1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][-a-zA-Z$._][-a-zA-Z$._0-9]*``'. Identifiers that require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves. The ``"\01"`` prefix
83 can be used on global variables to suppress mangling.
84 #. Unnamed values are represented as an unsigned numeric value with
85 their prefix. For example, ``%12``, ``@2``, ``%44``.
86 #. Constants, which are described in the section Constants_ below.
88 LLVM requires that values start with a prefix for two reasons: Compilers
89 don't need to worry about name clashes with reserved words, and the set
90 of reserved words may be expanded in the future without penalty.
91 Additionally, unnamed identifiers allow a compiler to quickly come up
92 with a temporary variable without having to avoid symbol table
95 Reserved words in LLVM are very similar to reserved words in other
96 languages. There are keywords for different opcodes ('``add``',
97 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
98 '``i32``', etc...), and others. These reserved words cannot conflict
99 with variable names, because none of them start with a prefix character
100 (``'%'`` or ``'@'``).
102 Here is an example of LLVM code to multiply the integer variable
109 %result = mul i32 %X, 8
111 After strength reduction:
115 %result = shl i32 %X, 3
121 %0 = add i32 %X, %X ; yields i32:%0
122 %1 = add i32 %0, %0 ; yields i32:%1
123 %result = add i32 %1, %1
125 This last way of multiplying ``%X`` by 8 illustrates several important
126 lexical features of LLVM:
128 #. Comments are delimited with a '``;``' and go until the end of line.
129 #. Unnamed temporaries are created when the result of a computation is
130 not assigned to a named value.
131 #. Unnamed temporaries are numbered sequentially (using a per-function
132 incrementing counter, starting with 0). Note that basic blocks and unnamed
133 function parameters are included in this numbering. For example, if the
134 entry basic block is not given a label name and all function parameters are
135 named, then it will get number 0.
137 It also shows a convention that we follow in this document. When
138 demonstrating instructions, we will follow an instruction with a comment
139 that defines the type and name of value produced.
147 LLVM programs are composed of ``Module``'s, each of which is a
148 translation unit of the input programs. Each module consists of
149 functions, global variables, and symbol table entries. Modules may be
150 combined together with the LLVM linker, which merges function (and
151 global variable) definitions, resolves forward declarations, and merges
152 symbol table entries. Here is an example of the "hello world" module:
156 ; Declare the string constant as a global constant.
157 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
159 ; External declaration of the puts function
160 declare i32 @puts(i8* nocapture) nounwind
162 ; Definition of main function
163 define i32 @main() { ; i32()*
164 ; Convert [13 x i8]* to i8 *...
165 %cast210 = getelementptr [13 x i8], [13 x i8]* @.str, i64 0, i64 0
167 ; Call puts function to write out the string to stdout.
168 call i32 @puts(i8* %cast210)
173 !0 = !{i32 42, null, !"string"}
176 This example is made up of a :ref:`global variable <globalvars>` named
177 "``.str``", an external declaration of the "``puts``" function, a
178 :ref:`function definition <functionstructure>` for "``main``" and
179 :ref:`named metadata <namedmetadatastructure>` "``foo``".
181 In general, a module is made up of a list of global values (where both
182 functions and global variables are global values). Global values are
183 represented by a pointer to a memory location (in this case, a pointer
184 to an array of char, and a pointer to a function), and have one of the
185 following :ref:`linkage types <linkage>`.
192 All Global Variables and Functions have one of the following types of
196 Global values with "``private``" linkage are only directly
197 accessible by objects in the current module. In particular, linking
198 code into a module with an private global value may cause the
199 private to be renamed as necessary to avoid collisions. Because the
200 symbol is private to the module, all references can be updated. This
201 doesn't show up in any symbol table in the object file.
203 Similar to private, but the value shows as a local symbol
204 (``STB_LOCAL`` in the case of ELF) in the object file. This
205 corresponds to the notion of the '``static``' keyword in C.
206 ``available_externally``
207 Globals with "``available_externally``" linkage are never emitted
208 into the object file corresponding to the LLVM module. They exist to
209 allow inlining and other optimizations to take place given knowledge
210 of the definition of the global, which is known to be somewhere
211 outside the module. Globals with ``available_externally`` linkage
212 are allowed to be discarded at will, and are otherwise the same as
213 ``linkonce_odr``. This linkage type is only allowed on definitions,
216 Globals with "``linkonce``" linkage are merged with other globals of
217 the same name when linkage occurs. This can be used to implement
218 some forms of inline functions, templates, or other code which must
219 be generated in each translation unit that uses it, but where the
220 body may be overridden with a more definitive definition later.
