2673 control transfers to one of the successors, and may not be executed
2674 when control is transferred to another.
2675 - Additionally, an instruction also *control-depends* on a terminator
2676 instruction if the set of instructions it otherwise depends on would
2677 be different if the terminator had transferred control to a different
2679 - Dependence is transitive.
2681 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2682 with the additional effect that any instruction that has a *dependence*
2683 on a poison value has undefined behavior.
2685 Here are some examples:
2687 .. code-block:: llvm
2690 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2691 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2692 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2693 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2695 store i32 %poison, i32* @g ; Poison value stored to memory.
2696 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2698 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2700 %narrowaddr = bitcast i32* @g to i16*
2701 %wideaddr = bitcast i32* @g to i64*
2702 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2703 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2705 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2706 br i1 %cmp, label %true, label %end ; Branch to either destination.
2709 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2710 ; it has undefined behavior.
2714 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2715 ; Both edges into this PHI are
2716 ; control-dependent on %cmp, so this
2717 ; always results in a poison value.
2719 store volatile i32 0, i32* @g ; This would depend on the store in %true
2720 ; if %cmp is true, or the store in %entry
2721 ; otherwise, so this is undefined behavior.
2723 br i1 %cmp, label %second_true, label %second_end
2724 ; The same branch again, but this time the
2725 ; true block doesn't have side effects.
2732 store volatile i32 0, i32* @g ; This time, the instruction always depends
2733 ; on the store in %end. Also, it is
2734 ; control-equivalent to %end, so this is
2735 ; well-defined (ignoring earlier undefined
2736 ; behavior in this example).
2740 Addresses of Basic Blocks
2741 -------------------------
2743 ``blockaddress(@function, %block)``
2745 The '``blockaddress``' constant computes the address of the specified
2746 basic block in the specified function, and always has an ``i8*`` type.
2747 Taking the address of the entry block is illegal.
2749 This value only has defined behavior when used as an operand to the
2750 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2751 against null. Pointer equality tests between labels addresses results in
2752 undefined behavior --- though, again, comparison against null is ok, and
2753 no label is equal to the null pointer. This may be passed around as an
2754 opaque pointer sized value as long as the bits are not inspected. This
2755 allows ``ptrtoint`` and arithmetic to be performed on these values so
2756 long as the original value is reconstituted before the ``indirectbr``
2759 Finally, some targets may provide defined semantics when using the value
2760 as the operand to an inline assembly, but that is target specific.
2764 Constant Expressions
2765 --------------------
2767 Constant expressions are used to allow expressions involving other
2768 constants to be used as constants. Constant expressions may be of any
2769 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2770 that does not have side effects (e.g. load and call are not supported).
2771 The following is the syntax for constant expressions:
2773 ``trunc (CST to TYPE)``
2774 Truncate a constant to another type. The bit size of CST must be
2775 larger than the bit size of TYPE. Both types must be integers.
2776 ``zext (CST to TYPE)``
2777 Zero extend a constant to another type. The bit size of CST must be
2778 smaller than the bit size of TYPE. Both types must be integers.
2779 ``sext (CST to TYPE)``
2780 Sign extend a constant to another type. The bit size of CST must be
2781 smaller than the bit size of TYPE. Both types must be integers.
2782 ``fptrunc (CST to TYPE)``
2783 Truncate a floating point constant to another floating point type.
2784 The size of CST must be larger than the size of TYPE. Both types
2785 must be floating point.
2786 ``fpext (CST to TYPE)``
2787 Floating point extend a constant to another type. The size of CST
2788 must be smaller or equal to the size of TYPE. Both types must be
2790 ``fptoui (CST to TYPE)``
2791 Convert a floating point constant to the corresponding unsigned
2792 integer constant. TYPE must be a scalar or vector integer type. CST
2793 must be of scalar or vector floating point type. Both CST and TYPE
2794 must be scalars, or vectors of the same number of elements. If the
2795 value won't fit in the integer type, the results are undefined.
2796 ``fptosi (CST to TYPE)``
2797 Convert a floating point constant to the corresponding signed
2798 integer constant. TYPE must be a scalar or vector integer type. CST
2799 must be of scalar or vector floating point type. Both CST and TYPE
2800 must be scalars, or vectors of the same number of elements. If the
2801 value won't fit in the integer type, the results are undefined.
2802 ``uitofp (CST to TYPE)``
2803 Convert an unsigned integer constant to the corresponding floating
2804 point constant. TYPE must be a scalar or vector floating point type.
2805 CST must be of scalar or vector integer type. Both CST and TYPE must
2806 be scalars, or vectors of the same number of elements. If the value
2807 won't fit in the floating point type, the results are undefined.
2808 ``sitofp (CST to TYPE)``
2809 Convert a signed integer constant to the corresponding floating
2810 point constant. TYPE must be a scalar or vector floating point type.
2811 CST must be of scalar or vector integer type. Both CST and TYPE must
2812 be scalars, or vectors of the same number of elements. If the value
2813 won't fit in the floating point type, the results are undefined.
2814 ``ptrtoint (CST to TYPE)``
2815 Convert a pointer typed constant to the corresponding integer
2816 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2817 pointer type. The ``CST`` value is zero extended, truncated, or
2818 unchanged to make it fit in ``TYPE``.
2819 ``inttoptr (CST to TYPE)``
2820 Convert an integer constant to a pointer constant. TYPE must be a
2821 pointer type. CST must be of integer type. The CST value is zero
2822 extended, truncated, or unchanged to make it fit in a pointer size.
2823 This one is *really* dangerous!
2824 ``bitcast (CST to TYPE)``
2825 Convert a constant, CST, to another TYPE. The constraints of the
2826 operands are the same as those for the :ref:`bitcast
2827 instruction <i_bitcast>`.
2828 ``addrspacecast (CST to TYPE)``
2829 Convert a constant pointer or constant vector of pointer, CST, to another
2830 TYPE in a different address space. The constraints of the operands are the
2831 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2832 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2833 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2834 constants. As with the :ref:`getelementptr <i_getelementptr>`
2835 instruction, the index list may have zero or more indexes, which are
2836 required to make sense for the type of "pointer to TY".
2837 ``select (COND, VAL1, VAL2)``
2838 Perform the :ref:`select operation <i_select>` on constants.
2839 ``icmp COND (VAL1, VAL2)``
2840 Performs the :ref:`icmp operation <i_icmp>` on constants.
2841 ``fcmp COND (VAL1, VAL2)``
2842 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2843 ``extractelement (VAL, IDX)``
2844 Perform the :ref:`extractelement operation <i_extractelement>` on
2846 ``insertelement (VAL, ELT, IDX)``
2847 Perform the :ref:`insertelement operation <i_insertelement>` on
2849 ``shufflevector (VEC1, VEC2, IDXMASK)``
2850 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2852 ``extractvalue (VAL, IDX0, IDX1, ...)``
2853 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2854 constants. The index list is interpreted in a similar manner as
2855 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2856 least one index value must be specified.
2857 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2858 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2859 The index list is interpreted in a similar manner as indices in a
2860 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2861 value must be specified.
2862 ``OPCODE (LHS, RHS)``
2863 Perform the specified operation of the LHS and RHS constants. OPCODE
2864 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2865 binary <bitwiseops>` operations. The constraints on operands are
2866 the same as those for the corresponding instruction (e.g. no bitwise
2867 operations on floating point values are allowed).
2874 Inline Assembler Expressions
2875 ----------------------------
2877 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2878 Inline Assembly <moduleasm>`) through the use of a special value. This value
2879 represents the inline assembler as a template string (containing the
2880 instructions to emit), a list of operand constraints (stored as a string), a
2881 flag that indicates whether or not the inline asm expression has side effects,
2882 and a flag indicating whether the function containing the asm needs to align its
2883 stack conservatively.
2885 The template string supports argument substitution of the operands using "``$``"
2886 followed by a number, to indicate substitution of the given register/memory
2887 location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
2888 be used, where ``MODIFIER`` is a target-specific annotation for how to print the
2889 operand (See :ref:`inline-asm-modifiers`).
2891 A literal "``$``" may be included by using "``$$``" in the template. To include
2892 other special characters into the output, the usual "``\XX``" escapes may be
2893 used, just as in other strings. Note that after template substitution, the
2894 resulting assembly string is parsed by LLVM's integrated assembler unless it is
2895 disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
2896 syntax known to LLVM.
2898 LLVM's support for inline asm is modeled closely on the requirements of Clang's
2899 GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
2900 modifier codes listed here are similar or identical to those in GCC's inline asm
2901 support. However, to be clear, the syntax of the template and constraint strings
2902 described here is *not* the same as the syntax accepted by GCC and Clang, and,
2903 while most constraint letters are passed through as-is by Clang, some get
2904 translated to other codes when converting from the C source to the LLVM
2907 An example inline assembler expression is:
2909 .. code-block:: llvm
2911 i32 (i32) asm "bswap $0", "=r,r"
2913 Inline assembler expressions may **only** be used as the callee operand
2914 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2915 Thus, typically we have:
2917 .. code-block:: llvm
2919 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2921 Inline asms with side effects not visible in the constraint list must be
2922 marked as having side effects. This is done through the use of the
2923 '``sideeffect``' keyword, like so:
2925 .. code-block:: llvm
2927 call void asm sideeffect "eieio", ""()
2929 In some cases inline asms will contain code that will not work unless
2930 the stack is aligned in some way, such as calls or SSE instructions on
2931 x86, yet will not contain code that does that alignment within the asm.
2932 The compiler should make conservative assumptions about what the asm
2933 might contain and should generate its usual stack alignment code in the
2934 prologue if the '``alignstack``' keyword is present:
2936 .. code-block:: llvm
2938 call void asm alignstack "eieio", ""()
2940 Inline asms also support using non-standard assembly dialects. The
2941 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2942 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2943 the only supported dialects. An example is:
2945 .. code-block:: llvm
2947 call void asm inteldialect "eieio", ""()
2949 If multiple keywords appear the '``sideeffect``' keyword must come
2950 first, the '``alignstack``' keyword second and the '``inteldialect``'
2953 Inline Asm Constraint String
2954 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2956 The constraint list is a comma-separated string, each element containing one or
2957 more constraint codes.
2959 For each element in the constraint list an appropriate register or memory
2960 operand will be chosen, and it will be made available to assembly template
2961 string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
2964 There are three different types of constraints, which are distinguished by a
2965 prefix symbol in front of the constraint code: Output, Input, and Clobber. The
2966 constraints must always be given in that order: outputs first, then inputs, then
2967 clobbers. They cannot be intermingled.
2969 There are also three different categories of constraint codes:
2971 - Register constraint. This is either a register class, or a fixed physical
2972 register. This kind of constraint will allocate a register, and if necessary,
2973 bitcast the argument or result to the appropriate type.
2974 - Memory constraint. This kind of constraint is for use with an instruction
2975 taking a memory operand. Different constraints allow for different addressing
2976 modes used by the target.
2977 - Immediate value constraint. This kind of constraint is for an integer or other
2978 immediate value which can be rendered directly into an instruction. The
2979 various target-specific constraints allow the selection of a value in the
2980 proper range for the instruction you wish to use it with.
2985 Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
2986 indicates that the assembly will write to this operand, and the operand will
2987 then be made available as a return value of the ``asm`` expression. Output
2988 constraints do not consume an argument from the call instruction. (Except, see
2989 below about indirect outputs).
2991 Normally, it is expected that no output locations are written to by the assembly
2992 expression until *all* of the inputs have been read. As such, LLVM may assign
2993 the same register to an output and an input. If this is not safe (e.g. if the
2994 assembly contains two instructions, where the first writes to one output, and
2995 the second reads an input and writes to a second output), then the "``&``"
2996 modifier must be used (e.g. "``=&r``") to specify that the output is an
2997 "early-clobber" output. Marking an ouput as "early-clobber" ensures that LLVM
2998 will not use the same register for any inputs (other than an input tied to this
3004 Input constraints do not have a prefix -- just the constraint codes. Each input
3005 constraint will consume one argument from the call instruction. It is not
3006 permitted for the asm to write to any input register or memory location (unless
3007 that input is tied to an output). Note also that multiple inputs may all be
3008 assigned to the same register, if LLVM can determine that they necessarily all
3009 contain the same value.
3011 Instead of providing a Constraint Code, input constraints may also "tie"
3012 themselves to an output constraint, by providing an integer as the constraint
3013 string. Tied inputs still consume an argument from the call instruction, and
3014 take up a position in the asm template numbering as is usual -- they will simply
3015 be constrained to always use the same register as the output they've been tied
3016 to. For example, a constraint string of "``=r,0``" says to assign a register for
3017 output, and use that register as an input as well (it being the 0'th
3020 It is permitted to tie an input to an "early-clobber" output. In that case, no
3021 *other* input may share the same register as the input tied to the early-clobber
3022 (even when the other input has the same value).
3024 You may only tie an input to an output which has a register constraint, not a
3025 memory constraint. Only a single input may be tied to an output.
3027 There is also an "interesting" feature which deserves a bit of explanation: if a
3028 register class constraint allocates a register which is too small for the value
3029 type operand provided as input, the input value will be split into multiple
3030 registers, and all of them passed to the inline asm.
3032 However, this feature is often not as useful as you might think.
3034 Firstly, the registers are *not* guaranteed to be consecutive. So, on those
3035 architectures that have instructions which operate on multiple consecutive
3036 instructions, this is not an appropriate way to support them. (e.g. the 32-bit
3037 SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
3038 hardware then loads into both the named register, and the next register. This
3039 feature of inline asm would not be useful to support that.)
3041 A few of the targets provide a template string modifier allowing explicit access
3042 to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
3043 ``D``). On such an architecture, you can actually access the second allocated
3044 register (yet, still, not any subsequent ones). But, in that case, you're still
3045 probably better off simply splitting the value into two separate operands, for
3046 clarity. (e.g. see the description of the ``A`` constraint on X86, which,
3047 despite existing only for use with this feature, is not really a good idea to
3050 Indirect inputs and outputs
3051 """""""""""""""""""""""""""
3053 Indirect output or input constraints can be specified by the "``*``" modifier
3054 (which goes after the "``=``" in case of an output). This indicates that the asm
3055 will write to or read from the contents of an *address* provided as an input
3056 argument. (Note that in this way, indirect outputs act more like an *input* than
3057 an output: just like an input, they consume an argument of the call expression,
3058 rather than producing a return value. An indirect output constraint is an
3059 "output" only in that the asm is expected to write to the contents of the input
3060 memory location, instead of just read from it).
3062 This is most typically used for memory constraint, e.g. "``=*m``", to pass the
3063 address of a variable as a value.
3065 It is also possible to use an indirect *register* constraint, but only on output
3066 (e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3067 value normally, and then, separately emit a store to the address provided as
3068 input, after the provided inline asm. (It's not clear what value this
3069 functionality provides, compared to writing the store explicitly after the asm
3070 statement, and it can only produce worse code, since it bypasses many
3071 optimization passes. I would recommend not using it.)
3077 A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3078 consume an input operand, nor generate an output. Clobbers cannot use any of the
3079 general constraint code letters -- they may use only explicit register
3080 constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3081 "``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3082 memory locations -- not only the memory pointed to by a declared indirect
3088 After a potential prefix comes constraint code, or codes.
3090 A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3091 followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3094 The one and two letter constraint codes are typically chosen to be the same as
3095 GCC's constraint codes.
3097 A single constraint may include one or more than constraint code in it, leaving
3098 it up to LLVM to choose which one to use. This is included mainly for
3099 compatibility with the translation of GCC inline asm coming from clang.
3101 There are two ways to specify alternatives, and either or both may be used in an
3102 inline asm constraint list:
3104 1) Append the codes to each other, making a constraint code set. E.g. "``im``"
3105 or "``{eax}m``". This means "choose any of the options in the set". The
3106 choice of constraint is made independently for each constraint in the
3109 2) Use "``|``" between constraint code sets, creating alternatives. Every
3110 constraint in the constraint list must have the same number of alternative
3111 sets. With this syntax, the same alternative in *all* of the items in the
3112 constraint list will be chosen together.
3114 Putting those together, you might have a two operand constraint string like
3115 ``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3116 operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3117 may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3119 However, the use of either of the alternatives features is *NOT* recommended, as
3120 LLVM is not able to make an intelligent choice about which one to use. (At the
3121 point it currently needs to choose, not enough information is available to do so
3122 in a smart way.) Thus, it simply tries to make a choice that's most likely to
3123 compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3124 always choose to use memory, not registers). And, if given multiple registers,
3125 or multiple register classes, it will simply choose the first one. (In fact, it
3126 doesn't currently even ensure explicitly specified physical registers are
3127 unique, so specifying multiple physical registers as alternatives, like
3128 ``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3131 Supported Constraint Code List
3132 """"""""""""""""""""""""""""""
3134 The constraint codes are, in general, expected to behave the same way they do in
3135 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3136 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3137 and GCC likely indicates a bug in LLVM.
3139 Some constraint codes are typically supported by all targets:
3141 - ``r``: A register in the target's general purpose register class.
3142 - ``m``: A memory address operand. It is target-specific what addressing modes
3143 are supported, typical examples are register, or register + register offset,
3144 or register + immediate offset (of some target-specific size).
3145 - ``i``: An integer constant (of target-specific width). Allows either a simple
3146 immediate, or a relocatable value.
3147 - ``n``: An integer constant -- *not* including relocatable values.
3148 - ``s``: An integer constant, but allowing *only* relocatable values.
3149 - ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3150 useful to pass a label for an asm branch or call.
3152 .. FIXME: but that surely isn't actually okay to jump out of an asm
3153 block without telling llvm about the control transfer???)
3155 - ``{register-name}``: Requires exactly the named physical register.
3157 Other constraints are target-specific:
3161 - ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3162 - ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3163 i.e. 0 to 4095 with optional shift by 12.
3164 - ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3165 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3166 - ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3167 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3168 - ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3169 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3170 - ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3171 32-bit register. This is a superset of ``K``: in addition to the bitmask
3172 immediate, also allows immediate integers which can be loaded with a single
3173 ``MOVZ`` or ``MOVL`` instruction.
3174 - ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3175 64-bit register. This is a superset of ``L``.
3176 - ``Q``: Memory address operand must be in a single register (no
3177 offsets). (However, LLVM currently does this for the ``m`` constraint as
3179 - ``r``: A 32 or 64-bit integer register (W* or X*).
3180 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3181 - ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3185 - ``r``: A 32 or 64-bit integer register.
3186 - ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3187 - ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3192 - ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3193 operand. Treated the same as operand ``m``, at the moment.
3195 ARM and ARM's Thumb2 mode:
3197 - ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3198 - ``I``: An immediate integer valid for a data-processing instruction.
3199 - ``J``: An immediate integer between -4095 and 4095.
3200 - ``K``: An immediate integer whose bitwise inverse is valid for a
3201 data-processing instruction. (Can be used with template modifier "``B``" to
3202 print the inverted value).
3203 - ``L``: An immediate integer whose negation is valid for a data-processing
3204 instruction. (Can be used with template modifier "``n``" to print the negated
3206 - ``M``: A power of two or a integer between 0 and 32.
3207 - ``N``: Invalid immediate constraint.
3208 - ``O``: Invalid immediate constraint.
3209 - ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3210 - ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3212 - ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3214 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3215 ``d0-d31``, or ``q0-q15``.
3216 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3217 ``d0-d7``, or ``q0-q3``.
3218 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3223 - ``I``: An immediate integer between 0 and 255.
3224 - ``J``: An immediate integer between -255 and -1.
3225 - ``K``: An immediate integer between 0 and 255, with optional left-shift by
3227 - ``L``: An immediate integer between -7 and 7.
3228 - ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3229 - ``N``: An immediate integer between 0 and 31.
3230 - ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3231 - ``r``: A low 32-bit GPR register (``r0-r7``).
3232 - ``l``: A low 32-bit GPR register (``r0-r7``).
3233 - ``h``: A high GPR register (``r0-r7``).
3234 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3235 ``d0-d31``, or ``q0-q15``.
3236 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3237 ``d0-d7``, or ``q0-q3``.
3238 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3244 - ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3246 - ``r``: A 32 or 64-bit register.
3250 - ``r``: An 8 or 16-bit register.
3254 - ``I``: An immediate signed 16-bit integer.
3255 - ``J``: An immediate integer zero.
3256 - ``K``: An immediate unsigned 16-bit integer.
3257 - ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3258 - ``N``: An immediate integer between -65535 and -1.
3259 - ``O``: An immediate signed 15-bit integer.
3260 - ``P``: An immediate integer between 1 and 65535.
3261 - ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3262 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3263 - ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3264 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3266 - ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3267 ``sc`` instruction on the given subtarget (details vary).
3268 - ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3269 - ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
3270 (``W0-W31``). In the case of MSA registers, it is recommended to use the ``w``
3271 argument modifier for compatibility with GCC.
3272 - ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3274 - ``l``: The ``lo`` register, 32 or 64-bit.
3279 - ``b``: A 1-bit integer register.
3280 - ``c`` or ``h``: A 16-bit integer register.
3281 - ``r``: A 32-bit integer register.
3282 - ``l`` or ``N``: A 64-bit integer register.
3283 - ``f``: A 32-bit float register.
3284 - ``d``: A 64-bit float register.
3289 - ``I``: An immediate signed 16-bit integer.
3290 - ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3291 - ``K``: An immediate unsigned 16-bit integer.
3292 - ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3293 - ``M``: An immediate integer greater than 31.
3294 - ``N``: An immediate integer that is an exact power of 2.
3295 - ``O``: The immediate integer constant 0.
3296 - ``P``: An immediate integer constant whose negation is a signed 16-bit
3298 - ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3299 treated the same as ``m``.
3300 - ``r``: A 32 or 64-bit integer register.
3301 - ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3303 - ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3304 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3305 - ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3306 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3307 altivec vector register (``V0-V31``).
3309 .. FIXME: is this a bug that v accepts QPX registers? I think this
3310 is supposed to only use the altivec vector registers?
3312 - ``y``: Condition register (``CR0-CR7``).
3313 - ``wc``: An individual CR bit in a CR register.
3314 - ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3315 register set (overlapping both the floating-point and vector register files).
3316 - ``ws``: A 32 or 64-bit floating point register, from the full VSX register
3321 - ``I``: An immediate 13-bit signed integer.
3322 - ``r``: A 32-bit integer register.
3326 - ``I``: An immediate unsigned 8-bit integer.
3327 - ``J``: An immediate unsigned 12-bit integer.
3328 - ``K``: An immediate signed 16-bit integer.
3329 - ``L``: An immediate signed 20-bit integer.
3330 - ``M``: An immediate integer 0x7fffffff.
3331 - ``Q``, ``R``, ``S``, ``T``: A memory address operand, treated the same as
3332 ``m``, at the moment.
3333 - ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3334 - ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3335 address context evaluates as zero).
3336 - ``h``: A 32-bit value in the high part of a 64bit data register
3338 - ``f``: A 32, 64, or 128-bit floating point register.
3342 - ``I``: An immediate integer between 0 and 31.
3343 - ``J``: An immediate integer between 0 and 64.
3344 - ``K``: An immediate signed 8-bit integer.
3345 - ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3347 - ``M``: An immediate integer between 0 and 3.
3348 - ``N``: An immediate unsigned 8-bit integer.
3349 - ``O``: An immediate integer between 0 and 127.
3350 - ``e``: An immediate 32-bit signed integer.
3351 - ``Z``: An immediate 32-bit unsigned integer.
3352 - ``o``, ``v``: Treated the same as ``m``, at the moment.
3353 - ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3354 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3355 registers, and on X86-64, it is all of the integer registers.
3356 - ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3357 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3358 - ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3359 - ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3360 existed since i386, and can be accessed without the REX prefix.
3361 - ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3362 - ``y``: A 64-bit MMX register, if MMX is enabled.
3363 - ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3364 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3365 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3366 512-bit vector operand in an AVX512 register, Otherwise, an error.
3367 - ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3368 - ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3369 32-bit mode, a 64-bit integer operand will get split into two registers). It
3370 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3371 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3372 you're better off splitting it yourself, before passing it to the asm
3377 - ``r``: A 32-bit integer register.
3380 .. _inline-asm-modifiers:
3382 Asm template argument modifiers
3383 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3385 In the asm template string, modifiers can be used on the operand reference, like
3388 The modifiers are, in general, expected to behave the same way they do in
3389 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3390 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3391 and GCC likely indicates a bug in LLVM.
3395 - ``c``: Print an immediate integer constant unadorned, without
3396 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3397 - ``n``: Negate and print immediate integer constant unadorned, without the
3398 target-specific immediate punctuation (e.g. no ``$`` prefix).
3399 - ``l``: Print as an unadorned label, without the target-specific label
3400 punctuation (e.g. no ``$`` prefix).
3404 - ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3405 instead of ``x30``, print ``w30``.
3406 - ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3407 - ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3408 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3417 - ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3421 - ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3422 as ``d4[1]`` instead of ``s9``)
3423 - ``B``: Bitwise invert and print an immediate integer constant without ``#``
3425 - ``L``: Print the low 16-bits of an immediate integer constant.
3426 - ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3427 register operands subsequent to the specified one (!), so use carefully.
3428 - ``Q``: Print the low-order register of a register-pair, or the low-order
3429 register of a two-register operand.
3430 - ``R``: Print the high-order register of a register-pair, or the high-order
3431 register of a two-register operand.
3432 - ``H``: Print the second register of a register-pair. (On a big-endian system,
3433 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3436 .. FIXME: H doesn't currently support printing the second register
3437 of a two-register operand.
3439 - ``e``: Print the low doubleword register of a NEON quad register.
3440 - ``f``: Print the high doubleword register of a NEON quad register.
3441 - ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3446 - ``L``: Print the second register of a two-register operand. Requires that it
3447 has been allocated consecutively to the first.
3449 .. FIXME: why is it restricted to consecutive ones? And there's
3450 nothing that ensures that happens, is there?
3452 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3453 nothing. Used to print 'addi' vs 'add' instructions.
3457 No additional modifiers.
3461 - ``X``: Print an immediate integer as hexadecimal
3462 - ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3463 - ``d``: Print an immediate integer as decimal.
3464 - ``m``: Subtract one and print an immediate integer as decimal.
3465 - ``z``: Print $0 if an immediate zero, otherwise print normally.
3466 - ``L``: Print the low-order register of a two-register operand, or prints the
3467 address of the low-order word of a double-word memory operand.
3469 .. FIXME: L seems to be missing memory operand support.
3471 - ``M``: Print the high-order register of a two-register operand, or prints the
3472 address of the high-order word of a double-word memory operand.
3474 .. FIXME: M seems to be missing memory operand support.
3476 - ``D``: Print the second register of a two-register operand, or prints the
3477 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3478 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3480 - ``w``: No effect. Provided for compatibility with GCC which requires this
3481 modifier in order to print MSA registers (``W0-W31``) with the ``f``
3490 - ``L``: Print the second register of a two-register operand. Requires that it
3491 has been allocated consecutively to the first.
