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``), unless they're overriden with
1481 callsite specific attributes.
1482 - An operand bundle at a call site cannot change the implementation
1483 of the called function. Inter-procedural optimizations work as
1484 usual as long as they take into account the first two properties.
1488 Module-Level Inline Assembly
1489 ----------------------------
1491 Modules may contain "module-level inline asm" blocks, which corresponds
1492 to the GCC "file scope inline asm" blocks. These blocks are internally
1493 concatenated by LLVM and treated as a single unit, but may be separated
1494 in the ``.ll`` file if desired. The syntax is very simple:
1496 .. code-block:: llvm
1498 module asm "inline asm code goes here"
1499 module asm "more can go here"
1501 The strings can contain any character by escaping non-printable
1502 characters. The escape sequence used is simply "\\xx" where "xx" is the
1503 two digit hex code for the number.
1505 Note that the assembly string *must* be parseable by LLVM's integrated assembler
1506 (unless it is disabled), even when emitting a ``.s`` file.
1508 .. _langref_datalayout:
1513 A module may specify a target specific data layout string that specifies
1514 how data is to be laid out in memory. The syntax for the data layout is
1517 .. code-block:: llvm
1519 target datalayout = "layout specification"
1521 The *layout specification* consists of a list of specifications
1522 separated by the minus sign character ('-'). Each specification starts
1523 with a letter and may include other information after the letter to
1524 define some aspect of the data layout. The specifications accepted are
1528 Specifies that the target lays out data in big-endian form. That is,
1529 the bits with the most significance have the lowest address
1532 Specifies that the target lays out data in little-endian form. That
1533 is, the bits with the least significance have the lowest address
1536 Specifies the natural alignment of the stack in bits. Alignment
1537 promotion of stack variables is limited to the natural stack
1538 alignment to avoid dynamic stack realignment. The stack alignment
1539 must be a multiple of 8-bits. If omitted, the natural stack
1540 alignment defaults to "unspecified", which does not prevent any
1541 alignment promotions.
1542 ``p[n]:<size>:<abi>:<pref>``
1543 This specifies the *size* of a pointer and its ``<abi>`` and
1544 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1545 bits. The address space, ``n``, is optional, and if not specified,
1546 denotes the default address space 0. The value of ``n`` must be
1547 in the range [1,2^23).
1548 ``i<size>:<abi>:<pref>``
1549 This specifies the alignment for an integer type of a given bit
1550 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1551 ``v<size>:<abi>:<pref>``
1552 This specifies the alignment for a vector type of a given bit
1554 ``f<size>:<abi>:<pref>``
1555 This specifies the alignment for a floating point type of a given bit
1556 ``<size>``. Only values of ``<size>`` that are supported by the target
1557 will work. 32 (float) and 64 (double) are supported on all targets; 80
1558 or 128 (different flavors of long double) are also supported on some
1561 This specifies the alignment for an object of aggregate type.
1563 If present, specifies that llvm names are mangled in the output. The
1566 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1567 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1568 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1569 symbols get a ``_`` prefix.
1570 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1571 functions also get a suffix based on the frame size.
1572 * ``x``: Windows x86 COFF prefix: Similar to Windows COFF, but use a ``_``
1573 prefix for ``__cdecl`` functions.
1574 ``n<size1>:<size2>:<size3>...``
1575 This specifies a set of native integer widths for the target CPU in
1576 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1577 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1578 this set are considered to support most general arithmetic operations
1581 On every specification that takes a ``<abi>:<pref>``, specifying the
1582 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1583 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1585 When constructing the data layout for a given target, LLVM starts with a
1586 default set of specifications which are then (possibly) overridden by
1587 the specifications in the ``datalayout`` keyword. The default
1588 specifications are given in this list:
1590 - ``E`` - big endian
1591 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1592 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1593 same as the default address space.
1594 - ``S0`` - natural stack alignment is unspecified
1595 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1596 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1597 - ``i16:16:16`` - i16 is 16-bit aligned
1598 - ``i32:32:32`` - i32 is 32-bit aligned
1599 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1600 alignment of 64-bits
1601 - ``f16:16:16`` - half is 16-bit aligned
1602 - ``f32:32:32`` - float is 32-bit aligned
1603 - ``f64:64:64`` - double is 64-bit aligned
1604 - ``f128:128:128`` - quad is 128-bit aligned
1605 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1606 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1607 - ``a:0:64`` - aggregates are 64-bit aligned
1609 When LLVM is determining the alignment for a given type, it uses the
1612 #. If the type sought is an exact match for one of the specifications,
1613 that specification is used.
1614 #. If no match is found, and the type sought is an integer type, then
1615 the smallest integer type that is larger than the bitwidth of the
1616 sought type is used. If none of the specifications are larger than
1617 the bitwidth then the largest integer type is used. For example,
1618 given the default specifications above, the i7 type will use the
1619 alignment of i8 (next largest) while both i65 and i256 will use the
1620 alignment of i64 (largest specified).
1621 #. If no match is found, and the type sought is a vector type, then the
1622 largest vector type that is smaller than the sought vector type will
1623 be used as a fall back. This happens because <128 x double> can be
1624 implemented in terms of 64 <2 x double>, for example.
1626 The function of the data layout string may not be what you expect.
1627 Notably, this is not a specification from the frontend of what alignment
1628 the code generator should use.
1630 Instead, if specified, the target data layout is required to match what
1631 the ultimate *code generator* expects. This string is used by the
1632 mid-level optimizers to improve code, and this only works if it matches
1633 what the ultimate code generator uses. There is no way to generate IR
1634 that does not embed this target-specific detail into the IR. If you
1635 don't specify the string, the default specifications will be used to
1636 generate a Data Layout and the optimization phases will operate
1637 accordingly and introduce target specificity into the IR with respect to
1638 these default specifications.
1645 A module may specify a target triple string that describes the target
1646 host. The syntax for the target triple is simply:
1648 .. code-block:: llvm
1650 target triple = "x86_64-apple-macosx10.7.0"
1652 The *target triple* string consists of a series of identifiers delimited
1653 by the minus sign character ('-'). The canonical forms are:
1657 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1658 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1660 This information is passed along to the backend so that it generates
1661 code for the proper architecture. It's possible to override this on the
1662 command line with the ``-mtriple`` command line option.
1664 .. _pointeraliasing:
1666 Pointer Aliasing Rules
1667 ----------------------
1669 Any memory access must be done through a pointer value associated with
1670 an address range of the memory access, otherwise the behavior is
1671 undefined. Pointer values are associated with address ranges according
1672 to the following rules:
1674 - A pointer value is associated with the addresses associated with any
1675 value it is *based* on.
1676 - An address of a global variable is associated with the address range
1677 of the variable's storage.
1678 - The result value of an allocation instruction is associated with the
1679 address range of the allocated storage.
1680 - A null pointer in the default address-space is associated with no
1682 - An integer constant other than zero or a pointer value returned from
1683 a function not defined within LLVM may be associated with address
1684 ranges allocated through mechanisms other than those provided by
1685 LLVM. Such ranges shall not overlap with any ranges of addresses
1686 allocated by mechanisms provided by LLVM.
1688 A pointer value is *based* on another pointer value according to the
1691 - A pointer value formed from a ``getelementptr`` operation is *based*
1692 on the first value operand of the ``getelementptr``.
1693 - The result value of a ``bitcast`` is *based* on the operand of the
1695 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1696 values that contribute (directly or indirectly) to the computation of
1697 the pointer's value.
1698 - The "*based* on" relationship is transitive.
1700 Note that this definition of *"based"* is intentionally similar to the
1701 definition of *"based"* in C99, though it is slightly weaker.
1703 LLVM IR does not associate types with memory. The result type of a
1704 ``load`` merely indicates the size and alignment of the memory from
1705 which to load, as well as the interpretation of the value. The first
1706 operand type of a ``store`` similarly only indicates the size and
1707 alignment of the store.
1709 Consequently, type-based alias analysis, aka TBAA, aka
1710 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1711 :ref:`Metadata <metadata>` may be used to encode additional information
1712 which specialized optimization passes may use to implement type-based
1717 Volatile Memory Accesses
1718 ------------------------
1720 Certain memory accesses, such as :ref:`load <i_load>`'s,
1721 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1722 marked ``volatile``. The optimizers must not change the number of
1723 volatile operations or change their order of execution relative to other
1724 volatile operations. The optimizers *may* change the order of volatile
1725 operations relative to non-volatile operations. This is not Java's
1726 "volatile" and has no cross-thread synchronization behavior.
1728 IR-level volatile loads and stores cannot safely be optimized into
1729 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1730 flagged volatile. Likewise, the backend should never split or merge
1731 target-legal volatile load/store instructions.
1733 .. admonition:: Rationale
1735 Platforms may rely on volatile loads and stores of natively supported
1736 data width to be executed as single instruction. For example, in C
1737 this holds for an l-value of volatile primitive type with native
1738 hardware support, but not necessarily for aggregate types. The
1739 frontend upholds these expectations, which are intentionally
1740 unspecified in the IR. The rules above ensure that IR transformations
1741 do not violate the frontend's contract with the language.
1745 Memory Model for Concurrent Operations
1746 --------------------------------------
1748 The LLVM IR does not define any way to start parallel threads of
1749 execution or to register signal handlers. Nonetheless, there are
1750 platform-specific ways to create them, and we define LLVM IR's behavior
1751 in their presence. This model is inspired by the C++0x memory model.
1753 For a more informal introduction to this model, see the :doc:`Atomics`.
1755 We define a *happens-before* partial order as the least partial order
1758 - Is a superset of single-thread program order, and
1759 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1760 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1761 techniques, like pthread locks, thread creation, thread joining,
1762 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1763 Constraints <ordering>`).
1765 Note that program order does not introduce *happens-before* edges
1766 between a thread and signals executing inside that thread.
1768 Every (defined) read operation (load instructions, memcpy, atomic
1769 loads/read-modify-writes, etc.) R reads a series of bytes written by
1770 (defined) write operations (store instructions, atomic
1771 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1772 section, initialized globals are considered to have a write of the
1773 initializer which is atomic and happens before any other read or write
1774 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1775 may see any write to the same byte, except:
1777 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1778 write\ :sub:`2` happens before R\ :sub:`byte`, then
1779 R\ :sub:`byte` does not see write\ :sub:`1`.
1780 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1781 R\ :sub:`byte` does not see write\ :sub:`3`.
1783 Given that definition, R\ :sub:`byte` is defined as follows:
1785 - If R is volatile, the result is target-dependent. (Volatile is
1786 supposed to give guarantees which can support ``sig_atomic_t`` in
1787 C/C++, and may be used for accesses to addresses that do not behave
1788 like normal memory. It does not generally provide cross-thread
1790 - Otherwise, if there is no write to the same byte that happens before
1791 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1792 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1793 R\ :sub:`byte` returns the value written by that write.
1794 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1795 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1796 Memory Ordering Constraints <ordering>` section for additional
1797 constraints on how the choice is made.
1798 - Otherwise R\ :sub:`byte` returns ``undef``.
1800 R returns the value composed of the series of bytes it read. This
1801 implies that some bytes within the value may be ``undef`` **without**
1802 the entire value being ``undef``. Note that this only defines the
1803 semantics of the operation; it doesn't mean that targets will emit more
1804 than one instruction to read the series of bytes.
1806 Note that in cases where none of the atomic intrinsics are used, this
1807 model places only one restriction on IR transformations on top of what
1808 is required for single-threaded execution: introducing a store to a byte
1809 which might not otherwise be stored is not allowed in general.
1810 (Specifically, in the case where another thread might write to and read
1811 from an address, introducing a store can change a load that may see
1812 exactly one write into a load that may see multiple writes.)
1816 Atomic Memory Ordering Constraints
1817 ----------------------------------
1819 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1820 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1821 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1822 ordering parameters that determine which other atomic instructions on
1823 the same address they *synchronize with*. These semantics are borrowed
1824 from Java and C++0x, but are somewhat more colloquial. If these
1825 descriptions aren't precise enough, check those specs (see spec
1826 references in the :doc:`atomics guide <Atomics>`).
1827 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1828 differently since they don't take an address. See that instruction's
1829 documentation for details.
1831 For a simpler introduction to the ordering constraints, see the
1835 The set of values that can be read is governed by the happens-before
1836 partial order. A value cannot be read unless some operation wrote
1837 it. This is intended to provide a guarantee strong enough to model
1838 Java's non-volatile shared variables. This ordering cannot be
1839 specified for read-modify-write operations; it is not strong enough
1840 to make them atomic in any interesting way.
1842 In addition to the guarantees of ``unordered``, there is a single
1843 total order for modifications by ``monotonic`` operations on each
1844 address. All modification orders must be compatible with the
1845 happens-before order. There is no guarantee that the modification
1846 orders can be combined to a global total order for the whole program
1847 (and this often will not be possible). The read in an atomic
1848 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1849 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1850 order immediately before the value it writes. If one atomic read
1851 happens before another atomic read of the same address, the later
1852 read must see the same value or a later value in the address's
1853 modification order. This disallows reordering of ``monotonic`` (or
1854 stronger) operations on the same address. If an address is written
1855 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1856 read that address repeatedly, the other threads must eventually see
1857 the write. This corresponds to the C++0x/C1x
1858 ``memory_order_relaxed``.
1860 In addition to the guarantees of ``monotonic``, a
1861 *synchronizes-with* edge may be formed with a ``release`` operation.
1862 This is intended to model C++'s ``memory_order_acquire``.
1864 In addition to the guarantees of ``monotonic``, if this operation
1865 writes a value which is subsequently read by an ``acquire``
1866 operation, it *synchronizes-with* that operation. (This isn't a
1867 complete description; see the C++0x definition of a release
1868 sequence.) This corresponds to the C++0x/C1x
1869 ``memory_order_release``.
1870 ``acq_rel`` (acquire+release)
1871 Acts as both an ``acquire`` and ``release`` operation on its
1872 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1873 ``seq_cst`` (sequentially consistent)
1874 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1875 operation that only reads, ``release`` for an operation that only
1876 writes), there is a global total order on all
1877 sequentially-consistent operations on all addresses, which is
1878 consistent with the *happens-before* partial order and with the
1879 modification orders of all the affected addresses. Each
1880 sequentially-consistent read sees the last preceding write to the
1881 same address in this global order. This corresponds to the C++0x/C1x
1882 ``memory_order_seq_cst`` and Java volatile.
1886 If an atomic operation is marked ``singlethread``, it only *synchronizes
1887 with* or participates in modification and seq\_cst total orderings with
1888 other operations running in the same thread (for example, in signal
1896 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1897 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1898 :ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) have the following flags that can
1899 be set to enable otherwise unsafe floating point operations
1902 No NaNs - Allow optimizations to assume the arguments and result are not
1903 NaN. Such optimizations are required to retain defined behavior over
1904 NaNs, but the value of the result is undefined.
1907 No Infs - Allow optimizations to assume the arguments and result are not
1908 +/-Inf. Such optimizations are required to retain defined behavior over
1909 +/-Inf, but the value of the result is undefined.
1912 No Signed Zeros - Allow optimizations to treat the sign of a zero
1913 argument or result as insignificant.
1916 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1917 argument rather than perform division.
1920 Fast - Allow algebraically equivalent transformations that may
1921 dramatically change results in floating point (e.g. reassociate). This
1922 flag implies all the others.
1926 Use-list Order Directives
1927 -------------------------
1929 Use-list directives encode the in-memory order of each use-list, allowing the
1930 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1931 indexes that are assigned to the referenced value's uses. The referenced
1932 value's use-list is immediately sorted by these indexes.
1934 Use-list directives may appear at function scope or global scope. They are not
1935 instructions, and have no effect on the semantics of the IR. When they're at
1936 function scope, they must appear after the terminator of the final basic block.
1938 If basic blocks have their address taken via ``blockaddress()`` expressions,
1939 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1946 uselistorder <ty> <value>, { <order-indexes> }
1947 uselistorder_bb @function, %block { <order-indexes> }
1953 define void @foo(i32 %arg1, i32 %arg2) {
1955 ; ... instructions ...
1957 ; ... instructions ...
1959 ; At function scope.
1960 uselistorder i32 %arg1, { 1, 0, 2 }
1961 uselistorder label %bb, { 1, 0 }
1965 uselistorder i32* @global, { 1, 2, 0 }
1966 uselistorder i32 7, { 1, 0 }
1967 uselistorder i32 (i32) @bar, { 1, 0 }
1968 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1975 The LLVM type system is one of the most important features of the
1976 intermediate representation. Being typed enables a number of
1977 optimizations to be performed on the intermediate representation
1978 directly, without having to do extra analyses on the side before the
1979 transformation. A strong type system makes it easier to read the
1980 generated code and enables novel analyses and transformations that are
1981 not feasible to perform on normal three address code representations.
1991 The void type does not represent any value and has no size.
2009 The function type can be thought of as a function signature. It consists of a
2010 return type and a list of formal parameter types. The return type of a function
2011 type is a void type or first class type --- except for :ref:`label <t_label>`
2012 and :ref:`metadata <t_metadata>` types.
2018 <returntype> (<parameter list>)
2020 ...where '``<parameter list>``' is a comma-separated list of type
2021 specifiers. Optionally, the parameter list may include a type ``...``, which
2022 indicates that the function takes a variable number of arguments. Variable
2023 argument functions can access their arguments with the :ref:`variable argument
2024 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
2025 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
2029 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2030 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
2031 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2032 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
2033 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2034 | ``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. |
2035 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2036 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
2037 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2044 The :ref:`first class <t_firstclass>` types are perhaps the most important.
2045 Values of these types are the only ones which can be produced by
2053 These are the types that are valid in registers from CodeGen's perspective.
2062 The integer type is a very simple type that simply specifies an
2063 arbitrary bit width for the integer type desired. Any bit width from 1
2064 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2072 The number of bits the integer will occupy is specified by the ``N``
2078 +----------------+------------------------------------------------+
2079 | ``i1`` | a single-bit integer. |
2080 +----------------+------------------------------------------------+
2081 | ``i32`` | a 32-bit integer. |
2082 +----------------+------------------------------------------------+
2083 | ``i1942652`` | a really big integer of over 1 million bits. |
2084 +----------------+------------------------------------------------+
2088 Floating Point Types
2089 """"""""""""""""""""
2098 - 16-bit floating point value
2101 - 32-bit floating point value
2104 - 64-bit floating point value
2107 - 128-bit floating point value (112-bit mantissa)
2110 - 80-bit floating point value (X87)
2113 - 128-bit floating point value (two 64-bits)
2120 The x86_mmx type represents a value held in an MMX register on an x86
2121 machine. The operations allowed on it are quite limited: parameters and
2122 return values, load and store, and bitcast. User-specified MMX
2123 instructions are represented as intrinsic or asm calls with arguments
2124 and/or results of this type. There are no arrays, vectors or constants
2141 The pointer type is used to specify memory locations. Pointers are
2142 commonly used to reference objects in memory.
2144 Pointer types may have an optional address space attribute defining the
2145 numbered address space where the pointed-to object resides. The default
2146 address space is number zero. The semantics of non-zero address spaces
2147 are target-specific.
2149 Note that LLVM does not permit pointers to void (``void*``) nor does it
2150 permit pointers to labels (``label*``). Use ``i8*`` instead.
2160 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2161 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2162 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2163 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2164 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2165 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2166 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2175 A vector type is a simple derived type that represents a vector of
2176 elements. Vector types are used when multiple primitive data are
2177 operated in parallel using a single instruction (SIMD). A vector type
2178 requires a size (number of elements) and an underlying primitive data
2179 type. Vector types are considered :ref:`first class <t_firstclass>`.
2185 < <# elements> x <elementtype> >
2187 The number of elements is a constant integer value larger than 0;
2188 elementtype may be any integer, floating point or pointer type. Vectors
2189 of size zero are not allowed.
2193 +-------------------+--------------------------------------------------+
2194 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2195 +-------------------+--------------------------------------------------+
2196 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2197 +-------------------+--------------------------------------------------+
2198 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2199 +-------------------+--------------------------------------------------+
2200 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2201 +-------------------+--------------------------------------------------+
2210 The label type represents code labels.
2225 The token type is used when a value is associated with an instruction
2226 but all uses of the value must not attempt to introspect or obscure it.
2227 As such, it is not appropriate to have a :ref:`phi <i_phi>` or
2228 :ref:`select <i_select>` of type token.
2245 The metadata type represents embedded metadata. No derived types may be
2246 created from metadata except for :ref:`function <t_function>` arguments.
2259 Aggregate Types are a subset of derived types that can contain multiple
2260 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2261 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2271 The array type is a very simple derived type that arranges elements
2272 sequentially in memory. The array type requires a size (number of
2273 elements) and an underlying data type.
2279 [<# elements> x <elementtype>]
2281 The number of elements is a constant integer value; ``elementtype`` may
2282 be any type with a size.
2286 +------------------+--------------------------------------+
2287 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2288 +------------------+--------------------------------------+
2289 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2290 +------------------+--------------------------------------+
2291 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2292 +------------------+--------------------------------------+
2294 Here are some examples of multidimensional arrays:
2296 +-----------------------------+----------------------------------------------------------+
2297 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2298 +-----------------------------+----------------------------------------------------------+
2299 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2300 +-----------------------------+----------------------------------------------------------+
2301 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2302 +-----------------------------+----------------------------------------------------------+
2304 There is no restriction on indexing beyond the end of the array implied
2305 by a static type (though there are restrictions on indexing beyond the
2306 bounds of an allocated object in some cases). This means that
2307 single-dimension 'variable sized array' addressing can be implemented in
2308 LLVM with a zero length array type. An implementation of 'pascal style
2309 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2319 The structure type is used to represent a collection of data members
2320 together in memory. The elements of a structure may be any type that has
2323 Structures in memory are accessed using '``load``' and '``store``' by
2324 getting a pointer to a field with the '``getelementptr``' instruction.
2325 Structures in registers are accessed using the '``extractvalue``' and
2326 '``insertvalue``' instructions.
2328 Structures may optionally be "packed" structures, which indicate that
2329 the alignment of the struct is one byte, and that there is no padding
2330 between the elements. In non-packed structs, padding between field types
2331 is inserted as defined by the DataLayout string in the module, which is
2332 required to match what the underlying code generator expects.
2334 Structures can either be "literal" or "identified". A literal structure
2335 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2336 identified types are always defined at the top level with a name.
2337 Literal types are uniqued by their contents and can never be recursive
2338 or opaque since there is no way to write one. Identified types can be
2339 recursive, can be opaqued, and are never uniqued.
2345 %T1 = type { <type list> } ; Identified normal struct type
2346 %T2 = type <{ <type list> }> ; Identified packed struct type
2350 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2351 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2352 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2353 | ``{ 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``. |
2354 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2355 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2356 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2360 Opaque Structure Types
2361 """"""""""""""""""""""
2365 Opaque structure types are used to represent named structure types that
2366 do not have a body specified. This corresponds (for example) to the C
2367 notion of a forward declared structure.
2378 +--------------+-------------------+
2379 | ``opaque`` | An opaque type. |
2380 +--------------+-------------------+
2387 LLVM has several different basic types of constants. This section
2388 describes them all and their syntax.
2393 **Boolean constants**
2394 The two strings '``true``' and '``false``' are both valid constants
2396 **Integer constants**
2397 Standard integers (such as '4') are constants of the
2398 :ref:`integer <t_integer>` type. Negative numbers may be used with
2400 **Floating point constants**
2401 Floating point constants use standard decimal notation (e.g.
2402 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2403 hexadecimal notation (see below). The assembler requires the exact
2404 decimal value of a floating-point constant. For example, the
2405 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2406 decimal in binary. Floating point constants must have a :ref:`floating
2407 point <t_floating>` type.
2408 **Null pointer constants**
2409 The identifier '``null``' is recognized as a null pointer constant
2410 and must be of :ref:`pointer type <t_pointer>`.
2412 The one non-intuitive notation for constants is the hexadecimal form of
2413 floating point constants. For example, the form
2414 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2415 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2416 constants are required (and the only time that they are generated by the
2417 disassembler) is when a floating point constant must be emitted but it
2418 cannot be represented as a decimal floating point number in a reasonable
2419 number of digits. For example, NaN's, infinities, and other special
2420 values are represented in their IEEE hexadecimal format so that assembly
2421 and disassembly do not cause any bits to change in the constants.
2423 When using the hexadecimal form, constants of types half, float, and
2424 double are represented using the 16-digit form shown above (which
2425 matches the IEEE754 representation for double); half and float values
2426 must, however, be exactly representable as IEEE 754 half and single
2427 precision, respectively. Hexadecimal format is always used for long
2428 double, and there are three forms of long double. The 80-bit format used
2429 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2430 128-bit format used by PowerPC (two adjacent doubles) is represented by
2431 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2432 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2433 will only work if they match the long double format on your target.
2434 The IEEE 16-bit format (half precision) is represented by ``0xH``
2435 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2436 (sign bit at the left).
2438 There are no constants of type x86_mmx.
2440 .. _complexconstants:
2445 Complex constants are a (potentially recursive) combination of simple
2446 constants and smaller complex constants.
2448 **Structure constants**
2449 Structure constants are represented with notation similar to
2450 structure type definitions (a comma separated list of elements,
2451 surrounded by braces (``{}``)). For example:
2452 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2453 "``@G = external global i32``". Structure constants must have
2454 :ref:`structure type <t_struct>`, and the number and types of elements
2455 must match those specified by the type.
2457 Array constants are represented with notation similar to array type
2458 definitions (a comma separated list of elements, surrounded by
2459 square brackets (``[]``)). For example:
2460 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2461 :ref:`array type <t_array>`, and the number and types of elements must
2462 match those specified by the type. As a special case, character array
2463 constants may also be represented as a double-quoted string using the ``c``
2464 prefix. For example: "``c"Hello World\0A\00"``".
2465 **Vector constants**
2466 Vector constants are represented with notation similar to vector
2467 type definitions (a comma separated list of elements, surrounded by
2468 less-than/greater-than's (``<>``)). For example:
2469 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2470 must have :ref:`vector type <t_vector>`, and the number and types of
2471 elements must match those specified by the type.
2472 **Zero initialization**
2473 The string '``zeroinitializer``' can be used to zero initialize a
2474 value to zero of *any* type, including scalar and
2475 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2476 having to print large zero initializers (e.g. for large arrays) and
2477 is always exactly equivalent to using explicit zero initializers.
2479 A metadata node is a constant tuple without types. For example:
2480 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2481 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2482 Unlike other typed constants that are meant to be interpreted as part of
2483 the instruction stream, metadata is a place to attach additional
2484 information such as debug info.
2486 Global Variable and Function Addresses
2487 --------------------------------------
2489 The addresses of :ref:`global variables <globalvars>` and
2490 :ref:`functions <functionstructure>` are always implicitly valid
2491 (link-time) constants. These constants are explicitly referenced when
2492 the :ref:`identifier for the global <identifiers>` is used and always have
2493 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2496 .. code-block:: llvm
2500 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2507 The string '``undef``' can be used anywhere a constant is expected, and
2508 indicates that the user of the value may receive an unspecified
2509 bit-pattern. Undefined values may be of any type (other than '``label``'
2510 or '``void``') and be used anywhere a constant is permitted.
2512 Undefined values are useful because they indicate to the compiler that
2513 the program is well defined no matter what value is used. This gives the
2514 compiler more freedom to optimize. Here are some examples of
2515 (potentially surprising) transformations that are valid (in pseudo IR):
2517 .. code-block:: llvm
2527 This is safe because all of the output bits are affected by the undef
2528 bits. Any output bit can have a zero or one depending on the input bits.
2530 .. code-block:: llvm
2541 These logical operations have bits that are not always affected by the
2542 input. For example, if ``%X`` has a zero bit, then the output of the
2543 '``and``' operation will always be a zero for that bit, no matter what
2544 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2545 optimize or assume that the result of the '``and``' is '``undef``'.
2546 However, it is safe to assume that all bits of the '``undef``' could be
2547 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2548 all the bits of the '``undef``' operand to the '``or``' could be set,
2549 allowing the '``or``' to be folded to -1.
2551 .. code-block:: llvm
2553 %A = select undef, %X, %Y
2554 %B = select undef, 42, %Y
2555 %C = select %X, %Y, undef
2565 This set of examples shows that undefined '``select``' (and conditional
2566 branch) conditions can go *either way*, but they have to come from one
2567 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2568 both known to have a clear low bit, then ``%A`` would have to have a
2569 cleared low bit. However, in the ``%C`` example, the optimizer is
2570 allowed to assume that the '``undef``' operand could be the same as
2571 ``%Y``, allowing the whole '``select``' to be eliminated.
2573 .. code-block:: llvm
2575 %A = xor undef, undef
2592 This example points out that two '``undef``' operands are not
2593 necessarily the same. This can be surprising to people (and also matches
2594 C semantics) where they assume that "``X^X``" is always zero, even if
2595 ``X`` is undefined. This isn't true for a number of reasons, but the
2596 short answer is that an '``undef``' "variable" can arbitrarily change
2597 its value over its "live range". This is true because the variable
2598 doesn't actually *have a live range*. Instead, the value is logically
2599 read from arbitrary registers that happen to be around when needed, so
2600 the value is not necessarily consistent over time. In fact, ``%A`` and
2601 ``%C`` need to have the same semantics or the core LLVM "replace all
2602 uses with" concept would not hold.
2604 .. code-block:: llvm
2612 These examples show the crucial difference between an *undefined value*
2613 and *undefined behavior*. An undefined value (like '``undef``') is
2614 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2615 operation can be constant folded to '``undef``', because the '``undef``'
2616 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2617 However, in the second example, we can make a more aggressive
2618 assumption: because the ``undef`` is allowed to be an arbitrary value,
2619 we are allowed to assume that it could be zero. Since a divide by zero
2620 has *undefined behavior*, we are allowed to assume that the operation
2621 does not execute at all. This allows us to delete the divide and all
2622 code after it. Because the undefined operation "can't happen", the
2623 optimizer can assume that it occurs in dead code.
2625 .. code-block:: llvm
2627 a: store undef -> %X
2628 b: store %X -> undef
2633 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2634 value can be assumed to not have any effect; we can assume that the
2635 value is overwritten with bits that happen to match what was already
2636 there. However, a store *to* an undefined location could clobber
2637 arbitrary memory, therefore, it has undefined behavior.
2644 Poison values are similar to :ref:`undef values <undefvalues>`, however
2645 they also represent the fact that an instruction or constant expression
2646 that cannot evoke side effects has nevertheless detected a condition
2647 that results in undefined behavior.
2649 There is currently no way of representing a poison value in the IR; they
2650 only exist when produced by operations such as :ref:`add <i_add>` with
2653 Poison value behavior is defined in terms of value *dependence*:
2655 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2656 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2657 their dynamic predecessor basic block.