221 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
222 that ``linkonce`` linkage does not actually allow the optimizer to
223 inline the body of this function into callers because it doesn't
224 know if this definition of the function is the definitive definition
225 within the program or whether it will be overridden by a stronger
226 definition. To enable inlining and other optimizations, use
227 "``linkonce_odr``" linkage.
229 "``weak``" linkage has the same merging semantics as ``linkonce``
230 linkage, except that unreferenced globals with ``weak`` linkage may
231 not be discarded. This is used for globals that are declared "weak"
234 "``common``" linkage is most similar to "``weak``" linkage, but they
235 are used for tentative definitions in C, such as "``int X;``" at
236 global scope. Symbols with "``common``" linkage are merged in the
237 same way as ``weak symbols``, and they may not be deleted if
238 unreferenced. ``common`` symbols may not have an explicit section,
239 must have a zero initializer, and may not be marked
240 ':ref:`constant <globalvars>`'. Functions and aliases may not have
243 .. _linkage_appending:
246 "``appending``" linkage may only be applied to global variables of
247 pointer to array type. When two global variables with appending
248 linkage are linked together, the two global arrays are appended
249 together. This is the LLVM, typesafe, equivalent of having the
250 system linker append together "sections" with identical names when
253 The semantics of this linkage follow the ELF object file model: the
254 symbol is weak until linked, if not linked, the symbol becomes null
255 instead of being an undefined reference.
256 ``linkonce_odr``, ``weak_odr``
257 Some languages allow differing globals to be merged, such as two
258 functions with different semantics. Other languages, such as
259 ``C++``, ensure that only equivalent globals are ever merged (the
260 "one definition rule" --- "ODR"). Such languages can use the
261 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
262 global will only be merged with equivalent globals. These linkage
263 types are otherwise the same as their non-``odr`` versions.
265 If none of the above identifiers are used, the global is externally
266 visible, meaning that it participates in linkage and can be used to
267 resolve external symbol references.
269 It is illegal for a function *declaration* to have any linkage type
270 other than ``external`` or ``extern_weak``.
277 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
278 :ref:`invokes <i_invoke>` can all have an optional calling convention
279 specified for the call. The calling convention of any pair of dynamic
280 caller/callee must match, or the behavior of the program is undefined.
281 The following calling conventions are supported by LLVM, and more may be
284 "``ccc``" - The C calling convention
285 This calling convention (the default if no other calling convention
286 is specified) matches the target C calling conventions. This calling
287 convention supports varargs function calls and tolerates some
288 mismatch in the declared prototype and implemented declaration of
289 the function (as does normal C).
290 "``fastcc``" - The fast calling convention
291 This calling convention attempts to make calls as fast as possible
292 (e.g. by passing things in registers). This calling convention
293 allows the target to use whatever tricks it wants to produce fast
294 code for the target, without having to conform to an externally
295 specified ABI (Application Binary Interface). `Tail calls can only
296 be optimized when this, the GHC or the HiPE convention is
297 used. <CodeGenerator.html#id80>`_ This calling convention does not
298 support varargs and requires the prototype of all callees to exactly
299 match the prototype of the function definition.
300 "``coldcc``" - The cold calling convention
301 This calling convention attempts to make code in the caller as
302 efficient as possible under the assumption that the call is not
303 commonly executed. As such, these calls often preserve all registers
304 so that the call does not break any live ranges in the caller side.
305 This calling convention does not support varargs and requires the
306 prototype of all callees to exactly match the prototype of the
307 function definition. Furthermore the inliner doesn't consider such function
309 "``cc 10``" - GHC convention
310 This calling convention has been implemented specifically for use by
311 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
312 It passes everything in registers, going to extremes to achieve this
313 by disabling callee save registers. This calling convention should
314 not be used lightly but only for specific situations such as an
315 alternative to the *register pinning* performance technique often
316 used when implementing functional programming languages. At the
317 moment only X86 supports this convention and it has the following
320 - On *X86-32* only supports up to 4 bit type parameters. No
321 floating point types are supported.
322 - On *X86-64* only supports up to 10 bit type parameters and 6
323 floating point parameters.
325 This calling convention supports `tail call
326 optimization <CodeGenerator.html#id80>`_ but requires both the
327 caller and callee are using it.