3493 .. FIXME: why is it restricted to consecutive ones? And there's
3494 nothing that ensures that happens, is there?
3496 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3497 nothing. Used to print 'addi' vs 'add' instructions.
3498 - ``y``: For a memory operand, prints formatter for a two-register X-form
3499 instruction. (Currently always prints ``r0,OPERAND``).
3500 - ``U``: Prints 'u' if the memory operand is an update form, and nothing
3501 otherwise. (NOTE: LLVM does not support update form, so this will currently
3502 always print nothing)
3503 - ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
3504 not support indexed form, so this will currently always print nothing)
3512 SystemZ implements only ``n``, and does *not* support any of the other
3513 target-independent modifiers.
3517 - ``c``: Print an unadorned integer or symbol name. (The latter is
3518 target-specific behavior for this typically target-independent modifier).
3519 - ``A``: Print a register name with a '``*``' before it.
3520 - ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
3522 - ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
3524 - ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
3526 - ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
3528 - ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
3529 available, otherwise the 32-bit register name; do nothing on a memory operand.
3530 - ``n``: Negate and print an unadorned integer, or, for operands other than an
3531 immediate integer (e.g. a relocatable symbol expression), print a '-' before
3532 the operand. (The behavior for relocatable symbol expressions is a
3533 target-specific behavior for this typically target-independent modifier)
3534 - ``H``: Print a memory reference with additional offset +8.
3535 - ``P``: Print a memory reference or operand for use as the argument of a call
3536 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
3540 No additional modifiers.
3546 The call instructions that wrap inline asm nodes may have a
3547 "``!srcloc``" MDNode attached to it that contains a list of constant
3548 integers. If present, the code generator will use the integer as the
3549 location cookie value when report errors through the ``LLVMContext``
3550 error reporting mechanisms. This allows a front-end to correlate backend
3551 errors that occur with inline asm back to the source code that produced
3554 .. code-block:: llvm
3556 call void asm sideeffect "something bad", ""(), !srcloc !42
3558 !42 = !{ i32 1234567 }
3560 It is up to the front-end to make sense of the magic numbers it places
3561 in the IR. If the MDNode contains multiple constants, the code generator
3562 will use the one that corresponds to the line of the asm that the error
3570 LLVM IR allows metadata to be attached to instructions in the program
3571 that can convey extra information about the code to the optimizers and
3572 code generator. One example application of metadata is source-level
3573 debug information. There are two metadata primitives: strings and nodes.
3575 Metadata does not have a type, and is not a value. If referenced from a
3576 ``call`` instruction, it uses the ``metadata`` type.
3578 All metadata are identified in syntax by a exclamation point ('``!``').
3580 .. _metadata-string:
3582 Metadata Nodes and Metadata Strings
3583 -----------------------------------
3585 A metadata string is a string surrounded by double quotes. It can
3586 contain any character by escaping non-printable characters with
3587 "``\xx``" where "``xx``" is the two digit hex code. For example:
3590 Metadata nodes are represented with notation similar to structure
3591 constants (a comma separated list of elements, surrounded by braces and
3592 preceded by an exclamation point). Metadata nodes can have any values as
3593 their operand. For example:
3595 .. code-block:: llvm
3597 !{ !"test\00", i32 10}
3599 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
3601 .. code-block:: llvm
3603 !0 = distinct !{!"test\00", i32 10}
3605 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
3606 content. They can also occur when transformations cause uniquing collisions
3607 when metadata operands change.
3609 A :ref:`named metadata <namedmetadatastructure>` is a collection of
3610 metadata nodes, which can be looked up in the module symbol table. For
3613 .. code-block:: llvm
3617 Metadata can be used as function arguments. Here ``llvm.dbg.value``
3618 function is using two metadata arguments:
3620 .. code-block:: llvm
3622 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3624 Metadata can be attached with an instruction. Here metadata ``!21`` is
3625 attached to the ``add`` instruction using the ``!dbg`` identifier:
3627 .. code-block:: llvm
3629 %indvar.next = add i64 %indvar, 1, !dbg !21
3631 More information about specific metadata nodes recognized by the
3632 optimizers and code generator is found below.
3634 .. _specialized-metadata:
3636 Specialized Metadata Nodes
3637 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3639 Specialized metadata nodes are custom data structures in metadata (as opposed
3640 to generic tuples). Their fields are labelled, and can be specified in any
3643 These aren't inherently debug info centric, but currently all the specialized
3644 metadata nodes are related to debug info.
3651 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
3652 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
3653 tuples containing the debug info to be emitted along with the compile unit,
3654 regardless of code optimizations (some nodes are only emitted if there are
3655 references to them from instructions).
3657 .. code-block:: llvm
3659 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
3660 isOptimized: true, flags: "-O2", runtimeVersion: 2,
3661 splitDebugFilename: "abc.debug", emissionKind: 1,
3662 enums: !2, retainedTypes: !3, subprograms: !4,
3663 globals: !5, imports: !6)
3665 Compile unit descriptors provide the root scope for objects declared in a
3666 specific compilation unit. File descriptors are defined using this scope.
3667 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
3668 keep track of subprograms, global variables, type information, and imported
3669 entities (declarations and namespaces).
3676 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
3678 .. code-block:: llvm
3680 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
3682 Files are sometimes used in ``scope:`` fields, and are the only valid target
3683 for ``file:`` fields.
3690 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
3691 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
3693 .. code-block:: llvm
3695 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3696 encoding: DW_ATE_unsigned_char)
3697 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
3699 The ``encoding:`` describes the details of the type. Usually it's one of the
3702 .. code-block:: llvm
3708 DW_ATE_signed_char = 6
3710 DW_ATE_unsigned_char = 8
3712 .. _DISubroutineType:
3717 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3718 refers to a tuple; the first operand is the return type, while the rest are the
3719 types of the formal arguments in order. If the first operand is ``null``, that
3720 represents a function with no return value (such as ``void foo() {}`` in C++).
3722 .. code-block:: llvm
3724 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3725 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3726 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3733 ``DIDerivedType`` nodes represent types derived from other types, such as
3736 .. code-block:: llvm
3738 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3739 encoding: DW_ATE_unsigned_char)
3740 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3743 The following ``tag:`` values are valid:
3745 .. code-block:: llvm
3747 DW_TAG_formal_parameter = 5
3749 DW_TAG_pointer_type = 15
3750 DW_TAG_reference_type = 16
3752 DW_TAG_ptr_to_member_type = 31
3753 DW_TAG_const_type = 38
3754 DW_TAG_volatile_type = 53
3755 DW_TAG_restrict_type = 55
3757 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3758 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3759 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3760 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3761 argument of a subprogram.
3763 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3765 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3766 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3769 Note that the ``void *`` type is expressed as a type derived from NULL.
3771 .. _DICompositeType:
3776 ``DICompositeType`` nodes represent types composed of other types, like
3777 structures and unions. ``elements:`` points to a tuple of the composed types.
3779 If the source language supports ODR, the ``identifier:`` field gives the unique
3780 identifier used for type merging between modules. When specified, other types
3781 can refer to composite types indirectly via a :ref:`metadata string
3782 <metadata-string>` that matches their identifier.
3784 .. code-block:: llvm
3786 !0 = !DIEnumerator(name: "SixKind", value: 7)
3787 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3788 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3789 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3790 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3791 elements: !{!0, !1, !2})
3793 The following ``tag:`` values are valid:
3795 .. code-block:: llvm
3797 DW_TAG_array_type = 1
3798 DW_TAG_class_type = 2
3799 DW_TAG_enumeration_type = 4
3800 DW_TAG_structure_type = 19
3801 DW_TAG_union_type = 23
3802 DW_TAG_subroutine_type = 21
3803 DW_TAG_inheritance = 28
3806 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3807 descriptors <DISubrange>`, each representing the range of subscripts at that
3808 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3809 array type is a native packed vector.
3811 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3812 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3813 value for the set. All enumeration type descriptors are collected in the
3814 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3816 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3817 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3818 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3825 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3826 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3828 .. code-block:: llvm
3830 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3831 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3832 !2 = !DISubrange(count: -1) ; empty array.
3839 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3840 variants of :ref:`DICompositeType`.
3842 .. code-block:: llvm
3844 !0 = !DIEnumerator(name: "SixKind", value: 7)
3845 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3846 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3848 DITemplateTypeParameter
3849 """""""""""""""""""""""
3851 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3852 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3853 :ref:`DISubprogram` ``templateParams:`` fields.
3855 .. code-block:: llvm
3857 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3859 DITemplateValueParameter
3860 """"""""""""""""""""""""
3862 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3863 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3864 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3865 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3866 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3868 .. code-block:: llvm
3870 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3875 ``DINamespace`` nodes represent namespaces in the source language.
3877 .. code-block:: llvm
3879 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3884 ``DIGlobalVariable`` nodes represent global variables in the source language.
3886 .. code-block:: llvm
3888 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3889 file: !2, line: 7, type: !3, isLocal: true,
3890 isDefinition: false, variable: i32* @foo,
3893 All global variables should be referenced by the `globals:` field of a
3894 :ref:`compile unit <DICompileUnit>`.
3901 ``DISubprogram`` nodes represent functions from the source language. The
3902 ``variables:`` field points at :ref:`variables <DILocalVariable>` that must be
3903 retained, even if their IR counterparts are optimized out of the IR. The
3904 ``type:`` field must point at an :ref:`DISubroutineType`.
3906 .. code-block:: llvm
3908 !0 = !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3909 file: !2, line: 7, type: !3, isLocal: true,
3910 isDefinition: false, scopeLine: 8, containingType: !4,
3911 virtuality: DW_VIRTUALITY_pure_virtual, virtualIndex: 10,
3912 flags: DIFlagPrototyped, isOptimized: true,
3913 function: void ()* @_Z3foov,
3914 templateParams: !5, declaration: !6, variables: !7)
3921 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3922 <DISubprogram>`. The line number and column numbers are used to distinguish
3923 two lexical blocks at same depth. They are valid targets for ``scope:``
3926 .. code-block:: llvm
3928 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3930 Usually lexical blocks are ``distinct`` to prevent node merging based on
3933 .. _DILexicalBlockFile:
3938 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3939 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
3940 indicate textual inclusion, or the ``discriminator:`` field can be used to
3941 discriminate between control flow within a single block in the source language.
3943 .. code-block:: llvm
3945 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3946 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3947 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3954 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
3955 mandatory, and points at an :ref:`DILexicalBlockFile`, an
3956 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
3958 .. code-block:: llvm
3960 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3962 .. _DILocalVariable:
3967 ``DILocalVariable`` nodes represent local variables in the source language. If
3968 the ``arg:`` field is set to non-zero, then this variable is a subprogram
3969 parameter, and it will be included in the ``variables:`` field of its
3970 :ref:`DISubprogram`.
3972 .. code-block:: llvm
3974 !0 = !DILocalVariable(name: "this", arg: 1, scope: !3, file: !2, line: 7,
3975 type: !3, flags: DIFlagArtificial)
3976 !1 = !DILocalVariable(name: "x", arg: 2, scope: !4, file: !2, line: 7,
3978 !2 = !DILocalVariable(name: "y", scope: !5, file: !2, line: 7, type: !3)
3983 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
3984 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
3985 describe how the referenced LLVM variable relates to the source language
3988 The current supported vocabulary is limited:
3990 - ``DW_OP_deref`` dereferences the working expression.
3991 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
3992 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
3993 here, respectively) of the variable piece from the working expression.
3995 .. code-block:: llvm
3997 !0 = !DIExpression(DW_OP_deref)
3998 !1 = !DIExpression(DW_OP_plus, 3)
3999 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
4000 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
4005 ``DIObjCProperty`` nodes represent Objective-C property nodes.
4007 .. code-block:: llvm
4009 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
4010 getter: "getFoo", attributes: 7, type: !2)
4015 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
4018 .. code-block:: llvm
4020 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
4021 entity: !1, line: 7)
4026 In LLVM IR, memory does not have types, so LLVM's own type system is not
4027 suitable for doing TBAA. Instead, metadata is added to the IR to
4028 describe a type system of a higher level language. This can be used to
4029 implement typical C/C++ TBAA, but it can also be used to implement
4030 custom alias analysis behavior for other languages.
4032 The current metadata format is very simple. TBAA metadata nodes have up
4033 to three fields, e.g.:
4035 .. code-block:: llvm
4037 !0 = !{ !"an example type tree" }
4038 !1 = !{ !"int", !0 }
4039 !2 = !{ !"float", !0 }
4040 !3 = !{ !"const float", !2, i64 1 }
4042 The first field is an identity field. It can be any value, usually a
4043 metadata string, which uniquely identifies the type. The most important
4044 name in the tree is the name of the root node. Two trees with different
4045 root node names are entirely disjoint, even if they have leaves with
4048 The second field identifies the type's parent node in the tree, or is
4049 null or omitted for a root node. A type is considered to alias all of
4050 its descendants and all of its ancestors in the tree. Also, a type is
4051 considered to alias all types in other trees, so that bitcode produced
4052 from multiple front-ends is handled conservatively.
4054 If the third field is present, it's an integer which if equal to 1
4055 indicates that the type is "constant" (meaning
4056 ``pointsToConstantMemory`` should return true; see `other useful
4057 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
4059 '``tbaa.struct``' Metadata
4060 ^^^^^^^^^^^^^^^^^^^^^^^^^^
4062 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
4063 aggregate assignment operations in C and similar languages, however it
4064 is defined to copy a contiguous region of memory, which is more than
4065 strictly necessary for aggregate types which contain holes due to
4066 padding. Also, it doesn't contain any TBAA information about the fields
4069 ``!tbaa.struct`` metadata can describe which memory subregions in a
4070 memcpy are padding and what the TBAA tags of the struct are.
4072 The current metadata format is very simple. ``!tbaa.struct`` metadata
4073 nodes are a list of operands which are in conceptual groups of three.
4074 For each group of three, the first operand gives the byte offset of a
4075 field in bytes, the second gives its size in bytes, and the third gives
4078 .. code-block:: llvm
4080 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
4082 This describes a struct with two fields. The first is at offset 0 bytes
4083 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4084 and has size 4 bytes and has tbaa tag !2.
4086 Note that the fields need not be contiguous. In this example, there is a
4087 4 byte gap between the two fields. This gap represents padding which
4088 does not carry useful data and need not be preserved.
4090 '``noalias``' and '``alias.scope``' Metadata
4091 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4093 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4094 noalias memory-access sets. This means that some collection of memory access
4095 instructions (loads, stores, memory-accessing calls, etc.) that carry
4096 ``noalias`` metadata can specifically be specified not to alias with some other
4097 collection of memory access instructions that carry ``alias.scope`` metadata.
4098 Each type of metadata specifies a list of scopes where each scope has an id and
4099 a domain. When evaluating an aliasing query, if for some domain, the set
4100 of scopes with that domain in one instruction's ``alias.scope`` list is a
4101 subset of (or equal to) the set of scopes for that domain in another
4102 instruction's ``noalias`` list, then the two memory accesses are assumed not to
4105 The metadata identifying each domain is itself a list containing one or two
4106 entries. The first entry is the name of the domain. Note that if the name is a
4107 string then it can be combined across functions and translation units. A
4108 self-reference can be used to create globally unique domain names. A
4109 descriptive string may optionally be provided as a second list entry.
4111 The metadata identifying each scope is also itself a list containing two or
4112 three entries. The first entry is the name of the scope. Note that if the name
4113 is a string then it can be combined across functions and translation units. A
4114 self-reference can be used to create globally unique scope names. A metadata
4115 reference to the scope's domain is the second entry. A descriptive string may
4116 optionally be provided as a third list entry.
4120 .. code-block:: llvm
4122 ; Two scope domains:
4126 ; Some scopes in these domains:
4132 !5 = !{!4} ; A list containing only scope !4
4136 ; These two instructions don't alias:
4137 %0 = load float, float* %c, align 4, !alias.scope !5
4138 store float %0, float* %arrayidx.i, align 4, !noalias !5
4140 ; These two instructions also don't alias (for domain !1, the set of scopes
4141 ; in the !alias.scope equals that in the !noalias list):
4142 %2 = load float, float* %c, align 4, !alias.scope !5
4143 store float %2, float* %arrayidx.i2, align 4, !noalias !6
4145 ; These two instructions may alias (for domain !0, the set of scopes in
4146 ; the !noalias list is not a superset of, or equal to, the scopes in the
4147 ; !alias.scope list):
4148 %2 = load float, float* %c, align 4, !alias.scope !6
4149 store float %0, float* %arrayidx.i, align 4, !noalias !7
4151 '``fpmath``' Metadata
4152 ^^^^^^^^^^^^^^^^^^^^^
4154 ``fpmath`` metadata may be attached to any instruction of floating point
4155 type. It can be used to express the maximum acceptable error in the
4156 result of that instruction, in ULPs, thus potentially allowing the
4157 compiler to use a more efficient but less accurate method of computing
4158 it. ULP is defined as follows:
4160 If ``x`` is a real number that lies between two finite consecutive
4161 floating-point numbers ``a`` and ``b``, without being equal to one
4162 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4163 distance between the two non-equal finite floating-point numbers
4164 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4166 The metadata node shall consist of a single positive floating point
4167 number representing the maximum relative error, for example:
4169 .. code-block:: llvm
4171 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
4175 '``range``' Metadata
4176 ^^^^^^^^^^^^^^^^^^^^
4178 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4179 integer types. It expresses the possible ranges the loaded value or the value
4180 returned by the called function at this call site is in. The ranges are
4181 represented with a flattened list of integers. The loaded value or the value
4182 returned is known to be in the union of the ranges defined by each consecutive
4183 pair. Each pair has the following properties:
4185 - The type must match the type loaded by the instruction.
4186 - The pair ``a,b`` represents the range ``[a,b)``.
4187 - Both ``a`` and ``b`` are constants.
4188 - The range is allowed to wrap.
4189 - The range should not represent the full or empty set. That is,
4192 In addition, the pairs must be in signed order of the lower bound and
4193 they must be non-contiguous.
4197 .. code-block:: llvm
4199 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4200 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
4201 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4202 %d = invoke i8 @bar() to label %cont
4203 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
4205 !0 = !{ i8 0, i8 2 }
4206 !1 = !{ i8 255, i8 2 }
4207 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4208 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
4210 '``unpredictable``' Metadata
4211 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4213 ``unpredictable`` metadata may be attached to any branch or switch
4214 instruction. It can be used to express the unpredictability of control
4215 flow. Similar to the llvm.expect intrinsic, it may be used to alter
4216 optimizations related to compare and branch instructions. The metadata
4217 is treated as a boolean value; if it exists, it signals that the branch
4218 or switch that it is attached to is completely unpredictable.
4223 It is sometimes useful to attach information to loop constructs. Currently,
4224 loop metadata is implemented as metadata attached to the branch instruction
4225 in the loop latch block. This type of metadata refer to a metadata node that is
4226 guaranteed to be separate for each loop. The loop identifier metadata is
4227 specified with the name ``llvm.loop``.
4229 The loop identifier metadata is implemented using a metadata that refers to
4230 itself to avoid merging it with any other identifier metadata, e.g.,
4231 during module linkage or function inlining. That is, each loop should refer
4232 to their own identification metadata even if they reside in separate functions.
4233 The following example contains loop identifier metadata for two separate loop
4236 .. code-block:: llvm
4241 The loop identifier metadata can be used to specify additional
4242 per-loop metadata. Any operands after the first operand can be treated
4243 as user-defined metadata. For example the ``llvm.loop.unroll.count``
4244 suggests an unroll factor to the loop unroller:
4246 .. code-block:: llvm
4248 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
4251 !1 = !{!"llvm.loop.unroll.count", i32 4}
4253 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
4254 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4256 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
4257 used to control per-loop vectorization and interleaving parameters such as
4258 vectorization width and interleave count. These metadata should be used in
4259 conjunction with ``llvm.loop`` loop identification metadata. The
4260 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
4261 optimization hints and the optimizer will only interleave and vectorize loops if
4262 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
4263 which contains information about loop-carried memory dependencies can be helpful
4264 in determining the safety of these transformations.
4266 '``llvm.loop.interleave.count``' Metadata
4267 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4269 This metadata suggests an interleave count to the loop interleaver.
4270 The first operand is the string ``llvm.loop.interleave.count`` and the
4271 second operand is an integer specifying the interleave count. For
4274 .. code-block:: llvm
4276 !0 = !{!"llvm.loop.interleave.count", i32 4}
4278 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
4279 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
4280 then the interleave count will be determined automatically.
4282 '``llvm.loop.vectorize.enable``' Metadata
4283 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4285 This metadata selectively enables or disables vectorization for the loop. The
4286 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
4287 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
4288 0 disables vectorization:
4290 .. code-block:: llvm
4292 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
4293 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
4295 '``llvm.loop.vectorize.width``' Metadata
4296 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4298 This metadata sets the target width of the vectorizer. The first
4299 operand is the string ``llvm.loop.vectorize.width`` and the second
4300 operand is an integer specifying the width. For example:
4302 .. code-block:: llvm
4304 !0 = !{!"llvm.loop.vectorize.width", i32 4}
4306 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
4307 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
4308 0 or if the loop does not have this metadata the width will be
4309 determined automatically.
4311 '``llvm.loop.unroll``'
4312 ^^^^^^^^^^^^^^^^^^^^^^
4314 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
4315 optimization hints such as the unroll factor. ``llvm.loop.unroll``
4316 metadata should be used in conjunction with ``llvm.loop`` loop
4317 identification metadata. The ``llvm.loop.unroll`` metadata are only
4318 optimization hints and the unrolling will only be performed if the
4319 optimizer believes it is safe to do so.
4321 '``llvm.loop.unroll.count``' Metadata
4322 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4324 This metadata suggests an unroll factor to the loop unroller. The
4325 first operand is the string ``llvm.loop.unroll.count`` and the second
4326 operand is a positive integer specifying the unroll factor. For
4329 .. code-block:: llvm
4331 !0 = !{!"llvm.loop.unroll.count", i32 4}
4333 If the trip count of the loop is less than the unroll count the loop
4334 will be partially unrolled.
4336 '``llvm.loop.unroll.disable``' Metadata
4337 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4339 This metadata disables loop unrolling. The metadata has a single operand
4340 which is the string ``llvm.loop.unroll.disable``. For example:
4342 .. code-block:: llvm
4344 !0 = !{!"llvm.loop.unroll.disable"}
4346 '``llvm.loop.unroll.runtime.disable``' Metadata
4347 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4349 This metadata disables runtime loop unrolling. The metadata has a single
4350 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
4352 .. code-block:: llvm
4354 !0 = !{!"llvm.loop.unroll.runtime.disable"}
4356 '``llvm.loop.unroll.enable``' Metadata
4357 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4359 This metadata suggests that the loop should be fully unrolled if the trip count
4360 is known at compile time and partially unrolled if the trip count is not known
4361 at compile time. The metadata has a single operand which is the string
4362 ``llvm.loop.unroll.enable``. For example:
4364 .. code-block:: llvm
4366 !0 = !{!"llvm.loop.unroll.enable"}
4368 '``llvm.loop.unroll.full``' Metadata
4369 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4371 This metadata suggests that the loop should be unrolled fully. The
4372 metadata has a single operand which is the string ``llvm.loop.unroll.full``.
4375 .. code-block:: llvm
4377 !0 = !{!"llvm.loop.unroll.full"}
4382 Metadata types used to annotate memory accesses with information helpful
4383 for optimizations are prefixed with ``llvm.mem``.
4385 '``llvm.mem.parallel_loop_access``' Metadata
4386 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4388 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
4389 or metadata containing a list of loop identifiers for nested loops.
4390 The metadata is attached to memory accessing instructions and denotes that
4391 no loop carried memory dependence exist between it and other instructions denoted
4392 with the same loop identifier.
4394 Precisely, given two instructions ``m1`` and ``m2`` that both have the
4395 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
4396 set of loops associated with that metadata, respectively, then there is no loop
4397 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
4400 As a special case, if all memory accessing instructions in a loop have
4401 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
4402 loop has no loop carried memory dependences and is considered to be a parallel
4405 Note that if not all memory access instructions have such metadata referring to
4406 the loop, then the loop is considered not being trivially parallel. Additional
4407 memory dependence analysis is required to make that determination. As a fail
4408 safe mechanism, this causes loops that were originally parallel to be considered
4409 sequential (if optimization passes that are unaware of the parallel semantics
4410 insert new memory instructions into the loop body).
4412 Example of a loop that is considered parallel due to its correct use of
4413 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
4414 metadata types that refer to the same loop identifier metadata.
4416 .. code-block:: llvm
4420 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
4422 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4424 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
4430 It is also possible to have nested parallel loops. In that case the
4431 memory accesses refer to a list of loop identifier metadata nodes instead of
4432 the loop identifier metadata node directly:
4434 .. code-block:: llvm
4438 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
4440 br label %inner.for.body
4444 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4446 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
4448 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
4452 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
4454 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
4456 outer.for.end: ; preds = %for.body
4458 !0 = !{!1, !2} ; a list of loop identifiers
4459 !1 = !{!1} ; an identifier for the inner loop
4460 !2 = !{!2} ; an identifier for the outer loop
4465 The ``llvm.bitsets`` global metadata is used to implement
4466 :doc:`bitsets <BitSets>`.
4468 '``invariant.group``' Metadata
4469 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4471 The ``invariant.group`` metadata may be attached to ``load``/``store`` instructions.
4472 The existence of the ``invariant.group`` metadata on the instruction tells
4473 the optimizer that every ``load`` and ``store`` to the same pointer operand
4474 within the same invariant group can be assumed to load or store the same
4475 value (but see the ``llvm.invariant.group.barrier`` intrinsic which affects
4476 when two pointers are considered the same).
4480 .. code-block:: llvm
4482 @unknownPtr = external global i8
4485 store i8 42, i8* %ptr, !invariant.group !0
4486 call void @foo(i8* %ptr)
4488 %a = load i8, i8* %ptr, !invariant.group !0 ; Can assume that value under %ptr didn't change
4489 call void @foo(i8* %ptr)
4490 %b = load i8, i8* %ptr, !invariant.group !1 ; Can't assume anything, because group changed
4492 %newPtr = call i8* @getPointer(i8* %ptr)
4493 %c = load i8, i8* %newPtr, !invariant.group !0 ; Can't assume anything, because we only have information about %ptr
4495 %unknownValue = load i8, i8* @unknownPtr
4496 store i8 %unknownValue, i8* %ptr, !invariant.group !0 ; Can assume that %unknownValue == 42
4498 call void @foo(i8* %ptr)
4499 %newPtr2 = call i8* @llvm.invariant.group.barrier(i8* %ptr)
4500 %d = load i8, i8* %newPtr2, !invariant.group !0 ; Can't step through invariant.group.barrier to get value of %ptr
4503 declare void @foo(i8*)
4504 declare i8* @getPointer(i8*)
4505 declare i8* @llvm.invariant.group.barrier(i8*)
4507 !0 = !{!"magic ptr"}
4508 !1 = !{!"other ptr"}
4512 Module Flags Metadata
4513 =====================
4515 Information about the module as a whole is difficult to convey to LLVM's
4516 subsystems. The LLVM IR isn't sufficient to transmit this information.