2658 - Function arguments depend on the corresponding actual argument values
2659 in the dynamic callers of their functions.
2660 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2661 instructions that dynamically transfer control back to them.
2662 - :ref:`Invoke <i_invoke>` instructions depend on the
2663 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2664 call instructions that dynamically transfer control back to them.
2665 - Non-volatile loads and stores depend on the most recent stores to all
2666 of the referenced memory addresses, following the order in the IR
2667 (including loads and stores implied by intrinsics such as
2668 :ref:`@llvm.memcpy <int_memcpy>`.)
2669 - An instruction with externally visible side effects depends on the
2670 most recent preceding instruction with externally visible side
2671 effects, following the order in the IR. (This includes :ref:`volatile
2672 operations <volatile>`.)
2673 - An instruction *control-depends* on a :ref:`terminator
2674 instruction <terminators>` if the terminator instruction has
2675 multiple successors and the instruction is always executed when
2676 control transfers to one of the successors, and may not be executed
2677 when control is transferred to another.
2678 - Additionally, an instruction also *control-depends* on a terminator
2679 instruction if the set of instructions it otherwise depends on would
2680 be different if the terminator had transferred control to a different
2682 - Dependence is transitive.
2684 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2685 with the additional effect that any instruction that has a *dependence*
2686 on a poison value has undefined behavior.
2688 Here are some examples:
2690 .. code-block:: llvm
2693 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2694 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2695 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2696 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2698 store i32 %poison, i32* @g ; Poison value stored to memory.
2699 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2701 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2703 %narrowaddr = bitcast i32* @g to i16*
2704 %wideaddr = bitcast i32* @g to i64*
2705 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2706 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2708 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2709 br i1 %cmp, label %true, label %end ; Branch to either destination.
2712 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2713 ; it has undefined behavior.
2717 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2718 ; Both edges into this PHI are
2719 ; control-dependent on %cmp, so this
2720 ; always results in a poison value.
2722 store volatile i32 0, i32* @g ; This would depend on the store in %true
2723 ; if %cmp is true, or the store in %entry
2724 ; otherwise, so this is undefined behavior.
2726 br i1 %cmp, label %second_true, label %second_end
2727 ; The same branch again, but this time the
2728 ; true block doesn't have side effects.
2735 store volatile i32 0, i32* @g ; This time, the instruction always depends
2736 ; on the store in %end. Also, it is
2737 ; control-equivalent to %end, so this is
2738 ; well-defined (ignoring earlier undefined
2739 ; behavior in this example).
2743 Addresses of Basic Blocks
2744 -------------------------
2746 ``blockaddress(@function, %block)``
2748 The '``blockaddress``' constant computes the address of the specified
2749 basic block in the specified function, and always has an ``i8*`` type.
2750 Taking the address of the entry block is illegal.
2752 This value only has defined behavior when used as an operand to the
2753 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2754 against null. Pointer equality tests between labels addresses results in
2755 undefined behavior --- though, again, comparison against null is ok, and
2756 no label is equal to the null pointer. This may be passed around as an
2757 opaque pointer sized value as long as the bits are not inspected. This
2758 allows ``ptrtoint`` and arithmetic to be performed on these values so
2759 long as the original value is reconstituted before the ``indirectbr``
2762 Finally, some targets may provide defined semantics when using the value
2763 as the operand to an inline assembly, but that is target specific.
2767 Constant Expressions
2768 --------------------
2770 Constant expressions are used to allow expressions involving other
2771 constants to be used as constants. Constant expressions may be of any
2772 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2773 that does not have side effects (e.g. load and call are not supported).
2774 The following is the syntax for constant expressions:
2776 ``trunc (CST to TYPE)``
2777 Truncate a constant to another type. The bit size of CST must be
2778 larger than the bit size of TYPE. Both types must be integers.
2779 ``zext (CST to TYPE)``
2780 Zero 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 ``sext (CST to TYPE)``
2783 Sign extend a constant to another type. The bit size of CST must be
2784 smaller than the bit size of TYPE. Both types must be integers.
2785 ``fptrunc (CST to TYPE)``
2786 Truncate a floating point constant to another floating point type.
2787 The size of CST must be larger than the size of TYPE. Both types
2788 must be floating point.
2789 ``fpext (CST to TYPE)``
2790 Floating point extend a constant to another type. The size of CST
2791 must be smaller or equal to the size of TYPE. Both types must be
2793 ``fptoui (CST to TYPE)``
2794 Convert a floating point constant to the corresponding unsigned
2795 integer constant. TYPE must be a scalar or vector integer type. CST
2796 must be of scalar or vector floating point type. Both CST and TYPE
2797 must be scalars, or vectors of the same number of elements. If the
2798 value won't fit in the integer type, the results are undefined.
2799 ``fptosi (CST to TYPE)``
2800 Convert a floating point constant to the corresponding signed
2801 integer constant. TYPE must be a scalar or vector integer type. CST
2802 must be of scalar or vector floating point type. Both CST and TYPE
2803 must be scalars, or vectors of the same number of elements. If the
2804 value won't fit in the integer type, the results are undefined.
2805 ``uitofp (CST to TYPE)``
2806 Convert an unsigned integer constant to the corresponding floating
2807 point constant. TYPE must be a scalar or vector floating point type.
2808 CST must be of scalar or vector integer type. Both CST and TYPE must
2809 be scalars, or vectors of the same number of elements. If the value
2810 won't fit in the floating point type, the results are undefined.
2811 ``sitofp (CST to TYPE)``
2812 Convert a signed integer constant to the corresponding floating
2813 point constant. TYPE must be a scalar or vector floating point type.
2814 CST must be of scalar or vector integer type. Both CST and TYPE must
2815 be scalars, or vectors of the same number of elements. If the value
2816 won't fit in the floating point type, the results are undefined.
2817 ``ptrtoint (CST to TYPE)``
2818 Convert a pointer typed constant to the corresponding integer
2819 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2820 pointer type. The ``CST`` value is zero extended, truncated, or
2821 unchanged to make it fit in ``TYPE``.
2822 ``inttoptr (CST to TYPE)``
2823 Convert an integer constant to a pointer constant. TYPE must be a
2824 pointer type. CST must be of integer type. The CST value is zero
2825 extended, truncated, or unchanged to make it fit in a pointer size.
2826 This one is *really* dangerous!
2827 ``bitcast (CST to TYPE)``
2828 Convert a constant, CST, to another TYPE. The constraints of the
2829 operands are the same as those for the :ref:`bitcast
2830 instruction <i_bitcast>`.
2831 ``addrspacecast (CST to TYPE)``
2832 Convert a constant pointer or constant vector of pointer, CST, to another
2833 TYPE in a different address space. The constraints of the operands are the
2834 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2835 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2836 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2837 constants. As with the :ref:`getelementptr <i_getelementptr>`
2838 instruction, the index list may have zero or more indexes, which are
2839 required to make sense for the type of "pointer to TY".
2840 ``select (COND, VAL1, VAL2)``
2841 Perform the :ref:`select operation <i_select>` on constants.
2842 ``icmp COND (VAL1, VAL2)``
2843 Performs the :ref:`icmp operation <i_icmp>` on constants.
2844 ``fcmp COND (VAL1, VAL2)``
2845 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2846 ``extractelement (VAL, IDX)``
2847 Perform the :ref:`extractelement operation <i_extractelement>` on
2849 ``insertelement (VAL, ELT, IDX)``
2850 Perform the :ref:`insertelement operation <i_insertelement>` on
2852 ``shufflevector (VEC1, VEC2, IDXMASK)``
2853 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2855 ``extractvalue (VAL, IDX0, IDX1, ...)``
2856 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2857 constants. The index list is interpreted in a similar manner as
2858 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2859 least one index value must be specified.
2860 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2861 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2862 The index list is interpreted in a similar manner as indices in a
2863 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2864 value must be specified.
2865 ``OPCODE (LHS, RHS)``
2866 Perform the specified operation of the LHS and RHS constants. OPCODE
2867 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2868 binary <bitwiseops>` operations. The constraints on operands are
2869 the same as those for the corresponding instruction (e.g. no bitwise
2870 operations on floating point values are allowed).
2877 Inline Assembler Expressions
2878 ----------------------------
2880 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2881 Inline Assembly <moduleasm>`) through the use of a special value. This value
2882 represents the inline assembler as a template string (containing the
2883 instructions to emit), a list of operand constraints (stored as a string), a
2884 flag that indicates whether or not the inline asm expression has side effects,
2885 and a flag indicating whether the function containing the asm needs to align its
2886 stack conservatively.
2888 The template string supports argument substitution of the operands using "``$``"
2889 followed by a number, to indicate substitution of the given register/memory
2890 location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
2891 be used, where ``MODIFIER`` is a target-specific annotation for how to print the
2892 operand (See :ref:`inline-asm-modifiers`).
2894 A literal "``$``" may be included by using "``$$``" in the template. To include
2895 other special characters into the output, the usual "``\XX``" escapes may be
2896 used, just as in other strings. Note that after template substitution, the
2897 resulting assembly string is parsed by LLVM's integrated assembler unless it is
2898 disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
2899 syntax known to LLVM.
2901 LLVM's support for inline asm is modeled closely on the requirements of Clang's
2902 GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
2903 modifier codes listed here are similar or identical to those in GCC's inline asm
2904 support. However, to be clear, the syntax of the template and constraint strings
2905 described here is *not* the same as the syntax accepted by GCC and Clang, and,
2906 while most constraint letters are passed through as-is by Clang, some get
2907 translated to other codes when converting from the C source to the LLVM
2910 An example inline assembler expression is:
2912 .. code-block:: llvm
2914 i32 (i32) asm "bswap $0", "=r,r"
2916 Inline assembler expressions may **only** be used as the callee operand
2917 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2918 Thus, typically we have:
2920 .. code-block:: llvm
2922 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2924 Inline asms with side effects not visible in the constraint list must be
2925 marked as having side effects. This is done through the use of the
2926 '``sideeffect``' keyword, like so:
2928 .. code-block:: llvm
2930 call void asm sideeffect "eieio", ""()
2932 In some cases inline asms will contain code that will not work unless
2933 the stack is aligned in some way, such as calls or SSE instructions on
2934 x86, yet will not contain code that does that alignment within the asm.
2935 The compiler should make conservative assumptions about what the asm
2936 might contain and should generate its usual stack alignment code in the
2937 prologue if the '``alignstack``' keyword is present:
2939 .. code-block:: llvm
2941 call void asm alignstack "eieio", ""()
2943 Inline asms also support using non-standard assembly dialects. The
2944 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2945 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2946 the only supported dialects. An example is:
2948 .. code-block:: llvm
2950 call void asm inteldialect "eieio", ""()
2952 If multiple keywords appear the '``sideeffect``' keyword must come
2953 first, the '``alignstack``' keyword second and the '``inteldialect``'
2956 Inline Asm Constraint String
2957 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2959 The constraint list is a comma-separated string, each element containing one or
2960 more constraint codes.
2962 For each element in the constraint list an appropriate register or memory
2963 operand will be chosen, and it will be made available to assembly template
2964 string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
2967 There are three different types of constraints, which are distinguished by a
2968 prefix symbol in front of the constraint code: Output, Input, and Clobber. The
2969 constraints must always be given in that order: outputs first, then inputs, then
2970 clobbers. They cannot be intermingled.
2972 There are also three different categories of constraint codes:
2974 - Register constraint. This is either a register class, or a fixed physical
2975 register. This kind of constraint will allocate a register, and if necessary,
2976 bitcast the argument or result to the appropriate type.
2977 - Memory constraint. This kind of constraint is for use with an instruction
2978 taking a memory operand. Different constraints allow for different addressing
2979 modes used by the target.
2980 - Immediate value constraint. This kind of constraint is for an integer or other
2981 immediate value which can be rendered directly into an instruction. The
2982 various target-specific constraints allow the selection of a value in the
2983 proper range for the instruction you wish to use it with.
2988 Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
2989 indicates that the assembly will write to this operand, and the operand will
2990 then be made available as a return value of the ``asm`` expression. Output
2991 constraints do not consume an argument from the call instruction. (Except, see
2992 below about indirect outputs).
2994 Normally, it is expected that no output locations are written to by the assembly
2995 expression until *all* of the inputs have been read. As such, LLVM may assign
2996 the same register to an output and an input. If this is not safe (e.g. if the
2997 assembly contains two instructions, where the first writes to one output, and
2998 the second reads an input and writes to a second output), then the "``&``"
2999 modifier must be used (e.g. "``=&r``") to specify that the output is an
3000 "early-clobber" output. Marking an ouput as "early-clobber" ensures that LLVM
3001 will not use the same register for any inputs (other than an input tied to this
3007 Input constraints do not have a prefix -- just the constraint codes. Each input
3008 constraint will consume one argument from the call instruction. It is not
3009 permitted for the asm to write to any input register or memory location (unless
3010 that input is tied to an output). Note also that multiple inputs may all be
3011 assigned to the same register, if LLVM can determine that they necessarily all
3012 contain the same value.
3014 Instead of providing a Constraint Code, input constraints may also "tie"
3015 themselves to an output constraint, by providing an integer as the constraint
3016 string. Tied inputs still consume an argument from the call instruction, and
3017 take up a position in the asm template numbering as is usual -- they will simply
3018 be constrained to always use the same register as the output they've been tied
3019 to. For example, a constraint string of "``=r,0``" says to assign a register for
3020 output, and use that register as an input as well (it being the 0'th
3023 It is permitted to tie an input to an "early-clobber" output. In that case, no
3024 *other* input may share the same register as the input tied to the early-clobber
3025 (even when the other input has the same value).
3027 You may only tie an input to an output which has a register constraint, not a
3028 memory constraint. Only a single input may be tied to an output.
3030 There is also an "interesting" feature which deserves a bit of explanation: if a
3031 register class constraint allocates a register which is too small for the value
3032 type operand provided as input, the input value will be split into multiple
3033 registers, and all of them passed to the inline asm.
3035 However, this feature is often not as useful as you might think.
3037 Firstly, the registers are *not* guaranteed to be consecutive. So, on those
3038 architectures that have instructions which operate on multiple consecutive
3039 instructions, this is not an appropriate way to support them. (e.g. the 32-bit
3040 SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
3041 hardware then loads into both the named register, and the next register. This
3042 feature of inline asm would not be useful to support that.)
3044 A few of the targets provide a template string modifier allowing explicit access
3045 to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
3046 ``D``). On such an architecture, you can actually access the second allocated
3047 register (yet, still, not any subsequent ones). But, in that case, you're still
3048 probably better off simply splitting the value into two separate operands, for
3049 clarity. (e.g. see the description of the ``A`` constraint on X86, which,
3050 despite existing only for use with this feature, is not really a good idea to
3053 Indirect inputs and outputs
3054 """""""""""""""""""""""""""
3056 Indirect output or input constraints can be specified by the "``*``" modifier
3057 (which goes after the "``=``" in case of an output). This indicates that the asm
3058 will write to or read from the contents of an *address* provided as an input
3059 argument. (Note that in this way, indirect outputs act more like an *input* than
3060 an output: just like an input, they consume an argument of the call expression,
3061 rather than producing a return value. An indirect output constraint is an
3062 "output" only in that the asm is expected to write to the contents of the input
3063 memory location, instead of just read from it).
3065 This is most typically used for memory constraint, e.g. "``=*m``", to pass the
3066 address of a variable as a value.
3068 It is also possible to use an indirect *register* constraint, but only on output
3069 (e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3070 value normally, and then, separately emit a store to the address provided as
3071 input, after the provided inline asm. (It's not clear what value this
3072 functionality provides, compared to writing the store explicitly after the asm
3073 statement, and it can only produce worse code, since it bypasses many
3074 optimization passes. I would recommend not using it.)
3080 A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3081 consume an input operand, nor generate an output. Clobbers cannot use any of the
3082 general constraint code letters -- they may use only explicit register
3083 constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3084 "``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3085 memory locations -- not only the memory pointed to by a declared indirect
3091 After a potential prefix comes constraint code, or codes.
3093 A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3094 followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3097 The one and two letter constraint codes are typically chosen to be the same as
3098 GCC's constraint codes.
3100 A single constraint may include one or more than constraint code in it, leaving
3101 it up to LLVM to choose which one to use. This is included mainly for
3102 compatibility with the translation of GCC inline asm coming from clang.
3104 There are two ways to specify alternatives, and either or both may be used in an
3105 inline asm constraint list:
3107 1) Append the codes to each other, making a constraint code set. E.g. "``im``"
3108 or "``{eax}m``". This means "choose any of the options in the set". The
3109 choice of constraint is made independently for each constraint in the
3112 2) Use "``|``" between constraint code sets, creating alternatives. Every
3113 constraint in the constraint list must have the same number of alternative
3114 sets. With this syntax, the same alternative in *all* of the items in the
3115 constraint list will be chosen together.
3117 Putting those together, you might have a two operand constraint string like
3118 ``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3119 operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3120 may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3122 However, the use of either of the alternatives features is *NOT* recommended, as
3123 LLVM is not able to make an intelligent choice about which one to use. (At the
3124 point it currently needs to choose, not enough information is available to do so
3125 in a smart way.) Thus, it simply tries to make a choice that's most likely to
3126 compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3127 always choose to use memory, not registers). And, if given multiple registers,
3128 or multiple register classes, it will simply choose the first one. (In fact, it
3129 doesn't currently even ensure explicitly specified physical registers are
3130 unique, so specifying multiple physical registers as alternatives, like
3131 ``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3134 Supported Constraint Code List
3135 """"""""""""""""""""""""""""""
3137 The constraint codes are, in general, expected to behave the same way they do in
3138 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3139 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3140 and GCC likely indicates a bug in LLVM.
3142 Some constraint codes are typically supported by all targets:
3144 - ``r``: A register in the target's general purpose register class.
3145 - ``m``: A memory address operand. It is target-specific what addressing modes
3146 are supported, typical examples are register, or register + register offset,
3147 or register + immediate offset (of some target-specific size).
3148 - ``i``: An integer constant (of target-specific width). Allows either a simple
3149 immediate, or a relocatable value.
3150 - ``n``: An integer constant -- *not* including relocatable values.
3151 - ``s``: An integer constant, but allowing *only* relocatable values.
3152 - ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3153 useful to pass a label for an asm branch or call.
3155 .. FIXME: but that surely isn't actually okay to jump out of an asm
3156 block without telling llvm about the control transfer???)
3158 - ``{register-name}``: Requires exactly the named physical register.
3160 Other constraints are target-specific:
3164 - ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3165 - ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3166 i.e. 0 to 4095 with optional shift by 12.
3167 - ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3168 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3169 - ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3170 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3171 - ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3172 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3173 - ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3174 32-bit register. This is a superset of ``K``: in addition to the bitmask
3175 immediate, also allows immediate integers which can be loaded with a single
3176 ``MOVZ`` or ``MOVL`` instruction.
3177 - ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3178 64-bit register. This is a superset of ``L``.
3179 - ``Q``: Memory address operand must be in a single register (no
3180 offsets). (However, LLVM currently does this for the ``m`` constraint as
3182 - ``r``: A 32 or 64-bit integer register (W* or X*).
3183 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3184 - ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3188 - ``r``: A 32 or 64-bit integer register.
3189 - ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3190 - ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3195 - ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3196 operand. Treated the same as operand ``m``, at the moment.
3198 ARM and ARM's Thumb2 mode:
3200 - ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3201 - ``I``: An immediate integer valid for a data-processing instruction.
3202 - ``J``: An immediate integer between -4095 and 4095.
3203 - ``K``: An immediate integer whose bitwise inverse is valid for a
3204 data-processing instruction. (Can be used with template modifier "``B``" to
3205 print the inverted value).
3206 - ``L``: An immediate integer whose negation is valid for a data-processing
3207 instruction. (Can be used with template modifier "``n``" to print the negated
3209 - ``M``: A power of two or a integer between 0 and 32.
3210 - ``N``: Invalid immediate constraint.
3211 - ``O``: Invalid immediate constraint.
3212 - ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3213 - ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3215 - ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3217 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3218 ``d0-d31``, or ``q0-q15``.
3219 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3220 ``d0-d7``, or ``q0-q3``.
3221 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3226 - ``I``: An immediate integer between 0 and 255.
3227 - ``J``: An immediate integer between -255 and -1.
3228 - ``K``: An immediate integer between 0 and 255, with optional left-shift by
3230 - ``L``: An immediate integer between -7 and 7.
3231 - ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3232 - ``N``: An immediate integer between 0 and 31.
3233 - ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3234 - ``r``: A low 32-bit GPR register (``r0-r7``).
3235 - ``l``: A low 32-bit GPR register (``r0-r7``).
3236 - ``h``: A high GPR register (``r0-r7``).
3237 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3238 ``d0-d31``, or ``q0-q15``.
3239 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3240 ``d0-d7``, or ``q0-q3``.
3241 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3247 - ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3249 - ``r``: A 32 or 64-bit register.
3253 - ``r``: An 8 or 16-bit register.
3257 - ``I``: An immediate signed 16-bit integer.
3258 - ``J``: An immediate integer zero.
3259 - ``K``: An immediate unsigned 16-bit integer.
3260 - ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3261 - ``N``: An immediate integer between -65535 and -1.
3262 - ``O``: An immediate signed 15-bit integer.
3263 - ``P``: An immediate integer between 1 and 65535.
3264 - ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3265 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3266 - ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3267 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3269 - ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3270 ``sc`` instruction on the given subtarget (details vary).
3271 - ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3272 - ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
3273 (``W0-W31``). In the case of MSA registers, it is recommended to use the ``w``
3274 argument modifier for compatibility with GCC.
3275 - ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3277 - ``l``: The ``lo`` register, 32 or 64-bit.
3282 - ``b``: A 1-bit integer register.
3283 - ``c`` or ``h``: A 16-bit integer register.
3284 - ``r``: A 32-bit integer register.
3285 - ``l`` or ``N``: A 64-bit integer register.
3286 - ``f``: A 32-bit float register.
3287 - ``d``: A 64-bit float register.
3292 - ``I``: An immediate signed 16-bit integer.
3293 - ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3294 - ``K``: An immediate unsigned 16-bit integer.
3295 - ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3296 - ``M``: An immediate integer greater than 31.
3297 - ``N``: An immediate integer that is an exact power of 2.
3298 - ``O``: The immediate integer constant 0.
3299 - ``P``: An immediate integer constant whose negation is a signed 16-bit
3301 - ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3302 treated the same as ``m``.
3303 - ``r``: A 32 or 64-bit integer register.
3304 - ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3306 - ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3307 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3308 - ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3309 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3310 altivec vector register (``V0-V31``).
3312 .. FIXME: is this a bug that v accepts QPX registers? I think this
3313 is supposed to only use the altivec vector registers?
3315 - ``y``: Condition register (``CR0-CR7``).
3316 - ``wc``: An individual CR bit in a CR register.
3317 - ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3318 register set (overlapping both the floating-point and vector register files).
3319 - ``ws``: A 32 or 64-bit floating point register, from the full VSX register
3324 - ``I``: An immediate 13-bit signed integer.
3325 - ``r``: A 32-bit integer register.
3329 - ``I``: An immediate unsigned 8-bit integer.
3330 - ``J``: An immediate unsigned 12-bit integer.
3331 - ``K``: An immediate signed 16-bit integer.
3332 - ``L``: An immediate signed 20-bit integer.
3333 - ``M``: An immediate integer 0x7fffffff.
3334 - ``Q``, ``R``, ``S``, ``T``: A memory address operand, treated the same as
3335 ``m``, at the moment.
3336 - ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3337 - ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3338 address context evaluates as zero).
3339 - ``h``: A 32-bit value in the high part of a 64bit data register
3341 - ``f``: A 32, 64, or 128-bit floating point register.
3345 - ``I``: An immediate integer between 0 and 31.
3346 - ``J``: An immediate integer between 0 and 64.
3347 - ``K``: An immediate signed 8-bit integer.
3348 - ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3350 - ``M``: An immediate integer between 0 and 3.
3351 - ``N``: An immediate unsigned 8-bit integer.
3352 - ``O``: An immediate integer between 0 and 127.
3353 - ``e``: An immediate 32-bit signed integer.
3354 - ``Z``: An immediate 32-bit unsigned integer.
3355 - ``o``, ``v``: Treated the same as ``m``, at the moment.
3356 - ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3357 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3358 registers, and on X86-64, it is all of the integer registers.
3359 - ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3360 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3361 - ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3362 - ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3363 existed since i386, and can be accessed without the REX prefix.
3364 - ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3365 - ``y``: A 64-bit MMX register, if MMX is enabled.
3366 - ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3367 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3368 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3369 512-bit vector operand in an AVX512 register, Otherwise, an error.
3370 - ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3371 - ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3372 32-bit mode, a 64-bit integer operand will get split into two registers). It
3373 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3374 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3375 you're better off splitting it yourself, before passing it to the asm
3380 - ``r``: A 32-bit integer register.
3383 .. _inline-asm-modifiers:
3385 Asm template argument modifiers
3386 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3388 In the asm template string, modifiers can be used on the operand reference, like
3391 The modifiers are, in general, expected to behave the same way they do in
3392 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3393 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3394 and GCC likely indicates a bug in LLVM.
3398 - ``c``: Print an immediate integer constant unadorned, without
3399 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3400 - ``n``: Negate and print immediate integer constant unadorned, without the
3401 target-specific immediate punctuation (e.g. no ``$`` prefix).
3402 - ``l``: Print as an unadorned label, without the target-specific label
3403 punctuation (e.g. no ``$`` prefix).
3407 - ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3408 instead of ``x30``, print ``w30``.
3409 - ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3410 - ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3411 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3420 - ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3424 - ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3425 as ``d4[1]`` instead of ``s9``)
3426 - ``B``: Bitwise invert and print an immediate integer constant without ``#``
3428 - ``L``: Print the low 16-bits of an immediate integer constant.
3429 - ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3430 register operands subsequent to the specified one (!), so use carefully.
3431 - ``Q``: Print the low-order register of a register-pair, or the low-order
3432 register of a two-register operand.
3433 - ``R``: Print the high-order register of a register-pair, or the high-order
3434 register of a two-register operand.
3435 - ``H``: Print the second register of a register-pair. (On a big-endian system,
3436 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3439 .. FIXME: H doesn't currently support printing the second register
3440 of a two-register operand.
3442 - ``e``: Print the low doubleword register of a NEON quad register.
3443 - ``f``: Print the high doubleword register of a NEON quad register.
3444 - ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3449 - ``L``: Print the second register of a two-register operand. Requires that it
3450 has been allocated consecutively to the first.
3452 .. FIXME: why is it restricted to consecutive ones? And there's
3453 nothing that ensures that happens, is there?
3455 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3456 nothing. Used to print 'addi' vs 'add' instructions.
3460 No additional modifiers.
3464 - ``X``: Print an immediate integer as hexadecimal
3465 - ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3466 - ``d``: Print an immediate integer as decimal.
3467 - ``m``: Subtract one and print an immediate integer as decimal.
3468 - ``z``: Print $0 if an immediate zero, otherwise print normally.
3469 - ``L``: Print the low-order register of a two-register operand, or prints the
3470 address of the low-order word of a double-word memory operand.
3472 .. FIXME: L seems to be missing memory operand support.
3474 - ``M``: Print the high-order register of a two-register operand, or prints the
3475 address of the high-order word of a double-word memory operand.
3477 .. FIXME: M seems to be missing memory operand support.
3479 - ``D``: Print the second register of a two-register operand, or prints the
3480 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3481 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3483 - ``w``: No effect. Provided for compatibility with GCC which requires this
3484 modifier in order to print MSA registers (``W0-W31``) with the ``f``
3493 - ``L``: Print the second register of a two-register operand. Requires that it
3494 has been allocated consecutively to the first.
3496 .. FIXME: why is it restricted to consecutive ones? And there's
3497 nothing that ensures that happens, is there?
3499 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3500 nothing. Used to print 'addi' vs 'add' instructions.
3501 - ``y``: For a memory operand, prints formatter for a two-register X-form
3502 instruction. (Currently always prints ``r0,OPERAND``).
3503 - ``U``: Prints 'u' if the memory operand is an update form, and nothing
3504 otherwise. (NOTE: LLVM does not support update form, so this will currently
3505 always print nothing)
3506 - ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
3507 not support indexed form, so this will currently always print nothing)
3515 SystemZ implements only ``n``, and does *not* support any of the other
3516 target-independent modifiers.
3520 - ``c``: Print an unadorned integer or symbol name. (The latter is
3521 target-specific behavior for this typically target-independent modifier).
3522 - ``A``: Print a register name with a '``*``' before it.
3523 - ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
3525 - ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
3527 - ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
3529 - ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
3531 - ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
3532 available, otherwise the 32-bit register name; do nothing on a memory operand.
3533 - ``n``: Negate and print an unadorned integer, or, for operands other than an
3534 immediate integer (e.g. a relocatable symbol expression), print a '-' before
3535 the operand. (The behavior for relocatable symbol expressions is a
3536 target-specific behavior for this typically target-independent modifier)
3537 - ``H``: Print a memory reference with additional offset +8.
3538 - ``P``: Print a memory reference or operand for use as the argument of a call
3539 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
3543 No additional modifiers.
3549 The call instructions that wrap inline asm nodes may have a
3550 "``!srcloc``" MDNode attached to it that contains a list of constant
3551 integers. If present, the code generator will use the integer as the
3552 location cookie value when report errors through the ``LLVMContext``
3553 error reporting mechanisms. This allows a front-end to correlate backend
3554 errors that occur with inline asm back to the source code that produced
3557 .. code-block:: llvm
3559 call void asm sideeffect "something bad", ""(), !srcloc !42
3561 !42 = !{ i32 1234567 }
3563 It is up to the front-end to make sense of the magic numbers it places
3564 in the IR. If the MDNode contains multiple constants, the code generator
3565 will use the one that corresponds to the line of the asm that the error
3573 LLVM IR allows metadata to be attached to instructions in the program
3574 that can convey extra information about the code to the optimizers and
3575 code generator. One example application of metadata is source-level
3576 debug information. There are two metadata primitives: strings and nodes.
3578 Metadata does not have a type, and is not a value. If referenced from a
3579 ``call`` instruction, it uses the ``metadata`` type.
3581 All metadata are identified in syntax by a exclamation point ('``!``').
3583 .. _metadata-string:
3585 Metadata Nodes and Metadata Strings
3586 -----------------------------------
3588 A metadata string is a string surrounded by double quotes. It can
3589 contain any character by escaping non-printable characters with
3590 "``\xx``" where "``xx``" is the two digit hex code. For example:
3593 Metadata nodes are represented with notation similar to structure
3594 constants (a comma separated list of elements, surrounded by braces and
3595 preceded by an exclamation point). Metadata nodes can have any values as
3596 their operand. For example:
3598 .. code-block:: llvm
3600 !{ !"test\00", i32 10}
3602 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
3604 .. code-block:: llvm
3606 !0 = distinct !{!"test\00", i32 10}
3608 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
3609 content. They can also occur when transformations cause uniquing collisions
3610 when metadata operands change.