328 "``cc 11``" - The HiPE calling convention
329 This calling convention has been implemented specifically for use by
330 the `High-Performance Erlang
331 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
332 native code compiler of the `Ericsson's Open Source Erlang/OTP
333 system <http://www.erlang.org/download.shtml>`_. It uses more
334 registers for argument passing than the ordinary C calling
335 convention and defines no callee-saved registers. The calling
336 convention properly supports `tail call
337 optimization <CodeGenerator.html#id80>`_ but requires that both the
338 caller and the callee use it. It uses a *register pinning*
339 mechanism, similar to GHC's convention, for keeping frequently
340 accessed runtime components pinned to specific hardware registers.
341 At the moment only X86 supports this convention (both 32 and 64
343 "``webkit_jscc``" - WebKit's JavaScript calling convention
344 This calling convention has been implemented for `WebKit FTL JIT
345 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
346 stack right to left (as cdecl does), and returns a value in the
347 platform's customary return register.
348 "``anyregcc``" - Dynamic calling convention for code patching
349 This is a special convention that supports patching an arbitrary code
350 sequence in place of a call site. This convention forces the call
351 arguments into registers but allows them to be dynamically
352 allocated. This can currently only be used with calls to
353 llvm.experimental.patchpoint because only this intrinsic records
354 the location of its arguments in a side table. See :doc:`StackMaps`.
355 "``preserve_mostcc``" - The `PreserveMost` calling convention
356 This calling convention attempts to make the code in the caller as
357 unintrusive as possible. This convention behaves identically to the `C`
358 calling convention on how arguments and return values are passed, but it
359 uses a different set of caller/callee-saved registers. This alleviates the
360 burden of saving and recovering a large register set before and after the
361 call in the caller. If the arguments are passed in callee-saved registers,
362 then they will be preserved by the callee across the call. This doesn't
363 apply for values returned in callee-saved registers.
365 - On X86-64 the callee preserves all general purpose registers, except for
366 R11. R11 can be used as a scratch register. Floating-point registers
367 (XMMs/YMMs) are not preserved and need to be saved by the caller.
369 The idea behind this convention is to support calls to runtime functions
370 that have a hot path and a cold path. The hot path is usually a small piece
371 of code that doesn't use many registers. The cold path might need to call out to
372 another function and therefore only needs to preserve the caller-saved
373 registers, which haven't already been saved by the caller. The
374 `PreserveMost` calling convention is very similar to the `cold` calling
375 convention in terms of caller/callee-saved registers, but they are used for
376 different types of function calls. `coldcc` is for function calls that are
377 rarely executed, whereas `preserve_mostcc` function calls are intended to be
378 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
379 doesn't prevent the inliner from inlining the function call.
381 This calling convention will be used by a future version of the ObjectiveC
382 runtime and should therefore still be considered experimental at this time.
383 Although this convention was created to optimize certain runtime calls to
384 the ObjectiveC runtime, it is not limited to this runtime and might be used
385 by other runtimes in the future too. The current implementation only
386 supports X86-64, but the intention is to support more architectures in the
388 "``preserve_allcc``" - The `PreserveAll` calling convention
389 This calling convention attempts to make the code in the caller even less
390 intrusive than the `PreserveMost` calling convention. This calling
391 convention also behaves identical to the `C` calling convention on how
392 arguments and return values are passed, but it uses a different set of
393 caller/callee-saved registers. This removes the burden of saving and
394 recovering a large register set before and after the call in the caller. If
395 the arguments are passed in callee-saved registers, then they will be
396 preserved by the callee across the call. This doesn't apply for values
397 returned in callee-saved registers.
399 - On X86-64 the callee preserves all general purpose registers, except for
400 R11. R11 can be used as a scratch register. Furthermore it also preserves
401 all floating-point registers (XMMs/YMMs).
403 The idea behind this convention is to support calls to runtime functions
404 that don't need to call out to any other functions.
406 This calling convention, like the `PreserveMost` calling convention, will be
407 used by a future version of the ObjectiveC runtime and should be considered
408 experimental at this time.
409 "``cc <n>``" - Numbered convention
410 Any calling convention may be specified by number, allowing
411 target-specific calling conventions to be used. Target specific
412 calling conventions start at 64.
414 More calling conventions can be added/defined on an as-needed basis, to
415 support Pascal conventions or any other well-known target-independent
418 .. _visibilitystyles:
423 All Global Variables and Functions have one of the following visibility
426 "``default``" - Default style
427 On targets that use the ELF object file format, default visibility
428 means that the declaration is visible to other modules and, in
429 shared libraries, means that the declared entity may be overridden.