4517 The ``llvm.module.flags`` named metadata exists in order to facilitate
4518 this. These flags are in the form of key / value pairs --- much like a
4519 dictionary --- making it easy for any subsystem who cares about a flag to
4522 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
4523 Each triplet has the following form:
4525 - The first element is a *behavior* flag, which specifies the behavior
4526 when two (or more) modules are merged together, and it encounters two
4527 (or more) metadata with the same ID. The supported behaviors are
4529 - The second element is a metadata string that is a unique ID for the
4530 metadata. Each module may only have one flag entry for each unique ID (not
4531 including entries with the **Require** behavior).
4532 - The third element is the value of the flag.
4534 When two (or more) modules are merged together, the resulting
4535 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
4536 each unique metadata ID string, there will be exactly one entry in the merged
4537 modules ``llvm.module.flags`` metadata table, and the value for that entry will
4538 be determined by the merge behavior flag, as described below. The only exception
4539 is that entries with the *Require* behavior are always preserved.
4541 The following behaviors are supported:
4552 Emits an error if two values disagree, otherwise the resulting value
4553 is that of the operands.
4557 Emits a warning if two values disagree. The result value will be the
4558 operand for the flag from the first module being linked.
4562 Adds a requirement that another module flag be present and have a
4563 specified value after linking is performed. The value must be a
4564 metadata pair, where the first element of the pair is the ID of the
4565 module flag to be restricted, and the second element of the pair is
4566 the value the module flag should be restricted to. This behavior can
4567 be used to restrict the allowable results (via triggering of an
4568 error) of linking IDs with the **Override** behavior.
4572 Uses the specified value, regardless of the behavior or value of the
4573 other module. If both modules specify **Override**, but the values
4574 differ, an error will be emitted.
4578 Appends the two values, which are required to be metadata nodes.
4582 Appends the two values, which are required to be metadata
4583 nodes. However, duplicate entries in the second list are dropped
4584 during the append operation.
4586 It is an error for a particular unique flag ID to have multiple behaviors,
4587 except in the case of **Require** (which adds restrictions on another metadata
4588 value) or **Override**.
4590 An example of module flags:
4592 .. code-block:: llvm
4594 !0 = !{ i32 1, !"foo", i32 1 }
4595 !1 = !{ i32 4, !"bar", i32 37 }
4596 !2 = !{ i32 2, !"qux", i32 42 }
4597 !3 = !{ i32 3, !"qux",
4602 !llvm.module.flags = !{ !0, !1, !2, !3 }
4604 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
4605 if two or more ``!"foo"`` flags are seen is to emit an error if their
4606 values are not equal.
4608 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
4609 behavior if two or more ``!"bar"`` flags are seen is to use the value
4612 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
4613 behavior if two or more ``!"qux"`` flags are seen is to emit a
4614 warning if their values are not equal.
4616 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
4622 The behavior is to emit an error if the ``llvm.module.flags`` does not
4623 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
4626 Objective-C Garbage Collection Module Flags Metadata
4627 ----------------------------------------------------
4629 On the Mach-O platform, Objective-C stores metadata about garbage
4630 collection in a special section called "image info". The metadata
4631 consists of a version number and a bitmask specifying what types of
4632 garbage collection are supported (if any) by the file. If two or more
4633 modules are linked together their garbage collection metadata needs to
4634 be merged rather than appended together.
4636 The Objective-C garbage collection module flags metadata consists of the
4637 following key-value pairs:
4646 * - ``Objective-C Version``
4647 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
4649 * - ``Objective-C Image Info Version``
4650 - **[Required]** --- The version of the image info section. Currently
4653 * - ``Objective-C Image Info Section``
4654 - **[Required]** --- The section to place the metadata. Valid values are
4655 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
4656 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
4657 Objective-C ABI version 2.
4659 * - ``Objective-C Garbage Collection``
4660 - **[Required]** --- Specifies whether garbage collection is supported or
4661 not. Valid values are 0, for no garbage collection, and 2, for garbage
4662 collection supported.
4664 * - ``Objective-C GC Only``
4665 - **[Optional]** --- Specifies that only garbage collection is supported.
4666 If present, its value must be 6. This flag requires that the
4667 ``Objective-C Garbage Collection`` flag have the value 2.
4669 Some important flag interactions:
4671 - If a module with ``Objective-C Garbage Collection`` set to 0 is
4672 merged with a module with ``Objective-C Garbage Collection`` set to
4673 2, then the resulting module has the
4674 ``Objective-C Garbage Collection`` flag set to 0.
4675 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
4676 merged with a module with ``Objective-C GC Only`` set to 6.
4678 Automatic Linker Flags Module Flags Metadata
4679 --------------------------------------------
4681 Some targets support embedding flags to the linker inside individual object
4682 files. Typically this is used in conjunction with language extensions which
4683 allow source files to explicitly declare the libraries they depend on, and have
4684 these automatically be transmitted to the linker via object files.
4686 These flags are encoded in the IR using metadata in the module flags section,
4687 using the ``Linker Options`` key. The merge behavior for this flag is required
4688 to be ``AppendUnique``, and the value for the key is expected to be a metadata
4689 node which should be a list of other metadata nodes, each of which should be a
4690 list of metadata strings defining linker options.
4692 For example, the following metadata section specifies two separate sets of
4693 linker options, presumably to link against ``libz`` and the ``Cocoa``
4696 !0 = !{ i32 6, !"Linker Options",
4699 !{ !"-framework", !"Cocoa" } } }
4700 !llvm.module.flags = !{ !0 }
4702 The metadata encoding as lists of lists of options, as opposed to a collapsed
4703 list of options, is chosen so that the IR encoding can use multiple option
4704 strings to specify e.g., a single library, while still having that specifier be
4705 preserved as an atomic element that can be recognized by a target specific
4706 assembly writer or object file emitter.
4708 Each individual option is required to be either a valid option for the target's
4709 linker, or an option that is reserved by the target specific assembly writer or
4710 object file emitter. No other aspect of these options is defined by the IR.
4712 C type width Module Flags Metadata
4713 ----------------------------------
4715 The ARM backend emits a section into each generated object file describing the
4716 options that it was compiled with (in a compiler-independent way) to prevent
4717 linking incompatible objects, and to allow automatic library selection. Some
4718 of these options are not visible at the IR level, namely wchar_t width and enum
4721 To pass this information to the backend, these options are encoded in module
4722 flags metadata, using the following key-value pairs:
4732 - * 0 --- sizeof(wchar_t) == 4
4733 * 1 --- sizeof(wchar_t) == 2
4736 - * 0 --- Enums are at least as large as an ``int``.
4737 * 1 --- Enums are stored in the smallest integer type which can
4738 represent all of its values.
4740 For example, the following metadata section specifies that the module was
4741 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
4742 enum is the smallest type which can represent all of its values::
4744 !llvm.module.flags = !{!0, !1}
4745 !0 = !{i32 1, !"short_wchar", i32 1}
4746 !1 = !{i32 1, !"short_enum", i32 0}
4748 .. _intrinsicglobalvariables:
4750 Intrinsic Global Variables
4751 ==========================
4753 LLVM has a number of "magic" global variables that contain data that
4754 affect code generation or other IR semantics. These are documented here.
4755 All globals of this sort should have a section specified as
4756 "``llvm.metadata``". This section and all globals that start with
4757 "``llvm.``" are reserved for use by LLVM.
4761 The '``llvm.used``' Global Variable
4762 -----------------------------------
4764 The ``@llvm.used`` global is an array which has
4765 :ref:`appending linkage <linkage_appending>`. This array contains a list of
4766 pointers to named global variables, functions and aliases which may optionally
4767 have a pointer cast formed of bitcast or getelementptr. For example, a legal
4770 .. code-block:: llvm
4775 @llvm.used = appending global [2 x i8*] [
4777 i8* bitcast (i32* @Y to i8*)
4778 ], section "llvm.metadata"
4780 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4781 and linker are required to treat the symbol as if there is a reference to the
4782 symbol that it cannot see (which is why they have to be named). For example, if
4783 a variable has internal linkage and no references other than that from the
4784 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4785 references from inline asms and other things the compiler cannot "see", and
4786 corresponds to "``attribute((used))``" in GNU C.
4788 On some targets, the code generator must emit a directive to the
4789 assembler or object file to prevent the assembler and linker from
4790 molesting the symbol.
4792 .. _gv_llvmcompilerused:
4794 The '``llvm.compiler.used``' Global Variable
4795 --------------------------------------------
4797 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4798 directive, except that it only prevents the compiler from touching the
4799 symbol. On targets that support it, this allows an intelligent linker to
4800 optimize references to the symbol without being impeded as it would be
4803 This is a rare construct that should only be used in rare circumstances,
4804 and should not be exposed to source languages.
4806 .. _gv_llvmglobalctors:
4808 The '``llvm.global_ctors``' Global Variable
4809 -------------------------------------------
4811 .. code-block:: llvm
4813 %0 = type { i32, void ()*, i8* }
4814 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4816 The ``@llvm.global_ctors`` array contains a list of constructor
4817 functions, priorities, and an optional associated global or function.
4818 The functions referenced by this array will be called in ascending order
4819 of priority (i.e. lowest first) when the module is loaded. The order of
4820 functions with the same priority is not defined.
4822 If the third field is present, non-null, and points to a global variable
4823 or function, the initializer function will only run if the associated
4824 data from the current module is not discarded.
4826 .. _llvmglobaldtors:
4828 The '``llvm.global_dtors``' Global Variable
4829 -------------------------------------------
4831 .. code-block:: llvm
4833 %0 = type { i32, void ()*, i8* }
4834 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4836 The ``@llvm.global_dtors`` array contains a list of destructor
4837 functions, priorities, and an optional associated global or function.
4838 The functions referenced by this array will be called in descending
4839 order of priority (i.e. highest first) when the module is unloaded. The
4840 order of functions with the same priority is not defined.
4842 If the third field is present, non-null, and points to a global variable
4843 or function, the destructor function will only run if the associated
4844 data from the current module is not discarded.
4846 Instruction Reference
4847 =====================
4849 The LLVM instruction set consists of several different classifications
4850 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4851 instructions <binaryops>`, :ref:`bitwise binary
4852 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4853 :ref:`other instructions <otherops>`.
4857 Terminator Instructions
4858 -----------------------
4860 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4861 program ends with a "Terminator" instruction, which indicates which
4862 block should be executed after the current block is finished. These
4863 terminator instructions typically yield a '``void``' value: they produce
4864 control flow, not values (the one exception being the
4865 ':ref:`invoke <i_invoke>`' instruction).
4867 The terminator instructions are: ':ref:`ret <i_ret>`',
4868 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4869 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4870 ':ref:`resume <i_resume>`', ':ref:`catchpad <i_catchpad>`',
4871 ':ref:`catchendpad <i_catchendpad>`',
4872 ':ref:`catchret <i_catchret>`',
4873 ':ref:`cleanupendpad <i_cleanupendpad>`',
4874 ':ref:`cleanupret <i_cleanupret>`',
4875 ':ref:`terminatepad <i_terminatepad>`',
4876 and ':ref:`unreachable <i_unreachable>`'.
4880 '``ret``' Instruction
4881 ^^^^^^^^^^^^^^^^^^^^^
4888 ret <type> <value> ; Return a value from a non-void function
4889 ret void ; Return from void function
4894 The '``ret``' instruction is used to return control flow (and optionally
4895 a value) from a function back to the caller.
4897 There are two forms of the '``ret``' instruction: one that returns a
4898 value and then causes control flow, and one that just causes control
4904 The '``ret``' instruction optionally accepts a single argument, the
4905 return value. The type of the return value must be a ':ref:`first
4906 class <t_firstclass>`' type.
4908 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4909 return type and contains a '``ret``' instruction with no return value or
4910 a return value with a type that does not match its type, or if it has a
4911 void return type and contains a '``ret``' instruction with a return
4917 When the '``ret``' instruction is executed, control flow returns back to
4918 the calling function's context. If the caller is a
4919 ":ref:`call <i_call>`" instruction, execution continues at the
4920 instruction after the call. If the caller was an
4921 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4922 beginning of the "normal" destination block. If the instruction returns
4923 a value, that value shall set the call or invoke instruction's return
4929 .. code-block:: llvm
4931 ret i32 5 ; Return an integer value of 5
4932 ret void ; Return from a void function
4933 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4937 '``br``' Instruction
4938 ^^^^^^^^^^^^^^^^^^^^
4945 br i1 <cond>, label <iftrue>, label <iffalse>
4946 br label <dest> ; Unconditional branch
4951 The '``br``' instruction is used to cause control flow to transfer to a
4952 different basic block in the current function. There are two forms of
4953 this instruction, corresponding to a conditional branch and an
4954 unconditional branch.
4959 The conditional branch form of the '``br``' instruction takes a single
4960 '``i1``' value and two '``label``' values. The unconditional form of the
4961 '``br``' instruction takes a single '``label``' value as a target.
4966 Upon execution of a conditional '``br``' instruction, the '``i1``'
4967 argument is evaluated. If the value is ``true``, control flows to the
4968 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4969 to the '``iffalse``' ``label`` argument.
4974 .. code-block:: llvm
4977 %cond = icmp eq i32 %a, %b
4978 br i1 %cond, label %IfEqual, label %IfUnequal
4986 '``switch``' Instruction
4987 ^^^^^^^^^^^^^^^^^^^^^^^^
4994 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
4999 The '``switch``' instruction is used to transfer control flow to one of
5000 several different places. It is a generalization of the '``br``'
5001 instruction, allowing a branch to occur to one of many possible
5007 The '``switch``' instruction uses three parameters: an integer
5008 comparison value '``value``', a default '``label``' destination, and an
5009 array of pairs of comparison value constants and '``label``'s. The table
5010 is not allowed to contain duplicate constant entries.
5015 The ``switch`` instruction specifies a table of values and destinations.
5016 When the '``switch``' instruction is executed, this table is searched
5017 for the given value. If the value is found, control flow is transferred
5018 to the corresponding destination; otherwise, control flow is transferred
5019 to the default destination.
5024 Depending on properties of the target machine and the particular
5025 ``switch`` instruction, this instruction may be code generated in
5026 different ways. For example, it could be generated as a series of
5027 chained conditional branches or with a lookup table.
5032 .. code-block:: llvm
5034 ; Emulate a conditional br instruction
5035 %Val = zext i1 %value to i32
5036 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
5038 ; Emulate an unconditional br instruction
5039 switch i32 0, label %dest [ ]
5041 ; Implement a jump table:
5042 switch i32 %val, label %otherwise [ i32 0, label %onzero
5044 i32 2, label %ontwo ]
5048 '``indirectbr``' Instruction
5049 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5056 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
5061 The '``indirectbr``' instruction implements an indirect branch to a
5062 label within the current function, whose address is specified by
5063 "``address``". Address must be derived from a
5064 :ref:`blockaddress <blockaddress>` constant.
5069 The '``address``' argument is the address of the label to jump to. The
5070 rest of the arguments indicate the full set of possible destinations
5071 that the address may point to. Blocks are allowed to occur multiple
5072 times in the destination list, though this isn't particularly useful.
5074 This destination list is required so that dataflow analysis has an
5075 accurate understanding of the CFG.
5080 Control transfers to the block specified in the address argument. All
5081 possible destination blocks must be listed in the label list, otherwise
5082 this instruction has undefined behavior. This implies that jumps to
5083 labels defined in other functions have undefined behavior as well.
5088 This is typically implemented with a jump through a register.
5093 .. code-block:: llvm
5095 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
5099 '``invoke``' Instruction
5100 ^^^^^^^^^^^^^^^^^^^^^^^^
5107 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
5108 [operand bundles] to label <normal label> unwind label <exception label>
5113 The '``invoke``' instruction causes control to transfer to a specified
5114 function, with the possibility of control flow transfer to either the
5115 '``normal``' label or the '``exception``' label. If the callee function
5116 returns with the "``ret``" instruction, control flow will return to the
5117 "normal" label. If the callee (or any indirect callees) returns via the
5118 ":ref:`resume <i_resume>`" instruction or other exception handling
5119 mechanism, control is interrupted and continued at the dynamically
5120 nearest "exception" label.
5122 The '``exception``' label is a `landing
5123 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
5124 '``exception``' label is required to have the
5125 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
5126 information about the behavior of the program after unwinding happens,
5127 as its first non-PHI instruction. The restrictions on the
5128 "``landingpad``" instruction's tightly couples it to the "``invoke``"
5129 instruction, so that the important information contained within the
5130 "``landingpad``" instruction can't be lost through normal code motion.
5135 This instruction requires several arguments:
5137 #. The optional "cconv" marker indicates which :ref:`calling
5138 convention <callingconv>` the call should use. If none is
5139 specified, the call defaults to using C calling conventions.
5140 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5141 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5143 #. '``ptr to function ty``': shall be the signature of the pointer to
5144 function value being invoked. In most cases, this is a direct
5145 function invocation, but indirect ``invoke``'s are just as possible,
5146 branching off an arbitrary pointer to function value.
5147 #. '``function ptr val``': An LLVM value containing a pointer to a
5148 function to be invoked.
5149 #. '``function args``': argument list whose types match the function
5150 signature argument types and parameter attributes. All arguments must
5151 be of :ref:`first class <t_firstclass>` type. If the function signature
5152 indicates the function accepts a variable number of arguments, the
5153 extra arguments can be specified.
5154 #. '``normal label``': the label reached when the called function
5155 executes a '``ret``' instruction.
5156 #. '``exception label``': the label reached when a callee returns via
5157 the :ref:`resume <i_resume>` instruction or other exception handling
5159 #. The optional :ref:`function attributes <fnattrs>` list. Only
5160 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5161 attributes are valid here.
5162 #. The optional :ref:`operand bundles <opbundles>` list.
5167 This instruction is designed to operate as a standard '``call``'
5168 instruction in most regards. The primary difference is that it
5169 establishes an association with a label, which is used by the runtime
5170 library to unwind the stack.
5172 This instruction is used in languages with destructors to ensure that
5173 proper cleanup is performed in the case of either a ``longjmp`` or a
5174 thrown exception. Additionally, this is important for implementation of
5175 '``catch``' clauses in high-level languages that support them.
5177 For the purposes of the SSA form, the definition of the value returned
5178 by the '``invoke``' instruction is deemed to occur on the edge from the
5179 current block to the "normal" label. If the callee unwinds then no
5180 return value is available.
5185 .. code-block:: llvm
5187 %retval = invoke i32 @Test(i32 15) to label %Continue
5188 unwind label %TestCleanup ; i32:retval set
5189 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
5190 unwind label %TestCleanup ; i32:retval set
5194 '``resume``' Instruction
5195 ^^^^^^^^^^^^^^^^^^^^^^^^
5202 resume <type> <value>
5207 The '``resume``' instruction is a terminator instruction that has no
5213 The '``resume``' instruction requires one argument, which must have the
5214 same type as the result of any '``landingpad``' instruction in the same
5220 The '``resume``' instruction resumes propagation of an existing
5221 (in-flight) exception whose unwinding was interrupted with a
5222 :ref:`landingpad <i_landingpad>` instruction.
5227 .. code-block:: llvm
5229 resume { i8*, i32 } %exn
5233 '``catchpad``' Instruction
5234 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5241 <resultval> = catchpad [<args>*]
5242 to label <normal label> unwind label <exception label>
5247 The '``catchpad``' instruction is used by `LLVM's exception handling
5248 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5249 is a catch block --- one where a personality routine attempts to transfer
5250 control to catch an exception.
5251 The ``args`` correspond to whatever information the personality
5252 routine requires to know if this is an appropriate place to catch the
5253 exception. Control is transfered to the ``exception`` label if the
5254 ``catchpad`` is not an appropriate handler for the in-flight exception.
5255 The ``normal`` label should contain the code found in the ``catch``
5256 portion of a ``try``/``catch`` sequence. The ``resultval`` has the type
5257 :ref:`token <t_token>` and is used to match the ``catchpad`` to
5258 corresponding :ref:`catchrets <i_catchret>`.
5263 The instruction takes a list of arbitrary values which are interpreted
5264 by the :ref:`personality function <personalityfn>`.
5266 The ``catchpad`` must be provided a ``normal`` label to transfer control
5267 to if the ``catchpad`` matches the exception and an ``exception``
5268 label to transfer control to if it doesn't.
5273 When the call stack is being unwound due to an exception being thrown,
5274 the exception is compared against the ``args``. If it doesn't match,
5275 then control is transfered to the ``exception`` basic block.
5276 As with calling conventions, how the personality function results are
5277 represented in LLVM IR is target specific.
5279 The ``catchpad`` instruction has several restrictions:
5281 - A catch block is a basic block which is the unwind destination of
5282 an exceptional instruction.
5283 - A catch block must have a '``catchpad``' instruction as its
5284 first non-PHI instruction.
5285 - A catch block's ``exception`` edge must refer to a catch block or a
5287 - There can be only one '``catchpad``' instruction within the
5289 - A basic block that is not a catch block may not include a
5290 '``catchpad``' instruction.
5291 - A catch block which has another catch block as a predecessor may not have
5292 any other predecessors.
5293 - It is undefined behavior for control to transfer from a ``catchpad`` to a
5294 ``ret`` without first executing a ``catchret`` that consumes the
5295 ``catchpad`` or unwinding through its ``catchendpad``.
5296 - It is undefined behavior for control to transfer from a ``catchpad`` to
5297 itself without first executing a ``catchret`` that consumes the
5298 ``catchpad`` or unwinding through its ``catchendpad``.
5303 .. code-block:: llvm
5305 ;; A catch block which can catch an integer.
5306 %tok = catchpad [i8** @_ZTIi]
5307 to label %int.handler unwind label %terminate
5311 '``catchendpad``' Instruction
5312 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5319 catchendpad unwind label <nextaction>
5320 catchendpad unwind to caller
5325 The '``catchendpad``' instruction is used by `LLVM's exception handling
5326 system <ExceptionHandling.html#overview>`_ to communicate to the
5327 :ref:`personality function <personalityfn>` which invokes are associated
5328 with a chain of :ref:`catchpad <i_catchpad>` instructions; propagating an
5329 exception out of a catch handler is represented by unwinding through its
5330 ``catchendpad``. Unwinding to the outer scope when a chain of catch handlers
5331 do not handle an exception is also represented by unwinding through their
5334 The ``nextaction`` label indicates where control should transfer to if
5335 none of the ``catchpad`` instructions are suitable for catching the
5336 in-flight exception.
5338 If a ``nextaction`` label is not present, the instruction unwinds out of
5339 its parent function. The
5340 :ref:`personality function <personalityfn>` will continue processing
5341 exception handling actions in the caller.
5346 The instruction optionally takes a label, ``nextaction``, indicating
5347 where control should transfer to if none of the preceding
5348 ``catchpad`` instructions are suitable for the in-flight exception.
5353 When the call stack is being unwound due to an exception being thrown
5354 and none of the constituent ``catchpad`` instructions match, then
5355 control is transfered to ``nextaction`` if it is present. If it is not
5356 present, control is transfered to the caller.
5358 The ``catchendpad`` instruction has several restrictions:
5360 - A catch-end block is a basic block which is the unwind destination of
5361 an exceptional instruction.
5362 - A catch-end block must have a '``catchendpad``' instruction as its
5363 first non-PHI instruction.
5364 - There can be only one '``catchendpad``' instruction within the
5366 - A basic block that is not a catch-end block may not include a
5367 '``catchendpad``' instruction.
5368 - Exactly one catch block may unwind to a ``catchendpad``.
5369 - It is undefined behavior to execute a ``catchendpad`` if none of the
5370 '``catchpad``'s chained to it have been executed.
5371 - It is undefined behavior to execute a ``catchendpad`` twice without an
5372 intervening execution of one or more of the '``catchpad``'s chained to it.
5373 - It is undefined behavior to execute a ``catchendpad`` if, after the most
5374 recent execution of the normal successor edge of any ``catchpad`` chained
5375 to it, some ``catchret`` consuming that ``catchpad`` has already been
5377 - It is undefined behavior to execute a ``catchendpad`` if, after the most
5378 recent execution of the normal successor edge of any ``catchpad`` chained
5379 to it, any other ``catchpad`` or ``cleanuppad`` has been executed but has
5380 not had a corresponding
5381 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5386 .. code-block:: llvm
5388 catchendpad unwind label %terminate
5389 catchendpad unwind to caller
5393 '``catchret``' Instruction
5394 ^^^^^^^^^^^^^^^^^^^^^^^^^^
5401 catchret <value> to label <normal>
5406 The '``catchret``' instruction is a terminator instruction that has a
5413 The first argument to a '``catchret``' indicates which ``catchpad`` it
5414 exits. It must be a :ref:`catchpad <i_catchpad>`.
5415 The second argument to a '``catchret``' specifies where control will
5421 The '``catchret``' instruction ends the existing (in-flight) exception
5422 whose unwinding was interrupted with a
5423 :ref:`catchpad <i_catchpad>` instruction.
5424 The :ref:`personality function <personalityfn>` gets a chance to execute
5425 arbitrary code to, for example, run a C++ destructor.
5426 Control then transfers to ``normal``.
5427 It may be passed an optional, personality specific, value.
5429 It is undefined behavior to execute a ``catchret`` whose ``catchpad`` has
5432 It is undefined behavior to execute a ``catchret`` if, after the most recent
5433 execution of its ``catchpad``, some ``catchret`` or ``catchendpad`` linked
5434 to the same ``catchpad`` has already been executed.
5436 It is undefined behavior to execute a ``catchret`` if, after the most recent
5437 execution of its ``catchpad``, any other ``catchpad`` or ``cleanuppad`` has
5438 been executed but has not had a corresponding
5439 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5444 .. code-block:: llvm
5446 catchret %catch label %continue
5448 .. _i_cleanupendpad:
5450 '``cleanupendpad``' Instruction
5451 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5458 cleanupendpad <value> unwind label <nextaction>
5459 cleanupendpad <value> unwind to caller
5464 The '``cleanupendpad``' instruction is used by `LLVM's exception handling
5465 system <ExceptionHandling.html#overview>`_ to communicate to the
5466 :ref:`personality function <personalityfn>` which invokes are associated
5467 with a :ref:`cleanuppad <i_cleanuppad>` instructions; propagating an exception
5468 out of a cleanup is represented by unwinding through its ``cleanupendpad``.
5470 The ``nextaction`` label indicates where control should unwind to next, in the
5471 event that a cleanup is exited by means of an(other) exception being raised.