3612 A :ref:`named metadata <namedmetadatastructure>` is a collection of
3613 metadata nodes, which can be looked up in the module symbol table. For
3616 .. code-block:: llvm
3620 Metadata can be used as function arguments. Here ``llvm.dbg.value``
3621 function is using two metadata arguments:
3623 .. code-block:: llvm
3625 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3627 Metadata can be attached with an instruction. Here metadata ``!21`` is
3628 attached to the ``add`` instruction using the ``!dbg`` identifier:
3630 .. code-block:: llvm
3632 %indvar.next = add i64 %indvar, 1, !dbg !21
3634 More information about specific metadata nodes recognized by the
3635 optimizers and code generator is found below.
3637 .. _specialized-metadata:
3639 Specialized Metadata Nodes
3640 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3642 Specialized metadata nodes are custom data structures in metadata (as opposed
3643 to generic tuples). Their fields are labelled, and can be specified in any
3646 These aren't inherently debug info centric, but currently all the specialized
3647 metadata nodes are related to debug info.
3654 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
3655 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
3656 tuples containing the debug info to be emitted along with the compile unit,
3657 regardless of code optimizations (some nodes are only emitted if there are
3658 references to them from instructions).
3660 .. code-block:: llvm
3662 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
3663 isOptimized: true, flags: "-O2", runtimeVersion: 2,
3664 splitDebugFilename: "abc.debug", emissionKind: 1,
3665 enums: !2, retainedTypes: !3, subprograms: !4,
3666 globals: !5, imports: !6)
3668 Compile unit descriptors provide the root scope for objects declared in a
3669 specific compilation unit. File descriptors are defined using this scope.
3670 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
3671 keep track of subprograms, global variables, type information, and imported
3672 entities (declarations and namespaces).
3679 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
3681 .. code-block:: llvm
3683 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
3685 Files are sometimes used in ``scope:`` fields, and are the only valid target
3686 for ``file:`` fields.
3693 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
3694 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
3696 .. code-block:: llvm
3698 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3699 encoding: DW_ATE_unsigned_char)
3700 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
3702 The ``encoding:`` describes the details of the type. Usually it's one of the
3705 .. code-block:: llvm
3711 DW_ATE_signed_char = 6
3713 DW_ATE_unsigned_char = 8
3715 .. _DISubroutineType:
3720 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3721 refers to a tuple; the first operand is the return type, while the rest are the
3722 types of the formal arguments in order. If the first operand is ``null``, that
3723 represents a function with no return value (such as ``void foo() {}`` in C++).
3725 .. code-block:: llvm
3727 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3728 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3729 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3736 ``DIDerivedType`` nodes represent types derived from other types, such as
3739 .. code-block:: llvm
3741 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3742 encoding: DW_ATE_unsigned_char)
3743 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3746 The following ``tag:`` values are valid:
3748 .. code-block:: llvm
3750 DW_TAG_formal_parameter = 5
3752 DW_TAG_pointer_type = 15
3753 DW_TAG_reference_type = 16
3755 DW_TAG_ptr_to_member_type = 31
3756 DW_TAG_const_type = 38
3757 DW_TAG_volatile_type = 53
3758 DW_TAG_restrict_type = 55
3760 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3761 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3762 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3763 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3764 argument of a subprogram.
3766 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3768 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3769 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3772 Note that the ``void *`` type is expressed as a type derived from NULL.
3774 .. _DICompositeType:
3779 ``DICompositeType`` nodes represent types composed of other types, like
3780 structures and unions. ``elements:`` points to a tuple of the composed types.
3782 If the source language supports ODR, the ``identifier:`` field gives the unique
3783 identifier used for type merging between modules. When specified, other types
3784 can refer to composite types indirectly via a :ref:`metadata string
3785 <metadata-string>` that matches their identifier.
3787 .. code-block:: llvm
3789 !0 = !DIEnumerator(name: "SixKind", value: 7)
3790 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3791 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3792 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3793 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3794 elements: !{!0, !1, !2})
3796 The following ``tag:`` values are valid:
3798 .. code-block:: llvm
3800 DW_TAG_array_type = 1
3801 DW_TAG_class_type = 2
3802 DW_TAG_enumeration_type = 4
3803 DW_TAG_structure_type = 19
3804 DW_TAG_union_type = 23
3805 DW_TAG_subroutine_type = 21
3806 DW_TAG_inheritance = 28
3809 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3810 descriptors <DISubrange>`, each representing the range of subscripts at that
3811 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3812 array type is a native packed vector.
3814 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3815 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3816 value for the set. All enumeration type descriptors are collected in the
3817 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3819 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3820 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3821 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3828 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3829 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3831 .. code-block:: llvm
3833 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3834 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3835 !2 = !DISubrange(count: -1) ; empty array.
3842 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3843 variants of :ref:`DICompositeType`.
3845 .. code-block:: llvm
3847 !0 = !DIEnumerator(name: "SixKind", value: 7)
3848 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3849 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3851 DITemplateTypeParameter
3852 """""""""""""""""""""""
3854 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3855 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3856 :ref:`DISubprogram` ``templateParams:`` fields.
3858 .. code-block:: llvm
3860 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3862 DITemplateValueParameter
3863 """"""""""""""""""""""""
3865 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3866 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3867 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3868 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3869 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3871 .. code-block:: llvm
3873 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3878 ``DINamespace`` nodes represent namespaces in the source language.
3880 .. code-block:: llvm
3882 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3887 ``DIGlobalVariable`` nodes represent global variables in the source language.
3889 .. code-block:: llvm
3891 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3892 file: !2, line: 7, type: !3, isLocal: true,
3893 isDefinition: false, variable: i32* @foo,
3896 All global variables should be referenced by the `globals:` field of a
3897 :ref:`compile unit <DICompileUnit>`.
3904 ``DISubprogram`` nodes represent functions from the source language. The
3905 ``variables:`` field points at :ref:`variables <DILocalVariable>` that must be
3906 retained, even if their IR counterparts are optimized out of the IR. The
3907 ``type:`` field must point at an :ref:`DISubroutineType`.
3909 .. code-block:: llvm
3911 !0 = !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3912 file: !2, line: 7, type: !3, isLocal: true,
3913 isDefinition: false, scopeLine: 8, containingType: !4,
3914 virtuality: DW_VIRTUALITY_pure_virtual, virtualIndex: 10,
3915 flags: DIFlagPrototyped, isOptimized: true,
3916 function: void ()* @_Z3foov,
3917 templateParams: !5, declaration: !6, variables: !7)
3924 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3925 <DISubprogram>`. The line number and column numbers are used to distinguish
3926 two lexical blocks at same depth. They are valid targets for ``scope:``
3929 .. code-block:: llvm
3931 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3933 Usually lexical blocks are ``distinct`` to prevent node merging based on
3936 .. _DILexicalBlockFile:
3941 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3942 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
3943 indicate textual inclusion, or the ``discriminator:`` field can be used to
3944 discriminate between control flow within a single block in the source language.
3946 .. code-block:: llvm
3948 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3949 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3950 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3957 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
3958 mandatory, and points at an :ref:`DILexicalBlockFile`, an
3959 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
3961 .. code-block:: llvm
3963 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3965 .. _DILocalVariable:
3970 ``DILocalVariable`` nodes represent local variables in the source language. If
3971 the ``arg:`` field is set to non-zero, then this variable is a subprogram
3972 parameter, and it will be included in the ``variables:`` field of its
3973 :ref:`DISubprogram`.
3975 .. code-block:: llvm
3977 !0 = !DILocalVariable(name: "this", arg: 1, scope: !3, file: !2, line: 7,
3978 type: !3, flags: DIFlagArtificial)
3979 !1 = !DILocalVariable(name: "x", arg: 2, scope: !4, file: !2, line: 7,
3981 !2 = !DILocalVariable(name: "y", scope: !5, file: !2, line: 7, type: !3)
3986 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
3987 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
3988 describe how the referenced LLVM variable relates to the source language
3991 The current supported vocabulary is limited:
3993 - ``DW_OP_deref`` dereferences the working expression.
3994 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
3995 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
3996 here, respectively) of the variable piece from the working expression.
3998 .. code-block:: llvm
4000 !0 = !DIExpression(DW_OP_deref)
4001 !1 = !DIExpression(DW_OP_plus, 3)
4002 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
4003 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
4008 ``DIObjCProperty`` nodes represent Objective-C property nodes.
4010 .. code-block:: llvm
4012 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
4013 getter: "getFoo", attributes: 7, type: !2)
4018 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
4021 .. code-block:: llvm
4023 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
4024 entity: !1, line: 7)
4029 In LLVM IR, memory does not have types, so LLVM's own type system is not
4030 suitable for doing TBAA. Instead, metadata is added to the IR to
4031 describe a type system of a higher level language. This can be used to
4032 implement typical C/C++ TBAA, but it can also be used to implement
4033 custom alias analysis behavior for other languages.
4035 The current metadata format is very simple. TBAA metadata nodes have up
4036 to three fields, e.g.:
4038 .. code-block:: llvm
4040 !0 = !{ !"an example type tree" }
4041 !1 = !{ !"int", !0 }
4042 !2 = !{ !"float", !0 }
4043 !3 = !{ !"const float", !2, i64 1 }
4045 The first field is an identity field. It can be any value, usually a
4046 metadata string, which uniquely identifies the type. The most important
4047 name in the tree is the name of the root node. Two trees with different
4048 root node names are entirely disjoint, even if they have leaves with
4051 The second field identifies the type's parent node in the tree, or is
4052 null or omitted for a root node. A type is considered to alias all of
4053 its descendants and all of its ancestors in the tree. Also, a type is
4054 considered to alias all types in other trees, so that bitcode produced
4055 from multiple front-ends is handled conservatively.
4057 If the third field is present, it's an integer which if equal to 1
4058 indicates that the type is "constant" (meaning
4059 ``pointsToConstantMemory`` should return true; see `other useful
4060 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
4062 '``tbaa.struct``' Metadata
4063 ^^^^^^^^^^^^^^^^^^^^^^^^^^
4065 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
4066 aggregate assignment operations in C and similar languages, however it
4067 is defined to copy a contiguous region of memory, which is more than
4068 strictly necessary for aggregate types which contain holes due to
4069 padding. Also, it doesn't contain any TBAA information about the fields
4072 ``!tbaa.struct`` metadata can describe which memory subregions in a
4073 memcpy are padding and what the TBAA tags of the struct are.
4075 The current metadata format is very simple. ``!tbaa.struct`` metadata
4076 nodes are a list of operands which are in conceptual groups of three.
4077 For each group of three, the first operand gives the byte offset of a
4078 field in bytes, the second gives its size in bytes, and the third gives
4081 .. code-block:: llvm
4083 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
4085 This describes a struct with two fields. The first is at offset 0 bytes
4086 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4087 and has size 4 bytes and has tbaa tag !2.
4089 Note that the fields need not be contiguous. In this example, there is a
4090 4 byte gap between the two fields. This gap represents padding which
4091 does not carry useful data and need not be preserved.
4093 '``noalias``' and '``alias.scope``' Metadata
4094 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4096 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4097 noalias memory-access sets. This means that some collection of memory access
4098 instructions (loads, stores, memory-accessing calls, etc.) that carry
4099 ``noalias`` metadata can specifically be specified not to alias with some other
4100 collection of memory access instructions that carry ``alias.scope`` metadata.
4101 Each type of metadata specifies a list of scopes where each scope has an id and
4102 a domain. When evaluating an aliasing query, if for some domain, the set
4103 of scopes with that domain in one instruction's ``alias.scope`` list is a
4104 subset of (or equal to) the set of scopes for that domain in another
4105 instruction's ``noalias`` list, then the two memory accesses are assumed not to
4108 The metadata identifying each domain is itself a list containing one or two
4109 entries. The first entry is the name of the domain. Note that if the name is a
4110 string then it can be combined across functions and translation units. A
4111 self-reference can be used to create globally unique domain names. A
4112 descriptive string may optionally be provided as a second list entry.
4114 The metadata identifying each scope is also itself a list containing two or
4115 three entries. The first entry is the name of the scope. Note that if the name
4116 is a string then it can be combined across functions and translation units. A
4117 self-reference can be used to create globally unique scope names. A metadata
4118 reference to the scope's domain is the second entry. A descriptive string may
4119 optionally be provided as a third list entry.
4123 .. code-block:: llvm
4125 ; Two scope domains:
4129 ; Some scopes in these domains:
4135 !5 = !{!4} ; A list containing only scope !4
4139 ; These two instructions don't alias:
4140 %0 = load float, float* %c, align 4, !alias.scope !5
4141 store float %0, float* %arrayidx.i, align 4, !noalias !5
4143 ; These two instructions also don't alias (for domain !1, the set of scopes
4144 ; in the !alias.scope equals that in the !noalias list):
4145 %2 = load float, float* %c, align 4, !alias.scope !5
4146 store float %2, float* %arrayidx.i2, align 4, !noalias !6
4148 ; These two instructions may alias (for domain !0, the set of scopes in
4149 ; the !noalias list is not a superset of, or equal to, the scopes in the
4150 ; !alias.scope list):
4151 %2 = load float, float* %c, align 4, !alias.scope !6
4152 store float %0, float* %arrayidx.i, align 4, !noalias !7
4154 '``fpmath``' Metadata
4155 ^^^^^^^^^^^^^^^^^^^^^
4157 ``fpmath`` metadata may be attached to any instruction of floating point
4158 type. It can be used to express the maximum acceptable error in the
4159 result of that instruction, in ULPs, thus potentially allowing the
4160 compiler to use a more efficient but less accurate method of computing
4161 it. ULP is defined as follows:
4163 If ``x`` is a real number that lies between two finite consecutive
4164 floating-point numbers ``a`` and ``b``, without being equal to one
4165 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4166 distance between the two non-equal finite floating-point numbers
4167 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4169 The metadata node shall consist of a single positive floating point
4170 number representing the maximum relative error, for example:
4172 .. code-block:: llvm
4174 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
4178 '``range``' Metadata
4179 ^^^^^^^^^^^^^^^^^^^^
4181 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4182 integer types. It expresses the possible ranges the loaded value or the value
4183 returned by the called function at this call site is in. The ranges are
4184 represented with a flattened list of integers. The loaded value or the value
4185 returned is known to be in the union of the ranges defined by each consecutive
4186 pair. Each pair has the following properties:
4188 - The type must match the type loaded by the instruction.
4189 - The pair ``a,b`` represents the range ``[a,b)``.
4190 - Both ``a`` and ``b`` are constants.
4191 - The range is allowed to wrap.
4192 - The range should not represent the full or empty set. That is,
4195 In addition, the pairs must be in signed order of the lower bound and
4196 they must be non-contiguous.
4200 .. code-block:: llvm
4202 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4203 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
4204 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4205 %d = invoke i8 @bar() to label %cont
4206 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
4208 !0 = !{ i8 0, i8 2 }
4209 !1 = !{ i8 255, i8 2 }
4210 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4211 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
4213 '``unpredictable``' Metadata
4214 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4216 ``unpredictable`` metadata may be attached to any branch or switch
4217 instruction. It can be used to express the unpredictability of control
4218 flow. Similar to the llvm.expect intrinsic, it may be used to alter
4219 optimizations related to compare and branch instructions. The metadata
4220 is treated as a boolean value; if it exists, it signals that the branch
4221 or switch that it is attached to is completely unpredictable.
4226 It is sometimes useful to attach information to loop constructs. Currently,
4227 loop metadata is implemented as metadata attached to the branch instruction
4228 in the loop latch block. This type of metadata refer to a metadata node that is
4229 guaranteed to be separate for each loop. The loop identifier metadata is
4230 specified with the name ``llvm.loop``.
4232 The loop identifier metadata is implemented using a metadata that refers to
4233 itself to avoid merging it with any other identifier metadata, e.g.,
4234 during module linkage or function inlining. That is, each loop should refer
4235 to their own identification metadata even if they reside in separate functions.
4236 The following example contains loop identifier metadata for two separate loop
4239 .. code-block:: llvm
4244 The loop identifier metadata can be used to specify additional
4245 per-loop metadata. Any operands after the first operand can be treated
4246 as user-defined metadata. For example the ``llvm.loop.unroll.count``
4247 suggests an unroll factor to the loop unroller:
4249 .. code-block:: llvm
4251 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
4254 !1 = !{!"llvm.loop.unroll.count", i32 4}
4256 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
4257 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4259 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
4260 used to control per-loop vectorization and interleaving parameters such as
4261 vectorization width and interleave count. These metadata should be used in
4262 conjunction with ``llvm.loop`` loop identification metadata. The
4263 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
4264 optimization hints and the optimizer will only interleave and vectorize loops if
4265 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
4266 which contains information about loop-carried memory dependencies can be helpful
4267 in determining the safety of these transformations.
4269 '``llvm.loop.interleave.count``' Metadata
4270 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4272 This metadata suggests an interleave count to the loop interleaver.
4273 The first operand is the string ``llvm.loop.interleave.count`` and the
4274 second operand is an integer specifying the interleave count. For
4277 .. code-block:: llvm
4279 !0 = !{!"llvm.loop.interleave.count", i32 4}
4281 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
4282 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
4283 then the interleave count will be determined automatically.
4285 '``llvm.loop.vectorize.enable``' Metadata
4286 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4288 This metadata selectively enables or disables vectorization for the loop. The
4289 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
4290 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
4291 0 disables vectorization:
4293 .. code-block:: llvm
4295 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
4296 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
4298 '``llvm.loop.vectorize.width``' Metadata
4299 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4301 This metadata sets the target width of the vectorizer. The first
4302 operand is the string ``llvm.loop.vectorize.width`` and the second
4303 operand is an integer specifying the width. For example:
4305 .. code-block:: llvm
4307 !0 = !{!"llvm.loop.vectorize.width", i32 4}
4309 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
4310 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
4311 0 or if the loop does not have this metadata the width will be
4312 determined automatically.
4314 '``llvm.loop.unroll``'
4315 ^^^^^^^^^^^^^^^^^^^^^^
4317 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
4318 optimization hints such as the unroll factor. ``llvm.loop.unroll``
4319 metadata should be used in conjunction with ``llvm.loop`` loop
4320 identification metadata. The ``llvm.loop.unroll`` metadata are only
4321 optimization hints and the unrolling will only be performed if the
4322 optimizer believes it is safe to do so.
4324 '``llvm.loop.unroll.count``' Metadata
4325 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4327 This metadata suggests an unroll factor to the loop unroller. The
4328 first operand is the string ``llvm.loop.unroll.count`` and the second
4329 operand is a positive integer specifying the unroll factor. For
4332 .. code-block:: llvm
4334 !0 = !{!"llvm.loop.unroll.count", i32 4}
4336 If the trip count of the loop is less than the unroll count the loop
4337 will be partially unrolled.
4339 '``llvm.loop.unroll.disable``' Metadata
4340 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4342 This metadata disables loop unrolling. The metadata has a single operand
4343 which is the string ``llvm.loop.unroll.disable``. For example:
4345 .. code-block:: llvm
4347 !0 = !{!"llvm.loop.unroll.disable"}
4349 '``llvm.loop.unroll.runtime.disable``' Metadata
4350 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4352 This metadata disables runtime loop unrolling. The metadata has a single
4353 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
4355 .. code-block:: llvm
4357 !0 = !{!"llvm.loop.unroll.runtime.disable"}
4359 '``llvm.loop.unroll.enable``' Metadata
4360 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4362 This metadata suggests that the loop should be fully unrolled if the trip count
4363 is known at compile time and partially unrolled if the trip count is not known
4364 at compile time. The metadata has a single operand which is the string
4365 ``llvm.loop.unroll.enable``. For example:
4367 .. code-block:: llvm
4369 !0 = !{!"llvm.loop.unroll.enable"}
4371 '``llvm.loop.unroll.full``' Metadata
4372 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4374 This metadata suggests that the loop should be unrolled fully. The
4375 metadata has a single operand which is the string ``llvm.loop.unroll.full``.
4378 .. code-block:: llvm
4380 !0 = !{!"llvm.loop.unroll.full"}
4385 Metadata types used to annotate memory accesses with information helpful
4386 for optimizations are prefixed with ``llvm.mem``.
4388 '``llvm.mem.parallel_loop_access``' Metadata
4389 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4391 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
4392 or metadata containing a list of loop identifiers for nested loops.
4393 The metadata is attached to memory accessing instructions and denotes that
4394 no loop carried memory dependence exist between it and other instructions denoted
4395 with the same loop identifier.
4397 Precisely, given two instructions ``m1`` and ``m2`` that both have the
4398 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
4399 set of loops associated with that metadata, respectively, then there is no loop
4400 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
4403 As a special case, if all memory accessing instructions in a loop have
4404 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
4405 loop has no loop carried memory dependences and is considered to be a parallel
4408 Note that if not all memory access instructions have such metadata referring to
4409 the loop, then the loop is considered not being trivially parallel. Additional
4410 memory dependence analysis is required to make that determination. As a fail
4411 safe mechanism, this causes loops that were originally parallel to be considered
4412 sequential (if optimization passes that are unaware of the parallel semantics
4413 insert new memory instructions into the loop body).
4415 Example of a loop that is considered parallel due to its correct use of
4416 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
4417 metadata types that refer to the same loop identifier metadata.
4419 .. code-block:: llvm
4423 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
4425 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4427 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
4433 It is also possible to have nested parallel loops. In that case the
4434 memory accesses refer to a list of loop identifier metadata nodes instead of
4435 the loop identifier metadata node directly:
4437 .. code-block:: llvm
4441 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
4443 br label %inner.for.body
4447 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4449 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
4451 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
4455 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
4457 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
4459 outer.for.end: ; preds = %for.body
4461 !0 = !{!1, !2} ; a list of loop identifiers
4462 !1 = !{!1} ; an identifier for the inner loop
4463 !2 = !{!2} ; an identifier for the outer loop
4468 The ``llvm.bitsets`` global metadata is used to implement
4469 :doc:`bitsets <BitSets>`.
4471 '``invariant.group``' Metadata
4472 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4474 The ``invariant.group`` metadata may be attached to ``load``/``store`` instructions.
4475 The existence of the ``invariant.group`` metadata on the instruction tells
4476 the optimizer that every ``load`` and ``store`` to the same pointer operand
4477 within the same invariant group can be assumed to load or store the same
4478 value (but see the ``llvm.invariant.group.barrier`` intrinsic which affects
4479 when two pointers are considered the same).
4483 .. code-block:: llvm
4485 @unknownPtr = external global i8
4488 store i8 42, i8* %ptr, !invariant.group !0
4489 call void @foo(i8* %ptr)
4491 %a = load i8, i8* %ptr, !invariant.group !0 ; Can assume that value under %ptr didn't change
4492 call void @foo(i8* %ptr)
4493 %b = load i8, i8* %ptr, !invariant.group !1 ; Can't assume anything, because group changed
4495 %newPtr = call i8* @getPointer(i8* %ptr)
4496 %c = load i8, i8* %newPtr, !invariant.group !0 ; Can't assume anything, because we only have information about %ptr
4498 %unknownValue = load i8, i8* @unknownPtr
4499 store i8 %unknownValue, i8* %ptr, !invariant.group !0 ; Can assume that %unknownValue == 42
4501 call void @foo(i8* %ptr)
4502 %newPtr2 = call i8* @llvm.invariant.group.barrier(i8* %ptr)
4503 %d = load i8, i8* %newPtr2, !invariant.group !0 ; Can't step through invariant.group.barrier to get value of %ptr
4506 declare void @foo(i8*)
4507 declare i8* @getPointer(i8*)
4508 declare i8* @llvm.invariant.group.barrier(i8*)
4510 !0 = !{!"magic ptr"}
4511 !1 = !{!"other ptr"}
4515 Module Flags Metadata
4516 =====================
4518 Information about the module as a whole is difficult to convey to LLVM's
4519 subsystems. The LLVM IR isn't sufficient to transmit this information.
4520 The ``llvm.module.flags`` named metadata exists in order to facilitate
4521 this. These flags are in the form of key / value pairs --- much like a
4522 dictionary --- making it easy for any subsystem who cares about a flag to
4525 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
4526 Each triplet has the following form:
4528 - The first element is a *behavior* flag, which specifies the behavior
4529 when two (or more) modules are merged together, and it encounters two
4530 (or more) metadata with the same ID. The supported behaviors are
4532 - The second element is a metadata string that is a unique ID for the
4533 metadata. Each module may only have one flag entry for each unique ID (not
4534 including entries with the **Require** behavior).
4535 - The third element is the value of the flag.
4537 When two (or more) modules are merged together, the resulting
4538 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
4539 each unique metadata ID string, there will be exactly one entry in the merged
4540 modules ``llvm.module.flags`` metadata table, and the value for that entry will
4541 be determined by the merge behavior flag, as described below. The only exception
4542 is that entries with the *Require* behavior are always preserved.
4544 The following behaviors are supported:
4555 Emits an error if two values disagree, otherwise the resulting value
4556 is that of the operands.
4560 Emits a warning if two values disagree. The result value will be the
4561 operand for the flag from the first module being linked.
4565 Adds a requirement that another module flag be present and have a
4566 specified value after linking is performed. The value must be a
4567 metadata pair, where the first element of the pair is the ID of the
4568 module flag to be restricted, and the second element of the pair is
4569 the value the module flag should be restricted to. This behavior can
4570 be used to restrict the allowable results (via triggering of an
4571 error) of linking IDs with the **Override** behavior.
4575 Uses the specified value, regardless of the behavior or value of the
4576 other module. If both modules specify **Override**, but the values
4577 differ, an error will be emitted.
4581 Appends the two values, which are required to be metadata nodes.
4585 Appends the two values, which are required to be metadata
4586 nodes. However, duplicate entries in the second list are dropped
4587 during the append operation.
4589 It is an error for a particular unique flag ID to have multiple behaviors,
4590 except in the case of **Require** (which adds restrictions on another metadata
4591 value) or **Override**.
4593 An example of module flags:
4595 .. code-block:: llvm
4597 !0 = !{ i32 1, !"foo", i32 1 }
4598 !1 = !{ i32 4, !"bar", i32 37 }
4599 !2 = !{ i32 2, !"qux", i32 42 }
4600 !3 = !{ i32 3, !"qux",
4605 !llvm.module.flags = !{ !0, !1, !2, !3 }
4607 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
4608 if two or more ``!"foo"`` flags are seen is to emit an error if their
4609 values are not equal.
4611 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
4612 behavior if two or more ``!"bar"`` flags are seen is to use the value
4615 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
4616 behavior if two or more ``!"qux"`` flags are seen is to emit a
4617 warning if their values are not equal.
4619 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
4625 The behavior is to emit an error if the ``llvm.module.flags`` does not
4626 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
4629 Objective-C Garbage Collection Module Flags Metadata
4630 ----------------------------------------------------
4632 On the Mach-O platform, Objective-C stores metadata about garbage
4633 collection in a special section called "image info". The metadata
4634 consists of a version number and a bitmask specifying what types of
4635 garbage collection are supported (if any) by the file. If two or more
4636 modules are linked together their garbage collection metadata needs to
4637 be merged rather than appended together.
4639 The Objective-C garbage collection module flags metadata consists of the
4640 following key-value pairs:
4649 * - ``Objective-C Version``
4650 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
4652 * - ``Objective-C Image Info Version``
4653 - **[Required]** --- The version of the image info section. Currently
4656 * - ``Objective-C Image Info Section``
4657 - **[Required]** --- The section to place the metadata. Valid values are
4658 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
4659 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
4660 Objective-C ABI version 2.
4662 * - ``Objective-C Garbage Collection``
4663 - **[Required]** --- Specifies whether garbage collection is supported or
4664 not. Valid values are 0, for no garbage collection, and 2, for garbage
4665 collection supported.
4667 * - ``Objective-C GC Only``
4668 - **[Optional]** --- Specifies that only garbage collection is supported.
4669 If present, its value must be 6. This flag requires that the
4670 ``Objective-C Garbage Collection`` flag have the value 2.
4672 Some important flag interactions:
4674 - If a module with ``Objective-C Garbage Collection`` set to 0 is
4675 merged with a module with ``Objective-C Garbage Collection`` set to
4676 2, then the resulting module has the
4677 ``Objective-C Garbage Collection`` flag set to 0.
4678 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
4679 merged with a module with ``Objective-C GC Only`` set to 6.
4681 Automatic Linker Flags Module Flags Metadata
4682 --------------------------------------------
4684 Some targets support embedding flags to the linker inside individual object
4685 files. Typically this is used in conjunction with language extensions which
4686 allow source files to explicitly declare the libraries they depend on, and have
4687 these automatically be transmitted to the linker via object files.
4689 These flags are encoded in the IR using metadata in the module flags section,
4690 using the ``Linker Options`` key. The merge behavior for this flag is required
4691 to be ``AppendUnique``, and the value for the key is expected to be a metadata
4692 node which should be a list of other metadata nodes, each of which should be a
4693 list of metadata strings defining linker options.
4695 For example, the following metadata section specifies two separate sets of
4696 linker options, presumably to link against ``libz`` and the ``Cocoa``
4699 !0 = !{ i32 6, !"Linker Options",
4702 !{ !"-framework", !"Cocoa" } } }
4703 !llvm.module.flags = !{ !0 }
4705 The metadata encoding as lists of lists of options, as opposed to a collapsed
4706 list of options, is chosen so that the IR encoding can use multiple option
4707 strings to specify e.g., a single library, while still having that specifier be
4708 preserved as an atomic element that can be recognized by a target specific
4709 assembly writer or object file emitter.
4711 Each individual option is required to be either a valid option for the target's
4712 linker, or an option that is reserved by the target specific assembly writer or
4713 object file emitter. No other aspect of these options is defined by the IR.
4715 C type width Module Flags Metadata
4716 ----------------------------------
4718 The ARM backend emits a section into each generated object file describing the
4719 options that it was compiled with (in a compiler-independent way) to prevent
4720 linking incompatible objects, and to allow automatic library selection. Some
4721 of these options are not visible at the IR level, namely wchar_t width and enum
4724 To pass this information to the backend, these options are encoded in module
4725 flags metadata, using the following key-value pairs:
4735 - * 0 --- sizeof(wchar_t) == 4
4736 * 1 --- sizeof(wchar_t) == 2
4739 - * 0 --- Enums are at least as large as an ``int``.
4740 * 1 --- Enums are stored in the smallest integer type which can
4741 represent all of its values.