430 On Darwin, default visibility means that the declaration is visible
431 to other modules. Default visibility corresponds to "external
432 linkage" in the language.
433 "``hidden``" - Hidden style
434 Two declarations of an object with hidden visibility refer to the
435 same object if they are in the same shared object. Usually, hidden
436 visibility indicates that the symbol will not be placed into the
437 dynamic symbol table, so no other module (executable or shared
438 library) can reference it directly.
439 "``protected``" - Protected style
440 On ELF, protected visibility indicates that the symbol will be
441 placed in the dynamic symbol table, but that references within the
442 defining module will bind to the local symbol. That is, the symbol
443 cannot be overridden by another module.
445 A symbol with ``internal`` or ``private`` linkage must have ``default``
453 All Global Variables, Functions and Aliases can have one of the following
457 "``dllimport``" causes the compiler to reference a function or variable via
458 a global pointer to a pointer that is set up by the DLL exporting the
459 symbol. On Microsoft Windows targets, the pointer name is formed by
460 combining ``__imp_`` and the function or variable name.
462 "``dllexport``" causes the compiler to provide a global pointer to a pointer
463 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
464 Microsoft Windows targets, the pointer name is formed by combining
465 ``__imp_`` and the function or variable name. Since this storage class
466 exists for defining a dll interface, the compiler, assembler and linker know
467 it is externally referenced and must refrain from deleting the symbol.
471 Thread Local Storage Models
472 ---------------------------
474 A variable may be defined as ``thread_local``, which means that it will
475 not be shared by threads (each thread will have a separated copy of the
476 variable). Not all targets support thread-local variables. Optionally, a
477 TLS model may be specified:
480 For variables that are only used within the current shared library.
482 For variables in modules that will not be loaded dynamically.
484 For variables defined in the executable and only used within it.
486 If no explicit model is given, the "general dynamic" model is used.
488 The models correspond to the ELF TLS models; see `ELF Handling For
489 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
490 more information on under which circumstances the different models may
491 be used. The target may choose a different TLS model if the specified
492 model is not supported, or if a better choice of model can be made.
494 A model can also be specified in an alias, but then it only governs how
495 the alias is accessed. It will not have any effect in the aliasee.
497 For platforms without linker support of ELF TLS model, the -femulated-tls
498 flag can be used to generate GCC compatible emulated TLS code.
505 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
506 types <t_struct>`. Literal types are uniqued structurally, but identified types
507 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
508 to forward declare a type that is not yet available.
510 An example of an identified structure specification is:
514 %mytype = type { %mytype*, i32 }
516 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
517 literal types are uniqued in recent versions of LLVM.
524 Global variables define regions of memory allocated at compilation time
527 Global variable definitions must be initialized.
529 Global variables in other translation units can also be declared, in which
530 case they don't have an initializer.
532 Either global variable definitions or declarations may have an explicit section
533 to be placed in and may have an optional explicit alignment specified.
535 A variable may be defined as a global ``constant``, which indicates that
536 the contents of the variable will **never** be modified (enabling better
537 optimization, allowing the global data to be placed in the read-only
538 section of an executable, etc). Note that variables that need runtime
539 initialization cannot be marked ``constant`` as there is a store to the
542 LLVM explicitly allows *declarations* of global variables to be marked
543 constant, even if the final definition of the global is not. This
544 capability can be used to enable slightly better optimization of the
545 program, but requires the language definition to guarantee that
546 optimizations based on the 'constantness' are valid for the translation
547 units that do not include the definition.
549 As SSA values, global variables define pointer values that are in scope
550 (i.e. they dominate) all basic blocks in the program. Global variables
551 always define a pointer to their "content" type because they describe a
552 region of memory, and all memory objects in LLVM are accessed through
555 Global variables can be marked with ``unnamed_addr`` which indicates
556 that the address is not significant, only the content. Constants marked
557 like this can be merged with other constants if they have the same
558 initializer. Note that a constant with significant address *can* be
559 merged with a ``unnamed_addr`` constant, the result being a constant
560 whose address is significant.
562 A global variable may be declared to reside in a target-specific
563 numbered address space. For targets that support them, address spaces
564 may affect how optimizations are performed and/or what target
565 instructions are used to access the variable. The default address space
566 is zero. The address space qualifier must precede any other attributes.