5473 If a ``nextaction`` label is not present, the instruction unwinds out of
5474 its parent function. The
5475 :ref:`personality function <personalityfn>` will continue processing
5476 exception handling actions in the caller.
5481 The '``cleanupendpad``' instruction requires one argument, which indicates
5482 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5483 It also has an optional successor, ``nextaction``, indicating where control
5489 When and exception propagates to a ``cleanupendpad``, control is transfered to
5490 ``nextaction`` if it is present. If it is not present, control is transfered to
5493 The ``cleanupendpad`` instruction has several restrictions:
5495 - A cleanup-end block is a basic block which is the unwind destination of
5496 an exceptional instruction.
5497 - A cleanup-end block must have a '``cleanupendpad``' instruction as its
5498 first non-PHI instruction.
5499 - There can be only one '``cleanupendpad``' instruction within the
5501 - A basic block that is not a cleanup-end block may not include a
5502 '``cleanupendpad``' instruction.
5503 - It is undefined behavior to execute a ``cleanupendpad`` whose ``cleanuppad``
5504 has not been executed.
5505 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5506 recent execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5507 consuming the same ``cleanuppad`` has already been executed.
5508 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5509 recent execution of its ``cleanuppad``, any other ``cleanuppad`` or
5510 ``catchpad`` has been executed but has not had a corresponding
5511 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5516 .. code-block:: llvm
5518 cleanupendpad %cleanup unwind label %terminate
5519 cleanupendpad %cleanup unwind to caller
5523 '``cleanupret``' Instruction
5524 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5531 cleanupret <value> unwind label <continue>
5532 cleanupret <value> unwind to caller
5537 The '``cleanupret``' instruction is a terminator instruction that has
5538 an optional successor.
5544 The '``cleanupret``' instruction requires one argument, which indicates
5545 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5546 It also has an optional successor, ``continue``.
5551 The '``cleanupret``' instruction indicates to the
5552 :ref:`personality function <personalityfn>` that one
5553 :ref:`cleanuppad <i_cleanuppad>` it transferred control to has ended.
5554 It transfers control to ``continue`` or unwinds out of the function.
5556 It is undefined behavior to execute a ``cleanupret`` whose ``cleanuppad`` has
5559 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5560 execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5561 consuming the same ``cleanuppad`` has already been executed.
5563 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5564 execution of its ``cleanuppad``, any other ``cleanuppad`` or ``catchpad`` has
5565 been executed but has not had a corresponding
5566 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5571 .. code-block:: llvm
5573 cleanupret %cleanup unwind to caller
5574 cleanupret %cleanup unwind label %continue
5578 '``terminatepad``' Instruction
5579 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5586 terminatepad [<args>*] unwind label <exception label>
5587 terminatepad [<args>*] unwind to caller
5592 The '``terminatepad``' instruction is used by `LLVM's exception handling
5593 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5594 is a terminate block --- one where a personality routine may decide to
5595 terminate the program.
5596 The ``args`` correspond to whatever information the personality
5597 routine requires to know if this is an appropriate place to terminate the
5598 program. Control is transferred to the ``exception`` label if the
5599 personality routine decides not to terminate the program for the
5600 in-flight exception.
5605 The instruction takes a list of arbitrary values which are interpreted
5606 by the :ref:`personality function <personalityfn>`.
5608 The ``terminatepad`` may be given an ``exception`` label to
5609 transfer control to if the in-flight exception matches the ``args``.
5614 When the call stack is being unwound due to an exception being thrown,
5615 the exception is compared against the ``args``. If it matches,
5616 then control is transfered to the ``exception`` basic block. Otherwise,
5617 the program is terminated via personality-specific means. Typically,
5618 the first argument to ``terminatepad`` specifies what function the
5619 personality should defer to in order to terminate the program.
5621 The ``terminatepad`` instruction has several restrictions:
5623 - A terminate block is a basic block which is the unwind destination of
5624 an exceptional instruction.
5625 - A terminate block must have a '``terminatepad``' instruction as its
5626 first non-PHI instruction.
5627 - There can be only one '``terminatepad``' instruction within the
5629 - A basic block that is not a terminate block may not include a
5630 '``terminatepad``' instruction.
5635 .. code-block:: llvm
5637 ;; A terminate block which only permits integers.
5638 terminatepad [i8** @_ZTIi] unwind label %continue
5642 '``unreachable``' Instruction
5643 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5655 The '``unreachable``' instruction has no defined semantics. This
5656 instruction is used to inform the optimizer that a particular portion of
5657 the code is not reachable. This can be used to indicate that the code
5658 after a no-return function cannot be reached, and other facts.
5663 The '``unreachable``' instruction has no defined semantics.
5670 Binary operators are used to do most of the computation in a program.
5671 They require two operands of the same type, execute an operation on
5672 them, and produce a single value. The operands might represent multiple
5673 data, as is the case with the :ref:`vector <t_vector>` data type. The
5674 result value has the same type as its operands.
5676 There are several different binary operators:
5680 '``add``' Instruction
5681 ^^^^^^^^^^^^^^^^^^^^^
5688 <result> = add <ty> <op1>, <op2> ; yields ty:result
5689 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
5690 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
5691 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
5696 The '``add``' instruction returns the sum of its two operands.
5701 The two arguments to the '``add``' instruction must be
5702 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5703 arguments must have identical types.
5708 The value produced is the integer sum of the two operands.
5710 If the sum has unsigned overflow, the result returned is the
5711 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5714 Because LLVM integers use a two's complement representation, this
5715 instruction is appropriate for both signed and unsigned integers.
5717 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5718 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5719 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
5720 unsigned and/or signed overflow, respectively, occurs.
5725 .. code-block:: llvm
5727 <result> = add i32 4, %var ; yields i32:result = 4 + %var
5731 '``fadd``' Instruction
5732 ^^^^^^^^^^^^^^^^^^^^^^
5739 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5744 The '``fadd``' instruction returns the sum of its two operands.
5749 The two arguments to the '``fadd``' instruction must be :ref:`floating
5750 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5751 Both arguments must have identical types.
5756 The value produced is the floating point sum of the two operands. This
5757 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
5758 which are optimization hints to enable otherwise unsafe floating point
5764 .. code-block:: llvm
5766 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
5768 '``sub``' Instruction
5769 ^^^^^^^^^^^^^^^^^^^^^
5776 <result> = sub <ty> <op1>, <op2> ; yields ty:result
5777 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
5778 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
5779 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
5784 The '``sub``' instruction returns the difference of its two operands.
5786 Note that the '``sub``' instruction is used to represent the '``neg``'
5787 instruction present in most other intermediate representations.
5792 The two arguments to the '``sub``' instruction must be
5793 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5794 arguments must have identical types.
5799 The value produced is the integer difference of the two operands.
5801 If the difference has unsigned overflow, the result returned is the
5802 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5805 Because LLVM integers use a two's complement representation, this
5806 instruction is appropriate for both signed and unsigned integers.
5808 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5809 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5810 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
5811 unsigned and/or signed overflow, respectively, occurs.
5816 .. code-block:: llvm
5818 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
5819 <result> = sub i32 0, %val ; yields i32:result = -%var
5823 '``fsub``' Instruction
5824 ^^^^^^^^^^^^^^^^^^^^^^
5831 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5836 The '``fsub``' instruction returns the difference of its two operands.
5838 Note that the '``fsub``' instruction is used to represent the '``fneg``'
5839 instruction present in most other intermediate representations.
5844 The two arguments to the '``fsub``' instruction must be :ref:`floating
5845 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5846 Both arguments must have identical types.
5851 The value produced is the floating point difference of the two operands.
5852 This instruction can also take any number of :ref:`fast-math
5853 flags <fastmath>`, which are optimization hints to enable otherwise
5854 unsafe floating point optimizations:
5859 .. code-block:: llvm
5861 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
5862 <result> = fsub float -0.0, %val ; yields float:result = -%var
5864 '``mul``' Instruction
5865 ^^^^^^^^^^^^^^^^^^^^^
5872 <result> = mul <ty> <op1>, <op2> ; yields ty:result
5873 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
5874 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
5875 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
5880 The '``mul``' instruction returns the product of its two operands.
5885 The two arguments to the '``mul``' instruction must be
5886 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5887 arguments must have identical types.
5892 The value produced is the integer product of the two operands.
5894 If the result of the multiplication has unsigned overflow, the result
5895 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
5896 bit width of the result.
5898 Because LLVM integers use a two's complement representation, and the
5899 result is the same width as the operands, this instruction returns the
5900 correct result for both signed and unsigned integers. If a full product
5901 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
5902 sign-extended or zero-extended as appropriate to the width of the full
5905 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5906 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5907 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
5908 unsigned and/or signed overflow, respectively, occurs.
5913 .. code-block:: llvm
5915 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
5919 '``fmul``' Instruction
5920 ^^^^^^^^^^^^^^^^^^^^^^
5927 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5932 The '``fmul``' instruction returns the product of its two operands.
5937 The two arguments to the '``fmul``' instruction must be :ref:`floating
5938 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5939 Both arguments must have identical types.
5944 The value produced is the floating point product of the two operands.
5945 This instruction can also take any number of :ref:`fast-math
5946 flags <fastmath>`, which are optimization hints to enable otherwise
5947 unsafe floating point optimizations:
5952 .. code-block:: llvm
5954 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
5956 '``udiv``' Instruction
5957 ^^^^^^^^^^^^^^^^^^^^^^
5964 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
5965 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
5970 The '``udiv``' instruction returns the quotient of its two operands.
5975 The two arguments to the '``udiv``' instruction must be
5976 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5977 arguments must have identical types.
5982 The value produced is the unsigned integer quotient of the two operands.
5984 Note that unsigned integer division and signed integer division are
5985 distinct operations; for signed integer division, use '``sdiv``'.
5987 Division by zero leads to undefined behavior.
5989 If the ``exact`` keyword is present, the result value of the ``udiv`` is
5990 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
5991 such, "((a udiv exact b) mul b) == a").
5996 .. code-block:: llvm
5998 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
6000 '``sdiv``' Instruction
6001 ^^^^^^^^^^^^^^^^^^^^^^
6008 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
6009 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
6014 The '``sdiv``' instruction returns the quotient of its two operands.
6019 The two arguments to the '``sdiv``' instruction must be
6020 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6021 arguments must have identical types.
6026 The value produced is the signed integer quotient of the two operands
6027 rounded towards zero.
6029 Note that signed integer division and unsigned integer division are
6030 distinct operations; for unsigned integer division, use '``udiv``'.
6032 Division by zero leads to undefined behavior. Overflow also leads to
6033 undefined behavior; this is a rare case, but can occur, for example, by
6034 doing a 32-bit division of -2147483648 by -1.
6036 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
6037 a :ref:`poison value <poisonvalues>` if the result would be rounded.
6042 .. code-block:: llvm
6044 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
6048 '``fdiv``' Instruction
6049 ^^^^^^^^^^^^^^^^^^^^^^
6056 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6061 The '``fdiv``' instruction returns the quotient of its two operands.
6066 The two arguments to the '``fdiv``' instruction must be :ref:`floating
6067 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6068 Both arguments must have identical types.
6073 The value produced is the floating point quotient of the two operands.
6074 This instruction can also take any number of :ref:`fast-math
6075 flags <fastmath>`, which are optimization hints to enable otherwise
6076 unsafe floating point optimizations:
6081 .. code-block:: llvm
6083 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
6085 '``urem``' Instruction
6086 ^^^^^^^^^^^^^^^^^^^^^^
6093 <result> = urem <ty> <op1>, <op2> ; yields ty:result
6098 The '``urem``' instruction returns the remainder from the unsigned
6099 division of its two arguments.
6104 The two arguments to the '``urem``' instruction must be
6105 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6106 arguments must have identical types.
6111 This instruction returns the unsigned integer *remainder* of a division.
6112 This instruction always performs an unsigned division to get the
6115 Note that unsigned integer remainder and signed integer remainder are
6116 distinct operations; for signed integer remainder, use '``srem``'.
6118 Taking the remainder of a division by zero leads to undefined behavior.
6123 .. code-block:: llvm
6125 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
6127 '``srem``' Instruction
6128 ^^^^^^^^^^^^^^^^^^^^^^
6135 <result> = srem <ty> <op1>, <op2> ; yields ty:result
6140 The '``srem``' instruction returns the remainder from the signed
6141 division of its two operands. This instruction can also take
6142 :ref:`vector <t_vector>` versions of the values in which case the elements
6148 The two arguments to the '``srem``' instruction must be
6149 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6150 arguments must have identical types.
6155 This instruction returns the *remainder* of a division (where the result
6156 is either zero or has the same sign as the dividend, ``op1``), not the
6157 *modulo* operator (where the result is either zero or has the same sign
6158 as the divisor, ``op2``) of a value. For more information about the
6159 difference, see `The Math
6160 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
6161 table of how this is implemented in various languages, please see
6163 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
6165 Note that signed integer remainder and unsigned integer remainder are
6166 distinct operations; for unsigned integer remainder, use '``urem``'.
6168 Taking the remainder of a division by zero leads to undefined behavior.
6169 Overflow also leads to undefined behavior; this is a rare case, but can
6170 occur, for example, by taking the remainder of a 32-bit division of
6171 -2147483648 by -1. (The remainder doesn't actually overflow, but this
6172 rule lets srem be implemented using instructions that return both the
6173 result of the division and the remainder.)
6178 .. code-block:: llvm
6180 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
6184 '``frem``' Instruction
6185 ^^^^^^^^^^^^^^^^^^^^^^
6192 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6197 The '``frem``' instruction returns the remainder from the division of
6203 The two arguments to the '``frem``' instruction must be :ref:`floating
6204 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6205 Both arguments must have identical types.
6210 This instruction returns the *remainder* of a division. The remainder
6211 has the same sign as the dividend. This instruction can also take any
6212 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
6213 to enable otherwise unsafe floating point optimizations:
6218 .. code-block:: llvm
6220 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
6224 Bitwise Binary Operations
6225 -------------------------
6227 Bitwise binary operators are used to do various forms of bit-twiddling
6228 in a program. They are generally very efficient instructions and can
6229 commonly be strength reduced from other instructions. They require two
6230 operands of the same type, execute an operation on them, and produce a
6231 single value. The resulting value is the same type as its operands.
6233 '``shl``' Instruction
6234 ^^^^^^^^^^^^^^^^^^^^^
6241 <result> = shl <ty> <op1>, <op2> ; yields ty:result
6242 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
6243 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
6244 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
6249 The '``shl``' instruction returns the first operand shifted to the left
6250 a specified number of bits.
6255 Both arguments to the '``shl``' instruction must be the same
6256 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6257 '``op2``' is treated as an unsigned value.
6262 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
6263 where ``n`` is the width of the result. If ``op2`` is (statically or
6264 dynamically) equal to or larger than the number of bits in
6265 ``op1``, the result is undefined. If the arguments are vectors, each
6266 vector element of ``op1`` is shifted by the corresponding shift amount
6269 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
6270 value <poisonvalues>` if it shifts out any non-zero bits. If the
6271 ``nsw`` keyword is present, then the shift produces a :ref:`poison
6272 value <poisonvalues>` if it shifts out any bits that disagree with the
6273 resultant sign bit. As such, NUW/NSW have the same semantics as they
6274 would if the shift were expressed as a mul instruction with the same
6275 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
6280 .. code-block:: llvm
6282 <result> = shl i32 4, %var ; yields i32: 4 << %var
6283 <result> = shl i32 4, 2 ; yields i32: 16
6284 <result> = shl i32 1, 10 ; yields i32: 1024
6285 <result> = shl i32 1, 32 ; undefined
6286 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
6288 '``lshr``' Instruction
6289 ^^^^^^^^^^^^^^^^^^^^^^
6296 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
6297 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
6302 The '``lshr``' instruction (logical shift right) returns the first
6303 operand shifted to the right a specified number of bits with zero fill.
6308 Both arguments to the '``lshr``' instruction must be the same
6309 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6310 '``op2``' is treated as an unsigned value.
6315 This instruction always performs a logical shift right operation. The
6316 most significant bits of the result will be filled with zero bits after
6317 the shift. If ``op2`` is (statically or dynamically) equal to or larger
6318 than the number of bits in ``op1``, the result is undefined. If the
6319 arguments are vectors, each vector element of ``op1`` is shifted by the
6320 corresponding shift amount in ``op2``.
6322 If the ``exact`` keyword is present, the result value of the ``lshr`` is
6323 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6329 .. code-block:: llvm
6331 <result> = lshr i32 4, 1 ; yields i32:result = 2
6332 <result> = lshr i32 4, 2 ; yields i32:result = 1
6333 <result> = lshr i8 4, 3 ; yields i8:result = 0
6334 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
6335 <result> = lshr i32 1, 32 ; undefined
6336 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
6338 '``ashr``' Instruction
6339 ^^^^^^^^^^^^^^^^^^^^^^
6346 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
6347 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
6352 The '``ashr``' instruction (arithmetic shift right) returns the first
6353 operand shifted to the right a specified number of bits with sign
6359 Both arguments to the '``ashr``' instruction must be the same
6360 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6361 '``op2``' is treated as an unsigned value.
6366 This instruction always performs an arithmetic shift right operation,
6367 The most significant bits of the result will be filled with the sign bit
6368 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
6369 than the number of bits in ``op1``, the result is undefined. If the
6370 arguments are vectors, each vector element of ``op1`` is shifted by the
6371 corresponding shift amount in ``op2``.
6373 If the ``exact`` keyword is present, the result value of the ``ashr`` is
6374 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6380 .. code-block:: llvm
6382 <result> = ashr i32 4, 1 ; yields i32:result = 2
6383 <result> = ashr i32 4, 2 ; yields i32:result = 1
6384 <result> = ashr i8 4, 3 ; yields i8:result = 0
6385 <result> = ashr i8 -2, 1 ; yields i8:result = -1
6386 <result> = ashr i32 1, 32 ; undefined
6387 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
6389 '``and``' Instruction
6390 ^^^^^^^^^^^^^^^^^^^^^
6397 <result> = and <ty> <op1>, <op2> ; yields ty:result
6402 The '``and``' instruction returns the bitwise logical and of its two
6408 The two arguments to the '``and``' instruction must be
6409 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6410 arguments must have identical types.
6415 The truth table used for the '``and``' instruction is:
6432 .. code-block:: llvm
6434 <result> = and i32 4, %var ; yields i32:result = 4 & %var
6435 <result> = and i32 15, 40 ; yields i32:result = 8
6436 <result> = and i32 4, 8 ; yields i32:result = 0
6438 '``or``' Instruction
6439 ^^^^^^^^^^^^^^^^^^^^
6446 <result> = or <ty> <op1>, <op2> ; yields ty:result
6451 The '``or``' instruction returns the bitwise logical inclusive or of its
6457 The two arguments to the '``or``' instruction must be
6458 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6459 arguments must have identical types.
6464 The truth table used for the '``or``' instruction is:
6483 <result> = or i32 4, %var ; yields i32:result = 4 | %var
6484 <result> = or i32 15, 40 ; yields i32:result = 47
6485 <result> = or i32 4, 8 ; yields i32:result = 12
6487 '``xor``' Instruction
6488 ^^^^^^^^^^^^^^^^^^^^^
6495 <result> = xor <ty> <op1>, <op2> ; yields ty:result
6500 The '``xor``' instruction returns the bitwise logical exclusive or of
6501 its two operands. The ``xor`` is used to implement the "one's
6502 complement" operation, which is the "~" operator in C.
6507 The two arguments to the '``xor``' instruction must be
6508 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6509 arguments must have identical types.
6514 The truth table used for the '``xor``' instruction is:
6531 .. code-block:: llvm
6533 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
6534 <result> = xor i32 15, 40 ; yields i32:result = 39
6535 <result> = xor i32 4, 8 ; yields i32:result = 12
6536 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
6541 LLVM supports several instructions to represent vector operations in a
6542 target-independent manner. These instructions cover the element-access
6543 and vector-specific operations needed to process vectors effectively.
6544 While LLVM does directly support these vector operations, many
6545 sophisticated algorithms will want to use target-specific intrinsics to
6546 take full advantage of a specific target.
6548 .. _i_extractelement:
6550 '``extractelement``' Instruction
6551 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6558 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
6563 The '``extractelement``' instruction extracts a single scalar element
6564 from a vector at a specified index.
6569 The first operand of an '``extractelement``' instruction is a value of
6570 :ref:`vector <t_vector>` type. The second operand is an index indicating
6571 the position from which to extract the element. The index may be a
6572 variable of any integer type.
6577 The result is a scalar of the same type as the element type of ``val``.
6578 Its value is the value at position ``idx`` of ``val``. If ``idx``
6579 exceeds the length of ``val``, the results are undefined.
6584 .. code-block:: llvm
6586 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
6588 .. _i_insertelement:
6590 '``insertelement``' Instruction
6591 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6598 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
6603 The '``insertelement``' instruction inserts a scalar element into a
6604 vector at a specified index.
6609 The first operand of an '``insertelement``' instruction is a value of
6610 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
6611 type must equal the element type of the first operand. The third operand
6612 is an index indicating the position at which to insert the value. The
6613 index may be a variable of any integer type.
6618 The result is a vector of the same type as ``val``. Its element values
6619 are those of ``val`` except at position ``idx``, where it gets the value
6620 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
6626 .. code-block:: llvm
6628 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
6630 .. _i_shufflevector:
6632 '``shufflevector``' Instruction
6633 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6640 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
6645 The '``shufflevector``' instruction constructs a permutation of elements
6646 from two input vectors, returning a vector with the same element type as
6647 the input and length that is the same as the shuffle mask.
6652 The first two operands of a '``shufflevector``' instruction are vectors
6653 with the same type. The third argument is a shuffle mask whose element
6654 type is always 'i32'. The result of the instruction is a vector whose
6655 length is the same as the shuffle mask and whose element type is the
6656 same as the element type of the first two operands.
6658 The shuffle mask operand is required to be a constant vector with either
6659 constant integer or undef values.
6664 The elements of the two input vectors are numbered from left to right
6665 across both of the vectors. The shuffle mask operand specifies, for each
6666 element of the result vector, which element of the two input vectors the
6667 result element gets. The element selector may be undef (meaning "don't
6668 care") and the second operand may be undef if performing a shuffle from
6674 .. code-block:: llvm
6676 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6677 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
6678 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
6679 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
6680 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
6681 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
6682 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6683 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
6685 Aggregate Operations
6686 --------------------
6688 LLVM supports several instructions for working with
6689 :ref:`aggregate <t_aggregate>` values.
6693 '``extractvalue``' Instruction
6694 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6701 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
6706 The '``extractvalue``' instruction extracts the value of a member field
6707 from an :ref:`aggregate <t_aggregate>` value.
6712 The first operand of an '``extractvalue``' instruction is a value of
6713 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The other operands are
6714 constant indices to specify which value to extract in a similar manner
6715 as indices in a '``getelementptr``' instruction.
6717 The major differences to ``getelementptr`` indexing are:
6719 - Since the value being indexed is not a pointer, the first index is
6720 omitted and assumed to be zero.
6721 - At least one index must be specified.
6722 - Not only struct indices but also array indices must be in bounds.
6727 The result is the value at the position in the aggregate specified by
6733 .. code-block:: llvm
6735 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
6739 '``insertvalue``' Instruction
6740 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6747 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
6752 The '``insertvalue``' instruction inserts a value into a member field in
6753 an :ref:`aggregate <t_aggregate>` value.
6758 The first operand of an '``insertvalue``' instruction is a value of
6759 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
6760 a first-class value to insert. The following operands are constant
6761 indices indicating the position at which to insert the value in a
6762 similar manner as indices in a '``extractvalue``' instruction. The value
6763 to insert must have the same type as the value identified by the
6769 The result is an aggregate of the same type as ``val``. Its value is
6770 that of ``val`` except that the value at the position specified by the
6771 indices is that of ``elt``.
6776 .. code-block:: llvm
6778 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
6779 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
6780 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
6784 Memory Access and Addressing Operations
6785 ---------------------------------------
6787 A key design point of an SSA-based representation is how it represents
6788 memory. In LLVM, no memory locations are in SSA form, which makes things
6789 very simple. This section describes how to read, write, and allocate
6794 '``alloca``' Instruction
6795 ^^^^^^^^^^^^^^^^^^^^^^^^
6802 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
6807 The '``alloca``' instruction allocates memory on the stack frame of the
6808 currently executing function, to be automatically released when this
6809 function returns to its caller. The object is always allocated in the
6810 generic address space (address space zero).
6815 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
6816 bytes of memory on the runtime stack, returning a pointer of the
6817 appropriate type to the program. If "NumElements" is specified, it is
6818 the number of elements allocated, otherwise "NumElements" is defaulted
6819 to be one. If a constant alignment is specified, the value result of the
6820 allocation is guaranteed to be aligned to at least that boundary. The
6821 alignment may not be greater than ``1 << 29``. If not specified, or if
6822 zero, the target can choose to align the allocation on any convenient
6823 boundary compatible with the type.
6825 '``type``' may be any sized type.
6830 Memory is allocated; a pointer is returned. The operation is undefined
6831 if there is insufficient stack space for the allocation. '``alloca``'d
6832 memory is automatically released when the function returns. The
6833 '``alloca``' instruction is commonly used to represent automatic
6834 variables that must have an address available. When the function returns
6835 (either with the ``ret`` or ``resume`` instructions), the memory is
6836 reclaimed. Allocating zero bytes is legal, but the result is undefined.
6837 The order in which memory is allocated (ie., which way the stack grows)
6843 .. code-block:: llvm
6845 %ptr = alloca i32 ; yields i32*:ptr
6846 %ptr = alloca i32, i32 4 ; yields i32*:ptr
6847 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
6848 %ptr = alloca i32, align 1024 ; yields i32*:ptr
6852 '``load``' Instruction
6853 ^^^^^^^^^^^^^^^^^^^^^^
6860 <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>]
6861 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>]
6862 !<index> = !{ i32 1 }
6863 !<deref_bytes_node> = !{i64 <dereferenceable_bytes>}
6864 !<align_node> = !{ i64 <value_alignment> }
6869 The '``load``' instruction is used to read from memory.
6874 The argument to the ``load`` instruction specifies the memory address
6875 from which to load. The type specified must be a :ref:`first
6876 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
6877 then the optimizer is not allowed to modify the number or order of
6878 execution of this ``load`` with other :ref:`volatile
6879 operations <volatile>`.
6881 If the ``load`` is marked as ``atomic``, it takes an extra
6882 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6883 ``release`` and ``acq_rel`` orderings are not valid on ``load``
6884 instructions. Atomic loads produce :ref:`defined <memmodel>` results
6885 when they may see multiple atomic stores. The type of the pointee must
6886 be an integer type whose bit width is a power of two greater than or
6887 equal to eight and less than or equal to a target-specific size limit.