4743 For example, the following metadata section specifies that the module was
4744 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
4745 enum is the smallest type which can represent all of its values::
4747 !llvm.module.flags = !{!0, !1}
4748 !0 = !{i32 1, !"short_wchar", i32 1}
4749 !1 = !{i32 1, !"short_enum", i32 0}
4751 .. _intrinsicglobalvariables:
4753 Intrinsic Global Variables
4754 ==========================
4756 LLVM has a number of "magic" global variables that contain data that
4757 affect code generation or other IR semantics. These are documented here.
4758 All globals of this sort should have a section specified as
4759 "``llvm.metadata``". This section and all globals that start with
4760 "``llvm.``" are reserved for use by LLVM.
4764 The '``llvm.used``' Global Variable
4765 -----------------------------------
4767 The ``@llvm.used`` global is an array which has
4768 :ref:`appending linkage <linkage_appending>`. This array contains a list of
4769 pointers to named global variables, functions and aliases which may optionally
4770 have a pointer cast formed of bitcast or getelementptr. For example, a legal
4773 .. code-block:: llvm
4778 @llvm.used = appending global [2 x i8*] [
4780 i8* bitcast (i32* @Y to i8*)
4781 ], section "llvm.metadata"
4783 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4784 and linker are required to treat the symbol as if there is a reference to the
4785 symbol that it cannot see (which is why they have to be named). For example, if
4786 a variable has internal linkage and no references other than that from the
4787 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4788 references from inline asms and other things the compiler cannot "see", and
4789 corresponds to "``attribute((used))``" in GNU C.
4791 On some targets, the code generator must emit a directive to the
4792 assembler or object file to prevent the assembler and linker from
4793 molesting the symbol.
4795 .. _gv_llvmcompilerused:
4797 The '``llvm.compiler.used``' Global Variable
4798 --------------------------------------------
4800 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4801 directive, except that it only prevents the compiler from touching the
4802 symbol. On targets that support it, this allows an intelligent linker to
4803 optimize references to the symbol without being impeded as it would be
4806 This is a rare construct that should only be used in rare circumstances,
4807 and should not be exposed to source languages.
4809 .. _gv_llvmglobalctors:
4811 The '``llvm.global_ctors``' Global Variable
4812 -------------------------------------------
4814 .. code-block:: llvm
4816 %0 = type { i32, void ()*, i8* }
4817 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4819 The ``@llvm.global_ctors`` array contains a list of constructor
4820 functions, priorities, and an optional associated global or function.
4821 The functions referenced by this array will be called in ascending order
4822 of priority (i.e. lowest first) when the module is loaded. The order of
4823 functions with the same priority is not defined.
4825 If the third field is present, non-null, and points to a global variable
4826 or function, the initializer function will only run if the associated
4827 data from the current module is not discarded.
4829 .. _llvmglobaldtors:
4831 The '``llvm.global_dtors``' Global Variable
4832 -------------------------------------------
4834 .. code-block:: llvm
4836 %0 = type { i32, void ()*, i8* }
4837 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4839 The ``@llvm.global_dtors`` array contains a list of destructor
4840 functions, priorities, and an optional associated global or function.
4841 The functions referenced by this array will be called in descending
4842 order of priority (i.e. highest first) when the module is unloaded. The
4843 order of functions with the same priority is not defined.
4845 If the third field is present, non-null, and points to a global variable
4846 or function, the destructor function will only run if the associated
4847 data from the current module is not discarded.
4849 Instruction Reference
4850 =====================
4852 The LLVM instruction set consists of several different classifications
4853 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4854 instructions <binaryops>`, :ref:`bitwise binary
4855 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4856 :ref:`other instructions <otherops>`.
4860 Terminator Instructions
4861 -----------------------
4863 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4864 program ends with a "Terminator" instruction, which indicates which
4865 block should be executed after the current block is finished. These
4866 terminator instructions typically yield a '``void``' value: they produce
4867 control flow, not values (the one exception being the
4868 ':ref:`invoke <i_invoke>`' instruction).
4870 The terminator instructions are: ':ref:`ret <i_ret>`',
4871 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4872 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4873 ':ref:`resume <i_resume>`', ':ref:`catchpad <i_catchpad>`',
4874 ':ref:`catchendpad <i_catchendpad>`',
4875 ':ref:`catchret <i_catchret>`',
4876 ':ref:`cleanupendpad <i_cleanupendpad>`',
4877 ':ref:`cleanupret <i_cleanupret>`',
4878 ':ref:`terminatepad <i_terminatepad>`',
4879 and ':ref:`unreachable <i_unreachable>`'.
4883 '``ret``' Instruction
4884 ^^^^^^^^^^^^^^^^^^^^^
4891 ret <type> <value> ; Return a value from a non-void function
4892 ret void ; Return from void function
4897 The '``ret``' instruction is used to return control flow (and optionally
4898 a value) from a function back to the caller.
4900 There are two forms of the '``ret``' instruction: one that returns a
4901 value and then causes control flow, and one that just causes control
4907 The '``ret``' instruction optionally accepts a single argument, the
4908 return value. The type of the return value must be a ':ref:`first
4909 class <t_firstclass>`' type.
4911 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4912 return type and contains a '``ret``' instruction with no return value or
4913 a return value with a type that does not match its type, or if it has a
4914 void return type and contains a '``ret``' instruction with a return
4920 When the '``ret``' instruction is executed, control flow returns back to
4921 the calling function's context. If the caller is a
4922 ":ref:`call <i_call>`" instruction, execution continues at the
4923 instruction after the call. If the caller was an
4924 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4925 beginning of the "normal" destination block. If the instruction returns
4926 a value, that value shall set the call or invoke instruction's return
4932 .. code-block:: llvm
4934 ret i32 5 ; Return an integer value of 5
4935 ret void ; Return from a void function
4936 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4940 '``br``' Instruction
4941 ^^^^^^^^^^^^^^^^^^^^
4948 br i1 <cond>, label <iftrue>, label <iffalse>
4949 br label <dest> ; Unconditional branch
4954 The '``br``' instruction is used to cause control flow to transfer to a
4955 different basic block in the current function. There are two forms of
4956 this instruction, corresponding to a conditional branch and an
4957 unconditional branch.
4962 The conditional branch form of the '``br``' instruction takes a single
4963 '``i1``' value and two '``label``' values. The unconditional form of the
4964 '``br``' instruction takes a single '``label``' value as a target.
4969 Upon execution of a conditional '``br``' instruction, the '``i1``'
4970 argument is evaluated. If the value is ``true``, control flows to the
4971 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4972 to the '``iffalse``' ``label`` argument.
4977 .. code-block:: llvm
4980 %cond = icmp eq i32 %a, %b
4981 br i1 %cond, label %IfEqual, label %IfUnequal
4989 '``switch``' Instruction
4990 ^^^^^^^^^^^^^^^^^^^^^^^^
4997 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
5002 The '``switch``' instruction is used to transfer control flow to one of
5003 several different places. It is a generalization of the '``br``'
5004 instruction, allowing a branch to occur to one of many possible
5010 The '``switch``' instruction uses three parameters: an integer
5011 comparison value '``value``', a default '``label``' destination, and an
5012 array of pairs of comparison value constants and '``label``'s. The table
5013 is not allowed to contain duplicate constant entries.
5018 The ``switch`` instruction specifies a table of values and destinations.
5019 When the '``switch``' instruction is executed, this table is searched
5020 for the given value. If the value is found, control flow is transferred
5021 to the corresponding destination; otherwise, control flow is transferred
5022 to the default destination.
5027 Depending on properties of the target machine and the particular
5028 ``switch`` instruction, this instruction may be code generated in
5029 different ways. For example, it could be generated as a series of
5030 chained conditional branches or with a lookup table.
5035 .. code-block:: llvm
5037 ; Emulate a conditional br instruction
5038 %Val = zext i1 %value to i32
5039 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
5041 ; Emulate an unconditional br instruction
5042 switch i32 0, label %dest [ ]
5044 ; Implement a jump table:
5045 switch i32 %val, label %otherwise [ i32 0, label %onzero
5047 i32 2, label %ontwo ]
5051 '``indirectbr``' Instruction
5052 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5059 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
5064 The '``indirectbr``' instruction implements an indirect branch to a
5065 label within the current function, whose address is specified by
5066 "``address``". Address must be derived from a
5067 :ref:`blockaddress <blockaddress>` constant.
5072 The '``address``' argument is the address of the label to jump to. The
5073 rest of the arguments indicate the full set of possible destinations
5074 that the address may point to. Blocks are allowed to occur multiple
5075 times in the destination list, though this isn't particularly useful.
5077 This destination list is required so that dataflow analysis has an
5078 accurate understanding of the CFG.
5083 Control transfers to the block specified in the address argument. All
5084 possible destination blocks must be listed in the label list, otherwise
5085 this instruction has undefined behavior. This implies that jumps to
5086 labels defined in other functions have undefined behavior as well.
5091 This is typically implemented with a jump through a register.
5096 .. code-block:: llvm
5098 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
5102 '``invoke``' Instruction
5103 ^^^^^^^^^^^^^^^^^^^^^^^^
5110 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
5111 [operand bundles] to label <normal label> unwind label <exception label>
5116 The '``invoke``' instruction causes control to transfer to a specified
5117 function, with the possibility of control flow transfer to either the
5118 '``normal``' label or the '``exception``' label. If the callee function
5119 returns with the "``ret``" instruction, control flow will return to the
5120 "normal" label. If the callee (or any indirect callees) returns via the
5121 ":ref:`resume <i_resume>`" instruction or other exception handling
5122 mechanism, control is interrupted and continued at the dynamically
5123 nearest "exception" label.
5125 The '``exception``' label is a `landing
5126 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
5127 '``exception``' label is required to have the
5128 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
5129 information about the behavior of the program after unwinding happens,
5130 as its first non-PHI instruction. The restrictions on the
5131 "``landingpad``" instruction's tightly couples it to the "``invoke``"
5132 instruction, so that the important information contained within the
5133 "``landingpad``" instruction can't be lost through normal code motion.
5138 This instruction requires several arguments:
5140 #. The optional "cconv" marker indicates which :ref:`calling
5141 convention <callingconv>` the call should use. If none is
5142 specified, the call defaults to using C calling conventions.
5143 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5144 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5146 #. '``ptr to function ty``': shall be the signature of the pointer to
5147 function value being invoked. In most cases, this is a direct
5148 function invocation, but indirect ``invoke``'s are just as possible,
5149 branching off an arbitrary pointer to function value.
5150 #. '``function ptr val``': An LLVM value containing a pointer to a
5151 function to be invoked.
5152 #. '``function args``': argument list whose types match the function
5153 signature argument types and parameter attributes. All arguments must
5154 be of :ref:`first class <t_firstclass>` type. If the function signature
5155 indicates the function accepts a variable number of arguments, the
5156 extra arguments can be specified.
5157 #. '``normal label``': the label reached when the called function
5158 executes a '``ret``' instruction.
5159 #. '``exception label``': the label reached when a callee returns via
5160 the :ref:`resume <i_resume>` instruction or other exception handling
5162 #. The optional :ref:`function attributes <fnattrs>` list. Only
5163 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5164 attributes are valid here.
5165 #. The optional :ref:`operand bundles <opbundles>` list.
5170 This instruction is designed to operate as a standard '``call``'
5171 instruction in most regards. The primary difference is that it
5172 establishes an association with a label, which is used by the runtime
5173 library to unwind the stack.
5175 This instruction is used in languages with destructors to ensure that
5176 proper cleanup is performed in the case of either a ``longjmp`` or a
5177 thrown exception. Additionally, this is important for implementation of
5178 '``catch``' clauses in high-level languages that support them.
5180 For the purposes of the SSA form, the definition of the value returned
5181 by the '``invoke``' instruction is deemed to occur on the edge from the
5182 current block to the "normal" label. If the callee unwinds then no
5183 return value is available.
5188 .. code-block:: llvm
5190 %retval = invoke i32 @Test(i32 15) to label %Continue
5191 unwind label %TestCleanup ; i32:retval set
5192 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
5193 unwind label %TestCleanup ; i32:retval set
5197 '``resume``' Instruction
5198 ^^^^^^^^^^^^^^^^^^^^^^^^
5205 resume <type> <value>
5210 The '``resume``' instruction is a terminator instruction that has no
5216 The '``resume``' instruction requires one argument, which must have the
5217 same type as the result of any '``landingpad``' instruction in the same
5223 The '``resume``' instruction resumes propagation of an existing
5224 (in-flight) exception whose unwinding was interrupted with a
5225 :ref:`landingpad <i_landingpad>` instruction.
5230 .. code-block:: llvm
5232 resume { i8*, i32 } %exn
5236 '``catchpad``' Instruction
5237 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5244 <resultval> = catchpad [<args>*]
5245 to label <normal label> unwind label <exception label>
5250 The '``catchpad``' instruction is used by `LLVM's exception handling
5251 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5252 is a catch block --- one where a personality routine attempts to transfer
5253 control to catch an exception.
5254 The ``args`` correspond to whatever information the personality
5255 routine requires to know if this is an appropriate place to catch the
5256 exception. Control is transfered to the ``exception`` label if the
5257 ``catchpad`` is not an appropriate handler for the in-flight exception.
5258 The ``normal`` label should contain the code found in the ``catch``
5259 portion of a ``try``/``catch`` sequence. The ``resultval`` has the type
5260 :ref:`token <t_token>` and is used to match the ``catchpad`` to
5261 corresponding :ref:`catchrets <i_catchret>`.
5266 The instruction takes a list of arbitrary values which are interpreted
5267 by the :ref:`personality function <personalityfn>`.
5269 The ``catchpad`` must be provided a ``normal`` label to transfer control
5270 to if the ``catchpad`` matches the exception and an ``exception``
5271 label to transfer control to if it doesn't.
5276 When the call stack is being unwound due to an exception being thrown,
5277 the exception is compared against the ``args``. If it doesn't match,
5278 then control is transfered to the ``exception`` basic block.
5279 As with calling conventions, how the personality function results are
5280 represented in LLVM IR is target specific.
5282 The ``catchpad`` instruction has several restrictions:
5284 - A catch block is a basic block which is the unwind destination of
5285 an exceptional instruction.
5286 - A catch block must have a '``catchpad``' instruction as its
5287 first non-PHI instruction.
5288 - A catch block's ``exception`` edge must refer to a catch block or a
5290 - There can be only one '``catchpad``' instruction within the
5292 - A basic block that is not a catch block may not include a
5293 '``catchpad``' instruction.
5294 - A catch block which has another catch block as a predecessor may not have
5295 any other predecessors.
5296 - It is undefined behavior for control to transfer from a ``catchpad`` to a
5297 ``ret`` without first executing a ``catchret`` that consumes the
5298 ``catchpad`` or unwinding through its ``catchendpad``.
5299 - It is undefined behavior for control to transfer from a ``catchpad`` to
5300 itself without first executing a ``catchret`` that consumes the
5301 ``catchpad`` or unwinding through its ``catchendpad``.
5306 .. code-block:: llvm
5308 ;; A catch block which can catch an integer.
5309 %tok = catchpad [i8** @_ZTIi]
5310 to label %int.handler unwind label %terminate
5314 '``catchendpad``' Instruction
5315 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5322 catchendpad unwind label <nextaction>
5323 catchendpad unwind to caller
5328 The '``catchendpad``' instruction is used by `LLVM's exception handling
5329 system <ExceptionHandling.html#overview>`_ to communicate to the
5330 :ref:`personality function <personalityfn>` which invokes are associated
5331 with a chain of :ref:`catchpad <i_catchpad>` instructions; propagating an
5332 exception out of a catch handler is represented by unwinding through its
5333 ``catchendpad``. Unwinding to the outer scope when a chain of catch handlers
5334 do not handle an exception is also represented by unwinding through their
5337 The ``nextaction`` label indicates where control should transfer to if
5338 none of the ``catchpad`` instructions are suitable for catching the
5339 in-flight exception.
5341 If a ``nextaction`` label is not present, the instruction unwinds out of
5342 its parent function. The
5343 :ref:`personality function <personalityfn>` will continue processing
5344 exception handling actions in the caller.
5349 The instruction optionally takes a label, ``nextaction``, indicating
5350 where control should transfer to if none of the preceding
5351 ``catchpad`` instructions are suitable for the in-flight exception.
5356 When the call stack is being unwound due to an exception being thrown
5357 and none of the constituent ``catchpad`` instructions match, then
5358 control is transfered to ``nextaction`` if it is present. If it is not
5359 present, control is transfered to the caller.
5361 The ``catchendpad`` instruction has several restrictions:
5363 - A catch-end block is a basic block which is the unwind destination of
5364 an exceptional instruction.
5365 - A catch-end block must have a '``catchendpad``' instruction as its
5366 first non-PHI instruction.
5367 - There can be only one '``catchendpad``' instruction within the
5369 - A basic block that is not a catch-end block may not include a
5370 '``catchendpad``' instruction.
5371 - Exactly one catch block may unwind to a ``catchendpad``.
5372 - It is undefined behavior to execute a ``catchendpad`` if none of the
5373 '``catchpad``'s chained to it have been executed.
5374 - It is undefined behavior to execute a ``catchendpad`` twice without an
5375 intervening execution of one or more of the '``catchpad``'s chained to it.
5376 - It is undefined behavior to execute a ``catchendpad`` if, after the most
5377 recent execution of the normal successor edge of any ``catchpad`` chained
5378 to it, some ``catchret`` consuming that ``catchpad`` has already been
5380 - It is undefined behavior to execute a ``catchendpad`` if, after the most
5381 recent execution of the normal successor edge of any ``catchpad`` chained
5382 to it, any other ``catchpad`` or ``cleanuppad`` has been executed but has
5383 not had a corresponding
5384 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5389 .. code-block:: llvm
5391 catchendpad unwind label %terminate
5392 catchendpad unwind to caller
5396 '``catchret``' Instruction
5397 ^^^^^^^^^^^^^^^^^^^^^^^^^^
5404 catchret <value> to label <normal>
5409 The '``catchret``' instruction is a terminator instruction that has a
5416 The first argument to a '``catchret``' indicates which ``catchpad`` it
5417 exits. It must be a :ref:`catchpad <i_catchpad>`.
5418 The second argument to a '``catchret``' specifies where control will
5424 The '``catchret``' instruction ends the existing (in-flight) exception
5425 whose unwinding was interrupted with a
5426 :ref:`catchpad <i_catchpad>` instruction.
5427 The :ref:`personality function <personalityfn>` gets a chance to execute
5428 arbitrary code to, for example, run a C++ destructor.
5429 Control then transfers to ``normal``.
5430 It may be passed an optional, personality specific, value.
5432 It is undefined behavior to execute a ``catchret`` whose ``catchpad`` has
5435 It is undefined behavior to execute a ``catchret`` if, after the most recent
5436 execution of its ``catchpad``, some ``catchret`` or ``catchendpad`` linked
5437 to the same ``catchpad`` has already been executed.
5439 It is undefined behavior to execute a ``catchret`` if, after the most recent
5440 execution of its ``catchpad``, any other ``catchpad`` or ``cleanuppad`` has
5441 been executed but has not had a corresponding
5442 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5447 .. code-block:: llvm
5449 catchret %catch label %continue
5451 .. _i_cleanupendpad:
5453 '``cleanupendpad``' Instruction
5454 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5461 cleanupendpad <value> unwind label <nextaction>
5462 cleanupendpad <value> unwind to caller
5467 The '``cleanupendpad``' instruction is used by `LLVM's exception handling
5468 system <ExceptionHandling.html#overview>`_ to communicate to the
5469 :ref:`personality function <personalityfn>` which invokes are associated
5470 with a :ref:`cleanuppad <i_cleanuppad>` instructions; propagating an exception
5471 out of a cleanup is represented by unwinding through its ``cleanupendpad``.
5473 The ``nextaction`` label indicates where control should unwind to next, in the
5474 event that a cleanup is exited by means of an(other) exception being raised.
5476 If a ``nextaction`` label is not present, the instruction unwinds out of
5477 its parent function. The
5478 :ref:`personality function <personalityfn>` will continue processing
5479 exception handling actions in the caller.
5484 The '``cleanupendpad``' instruction requires one argument, which indicates
5485 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5486 It also has an optional successor, ``nextaction``, indicating where control
5492 When and exception propagates to a ``cleanupendpad``, control is transfered to
5493 ``nextaction`` if it is present. If it is not present, control is transfered to
5496 The ``cleanupendpad`` instruction has several restrictions:
5498 - A cleanup-end block is a basic block which is the unwind destination of
5499 an exceptional instruction.
5500 - A cleanup-end block must have a '``cleanupendpad``' instruction as its
5501 first non-PHI instruction.
5502 - There can be only one '``cleanupendpad``' instruction within the
5504 - A basic block that is not a cleanup-end block may not include a
5505 '``cleanupendpad``' instruction.
5506 - It is undefined behavior to execute a ``cleanupendpad`` whose ``cleanuppad``
5507 has not been executed.
5508 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5509 recent execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5510 consuming the same ``cleanuppad`` has already been executed.
5511 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5512 recent execution of its ``cleanuppad``, any other ``cleanuppad`` or
5513 ``catchpad`` has been executed but has not had a corresponding
5514 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5519 .. code-block:: llvm
5521 cleanupendpad %cleanup unwind label %terminate
5522 cleanupendpad %cleanup unwind to caller
5526 '``cleanupret``' Instruction
5527 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5534 cleanupret <value> unwind label <continue>
5535 cleanupret <value> unwind to caller
5540 The '``cleanupret``' instruction is a terminator instruction that has
5541 an optional successor.
5547 The '``cleanupret``' instruction requires one argument, which indicates
5548 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5549 It also has an optional successor, ``continue``.
5554 The '``cleanupret``' instruction indicates to the
5555 :ref:`personality function <personalityfn>` that one
5556 :ref:`cleanuppad <i_cleanuppad>` it transferred control to has ended.
5557 It transfers control to ``continue`` or unwinds out of the function.
5559 It is undefined behavior to execute a ``cleanupret`` whose ``cleanuppad`` has
5562 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5563 execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5564 consuming the same ``cleanuppad`` has already been executed.
5566 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5567 execution of its ``cleanuppad``, any other ``cleanuppad`` or ``catchpad`` has
5568 been executed but has not had a corresponding
5569 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5574 .. code-block:: llvm
5576 cleanupret %cleanup unwind to caller
5577 cleanupret %cleanup unwind label %continue
5581 '``terminatepad``' Instruction
5582 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5589 terminatepad [<args>*] unwind label <exception label>
5590 terminatepad [<args>*] unwind to caller
5595 The '``terminatepad``' instruction is used by `LLVM's exception handling
5596 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5597 is a terminate block --- one where a personality routine may decide to
5598 terminate the program.
5599 The ``args`` correspond to whatever information the personality
5600 routine requires to know if this is an appropriate place to terminate the
5601 program. Control is transferred to the ``exception`` label if the
5602 personality routine decides not to terminate the program for the
5603 in-flight exception.
5608 The instruction takes a list of arbitrary values which are interpreted
5609 by the :ref:`personality function <personalityfn>`.
5611 The ``terminatepad`` may be given an ``exception`` label to
5612 transfer control to if the in-flight exception matches the ``args``.
5617 When the call stack is being unwound due to an exception being thrown,
5618 the exception is compared against the ``args``. If it matches,
5619 then control is transfered to the ``exception`` basic block. Otherwise,
5620 the program is terminated via personality-specific means. Typically,
5621 the first argument to ``terminatepad`` specifies what function the
5622 personality should defer to in order to terminate the program.
5624 The ``terminatepad`` instruction has several restrictions:
5626 - A terminate block is a basic block which is the unwind destination of
5627 an exceptional instruction.
5628 - A terminate block must have a '``terminatepad``' instruction as its
5629 first non-PHI instruction.
5630 - There can be only one '``terminatepad``' instruction within the
5632 - A basic block that is not a terminate block may not include a
5633 '``terminatepad``' instruction.
5638 .. code-block:: llvm
5640 ;; A terminate block which only permits integers.
5641 terminatepad [i8** @_ZTIi] unwind label %continue
5645 '``unreachable``' Instruction
5646 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5658 The '``unreachable``' instruction has no defined semantics. This
5659 instruction is used to inform the optimizer that a particular portion of
5660 the code is not reachable. This can be used to indicate that the code
5661 after a no-return function cannot be reached, and other facts.
5666 The '``unreachable``' instruction has no defined semantics.
5673 Binary operators are used to do most of the computation in a program.
5674 They require two operands of the same type, execute an operation on
5675 them, and produce a single value. The operands might represent multiple
5676 data, as is the case with the :ref:`vector <t_vector>` data type. The
5677 result value has the same type as its operands.
5679 There are several different binary operators:
5683 '``add``' Instruction
5684 ^^^^^^^^^^^^^^^^^^^^^
5691 <result> = add <ty> <op1>, <op2> ; yields ty:result
5692 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
5693 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
5694 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
5699 The '``add``' instruction returns the sum of its two operands.
5704 The two arguments to the '``add``' instruction must be
5705 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5706 arguments must have identical types.
5711 The value produced is the integer sum of the two operands.
5713 If the sum has unsigned overflow, the result returned is the
5714 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5717 Because LLVM integers use a two's complement representation, this
5718 instruction is appropriate for both signed and unsigned integers.
5720 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5721 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5722 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
5723 unsigned and/or signed overflow, respectively, occurs.
5728 .. code-block:: llvm
5730 <result> = add i32 4, %var ; yields i32:result = 4 + %var
5734 '``fadd``' Instruction
5735 ^^^^^^^^^^^^^^^^^^^^^^
5742 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5747 The '``fadd``' instruction returns the sum of its two operands.
5752 The two arguments to the '``fadd``' instruction must be :ref:`floating
5753 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5754 Both arguments must have identical types.
5759 The value produced is the floating point sum of the two operands. This
5760 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
5761 which are optimization hints to enable otherwise unsafe floating point
5767 .. code-block:: llvm
5769 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
5771 '``sub``' Instruction
5772 ^^^^^^^^^^^^^^^^^^^^^
5779 <result> = sub <ty> <op1>, <op2> ; yields ty:result
5780 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
5781 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
5782 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
5787 The '``sub``' instruction returns the difference of its two operands.
5789 Note that the '``sub``' instruction is used to represent the '``neg``'
5790 instruction present in most other intermediate representations.
5795 The two arguments to the '``sub``' instruction must be
5796 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5797 arguments must have identical types.
5802 The value produced is the integer difference of the two operands.
5804 If the difference has unsigned overflow, the result returned is the
5805 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5808 Because LLVM integers use a two's complement representation, this
5809 instruction is appropriate for both signed and unsigned integers.
5811 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5812 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5813 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
5814 unsigned and/or signed overflow, respectively, occurs.
5819 .. code-block:: llvm
5821 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
5822 <result> = sub i32 0, %val ; yields i32:result = -%var
5826 '``fsub``' Instruction
5827 ^^^^^^^^^^^^^^^^^^^^^^
5834 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5839 The '``fsub``' instruction returns the difference of its two operands.
5841 Note that the '``fsub``' instruction is used to represent the '``fneg``'
5842 instruction present in most other intermediate representations.
5847 The two arguments to the '``fsub``' instruction must be :ref:`floating
5848 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5849 Both arguments must have identical types.
5854 The value produced is the floating point difference of the two operands.
5855 This instruction can also take any number of :ref:`fast-math
5856 flags <fastmath>`, which are optimization hints to enable otherwise
5857 unsafe floating point optimizations:
5862 .. code-block:: llvm
5864 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
5865 <result> = fsub float -0.0, %val ; yields float:result = -%var
5867 '``mul``' Instruction
5868 ^^^^^^^^^^^^^^^^^^^^^
5875 <result> = mul <ty> <op1>, <op2> ; yields ty:result
5876 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
5877 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
5878 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
5883 The '``mul``' instruction returns the product of its two operands.
5888 The two arguments to the '``mul``' instruction must be
5889 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5890 arguments must have identical types.
5895 The value produced is the integer product of the two operands.
5897 If the result of the multiplication has unsigned overflow, the result
5898 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
5899 bit width of the result.
5901 Because LLVM integers use a two's complement representation, and the
5902 result is the same width as the operands, this instruction returns the
5903 correct result for both signed and unsigned integers. If a full product
5904 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
5905 sign-extended or zero-extended as appropriate to the width of the full
5908 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5909 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5910 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
5911 unsigned and/or signed overflow, respectively, occurs.
5916 .. code-block:: llvm
5918 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
5922 '``fmul``' Instruction
5923 ^^^^^^^^^^^^^^^^^^^^^^
5930 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5935 The '``fmul``' instruction returns the product of its two operands.
5940 The two arguments to the '``fmul``' instruction must be :ref:`floating
5941 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5942 Both arguments must have identical types.
5947 The value produced is the floating point product of the two operands.
5948 This instruction can also take any number of :ref:`fast-math
5949 flags <fastmath>`, which are optimization hints to enable otherwise
5950 unsafe floating point optimizations:
5955 .. code-block:: llvm
5957 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
5959 '``udiv``' Instruction
5960 ^^^^^^^^^^^^^^^^^^^^^^
5967 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
5968 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
5973 The '``udiv``' instruction returns the quotient of its two operands.
5978 The two arguments to the '``udiv``' instruction must be
5979 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5980 arguments must have identical types.
5985 The value produced is the unsigned integer quotient of the two operands.
5987 Note that unsigned integer division and signed integer division are
5988 distinct operations; for signed integer division, use '``sdiv``'.
5990 Division by zero leads to undefined behavior.
5992 If the ``exact`` keyword is present, the result value of the ``udiv`` is
5993 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
5994 such, "((a udiv exact b) mul b) == a").
5999 .. code-block:: llvm
6001 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
6003 '``sdiv``' Instruction
6004 ^^^^^^^^^^^^^^^^^^^^^^
6011 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
6012 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
6017 The '``sdiv``' instruction returns the quotient of its two operands.
6022 The two arguments to the '``sdiv``' instruction must be
6023 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6024 arguments must have identical types.
6029 The value produced is the signed integer quotient of the two operands
6030 rounded towards zero.
6032 Note that signed integer division and unsigned integer division are
6033 distinct operations; for unsigned integer division, use '``udiv``'.
6035 Division by zero leads to undefined behavior. Overflow also leads to
6036 undefined behavior; this is a rare case, but can occur, for example, by
6037 doing a 32-bit division of -2147483648 by -1.
6039 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
6040 a :ref:`poison value <poisonvalues>` if the result would be rounded.
6045 .. code-block:: llvm
6047 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
6051 '``fdiv``' Instruction
6052 ^^^^^^^^^^^^^^^^^^^^^^
6059 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6064 The '``fdiv``' instruction returns the quotient of its two operands.
6069 The two arguments to the '``fdiv``' instruction must be :ref:`floating
6070 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6071 Both arguments must have identical types.