568 LLVM allows an explicit section to be specified for globals. If the
569 target supports it, it will emit globals to the section specified.
570 Additionally, the global can placed in a comdat if the target has the necessary
573 By default, global initializers are optimized by assuming that global
574 variables defined within the module are not modified from their
575 initial values before the start of the global initializer. This is
576 true even for variables potentially accessible from outside the
577 module, including those with external linkage or appearing in
578 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
579 by marking the variable with ``externally_initialized``.
581 An explicit alignment may be specified for a global, which must be a
582 power of 2. If not present, or if the alignment is set to zero, the
583 alignment of the global is set by the target to whatever it feels
584 convenient. If an explicit alignment is specified, the global is forced
585 to have exactly that alignment. Targets and optimizers are not allowed
586 to over-align the global if the global has an assigned section. In this
587 case, the extra alignment could be observable: for example, code could
588 assume that the globals are densely packed in their section and try to
589 iterate over them as an array, alignment padding would break this
590 iteration. The maximum alignment is ``1 << 29``.
592 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
594 Variables and aliases can have a
595 :ref:`Thread Local Storage Model <tls_model>`.
599 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
600 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
601 <global | constant> <Type> [<InitializerConstant>]
602 [, section "name"] [, comdat [($name)]]
603 [, align <Alignment>]
605 For example, the following defines a global in a numbered address space
606 with an initializer, section, and alignment:
610 @G = addrspace(5) constant float 1.0, section "foo", align 4
612 The following example just declares a global variable
616 @G = external global i32
618 The following example defines a thread-local global with the
619 ``initialexec`` TLS model:
623 @G = thread_local(initialexec) global i32 0, align 4
625 .. _functionstructure:
630 LLVM function definitions consist of the "``define``" keyword, an
631 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
632 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
633 an optional :ref:`calling convention <callingconv>`,
634 an optional ``unnamed_addr`` attribute, a return type, an optional
635 :ref:`parameter attribute <paramattrs>` for the return type, a function
636 name, a (possibly empty) argument list (each with optional :ref:`parameter
637 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
638 an optional section, an optional alignment,
639 an optional :ref:`comdat <langref_comdats>`,
640 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
641 an optional :ref:`prologue <prologuedata>`,
642 an optional :ref:`personality <personalityfn>`,
643 an opening curly brace, a list of basic blocks, and a closing curly brace.
645 LLVM function declarations consist of the "``declare``" keyword, an
646 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
647 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
648 an optional :ref:`calling convention <callingconv>`,
649 an optional ``unnamed_addr`` attribute, a return type, an optional
650 :ref:`parameter attribute <paramattrs>` for the return type, a function
651 name, a possibly empty list of arguments, an optional alignment, an optional
652 :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
653 and an optional :ref:`prologue <prologuedata>`.
655 A function definition contains a list of basic blocks, forming the CFG (Control
656 Flow Graph) for the function. Each basic block may optionally start with a label
657 (giving the basic block a symbol table entry), contains a list of instructions,
658 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
659 function return). If an explicit label is not provided, a block is assigned an
660 implicit numbered label, using the next value from the same counter as used for
661 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
662 entry block does not have an explicit label, it will be assigned label "%0",
663 then the first unnamed temporary in that block will be "%1", etc.
665 The first basic block in a function is special in two ways: it is
666 immediately executed on entrance to the function, and it is not allowed
667 to have predecessor basic blocks (i.e. there can not be any branches to
668 the entry block of a function). Because the block can have no
669 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
671 LLVM allows an explicit section to be specified for functions. If the
672 target supports it, it will emit functions to the section specified.
673 Additionally, the function can be placed in a COMDAT.
675 An explicit alignment may be specified for a function. If not present,
676 or if the alignment is set to zero, the alignment of the function is set
677 by the target to whatever it feels convenient. If an explicit alignment
678 is specified, the function is forced to have at least that much
679 alignment. All alignments must be a power of 2.
681 If the ``unnamed_addr`` attribute is given, the address is known to not
682 be significant and two identical functions can be merged.
686 define [linkage] [visibility] [DLLStorageClass]
688 <ResultType> @<FunctionName> ([argument list])
689 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
690 [align N] [gc] [prefix Constant] [prologue Constant]
691 [personality Constant] { ... }
693 The argument list is a comma separated sequence of arguments where each
694 argument is of the following form:
698 <type> [parameter Attrs] [name]
706 Aliases, unlike function or variables, don't create any new data. They
707 are just a new symbol and metadata for an existing position.