6888 ``align`` must be explicitly specified on atomic loads, and the load has
6889 undefined behavior if the alignment is not set to a value which is at
6890 least the size in bytes of the pointee. ``!nontemporal`` does not have
6891 any defined semantics for atomic loads.
6893 The optional constant ``align`` argument specifies the alignment of the
6894 operation (that is, the alignment of the memory address). A value of 0
6895 or an omitted ``align`` argument means that the operation has the ABI
6896 alignment for the target. It is the responsibility of the code emitter
6897 to ensure that the alignment information is correct. Overestimating the
6898 alignment results in undefined behavior. Underestimating the alignment
6899 may produce less efficient code. An alignment of 1 is always safe. The
6900 maximum possible alignment is ``1 << 29``.
6902 The optional ``!nontemporal`` metadata must reference a single
6903 metadata name ``<index>`` corresponding to a metadata node with one
6904 ``i32`` entry of value 1. The existence of the ``!nontemporal``
6905 metadata on the instruction tells the optimizer and code generator
6906 that this load is not expected to be reused in the cache. The code
6907 generator may select special instructions to save cache bandwidth, such
6908 as the ``MOVNT`` instruction on x86.
6910 The optional ``!invariant.load`` metadata must reference a single
6911 metadata name ``<index>`` corresponding to a metadata node with no
6912 entries. The existence of the ``!invariant.load`` metadata on the
6913 instruction tells the optimizer and code generator that the address
6914 operand to this load points to memory which can be assumed unchanged.
6915 Being invariant does not imply that a location is dereferenceable,
6916 but it does imply that once the location is known dereferenceable
6917 its value is henceforth unchanging.
6919 The optional ``!invariant.group`` metadata must reference a single metadata name
6920 ``<index>`` corresponding to a metadata node. See ``invariant.group`` metadata.
6922 The optional ``!nonnull`` metadata must reference a single
6923 metadata name ``<index>`` corresponding to a metadata node with no
6924 entries. The existence of the ``!nonnull`` metadata on the
6925 instruction tells the optimizer that the value loaded is known to
6926 never be null. This is analogous to the ``nonnull`` attribute
6927 on parameters and return values. This metadata can only be applied
6928 to loads of a pointer type.
6930 The optional ``!dereferenceable`` metadata must reference a single metadata
6931 name ``<deref_bytes_node>`` corresponding to a metadata node with one ``i64``
6932 entry. The existence of the ``!dereferenceable`` metadata on the instruction
6933 tells the optimizer that the value loaded is known to be dereferenceable.
6934 The number of bytes known to be dereferenceable is specified by the integer
6935 value in the metadata node. This is analogous to the ''dereferenceable''
6936 attribute on parameters and return values. This metadata can only be applied
6937 to loads of a pointer type.
6939 The optional ``!dereferenceable_or_null`` metadata must reference a single
6940 metadata name ``<deref_bytes_node>`` corresponding to a metadata node with one
6941 ``i64`` entry. The existence of the ``!dereferenceable_or_null`` metadata on the
6942 instruction tells the optimizer that the value loaded is known to be either
6943 dereferenceable or null.
6944 The number of bytes known to be dereferenceable is specified by the integer
6945 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
6946 attribute on parameters and return values. This metadata can only be applied
6947 to loads of a pointer type.
6949 The optional ``!align`` metadata must reference a single metadata name
6950 ``<align_node>`` corresponding to a metadata node with one ``i64`` entry.
6951 The existence of the ``!align`` metadata on the instruction tells the
6952 optimizer that the value loaded is known to be aligned to a boundary specified
6953 by the integer value in the metadata node. The alignment must be a power of 2.
6954 This is analogous to the ''align'' attribute on parameters and return values.
6955 This metadata can only be applied to loads of a pointer type.
6960 The location of memory pointed to is loaded. If the value being loaded
6961 is of scalar type then the number of bytes read does not exceed the
6962 minimum number of bytes needed to hold all bits of the type. For
6963 example, loading an ``i24`` reads at most three bytes. When loading a
6964 value of a type like ``i20`` with a size that is not an integral number
6965 of bytes, the result is undefined if the value was not originally
6966 written using a store of the same type.
6971 .. code-block:: llvm
6973 %ptr = alloca i32 ; yields i32*:ptr
6974 store i32 3, i32* %ptr ; yields void
6975 %val = load i32, i32* %ptr ; yields i32:val = i32 3
6979 '``store``' Instruction
6980 ^^^^^^^^^^^^^^^^^^^^^^^
6987 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.group !<index>] ; yields void
6988 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>] ; yields void
6993 The '``store``' instruction is used to write to memory.
6998 There are two arguments to the ``store`` instruction: a value to store
6999 and an address at which to store it. The type of the ``<pointer>``
7000 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
7001 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
7002 then the optimizer is not allowed to modify the number or order of
7003 execution of this ``store`` with other :ref:`volatile
7004 operations <volatile>`.
7006 If the ``store`` is marked as ``atomic``, it takes an extra
7007 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
7008 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
7009 instructions. Atomic loads produce :ref:`defined <memmodel>` results
7010 when they may see multiple atomic stores. The type of the pointee must
7011 be an integer type whose bit width is a power of two greater than or
7012 equal to eight and less than or equal to a target-specific size limit.
7013 ``align`` must be explicitly specified on atomic stores, and the store
7014 has undefined behavior if the alignment is not set to a value which is
7015 at least the size in bytes of the pointee. ``!nontemporal`` does not
7016 have any defined semantics for atomic stores.
7018 The optional constant ``align`` argument specifies the alignment of the
7019 operation (that is, the alignment of the memory address). A value of 0
7020 or an omitted ``align`` argument means that the operation has the ABI
7021 alignment for the target. It is the responsibility of the code emitter
7022 to ensure that the alignment information is correct. Overestimating the
7023 alignment results in undefined behavior. Underestimating the
7024 alignment may produce less efficient code. An alignment of 1 is always
7025 safe. The maximum possible alignment is ``1 << 29``.
7027 The optional ``!nontemporal`` metadata must reference a single metadata
7028 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
7029 value 1. The existence of the ``!nontemporal`` metadata on the instruction
7030 tells the optimizer and code generator that this load is not expected to
7031 be reused in the cache. The code generator may select special
7032 instructions to save cache bandwidth, such as the MOVNT instruction on
7035 The optional ``!invariant.group`` metadata must reference a
7036 single metadata name ``<index>``. See ``invariant.group`` metadata.
7041 The contents of memory are updated to contain ``<value>`` at the
7042 location specified by the ``<pointer>`` operand. If ``<value>`` is
7043 of scalar type then the number of bytes written does not exceed the
7044 minimum number of bytes needed to hold all bits of the type. For
7045 example, storing an ``i24`` writes at most three bytes. When writing a
7046 value of a type like ``i20`` with a size that is not an integral number
7047 of bytes, it is unspecified what happens to the extra bits that do not
7048 belong to the type, but they will typically be overwritten.
7053 .. code-block:: llvm
7055 %ptr = alloca i32 ; yields i32*:ptr
7056 store i32 3, i32* %ptr ; yields void
7057 %val = load i32, i32* %ptr ; yields i32:val = i32 3
7061 '``fence``' Instruction
7062 ^^^^^^^^^^^^^^^^^^^^^^^
7069 fence [singlethread] <ordering> ; yields void
7074 The '``fence``' instruction is used to introduce happens-before edges
7080 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
7081 defines what *synchronizes-with* edges they add. They can only be given
7082 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
7087 A fence A which has (at least) ``release`` ordering semantics
7088 *synchronizes with* a fence B with (at least) ``acquire`` ordering
7089 semantics if and only if there exist atomic operations X and Y, both
7090 operating on some atomic object M, such that A is sequenced before X, X
7091 modifies M (either directly or through some side effect of a sequence
7092 headed by X), Y is sequenced before B, and Y observes M. This provides a
7093 *happens-before* dependency between A and B. Rather than an explicit
7094 ``fence``, one (but not both) of the atomic operations X or Y might
7095 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
7096 still *synchronize-with* the explicit ``fence`` and establish the
7097 *happens-before* edge.
7099 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
7100 ``acquire`` and ``release`` semantics specified above, participates in
7101 the global program order of other ``seq_cst`` operations and/or fences.
7103 The optional ":ref:`singlethread <singlethread>`" argument specifies
7104 that the fence only synchronizes with other fences in the same thread.
7105 (This is useful for interacting with signal handlers.)
7110 .. code-block:: llvm
7112 fence acquire ; yields void
7113 fence singlethread seq_cst ; yields void
7117 '``cmpxchg``' Instruction
7118 ^^^^^^^^^^^^^^^^^^^^^^^^^
7125 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
7130 The '``cmpxchg``' instruction is used to atomically modify memory. It
7131 loads a value in memory and compares it to a given value. If they are
7132 equal, it tries to store a new value into the memory.
7137 There are three arguments to the '``cmpxchg``' instruction: an address
7138 to operate on, a value to compare to the value currently be at that
7139 address, and a new value to place at that address if the compared values
7140 are equal. The type of '<cmp>' must be an integer type whose bit width
7141 is a power of two greater than or equal to eight and less than or equal
7142 to a target-specific size limit. '<cmp>' and '<new>' must have the same
7143 type, and the type of '<pointer>' must be a pointer to that type. If the
7144 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
7145 to modify the number or order of execution of this ``cmpxchg`` with
7146 other :ref:`volatile operations <volatile>`.
7148 The success and failure :ref:`ordering <ordering>` arguments specify how this
7149 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
7150 must be at least ``monotonic``, the ordering constraint on failure must be no
7151 stronger than that on success, and the failure ordering cannot be either
7152 ``release`` or ``acq_rel``.
7154 The optional "``singlethread``" argument declares that the ``cmpxchg``
7155 is only atomic with respect to code (usually signal handlers) running in
7156 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
7157 respect to all other code in the system.
7159 The pointer passed into cmpxchg must have alignment greater than or
7160 equal to the size in memory of the operand.
7165 The contents of memory at the location specified by the '``<pointer>``' operand
7166 is read and compared to '``<cmp>``'; if the read value is the equal, the
7167 '``<new>``' is written. The original value at the location is returned, together
7168 with a flag indicating success (true) or failure (false).
7170 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
7171 permitted: the operation may not write ``<new>`` even if the comparison
7174 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
7175 if the value loaded equals ``cmp``.
7177 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
7178 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
7179 load with an ordering parameter determined the second ordering parameter.
7184 .. code-block:: llvm
7187 %orig = atomic load i32, i32* %ptr unordered ; yields i32
7191 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
7192 %squared = mul i32 %cmp, %cmp
7193 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
7194 %value_loaded = extractvalue { i32, i1 } %val_success, 0
7195 %success = extractvalue { i32, i1 } %val_success, 1
7196 br i1 %success, label %done, label %loop
7203 '``atomicrmw``' Instruction
7204 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7211 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
7216 The '``atomicrmw``' instruction is used to atomically modify memory.
7221 There are three arguments to the '``atomicrmw``' instruction: an
7222 operation to apply, an address whose value to modify, an argument to the
7223 operation. The operation must be one of the following keywords:
7237 The type of '<value>' must be an integer type whose bit width is a power
7238 of two greater than or equal to eight and less than or equal to a
7239 target-specific size limit. The type of the '``<pointer>``' operand must
7240 be a pointer to that type. If the ``atomicrmw`` is marked as
7241 ``volatile``, then the optimizer is not allowed to modify the number or
7242 order of execution of this ``atomicrmw`` with other :ref:`volatile
7243 operations <volatile>`.
7248 The contents of memory at the location specified by the '``<pointer>``'
7249 operand are atomically read, modified, and written back. The original
7250 value at the location is returned. The modification is specified by the
7253 - xchg: ``*ptr = val``
7254 - add: ``*ptr = *ptr + val``
7255 - sub: ``*ptr = *ptr - val``
7256 - and: ``*ptr = *ptr & val``
7257 - nand: ``*ptr = ~(*ptr & val)``
7258 - or: ``*ptr = *ptr | val``
7259 - xor: ``*ptr = *ptr ^ val``
7260 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
7261 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
7262 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
7264 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
7270 .. code-block:: llvm
7272 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
7274 .. _i_getelementptr:
7276 '``getelementptr``' Instruction
7277 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7284 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7285 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7286 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
7291 The '``getelementptr``' instruction is used to get the address of a
7292 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
7293 address calculation only and does not access memory. The instruction can also
7294 be used to calculate a vector of such addresses.
7299 The first argument is always a type used as the basis for the calculations.
7300 The second argument is always a pointer or a vector of pointers, and is the
7301 base address to start from. The remaining arguments are indices
7302 that indicate which of the elements of the aggregate object are indexed.
7303 The interpretation of each index is dependent on the type being indexed
7304 into. The first index always indexes the pointer value given as the
7305 first argument, the second index indexes a value of the type pointed to
7306 (not necessarily the value directly pointed to, since the first index
7307 can be non-zero), etc. The first type indexed into must be a pointer
7308 value, subsequent types can be arrays, vectors, and structs. Note that
7309 subsequent types being indexed into can never be pointers, since that
7310 would require loading the pointer before continuing calculation.
7312 The type of each index argument depends on the type it is indexing into.
7313 When indexing into a (optionally packed) structure, only ``i32`` integer
7314 **constants** are allowed (when using a vector of indices they must all
7315 be the **same** ``i32`` integer constant). When indexing into an array,
7316 pointer or vector, integers of any width are allowed, and they are not
7317 required to be constant. These integers are treated as signed values
7320 For example, let's consider a C code fragment and how it gets compiled
7336 int *foo(struct ST *s) {
7337 return &s[1].Z.B[5][13];
7340 The LLVM code generated by Clang is:
7342 .. code-block:: llvm
7344 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
7345 %struct.ST = type { i32, double, %struct.RT }
7347 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
7349 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
7356 In the example above, the first index is indexing into the
7357 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
7358 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
7359 indexes into the third element of the structure, yielding a
7360 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
7361 structure. The third index indexes into the second element of the
7362 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
7363 dimensions of the array are subscripted into, yielding an '``i32``'
7364 type. The '``getelementptr``' instruction returns a pointer to this
7365 element, thus computing a value of '``i32*``' type.
7367 Note that it is perfectly legal to index partially through a structure,
7368 returning a pointer to an inner element. Because of this, the LLVM code
7369 for the given testcase is equivalent to:
7371 .. code-block:: llvm
7373 define i32* @foo(%struct.ST* %s) {
7374 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
7375 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
7376 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
7377 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
7378 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
7382 If the ``inbounds`` keyword is present, the result value of the
7383 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
7384 pointer is not an *in bounds* address of an allocated object, or if any
7385 of the addresses that would be formed by successive addition of the
7386 offsets implied by the indices to the base address with infinitely
7387 precise signed arithmetic are not an *in bounds* address of that
7388 allocated object. The *in bounds* addresses for an allocated object are
7389 all the addresses that point into the object, plus the address one byte
7390 past the end. In cases where the base is a vector of pointers the
7391 ``inbounds`` keyword applies to each of the computations element-wise.
7393 If the ``inbounds`` keyword is not present, the offsets are added to the
7394 base address with silently-wrapping two's complement arithmetic. If the
7395 offsets have a different width from the pointer, they are sign-extended
7396 or truncated to the width of the pointer. The result value of the
7397 ``getelementptr`` may be outside the object pointed to by the base
7398 pointer. The result value may not necessarily be used to access memory
7399 though, even if it happens to point into allocated storage. See the
7400 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
7403 The getelementptr instruction is often confusing. For some more insight
7404 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
7409 .. code-block:: llvm
7411 ; yields [12 x i8]*:aptr
7412 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
7414 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
7416 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
7418 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
7423 The ``getelementptr`` returns a vector of pointers, instead of a single address,
7424 when one or more of its arguments is a vector. In such cases, all vector
7425 arguments should have the same number of elements, and every scalar argument
7426 will be effectively broadcast into a vector during address calculation.
7428 .. code-block:: llvm
7430 ; All arguments are vectors:
7431 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
7432 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
7434 ; Add the same scalar offset to each pointer of a vector:
7435 ; A[i] = ptrs[i] + offset*sizeof(i8)
7436 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
7438 ; Add distinct offsets to the same pointer:
7439 ; A[i] = ptr + offsets[i]*sizeof(i8)
7440 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
7442 ; In all cases described above the type of the result is <4 x i8*>
7444 The two following instructions are equivalent:
7446 .. code-block:: llvm
7448 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7449 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
7450 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
7452 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
7454 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7455 i32 2, i32 1, <4 x i32> %ind4, i64 13
7457 Let's look at the C code, where the vector version of ``getelementptr``
7462 // Let's assume that we vectorize the following loop:
7463 double *A, B; int *C;
7464 for (int i = 0; i < size; ++i) {
7468 .. code-block:: llvm
7470 ; get pointers for 8 elements from array B
7471 %ptrs = getelementptr double, double* %B, <8 x i32> %C
7472 ; load 8 elements from array B into A
7473 %A = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs,
7474 i32 8, <8 x i1> %mask, <8 x double> %passthru)
7476 Conversion Operations
7477 ---------------------
7479 The instructions in this category are the conversion instructions
7480 (casting) which all take a single operand and a type. They perform
7481 various bit conversions on the operand.
7483 '``trunc .. to``' Instruction
7484 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7491 <result> = trunc <ty> <value> to <ty2> ; yields ty2
7496 The '``trunc``' instruction truncates its operand to the type ``ty2``.
7501 The '``trunc``' instruction takes a value to trunc, and a type to trunc
7502 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
7503 of the same number of integers. The bit size of the ``value`` must be
7504 larger than the bit size of the destination type, ``ty2``. Equal sized
7505 types are not allowed.
7510 The '``trunc``' instruction truncates the high order bits in ``value``
7511 and converts the remaining bits to ``ty2``. Since the source size must
7512 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
7513 It will always truncate bits.
7518 .. code-block:: llvm
7520 %X = trunc i32 257 to i8 ; yields i8:1
7521 %Y = trunc i32 123 to i1 ; yields i1:true
7522 %Z = trunc i32 122 to i1 ; yields i1:false
7523 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
7525 '``zext .. to``' Instruction
7526 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7533 <result> = zext <ty> <value> to <ty2> ; yields ty2
7538 The '``zext``' instruction zero extends its operand to type ``ty2``.
7543 The '``zext``' instruction takes a value to cast, and a type to cast it
7544 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7545 the same number of integers. The bit size of the ``value`` must be
7546 smaller than the bit size of the destination type, ``ty2``.
7551 The ``zext`` fills the high order bits of the ``value`` with zero bits
7552 until it reaches the size of the destination type, ``ty2``.
7554 When zero extending from i1, the result will always be either 0 or 1.
7559 .. code-block:: llvm
7561 %X = zext i32 257 to i64 ; yields i64:257
7562 %Y = zext i1 true to i32 ; yields i32:1
7563 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7565 '``sext .. to``' Instruction
7566 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7573 <result> = sext <ty> <value> to <ty2> ; yields ty2
7578 The '``sext``' sign extends ``value`` to the type ``ty2``.
7583 The '``sext``' instruction takes a value to cast, and a type to cast it
7584 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7585 the same number of integers. The bit size of the ``value`` must be
7586 smaller than the bit size of the destination type, ``ty2``.
7591 The '``sext``' instruction performs a sign extension by copying the sign
7592 bit (highest order bit) of the ``value`` until it reaches the bit size
7593 of the type ``ty2``.
7595 When sign extending from i1, the extension always results in -1 or 0.
7600 .. code-block:: llvm
7602 %X = sext i8 -1 to i16 ; yields i16 :65535
7603 %Y = sext i1 true to i32 ; yields i32:-1
7604 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7606 '``fptrunc .. to``' Instruction
7607 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7614 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
7619 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
7624 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
7625 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
7626 The size of ``value`` must be larger than the size of ``ty2``. This
7627 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
7632 The '``fptrunc``' instruction casts a ``value`` from a larger
7633 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
7634 point <t_floating>` type. If the value cannot fit (i.e. overflows) within the
7635 destination type, ``ty2``, then the results are undefined. If the cast produces
7636 an inexact result, how rounding is performed (e.g. truncation, also known as
7637 round to zero) is undefined.
7642 .. code-block:: llvm
7644 %X = fptrunc double 123.0 to float ; yields float:123.0
7645 %Y = fptrunc double 1.0E+300 to float ; yields undefined
7647 '``fpext .. to``' Instruction
7648 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7655 <result> = fpext <ty> <value> to <ty2> ; yields ty2
7660 The '``fpext``' extends a floating point ``value`` to a larger floating
7666 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
7667 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
7668 to. The source type must be smaller than the destination type.
7673 The '``fpext``' instruction extends the ``value`` from a smaller
7674 :ref:`floating point <t_floating>` type to a larger :ref:`floating
7675 point <t_floating>` type. The ``fpext`` cannot be used to make a
7676 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
7677 *no-op cast* for a floating point cast.
7682 .. code-block:: llvm
7684 %X = fpext float 3.125 to double ; yields double:3.125000e+00
7685 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
7687 '``fptoui .. to``' Instruction
7688 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7695 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
7700 The '``fptoui``' converts a floating point ``value`` to its unsigned
7701 integer equivalent of type ``ty2``.
7706 The '``fptoui``' instruction takes a value to cast, which must be a
7707 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7708 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7709 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7710 type with the same number of elements as ``ty``
7715 The '``fptoui``' instruction converts its :ref:`floating
7716 point <t_floating>` operand into the nearest (rounding towards zero)
7717 unsigned integer value. If the value cannot fit in ``ty2``, the results
7723 .. code-block:: llvm
7725 %X = fptoui double 123.0 to i32 ; yields i32:123
7726 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
7727 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
7729 '``fptosi .. to``' Instruction
7730 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7737 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
7742 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
7743 ``value`` to type ``ty2``.
7748 The '``fptosi``' instruction takes a value to cast, which must be a
7749 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7750 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7751 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7752 type with the same number of elements as ``ty``
7757 The '``fptosi``' instruction converts its :ref:`floating
7758 point <t_floating>` operand into the nearest (rounding towards zero)
7759 signed integer value. If the value cannot fit in ``ty2``, the results
7765 .. code-block:: llvm
7767 %X = fptosi double -123.0 to i32 ; yields i32:-123
7768 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
7769 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
7771 '``uitofp .. to``' Instruction
7772 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7779 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
7784 The '``uitofp``' instruction regards ``value`` as an unsigned integer
7785 and converts that value to the ``ty2`` type.
7790 The '``uitofp``' instruction takes a value to cast, which must be a
7791 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7792 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7793 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7794 type with the same number of elements as ``ty``
7799 The '``uitofp``' instruction interprets its operand as an unsigned
7800 integer quantity and converts it to the corresponding floating point
7801 value. If the value cannot fit in the floating point value, the results
7807 .. code-block:: llvm
7809 %X = uitofp i32 257 to float ; yields float:257.0
7810 %Y = uitofp i8 -1 to double ; yields double:255.0
7812 '``sitofp .. to``' Instruction
7813 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7820 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
7825 The '``sitofp``' instruction regards ``value`` as a signed integer and
7826 converts that value to the ``ty2`` type.
7831 The '``sitofp``' instruction takes a value to cast, which must be a
7832 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7833 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7834 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7835 type with the same number of elements as ``ty``
7840 The '``sitofp``' instruction interprets its operand as a signed integer
7841 quantity and converts it to the corresponding floating point value. If
7842 the value cannot fit in the floating point value, the results are
7848 .. code-block:: llvm
7850 %X = sitofp i32 257 to float ; yields float:257.0
7851 %Y = sitofp i8 -1 to double ; yields double:-1.0
7855 '``ptrtoint .. to``' Instruction
7856 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7863 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
7868 The '``ptrtoint``' instruction converts the pointer or a vector of
7869 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
7874 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
7875 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
7876 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
7877 a vector of integers type.
7882 The '``ptrtoint``' instruction converts ``value`` to integer type
7883 ``ty2`` by interpreting the pointer value as an integer and either
7884 truncating or zero extending that value to the size of the integer type.
7885 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
7886 ``value`` is larger than ``ty2`` then a truncation is done. If they are
7887 the same size, then nothing is done (*no-op cast*) other than a type
7893 .. code-block:: llvm
7895 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
7896 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
7897 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
7901 '``inttoptr .. to``' Instruction
7902 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7909 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
7914 The '``inttoptr``' instruction converts an integer ``value`` to a
7915 pointer type, ``ty2``.
7920 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
7921 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
7927 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
7928 applying either a zero extension or a truncation depending on the size
7929 of the integer ``value``. If ``value`` is larger than the size of a
7930 pointer then a truncation is done. If ``value`` is smaller than the size
7931 of a pointer then a zero extension is done. If they are the same size,
7932 nothing is done (*no-op cast*).
7937 .. code-block:: llvm
7939 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
7940 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
7941 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
7942 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
7946 '``bitcast .. to``' Instruction
7947 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7954 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
7959 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
7965 The '``bitcast``' instruction takes a value to cast, which must be a
7966 non-aggregate first class value, and a type to cast it to, which must
7967 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
7968 bit sizes of ``value`` and the destination type, ``ty2``, must be
7969 identical. If the source type is a pointer, the destination type must
7970 also be a pointer of the same size. This instruction supports bitwise
7971 conversion of vectors to integers and to vectors of other types (as
7972 long as they have the same size).
7977 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
7978 is always a *no-op cast* because no bits change with this
7979 conversion. The conversion is done as if the ``value`` had been stored
7980 to memory and read back as type ``ty2``. Pointer (or vector of
7981 pointers) types may only be converted to other pointer (or vector of
7982 pointers) types with the same address space through this instruction.
7983 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
7984 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
7989 .. code-block:: llvm
7991 %X = bitcast i8 255 to i8 ; yields i8 :-1
7992 %Y = bitcast i32* %x to sint* ; yields sint*:%x
7993 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
7994 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
7996 .. _i_addrspacecast:
7998 '``addrspacecast .. to``' Instruction
7999 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8006 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
8011 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
8012 address space ``n`` to type ``pty2`` in address space ``m``.
8017 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
8018 to cast and a pointer type to cast it to, which must have a different
8024 The '``addrspacecast``' instruction converts the pointer value
8025 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
8026 value modification, depending on the target and the address space
8027 pair. Pointer conversions within the same address space must be
8028 performed with the ``bitcast`` instruction. Note that if the address space
8029 conversion is legal then both result and operand refer to the same memory
8035 .. code-block:: llvm
8037 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
8038 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
8039 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
8046 The instructions in this category are the "miscellaneous" instructions,
8047 which defy better classification.
8051 '``icmp``' Instruction
8052 ^^^^^^^^^^^^^^^^^^^^^^
8059 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8064 The '``icmp``' instruction returns a boolean value or a vector of
8065 boolean values based on comparison of its two integer, integer vector,
8066 pointer, or pointer vector operands.