6076 The value produced is the floating point quotient of the two operands.
6077 This instruction can also take any number of :ref:`fast-math
6078 flags <fastmath>`, which are optimization hints to enable otherwise
6079 unsafe floating point optimizations:
6084 .. code-block:: llvm
6086 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
6088 '``urem``' Instruction
6089 ^^^^^^^^^^^^^^^^^^^^^^
6096 <result> = urem <ty> <op1>, <op2> ; yields ty:result
6101 The '``urem``' instruction returns the remainder from the unsigned
6102 division of its two arguments.
6107 The two arguments to the '``urem``' instruction must be
6108 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6109 arguments must have identical types.
6114 This instruction returns the unsigned integer *remainder* of a division.
6115 This instruction always performs an unsigned division to get the
6118 Note that unsigned integer remainder and signed integer remainder are
6119 distinct operations; for signed integer remainder, use '``srem``'.
6121 Taking the remainder of a division by zero leads to undefined behavior.
6126 .. code-block:: llvm
6128 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
6130 '``srem``' Instruction
6131 ^^^^^^^^^^^^^^^^^^^^^^
6138 <result> = srem <ty> <op1>, <op2> ; yields ty:result
6143 The '``srem``' instruction returns the remainder from the signed
6144 division of its two operands. This instruction can also take
6145 :ref:`vector <t_vector>` versions of the values in which case the elements
6151 The two arguments to the '``srem``' instruction must be
6152 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6153 arguments must have identical types.
6158 This instruction returns the *remainder* of a division (where the result
6159 is either zero or has the same sign as the dividend, ``op1``), not the
6160 *modulo* operator (where the result is either zero or has the same sign
6161 as the divisor, ``op2``) of a value. For more information about the
6162 difference, see `The Math
6163 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
6164 table of how this is implemented in various languages, please see
6166 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
6168 Note that signed integer remainder and unsigned integer remainder are
6169 distinct operations; for unsigned integer remainder, use '``urem``'.
6171 Taking the remainder of a division by zero leads to undefined behavior.
6172 Overflow also leads to undefined behavior; this is a rare case, but can
6173 occur, for example, by taking the remainder of a 32-bit division of
6174 -2147483648 by -1. (The remainder doesn't actually overflow, but this
6175 rule lets srem be implemented using instructions that return both the
6176 result of the division and the remainder.)
6181 .. code-block:: llvm
6183 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
6187 '``frem``' Instruction
6188 ^^^^^^^^^^^^^^^^^^^^^^
6195 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6200 The '``frem``' instruction returns the remainder from the division of
6206 The two arguments to the '``frem``' instruction must be :ref:`floating
6207 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6208 Both arguments must have identical types.
6213 This instruction returns the *remainder* of a division. The remainder
6214 has the same sign as the dividend. This instruction can also take any
6215 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
6216 to enable otherwise unsafe floating point optimizations:
6221 .. code-block:: llvm
6223 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
6227 Bitwise Binary Operations
6228 -------------------------
6230 Bitwise binary operators are used to do various forms of bit-twiddling
6231 in a program. They are generally very efficient instructions and can
6232 commonly be strength reduced from other instructions. They require two
6233 operands of the same type, execute an operation on them, and produce a
6234 single value. The resulting value is the same type as its operands.
6236 '``shl``' Instruction
6237 ^^^^^^^^^^^^^^^^^^^^^
6244 <result> = shl <ty> <op1>, <op2> ; yields ty:result
6245 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
6246 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
6247 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
6252 The '``shl``' instruction returns the first operand shifted to the left
6253 a specified number of bits.
6258 Both arguments to the '``shl``' instruction must be the same
6259 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6260 '``op2``' is treated as an unsigned value.
6265 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
6266 where ``n`` is the width of the result. If ``op2`` is (statically or
6267 dynamically) equal to or larger than the number of bits in
6268 ``op1``, the result is undefined. If the arguments are vectors, each
6269 vector element of ``op1`` is shifted by the corresponding shift amount
6272 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
6273 value <poisonvalues>` if it shifts out any non-zero bits. If the
6274 ``nsw`` keyword is present, then the shift produces a :ref:`poison
6275 value <poisonvalues>` if it shifts out any bits that disagree with the
6276 resultant sign bit. As such, NUW/NSW have the same semantics as they
6277 would if the shift were expressed as a mul instruction with the same
6278 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
6283 .. code-block:: llvm
6285 <result> = shl i32 4, %var ; yields i32: 4 << %var
6286 <result> = shl i32 4, 2 ; yields i32: 16
6287 <result> = shl i32 1, 10 ; yields i32: 1024
6288 <result> = shl i32 1, 32 ; undefined
6289 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
6291 '``lshr``' Instruction
6292 ^^^^^^^^^^^^^^^^^^^^^^
6299 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
6300 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
6305 The '``lshr``' instruction (logical shift right) returns the first
6306 operand shifted to the right a specified number of bits with zero fill.
6311 Both arguments to the '``lshr``' instruction must be the same
6312 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6313 '``op2``' is treated as an unsigned value.
6318 This instruction always performs a logical shift right operation. The
6319 most significant bits of the result will be filled with zero bits after
6320 the shift. If ``op2`` is (statically or dynamically) equal to or larger
6321 than the number of bits in ``op1``, the result is undefined. If the
6322 arguments are vectors, each vector element of ``op1`` is shifted by the
6323 corresponding shift amount in ``op2``.
6325 If the ``exact`` keyword is present, the result value of the ``lshr`` is
6326 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6332 .. code-block:: llvm
6334 <result> = lshr i32 4, 1 ; yields i32:result = 2
6335 <result> = lshr i32 4, 2 ; yields i32:result = 1
6336 <result> = lshr i8 4, 3 ; yields i8:result = 0
6337 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
6338 <result> = lshr i32 1, 32 ; undefined
6339 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
6341 '``ashr``' Instruction
6342 ^^^^^^^^^^^^^^^^^^^^^^
6349 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
6350 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
6355 The '``ashr``' instruction (arithmetic shift right) returns the first
6356 operand shifted to the right a specified number of bits with sign
6362 Both arguments to the '``ashr``' instruction must be the same
6363 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6364 '``op2``' is treated as an unsigned value.
6369 This instruction always performs an arithmetic shift right operation,
6370 The most significant bits of the result will be filled with the sign bit
6371 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
6372 than the number of bits in ``op1``, the result is undefined. If the
6373 arguments are vectors, each vector element of ``op1`` is shifted by the
6374 corresponding shift amount in ``op2``.
6376 If the ``exact`` keyword is present, the result value of the ``ashr`` is
6377 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6383 .. code-block:: llvm
6385 <result> = ashr i32 4, 1 ; yields i32:result = 2
6386 <result> = ashr i32 4, 2 ; yields i32:result = 1
6387 <result> = ashr i8 4, 3 ; yields i8:result = 0
6388 <result> = ashr i8 -2, 1 ; yields i8:result = -1
6389 <result> = ashr i32 1, 32 ; undefined
6390 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
6392 '``and``' Instruction
6393 ^^^^^^^^^^^^^^^^^^^^^
6400 <result> = and <ty> <op1>, <op2> ; yields ty:result
6405 The '``and``' instruction returns the bitwise logical and of its two
6411 The two arguments to the '``and``' instruction must be
6412 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6413 arguments must have identical types.
6418 The truth table used for the '``and``' instruction is:
6435 .. code-block:: llvm
6437 <result> = and i32 4, %var ; yields i32:result = 4 & %var
6438 <result> = and i32 15, 40 ; yields i32:result = 8
6439 <result> = and i32 4, 8 ; yields i32:result = 0
6441 '``or``' Instruction
6442 ^^^^^^^^^^^^^^^^^^^^
6449 <result> = or <ty> <op1>, <op2> ; yields ty:result
6454 The '``or``' instruction returns the bitwise logical inclusive or of its
6460 The two arguments to the '``or``' instruction must be
6461 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6462 arguments must have identical types.
6467 The truth table used for the '``or``' instruction is:
6486 <result> = or i32 4, %var ; yields i32:result = 4 | %var
6487 <result> = or i32 15, 40 ; yields i32:result = 47
6488 <result> = or i32 4, 8 ; yields i32:result = 12
6490 '``xor``' Instruction
6491 ^^^^^^^^^^^^^^^^^^^^^
6498 <result> = xor <ty> <op1>, <op2> ; yields ty:result
6503 The '``xor``' instruction returns the bitwise logical exclusive or of
6504 its two operands. The ``xor`` is used to implement the "one's
6505 complement" operation, which is the "~" operator in C.
6510 The two arguments to the '``xor``' instruction must be
6511 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6512 arguments must have identical types.
6517 The truth table used for the '``xor``' instruction is:
6534 .. code-block:: llvm
6536 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
6537 <result> = xor i32 15, 40 ; yields i32:result = 39
6538 <result> = xor i32 4, 8 ; yields i32:result = 12
6539 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
6544 LLVM supports several instructions to represent vector operations in a
6545 target-independent manner. These instructions cover the element-access
6546 and vector-specific operations needed to process vectors effectively.
6547 While LLVM does directly support these vector operations, many
6548 sophisticated algorithms will want to use target-specific intrinsics to
6549 take full advantage of a specific target.
6551 .. _i_extractelement:
6553 '``extractelement``' Instruction
6554 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6561 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
6566 The '``extractelement``' instruction extracts a single scalar element
6567 from a vector at a specified index.
6572 The first operand of an '``extractelement``' instruction is a value of
6573 :ref:`vector <t_vector>` type. The second operand is an index indicating
6574 the position from which to extract the element. The index may be a
6575 variable of any integer type.
6580 The result is a scalar of the same type as the element type of ``val``.
6581 Its value is the value at position ``idx`` of ``val``. If ``idx``
6582 exceeds the length of ``val``, the results are undefined.
6587 .. code-block:: llvm
6589 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
6591 .. _i_insertelement:
6593 '``insertelement``' Instruction
6594 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6601 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
6606 The '``insertelement``' instruction inserts a scalar element into a
6607 vector at a specified index.
6612 The first operand of an '``insertelement``' instruction is a value of
6613 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
6614 type must equal the element type of the first operand. The third operand
6615 is an index indicating the position at which to insert the value. The
6616 index may be a variable of any integer type.
6621 The result is a vector of the same type as ``val``. Its element values
6622 are those of ``val`` except at position ``idx``, where it gets the value
6623 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
6629 .. code-block:: llvm
6631 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
6633 .. _i_shufflevector:
6635 '``shufflevector``' Instruction
6636 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6643 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
6648 The '``shufflevector``' instruction constructs a permutation of elements
6649 from two input vectors, returning a vector with the same element type as
6650 the input and length that is the same as the shuffle mask.
6655 The first two operands of a '``shufflevector``' instruction are vectors
6656 with the same type. The third argument is a shuffle mask whose element
6657 type is always 'i32'. The result of the instruction is a vector whose
6658 length is the same as the shuffle mask and whose element type is the
6659 same as the element type of the first two operands.
6661 The shuffle mask operand is required to be a constant vector with either
6662 constant integer or undef values.
6667 The elements of the two input vectors are numbered from left to right
6668 across both of the vectors. The shuffle mask operand specifies, for each
6669 element of the result vector, which element of the two input vectors the
6670 result element gets. The element selector may be undef (meaning "don't
6671 care") and the second operand may be undef if performing a shuffle from
6677 .. code-block:: llvm
6679 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6680 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
6681 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
6682 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
6683 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
6684 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
6685 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6686 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
6688 Aggregate Operations
6689 --------------------
6691 LLVM supports several instructions for working with
6692 :ref:`aggregate <t_aggregate>` values.
6696 '``extractvalue``' Instruction
6697 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6704 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
6709 The '``extractvalue``' instruction extracts the value of a member field
6710 from an :ref:`aggregate <t_aggregate>` value.
6715 The first operand of an '``extractvalue``' instruction is a value of
6716 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The other operands are
6717 constant indices to specify which value to extract in a similar manner
6718 as indices in a '``getelementptr``' instruction.
6720 The major differences to ``getelementptr`` indexing are:
6722 - Since the value being indexed is not a pointer, the first index is
6723 omitted and assumed to be zero.
6724 - At least one index must be specified.
6725 - Not only struct indices but also array indices must be in bounds.
6730 The result is the value at the position in the aggregate specified by
6736 .. code-block:: llvm
6738 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
6742 '``insertvalue``' Instruction
6743 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6750 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
6755 The '``insertvalue``' instruction inserts a value into a member field in
6756 an :ref:`aggregate <t_aggregate>` value.
6761 The first operand of an '``insertvalue``' instruction is a value of
6762 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
6763 a first-class value to insert. The following operands are constant
6764 indices indicating the position at which to insert the value in a
6765 similar manner as indices in a '``extractvalue``' instruction. The value
6766 to insert must have the same type as the value identified by the
6772 The result is an aggregate of the same type as ``val``. Its value is
6773 that of ``val`` except that the value at the position specified by the
6774 indices is that of ``elt``.
6779 .. code-block:: llvm
6781 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
6782 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
6783 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
6787 Memory Access and Addressing Operations
6788 ---------------------------------------
6790 A key design point of an SSA-based representation is how it represents
6791 memory. In LLVM, no memory locations are in SSA form, which makes things
6792 very simple. This section describes how to read, write, and allocate
6797 '``alloca``' Instruction
6798 ^^^^^^^^^^^^^^^^^^^^^^^^
6805 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
6810 The '``alloca``' instruction allocates memory on the stack frame of the
6811 currently executing function, to be automatically released when this
6812 function returns to its caller. The object is always allocated in the
6813 generic address space (address space zero).
6818 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
6819 bytes of memory on the runtime stack, returning a pointer of the
6820 appropriate type to the program. If "NumElements" is specified, it is
6821 the number of elements allocated, otherwise "NumElements" is defaulted
6822 to be one. If a constant alignment is specified, the value result of the
6823 allocation is guaranteed to be aligned to at least that boundary. The
6824 alignment may not be greater than ``1 << 29``. If not specified, or if
6825 zero, the target can choose to align the allocation on any convenient
6826 boundary compatible with the type.
6828 '``type``' may be any sized type.
6833 Memory is allocated; a pointer is returned. The operation is undefined
6834 if there is insufficient stack space for the allocation. '``alloca``'d
6835 memory is automatically released when the function returns. The
6836 '``alloca``' instruction is commonly used to represent automatic
6837 variables that must have an address available. When the function returns
6838 (either with the ``ret`` or ``resume`` instructions), the memory is
6839 reclaimed. Allocating zero bytes is legal, but the result is undefined.
6840 The order in which memory is allocated (ie., which way the stack grows)
6846 .. code-block:: llvm
6848 %ptr = alloca i32 ; yields i32*:ptr
6849 %ptr = alloca i32, i32 4 ; yields i32*:ptr
6850 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
6851 %ptr = alloca i32, align 1024 ; yields i32*:ptr
6855 '``load``' Instruction
6856 ^^^^^^^^^^^^^^^^^^^^^^
6863 <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>]
6864 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>]
6865 !<index> = !{ i32 1 }
6866 !<deref_bytes_node> = !{i64 <dereferenceable_bytes>}
6867 !<align_node> = !{ i64 <value_alignment> }
6872 The '``load``' instruction is used to read from memory.
6877 The argument to the ``load`` instruction specifies the memory address
6878 from which to load. The type specified must be a :ref:`first
6879 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
6880 then the optimizer is not allowed to modify the number or order of
6881 execution of this ``load`` with other :ref:`volatile
6882 operations <volatile>`.
6884 If the ``load`` is marked as ``atomic``, it takes an extra
6885 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6886 ``release`` and ``acq_rel`` orderings are not valid on ``load``
6887 instructions. Atomic loads produce :ref:`defined <memmodel>` results
6888 when they may see multiple atomic stores. The type of the pointee must
6889 be an integer type whose bit width is a power of two greater than or
6890 equal to eight and less than or equal to a target-specific size limit.
6891 ``align`` must be explicitly specified on atomic loads, and the load has
6892 undefined behavior if the alignment is not set to a value which is at
6893 least the size in bytes of the pointee. ``!nontemporal`` does not have
6894 any defined semantics for atomic loads.
6896 The optional constant ``align`` argument specifies the alignment of the
6897 operation (that is, the alignment of the memory address). A value of 0
6898 or an omitted ``align`` argument means that the operation has the ABI
6899 alignment for the target. It is the responsibility of the code emitter
6900 to ensure that the alignment information is correct. Overestimating the
6901 alignment results in undefined behavior. Underestimating the alignment
6902 may produce less efficient code. An alignment of 1 is always safe. The
6903 maximum possible alignment is ``1 << 29``.
6905 The optional ``!nontemporal`` metadata must reference a single
6906 metadata name ``<index>`` corresponding to a metadata node with one
6907 ``i32`` entry of value 1. The existence of the ``!nontemporal``
6908 metadata on the instruction tells the optimizer and code generator
6909 that this load is not expected to be reused in the cache. The code
6910 generator may select special instructions to save cache bandwidth, such
6911 as the ``MOVNT`` instruction on x86.
6913 The optional ``!invariant.load`` metadata must reference a single
6914 metadata name ``<index>`` corresponding to a metadata node with no
6915 entries. The existence of the ``!invariant.load`` metadata on the
6916 instruction tells the optimizer and code generator that the address
6917 operand to this load points to memory which can be assumed unchanged.
6918 Being invariant does not imply that a location is dereferenceable,
6919 but it does imply that once the location is known dereferenceable
6920 its value is henceforth unchanging.
6922 The optional ``!invariant.group`` metadata must reference a single metadata name
6923 ``<index>`` corresponding to a metadata node. See ``invariant.group`` metadata.
6925 The optional ``!nonnull`` metadata must reference a single
6926 metadata name ``<index>`` corresponding to a metadata node with no
6927 entries. The existence of the ``!nonnull`` metadata on the
6928 instruction tells the optimizer that the value loaded is known to
6929 never be null. This is analogous to the ``nonnull`` attribute
6930 on parameters and return values. This metadata can only be applied
6931 to loads of a pointer type.
6933 The optional ``!dereferenceable`` metadata must reference a single metadata
6934 name ``<deref_bytes_node>`` corresponding to a metadata node with one ``i64``
6935 entry. The existence of the ``!dereferenceable`` metadata on the instruction
6936 tells the optimizer that the value loaded is known to be dereferenceable.
6937 The number of bytes known to be dereferenceable is specified by the integer
6938 value in the metadata node. This is analogous to the ''dereferenceable''
6939 attribute on parameters and return values. This metadata can only be applied
6940 to loads of a pointer type.
6942 The optional ``!dereferenceable_or_null`` metadata must reference a single
6943 metadata name ``<deref_bytes_node>`` corresponding to a metadata node with one
6944 ``i64`` entry. The existence of the ``!dereferenceable_or_null`` metadata on the
6945 instruction tells the optimizer that the value loaded is known to be either
6946 dereferenceable or null.
6947 The number of bytes known to be dereferenceable is specified by the integer
6948 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
6949 attribute on parameters and return values. This metadata can only be applied
6950 to loads of a pointer type.
6952 The optional ``!align`` metadata must reference a single metadata name
6953 ``<align_node>`` corresponding to a metadata node with one ``i64`` entry.
6954 The existence of the ``!align`` metadata on the instruction tells the
6955 optimizer that the value loaded is known to be aligned to a boundary specified
6956 by the integer value in the metadata node. The alignment must be a power of 2.
6957 This is analogous to the ''align'' attribute on parameters and return values.
6958 This metadata can only be applied to loads of a pointer type.
6963 The location of memory pointed to is loaded. If the value being loaded
6964 is of scalar type then the number of bytes read does not exceed the
6965 minimum number of bytes needed to hold all bits of the type. For
6966 example, loading an ``i24`` reads at most three bytes. When loading a
6967 value of a type like ``i20`` with a size that is not an integral number
6968 of bytes, the result is undefined if the value was not originally
6969 written using a store of the same type.
6974 .. code-block:: llvm
6976 %ptr = alloca i32 ; yields i32*:ptr
6977 store i32 3, i32* %ptr ; yields void
6978 %val = load i32, i32* %ptr ; yields i32:val = i32 3
6982 '``store``' Instruction
6983 ^^^^^^^^^^^^^^^^^^^^^^^
6990 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.group !<index>] ; yields void
6991 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>] ; yields void
6996 The '``store``' instruction is used to write to memory.
7001 There are two arguments to the ``store`` instruction: a value to store
7002 and an address at which to store it. The type of the ``<pointer>``
7003 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
7004 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
7005 then the optimizer is not allowed to modify the number or order of
7006 execution of this ``store`` with other :ref:`volatile
7007 operations <volatile>`.
7009 If the ``store`` is marked as ``atomic``, it takes an extra
7010 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
7011 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
7012 instructions. Atomic loads produce :ref:`defined <memmodel>` results
7013 when they may see multiple atomic stores. The type of the pointee must
7014 be an integer type whose bit width is a power of two greater than or
7015 equal to eight and less than or equal to a target-specific size limit.
7016 ``align`` must be explicitly specified on atomic stores, and the store
7017 has undefined behavior if the alignment is not set to a value which is
7018 at least the size in bytes of the pointee. ``!nontemporal`` does not
7019 have any defined semantics for atomic stores.
7021 The optional constant ``align`` argument specifies the alignment of the
7022 operation (that is, the alignment of the memory address). A value of 0
7023 or an omitted ``align`` argument means that the operation has the ABI
7024 alignment for the target. It is the responsibility of the code emitter
7025 to ensure that the alignment information is correct. Overestimating the
7026 alignment results in undefined behavior. Underestimating the
7027 alignment may produce less efficient code. An alignment of 1 is always
7028 safe. The maximum possible alignment is ``1 << 29``.
7030 The optional ``!nontemporal`` metadata must reference a single metadata
7031 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
7032 value 1. The existence of the ``!nontemporal`` metadata on the instruction
7033 tells the optimizer and code generator that this load is not expected to
7034 be reused in the cache. The code generator may select special
7035 instructions to save cache bandwidth, such as the MOVNT instruction on
7038 The optional ``!invariant.group`` metadata must reference a
7039 single metadata name ``<index>``. See ``invariant.group`` metadata.
7044 The contents of memory are updated to contain ``<value>`` at the
7045 location specified by the ``<pointer>`` operand. If ``<value>`` is
7046 of scalar type then the number of bytes written does not exceed the
7047 minimum number of bytes needed to hold all bits of the type. For
7048 example, storing an ``i24`` writes at most three bytes. When writing a
7049 value of a type like ``i20`` with a size that is not an integral number
7050 of bytes, it is unspecified what happens to the extra bits that do not
7051 belong to the type, but they will typically be overwritten.
7056 .. code-block:: llvm
7058 %ptr = alloca i32 ; yields i32*:ptr
7059 store i32 3, i32* %ptr ; yields void
7060 %val = load i32, i32* %ptr ; yields i32:val = i32 3
7064 '``fence``' Instruction
7065 ^^^^^^^^^^^^^^^^^^^^^^^
7072 fence [singlethread] <ordering> ; yields void
7077 The '``fence``' instruction is used to introduce happens-before edges
7083 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
7084 defines what *synchronizes-with* edges they add. They can only be given
7085 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
7090 A fence A which has (at least) ``release`` ordering semantics
7091 *synchronizes with* a fence B with (at least) ``acquire`` ordering
7092 semantics if and only if there exist atomic operations X and Y, both
7093 operating on some atomic object M, such that A is sequenced before X, X
7094 modifies M (either directly or through some side effect of a sequence
7095 headed by X), Y is sequenced before B, and Y observes M. This provides a
7096 *happens-before* dependency between A and B. Rather than an explicit
7097 ``fence``, one (but not both) of the atomic operations X or Y might
7098 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
7099 still *synchronize-with* the explicit ``fence`` and establish the
7100 *happens-before* edge.
7102 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
7103 ``acquire`` and ``release`` semantics specified above, participates in
7104 the global program order of other ``seq_cst`` operations and/or fences.
7106 The optional ":ref:`singlethread <singlethread>`" argument specifies
7107 that the fence only synchronizes with other fences in the same thread.
7108 (This is useful for interacting with signal handlers.)
7113 .. code-block:: llvm
7115 fence acquire ; yields void
7116 fence singlethread seq_cst ; yields void
7120 '``cmpxchg``' Instruction
7121 ^^^^^^^^^^^^^^^^^^^^^^^^^
7128 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
7133 The '``cmpxchg``' instruction is used to atomically modify memory. It
7134 loads a value in memory and compares it to a given value. If they are
7135 equal, it tries to store a new value into the memory.
7140 There are three arguments to the '``cmpxchg``' instruction: an address
7141 to operate on, a value to compare to the value currently be at that
7142 address, and a new value to place at that address if the compared values
7143 are equal. The type of '<cmp>' must be an integer type whose bit width
7144 is a power of two greater than or equal to eight and less than or equal
7145 to a target-specific size limit. '<cmp>' and '<new>' must have the same
7146 type, and the type of '<pointer>' must be a pointer to that type. If the
7147 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
7148 to modify the number or order of execution of this ``cmpxchg`` with
7149 other :ref:`volatile operations <volatile>`.
7151 The success and failure :ref:`ordering <ordering>` arguments specify how this
7152 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
7153 must be at least ``monotonic``, the ordering constraint on failure must be no
7154 stronger than that on success, and the failure ordering cannot be either
7155 ``release`` or ``acq_rel``.
7157 The optional "``singlethread``" argument declares that the ``cmpxchg``
7158 is only atomic with respect to code (usually signal handlers) running in
7159 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
7160 respect to all other code in the system.
7162 The pointer passed into cmpxchg must have alignment greater than or
7163 equal to the size in memory of the operand.
7168 The contents of memory at the location specified by the '``<pointer>``' operand
7169 is read and compared to '``<cmp>``'; if the read value is the equal, the
7170 '``<new>``' is written. The original value at the location is returned, together
7171 with a flag indicating success (true) or failure (false).
7173 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
7174 permitted: the operation may not write ``<new>`` even if the comparison
7177 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
7178 if the value loaded equals ``cmp``.
7180 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
7181 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
7182 load with an ordering parameter determined the second ordering parameter.
7187 .. code-block:: llvm
7190 %orig = atomic load i32, i32* %ptr unordered ; yields i32
7194 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
7195 %squared = mul i32 %cmp, %cmp
7196 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
7197 %value_loaded = extractvalue { i32, i1 } %val_success, 0
7198 %success = extractvalue { i32, i1 } %val_success, 1
7199 br i1 %success, label %done, label %loop
7206 '``atomicrmw``' Instruction
7207 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7214 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
7219 The '``atomicrmw``' instruction is used to atomically modify memory.
7224 There are three arguments to the '``atomicrmw``' instruction: an
7225 operation to apply, an address whose value to modify, an argument to the
7226 operation. The operation must be one of the following keywords:
7240 The type of '<value>' must be an integer type whose bit width is a power
7241 of two greater than or equal to eight and less than or equal to a
7242 target-specific size limit. The type of the '``<pointer>``' operand must
7243 be a pointer to that type. If the ``atomicrmw`` is marked as
7244 ``volatile``, then the optimizer is not allowed to modify the number or
7245 order of execution of this ``atomicrmw`` with other :ref:`volatile
7246 operations <volatile>`.
7251 The contents of memory at the location specified by the '``<pointer>``'
7252 operand are atomically read, modified, and written back. The original
7253 value at the location is returned. The modification is specified by the
7256 - xchg: ``*ptr = val``
7257 - add: ``*ptr = *ptr + val``
7258 - sub: ``*ptr = *ptr - val``
7259 - and: ``*ptr = *ptr & val``
7260 - nand: ``*ptr = ~(*ptr & val)``
7261 - or: ``*ptr = *ptr | val``
7262 - xor: ``*ptr = *ptr ^ val``
7263 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
7264 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
7265 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
7267 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
7273 .. code-block:: llvm
7275 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
7277 .. _i_getelementptr:
7279 '``getelementptr``' Instruction
7280 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7287 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7288 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7289 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
7294 The '``getelementptr``' instruction is used to get the address of a
7295 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
7296 address calculation only and does not access memory. The instruction can also
7297 be used to calculate a vector of such addresses.
7302 The first argument is always a type used as the basis for the calculations.
7303 The second argument is always a pointer or a vector of pointers, and is the
7304 base address to start from. The remaining arguments are indices
7305 that indicate which of the elements of the aggregate object are indexed.
7306 The interpretation of each index is dependent on the type being indexed
7307 into. The first index always indexes the pointer value given as the
7308 first argument, the second index indexes a value of the type pointed to
7309 (not necessarily the value directly pointed to, since the first index
7310 can be non-zero), etc. The first type indexed into must be a pointer
7311 value, subsequent types can be arrays, vectors, and structs. Note that
7312 subsequent types being indexed into can never be pointers, since that
7313 would require loading the pointer before continuing calculation.
7315 The type of each index argument depends on the type it is indexing into.
7316 When indexing into a (optionally packed) structure, only ``i32`` integer
7317 **constants** are allowed (when using a vector of indices they must all
7318 be the **same** ``i32`` integer constant). When indexing into an array,
7319 pointer or vector, integers of any width are allowed, and they are not
7320 required to be constant. These integers are treated as signed values
7323 For example, let's consider a C code fragment and how it gets compiled
7339 int *foo(struct ST *s) {
7340 return &s[1].Z.B[5][13];
7343 The LLVM code generated by Clang is:
7345 .. code-block:: llvm
7347 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
7348 %struct.ST = type { i32, double, %struct.RT }
7350 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
7352 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
7359 In the example above, the first index is indexing into the
7360 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
7361 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
7362 indexes into the third element of the structure, yielding a
7363 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
7364 structure. The third index indexes into the second element of the
7365 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
7366 dimensions of the array are subscripted into, yielding an '``i32``'
7367 type. The '``getelementptr``' instruction returns a pointer to this
7368 element, thus computing a value of '``i32*``' type.
7370 Note that it is perfectly legal to index partially through a structure,
7371 returning a pointer to an inner element. Because of this, the LLVM code
7372 for the given testcase is equivalent to:
7374 .. code-block:: llvm
7376 define i32* @foo(%struct.ST* %s) {
7377 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
7378 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
7379 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
7380 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
7381 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
7385 If the ``inbounds`` keyword is present, the result value of the
7386 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
7387 pointer is not an *in bounds* address of an allocated object, or if any
7388 of the addresses that would be formed by successive addition of the
7389 offsets implied by the indices to the base address with infinitely
7390 precise signed arithmetic are not an *in bounds* address of that
7391 allocated object. The *in bounds* addresses for an allocated object are
7392 all the addresses that point into the object, plus the address one byte
7393 past the end. In cases where the base is a vector of pointers the
7394 ``inbounds`` keyword applies to each of the computations element-wise.