709 Aliases have a name and an aliasee that is either a global value or a
712 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
713 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
714 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
718 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy>, <AliaseeTy>* @<Aliasee>
720 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
721 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
722 might not correctly handle dropping a weak symbol that is aliased.
724 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
725 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
728 Since aliases are only a second name, some restrictions apply, of which
729 some can only be checked when producing an object file:
731 * The expression defining the aliasee must be computable at assembly
732 time. Since it is just a name, no relocations can be used.
734 * No alias in the expression can be weak as the possibility of the
735 intermediate alias being overridden cannot be represented in an
738 * No global value in the expression can be a declaration, since that
739 would require a relocation, which is not possible.
746 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
748 Comdats have a name which represents the COMDAT key. All global objects that
749 specify this key will only end up in the final object file if the linker chooses
750 that key over some other key. Aliases are placed in the same COMDAT that their
751 aliasee computes to, if any.
753 Comdats have a selection kind to provide input on how the linker should
754 choose between keys in two different object files.
758 $<Name> = comdat SelectionKind
760 The selection kind must be one of the following:
763 The linker may choose any COMDAT key, the choice is arbitrary.
765 The linker may choose any COMDAT key but the sections must contain the
768 The linker will choose the section containing the largest COMDAT key.
770 The linker requires that only section with this COMDAT key exist.
772 The linker may choose any COMDAT key but the sections must contain the
775 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
776 ``any`` as a selection kind.
778 Here is an example of a COMDAT group where a function will only be selected if
779 the COMDAT key's section is the largest:
783 $foo = comdat largest
784 @foo = global i32 2, comdat($foo)
786 define void @bar() comdat($foo) {
790 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
796 @foo = global i32 2, comdat
799 In a COFF object file, this will create a COMDAT section with selection kind
800 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
801 and another COMDAT section with selection kind
802 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
803 section and contains the contents of the ``@bar`` symbol.
805 There are some restrictions on the properties of the global object.
806 It, or an alias to it, must have the same name as the COMDAT group when
808 The contents and size of this object may be used during link-time to determine
809 which COMDAT groups get selected depending on the selection kind.
810 Because the name of the object must match the name of the COMDAT group, the
811 linkage of the global object must not be local; local symbols can get renamed
812 if a collision occurs in the symbol table.
814 The combined use of COMDATS and section attributes may yield surprising results.
821 @g1 = global i32 42, section "sec", comdat($foo)
822 @g2 = global i32 42, section "sec", comdat($bar)
824 From the object file perspective, this requires the creation of two sections
825 with the same name. This is necessary because both globals belong to different
826 COMDAT groups and COMDATs, at the object file level, are represented by
829 Note that certain IR constructs like global variables and functions may
830 create COMDATs in the object file in addition to any which are specified using
831 COMDAT IR. This arises when the code generator is configured to emit globals
832 in individual sections (e.g. when `-data-sections` or `-function-sections`
833 is supplied to `llc`).
835 .. _namedmetadatastructure:
840 Named metadata is a collection of metadata. :ref:`Metadata
841 nodes <metadata>` (but not metadata strings) are the only valid
842 operands for a named metadata.
844 #. Named metadata are represented as a string of characters with the
845 metadata prefix. The rules for metadata names are the same as for
846 identifiers, but quoted names are not allowed. ``"\xx"`` type escapes
847 are still valid, which allows any character to be part of a name.
851 ; Some unnamed metadata nodes, which are referenced by the named metadata.
856 !name = !{!0, !1, !2}
863 The return type and each parameter of a function type may have a set of
864 *parameter attributes* associated with them. Parameter attributes are
865 used to communicate additional information about the result or
866 parameters of a function. Parameter attributes are considered to be part
867 of the function, not of the function type, so functions with different
868 parameter attributes can have the same function type.
870 Parameter attributes are simple keywords that follow the type specified.
871 If multiple parameter attributes are needed, they are space separated.
876 declare i32 @printf(i8* noalias nocapture, ...)
877 declare i32 @atoi(i8 zeroext)
878 declare signext i8 @returns_signed_char()
880 Note that any attributes for the function result (``nounwind``,
881 ``readonly``) come immediately after the argument list.
883 Currently, only the following parameter attributes are defined:
886 This indicates to the code generator that the parameter or return
887 value should be zero-extended to the extent required by the target's
888 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
889 the caller (for a parameter) or the callee (for a return value).