8071 The '``icmp``' instruction takes three operands. The first operand is
8072 the condition code indicating the kind of comparison to perform. It is
8073 not a value, just a keyword. The possible condition code are:
8076 #. ``ne``: not equal
8077 #. ``ugt``: unsigned greater than
8078 #. ``uge``: unsigned greater or equal
8079 #. ``ult``: unsigned less than
8080 #. ``ule``: unsigned less or equal
8081 #. ``sgt``: signed greater than
8082 #. ``sge``: signed greater or equal
8083 #. ``slt``: signed less than
8084 #. ``sle``: signed less or equal
8086 The remaining two arguments must be :ref:`integer <t_integer>` or
8087 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
8088 must also be identical types.
8093 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
8094 code given as ``cond``. The comparison performed always yields either an
8095 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
8097 #. ``eq``: yields ``true`` if the operands are equal, ``false``
8098 otherwise. No sign interpretation is necessary or performed.
8099 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
8100 otherwise. No sign interpretation is necessary or performed.
8101 #. ``ugt``: interprets the operands as unsigned values and yields
8102 ``true`` if ``op1`` is greater than ``op2``.
8103 #. ``uge``: interprets the operands as unsigned values and yields
8104 ``true`` if ``op1`` is greater than or equal to ``op2``.
8105 #. ``ult``: interprets the operands as unsigned values and yields
8106 ``true`` if ``op1`` is less than ``op2``.
8107 #. ``ule``: interprets the operands as unsigned values and yields
8108 ``true`` if ``op1`` is less than or equal to ``op2``.
8109 #. ``sgt``: interprets the operands as signed values and yields ``true``
8110 if ``op1`` is greater than ``op2``.
8111 #. ``sge``: interprets the operands as signed values and yields ``true``
8112 if ``op1`` is greater than or equal to ``op2``.
8113 #. ``slt``: interprets the operands as signed values and yields ``true``
8114 if ``op1`` is less than ``op2``.
8115 #. ``sle``: interprets the operands as signed values and yields ``true``
8116 if ``op1`` is less than or equal to ``op2``.
8118 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
8119 are compared as if they were integers.
8121 If the operands are integer vectors, then they are compared element by
8122 element. The result is an ``i1`` vector with the same number of elements
8123 as the values being compared. Otherwise, the result is an ``i1``.
8128 .. code-block:: llvm
8130 <result> = icmp eq i32 4, 5 ; yields: result=false
8131 <result> = icmp ne float* %X, %X ; yields: result=false
8132 <result> = icmp ult i16 4, 5 ; yields: result=true
8133 <result> = icmp sgt i16 4, 5 ; yields: result=false
8134 <result> = icmp ule i16 -4, 5 ; yields: result=false
8135 <result> = icmp sge i16 4, 5 ; yields: result=false
8137 Note that the code generator does not yet support vector types with the
8138 ``icmp`` instruction.
8142 '``fcmp``' Instruction
8143 ^^^^^^^^^^^^^^^^^^^^^^
8150 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8155 The '``fcmp``' instruction returns a boolean value or vector of boolean
8156 values based on comparison of its operands.
8158 If the operands are floating point scalars, then the result type is a
8159 boolean (:ref:`i1 <t_integer>`).
8161 If the operands are floating point vectors, then the result type is a
8162 vector of boolean with the same number of elements as the operands being
8168 The '``fcmp``' instruction takes three operands. The first operand is
8169 the condition code indicating the kind of comparison to perform. It is
8170 not a value, just a keyword. The possible condition code are:
8172 #. ``false``: no comparison, always returns false
8173 #. ``oeq``: ordered and equal
8174 #. ``ogt``: ordered and greater than
8175 #. ``oge``: ordered and greater than or equal
8176 #. ``olt``: ordered and less than
8177 #. ``ole``: ordered and less than or equal
8178 #. ``one``: ordered and not equal
8179 #. ``ord``: ordered (no nans)
8180 #. ``ueq``: unordered or equal
8181 #. ``ugt``: unordered or greater than
8182 #. ``uge``: unordered or greater than or equal
8183 #. ``ult``: unordered or less than
8184 #. ``ule``: unordered or less than or equal
8185 #. ``une``: unordered or not equal
8186 #. ``uno``: unordered (either nans)
8187 #. ``true``: no comparison, always returns true
8189 *Ordered* means that neither operand is a QNAN while *unordered* means
8190 that either operand may be a QNAN.
8192 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
8193 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
8194 type. They must have identical types.
8199 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
8200 condition code given as ``cond``. If the operands are vectors, then the
8201 vectors are compared element by element. Each comparison performed
8202 always yields an :ref:`i1 <t_integer>` result, as follows:
8204 #. ``false``: always yields ``false``, regardless of operands.
8205 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
8206 is equal to ``op2``.
8207 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
8208 is greater than ``op2``.
8209 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
8210 is greater than or equal to ``op2``.
8211 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
8212 is less than ``op2``.
8213 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
8214 is less than or equal to ``op2``.
8215 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
8216 is not equal to ``op2``.
8217 #. ``ord``: yields ``true`` if both operands are not a QNAN.
8218 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
8220 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
8221 greater than ``op2``.
8222 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
8223 greater than or equal to ``op2``.
8224 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
8226 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
8227 less than or equal to ``op2``.
8228 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
8229 not equal to ``op2``.
8230 #. ``uno``: yields ``true`` if either operand is a QNAN.
8231 #. ``true``: always yields ``true``, regardless of operands.
8233 The ``fcmp`` instruction can also optionally take any number of
8234 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8235 otherwise unsafe floating point optimizations.
8237 Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
8238 only flags that have any effect on its semantics are those that allow
8239 assumptions to be made about the values of input arguments; namely
8240 ``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
8245 .. code-block:: llvm
8247 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
8248 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
8249 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
8250 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
8252 Note that the code generator does not yet support vector types with the
8253 ``fcmp`` instruction.
8257 '``phi``' Instruction
8258 ^^^^^^^^^^^^^^^^^^^^^
8265 <result> = phi <ty> [ <val0>, <label0>], ...
8270 The '``phi``' instruction is used to implement the φ node in the SSA
8271 graph representing the function.
8276 The type of the incoming values is specified with the first type field.
8277 After this, the '``phi``' instruction takes a list of pairs as
8278 arguments, with one pair for each predecessor basic block of the current
8279 block. Only values of :ref:`first class <t_firstclass>` type may be used as
8280 the value arguments to the PHI node. Only labels may be used as the
8283 There must be no non-phi instructions between the start of a basic block
8284 and the PHI instructions: i.e. PHI instructions must be first in a basic
8287 For the purposes of the SSA form, the use of each incoming value is
8288 deemed to occur on the edge from the corresponding predecessor block to
8289 the current block (but after any definition of an '``invoke``'
8290 instruction's return value on the same edge).
8295 At runtime, the '``phi``' instruction logically takes on the value
8296 specified by the pair corresponding to the predecessor basic block that
8297 executed just prior to the current block.
8302 .. code-block:: llvm
8304 Loop: ; Infinite loop that counts from 0 on up...
8305 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
8306 %nextindvar = add i32 %indvar, 1
8311 '``select``' Instruction
8312 ^^^^^^^^^^^^^^^^^^^^^^^^
8319 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
8321 selty is either i1 or {<N x i1>}
8326 The '``select``' instruction is used to choose one value based on a
8327 condition, without IR-level branching.
8332 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
8333 values indicating the condition, and two values of the same :ref:`first
8334 class <t_firstclass>` type.
8339 If the condition is an i1 and it evaluates to 1, the instruction returns
8340 the first value argument; otherwise, it returns the second value
8343 If the condition is a vector of i1, then the value arguments must be
8344 vectors of the same size, and the selection is done element by element.
8346 If the condition is an i1 and the value arguments are vectors of the
8347 same size, then an entire vector is selected.
8352 .. code-block:: llvm
8354 %X = select i1 true, i8 17, i8 42 ; yields i8:17
8358 '``call``' Instruction
8359 ^^^^^^^^^^^^^^^^^^^^^^
8366 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
8372 The '``call``' instruction represents a simple function call.
8377 This instruction requires several arguments:
8379 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
8380 should perform tail call optimization. The ``tail`` marker is a hint that
8381 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
8382 means that the call must be tail call optimized in order for the program to
8383 be correct. The ``musttail`` marker provides these guarantees:
8385 #. The call will not cause unbounded stack growth if it is part of a
8386 recursive cycle in the call graph.
8387 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
8390 Both markers imply that the callee does not access allocas or varargs from
8391 the caller. Calls marked ``musttail`` must obey the following additional
8394 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
8395 or a pointer bitcast followed by a ret instruction.
8396 - The ret instruction must return the (possibly bitcasted) value
8397 produced by the call or void.
8398 - The caller and callee prototypes must match. Pointer types of
8399 parameters or return types may differ in pointee type, but not
8401 - The calling conventions of the caller and callee must match.
8402 - All ABI-impacting function attributes, such as sret, byval, inreg,
8403 returned, and inalloca, must match.
8404 - The callee must be varargs iff the caller is varargs. Bitcasting a
8405 non-varargs function to the appropriate varargs type is legal so
8406 long as the non-varargs prefixes obey the other rules.
8408 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
8409 the following conditions are met:
8411 - Caller and callee both have the calling convention ``fastcc``.
8412 - The call is in tail position (ret immediately follows call and ret
8413 uses value of call or is void).
8414 - Option ``-tailcallopt`` is enabled, or
8415 ``llvm::GuaranteedTailCallOpt`` is ``true``.
8416 - `Platform-specific constraints are
8417 met. <CodeGenerator.html#tailcallopt>`_
8419 #. The optional "cconv" marker indicates which :ref:`calling
8420 convention <callingconv>` the call should use. If none is
8421 specified, the call defaults to using C calling conventions. The
8422 calling convention of the call must match the calling convention of
8423 the target function, or else the behavior is undefined.
8424 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
8425 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
8427 #. '``ty``': the type of the call instruction itself which is also the
8428 type of the return value. Functions that return no value are marked
8430 #. '``fnty``': shall be the signature of the pointer to function value
8431 being invoked. The argument types must match the types implied by
8432 this signature. This type can be omitted if the function is not
8433 varargs and if the function type does not return a pointer to a
8435 #. '``fnptrval``': An LLVM value containing a pointer to a function to
8436 be invoked. In most cases, this is a direct function invocation, but
8437 indirect ``call``'s are just as possible, calling an arbitrary pointer
8439 #. '``function args``': argument list whose types match the function
8440 signature argument types and parameter attributes. All arguments must
8441 be of :ref:`first class <t_firstclass>` type. If the function signature
8442 indicates the function accepts a variable number of arguments, the
8443 extra arguments can be specified.
8444 #. The optional :ref:`function attributes <fnattrs>` list. Only
8445 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
8446 attributes are valid here.
8447 #. The optional :ref:`operand bundles <opbundles>` list.
8452 The '``call``' instruction is used to cause control flow to transfer to
8453 a specified function, with its incoming arguments bound to the specified
8454 values. Upon a '``ret``' instruction in the called function, control
8455 flow continues with the instruction after the function call, and the
8456 return value of the function is bound to the result argument.
8461 .. code-block:: llvm
8463 %retval = call i32 @test(i32 %argc)
8464 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
8465 %X = tail call i32 @foo() ; yields i32
8466 %Y = tail call fastcc i32 @foo() ; yields i32
8467 call void %foo(i8 97 signext)
8469 %struct.A = type { i32, i8 }
8470 %r = call %struct.A @foo() ; yields { i32, i8 }
8471 %gr = extractvalue %struct.A %r, 0 ; yields i32
8472 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
8473 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
8474 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
8476 llvm treats calls to some functions with names and arguments that match
8477 the standard C99 library as being the C99 library functions, and may
8478 perform optimizations or generate code for them under that assumption.
8479 This is something we'd like to change in the future to provide better
8480 support for freestanding environments and non-C-based languages.
8484 '``va_arg``' Instruction
8485 ^^^^^^^^^^^^^^^^^^^^^^^^
8492 <resultval> = va_arg <va_list*> <arglist>, <argty>
8497 The '``va_arg``' instruction is used to access arguments passed through
8498 the "variable argument" area of a function call. It is used to implement
8499 the ``va_arg`` macro in C.
8504 This instruction takes a ``va_list*`` value and the type of the
8505 argument. It returns a value of the specified argument type and
8506 increments the ``va_list`` to point to the next argument. The actual
8507 type of ``va_list`` is target specific.
8512 The '``va_arg``' instruction loads an argument of the specified type
8513 from the specified ``va_list`` and causes the ``va_list`` to point to
8514 the next argument. For more information, see the variable argument
8515 handling :ref:`Intrinsic Functions <int_varargs>`.
8517 It is legal for this instruction to be called in a function which does
8518 not take a variable number of arguments, for example, the ``vfprintf``
8521 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
8522 function <intrinsics>` because it takes a type as an argument.
8527 See the :ref:`variable argument processing <int_varargs>` section.
8529 Note that the code generator does not yet fully support va\_arg on many
8530 targets. Also, it does not currently support va\_arg with aggregate
8531 types on any target.
8535 '``landingpad``' Instruction
8536 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8543 <resultval> = landingpad <resultty> <clause>+
8544 <resultval> = landingpad <resultty> cleanup <clause>*
8546 <clause> := catch <type> <value>
8547 <clause> := filter <array constant type> <array constant>
8552 The '``landingpad``' instruction is used by `LLVM's exception handling
8553 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8554 is a landing pad --- one where the exception lands, and corresponds to the
8555 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
8556 defines values supplied by the :ref:`personality function <personalityfn>` upon
8557 re-entry to the function. The ``resultval`` has the type ``resultty``.
8563 ``cleanup`` flag indicates that the landing pad block is a cleanup.
8565 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
8566 contains the global variable representing the "type" that may be caught
8567 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
8568 clause takes an array constant as its argument. Use
8569 "``[0 x i8**] undef``" for a filter which cannot throw. The
8570 '``landingpad``' instruction must contain *at least* one ``clause`` or
8571 the ``cleanup`` flag.
8576 The '``landingpad``' instruction defines the values which are set by the
8577 :ref:`personality function <personalityfn>` upon re-entry to the function, and
8578 therefore the "result type" of the ``landingpad`` instruction. As with
8579 calling conventions, how the personality function results are
8580 represented in LLVM IR is target specific.
8582 The clauses are applied in order from top to bottom. If two
8583 ``landingpad`` instructions are merged together through inlining, the
8584 clauses from the calling function are appended to the list of clauses.
8585 When the call stack is being unwound due to an exception being thrown,
8586 the exception is compared against each ``clause`` in turn. If it doesn't
8587 match any of the clauses, and the ``cleanup`` flag is not set, then
8588 unwinding continues further up the call stack.
8590 The ``landingpad`` instruction has several restrictions:
8592 - A landing pad block is a basic block which is the unwind destination
8593 of an '``invoke``' instruction.
8594 - A landing pad block must have a '``landingpad``' instruction as its
8595 first non-PHI instruction.
8596 - There can be only one '``landingpad``' instruction within the landing
8598 - A basic block that is not a landing pad block may not include a
8599 '``landingpad``' instruction.
8604 .. code-block:: llvm
8606 ;; A landing pad which can catch an integer.
8607 %res = landingpad { i8*, i32 }
8609 ;; A landing pad that is a cleanup.
8610 %res = landingpad { i8*, i32 }
8612 ;; A landing pad which can catch an integer and can only throw a double.
8613 %res = landingpad { i8*, i32 }
8615 filter [1 x i8**] [@_ZTId]
8619 '``cleanuppad``' Instruction
8620 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8627 <resultval> = cleanuppad [<args>*]
8632 The '``cleanuppad``' instruction is used by `LLVM's exception handling
8633 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8634 is a cleanup block --- one where a personality routine attempts to
8635 transfer control to run cleanup actions.
8636 The ``args`` correspond to whatever additional
8637 information the :ref:`personality function <personalityfn>` requires to
8638 execute the cleanup.
8639 The ``resultval`` has the type :ref:`token <t_token>` and is used to
8640 match the ``cleanuppad`` to corresponding :ref:`cleanuprets <i_cleanupret>`
8641 and :ref:`cleanupendpads <i_cleanupendpad>`.
8646 The instruction takes a list of arbitrary values which are interpreted
8647 by the :ref:`personality function <personalityfn>`.
8652 When the call stack is being unwound due to an exception being thrown,
8653 the :ref:`personality function <personalityfn>` transfers control to the
8654 ``cleanuppad`` with the aid of the personality-specific arguments.
8655 As with calling conventions, how the personality function results are
8656 represented in LLVM IR is target specific.
8658 The ``cleanuppad`` instruction has several restrictions:
8660 - A cleanup block is a basic block which is the unwind destination of
8661 an exceptional instruction.
8662 - A cleanup block must have a '``cleanuppad``' instruction as its
8663 first non-PHI instruction.
8664 - There can be only one '``cleanuppad``' instruction within the
8666 - A basic block that is not a cleanup block may not include a
8667 '``cleanuppad``' instruction.
8668 - All '``cleanupret``'s and '``cleanupendpad``'s which consume a ``cleanuppad``
8669 must have the same exceptional successor.
8670 - It is undefined behavior for control to transfer from a ``cleanuppad`` to a
8671 ``ret`` without first executing a ``cleanupret`` or ``cleanupendpad`` that
8672 consumes the ``cleanuppad``.
8673 - It is undefined behavior for control to transfer from a ``cleanuppad`` to
8674 itself without first executing a ``cleanupret`` or ``cleanupendpad`` that
8675 consumes the ``cleanuppad``.
8680 .. code-block:: llvm
8682 %tok = cleanuppad []
8689 LLVM supports the notion of an "intrinsic function". These functions
8690 have well known names and semantics and are required to follow certain
8691 restrictions. Overall, these intrinsics represent an extension mechanism
8692 for the LLVM language that does not require changing all of the
8693 transformations in LLVM when adding to the language (or the bitcode
8694 reader/writer, the parser, etc...).
8696 Intrinsic function names must all start with an "``llvm.``" prefix. This
8697 prefix is reserved in LLVM for intrinsic names; thus, function names may
8698 not begin with this prefix. Intrinsic functions must always be external
8699 functions: you cannot define the body of intrinsic functions. Intrinsic
8700 functions may only be used in call or invoke instructions: it is illegal
8701 to take the address of an intrinsic function. Additionally, because
8702 intrinsic functions are part of the LLVM language, it is required if any
8703 are added that they be documented here.
8705 Some intrinsic functions can be overloaded, i.e., the intrinsic
8706 represents a family of functions that perform the same operation but on
8707 different data types. Because LLVM can represent over 8 million
8708 different integer types, overloading is used commonly to allow an
8709 intrinsic function to operate on any integer type. One or more of the
8710 argument types or the result type can be overloaded to accept any
8711 integer type. Argument types may also be defined as exactly matching a
8712 previous argument's type or the result type. This allows an intrinsic
8713 function which accepts multiple arguments, but needs all of them to be
8714 of the same type, to only be overloaded with respect to a single
8715 argument or the result.
8717 Overloaded intrinsics will have the names of its overloaded argument
8718 types encoded into its function name, each preceded by a period. Only
8719 those types which are overloaded result in a name suffix. Arguments
8720 whose type is matched against another type do not. For example, the
8721 ``llvm.ctpop`` function can take an integer of any width and returns an
8722 integer of exactly the same integer width. This leads to a family of
8723 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
8724 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
8725 overloaded, and only one type suffix is required. Because the argument's
8726 type is matched against the return type, it does not require its own
8729 To learn how to add an intrinsic function, please see the `Extending
8730 LLVM Guide <ExtendingLLVM.html>`_.
8734 Variable Argument Handling Intrinsics
8735 -------------------------------------
8737 Variable argument support is defined in LLVM with the
8738 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
8739 functions. These functions are related to the similarly named macros
8740 defined in the ``<stdarg.h>`` header file.
8742 All of these functions operate on arguments that use a target-specific
8743 value type "``va_list``". The LLVM assembly language reference manual
8744 does not define what this type is, so all transformations should be
8745 prepared to handle these functions regardless of the type used.
8747 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
8748 variable argument handling intrinsic functions are used.
8750 .. code-block:: llvm
8752 ; This struct is different for every platform. For most platforms,
8753 ; it is merely an i8*.
8754 %struct.va_list = type { i8* }
8756 ; For Unix x86_64 platforms, va_list is the following struct:
8757 ; %struct.va_list = type { i32, i32, i8*, i8* }
8759 define i32 @test(i32 %X, ...) {
8760 ; Initialize variable argument processing
8761 %ap = alloca %struct.va_list
8762 %ap2 = bitcast %struct.va_list* %ap to i8*
8763 call void @llvm.va_start(i8* %ap2)
8765 ; Read a single integer argument
8766 %tmp = va_arg i8* %ap2, i32
8768 ; Demonstrate usage of llvm.va_copy and llvm.va_end
8770 %aq2 = bitcast i8** %aq to i8*
8771 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
8772 call void @llvm.va_end(i8* %aq2)
8774 ; Stop processing of arguments.
8775 call void @llvm.va_end(i8* %ap2)
8779 declare void @llvm.va_start(i8*)
8780 declare void @llvm.va_copy(i8*, i8*)
8781 declare void @llvm.va_end(i8*)
8785 '``llvm.va_start``' Intrinsic
8786 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8793 declare void @llvm.va_start(i8* <arglist>)
8798 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
8799 subsequent use by ``va_arg``.
8804 The argument is a pointer to a ``va_list`` element to initialize.
8809 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
8810 available in C. In a target-dependent way, it initializes the
8811 ``va_list`` element to which the argument points, so that the next call
8812 to ``va_arg`` will produce the first variable argument passed to the
8813 function. Unlike the C ``va_start`` macro, this intrinsic does not need
8814 to know the last argument of the function as the compiler can figure
8817 '``llvm.va_end``' Intrinsic
8818 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8825 declare void @llvm.va_end(i8* <arglist>)
8830 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
8831 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
8836 The argument is a pointer to a ``va_list`` to destroy.
8841 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
8842 available in C. In a target-dependent way, it destroys the ``va_list``
8843 element to which the argument points. Calls to
8844 :ref:`llvm.va_start <int_va_start>` and
8845 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
8850 '``llvm.va_copy``' Intrinsic
8851 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8858 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
8863 The '``llvm.va_copy``' intrinsic copies the current argument position
8864 from the source argument list to the destination argument list.
8869 The first argument is a pointer to a ``va_list`` element to initialize.
8870 The second argument is a pointer to a ``va_list`` element to copy from.
8875 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
8876 available in C. In a target-dependent way, it copies the source
8877 ``va_list`` element into the destination ``va_list`` element. This
8878 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
8879 arbitrarily complex and require, for example, memory allocation.
8881 Accurate Garbage Collection Intrinsics
8882 --------------------------------------
8884 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
8885 (GC) requires the frontend to generate code containing appropriate intrinsic
8886 calls and select an appropriate GC strategy which knows how to lower these
8887 intrinsics in a manner which is appropriate for the target collector.
8889 These intrinsics allow identification of :ref:`GC roots on the
8890 stack <int_gcroot>`, as well as garbage collector implementations that
8891 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
8892 Frontends for type-safe garbage collected languages should generate
8893 these intrinsics to make use of the LLVM garbage collectors. For more
8894 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
8896 Experimental Statepoint Intrinsics
8897 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8899 LLVM provides an second experimental set of intrinsics for describing garbage
8900 collection safepoints in compiled code. These intrinsics are an alternative
8901 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
8902 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
8903 differences in approach are covered in the `Garbage Collection with LLVM
8904 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
8905 described in :doc:`Statepoints`.
8909 '``llvm.gcroot``' Intrinsic
8910 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8917 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
8922 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
8923 the code generator, and allows some metadata to be associated with it.
8928 The first argument specifies the address of a stack object that contains
8929 the root pointer. The second pointer (which must be either a constant or
8930 a global value address) contains the meta-data to be associated with the
8936 At runtime, a call to this intrinsic stores a null pointer into the
8937 "ptrloc" location. At compile-time, the code generator generates
8938 information to allow the runtime to find the pointer at GC safe points.
8939 The '``llvm.gcroot``' intrinsic may only be used in a function which
8940 :ref:`specifies a GC algorithm <gc>`.
8944 '``llvm.gcread``' Intrinsic
8945 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8952 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
8957 The '``llvm.gcread``' intrinsic identifies reads of references from heap
8958 locations, allowing garbage collector implementations that require read
8964 The second argument is the address to read from, which should be an
8965 address allocated from the garbage collector. The first object is a
8966 pointer to the start of the referenced object, if needed by the language
8967 runtime (otherwise null).
8972 The '``llvm.gcread``' intrinsic has the same semantics as a load
8973 instruction, but may be replaced with substantially more complex code by
8974 the garbage collector runtime, as needed. The '``llvm.gcread``'
8975 intrinsic may only be used in a function which :ref:`specifies a GC
8980 '``llvm.gcwrite``' Intrinsic
8981 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8988 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
8993 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
8994 locations, allowing garbage collector implementations that require write
8995 barriers (such as generational or reference counting collectors).
9000 The first argument is the reference to store, the second is the start of
9001 the object to store it to, and the third is the address of the field of
9002 Obj to store to. If the runtime does not require a pointer to the
9003 object, Obj may be null.
9008 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
9009 instruction, but may be replaced with substantially more complex code by
9010 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
9011 intrinsic may only be used in a function which :ref:`specifies a GC
9014 Code Generator Intrinsics
9015 -------------------------
9017 These intrinsics are provided by LLVM to expose special features that
9018 may only be implemented with code generator support.
9020 '``llvm.returnaddress``' Intrinsic
9021 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9028 declare i8 *@llvm.returnaddress(i32 <level>)
9033 The '``llvm.returnaddress``' intrinsic attempts to compute a
9034 target-specific value indicating the return address of the current
9035 function or one of its callers.
9040 The argument to this intrinsic indicates which function to return the
9041 address for. Zero indicates the calling function, one indicates its
9042 caller, etc. The argument is **required** to be a constant integer
9048 The '``llvm.returnaddress``' intrinsic either returns a pointer
9049 indicating the return address of the specified call frame, or zero if it
9050 cannot be identified. The value returned by this intrinsic is likely to
9051 be incorrect or 0 for arguments other than zero, so it should only be
9052 used for debugging purposes.
9054 Note that calling this intrinsic does not prevent function inlining or
9055 other aggressive transformations, so the value returned may not be that
9056 of the obvious source-language caller.
9058 '``llvm.frameaddress``' Intrinsic
9059 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9066 declare i8* @llvm.frameaddress(i32 <level>)
9071 The '``llvm.frameaddress``' intrinsic attempts to return the
9072 target-specific frame pointer value for the specified stack frame.
9077 The argument to this intrinsic indicates which function to return the
9078 frame pointer for. Zero indicates the calling function, one indicates
9079 its caller, etc. The argument is **required** to be a constant integer
9085 The '``llvm.frameaddress``' intrinsic either returns a pointer
9086 indicating the frame address of the specified call frame, or zero if it
9087 cannot be identified. The value returned by this intrinsic is likely to
9088 be incorrect or 0 for arguments other than zero, so it should only be
9089 used for debugging purposes.