7396 If the ``inbounds`` keyword is not present, the offsets are added to the
7397 base address with silently-wrapping two's complement arithmetic. If the
7398 offsets have a different width from the pointer, they are sign-extended
7399 or truncated to the width of the pointer. The result value of the
7400 ``getelementptr`` may be outside the object pointed to by the base
7401 pointer. The result value may not necessarily be used to access memory
7402 though, even if it happens to point into allocated storage. See the
7403 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
7406 The getelementptr instruction is often confusing. For some more insight
7407 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
7412 .. code-block:: llvm
7414 ; yields [12 x i8]*:aptr
7415 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
7417 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
7419 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
7421 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
7426 The ``getelementptr`` returns a vector of pointers, instead of a single address,
7427 when one or more of its arguments is a vector. In such cases, all vector
7428 arguments should have the same number of elements, and every scalar argument
7429 will be effectively broadcast into a vector during address calculation.
7431 .. code-block:: llvm
7433 ; All arguments are vectors:
7434 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
7435 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
7437 ; Add the same scalar offset to each pointer of a vector:
7438 ; A[i] = ptrs[i] + offset*sizeof(i8)
7439 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
7441 ; Add distinct offsets to the same pointer:
7442 ; A[i] = ptr + offsets[i]*sizeof(i8)
7443 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
7445 ; In all cases described above the type of the result is <4 x i8*>
7447 The two following instructions are equivalent:
7449 .. code-block:: llvm
7451 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7452 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
7453 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
7455 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
7457 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7458 i32 2, i32 1, <4 x i32> %ind4, i64 13
7460 Let's look at the C code, where the vector version of ``getelementptr``
7465 // Let's assume that we vectorize the following loop:
7466 double *A, B; int *C;
7467 for (int i = 0; i < size; ++i) {
7471 .. code-block:: llvm
7473 ; get pointers for 8 elements from array B
7474 %ptrs = getelementptr double, double* %B, <8 x i32> %C
7475 ; load 8 elements from array B into A
7476 %A = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs,
7477 i32 8, <8 x i1> %mask, <8 x double> %passthru)
7479 Conversion Operations
7480 ---------------------
7482 The instructions in this category are the conversion instructions
7483 (casting) which all take a single operand and a type. They perform
7484 various bit conversions on the operand.
7486 '``trunc .. to``' Instruction
7487 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7494 <result> = trunc <ty> <value> to <ty2> ; yields ty2
7499 The '``trunc``' instruction truncates its operand to the type ``ty2``.
7504 The '``trunc``' instruction takes a value to trunc, and a type to trunc
7505 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
7506 of the same number of integers. The bit size of the ``value`` must be
7507 larger than the bit size of the destination type, ``ty2``. Equal sized
7508 types are not allowed.
7513 The '``trunc``' instruction truncates the high order bits in ``value``
7514 and converts the remaining bits to ``ty2``. Since the source size must
7515 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
7516 It will always truncate bits.
7521 .. code-block:: llvm
7523 %X = trunc i32 257 to i8 ; yields i8:1
7524 %Y = trunc i32 123 to i1 ; yields i1:true
7525 %Z = trunc i32 122 to i1 ; yields i1:false
7526 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
7528 '``zext .. to``' Instruction
7529 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7536 <result> = zext <ty> <value> to <ty2> ; yields ty2
7541 The '``zext``' instruction zero extends its operand to type ``ty2``.
7546 The '``zext``' instruction takes a value to cast, and a type to cast it
7547 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7548 the same number of integers. The bit size of the ``value`` must be
7549 smaller than the bit size of the destination type, ``ty2``.
7554 The ``zext`` fills the high order bits of the ``value`` with zero bits
7555 until it reaches the size of the destination type, ``ty2``.
7557 When zero extending from i1, the result will always be either 0 or 1.
7562 .. code-block:: llvm
7564 %X = zext i32 257 to i64 ; yields i64:257
7565 %Y = zext i1 true to i32 ; yields i32:1
7566 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7568 '``sext .. to``' Instruction
7569 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7576 <result> = sext <ty> <value> to <ty2> ; yields ty2
7581 The '``sext``' sign extends ``value`` to the type ``ty2``.
7586 The '``sext``' instruction takes a value to cast, and a type to cast it
7587 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7588 the same number of integers. The bit size of the ``value`` must be
7589 smaller than the bit size of the destination type, ``ty2``.
7594 The '``sext``' instruction performs a sign extension by copying the sign
7595 bit (highest order bit) of the ``value`` until it reaches the bit size
7596 of the type ``ty2``.
7598 When sign extending from i1, the extension always results in -1 or 0.
7603 .. code-block:: llvm
7605 %X = sext i8 -1 to i16 ; yields i16 :65535
7606 %Y = sext i1 true to i32 ; yields i32:-1
7607 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7609 '``fptrunc .. to``' Instruction
7610 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7617 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
7622 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
7627 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
7628 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
7629 The size of ``value`` must be larger than the size of ``ty2``. This
7630 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
7635 The '``fptrunc``' instruction casts a ``value`` from a larger
7636 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
7637 point <t_floating>` type. If the value cannot fit (i.e. overflows) within the
7638 destination type, ``ty2``, then the results are undefined. If the cast produces
7639 an inexact result, how rounding is performed (e.g. truncation, also known as
7640 round to zero) is undefined.
7645 .. code-block:: llvm
7647 %X = fptrunc double 123.0 to float ; yields float:123.0
7648 %Y = fptrunc double 1.0E+300 to float ; yields undefined
7650 '``fpext .. to``' Instruction
7651 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7658 <result> = fpext <ty> <value> to <ty2> ; yields ty2
7663 The '``fpext``' extends a floating point ``value`` to a larger floating
7669 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
7670 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
7671 to. The source type must be smaller than the destination type.
7676 The '``fpext``' instruction extends the ``value`` from a smaller
7677 :ref:`floating point <t_floating>` type to a larger :ref:`floating
7678 point <t_floating>` type. The ``fpext`` cannot be used to make a
7679 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
7680 *no-op cast* for a floating point cast.
7685 .. code-block:: llvm
7687 %X = fpext float 3.125 to double ; yields double:3.125000e+00
7688 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
7690 '``fptoui .. to``' Instruction
7691 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7698 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
7703 The '``fptoui``' converts a floating point ``value`` to its unsigned
7704 integer equivalent of type ``ty2``.
7709 The '``fptoui``' instruction takes a value to cast, which must be a
7710 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7711 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7712 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7713 type with the same number of elements as ``ty``
7718 The '``fptoui``' instruction converts its :ref:`floating
7719 point <t_floating>` operand into the nearest (rounding towards zero)
7720 unsigned integer value. If the value cannot fit in ``ty2``, the results
7726 .. code-block:: llvm
7728 %X = fptoui double 123.0 to i32 ; yields i32:123
7729 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
7730 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
7732 '``fptosi .. to``' Instruction
7733 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7740 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
7745 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
7746 ``value`` to type ``ty2``.
7751 The '``fptosi``' instruction takes a value to cast, which must be a
7752 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7753 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7754 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7755 type with the same number of elements as ``ty``
7760 The '``fptosi``' instruction converts its :ref:`floating
7761 point <t_floating>` operand into the nearest (rounding towards zero)
7762 signed integer value. If the value cannot fit in ``ty2``, the results
7768 .. code-block:: llvm
7770 %X = fptosi double -123.0 to i32 ; yields i32:-123
7771 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
7772 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
7774 '``uitofp .. to``' Instruction
7775 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7782 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
7787 The '``uitofp``' instruction regards ``value`` as an unsigned integer
7788 and converts that value to the ``ty2`` type.
7793 The '``uitofp``' instruction takes a value to cast, which must be a
7794 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7795 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7796 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7797 type with the same number of elements as ``ty``
7802 The '``uitofp``' instruction interprets its operand as an unsigned
7803 integer quantity and converts it to the corresponding floating point
7804 value. If the value cannot fit in the floating point value, the results
7810 .. code-block:: llvm
7812 %X = uitofp i32 257 to float ; yields float:257.0
7813 %Y = uitofp i8 -1 to double ; yields double:255.0
7815 '``sitofp .. to``' Instruction
7816 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7823 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
7828 The '``sitofp``' instruction regards ``value`` as a signed integer and
7829 converts that value to the ``ty2`` type.
7834 The '``sitofp``' instruction takes a value to cast, which must be a
7835 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7836 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7837 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7838 type with the same number of elements as ``ty``
7843 The '``sitofp``' instruction interprets its operand as a signed integer
7844 quantity and converts it to the corresponding floating point value. If
7845 the value cannot fit in the floating point value, the results are
7851 .. code-block:: llvm
7853 %X = sitofp i32 257 to float ; yields float:257.0
7854 %Y = sitofp i8 -1 to double ; yields double:-1.0
7858 '``ptrtoint .. to``' Instruction
7859 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7866 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
7871 The '``ptrtoint``' instruction converts the pointer or a vector of
7872 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
7877 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
7878 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
7879 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
7880 a vector of integers type.
7885 The '``ptrtoint``' instruction converts ``value`` to integer type
7886 ``ty2`` by interpreting the pointer value as an integer and either
7887 truncating or zero extending that value to the size of the integer type.
7888 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
7889 ``value`` is larger than ``ty2`` then a truncation is done. If they are
7890 the same size, then nothing is done (*no-op cast*) other than a type
7896 .. code-block:: llvm
7898 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
7899 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
7900 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
7904 '``inttoptr .. to``' Instruction
7905 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7912 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
7917 The '``inttoptr``' instruction converts an integer ``value`` to a
7918 pointer type, ``ty2``.
7923 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
7924 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
7930 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
7931 applying either a zero extension or a truncation depending on the size
7932 of the integer ``value``. If ``value`` is larger than the size of a
7933 pointer then a truncation is done. If ``value`` is smaller than the size
7934 of a pointer then a zero extension is done. If they are the same size,
7935 nothing is done (*no-op cast*).
7940 .. code-block:: llvm
7942 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
7943 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
7944 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
7945 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
7949 '``bitcast .. to``' Instruction
7950 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7957 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
7962 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
7968 The '``bitcast``' instruction takes a value to cast, which must be a
7969 non-aggregate first class value, and a type to cast it to, which must
7970 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
7971 bit sizes of ``value`` and the destination type, ``ty2``, must be
7972 identical. If the source type is a pointer, the destination type must
7973 also be a pointer of the same size. This instruction supports bitwise
7974 conversion of vectors to integers and to vectors of other types (as
7975 long as they have the same size).
7980 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
7981 is always a *no-op cast* because no bits change with this
7982 conversion. The conversion is done as if the ``value`` had been stored
7983 to memory and read back as type ``ty2``. Pointer (or vector of
7984 pointers) types may only be converted to other pointer (or vector of
7985 pointers) types with the same address space through this instruction.
7986 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
7987 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
7992 .. code-block:: llvm
7994 %X = bitcast i8 255 to i8 ; yields i8 :-1
7995 %Y = bitcast i32* %x to sint* ; yields sint*:%x
7996 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
7997 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
7999 .. _i_addrspacecast:
8001 '``addrspacecast .. to``' Instruction
8002 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8009 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
8014 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
8015 address space ``n`` to type ``pty2`` in address space ``m``.
8020 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
8021 to cast and a pointer type to cast it to, which must have a different
8027 The '``addrspacecast``' instruction converts the pointer value
8028 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
8029 value modification, depending on the target and the address space
8030 pair. Pointer conversions within the same address space must be
8031 performed with the ``bitcast`` instruction. Note that if the address space
8032 conversion is legal then both result and operand refer to the same memory
8038 .. code-block:: llvm
8040 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
8041 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
8042 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
8049 The instructions in this category are the "miscellaneous" instructions,
8050 which defy better classification.
8054 '``icmp``' Instruction
8055 ^^^^^^^^^^^^^^^^^^^^^^
8062 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8067 The '``icmp``' instruction returns a boolean value or a vector of
8068 boolean values based on comparison of its two integer, integer vector,
8069 pointer, or pointer vector operands.
8074 The '``icmp``' instruction takes three operands. The first operand is
8075 the condition code indicating the kind of comparison to perform. It is
8076 not a value, just a keyword. The possible condition code are:
8079 #. ``ne``: not equal
8080 #. ``ugt``: unsigned greater than
8081 #. ``uge``: unsigned greater or equal
8082 #. ``ult``: unsigned less than
8083 #. ``ule``: unsigned less or equal
8084 #. ``sgt``: signed greater than
8085 #. ``sge``: signed greater or equal
8086 #. ``slt``: signed less than
8087 #. ``sle``: signed less or equal
8089 The remaining two arguments must be :ref:`integer <t_integer>` or
8090 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
8091 must also be identical types.
8096 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
8097 code given as ``cond``. The comparison performed always yields either an
8098 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
8100 #. ``eq``: yields ``true`` if the operands are equal, ``false``
8101 otherwise. No sign interpretation is necessary or performed.
8102 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
8103 otherwise. No sign interpretation is necessary or performed.
8104 #. ``ugt``: interprets the operands as unsigned values and yields
8105 ``true`` if ``op1`` is greater than ``op2``.
8106 #. ``uge``: interprets the operands as unsigned values and yields
8107 ``true`` if ``op1`` is greater than or equal to ``op2``.
8108 #. ``ult``: interprets the operands as unsigned values and yields
8109 ``true`` if ``op1`` is less than ``op2``.
8110 #. ``ule``: interprets the operands as unsigned values and yields
8111 ``true`` if ``op1`` is less than or equal to ``op2``.
8112 #. ``sgt``: interprets the operands as signed values and yields ``true``
8113 if ``op1`` is greater than ``op2``.
8114 #. ``sge``: interprets the operands as signed values and yields ``true``
8115 if ``op1`` is greater than or equal to ``op2``.
8116 #. ``slt``: interprets the operands as signed values and yields ``true``
8117 if ``op1`` is less than ``op2``.
8118 #. ``sle``: interprets the operands as signed values and yields ``true``
8119 if ``op1`` is less than or equal to ``op2``.
8121 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
8122 are compared as if they were integers.
8124 If the operands are integer vectors, then they are compared element by
8125 element. The result is an ``i1`` vector with the same number of elements
8126 as the values being compared. Otherwise, the result is an ``i1``.
8131 .. code-block:: llvm
8133 <result> = icmp eq i32 4, 5 ; yields: result=false
8134 <result> = icmp ne float* %X, %X ; yields: result=false
8135 <result> = icmp ult i16 4, 5 ; yields: result=true
8136 <result> = icmp sgt i16 4, 5 ; yields: result=false
8137 <result> = icmp ule i16 -4, 5 ; yields: result=false
8138 <result> = icmp sge i16 4, 5 ; yields: result=false
8140 Note that the code generator does not yet support vector types with the
8141 ``icmp`` instruction.
8145 '``fcmp``' Instruction
8146 ^^^^^^^^^^^^^^^^^^^^^^
8153 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8158 The '``fcmp``' instruction returns a boolean value or vector of boolean
8159 values based on comparison of its operands.
8161 If the operands are floating point scalars, then the result type is a
8162 boolean (:ref:`i1 <t_integer>`).
8164 If the operands are floating point vectors, then the result type is a
8165 vector of boolean with the same number of elements as the operands being
8171 The '``fcmp``' instruction takes three operands. The first operand is
8172 the condition code indicating the kind of comparison to perform. It is
8173 not a value, just a keyword. The possible condition code are:
8175 #. ``false``: no comparison, always returns false
8176 #. ``oeq``: ordered and equal
8177 #. ``ogt``: ordered and greater than
8178 #. ``oge``: ordered and greater than or equal
8179 #. ``olt``: ordered and less than
8180 #. ``ole``: ordered and less than or equal
8181 #. ``one``: ordered and not equal
8182 #. ``ord``: ordered (no nans)
8183 #. ``ueq``: unordered or equal
8184 #. ``ugt``: unordered or greater than
8185 #. ``uge``: unordered or greater than or equal
8186 #. ``ult``: unordered or less than
8187 #. ``ule``: unordered or less than or equal
8188 #. ``une``: unordered or not equal
8189 #. ``uno``: unordered (either nans)
8190 #. ``true``: no comparison, always returns true
8192 *Ordered* means that neither operand is a QNAN while *unordered* means
8193 that either operand may be a QNAN.
8195 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
8196 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
8197 type. They must have identical types.
8202 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
8203 condition code given as ``cond``. If the operands are vectors, then the
8204 vectors are compared element by element. Each comparison performed
8205 always yields an :ref:`i1 <t_integer>` result, as follows:
8207 #. ``false``: always yields ``false``, regardless of operands.
8208 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
8209 is equal to ``op2``.
8210 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
8211 is greater than ``op2``.
8212 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
8213 is greater than or equal to ``op2``.
8214 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
8215 is less than ``op2``.
8216 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
8217 is less than or equal to ``op2``.
8218 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
8219 is not equal to ``op2``.
8220 #. ``ord``: yields ``true`` if both operands are not a QNAN.
8221 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
8223 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
8224 greater than ``op2``.
8225 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
8226 greater than or equal to ``op2``.
8227 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
8229 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
8230 less than or equal to ``op2``.
8231 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
8232 not equal to ``op2``.
8233 #. ``uno``: yields ``true`` if either operand is a QNAN.
8234 #. ``true``: always yields ``true``, regardless of operands.
8236 The ``fcmp`` instruction can also optionally take any number of
8237 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8238 otherwise unsafe floating point optimizations.
8240 Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
8241 only flags that have any effect on its semantics are those that allow
8242 assumptions to be made about the values of input arguments; namely
8243 ``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
8248 .. code-block:: llvm
8250 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
8251 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
8252 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
8253 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
8255 Note that the code generator does not yet support vector types with the
8256 ``fcmp`` instruction.
8260 '``phi``' Instruction
8261 ^^^^^^^^^^^^^^^^^^^^^
8268 <result> = phi <ty> [ <val0>, <label0>], ...
8273 The '``phi``' instruction is used to implement the φ node in the SSA
8274 graph representing the function.
8279 The type of the incoming values is specified with the first type field.
8280 After this, the '``phi``' instruction takes a list of pairs as
8281 arguments, with one pair for each predecessor basic block of the current
8282 block. Only values of :ref:`first class <t_firstclass>` type may be used as
8283 the value arguments to the PHI node. Only labels may be used as the
8286 There must be no non-phi instructions between the start of a basic block
8287 and the PHI instructions: i.e. PHI instructions must be first in a basic
8290 For the purposes of the SSA form, the use of each incoming value is
8291 deemed to occur on the edge from the corresponding predecessor block to
8292 the current block (but after any definition of an '``invoke``'
8293 instruction's return value on the same edge).
8298 At runtime, the '``phi``' instruction logically takes on the value
8299 specified by the pair corresponding to the predecessor basic block that
8300 executed just prior to the current block.
8305 .. code-block:: llvm
8307 Loop: ; Infinite loop that counts from 0 on up...
8308 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
8309 %nextindvar = add i32 %indvar, 1
8314 '``select``' Instruction
8315 ^^^^^^^^^^^^^^^^^^^^^^^^
8322 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
8324 selty is either i1 or {<N x i1>}
8329 The '``select``' instruction is used to choose one value based on a
8330 condition, without IR-level branching.
8335 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
8336 values indicating the condition, and two values of the same :ref:`first
8337 class <t_firstclass>` type.
8342 If the condition is an i1 and it evaluates to 1, the instruction returns
8343 the first value argument; otherwise, it returns the second value
8346 If the condition is a vector of i1, then the value arguments must be
8347 vectors of the same size, and the selection is done element by element.
8349 If the condition is an i1 and the value arguments are vectors of the
8350 same size, then an entire vector is selected.
8355 .. code-block:: llvm
8357 %X = select i1 true, i8 17, i8 42 ; yields i8:17
8361 '``call``' Instruction
8362 ^^^^^^^^^^^^^^^^^^^^^^
8369 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
8375 The '``call``' instruction represents a simple function call.
8380 This instruction requires several arguments:
8382 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
8383 should perform tail call optimization. The ``tail`` marker is a hint that
8384 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
8385 means that the call must be tail call optimized in order for the program to
8386 be correct. The ``musttail`` marker provides these guarantees:
8388 #. The call will not cause unbounded stack growth if it is part of a
8389 recursive cycle in the call graph.
8390 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
8393 Both markers imply that the callee does not access allocas or varargs from
8394 the caller. Calls marked ``musttail`` must obey the following additional
8397 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
8398 or a pointer bitcast followed by a ret instruction.
8399 - The ret instruction must return the (possibly bitcasted) value
8400 produced by the call or void.
8401 - The caller and callee prototypes must match. Pointer types of
8402 parameters or return types may differ in pointee type, but not
8404 - The calling conventions of the caller and callee must match.
8405 - All ABI-impacting function attributes, such as sret, byval, inreg,
8406 returned, and inalloca, must match.
8407 - The callee must be varargs iff the caller is varargs. Bitcasting a
8408 non-varargs function to the appropriate varargs type is legal so
8409 long as the non-varargs prefixes obey the other rules.
8411 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
8412 the following conditions are met:
8414 - Caller and callee both have the calling convention ``fastcc``.
8415 - The call is in tail position (ret immediately follows call and ret
8416 uses value of call or is void).
8417 - Option ``-tailcallopt`` is enabled, or
8418 ``llvm::GuaranteedTailCallOpt`` is ``true``.
8419 - `Platform-specific constraints are
8420 met. <CodeGenerator.html#tailcallopt>`_
8422 #. The optional "cconv" marker indicates which :ref:`calling
8423 convention <callingconv>` the call should use. If none is
8424 specified, the call defaults to using C calling conventions. The
8425 calling convention of the call must match the calling convention of
8426 the target function, or else the behavior is undefined.
8427 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
8428 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
8430 #. '``ty``': the type of the call instruction itself which is also the
8431 type of the return value. Functions that return no value are marked
8433 #. '``fnty``': shall be the signature of the pointer to function value
8434 being invoked. The argument types must match the types implied by
8435 this signature. This type can be omitted if the function is not
8436 varargs and if the function type does not return a pointer to a
8438 #. '``fnptrval``': An LLVM value containing a pointer to a function to
8439 be invoked. In most cases, this is a direct function invocation, but
8440 indirect ``call``'s are just as possible, calling an arbitrary pointer
8442 #. '``function args``': argument list whose types match the function
8443 signature argument types and parameter attributes. All arguments must
8444 be of :ref:`first class <t_firstclass>` type. If the function signature
8445 indicates the function accepts a variable number of arguments, the
8446 extra arguments can be specified.
8447 #. The optional :ref:`function attributes <fnattrs>` list. Only
8448 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
8449 attributes are valid here.
8450 #. The optional :ref:`operand bundles <opbundles>` list.
8455 The '``call``' instruction is used to cause control flow to transfer to
8456 a specified function, with its incoming arguments bound to the specified
8457 values. Upon a '``ret``' instruction in the called function, control
8458 flow continues with the instruction after the function call, and the
8459 return value of the function is bound to the result argument.
8464 .. code-block:: llvm
8466 %retval = call i32 @test(i32 %argc)
8467 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
8468 %X = tail call i32 @foo() ; yields i32
8469 %Y = tail call fastcc i32 @foo() ; yields i32
8470 call void %foo(i8 97 signext)
8472 %struct.A = type { i32, i8 }
8473 %r = call %struct.A @foo() ; yields { i32, i8 }
8474 %gr = extractvalue %struct.A %r, 0 ; yields i32
8475 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
8476 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
8477 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
8479 llvm treats calls to some functions with names and arguments that match
8480 the standard C99 library as being the C99 library functions, and may
8481 perform optimizations or generate code for them under that assumption.
8482 This is something we'd like to change in the future to provide better
8483 support for freestanding environments and non-C-based languages.
8487 '``va_arg``' Instruction
8488 ^^^^^^^^^^^^^^^^^^^^^^^^
8495 <resultval> = va_arg <va_list*> <arglist>, <argty>
8500 The '``va_arg``' instruction is used to access arguments passed through
8501 the "variable argument" area of a function call. It is used to implement
8502 the ``va_arg`` macro in C.
8507 This instruction takes a ``va_list*`` value and the type of the
8508 argument. It returns a value of the specified argument type and
8509 increments the ``va_list`` to point to the next argument. The actual
8510 type of ``va_list`` is target specific.
8515 The '``va_arg``' instruction loads an argument of the specified type
8516 from the specified ``va_list`` and causes the ``va_list`` to point to
8517 the next argument. For more information, see the variable argument
8518 handling :ref:`Intrinsic Functions <int_varargs>`.
8520 It is legal for this instruction to be called in a function which does
8521 not take a variable number of arguments, for example, the ``vfprintf``
8524 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
8525 function <intrinsics>` because it takes a type as an argument.
8530 See the :ref:`variable argument processing <int_varargs>` section.
8532 Note that the code generator does not yet fully support va\_arg on many
8533 targets. Also, it does not currently support va\_arg with aggregate
8534 types on any target.
8538 '``landingpad``' Instruction
8539 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8546 <resultval> = landingpad <resultty> <clause>+
8547 <resultval> = landingpad <resultty> cleanup <clause>*
8549 <clause> := catch <type> <value>
8550 <clause> := filter <array constant type> <array constant>
8555 The '``landingpad``' instruction is used by `LLVM's exception handling
8556 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8557 is a landing pad --- one where the exception lands, and corresponds to the
8558 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
8559 defines values supplied by the :ref:`personality function <personalityfn>` upon
8560 re-entry to the function. The ``resultval`` has the type ``resultty``.
8566 ``cleanup`` flag indicates that the landing pad block is a cleanup.
8568 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
8569 contains the global variable representing the "type" that may be caught
8570 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
8571 clause takes an array constant as its argument. Use
8572 "``[0 x i8**] undef``" for a filter which cannot throw. The
8573 '``landingpad``' instruction must contain *at least* one ``clause`` or
8574 the ``cleanup`` flag.
8579 The '``landingpad``' instruction defines the values which are set by the
8580 :ref:`personality function <personalityfn>` upon re-entry to the function, and
8581 therefore the "result type" of the ``landingpad`` instruction. As with
8582 calling conventions, how the personality function results are
8583 represented in LLVM IR is target specific.
8585 The clauses are applied in order from top to bottom. If two
8586 ``landingpad`` instructions are merged together through inlining, the
8587 clauses from the calling function are appended to the list of clauses.
8588 When the call stack is being unwound due to an exception being thrown,
8589 the exception is compared against each ``clause`` in turn. If it doesn't
8590 match any of the clauses, and the ``cleanup`` flag is not set, then
8591 unwinding continues further up the call stack.
8593 The ``landingpad`` instruction has several restrictions:
8595 - A landing pad block is a basic block which is the unwind destination
8596 of an '``invoke``' instruction.
8597 - A landing pad block must have a '``landingpad``' instruction as its
8598 first non-PHI instruction.
8599 - There can be only one '``landingpad``' instruction within the landing
8601 - A basic block that is not a landing pad block may not include a
8602 '``landingpad``' instruction.
8607 .. code-block:: llvm
8609 ;; A landing pad which can catch an integer.
8610 %res = landingpad { i8*, i32 }
8612 ;; A landing pad that is a cleanup.
8613 %res = landingpad { i8*, i32 }
8615 ;; A landing pad which can catch an integer and can only throw a double.
8616 %res = landingpad { i8*, i32 }
8618 filter [1 x i8**] [@_ZTId]
8622 '``cleanuppad``' Instruction
8623 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8630 <resultval> = cleanuppad [<args>*]
8635 The '``cleanuppad``' instruction is used by `LLVM's exception handling
8636 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8637 is a cleanup block --- one where a personality routine attempts to
8638 transfer control to run cleanup actions.
8639 The ``args`` correspond to whatever additional
8640 information the :ref:`personality function <personalityfn>` requires to
8641 execute the cleanup.
8642 The ``resultval`` has the type :ref:`token <t_token>` and is used to
8643 match the ``cleanuppad`` to corresponding :ref:`cleanuprets <i_cleanupret>`
8644 and :ref:`cleanupendpads <i_cleanupendpad>`.
8649 The instruction takes a list of arbitrary values which are interpreted
8650 by the :ref:`personality function <personalityfn>`.
8655 When the call stack is being unwound due to an exception being thrown,
8656 the :ref:`personality function <personalityfn>` transfers control to the
8657 ``cleanuppad`` with the aid of the personality-specific arguments.
8658 As with calling conventions, how the personality function results are
8659 represented in LLVM IR is target specific.
8661 The ``cleanuppad`` instruction has several restrictions:
8663 - A cleanup block is a basic block which is the unwind destination of
8664 an exceptional instruction.
8665 - A cleanup block must have a '``cleanuppad``' instruction as its
8666 first non-PHI instruction.
8667 - There can be only one '``cleanuppad``' instruction within the
8669 - A basic block that is not a cleanup block may not include a
8670 '``cleanuppad``' instruction.
8671 - All '``cleanupret``'s and '``cleanupendpad``'s which consume a ``cleanuppad``
8672 must have the same exceptional successor.
8673 - It is undefined behavior for control to transfer from a ``cleanuppad`` to a
8674 ``ret`` without first executing a ``cleanupret`` or ``cleanupendpad`` that
8675 consumes the ``cleanuppad``.
8676 - It is undefined behavior for control to transfer from a ``cleanuppad`` to
8677 itself without first executing a ``cleanupret`` or ``cleanupendpad`` that
8678 consumes the ``cleanuppad``.
8683 .. code-block:: llvm
8685 %tok = cleanuppad []
8692 LLVM supports the notion of an "intrinsic function". These functions
8693 have well known names and semantics and are required to follow certain
8694 restrictions. Overall, these intrinsics represent an extension mechanism
8695 for the LLVM language that does not require changing all of the
8696 transformations in LLVM when adding to the language (or the bitcode
8697 reader/writer, the parser, etc...).
8699 Intrinsic function names must all start with an "``llvm.``" prefix. This
8700 prefix is reserved in LLVM for intrinsic names; thus, function names may
8701 not begin with this prefix. Intrinsic functions must always be external
8702 functions: you cannot define the body of intrinsic functions. Intrinsic
8703 functions may only be used in call or invoke instructions: it is illegal
8704 to take the address of an intrinsic function. Additionally, because
8705 intrinsic functions are part of the LLVM language, it is required if any
8706 are added that they be documented here.
8708 Some intrinsic functions can be overloaded, i.e., the intrinsic
8709 represents a family of functions that perform the same operation but on
8710 different data types. Because LLVM can represent over 8 million
8711 different integer types, overloading is used commonly to allow an
8712 intrinsic function to operate on any integer type. One or more of the
8713 argument types or the result type can be overloaded to accept any
8714 integer type. Argument types may also be defined as exactly matching a
8715 previous argument's type or the result type. This allows an intrinsic
8716 function which accepts multiple arguments, but needs all of them to be
8717 of the same type, to only be overloaded with respect to a single
8718 argument or the result.