891 This indicates to the code generator that the parameter or return
892 value should be sign-extended to the extent required by the target's
893 ABI (which is usually 32-bits) by the caller (for a parameter) or
894 the callee (for a return value).
896 This indicates that this parameter or return value should be treated
897 in a special target-dependent fashion while emitting code for
898 a function call or return (usually, by putting it in a register as
899 opposed to memory, though some targets use it to distinguish between
900 two different kinds of registers). Use of this attribute is
903 This indicates that the pointer parameter should really be passed by
904 value to the function. The attribute implies that a hidden copy of
905 the pointee is made between the caller and the callee, so the callee
906 is unable to modify the value in the caller. This attribute is only
907 valid on LLVM pointer arguments. It is generally used to pass
908 structs and arrays by value, but is also valid on pointers to
909 scalars. The copy is considered to belong to the caller not the
910 callee (for example, ``readonly`` functions should not write to
911 ``byval`` parameters). This is not a valid attribute for return
914 The byval attribute also supports specifying an alignment with the
915 align attribute. It indicates the alignment of the stack slot to
916 form and the known alignment of the pointer specified to the call
917 site. If the alignment is not specified, then the code generator
918 makes a target-specific assumption.
924 The ``inalloca`` argument attribute allows the caller to take the
925 address of outgoing stack arguments. An ``inalloca`` argument must
926 be a pointer to stack memory produced by an ``alloca`` instruction.
927 The alloca, or argument allocation, must also be tagged with the
928 inalloca keyword. Only the last argument may have the ``inalloca``
929 attribute, and that argument is guaranteed to be passed in memory.
931 An argument allocation may be used by a call at most once because
932 the call may deallocate it. The ``inalloca`` attribute cannot be
933 used in conjunction with other attributes that affect argument
934 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
935 ``inalloca`` attribute also disables LLVM's implicit lowering of
936 large aggregate return values, which means that frontend authors
937 must lower them with ``sret`` pointers.
939 When the call site is reached, the argument allocation must have
940 been the most recent stack allocation that is still live, or the
941 results are undefined. It is possible to allocate additional stack
942 space after an argument allocation and before its call site, but it
943 must be cleared off with :ref:`llvm.stackrestore
946 See :doc:`InAlloca` for more information on how to use this
950 This indicates that the pointer parameter specifies the address of a
951 structure that is the return value of the function in the source
952 program. This pointer must be guaranteed by the caller to be valid:
953 loads and stores to the structure may be assumed by the callee
954 not to trap and to be properly aligned. This may only be applied to
955 the first parameter. This is not a valid attribute for return
959 This indicates that the pointer value may be assumed by the optimizer to
960 have the specified alignment.
962 Note that this attribute has additional semantics when combined with the
968 This indicates that objects accessed via pointer values
969 :ref:`based <pointeraliasing>` on the argument or return value are not also
970 accessed, during the execution of the function, via pointer values not
971 *based* on the argument or return value. The attribute on a return value
972 also has additional semantics described below. The caller shares the
973 responsibility with the callee for ensuring that these requirements are met.
974 For further details, please see the discussion of the NoAlias response in
975 :ref:`alias analysis <Must, May, or No>`.
977 Note that this definition of ``noalias`` is intentionally similar
978 to the definition of ``restrict`` in C99 for function arguments.
980 For function return values, C99's ``restrict`` is not meaningful,
981 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
982 attribute on return values are stronger than the semantics of the attribute
983 when used on function arguments. On function return values, the ``noalias``
984 attribute indicates that the function acts like a system memory allocation
985 function, returning a pointer to allocated storage disjoint from the
986 storage for any other object accessible to the caller.
989 This indicates that the callee does not make any copies of the
990 pointer that outlive the callee itself. This is not a valid
991 attribute for return values.
996 This indicates that the pointer parameter can be excised using the
997 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
998 attribute for return values and can only be applied to one parameter.
1001 This indicates that the function always returns the argument as its return
1002 value. This is an optimization hint to the code generator when generating
1003 the caller, allowing tail call optimization and omission of register saves
1004 and restores in some cases; it is not checked or enforced when generating
1005 the callee. The parameter and the function return type must be valid
1006 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
1007 valid attribute for return values and can only be applied to one parameter.