9091 Note that calling this intrinsic does not prevent function inlining or
9092 other aggressive transformations, so the value returned may not be that
9093 of the obvious source-language caller.
9095 '``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
9096 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9103 declare void @llvm.localescape(...)
9104 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
9109 The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
9110 allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
9111 live frame pointer to recover the address of the allocation. The offset is
9112 computed during frame layout of the caller of ``llvm.localescape``.
9117 All arguments to '``llvm.localescape``' must be pointers to static allocas or
9118 casts of static allocas. Each function can only call '``llvm.localescape``'
9119 once, and it can only do so from the entry block.
9121 The ``func`` argument to '``llvm.localrecover``' must be a constant
9122 bitcasted pointer to a function defined in the current module. The code
9123 generator cannot determine the frame allocation offset of functions defined in
9126 The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
9127 call frame that is currently live. The return value of '``llvm.localaddress``'
9128 is one way to produce such a value, but various runtimes also expose a suitable
9129 pointer in platform-specific ways.
9131 The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
9132 '``llvm.localescape``' to recover. It is zero-indexed.
9137 These intrinsics allow a group of functions to share access to a set of local
9138 stack allocations of a one parent function. The parent function may call the
9139 '``llvm.localescape``' intrinsic once from the function entry block, and the
9140 child functions can use '``llvm.localrecover``' to access the escaped allocas.
9141 The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
9142 the escaped allocas are allocated, which would break attempts to use
9143 '``llvm.localrecover``'.
9145 .. _int_read_register:
9146 .. _int_write_register:
9148 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
9149 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9156 declare i32 @llvm.read_register.i32(metadata)
9157 declare i64 @llvm.read_register.i64(metadata)
9158 declare void @llvm.write_register.i32(metadata, i32 @value)
9159 declare void @llvm.write_register.i64(metadata, i64 @value)
9165 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
9166 provides access to the named register. The register must be valid on
9167 the architecture being compiled to. The type needs to be compatible
9168 with the register being read.
9173 The '``llvm.read_register``' intrinsic returns the current value of the
9174 register, where possible. The '``llvm.write_register``' intrinsic sets
9175 the current value of the register, where possible.
9177 This is useful to implement named register global variables that need
9178 to always be mapped to a specific register, as is common practice on
9179 bare-metal programs including OS kernels.
9181 The compiler doesn't check for register availability or use of the used
9182 register in surrounding code, including inline assembly. Because of that,
9183 allocatable registers are not supported.
9185 Warning: So far it only works with the stack pointer on selected
9186 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
9187 work is needed to support other registers and even more so, allocatable
9192 '``llvm.stacksave``' Intrinsic
9193 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9200 declare i8* @llvm.stacksave()
9205 The '``llvm.stacksave``' intrinsic is used to remember the current state
9206 of the function stack, for use with
9207 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
9208 implementing language features like scoped automatic variable sized
9214 This intrinsic returns a opaque pointer value that can be passed to
9215 :ref:`llvm.stackrestore <int_stackrestore>`. When an
9216 ``llvm.stackrestore`` intrinsic is executed with a value saved from
9217 ``llvm.stacksave``, it effectively restores the state of the stack to
9218 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
9219 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
9220 were allocated after the ``llvm.stacksave`` was executed.
9222 .. _int_stackrestore:
9224 '``llvm.stackrestore``' Intrinsic
9225 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9232 declare void @llvm.stackrestore(i8* %ptr)
9237 The '``llvm.stackrestore``' intrinsic is used to restore the state of
9238 the function stack to the state it was in when the corresponding
9239 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
9240 useful for implementing language features like scoped automatic variable
9241 sized arrays in C99.
9246 See the description for :ref:`llvm.stacksave <int_stacksave>`.
9248 '``llvm.prefetch``' Intrinsic
9249 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9256 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
9261 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
9262 insert a prefetch instruction if supported; otherwise, it is a noop.
9263 Prefetches have no effect on the behavior of the program but can change
9264 its performance characteristics.
9269 ``address`` is the address to be prefetched, ``rw`` is the specifier
9270 determining if the fetch should be for a read (0) or write (1), and
9271 ``locality`` is a temporal locality specifier ranging from (0) - no
9272 locality, to (3) - extremely local keep in cache. The ``cache type``
9273 specifies whether the prefetch is performed on the data (1) or
9274 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
9275 arguments must be constant integers.
9280 This intrinsic does not modify the behavior of the program. In
9281 particular, prefetches cannot trap and do not produce a value. On
9282 targets that support this intrinsic, the prefetch can provide hints to
9283 the processor cache for better performance.
9285 '``llvm.pcmarker``' Intrinsic
9286 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9293 declare void @llvm.pcmarker(i32 <id>)
9298 The '``llvm.pcmarker``' intrinsic is a method to export a Program
9299 Counter (PC) in a region of code to simulators and other tools. The
9300 method is target specific, but it is expected that the marker will use
9301 exported symbols to transmit the PC of the marker. The marker makes no
9302 guarantees that it will remain with any specific instruction after
9303 optimizations. It is possible that the presence of a marker will inhibit
9304 optimizations. The intended use is to be inserted after optimizations to
9305 allow correlations of simulation runs.
9310 ``id`` is a numerical id identifying the marker.
9315 This intrinsic does not modify the behavior of the program. Backends
9316 that do not support this intrinsic may ignore it.
9318 '``llvm.readcyclecounter``' Intrinsic
9319 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9326 declare i64 @llvm.readcyclecounter()
9331 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
9332 counter register (or similar low latency, high accuracy clocks) on those
9333 targets that support it. On X86, it should map to RDTSC. On Alpha, it
9334 should map to RPCC. As the backing counters overflow quickly (on the
9335 order of 9 seconds on alpha), this should only be used for small
9341 When directly supported, reading the cycle counter should not modify any
9342 memory. Implementations are allowed to either return a application
9343 specific value or a system wide value. On backends without support, this
9344 is lowered to a constant 0.
9346 Note that runtime support may be conditional on the privilege-level code is
9347 running at and the host platform.
9349 '``llvm.clear_cache``' Intrinsic
9350 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9357 declare void @llvm.clear_cache(i8*, i8*)
9362 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
9363 in the specified range to the execution unit of the processor. On
9364 targets with non-unified instruction and data cache, the implementation
9365 flushes the instruction cache.
9370 On platforms with coherent instruction and data caches (e.g. x86), this
9371 intrinsic is a nop. On platforms with non-coherent instruction and data
9372 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
9373 instructions or a system call, if cache flushing requires special
9376 The default behavior is to emit a call to ``__clear_cache`` from the run
9379 This instrinsic does *not* empty the instruction pipeline. Modifications
9380 of the current function are outside the scope of the intrinsic.
9382 '``llvm.instrprof_increment``' Intrinsic
9383 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9390 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
9391 i32 <num-counters>, i32 <index>)
9396 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
9397 frontend for use with instrumentation based profiling. These will be
9398 lowered by the ``-instrprof`` pass to generate execution counts of a
9404 The first argument is a pointer to a global variable containing the
9405 name of the entity being instrumented. This should generally be the
9406 (mangled) function name for a set of counters.
9408 The second argument is a hash value that can be used by the consumer
9409 of the profile data to detect changes to the instrumented source, and
9410 the third is the number of counters associated with ``name``. It is an
9411 error if ``hash`` or ``num-counters`` differ between two instances of
9412 ``instrprof_increment`` that refer to the same name.
9414 The last argument refers to which of the counters for ``name`` should
9415 be incremented. It should be a value between 0 and ``num-counters``.
9420 This intrinsic represents an increment of a profiling counter. It will
9421 cause the ``-instrprof`` pass to generate the appropriate data
9422 structures and the code to increment the appropriate value, in a
9423 format that can be written out by a compiler runtime and consumed via
9424 the ``llvm-profdata`` tool.
9426 Standard C Library Intrinsics
9427 -----------------------------
9429 LLVM provides intrinsics for a few important standard C library
9430 functions. These intrinsics allow source-language front-ends to pass
9431 information about the alignment of the pointer arguments to the code
9432 generator, providing opportunity for more efficient code generation.
9436 '``llvm.memcpy``' Intrinsic
9437 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9442 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
9443 integer bit width and for different address spaces. Not all targets
9444 support all bit widths however.
9448 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9449 i32 <len>, i32 <align>, i1 <isvolatile>)
9450 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9451 i64 <len>, i32 <align>, i1 <isvolatile>)
9456 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9457 source location to the destination location.
9459 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
9460 intrinsics do not return a value, takes extra alignment/isvolatile
9461 arguments and the pointers can be in specified address spaces.
9466 The first argument is a pointer to the destination, the second is a
9467 pointer to the source. The third argument is an integer argument
9468 specifying the number of bytes to copy, the fourth argument is the
9469 alignment of the source and destination locations, and the fifth is a
9470 boolean indicating a volatile access.
9472 If the call to this intrinsic has an alignment value that is not 0 or 1,
9473 then the caller guarantees that both the source and destination pointers
9474 are aligned to that boundary.
9476 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
9477 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9478 very cleanly specified and it is unwise to depend on it.
9483 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9484 source location to the destination location, which are not allowed to
9485 overlap. It copies "len" bytes of memory over. If the argument is known
9486 to be aligned to some boundary, this can be specified as the fourth
9487 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
9489 '``llvm.memmove``' Intrinsic
9490 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9495 This is an overloaded intrinsic. You can use llvm.memmove on any integer
9496 bit width and for different address space. Not all targets support all
9501 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9502 i32 <len>, i32 <align>, i1 <isvolatile>)
9503 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9504 i64 <len>, i32 <align>, i1 <isvolatile>)
9509 The '``llvm.memmove.*``' intrinsics move a block of memory from the
9510 source location to the destination location. It is similar to the
9511 '``llvm.memcpy``' intrinsic but allows the two memory locations to
9514 Note that, unlike the standard libc function, the ``llvm.memmove.*``
9515 intrinsics do not return a value, takes extra alignment/isvolatile
9516 arguments and the pointers can be in specified address spaces.
9521 The first argument is a pointer to the destination, the second is a
9522 pointer to the source. The third argument is an integer argument
9523 specifying the number of bytes to copy, the fourth argument is the
9524 alignment of the source and destination locations, and the fifth is a
9525 boolean indicating a volatile access.
9527 If the call to this intrinsic has an alignment value that is not 0 or 1,
9528 then the caller guarantees that the source and destination pointers are
9529 aligned to that boundary.
9531 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
9532 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
9533 not very cleanly specified and it is unwise to depend on it.
9538 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
9539 source location to the destination location, which may overlap. It
9540 copies "len" bytes of memory over. If the argument is known to be
9541 aligned to some boundary, this can be specified as the fourth argument,
9542 otherwise it should be set to 0 or 1 (both meaning no alignment).
9544 '``llvm.memset.*``' Intrinsics
9545 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9550 This is an overloaded intrinsic. You can use llvm.memset on any integer
9551 bit width and for different address spaces. However, not all targets
9552 support all bit widths.
9556 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
9557 i32 <len>, i32 <align>, i1 <isvolatile>)
9558 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
9559 i64 <len>, i32 <align>, i1 <isvolatile>)
9564 The '``llvm.memset.*``' intrinsics fill a block of memory with a
9565 particular byte value.
9567 Note that, unlike the standard libc function, the ``llvm.memset``
9568 intrinsic does not return a value and takes extra alignment/volatile
9569 arguments. Also, the destination can be in an arbitrary address space.
9574 The first argument is a pointer to the destination to fill, the second
9575 is the byte value with which to fill it, the third argument is an
9576 integer argument specifying the number of bytes to fill, and the fourth
9577 argument is the known alignment of the destination location.
9579 If the call to this intrinsic has an alignment value that is not 0 or 1,
9580 then the caller guarantees that the destination pointer is aligned to
9583 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
9584 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9585 very cleanly specified and it is unwise to depend on it.
9590 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
9591 at the destination location. If the argument is known to be aligned to
9592 some boundary, this can be specified as the fourth argument, otherwise
9593 it should be set to 0 or 1 (both meaning no alignment).
9595 '``llvm.sqrt.*``' Intrinsic
9596 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9601 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
9602 floating point or vector of floating point type. Not all targets support
9607 declare float @llvm.sqrt.f32(float %Val)
9608 declare double @llvm.sqrt.f64(double %Val)
9609 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
9610 declare fp128 @llvm.sqrt.f128(fp128 %Val)
9611 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
9616 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
9617 returning the same value as the libm '``sqrt``' functions would. Unlike
9618 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
9619 negative numbers other than -0.0 (which allows for better optimization,
9620 because there is no need to worry about errno being set).
9621 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
9626 The argument and return value are floating point numbers of the same
9632 This function returns the sqrt of the specified operand if it is a
9633 nonnegative floating point number.
9635 '``llvm.powi.*``' Intrinsic
9636 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9641 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
9642 floating point or vector of floating point type. Not all targets support
9647 declare float @llvm.powi.f32(float %Val, i32 %power)
9648 declare double @llvm.powi.f64(double %Val, i32 %power)
9649 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
9650 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
9651 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
9656 The '``llvm.powi.*``' intrinsics return the first operand raised to the
9657 specified (positive or negative) power. The order of evaluation of
9658 multiplications is not defined. When a vector of floating point type is
9659 used, the second argument remains a scalar integer value.
9664 The second argument is an integer power, and the first is a value to
9665 raise to that power.
9670 This function returns the first value raised to the second power with an
9671 unspecified sequence of rounding operations.
9673 '``llvm.sin.*``' Intrinsic
9674 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9679 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
9680 floating point or vector of floating point type. Not all targets support
9685 declare float @llvm.sin.f32(float %Val)
9686 declare double @llvm.sin.f64(double %Val)
9687 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
9688 declare fp128 @llvm.sin.f128(fp128 %Val)
9689 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
9694 The '``llvm.sin.*``' intrinsics return the sine of the operand.
9699 The argument and return value are floating point numbers of the same
9705 This function returns the sine of the specified operand, returning the
9706 same values as the libm ``sin`` functions would, and handles error
9707 conditions in the same way.
9709 '``llvm.cos.*``' Intrinsic
9710 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9715 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
9716 floating point or vector of floating point type. Not all targets support
9721 declare float @llvm.cos.f32(float %Val)
9722 declare double @llvm.cos.f64(double %Val)
9723 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
9724 declare fp128 @llvm.cos.f128(fp128 %Val)
9725 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
9730 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
9735 The argument and return value are floating point numbers of the same
9741 This function returns the cosine of the specified operand, returning the
9742 same values as the libm ``cos`` functions would, and handles error
9743 conditions in the same way.
9745 '``llvm.pow.*``' Intrinsic
9746 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9751 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
9752 floating point or vector of floating point type. Not all targets support
9757 declare float @llvm.pow.f32(float %Val, float %Power)
9758 declare double @llvm.pow.f64(double %Val, double %Power)
9759 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
9760 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
9761 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
9766 The '``llvm.pow.*``' intrinsics return the first operand raised to the
9767 specified (positive or negative) power.
9772 The second argument is a floating point power, and the first is a value
9773 to raise to that power.
9778 This function returns the first value raised to the second power,
9779 returning the same values as the libm ``pow`` functions would, and
9780 handles error conditions in the same way.
9782 '``llvm.exp.*``' Intrinsic
9783 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9788 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
9789 floating point or vector of floating point type. Not all targets support
9794 declare float @llvm.exp.f32(float %Val)
9795 declare double @llvm.exp.f64(double %Val)
9796 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
9797 declare fp128 @llvm.exp.f128(fp128 %Val)
9798 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
9803 The '``llvm.exp.*``' intrinsics perform the exp function.
9808 The argument and return value are floating point numbers of the same
9814 This function returns the same values as the libm ``exp`` functions
9815 would, and handles error conditions in the same way.
9817 '``llvm.exp2.*``' Intrinsic
9818 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9823 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
9824 floating point or vector of floating point type. Not all targets support
9829 declare float @llvm.exp2.f32(float %Val)
9830 declare double @llvm.exp2.f64(double %Val)
9831 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
9832 declare fp128 @llvm.exp2.f128(fp128 %Val)
9833 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
9838 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
9843 The argument and return value are floating point numbers of the same
9849 This function returns the same values as the libm ``exp2`` functions
9850 would, and handles error conditions in the same way.
9852 '``llvm.log.*``' Intrinsic
9853 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9858 This is an overloaded intrinsic. You can use ``llvm.log`` on any
9859 floating point or vector of floating point type. Not all targets support
9864 declare float @llvm.log.f32(float %Val)
9865 declare double @llvm.log.f64(double %Val)
9866 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
9867 declare fp128 @llvm.log.f128(fp128 %Val)
9868 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
9873 The '``llvm.log.*``' intrinsics perform the log function.
9878 The argument and return value are floating point numbers of the same
9884 This function returns the same values as the libm ``log`` functions
9885 would, and handles error conditions in the same way.
9887 '``llvm.log10.*``' Intrinsic
9888 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9893 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
9894 floating point or vector of floating point type. Not all targets support
9899 declare float @llvm.log10.f32(float %Val)
9900 declare double @llvm.log10.f64(double %Val)
9901 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
9902 declare fp128 @llvm.log10.f128(fp128 %Val)
9903 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
9908 The '``llvm.log10.*``' intrinsics perform the log10 function.
9913 The argument and return value are floating point numbers of the same
9919 This function returns the same values as the libm ``log10`` functions
9920 would, and handles error conditions in the same way.
9922 '``llvm.log2.*``' Intrinsic
9923 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9928 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
9929 floating point or vector of floating point type. Not all targets support
9934 declare float @llvm.log2.f32(float %Val)
9935 declare double @llvm.log2.f64(double %Val)
9936 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
9937 declare fp128 @llvm.log2.f128(fp128 %Val)
9938 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
9943 The '``llvm.log2.*``' intrinsics perform the log2 function.
9948 The argument and return value are floating point numbers of the same
9954 This function returns the same values as the libm ``log2`` functions
9955 would, and handles error conditions in the same way.
9957 '``llvm.fma.*``' Intrinsic
9958 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9963 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
9964 floating point or vector of floating point type. Not all targets support
9969 declare float @llvm.fma.f32(float %a, float %b, float %c)
9970 declare double @llvm.fma.f64(double %a, double %b, double %c)
9971 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
9972 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
9973 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
9978 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
9984 The argument and return value are floating point numbers of the same
9990 This function returns the same values as the libm ``fma`` functions
9991 would, and does not set errno.
9993 '``llvm.fabs.*``' Intrinsic
9994 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9999 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
10000 floating point or vector of floating point type. Not all targets support
10005 declare float @llvm.fabs.f32(float %Val)
10006 declare double @llvm.fabs.f64(double %Val)
10007 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
10008 declare fp128 @llvm.fabs.f128(fp128 %Val)
10009 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
10014 The '``llvm.fabs.*``' intrinsics return the absolute value of the
10020 The argument and return value are floating point numbers of the same
10026 This function returns the same values as the libm ``fabs`` functions
10027 would, and handles error conditions in the same way.
10029 '``llvm.minnum.*``' Intrinsic
10030 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10035 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
10036 floating point or vector of floating point type. Not all targets support
10041 declare float @llvm.minnum.f32(float %Val0, float %Val1)
10042 declare double @llvm.minnum.f64(double %Val0, double %Val1)
10043 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10044 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
10045 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10050 The '``llvm.minnum.*``' intrinsics return the minimum of the two
10057 The arguments and return value are floating point numbers of the same
10063 Follows the IEEE-754 semantics for minNum, which also match for libm's
10066 If either operand is a NaN, returns the other non-NaN operand. Returns
10067 NaN only if both operands are NaN. If the operands compare equal,
10068 returns a value that compares equal to both operands. This means that
10069 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10071 '``llvm.maxnum.*``' Intrinsic
10072 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10077 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
10078 floating point or vector of floating point type. Not all targets support
10083 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
10084 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
10085 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10086 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
10087 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10092 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
10099 The arguments and return value are floating point numbers of the same
10104 Follows the IEEE-754 semantics for maxNum, which also match for libm's
10107 If either operand is a NaN, returns the other non-NaN operand. Returns
10108 NaN only if both operands are NaN. If the operands compare equal,
10109 returns a value that compares equal to both operands. This means that
10110 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10112 '``llvm.copysign.*``' Intrinsic
10113 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10118 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
10119 floating point or vector of floating point type. Not all targets support
10124 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
10125 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
10126 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
10127 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
10128 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
10133 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
10134 first operand and the sign of the second operand.
10139 The arguments and return value are floating point numbers of the same
10145 This function returns the same values as the libm ``copysign``
10146 functions would, and handles error conditions in the same way.
10148 '``llvm.floor.*``' Intrinsic
10149 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10154 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
10155 floating point or vector of floating point type. Not all targets support
10160 declare float @llvm.floor.f32(float %Val)
10161 declare double @llvm.floor.f64(double %Val)
10162 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
10163 declare fp128 @llvm.floor.f128(fp128 %Val)
10164 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
10169 The '``llvm.floor.*``' intrinsics return the floor of the operand.
10174 The argument and return value are floating point numbers of the same
10180 This function returns the same values as the libm ``floor`` functions
10181 would, and handles error conditions in the same way.
10183 '``llvm.ceil.*``' Intrinsic
10184 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10189 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
10190 floating point or vector of floating point type. Not all targets support
10195 declare float @llvm.ceil.f32(float %Val)
10196 declare double @llvm.ceil.f64(double %Val)
10197 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
10198 declare fp128 @llvm.ceil.f128(fp128 %Val)
10199 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
10204 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
10209 The argument and return value are floating point numbers of the same
10215 This function returns the same values as the libm ``ceil`` functions
10216 would, and handles error conditions in the same way.
10218 '``llvm.trunc.*``' Intrinsic
10219 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10224 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
10225 floating point or vector of floating point type. Not all targets support
10230 declare float @llvm.trunc.f32(float %Val)
10231 declare double @llvm.trunc.f64(double %Val)
10232 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
10233 declare fp128 @llvm.trunc.f128(fp128 %Val)
10234 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
10239 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
10240 nearest integer not larger in magnitude than the operand.
10245 The argument and return value are floating point numbers of the same
10251 This function returns the same values as the libm ``trunc`` functions
10252 would, and handles error conditions in the same way.
10254 '``llvm.rint.*``' Intrinsic
10255 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10260 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
10261 floating point or vector of floating point type. Not all targets support
10266 declare float @llvm.rint.f32(float %Val)
10267 declare double @llvm.rint.f64(double %Val)
10268 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
10269 declare fp128 @llvm.rint.f128(fp128 %Val)
10270 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
10275 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
10276 nearest integer. It may raise an inexact floating-point exception if the
10277 operand isn't an integer.
10282 The argument and return value are floating point numbers of the same
10288 This function returns the same values as the libm ``rint`` functions
10289 would, and handles error conditions in the same way.
10291 '``llvm.nearbyint.*``' Intrinsic
10292 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10297 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
10298 floating point or vector of floating point type. Not all targets support
10303 declare float @llvm.nearbyint.f32(float %Val)
10304 declare double @llvm.nearbyint.f64(double %Val)
10305 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
10306 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
10307 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
10312 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
10318 The argument and return value are floating point numbers of the same
10324 This function returns the same values as the libm ``nearbyint``
10325 functions would, and handles error conditions in the same way.
10327 '``llvm.round.*``' Intrinsic
10328 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10333 This is an overloaded intrinsic. You can use ``llvm.round`` on any
10334 floating point or vector of floating point type. Not all targets support
10339 declare float @llvm.round.f32(float %Val)
10340 declare double @llvm.round.f64(double %Val)
10341 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
10342 declare fp128 @llvm.round.f128(fp128 %Val)
10343 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
10348 The '``llvm.round.*``' intrinsics returns the operand rounded to the
10354 The argument and return value are floating point numbers of the same
10360 This function returns the same values as the libm ``round``
10361 functions would, and handles error conditions in the same way.
10363 Bit Manipulation Intrinsics
10364 ---------------------------
10366 LLVM provides intrinsics for a few important bit manipulation
10367 operations. These allow efficient code generation for some algorithms.
10369 '``llvm.bswap.*``' Intrinsics
10370 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10375 This is an overloaded intrinsic function. You can use bswap on any
10376 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
10380 declare i16 @llvm.bswap.i16(i16 <id>)
10381 declare i32 @llvm.bswap.i32(i32 <id>)
10382 declare i64 @llvm.bswap.i64(i64 <id>)
10387 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
10388 values with an even number of bytes (positive multiple of 16 bits).
10389 These are useful for performing operations on data that is not in the
10390 target's native byte order.
10395 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
10396 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
10397 intrinsic returns an i32 value that has the four bytes of the input i32
10398 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
10399 returned i32 will have its bytes in 3, 2, 1, 0 order. The
10400 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
10401 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
10404 '``llvm.ctpop.*``' Intrinsic
10405 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10410 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
10411 bit width, or on any vector with integer elements. Not all targets
10412 support all bit widths or vector types, however.
10416 declare i8 @llvm.ctpop.i8(i8 <src>)
10417 declare i16 @llvm.ctpop.i16(i16 <src>)
10418 declare i32 @llvm.ctpop.i32(i32 <src>)
10419 declare i64 @llvm.ctpop.i64(i64 <src>)
10420 declare i256 @llvm.ctpop.i256(i256 <src>)
10421 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
10426 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
10432 The only argument is the value to be counted. The argument may be of any
10433 integer type, or a vector with integer elements. The return type must
10434 match the argument type.
10439 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
10440 each element of a vector.
10442 '``llvm.ctlz.*``' Intrinsic
10443 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10448 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
10449 integer bit width, or any vector whose elements are integers. Not all
10450 targets support all bit widths or vector types, however.
10454 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
10455 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
10456 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
10457 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
10458 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
10459 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10464 The '``llvm.ctlz``' family of intrinsic functions counts the number of
10465 leading zeros in a variable.
10470 The first argument is the value to be counted. This argument may be of
10471 any integer type, or a vector with integer element type. The return
10472 type must match the first argument type.
10474 The second argument must be a constant and is a flag to indicate whether
10475 the intrinsic should ensure that a zero as the first argument produces a
10476 defined result. Historically some architectures did not provide a
10477 defined result for zero values as efficiently, and many algorithms are
10478 now predicated on avoiding zero-value inputs.
10483 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
10484 zeros in a variable, or within each element of the vector. If
10485 ``src == 0`` then the result is the size in bits of the type of ``src``
10486 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10487 ``llvm.ctlz(i32 2) = 30``.
10489 '``llvm.cttz.*``' Intrinsic
10490 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10495 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
10496 integer bit width, or any vector of integer elements. Not all targets
10497 support all bit widths or vector types, however.
10501 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
10502 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
10503 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
10504 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
10505 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
10506 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10511 The '``llvm.cttz``' family of intrinsic functions counts the number of
10517 The first argument is the value to be counted. This argument may be of
10518 any integer type, or a vector with integer element type. The return
10519 type must match the first argument type.