8720 Overloaded intrinsics will have the names of its overloaded argument
8721 types encoded into its function name, each preceded by a period. Only
8722 those types which are overloaded result in a name suffix. Arguments
8723 whose type is matched against another type do not. For example, the
8724 ``llvm.ctpop`` function can take an integer of any width and returns an
8725 integer of exactly the same integer width. This leads to a family of
8726 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
8727 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
8728 overloaded, and only one type suffix is required. Because the argument's
8729 type is matched against the return type, it does not require its own
8732 To learn how to add an intrinsic function, please see the `Extending
8733 LLVM Guide <ExtendingLLVM.html>`_.
8737 Variable Argument Handling Intrinsics
8738 -------------------------------------
8740 Variable argument support is defined in LLVM with the
8741 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
8742 functions. These functions are related to the similarly named macros
8743 defined in the ``<stdarg.h>`` header file.
8745 All of these functions operate on arguments that use a target-specific
8746 value type "``va_list``". The LLVM assembly language reference manual
8747 does not define what this type is, so all transformations should be
8748 prepared to handle these functions regardless of the type used.
8750 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
8751 variable argument handling intrinsic functions are used.
8753 .. code-block:: llvm
8755 ; This struct is different for every platform. For most platforms,
8756 ; it is merely an i8*.
8757 %struct.va_list = type { i8* }
8759 ; For Unix x86_64 platforms, va_list is the following struct:
8760 ; %struct.va_list = type { i32, i32, i8*, i8* }
8762 define i32 @test(i32 %X, ...) {
8763 ; Initialize variable argument processing
8764 %ap = alloca %struct.va_list
8765 %ap2 = bitcast %struct.va_list* %ap to i8*
8766 call void @llvm.va_start(i8* %ap2)
8768 ; Read a single integer argument
8769 %tmp = va_arg i8* %ap2, i32
8771 ; Demonstrate usage of llvm.va_copy and llvm.va_end
8773 %aq2 = bitcast i8** %aq to i8*
8774 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
8775 call void @llvm.va_end(i8* %aq2)
8777 ; Stop processing of arguments.
8778 call void @llvm.va_end(i8* %ap2)
8782 declare void @llvm.va_start(i8*)
8783 declare void @llvm.va_copy(i8*, i8*)
8784 declare void @llvm.va_end(i8*)
8788 '``llvm.va_start``' Intrinsic
8789 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8796 declare void @llvm.va_start(i8* <arglist>)
8801 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
8802 subsequent use by ``va_arg``.
8807 The argument is a pointer to a ``va_list`` element to initialize.
8812 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
8813 available in C. In a target-dependent way, it initializes the
8814 ``va_list`` element to which the argument points, so that the next call
8815 to ``va_arg`` will produce the first variable argument passed to the
8816 function. Unlike the C ``va_start`` macro, this intrinsic does not need
8817 to know the last argument of the function as the compiler can figure
8820 '``llvm.va_end``' Intrinsic
8821 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8828 declare void @llvm.va_end(i8* <arglist>)
8833 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
8834 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
8839 The argument is a pointer to a ``va_list`` to destroy.
8844 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
8845 available in C. In a target-dependent way, it destroys the ``va_list``
8846 element to which the argument points. Calls to
8847 :ref:`llvm.va_start <int_va_start>` and
8848 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
8853 '``llvm.va_copy``' Intrinsic
8854 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8861 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
8866 The '``llvm.va_copy``' intrinsic copies the current argument position
8867 from the source argument list to the destination argument list.
8872 The first argument is a pointer to a ``va_list`` element to initialize.
8873 The second argument is a pointer to a ``va_list`` element to copy from.
8878 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
8879 available in C. In a target-dependent way, it copies the source
8880 ``va_list`` element into the destination ``va_list`` element. This
8881 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
8882 arbitrarily complex and require, for example, memory allocation.
8884 Accurate Garbage Collection Intrinsics
8885 --------------------------------------
8887 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
8888 (GC) requires the frontend to generate code containing appropriate intrinsic
8889 calls and select an appropriate GC strategy which knows how to lower these
8890 intrinsics in a manner which is appropriate for the target collector.
8892 These intrinsics allow identification of :ref:`GC roots on the
8893 stack <int_gcroot>`, as well as garbage collector implementations that
8894 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
8895 Frontends for type-safe garbage collected languages should generate
8896 these intrinsics to make use of the LLVM garbage collectors. For more
8897 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
8899 Experimental Statepoint Intrinsics
8900 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8902 LLVM provides an second experimental set of intrinsics for describing garbage
8903 collection safepoints in compiled code. These intrinsics are an alternative
8904 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
8905 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
8906 differences in approach are covered in the `Garbage Collection with LLVM
8907 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
8908 described in :doc:`Statepoints`.
8912 '``llvm.gcroot``' Intrinsic
8913 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8920 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
8925 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
8926 the code generator, and allows some metadata to be associated with it.
8931 The first argument specifies the address of a stack object that contains
8932 the root pointer. The second pointer (which must be either a constant or
8933 a global value address) contains the meta-data to be associated with the
8939 At runtime, a call to this intrinsic stores a null pointer into the
8940 "ptrloc" location. At compile-time, the code generator generates
8941 information to allow the runtime to find the pointer at GC safe points.
8942 The '``llvm.gcroot``' intrinsic may only be used in a function which
8943 :ref:`specifies a GC algorithm <gc>`.
8947 '``llvm.gcread``' Intrinsic
8948 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8955 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
8960 The '``llvm.gcread``' intrinsic identifies reads of references from heap
8961 locations, allowing garbage collector implementations that require read
8967 The second argument is the address to read from, which should be an
8968 address allocated from the garbage collector. The first object is a
8969 pointer to the start of the referenced object, if needed by the language
8970 runtime (otherwise null).
8975 The '``llvm.gcread``' intrinsic has the same semantics as a load
8976 instruction, but may be replaced with substantially more complex code by
8977 the garbage collector runtime, as needed. The '``llvm.gcread``'
8978 intrinsic may only be used in a function which :ref:`specifies a GC
8983 '``llvm.gcwrite``' Intrinsic
8984 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8991 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
8996 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
8997 locations, allowing garbage collector implementations that require write
8998 barriers (such as generational or reference counting collectors).
9003 The first argument is the reference to store, the second is the start of
9004 the object to store it to, and the third is the address of the field of
9005 Obj to store to. If the runtime does not require a pointer to the
9006 object, Obj may be null.
9011 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
9012 instruction, but may be replaced with substantially more complex code by
9013 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
9014 intrinsic may only be used in a function which :ref:`specifies a GC
9017 Code Generator Intrinsics
9018 -------------------------
9020 These intrinsics are provided by LLVM to expose special features that
9021 may only be implemented with code generator support.
9023 '``llvm.returnaddress``' Intrinsic
9024 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9031 declare i8 *@llvm.returnaddress(i32 <level>)
9036 The '``llvm.returnaddress``' intrinsic attempts to compute a
9037 target-specific value indicating the return address of the current
9038 function or one of its callers.
9043 The argument to this intrinsic indicates which function to return the
9044 address for. Zero indicates the calling function, one indicates its
9045 caller, etc. The argument is **required** to be a constant integer
9051 The '``llvm.returnaddress``' intrinsic either returns a pointer
9052 indicating the return address of the specified call frame, or zero if it
9053 cannot be identified. The value returned by this intrinsic is likely to
9054 be incorrect or 0 for arguments other than zero, so it should only be
9055 used for debugging purposes.
9057 Note that calling this intrinsic does not prevent function inlining or
9058 other aggressive transformations, so the value returned may not be that
9059 of the obvious source-language caller.
9061 '``llvm.frameaddress``' Intrinsic
9062 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9069 declare i8* @llvm.frameaddress(i32 <level>)
9074 The '``llvm.frameaddress``' intrinsic attempts to return the
9075 target-specific frame pointer value for the specified stack frame.
9080 The argument to this intrinsic indicates which function to return the
9081 frame pointer for. Zero indicates the calling function, one indicates
9082 its caller, etc. The argument is **required** to be a constant integer
9088 The '``llvm.frameaddress``' intrinsic either returns a pointer
9089 indicating the frame address of the specified call frame, or zero if it
9090 cannot be identified. The value returned by this intrinsic is likely to
9091 be incorrect or 0 for arguments other than zero, so it should only be
9092 used for debugging purposes.
9094 Note that calling this intrinsic does not prevent function inlining or
9095 other aggressive transformations, so the value returned may not be that
9096 of the obvious source-language caller.
9098 '``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
9099 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9106 declare void @llvm.localescape(...)
9107 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
9112 The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
9113 allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
9114 live frame pointer to recover the address of the allocation. The offset is
9115 computed during frame layout of the caller of ``llvm.localescape``.
9120 All arguments to '``llvm.localescape``' must be pointers to static allocas or
9121 casts of static allocas. Each function can only call '``llvm.localescape``'
9122 once, and it can only do so from the entry block.
9124 The ``func`` argument to '``llvm.localrecover``' must be a constant
9125 bitcasted pointer to a function defined in the current module. The code
9126 generator cannot determine the frame allocation offset of functions defined in
9129 The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
9130 call frame that is currently live. The return value of '``llvm.localaddress``'
9131 is one way to produce such a value, but various runtimes also expose a suitable
9132 pointer in platform-specific ways.
9134 The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
9135 '``llvm.localescape``' to recover. It is zero-indexed.
9140 These intrinsics allow a group of functions to share access to a set of local
9141 stack allocations of a one parent function. The parent function may call the
9142 '``llvm.localescape``' intrinsic once from the function entry block, and the
9143 child functions can use '``llvm.localrecover``' to access the escaped allocas.
9144 The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
9145 the escaped allocas are allocated, which would break attempts to use
9146 '``llvm.localrecover``'.
9148 .. _int_read_register:
9149 .. _int_write_register:
9151 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
9152 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9159 declare i32 @llvm.read_register.i32(metadata)
9160 declare i64 @llvm.read_register.i64(metadata)
9161 declare void @llvm.write_register.i32(metadata, i32 @value)
9162 declare void @llvm.write_register.i64(metadata, i64 @value)
9168 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
9169 provides access to the named register. The register must be valid on
9170 the architecture being compiled to. The type needs to be compatible
9171 with the register being read.
9176 The '``llvm.read_register``' intrinsic returns the current value of the
9177 register, where possible. The '``llvm.write_register``' intrinsic sets
9178 the current value of the register, where possible.
9180 This is useful to implement named register global variables that need
9181 to always be mapped to a specific register, as is common practice on
9182 bare-metal programs including OS kernels.
9184 The compiler doesn't check for register availability or use of the used
9185 register in surrounding code, including inline assembly. Because of that,
9186 allocatable registers are not supported.
9188 Warning: So far it only works with the stack pointer on selected
9189 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
9190 work is needed to support other registers and even more so, allocatable
9195 '``llvm.stacksave``' Intrinsic
9196 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9203 declare i8* @llvm.stacksave()
9208 The '``llvm.stacksave``' intrinsic is used to remember the current state
9209 of the function stack, for use with
9210 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
9211 implementing language features like scoped automatic variable sized
9217 This intrinsic returns a opaque pointer value that can be passed to
9218 :ref:`llvm.stackrestore <int_stackrestore>`. When an
9219 ``llvm.stackrestore`` intrinsic is executed with a value saved from
9220 ``llvm.stacksave``, it effectively restores the state of the stack to
9221 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
9222 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
9223 were allocated after the ``llvm.stacksave`` was executed.
9225 .. _int_stackrestore:
9227 '``llvm.stackrestore``' Intrinsic
9228 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9235 declare void @llvm.stackrestore(i8* %ptr)
9240 The '``llvm.stackrestore``' intrinsic is used to restore the state of
9241 the function stack to the state it was in when the corresponding
9242 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
9243 useful for implementing language features like scoped automatic variable
9244 sized arrays in C99.
9249 See the description for :ref:`llvm.stacksave <int_stacksave>`.
9251 '``llvm.prefetch``' Intrinsic
9252 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9259 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
9264 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
9265 insert a prefetch instruction if supported; otherwise, it is a noop.
9266 Prefetches have no effect on the behavior of the program but can change
9267 its performance characteristics.
9272 ``address`` is the address to be prefetched, ``rw`` is the specifier
9273 determining if the fetch should be for a read (0) or write (1), and
9274 ``locality`` is a temporal locality specifier ranging from (0) - no
9275 locality, to (3) - extremely local keep in cache. The ``cache type``
9276 specifies whether the prefetch is performed on the data (1) or
9277 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
9278 arguments must be constant integers.
9283 This intrinsic does not modify the behavior of the program. In
9284 particular, prefetches cannot trap and do not produce a value. On
9285 targets that support this intrinsic, the prefetch can provide hints to
9286 the processor cache for better performance.
9288 '``llvm.pcmarker``' Intrinsic
9289 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9296 declare void @llvm.pcmarker(i32 <id>)
9301 The '``llvm.pcmarker``' intrinsic is a method to export a Program
9302 Counter (PC) in a region of code to simulators and other tools. The
9303 method is target specific, but it is expected that the marker will use
9304 exported symbols to transmit the PC of the marker. The marker makes no
9305 guarantees that it will remain with any specific instruction after
9306 optimizations. It is possible that the presence of a marker will inhibit
9307 optimizations. The intended use is to be inserted after optimizations to
9308 allow correlations of simulation runs.
9313 ``id`` is a numerical id identifying the marker.
9318 This intrinsic does not modify the behavior of the program. Backends
9319 that do not support this intrinsic may ignore it.
9321 '``llvm.readcyclecounter``' Intrinsic
9322 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9329 declare i64 @llvm.readcyclecounter()
9334 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
9335 counter register (or similar low latency, high accuracy clocks) on those
9336 targets that support it. On X86, it should map to RDTSC. On Alpha, it
9337 should map to RPCC. As the backing counters overflow quickly (on the
9338 order of 9 seconds on alpha), this should only be used for small
9344 When directly supported, reading the cycle counter should not modify any
9345 memory. Implementations are allowed to either return a application
9346 specific value or a system wide value. On backends without support, this
9347 is lowered to a constant 0.
9349 Note that runtime support may be conditional on the privilege-level code is
9350 running at and the host platform.
9352 '``llvm.clear_cache``' Intrinsic
9353 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9360 declare void @llvm.clear_cache(i8*, i8*)
9365 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
9366 in the specified range to the execution unit of the processor. On
9367 targets with non-unified instruction and data cache, the implementation
9368 flushes the instruction cache.
9373 On platforms with coherent instruction and data caches (e.g. x86), this
9374 intrinsic is a nop. On platforms with non-coherent instruction and data
9375 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
9376 instructions or a system call, if cache flushing requires special
9379 The default behavior is to emit a call to ``__clear_cache`` from the run
9382 This instrinsic does *not* empty the instruction pipeline. Modifications
9383 of the current function are outside the scope of the intrinsic.
9385 '``llvm.instrprof_increment``' Intrinsic
9386 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9393 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
9394 i32 <num-counters>, i32 <index>)
9399 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
9400 frontend for use with instrumentation based profiling. These will be
9401 lowered by the ``-instrprof`` pass to generate execution counts of a
9407 The first argument is a pointer to a global variable containing the
9408 name of the entity being instrumented. This should generally be the
9409 (mangled) function name for a set of counters.
9411 The second argument is a hash value that can be used by the consumer
9412 of the profile data to detect changes to the instrumented source, and
9413 the third is the number of counters associated with ``name``. It is an
9414 error if ``hash`` or ``num-counters`` differ between two instances of
9415 ``instrprof_increment`` that refer to the same name.
9417 The last argument refers to which of the counters for ``name`` should
9418 be incremented. It should be a value between 0 and ``num-counters``.
9423 This intrinsic represents an increment of a profiling counter. It will
9424 cause the ``-instrprof`` pass to generate the appropriate data
9425 structures and the code to increment the appropriate value, in a
9426 format that can be written out by a compiler runtime and consumed via
9427 the ``llvm-profdata`` tool.
9429 Standard C Library Intrinsics
9430 -----------------------------
9432 LLVM provides intrinsics for a few important standard C library
9433 functions. These intrinsics allow source-language front-ends to pass
9434 information about the alignment of the pointer arguments to the code
9435 generator, providing opportunity for more efficient code generation.
9439 '``llvm.memcpy``' Intrinsic
9440 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9445 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
9446 integer bit width and for different address spaces. Not all targets
9447 support all bit widths however.
9451 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9452 i32 <len>, i32 <align>, i1 <isvolatile>)
9453 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9454 i64 <len>, i32 <align>, i1 <isvolatile>)
9459 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9460 source location to the destination location.
9462 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
9463 intrinsics do not return a value, takes extra alignment/isvolatile
9464 arguments and the pointers can be in specified address spaces.
9469 The first argument is a pointer to the destination, the second is a
9470 pointer to the source. The third argument is an integer argument
9471 specifying the number of bytes to copy, the fourth argument is the
9472 alignment of the source and destination locations, and the fifth is a
9473 boolean indicating a volatile access.
9475 If the call to this intrinsic has an alignment value that is not 0 or 1,
9476 then the caller guarantees that both the source and destination pointers
9477 are aligned to that boundary.
9479 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
9480 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9481 very cleanly specified and it is unwise to depend on it.
9486 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9487 source location to the destination location, which are not allowed to
9488 overlap. It copies "len" bytes of memory over. If the argument is known
9489 to be aligned to some boundary, this can be specified as the fourth
9490 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
9492 '``llvm.memmove``' Intrinsic
9493 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9498 This is an overloaded intrinsic. You can use llvm.memmove on any integer
9499 bit width and for different address space. Not all targets support all
9504 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9505 i32 <len>, i32 <align>, i1 <isvolatile>)
9506 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9507 i64 <len>, i32 <align>, i1 <isvolatile>)
9512 The '``llvm.memmove.*``' intrinsics move a block of memory from the
9513 source location to the destination location. It is similar to the
9514 '``llvm.memcpy``' intrinsic but allows the two memory locations to
9517 Note that, unlike the standard libc function, the ``llvm.memmove.*``
9518 intrinsics do not return a value, takes extra alignment/isvolatile
9519 arguments and the pointers can be in specified address spaces.
9524 The first argument is a pointer to the destination, the second is a
9525 pointer to the source. The third argument is an integer argument
9526 specifying the number of bytes to copy, the fourth argument is the
9527 alignment of the source and destination locations, and the fifth is a
9528 boolean indicating a volatile access.
9530 If the call to this intrinsic has an alignment value that is not 0 or 1,
9531 then the caller guarantees that the source and destination pointers are
9532 aligned to that boundary.
9534 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
9535 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
9536 not very cleanly specified and it is unwise to depend on it.
9541 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
9542 source location to the destination location, which may overlap. It
9543 copies "len" bytes of memory over. If the argument is known to be
9544 aligned to some boundary, this can be specified as the fourth argument,
9545 otherwise it should be set to 0 or 1 (both meaning no alignment).
9547 '``llvm.memset.*``' Intrinsics
9548 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9553 This is an overloaded intrinsic. You can use llvm.memset on any integer
9554 bit width and for different address spaces. However, not all targets
9555 support all bit widths.
9559 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
9560 i32 <len>, i32 <align>, i1 <isvolatile>)
9561 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
9562 i64 <len>, i32 <align>, i1 <isvolatile>)
9567 The '``llvm.memset.*``' intrinsics fill a block of memory with a
9568 particular byte value.
9570 Note that, unlike the standard libc function, the ``llvm.memset``
9571 intrinsic does not return a value and takes extra alignment/volatile
9572 arguments. Also, the destination can be in an arbitrary address space.
9577 The first argument is a pointer to the destination to fill, the second
9578 is the byte value with which to fill it, the third argument is an
9579 integer argument specifying the number of bytes to fill, and the fourth
9580 argument is the known alignment of the destination location.
9582 If the call to this intrinsic has an alignment value that is not 0 or 1,
9583 then the caller guarantees that the destination pointer is aligned to
9586 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
9587 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9588 very cleanly specified and it is unwise to depend on it.
9593 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
9594 at the destination location. If the argument is known to be aligned to
9595 some boundary, this can be specified as the fourth argument, otherwise
9596 it should be set to 0 or 1 (both meaning no alignment).
9598 '``llvm.sqrt.*``' Intrinsic
9599 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9604 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
9605 floating point or vector of floating point type. Not all targets support
9610 declare float @llvm.sqrt.f32(float %Val)
9611 declare double @llvm.sqrt.f64(double %Val)
9612 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
9613 declare fp128 @llvm.sqrt.f128(fp128 %Val)
9614 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
9619 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
9620 returning the same value as the libm '``sqrt``' functions would. Unlike
9621 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
9622 negative numbers other than -0.0 (which allows for better optimization,
9623 because there is no need to worry about errno being set).
9624 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
9629 The argument and return value are floating point numbers of the same
9635 This function returns the sqrt of the specified operand if it is a
9636 nonnegative floating point number.
9638 '``llvm.powi.*``' Intrinsic
9639 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9644 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
9645 floating point or vector of floating point type. Not all targets support
9650 declare float @llvm.powi.f32(float %Val, i32 %power)
9651 declare double @llvm.powi.f64(double %Val, i32 %power)
9652 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
9653 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
9654 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
9659 The '``llvm.powi.*``' intrinsics return the first operand raised to the
9660 specified (positive or negative) power. The order of evaluation of
9661 multiplications is not defined. When a vector of floating point type is
9662 used, the second argument remains a scalar integer value.
9667 The second argument is an integer power, and the first is a value to
9668 raise to that power.
9673 This function returns the first value raised to the second power with an
9674 unspecified sequence of rounding operations.
9676 '``llvm.sin.*``' Intrinsic
9677 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9682 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
9683 floating point or vector of floating point type. Not all targets support
9688 declare float @llvm.sin.f32(float %Val)
9689 declare double @llvm.sin.f64(double %Val)
9690 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
9691 declare fp128 @llvm.sin.f128(fp128 %Val)
9692 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
9697 The '``llvm.sin.*``' intrinsics return the sine of the operand.
9702 The argument and return value are floating point numbers of the same
9708 This function returns the sine of the specified operand, returning the
9709 same values as the libm ``sin`` functions would, and handles error
9710 conditions in the same way.
9712 '``llvm.cos.*``' Intrinsic
9713 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9718 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
9719 floating point or vector of floating point type. Not all targets support
9724 declare float @llvm.cos.f32(float %Val)
9725 declare double @llvm.cos.f64(double %Val)
9726 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
9727 declare fp128 @llvm.cos.f128(fp128 %Val)
9728 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
9733 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
9738 The argument and return value are floating point numbers of the same
9744 This function returns the cosine of the specified operand, returning the
9745 same values as the libm ``cos`` functions would, and handles error
9746 conditions in the same way.
9748 '``llvm.pow.*``' Intrinsic
9749 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9754 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
9755 floating point or vector of floating point type. Not all targets support
9760 declare float @llvm.pow.f32(float %Val, float %Power)
9761 declare double @llvm.pow.f64(double %Val, double %Power)
9762 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
9763 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
9764 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
9769 The '``llvm.pow.*``' intrinsics return the first operand raised to the
9770 specified (positive or negative) power.
9775 The second argument is a floating point power, and the first is a value
9776 to raise to that power.
9781 This function returns the first value raised to the second power,
9782 returning the same values as the libm ``pow`` functions would, and
9783 handles error conditions in the same way.
9785 '``llvm.exp.*``' Intrinsic
9786 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9791 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
9792 floating point or vector of floating point type. Not all targets support
9797 declare float @llvm.exp.f32(float %Val)
9798 declare double @llvm.exp.f64(double %Val)
9799 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
9800 declare fp128 @llvm.exp.f128(fp128 %Val)
9801 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
9806 The '``llvm.exp.*``' intrinsics perform the exp function.
9811 The argument and return value are floating point numbers of the same
9817 This function returns the same values as the libm ``exp`` functions
9818 would, and handles error conditions in the same way.
9820 '``llvm.exp2.*``' Intrinsic
9821 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9826 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
9827 floating point or vector of floating point type. Not all targets support
9832 declare float @llvm.exp2.f32(float %Val)
9833 declare double @llvm.exp2.f64(double %Val)
9834 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
9835 declare fp128 @llvm.exp2.f128(fp128 %Val)
9836 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
9841 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
9846 The argument and return value are floating point numbers of the same
9852 This function returns the same values as the libm ``exp2`` functions
9853 would, and handles error conditions in the same way.
9855 '``llvm.log.*``' Intrinsic
9856 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9861 This is an overloaded intrinsic. You can use ``llvm.log`` on any
9862 floating point or vector of floating point type. Not all targets support
9867 declare float @llvm.log.f32(float %Val)
9868 declare double @llvm.log.f64(double %Val)
9869 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
9870 declare fp128 @llvm.log.f128(fp128 %Val)
9871 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
9876 The '``llvm.log.*``' intrinsics perform the log function.
9881 The argument and return value are floating point numbers of the same
9887 This function returns the same values as the libm ``log`` functions
9888 would, and handles error conditions in the same way.
9890 '``llvm.log10.*``' Intrinsic
9891 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9896 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
9897 floating point or vector of floating point type. Not all targets support
9902 declare float @llvm.log10.f32(float %Val)
9903 declare double @llvm.log10.f64(double %Val)
9904 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
9905 declare fp128 @llvm.log10.f128(fp128 %Val)
9906 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
9911 The '``llvm.log10.*``' intrinsics perform the log10 function.
9916 The argument and return value are floating point numbers of the same
9922 This function returns the same values as the libm ``log10`` functions
9923 would, and handles error conditions in the same way.
9925 '``llvm.log2.*``' Intrinsic
9926 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9931 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
9932 floating point or vector of floating point type. Not all targets support
9937 declare float @llvm.log2.f32(float %Val)
9938 declare double @llvm.log2.f64(double %Val)
9939 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
9940 declare fp128 @llvm.log2.f128(fp128 %Val)
9941 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
9946 The '``llvm.log2.*``' intrinsics perform the log2 function.
9951 The argument and return value are floating point numbers of the same
9957 This function returns the same values as the libm ``log2`` functions
9958 would, and handles error conditions in the same way.
9960 '``llvm.fma.*``' Intrinsic
9961 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9966 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
9967 floating point or vector of floating point type. Not all targets support
9972 declare float @llvm.fma.f32(float %a, float %b, float %c)
9973 declare double @llvm.fma.f64(double %a, double %b, double %c)
9974 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
9975 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
9976 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
9981 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
9987 The argument and return value are floating point numbers of the same
9993 This function returns the same values as the libm ``fma`` functions
9994 would, and does not set errno.
9996 '``llvm.fabs.*``' Intrinsic
9997 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10002 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
10003 floating point or vector of floating point type. Not all targets support
10008 declare float @llvm.fabs.f32(float %Val)
10009 declare double @llvm.fabs.f64(double %Val)
10010 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
10011 declare fp128 @llvm.fabs.f128(fp128 %Val)
10012 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
10017 The '``llvm.fabs.*``' intrinsics return the absolute value of the
10023 The argument and return value are floating point numbers of the same
10029 This function returns the same values as the libm ``fabs`` functions
10030 would, and handles error conditions in the same way.
10032 '``llvm.minnum.*``' Intrinsic
10033 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10038 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
10039 floating point or vector of floating point type. Not all targets support
10044 declare float @llvm.minnum.f32(float %Val0, float %Val1)
10045 declare double @llvm.minnum.f64(double %Val0, double %Val1)
10046 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10047 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
10048 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10053 The '``llvm.minnum.*``' intrinsics return the minimum of the two
10060 The arguments and return value are floating point numbers of the same
10066 Follows the IEEE-754 semantics for minNum, which also match for libm's
10069 If either operand is a NaN, returns the other non-NaN operand. Returns
10070 NaN only if both operands are NaN. If the operands compare equal,
10071 returns a value that compares equal to both operands. This means that
10072 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10074 '``llvm.maxnum.*``' Intrinsic
10075 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10080 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
10081 floating point or vector of floating point type. Not all targets support
10086 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
10087 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
10088 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10089 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
10090 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10095 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
10102 The arguments and return value are floating point numbers of the same
10107 Follows the IEEE-754 semantics for maxNum, which also match for libm's
10110 If either operand is a NaN, returns the other non-NaN operand. Returns
10111 NaN only if both operands are NaN. If the operands compare equal,
10112 returns a value that compares equal to both operands. This means that
10113 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10115 '``llvm.copysign.*``' Intrinsic
10116 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10121 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
10122 floating point or vector of floating point type. Not all targets support
10127 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
10128 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
10129 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
10130 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
10131 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
10136 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
10137 first operand and the sign of the second operand.
10142 The arguments and return value are floating point numbers of the same
10148 This function returns the same values as the libm ``copysign``
10149 functions would, and handles error conditions in the same way.
10151 '``llvm.floor.*``' Intrinsic
10152 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10157 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
10158 floating point or vector of floating point type. Not all targets support
10163 declare float @llvm.floor.f32(float %Val)
10164 declare double @llvm.floor.f64(double %Val)
10165 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
10166 declare fp128 @llvm.floor.f128(fp128 %Val)
10167 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
10172 The '``llvm.floor.*``' intrinsics return the floor of the operand.
10177 The argument and return value are floating point numbers of the same
10183 This function returns the same values as the libm ``floor`` functions
10184 would, and handles error conditions in the same way.
10186 '``llvm.ceil.*``' Intrinsic
10187 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10192 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
10193 floating point or vector of floating point type. Not all targets support
10198 declare float @llvm.ceil.f32(float %Val)
10199 declare double @llvm.ceil.f64(double %Val)
10200 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
10201 declare fp128 @llvm.ceil.f128(fp128 %Val)
10202 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
10207 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
10212 The argument and return value are floating point numbers of the same
10218 This function returns the same values as the libm ``ceil`` functions
10219 would, and handles error conditions in the same way.
10221 '``llvm.trunc.*``' Intrinsic
10222 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10227 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
10228 floating point or vector of floating point type. Not all targets support
10233 declare float @llvm.trunc.f32(float %Val)
10234 declare double @llvm.trunc.f64(double %Val)
10235 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
10236 declare fp128 @llvm.trunc.f128(fp128 %Val)
10237 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
10242 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
10243 nearest integer not larger in magnitude than the operand.
10248 The argument and return value are floating point numbers of the same
10254 This function returns the same values as the libm ``trunc`` functions
10255 would, and handles error conditions in the same way.
10257 '``llvm.rint.*``' Intrinsic
10258 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10263 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
10264 floating point or vector of floating point type. Not all targets support
10269 declare float @llvm.rint.f32(float %Val)
10270 declare double @llvm.rint.f64(double %Val)
10271 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
10272 declare fp128 @llvm.rint.f128(fp128 %Val)
10273 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
10278 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
10279 nearest integer. It may raise an inexact floating-point exception if the
10280 operand isn't an integer.
10285 The argument and return value are floating point numbers of the same
10291 This function returns the same values as the libm ``rint`` functions
10292 would, and handles error conditions in the same way.