10521 The second argument must be a constant and is a flag to indicate whether
10522 the intrinsic should ensure that a zero as the first argument produces a
10523 defined result. Historically some architectures did not provide a
10524 defined result for zero values as efficiently, and many algorithms are
10525 now predicated on avoiding zero-value inputs.
10530 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
10531 zeros in a variable, or within each element of a vector. If ``src == 0``
10532 then the result is the size in bits of the type of ``src`` if
10533 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10534 ``llvm.cttz(2) = 1``.
10538 Arithmetic with Overflow Intrinsics
10539 -----------------------------------
10541 LLVM provides intrinsics for some arithmetic with overflow operations.
10543 '``llvm.sadd.with.overflow.*``' Intrinsics
10544 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10549 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
10550 on any integer bit width.
10554 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
10555 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10556 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
10561 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10562 a signed addition of the two arguments, and indicate whether an overflow
10563 occurred during the signed summation.
10568 The arguments (%a and %b) and the first element of the result structure
10569 may be of integer types of any bit width, but they must have the same
10570 bit width. The second element of the result structure must be of type
10571 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10577 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10578 a signed addition of the two variables. They return a structure --- the
10579 first element of which is the signed summation, and the second element
10580 of which is a bit specifying if the signed summation resulted in an
10586 .. code-block:: llvm
10588 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10589 %sum = extractvalue {i32, i1} %res, 0
10590 %obit = extractvalue {i32, i1} %res, 1
10591 br i1 %obit, label %overflow, label %normal
10593 '``llvm.uadd.with.overflow.*``' Intrinsics
10594 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10599 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
10600 on any integer bit width.
10604 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
10605 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10606 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
10611 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10612 an unsigned addition of the two arguments, and indicate whether a carry
10613 occurred during the unsigned summation.
10618 The arguments (%a and %b) and the first element of the result structure
10619 may be of integer types of any bit width, but they must have the same
10620 bit width. The second element of the result structure must be of type
10621 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10627 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10628 an unsigned addition of the two arguments. They return a structure --- the
10629 first element of which is the sum, and the second element of which is a
10630 bit specifying if the unsigned summation resulted in a carry.
10635 .. code-block:: llvm
10637 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10638 %sum = extractvalue {i32, i1} %res, 0
10639 %obit = extractvalue {i32, i1} %res, 1
10640 br i1 %obit, label %carry, label %normal
10642 '``llvm.ssub.with.overflow.*``' Intrinsics
10643 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10648 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
10649 on any integer bit width.
10653 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
10654 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10655 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
10660 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10661 a signed subtraction of the two arguments, and indicate whether an
10662 overflow occurred during the signed subtraction.
10667 The arguments (%a and %b) and the first element of the result structure
10668 may be of integer types of any bit width, but they must have the same
10669 bit width. The second element of the result structure must be of type
10670 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10676 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10677 a signed subtraction of the two arguments. They return a structure --- the
10678 first element of which is the subtraction, and the second element of
10679 which is a bit specifying if the signed subtraction resulted in an
10685 .. code-block:: llvm
10687 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10688 %sum = extractvalue {i32, i1} %res, 0
10689 %obit = extractvalue {i32, i1} %res, 1
10690 br i1 %obit, label %overflow, label %normal
10692 '``llvm.usub.with.overflow.*``' Intrinsics
10693 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10698 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
10699 on any integer bit width.
10703 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
10704 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10705 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
10710 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10711 an unsigned subtraction of the two arguments, and indicate whether an
10712 overflow occurred during the unsigned subtraction.
10717 The arguments (%a and %b) and the first element of the result structure
10718 may be of integer types of any bit width, but they must have the same
10719 bit width. The second element of the result structure must be of type
10720 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10726 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10727 an unsigned subtraction of the two arguments. They return a structure ---
10728 the first element of which is the subtraction, and the second element of
10729 which is a bit specifying if the unsigned subtraction resulted in an
10735 .. code-block:: llvm
10737 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10738 %sum = extractvalue {i32, i1} %res, 0
10739 %obit = extractvalue {i32, i1} %res, 1
10740 br i1 %obit, label %overflow, label %normal
10742 '``llvm.smul.with.overflow.*``' Intrinsics
10743 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10748 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
10749 on any integer bit width.
10753 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
10754 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10755 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
10760 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10761 a signed multiplication of the two arguments, and indicate whether an
10762 overflow occurred during the signed multiplication.
10767 The arguments (%a and %b) and the first element of the result structure
10768 may be of integer types of any bit width, but they must have the same
10769 bit width. The second element of the result structure must be of type
10770 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10776 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10777 a signed multiplication of the two arguments. They return a structure ---
10778 the first element of which is the multiplication, and the second element
10779 of which is a bit specifying if the signed multiplication resulted in an
10785 .. code-block:: llvm
10787 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10788 %sum = extractvalue {i32, i1} %res, 0
10789 %obit = extractvalue {i32, i1} %res, 1
10790 br i1 %obit, label %overflow, label %normal
10792 '``llvm.umul.with.overflow.*``' Intrinsics
10793 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10798 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
10799 on any integer bit width.
10803 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
10804 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10805 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
10810 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10811 a unsigned multiplication of the two arguments, and indicate whether an
10812 overflow occurred during the unsigned multiplication.
10817 The arguments (%a and %b) and the first element of the result structure
10818 may be of integer types of any bit width, but they must have the same
10819 bit width. The second element of the result structure must be of type
10820 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10826 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10827 an unsigned multiplication of the two arguments. They return a structure ---
10828 the first element of which is the multiplication, and the second
10829 element of which is a bit specifying if the unsigned multiplication
10830 resulted in an overflow.
10835 .. code-block:: llvm
10837 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10838 %sum = extractvalue {i32, i1} %res, 0
10839 %obit = extractvalue {i32, i1} %res, 1
10840 br i1 %obit, label %overflow, label %normal
10842 Specialised Arithmetic Intrinsics
10843 ---------------------------------
10845 '``llvm.canonicalize.*``' Intrinsic
10846 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10853 declare float @llvm.canonicalize.f32(float %a)
10854 declare double @llvm.canonicalize.f64(double %b)
10859 The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
10860 encoding of a floating point number. This canonicalization is useful for
10861 implementing certain numeric primitives such as frexp. The canonical encoding is
10862 defined by IEEE-754-2008 to be:
10866 2.1.8 canonical encoding: The preferred encoding of a floating-point
10867 representation in a format. Applied to declets, significands of finite
10868 numbers, infinities, and NaNs, especially in decimal formats.
10870 This operation can also be considered equivalent to the IEEE-754-2008
10871 conversion of a floating-point value to the same format. NaNs are handled
10872 according to section 6.2.
10874 Examples of non-canonical encodings:
10876 - x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
10877 converted to a canonical representation per hardware-specific protocol.
10878 - Many normal decimal floating point numbers have non-canonical alternative
10880 - Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
10881 These are treated as non-canonical encodings of zero and with be flushed to
10882 a zero of the same sign by this operation.
10884 Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
10885 default exception handling must signal an invalid exception, and produce a
10888 This function should always be implementable as multiplication by 1.0, provided
10889 that the compiler does not constant fold the operation. Likewise, division by
10890 1.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
10891 -0.0 is also sufficient provided that the rounding mode is not -Infinity.
10893 ``@llvm.canonicalize`` must preserve the equality relation. That is:
10895 - ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
10896 - ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
10899 Additionally, the sign of zero must be conserved:
10900 ``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
10902 The payload bits of a NaN must be conserved, with two exceptions.
10903 First, environments which use only a single canonical representation of NaN
10904 must perform said canonicalization. Second, SNaNs must be quieted per the
10907 The canonicalization operation may be optimized away if:
10909 - The input is known to be canonical. For example, it was produced by a
10910 floating-point operation that is required by the standard to be canonical.
10911 - The result is consumed only by (or fused with) other floating-point
10912 operations. That is, the bits of the floating point value are not examined.
10914 '``llvm.fmuladd.*``' Intrinsic
10915 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10922 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
10923 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
10928 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
10929 expressions that can be fused if the code generator determines that (a) the
10930 target instruction set has support for a fused operation, and (b) that the
10931 fused operation is more efficient than the equivalent, separate pair of mul
10932 and add instructions.
10937 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
10938 multiplicands, a and b, and an addend c.
10947 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
10949 is equivalent to the expression a \* b + c, except that rounding will
10950 not be performed between the multiplication and addition steps if the
10951 code generator fuses the operations. Fusion is not guaranteed, even if
10952 the target platform supports it. If a fused multiply-add is required the
10953 corresponding llvm.fma.\* intrinsic function should be used
10954 instead. This never sets errno, just as '``llvm.fma.*``'.
10959 .. code-block:: llvm
10961 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
10964 '``llvm.uabsdiff.*``' and '``llvm.sabsdiff.*``' Intrinsics
10965 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10969 This is an overloaded intrinsic. The loaded data is a vector of any integer bit width.
10971 .. code-block:: llvm
10973 declare <4 x integer> @llvm.uabsdiff.v4i32(<4 x integer> %a, <4 x integer> %b)
10979 The ``llvm.uabsdiff`` intrinsic returns a vector result of the absolute difference
10980 of the two operands, treating them both as unsigned integers. The intermediate
10981 calculations are computed using infinitely precise unsigned arithmetic. The final
10982 result will be truncated to the given type.
10984 The ``llvm.sabsdiff`` intrinsic returns a vector result of the absolute difference of
10985 the two operands, treating them both as signed integers. If the result overflows, the
10986 behavior is undefined.
10990 These intrinsics are primarily used during the code generation stage of compilation.
10991 They are generated by compiler passes such as the Loop and SLP vectorizers. It is not
10992 recommended for users to create them manually.
10997 Both intrinsics take two integer of the same bitwidth.
11004 call <4 x i32> @llvm.uabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11008 %1 = zext <4 x i32> %a to <4 x i64>
11009 %2 = zext <4 x i32> %b to <4 x i64>
11010 %sub = sub <4 x i64> %1, %2
11011 %trunc = trunc <4 x i64> to <4 x i32>
11013 and the expression::
11015 call <4 x i32> @llvm.sabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11019 %sub = sub nsw <4 x i32> %a, %b
11020 %ispos = icmp sge <4 x i32> %sub, zeroinitializer
11021 %neg = sub nsw <4 x i32> zeroinitializer, %sub
11022 %1 = select <4 x i1> %ispos, <4 x i32> %sub, <4 x i32> %neg
11025 Half Precision Floating Point Intrinsics
11026 ----------------------------------------
11028 For most target platforms, half precision floating point is a
11029 storage-only format. This means that it is a dense encoding (in memory)
11030 but does not support computation in the format.
11032 This means that code must first load the half-precision floating point
11033 value as an i16, then convert it to float with
11034 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
11035 then be performed on the float value (including extending to double
11036 etc). To store the value back to memory, it is first converted to float
11037 if needed, then converted to i16 with
11038 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
11041 .. _int_convert_to_fp16:
11043 '``llvm.convert.to.fp16``' Intrinsic
11044 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11051 declare i16 @llvm.convert.to.fp16.f32(float %a)
11052 declare i16 @llvm.convert.to.fp16.f64(double %a)
11057 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11058 conventional floating point type to half precision floating point format.
11063 The intrinsic function contains single argument - the value to be
11069 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11070 conventional floating point format to half precision floating point format. The
11071 return value is an ``i16`` which contains the converted number.
11076 .. code-block:: llvm
11078 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
11079 store i16 %res, i16* @x, align 2
11081 .. _int_convert_from_fp16:
11083 '``llvm.convert.from.fp16``' Intrinsic
11084 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11091 declare float @llvm.convert.from.fp16.f32(i16 %a)
11092 declare double @llvm.convert.from.fp16.f64(i16 %a)
11097 The '``llvm.convert.from.fp16``' intrinsic function performs a
11098 conversion from half precision floating point format to single precision
11099 floating point format.
11104 The intrinsic function contains single argument - the value to be
11110 The '``llvm.convert.from.fp16``' intrinsic function performs a
11111 conversion from half single precision floating point format to single
11112 precision floating point format. The input half-float value is
11113 represented by an ``i16`` value.
11118 .. code-block:: llvm
11120 %a = load i16, i16* @x, align 2
11121 %res = call float @llvm.convert.from.fp16(i16 %a)
11123 .. _dbg_intrinsics:
11125 Debugger Intrinsics
11126 -------------------
11128 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
11129 prefix), are described in the `LLVM Source Level
11130 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
11133 Exception Handling Intrinsics
11134 -----------------------------
11136 The LLVM exception handling intrinsics (which all start with
11137 ``llvm.eh.`` prefix), are described in the `LLVM Exception
11138 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
11140 .. _int_trampoline:
11142 Trampoline Intrinsics
11143 ---------------------
11145 These intrinsics make it possible to excise one parameter, marked with
11146 the :ref:`nest <nest>` attribute, from a function. The result is a
11147 callable function pointer lacking the nest parameter - the caller does
11148 not need to provide a value for it. Instead, the value to use is stored
11149 in advance in a "trampoline", a block of memory usually allocated on the
11150 stack, which also contains code to splice the nest value into the
11151 argument list. This is used to implement the GCC nested function address
11154 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
11155 then the resulting function pointer has signature ``i32 (i32, i32)*``.
11156 It can be created as follows:
11158 .. code-block:: llvm
11160 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
11161 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
11162 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
11163 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
11164 %fp = bitcast i8* %p to i32 (i32, i32)*
11166 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
11167 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
11171 '``llvm.init.trampoline``' Intrinsic
11172 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11179 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
11184 This fills the memory pointed to by ``tramp`` with executable code,
11185 turning it into a trampoline.
11190 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
11191 pointers. The ``tramp`` argument must point to a sufficiently large and
11192 sufficiently aligned block of memory; this memory is written to by the
11193 intrinsic. Note that the size and the alignment are target-specific -
11194 LLVM currently provides no portable way of determining them, so a
11195 front-end that generates this intrinsic needs to have some
11196 target-specific knowledge. The ``func`` argument must hold a function
11197 bitcast to an ``i8*``.
11202 The block of memory pointed to by ``tramp`` is filled with target
11203 dependent code, turning it into a function. Then ``tramp`` needs to be
11204 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
11205 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
11206 function's signature is the same as that of ``func`` with any arguments
11207 marked with the ``nest`` attribute removed. At most one such ``nest``
11208 argument is allowed, and it must be of pointer type. Calling the new
11209 function is equivalent to calling ``func`` with the same argument list,
11210 but with ``nval`` used for the missing ``nest`` argument. If, after
11211 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
11212 modified, then the effect of any later call to the returned function
11213 pointer is undefined.
11217 '``llvm.adjust.trampoline``' Intrinsic
11218 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11225 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
11230 This performs any required machine-specific adjustment to the address of
11231 a trampoline (passed as ``tramp``).
11236 ``tramp`` must point to a block of memory which already has trampoline
11237 code filled in by a previous call to
11238 :ref:`llvm.init.trampoline <int_it>`.
11243 On some architectures the address of the code to be executed needs to be
11244 different than the address where the trampoline is actually stored. This
11245 intrinsic returns the executable address corresponding to ``tramp``
11246 after performing the required machine specific adjustments. The pointer
11247 returned can then be :ref:`bitcast and executed <int_trampoline>`.
11249 .. _int_mload_mstore:
11251 Masked Vector Load and Store Intrinsics
11252 ---------------------------------------
11254 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.
11258 '``llvm.masked.load.*``' Intrinsics
11259 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11263 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
11267 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11268 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11273 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.
11279 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.
11285 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.
11286 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.
11291 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
11293 ;; The result of the two following instructions is identical aside from potential memory access exception
11294 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
11295 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
11299 '``llvm.masked.store.*``' Intrinsics
11300 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11304 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
11308 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
11309 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
11314 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.
11319 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.
11325 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.
11326 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.
11330 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
11332 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
11333 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
11334 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
11335 store <16 x float> %res, <16 x float>* %ptr, align 4
11338 Masked Vector Gather and Scatter Intrinsics
11339 -------------------------------------------
11341 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.
11345 '``llvm.masked.gather.*``' Intrinsics
11346 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11350 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.
11354 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11355 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11360 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.
11366 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.
11372 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.
11373 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.
11378 %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>)
11380 ;; The gather with all-true mask is equivalent to the following instruction sequence
11381 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
11382 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
11383 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
11384 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
11386 %val0 = load double, double* %ptr0, align 8
11387 %val1 = load double, double* %ptr1, align 8
11388 %val2 = load double, double* %ptr2, align 8
11389 %val3 = load double, double* %ptr3, align 8
11391 %vec0 = insertelement <4 x double>undef, %val0, 0
11392 %vec01 = insertelement <4 x double>%vec0, %val1, 1
11393 %vec012 = insertelement <4 x double>%vec01, %val2, 2
11394 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
11398 '``llvm.masked.scatter.*``' Intrinsics
11399 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11403 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.
11407 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
11408 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
11413 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.
11418 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.
11424 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.
11428 ;; This instruction unconditionaly stores data vector in multiple addresses
11429 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
11431 ;; It is equivalent to a list of scalar stores
11432 %val0 = extractelement <8 x i32> %value, i32 0
11433 %val1 = extractelement <8 x i32> %value, i32 1
11435 %val7 = extractelement <8 x i32> %value, i32 7
11436 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
11437 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
11439 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
11440 ;; Note: the order of the following stores is important when they overlap:
11441 store i32 %val0, i32* %ptr0, align 4
11442 store i32 %val1, i32* %ptr1, align 4
11444 store i32 %val7, i32* %ptr7, align 4
11450 This class of intrinsics provides information about the lifetime of
11451 memory objects and ranges where variables are immutable.
11455 '``llvm.lifetime.start``' Intrinsic
11456 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11463 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
11468 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
11474 The first argument is a constant integer representing the size of the
11475 object, or -1 if it is variable sized. The second argument is a pointer
11481 This intrinsic indicates that before this point in the code, the value
11482 of the memory pointed to by ``ptr`` is dead. This means that it is known
11483 to never be used and has an undefined value. A load from the pointer
11484 that precedes this intrinsic can be replaced with ``'undef'``.
11488 '``llvm.lifetime.end``' Intrinsic
11489 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11496 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
11501 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
11507 The first argument is a constant integer representing the size of the
11508 object, or -1 if it is variable sized. The second argument is a pointer
11514 This intrinsic indicates that after this point in the code, the value of
11515 the memory pointed to by ``ptr`` is dead. This means that it is known to
11516 never be used and has an undefined value. Any stores into the memory
11517 object following this intrinsic may be removed as dead.
11519 '``llvm.invariant.start``' Intrinsic
11520 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11527 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
11532 The '``llvm.invariant.start``' intrinsic specifies that the contents of
11533 a memory object will not change.
11538 The first argument is a constant integer representing the size of the
11539 object, or -1 if it is variable sized. The second argument is a pointer
11545 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
11546 the return value, the referenced memory location is constant and
11549 '``llvm.invariant.end``' Intrinsic
11550 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11557 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
11562 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
11563 memory object are mutable.
11568 The first argument is the matching ``llvm.invariant.start`` intrinsic.
11569 The second argument is a constant integer representing the size of the
11570 object, or -1 if it is variable sized and the third argument is a
11571 pointer to the object.
11576 This intrinsic indicates that the memory is mutable again.
11578 '``llvm.invariant.group.barrier``' Intrinsic
11579 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11586 declare i8* @llvm.invariant.group.barrier(i8* <ptr>)
11591 The '``llvm.invariant.group.barrier``' intrinsic can be used when an invariant
11592 established by invariant.group metadata no longer holds, to obtain a new pointer
11593 value that does not carry the invariant information.
11599 The ``llvm.invariant.group.barrier`` takes only one argument, which is
11600 the pointer to the memory for which the ``invariant.group`` no longer holds.
11605 Returns another pointer that aliases its argument but which is considered different
11606 for the purposes of ``load``/``store`` ``invariant.group`` metadata.
11611 This class of intrinsics is designed to be generic and has no specific
11614 '``llvm.var.annotation``' Intrinsic
11615 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11622 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11627 The '``llvm.var.annotation``' intrinsic.
11632 The first argument is a pointer to a value, the second is a pointer to a
11633 global string, the third is a pointer to a global string which is the
11634 source file name, and the last argument is the line number.
11639 This intrinsic allows annotation of local variables with arbitrary
11640 strings. This can be useful for special purpose optimizations that want
11641 to look for these annotations. These have no other defined use; they are
11642 ignored by code generation and optimization.
11644 '``llvm.ptr.annotation.*``' Intrinsic
11645 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11650 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
11651 pointer to an integer of any width. *NOTE* you must specify an address space for
11652 the pointer. The identifier for the default address space is the integer
11657 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11658 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
11659 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
11660 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
11661 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
11666 The '``llvm.ptr.annotation``' intrinsic.
11671 The first argument is a pointer to an integer value of arbitrary bitwidth
11672 (result of some expression), the second is a pointer to a global string, the
11673 third is a pointer to a global string which is the source file name, and the
11674 last argument is the line number. It returns the value of the first argument.
11679 This intrinsic allows annotation of a pointer to an integer with arbitrary
11680 strings. This can be useful for special purpose optimizations that want to look
11681 for these annotations. These have no other defined use; they are ignored by code
11682 generation and optimization.
11684 '``llvm.annotation.*``' Intrinsic
11685 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11690 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
11691 any integer bit width.
11695 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
11696 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
11697 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
11698 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
11699 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
11704 The '``llvm.annotation``' intrinsic.
11709 The first argument is an integer value (result of some expression), the
11710 second is a pointer to a global string, the third is a pointer to a
11711 global string which is the source file name, and the last argument is
11712 the line number. It returns the value of the first argument.
11717 This intrinsic allows annotations to be put on arbitrary expressions
11718 with arbitrary strings. This can be useful for special purpose
11719 optimizations that want to look for these annotations. These have no
11720 other defined use; they are ignored by code generation and optimization.
11722 '``llvm.trap``' Intrinsic
11723 ^^^^^^^^^^^^^^^^^^^^^^^^^
11730 declare void @llvm.trap() noreturn nounwind
11735 The '``llvm.trap``' intrinsic.
11745 This intrinsic is lowered to the target dependent trap instruction. If
11746 the target does not have a trap instruction, this intrinsic will be
11747 lowered to a call of the ``abort()`` function.
11749 '``llvm.debugtrap``' Intrinsic
11750 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11757 declare void @llvm.debugtrap() nounwind
11762 The '``llvm.debugtrap``' intrinsic.
11772 This intrinsic is lowered to code which is intended to cause an
11773 execution trap with the intention of requesting the attention of a
11776 '``llvm.stackprotector``' Intrinsic
11777 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11784 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
11789 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
11790 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
11791 is placed on the stack before local variables.
11796 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
11797 The first argument is the value loaded from the stack guard
11798 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
11799 enough space to hold the value of the guard.
11804 This intrinsic causes the prologue/epilogue inserter to force the position of
11805 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
11806 to ensure that if a local variable on the stack is overwritten, it will destroy
11807 the value of the guard. When the function exits, the guard on the stack is
11808 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
11809 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
11810 calling the ``__stack_chk_fail()`` function.
11812 '``llvm.stackprotectorcheck``' Intrinsic
11813 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11820 declare void @llvm.stackprotectorcheck(i8** <guard>)
11825 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
11826 created stack protector and if they are not equal calls the
11827 ``__stack_chk_fail()`` function.
11832 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
11833 the variable ``@__stack_chk_guard``.
11838 This intrinsic is provided to perform the stack protector check by comparing
11839 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
11840 values do not match call the ``__stack_chk_fail()`` function.
11842 The reason to provide this as an IR level intrinsic instead of implementing it
11843 via other IR operations is that in order to perform this operation at the IR
11844 level without an intrinsic, one would need to create additional basic blocks to
11845 handle the success/failure cases. This makes it difficult to stop the stack
11846 protector check from disrupting sibling tail calls in Codegen. With this
11847 intrinsic, we are able to generate the stack protector basic blocks late in
11848 codegen after the tail call decision has occurred.
11850 '``llvm.objectsize``' Intrinsic
11851 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11858 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
11859 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
11864 The ``llvm.objectsize`` intrinsic is designed to provide information to
11865 the optimizers to determine at compile time whether a) an operation
11866 (like memcpy) will overflow a buffer that corresponds to an object, or
11867 b) that a runtime check for overflow isn't necessary. An object in this
11868 context means an allocation of a specific class, structure, array, or
11874 The ``llvm.objectsize`` intrinsic takes two arguments. The first
11875 argument is a pointer to or into the ``object``. The second argument is
11876 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
11877 or -1 (if false) when the object size is unknown. The second argument
11878 only accepts constants.
11883 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
11884 the size of the object concerned. If the size cannot be determined at
11885 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
11886 on the ``min`` argument).
11888 '``llvm.expect``' Intrinsic
11889 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11894 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
11899 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
11900 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
11901 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
11906 The ``llvm.expect`` intrinsic provides information about expected (the
11907 most probable) value of ``val``, which can be used by optimizers.
11912 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
11913 a value. The second argument is an expected value, this needs to be a
11914 constant value, variables are not allowed.
11919 This intrinsic is lowered to the ``val``.
11923 '``llvm.assume``' Intrinsic
11924 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11931 declare void @llvm.assume(i1 %cond)
11936 The ``llvm.assume`` allows the optimizer to assume that the provided
11937 condition is true. This information can then be used in simplifying other parts
11943 The condition which the optimizer may assume is always true.
11948 The intrinsic allows the optimizer to assume that the provided condition is
11949 always true whenever the control flow reaches the intrinsic call. No code is
11950 generated for this intrinsic, and instructions that contribute only to the
11951 provided condition are not used for code generation. If the condition is
11952 violated during execution, the behavior is undefined.
11954 Note that the optimizer might limit the transformations performed on values
11955 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
11956 only used to form the intrinsic's input argument. This might prove undesirable
11957 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
11958 sufficient overall improvement in code quality. For this reason,
11959 ``llvm.assume`` should not be used to document basic mathematical invariants
11960 that the optimizer can otherwise deduce or facts that are of little use to the
11965 '``llvm.bitset.test``' Intrinsic
11966 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11973 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
11979 The first argument is a pointer to be tested. The second argument is a
11980 metadata object representing an identifier for a :doc:`bitset <BitSets>`.
11985 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
11986 member of the given bitset.
11988 '``llvm.donothing``' Intrinsic
11989 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11996 declare void @llvm.donothing() nounwind readnone
12001 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
12002 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
12003 with an invoke instruction.
12013 This intrinsic does nothing, and it's removed by optimizers and ignored
12016 Stack Map Intrinsics
12017 --------------------
12019 LLVM provides experimental intrinsics to support runtime patching
12020 mechanisms commonly desired in dynamic language JITs. These intrinsics
12021 are described in :doc:`StackMaps`.