10294 '``llvm.nearbyint.*``' Intrinsic
10295 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10300 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
10301 floating point or vector of floating point type. Not all targets support
10306 declare float @llvm.nearbyint.f32(float %Val)
10307 declare double @llvm.nearbyint.f64(double %Val)
10308 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
10309 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
10310 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
10315 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
10321 The argument and return value are floating point numbers of the same
10327 This function returns the same values as the libm ``nearbyint``
10328 functions would, and handles error conditions in the same way.
10330 '``llvm.round.*``' Intrinsic
10331 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10336 This is an overloaded intrinsic. You can use ``llvm.round`` on any
10337 floating point or vector of floating point type. Not all targets support
10342 declare float @llvm.round.f32(float %Val)
10343 declare double @llvm.round.f64(double %Val)
10344 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
10345 declare fp128 @llvm.round.f128(fp128 %Val)
10346 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
10351 The '``llvm.round.*``' intrinsics returns the operand rounded to the
10357 The argument and return value are floating point numbers of the same
10363 This function returns the same values as the libm ``round``
10364 functions would, and handles error conditions in the same way.
10366 Bit Manipulation Intrinsics
10367 ---------------------------
10369 LLVM provides intrinsics for a few important bit manipulation
10370 operations. These allow efficient code generation for some algorithms.
10372 '``llvm.bswap.*``' Intrinsics
10373 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10378 This is an overloaded intrinsic function. You can use bswap on any
10379 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
10383 declare i16 @llvm.bswap.i16(i16 <id>)
10384 declare i32 @llvm.bswap.i32(i32 <id>)
10385 declare i64 @llvm.bswap.i64(i64 <id>)
10390 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
10391 values with an even number of bytes (positive multiple of 16 bits).
10392 These are useful for performing operations on data that is not in the
10393 target's native byte order.
10398 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
10399 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
10400 intrinsic returns an i32 value that has the four bytes of the input i32
10401 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
10402 returned i32 will have its bytes in 3, 2, 1, 0 order. The
10403 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
10404 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
10407 '``llvm.ctpop.*``' Intrinsic
10408 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10413 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
10414 bit width, or on any vector with integer elements. Not all targets
10415 support all bit widths or vector types, however.
10419 declare i8 @llvm.ctpop.i8(i8 <src>)
10420 declare i16 @llvm.ctpop.i16(i16 <src>)
10421 declare i32 @llvm.ctpop.i32(i32 <src>)
10422 declare i64 @llvm.ctpop.i64(i64 <src>)
10423 declare i256 @llvm.ctpop.i256(i256 <src>)
10424 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
10429 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
10435 The only argument is the value to be counted. The argument may be of any
10436 integer type, or a vector with integer elements. The return type must
10437 match the argument type.
10442 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
10443 each element of a vector.
10445 '``llvm.ctlz.*``' Intrinsic
10446 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10451 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
10452 integer bit width, or any vector whose elements are integers. Not all
10453 targets support all bit widths or vector types, however.
10457 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
10458 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
10459 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
10460 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
10461 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
10462 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10467 The '``llvm.ctlz``' family of intrinsic functions counts the number of
10468 leading zeros in a variable.
10473 The first argument is the value to be counted. This argument may be of
10474 any integer type, or a vector with integer element type. The return
10475 type must match the first argument type.
10477 The second argument must be a constant and is a flag to indicate whether
10478 the intrinsic should ensure that a zero as the first argument produces a
10479 defined result. Historically some architectures did not provide a
10480 defined result for zero values as efficiently, and many algorithms are
10481 now predicated on avoiding zero-value inputs.
10486 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
10487 zeros in a variable, or within each element of the vector. If
10488 ``src == 0`` then the result is the size in bits of the type of ``src``
10489 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10490 ``llvm.ctlz(i32 2) = 30``.
10492 '``llvm.cttz.*``' Intrinsic
10493 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10498 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
10499 integer bit width, or any vector of integer elements. Not all targets
10500 support all bit widths or vector types, however.
10504 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
10505 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
10506 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
10507 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
10508 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
10509 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10514 The '``llvm.cttz``' family of intrinsic functions counts the number of
10520 The first argument is the value to be counted. This argument may be of
10521 any integer type, or a vector with integer element type. The return
10522 type must match the first argument type.
10524 The second argument must be a constant and is a flag to indicate whether
10525 the intrinsic should ensure that a zero as the first argument produces a
10526 defined result. Historically some architectures did not provide a
10527 defined result for zero values as efficiently, and many algorithms are
10528 now predicated on avoiding zero-value inputs.
10533 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
10534 zeros in a variable, or within each element of a vector. If ``src == 0``
10535 then the result is the size in bits of the type of ``src`` if
10536 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10537 ``llvm.cttz(2) = 1``.
10541 Arithmetic with Overflow Intrinsics
10542 -----------------------------------
10544 LLVM provides intrinsics for some arithmetic with overflow operations.
10546 '``llvm.sadd.with.overflow.*``' Intrinsics
10547 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10552 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
10553 on any integer bit width.
10557 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
10558 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10559 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
10564 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10565 a signed addition of the two arguments, and indicate whether an overflow
10566 occurred during the signed summation.
10571 The arguments (%a and %b) and the first element of the result structure
10572 may be of integer types of any bit width, but they must have the same
10573 bit width. The second element of the result structure must be of type
10574 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10580 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10581 a signed addition of the two variables. They return a structure --- the
10582 first element of which is the signed summation, and the second element
10583 of which is a bit specifying if the signed summation resulted in an
10589 .. code-block:: llvm
10591 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10592 %sum = extractvalue {i32, i1} %res, 0
10593 %obit = extractvalue {i32, i1} %res, 1
10594 br i1 %obit, label %overflow, label %normal
10596 '``llvm.uadd.with.overflow.*``' Intrinsics
10597 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10602 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
10603 on any integer bit width.
10607 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
10608 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10609 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
10614 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10615 an unsigned addition of the two arguments, and indicate whether a carry
10616 occurred during the unsigned summation.
10621 The arguments (%a and %b) and the first element of the result structure
10622 may be of integer types of any bit width, but they must have the same
10623 bit width. The second element of the result structure must be of type
10624 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10630 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10631 an unsigned addition of the two arguments. They return a structure --- the
10632 first element of which is the sum, and the second element of which is a
10633 bit specifying if the unsigned summation resulted in a carry.
10638 .. code-block:: llvm
10640 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10641 %sum = extractvalue {i32, i1} %res, 0
10642 %obit = extractvalue {i32, i1} %res, 1
10643 br i1 %obit, label %carry, label %normal
10645 '``llvm.ssub.with.overflow.*``' Intrinsics
10646 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10651 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
10652 on any integer bit width.
10656 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
10657 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10658 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
10663 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10664 a signed subtraction of the two arguments, and indicate whether an
10665 overflow occurred during the signed subtraction.
10670 The arguments (%a and %b) and the first element of the result structure
10671 may be of integer types of any bit width, but they must have the same
10672 bit width. The second element of the result structure must be of type
10673 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10679 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10680 a signed subtraction of the two arguments. They return a structure --- the
10681 first element of which is the subtraction, and the second element of
10682 which is a bit specifying if the signed subtraction resulted in an
10688 .. code-block:: llvm
10690 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10691 %sum = extractvalue {i32, i1} %res, 0
10692 %obit = extractvalue {i32, i1} %res, 1
10693 br i1 %obit, label %overflow, label %normal
10695 '``llvm.usub.with.overflow.*``' Intrinsics
10696 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10701 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
10702 on any integer bit width.
10706 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
10707 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10708 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
10713 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10714 an unsigned subtraction of the two arguments, and indicate whether an
10715 overflow occurred during the unsigned subtraction.
10720 The arguments (%a and %b) and the first element of the result structure
10721 may be of integer types of any bit width, but they must have the same
10722 bit width. The second element of the result structure must be of type
10723 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10729 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10730 an unsigned subtraction of the two arguments. They return a structure ---
10731 the first element of which is the subtraction, and the second element of
10732 which is a bit specifying if the unsigned subtraction resulted in an
10738 .. code-block:: llvm
10740 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10741 %sum = extractvalue {i32, i1} %res, 0
10742 %obit = extractvalue {i32, i1} %res, 1
10743 br i1 %obit, label %overflow, label %normal
10745 '``llvm.smul.with.overflow.*``' Intrinsics
10746 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10751 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
10752 on any integer bit width.
10756 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
10757 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10758 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
10763 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10764 a signed multiplication of the two arguments, and indicate whether an
10765 overflow occurred during the signed multiplication.
10770 The arguments (%a and %b) and the first element of the result structure
10771 may be of integer types of any bit width, but they must have the same
10772 bit width. The second element of the result structure must be of type
10773 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10779 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10780 a signed multiplication of the two arguments. They return a structure ---
10781 the first element of which is the multiplication, and the second element
10782 of which is a bit specifying if the signed multiplication resulted in an
10788 .. code-block:: llvm
10790 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10791 %sum = extractvalue {i32, i1} %res, 0
10792 %obit = extractvalue {i32, i1} %res, 1
10793 br i1 %obit, label %overflow, label %normal
10795 '``llvm.umul.with.overflow.*``' Intrinsics
10796 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10801 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
10802 on any integer bit width.
10806 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
10807 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10808 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
10813 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10814 a unsigned multiplication of the two arguments, and indicate whether an
10815 overflow occurred during the unsigned multiplication.
10820 The arguments (%a and %b) and the first element of the result structure
10821 may be of integer types of any bit width, but they must have the same
10822 bit width. The second element of the result structure must be of type
10823 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10829 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10830 an unsigned multiplication of the two arguments. They return a structure ---
10831 the first element of which is the multiplication, and the second
10832 element of which is a bit specifying if the unsigned multiplication
10833 resulted in an overflow.
10838 .. code-block:: llvm
10840 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10841 %sum = extractvalue {i32, i1} %res, 0
10842 %obit = extractvalue {i32, i1} %res, 1
10843 br i1 %obit, label %overflow, label %normal
10845 Specialised Arithmetic Intrinsics
10846 ---------------------------------
10848 '``llvm.canonicalize.*``' Intrinsic
10849 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10856 declare float @llvm.canonicalize.f32(float %a)
10857 declare double @llvm.canonicalize.f64(double %b)
10862 The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
10863 encoding of a floating point number. This canonicalization is useful for
10864 implementing certain numeric primitives such as frexp. The canonical encoding is
10865 defined by IEEE-754-2008 to be:
10869 2.1.8 canonical encoding: The preferred encoding of a floating-point
10870 representation in a format. Applied to declets, significands of finite
10871 numbers, infinities, and NaNs, especially in decimal formats.
10873 This operation can also be considered equivalent to the IEEE-754-2008
10874 conversion of a floating-point value to the same format. NaNs are handled
10875 according to section 6.2.
10877 Examples of non-canonical encodings:
10879 - x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
10880 converted to a canonical representation per hardware-specific protocol.
10881 - Many normal decimal floating point numbers have non-canonical alternative
10883 - Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
10884 These are treated as non-canonical encodings of zero and with be flushed to
10885 a zero of the same sign by this operation.
10887 Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
10888 default exception handling must signal an invalid exception, and produce a
10891 This function should always be implementable as multiplication by 1.0, provided
10892 that the compiler does not constant fold the operation. Likewise, division by
10893 1.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
10894 -0.0 is also sufficient provided that the rounding mode is not -Infinity.
10896 ``@llvm.canonicalize`` must preserve the equality relation. That is:
10898 - ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
10899 - ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
10902 Additionally, the sign of zero must be conserved:
10903 ``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
10905 The payload bits of a NaN must be conserved, with two exceptions.
10906 First, environments which use only a single canonical representation of NaN
10907 must perform said canonicalization. Second, SNaNs must be quieted per the
10910 The canonicalization operation may be optimized away if:
10912 - The input is known to be canonical. For example, it was produced by a
10913 floating-point operation that is required by the standard to be canonical.
10914 - The result is consumed only by (or fused with) other floating-point
10915 operations. That is, the bits of the floating point value are not examined.
10917 '``llvm.fmuladd.*``' Intrinsic
10918 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10925 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
10926 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
10931 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
10932 expressions that can be fused if the code generator determines that (a) the
10933 target instruction set has support for a fused operation, and (b) that the
10934 fused operation is more efficient than the equivalent, separate pair of mul
10935 and add instructions.
10940 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
10941 multiplicands, a and b, and an addend c.
10950 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
10952 is equivalent to the expression a \* b + c, except that rounding will
10953 not be performed between the multiplication and addition steps if the
10954 code generator fuses the operations. Fusion is not guaranteed, even if
10955 the target platform supports it. If a fused multiply-add is required the
10956 corresponding llvm.fma.\* intrinsic function should be used
10957 instead. This never sets errno, just as '``llvm.fma.*``'.
10962 .. code-block:: llvm
10964 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
10967 '``llvm.uabsdiff.*``' and '``llvm.sabsdiff.*``' Intrinsics
10968 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10972 This is an overloaded intrinsic. The loaded data is a vector of any integer bit width.
10974 .. code-block:: llvm
10976 declare <4 x integer> @llvm.uabsdiff.v4i32(<4 x integer> %a, <4 x integer> %b)
10982 The ``llvm.uabsdiff`` intrinsic returns a vector result of the absolute difference
10983 of the two operands, treating them both as unsigned integers. The intermediate
10984 calculations are computed using infinitely precise unsigned arithmetic. The final
10985 result will be truncated to the given type.
10987 The ``llvm.sabsdiff`` intrinsic returns a vector result of the absolute difference of
10988 the two operands, treating them both as signed integers. If the result overflows, the
10989 behavior is undefined.
10993 These intrinsics are primarily used during the code generation stage of compilation.
10994 They are generated by compiler passes such as the Loop and SLP vectorizers. It is not
10995 recommended for users to create them manually.
11000 Both intrinsics take two integer of the same bitwidth.
11007 call <4 x i32> @llvm.uabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11011 %1 = zext <4 x i32> %a to <4 x i64>
11012 %2 = zext <4 x i32> %b to <4 x i64>
11013 %sub = sub <4 x i64> %1, %2
11014 %trunc = trunc <4 x i64> to <4 x i32>
11016 and the expression::
11018 call <4 x i32> @llvm.sabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11022 %sub = sub nsw <4 x i32> %a, %b
11023 %ispos = icmp sge <4 x i32> %sub, zeroinitializer
11024 %neg = sub nsw <4 x i32> zeroinitializer, %sub
11025 %1 = select <4 x i1> %ispos, <4 x i32> %sub, <4 x i32> %neg
11028 Half Precision Floating Point Intrinsics
11029 ----------------------------------------
11031 For most target platforms, half precision floating point is a
11032 storage-only format. This means that it is a dense encoding (in memory)
11033 but does not support computation in the format.
11035 This means that code must first load the half-precision floating point
11036 value as an i16, then convert it to float with
11037 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
11038 then be performed on the float value (including extending to double
11039 etc). To store the value back to memory, it is first converted to float
11040 if needed, then converted to i16 with
11041 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
11044 .. _int_convert_to_fp16:
11046 '``llvm.convert.to.fp16``' Intrinsic
11047 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11054 declare i16 @llvm.convert.to.fp16.f32(float %a)
11055 declare i16 @llvm.convert.to.fp16.f64(double %a)
11060 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11061 conventional floating point type to half precision floating point format.
11066 The intrinsic function contains single argument - the value to be
11072 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11073 conventional floating point format to half precision floating point format. The
11074 return value is an ``i16`` which contains the converted number.
11079 .. code-block:: llvm
11081 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
11082 store i16 %res, i16* @x, align 2
11084 .. _int_convert_from_fp16:
11086 '``llvm.convert.from.fp16``' Intrinsic
11087 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11094 declare float @llvm.convert.from.fp16.f32(i16 %a)
11095 declare double @llvm.convert.from.fp16.f64(i16 %a)
11100 The '``llvm.convert.from.fp16``' intrinsic function performs a
11101 conversion from half precision floating point format to single precision
11102 floating point format.
11107 The intrinsic function contains single argument - the value to be
11113 The '``llvm.convert.from.fp16``' intrinsic function performs a
11114 conversion from half single precision floating point format to single
11115 precision floating point format. The input half-float value is
11116 represented by an ``i16`` value.
11121 .. code-block:: llvm
11123 %a = load i16, i16* @x, align 2
11124 %res = call float @llvm.convert.from.fp16(i16 %a)
11126 .. _dbg_intrinsics:
11128 Debugger Intrinsics
11129 -------------------
11131 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
11132 prefix), are described in the `LLVM Source Level
11133 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
11136 Exception Handling Intrinsics
11137 -----------------------------
11139 The LLVM exception handling intrinsics (which all start with
11140 ``llvm.eh.`` prefix), are described in the `LLVM Exception
11141 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
11143 .. _int_trampoline:
11145 Trampoline Intrinsics
11146 ---------------------
11148 These intrinsics make it possible to excise one parameter, marked with
11149 the :ref:`nest <nest>` attribute, from a function. The result is a
11150 callable function pointer lacking the nest parameter - the caller does
11151 not need to provide a value for it. Instead, the value to use is stored
11152 in advance in a "trampoline", a block of memory usually allocated on the
11153 stack, which also contains code to splice the nest value into the
11154 argument list. This is used to implement the GCC nested function address
11157 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
11158 then the resulting function pointer has signature ``i32 (i32, i32)*``.
11159 It can be created as follows:
11161 .. code-block:: llvm
11163 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
11164 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
11165 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
11166 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
11167 %fp = bitcast i8* %p to i32 (i32, i32)*
11169 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
11170 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
11174 '``llvm.init.trampoline``' Intrinsic
11175 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11182 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
11187 This fills the memory pointed to by ``tramp`` with executable code,
11188 turning it into a trampoline.
11193 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
11194 pointers. The ``tramp`` argument must point to a sufficiently large and
11195 sufficiently aligned block of memory; this memory is written to by the
11196 intrinsic. Note that the size and the alignment are target-specific -
11197 LLVM currently provides no portable way of determining them, so a
11198 front-end that generates this intrinsic needs to have some
11199 target-specific knowledge. The ``func`` argument must hold a function
11200 bitcast to an ``i8*``.
11205 The block of memory pointed to by ``tramp`` is filled with target
11206 dependent code, turning it into a function. Then ``tramp`` needs to be
11207 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
11208 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
11209 function's signature is the same as that of ``func`` with any arguments
11210 marked with the ``nest`` attribute removed. At most one such ``nest``
11211 argument is allowed, and it must be of pointer type. Calling the new
11212 function is equivalent to calling ``func`` with the same argument list,
11213 but with ``nval`` used for the missing ``nest`` argument. If, after
11214 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
11215 modified, then the effect of any later call to the returned function
11216 pointer is undefined.
11220 '``llvm.adjust.trampoline``' Intrinsic
11221 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11228 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
11233 This performs any required machine-specific adjustment to the address of
11234 a trampoline (passed as ``tramp``).
11239 ``tramp`` must point to a block of memory which already has trampoline
11240 code filled in by a previous call to
11241 :ref:`llvm.init.trampoline <int_it>`.
11246 On some architectures the address of the code to be executed needs to be
11247 different than the address where the trampoline is actually stored. This
11248 intrinsic returns the executable address corresponding to ``tramp``
11249 after performing the required machine specific adjustments. The pointer
11250 returned can then be :ref:`bitcast and executed <int_trampoline>`.
11252 .. _int_mload_mstore:
11254 Masked Vector Load and Store Intrinsics
11255 ---------------------------------------
11257 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.
11261 '``llvm.masked.load.*``' Intrinsics
11262 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11266 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
11270 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11271 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11276 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.
11282 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.
11288 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.
11289 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.
11294 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
11296 ;; The result of the two following instructions is identical aside from potential memory access exception
11297 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
11298 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
11302 '``llvm.masked.store.*``' Intrinsics
11303 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11307 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
11311 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
11312 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
11317 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.
11322 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.
11328 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.
11329 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.
11333 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
11335 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
11336 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
11337 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
11338 store <16 x float> %res, <16 x float>* %ptr, align 4
11341 Masked Vector Gather and Scatter Intrinsics
11342 -------------------------------------------
11344 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.
11348 '``llvm.masked.gather.*``' Intrinsics
11349 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11353 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.
11357 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11358 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11363 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.
11369 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.
11375 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.
11376 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.
11381 %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>)
11383 ;; The gather with all-true mask is equivalent to the following instruction sequence
11384 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
11385 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
11386 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
11387 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
11389 %val0 = load double, double* %ptr0, align 8
11390 %val1 = load double, double* %ptr1, align 8
11391 %val2 = load double, double* %ptr2, align 8
11392 %val3 = load double, double* %ptr3, align 8
11394 %vec0 = insertelement <4 x double>undef, %val0, 0
11395 %vec01 = insertelement <4 x double>%vec0, %val1, 1
11396 %vec012 = insertelement <4 x double>%vec01, %val2, 2
11397 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
11401 '``llvm.masked.scatter.*``' Intrinsics
11402 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11406 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.
11410 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
11411 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
11416 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.
11421 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.
11427 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.
11431 ;; This instruction unconditionaly stores data vector in multiple addresses
11432 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
11434 ;; It is equivalent to a list of scalar stores
11435 %val0 = extractelement <8 x i32> %value, i32 0
11436 %val1 = extractelement <8 x i32> %value, i32 1
11438 %val7 = extractelement <8 x i32> %value, i32 7
11439 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
11440 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
11442 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
11443 ;; Note: the order of the following stores is important when they overlap:
11444 store i32 %val0, i32* %ptr0, align 4
11445 store i32 %val1, i32* %ptr1, align 4
11447 store i32 %val7, i32* %ptr7, align 4
11453 This class of intrinsics provides information about the lifetime of
11454 memory objects and ranges where variables are immutable.
11458 '``llvm.lifetime.start``' Intrinsic
11459 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11466 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
11471 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
11477 The first argument is a constant integer representing the size of the
11478 object, or -1 if it is variable sized. The second argument is a pointer
11484 This intrinsic indicates that before this point in the code, the value
11485 of the memory pointed to by ``ptr`` is dead. This means that it is known
11486 to never be used and has an undefined value. A load from the pointer
11487 that precedes this intrinsic can be replaced with ``'undef'``.
11491 '``llvm.lifetime.end``' Intrinsic
11492 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11499 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
11504 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
11510 The first argument is a constant integer representing the size of the
11511 object, or -1 if it is variable sized. The second argument is a pointer
11517 This intrinsic indicates that after this point in the code, the value of
11518 the memory pointed to by ``ptr`` is dead. This means that it is known to
11519 never be used and has an undefined value. Any stores into the memory
11520 object following this intrinsic may be removed as dead.
11522 '``llvm.invariant.start``' Intrinsic
11523 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11530 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
11535 The '``llvm.invariant.start``' intrinsic specifies that the contents of
11536 a memory object will not change.
11541 The first argument is a constant integer representing the size of the
11542 object, or -1 if it is variable sized. The second argument is a pointer
11548 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
11549 the return value, the referenced memory location is constant and
11552 '``llvm.invariant.end``' Intrinsic
11553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11560 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
11565 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
11566 memory object are mutable.
11571 The first argument is the matching ``llvm.invariant.start`` intrinsic.
11572 The second argument is a constant integer representing the size of the
11573 object, or -1 if it is variable sized and the third argument is a
11574 pointer to the object.
11579 This intrinsic indicates that the memory is mutable again.
11581 '``llvm.invariant.group.barrier``' Intrinsic
11582 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11589 declare i8* @llvm.invariant.group.barrier(i8* <ptr>)
11594 The '``llvm.invariant.group.barrier``' intrinsic can be used when an invariant
11595 established by invariant.group metadata no longer holds, to obtain a new pointer
11596 value that does not carry the invariant information.
11602 The ``llvm.invariant.group.barrier`` takes only one argument, which is
11603 the pointer to the memory for which the ``invariant.group`` no longer holds.
11608 Returns another pointer that aliases its argument but which is considered different
11609 for the purposes of ``load``/``store`` ``invariant.group`` metadata.
11614 This class of intrinsics is designed to be generic and has no specific
11617 '``llvm.var.annotation``' Intrinsic
11618 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11625 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11630 The '``llvm.var.annotation``' intrinsic.
11635 The first argument is a pointer to a value, the second is a pointer to a
11636 global string, the third is a pointer to a global string which is the
11637 source file name, and the last argument is the line number.
11642 This intrinsic allows annotation of local variables with arbitrary
11643 strings. This can be useful for special purpose optimizations that want
11644 to look for these annotations. These have no other defined use; they are
11645 ignored by code generation and optimization.
11647 '``llvm.ptr.annotation.*``' Intrinsic
11648 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11653 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
11654 pointer to an integer of any width. *NOTE* you must specify an address space for
11655 the pointer. The identifier for the default address space is the integer
11660 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11661 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
11662 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
11663 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
11664 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
11669 The '``llvm.ptr.annotation``' intrinsic.
11674 The first argument is a pointer to an integer value of arbitrary bitwidth
11675 (result of some expression), the second is a pointer to a global string, the
11676 third is a pointer to a global string which is the source file name, and the
11677 last argument is the line number. It returns the value of the first argument.
11682 This intrinsic allows annotation of a pointer to an integer with arbitrary
11683 strings. This can be useful for special purpose optimizations that want to look
11684 for these annotations. These have no other defined use; they are ignored by code
11685 generation and optimization.
11687 '``llvm.annotation.*``' Intrinsic
11688 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11693 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
11694 any integer bit width.
11698 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
11699 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
11700 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
11701 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
11702 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
11707 The '``llvm.annotation``' intrinsic.
11712 The first argument is an integer value (result of some expression), the
11713 second is a pointer to a global string, the third is a pointer to a
11714 global string which is the source file name, and the last argument is
11715 the line number. It returns the value of the first argument.
11720 This intrinsic allows annotations to be put on arbitrary expressions
11721 with arbitrary strings. This can be useful for special purpose
11722 optimizations that want to look for these annotations. These have no
11723 other defined use; they are ignored by code generation and optimization.
11725 '``llvm.trap``' Intrinsic
11726 ^^^^^^^^^^^^^^^^^^^^^^^^^
11733 declare void @llvm.trap() noreturn nounwind
11738 The '``llvm.trap``' intrinsic.
11748 This intrinsic is lowered to the target dependent trap instruction. If
11749 the target does not have a trap instruction, this intrinsic will be
11750 lowered to a call of the ``abort()`` function.
11752 '``llvm.debugtrap``' Intrinsic
11753 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11760 declare void @llvm.debugtrap() nounwind
11765 The '``llvm.debugtrap``' intrinsic.
11775 This intrinsic is lowered to code which is intended to cause an
11776 execution trap with the intention of requesting the attention of a
11779 '``llvm.stackprotector``' Intrinsic
11780 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11787 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
11792 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
11793 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
11794 is placed on the stack before local variables.
11799 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
11800 The first argument is the value loaded from the stack guard
11801 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
11802 enough space to hold the value of the guard.
11807 This intrinsic causes the prologue/epilogue inserter to force the position of
11808 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
11809 to ensure that if a local variable on the stack is overwritten, it will destroy
11810 the value of the guard. When the function exits, the guard on the stack is
11811 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
11812 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
11813 calling the ``__stack_chk_fail()`` function.
11815 '``llvm.stackprotectorcheck``' Intrinsic
11816 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11823 declare void @llvm.stackprotectorcheck(i8** <guard>)
11828 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
11829 created stack protector and if they are not equal calls the
11830 ``__stack_chk_fail()`` function.
11835 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
11836 the variable ``@__stack_chk_guard``.
11841 This intrinsic is provided to perform the stack protector check by comparing
11842 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
11843 values do not match call the ``__stack_chk_fail()`` function.
11845 The reason to provide this as an IR level intrinsic instead of implementing it
11846 via other IR operations is that in order to perform this operation at the IR
11847 level without an intrinsic, one would need to create additional basic blocks to
11848 handle the success/failure cases. This makes it difficult to stop the stack
11849 protector check from disrupting sibling tail calls in Codegen. With this
11850 intrinsic, we are able to generate the stack protector basic blocks late in
11851 codegen after the tail call decision has occurred.
11853 '``llvm.objectsize``' Intrinsic
11854 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11861 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
11862 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
11867 The ``llvm.objectsize`` intrinsic is designed to provide information to
11868 the optimizers to determine at compile time whether a) an operation
11869 (like memcpy) will overflow a buffer that corresponds to an object, or
11870 b) that a runtime check for overflow isn't necessary. An object in this
11871 context means an allocation of a specific class, structure, array, or
11877 The ``llvm.objectsize`` intrinsic takes two arguments. The first
11878 argument is a pointer to or into the ``object``. The second argument is
11879 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
11880 or -1 (if false) when the object size is unknown. The second argument
11881 only accepts constants.
11886 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
11887 the size of the object concerned. If the size cannot be determined at
11888 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
11889 on the ``min`` argument).
11891 '``llvm.expect``' Intrinsic
11892 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11897 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
11902 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
11903 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
11904 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
11909 The ``llvm.expect`` intrinsic provides information about expected (the
11910 most probable) value of ``val``, which can be used by optimizers.
11915 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
11916 a value. The second argument is an expected value, this needs to be a
11917 constant value, variables are not allowed.
11922 This intrinsic is lowered to the ``val``.
11926 '``llvm.assume``' Intrinsic
11927 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11934 declare void @llvm.assume(i1 %cond)
11939 The ``llvm.assume`` allows the optimizer to assume that the provided
11940 condition is true. This information can then be used in simplifying other parts
11946 The condition which the optimizer may assume is always true.
11951 The intrinsic allows the optimizer to assume that the provided condition is
11952 always true whenever the control flow reaches the intrinsic call. No code is
11953 generated for this intrinsic, and instructions that contribute only to the
11954 provided condition are not used for code generation. If the condition is
11955 violated during execution, the behavior is undefined.
11957 Note that the optimizer might limit the transformations performed on values
11958 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
11959 only used to form the intrinsic's input argument. This might prove undesirable
11960 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
11961 sufficient overall improvement in code quality. For this reason,
11962 ``llvm.assume`` should not be used to document basic mathematical invariants
11963 that the optimizer can otherwise deduce or facts that are of little use to the
11968 '``llvm.bitset.test``' Intrinsic
11969 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11976 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
11982 The first argument is a pointer to be tested. The second argument is a
11983 metadata object representing an identifier for a :doc:`bitset <BitSets>`.
11988 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
11989 member of the given bitset.
11991 '``llvm.donothing``' Intrinsic
11992 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11999 declare void @llvm.donothing() nounwind readnone
12004 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
12005 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
12006 with an invoke instruction.
12016 This intrinsic does nothing, and it's removed by optimizers and ignored
12019 Stack Map Intrinsics
12020 --------------------
12022 LLVM provides experimental intrinsics to support runtime patching
12023 mechanisms commonly desired in dynamic language JITs. These intrinsics
12024 are described in :doc:`StackMaps`.