1010 This indicates that the parameter or return pointer is not null. This
1011 attribute may only be applied to pointer typed parameters. This is not
1012 checked or enforced by LLVM, the caller must ensure that the pointer
1013 passed in is non-null, or the callee must ensure that the returned pointer
1016 ``dereferenceable(<n>)``
1017 This indicates that the parameter or return pointer is dereferenceable. This
1018 attribute may only be applied to pointer typed parameters. A pointer that
1019 is dereferenceable can be loaded from speculatively without a risk of
1020 trapping. The number of bytes known to be dereferenceable must be provided
1021 in parentheses. It is legal for the number of bytes to be less than the
1022 size of the pointee type. The ``nonnull`` attribute does not imply
1023 dereferenceability (consider a pointer to one element past the end of an
1024 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1025 ``addrspace(0)`` (which is the default address space).
1027 ``dereferenceable_or_null(<n>)``
1028 This indicates that the parameter or return value isn't both
1029 non-null and non-dereferenceable (up to ``<n>`` bytes) at the same
1030 time. All non-null pointers tagged with
1031 ``dereferenceable_or_null(<n>)`` are ``dereferenceable(<n>)``.
1032 For address space 0 ``dereferenceable_or_null(<n>)`` implies that
1033 a pointer is exactly one of ``dereferenceable(<n>)`` or ``null``,
1034 and in other address spaces ``dereferenceable_or_null(<n>)``
1035 implies that a pointer is at least one of ``dereferenceable(<n>)``
1036 or ``null`` (i.e. it may be both ``null`` and
1037 ``dereferenceable(<n>)``). This attribute may only be applied to
1038 pointer typed parameters.
1042 Garbage Collector Strategy Names
1043 --------------------------------
1045 Each function may specify a garbage collector strategy name, which is simply a
1048 .. code-block:: llvm
1050 define void @f() gc "name" { ... }
1052 The supported values of *name* includes those :ref:`built in to LLVM
1053 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1054 strategy will cause the compiler to alter its output in order to support the
1055 named garbage collection algorithm. Note that LLVM itself does not contain a
1056 garbage collector, this functionality is restricted to generating machine code
1057 which can interoperate with a collector provided externally.
1064 Prefix data is data associated with a function which the code
1065 generator will emit immediately before the function's entrypoint.
1066 The purpose of this feature is to allow frontends to associate
1067 language-specific runtime metadata with specific functions and make it
1068 available through the function pointer while still allowing the
1069 function pointer to be called.
1071 To access the data for a given function, a program may bitcast the
1072 function pointer to a pointer to the constant's type and dereference
1073 index -1. This implies that the IR symbol points just past the end of
1074 the prefix data. For instance, take the example of a function annotated
1075 with a single ``i32``,
1077 .. code-block:: llvm
1079 define void @f() prefix i32 123 { ... }
1081 The prefix data can be referenced as,
1083 .. code-block:: llvm
1085 %0 = bitcast void* () @f to i32*
1086 %a = getelementptr inbounds i32, i32* %0, i32 -1
1087 %b = load i32, i32* %a
1089 Prefix data is laid out as if it were an initializer for a global variable
1090 of the prefix data's type. The function will be placed such that the
1091 beginning of the prefix data is aligned. This means that if the size
1092 of the prefix data is not a multiple of the alignment size, the
1093 function's entrypoint will not be aligned. If alignment of the
1094 function's entrypoint is desired, padding must be added to the prefix
1097 A function may have prefix data but no body. This has similar semantics
1098 to the ``available_externally`` linkage in that the data may be used by the
1099 optimizers but will not be emitted in the object file.
1106 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1107 be inserted prior to the function body. This can be used for enabling
1108 function hot-patching and instrumentation.
1110 To maintain the semantics of ordinary function calls, the prologue data must
1111 have a particular format. Specifically, it must begin with a sequence of
1112 bytes which decode to a sequence of machine instructions, valid for the
1113 module's target, which transfer control to the point immediately succeeding
1114 the prologue data, without performing any other visible action. This allows
1115 the inliner and other passes to reason about the semantics of the function
1116 definition without needing to reason about the prologue data. Obviously this
1117 makes the format of the prologue data highly target dependent.
1119 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1120 which encodes the ``nop`` instruction:
1122 .. code-block:: llvm
1124 define void @f() prologue i8 144 { ... }
1126 Generally prologue data can be formed by encoding a relative branch instruction
1127 which skips the metadata, as in this example of valid prologue data for the
1128 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1130 .. code-block:: llvm
1132 %0 = type <{ i8, i8, i8* }>
1134 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1136 A function may have prologue data but no body. This has similar semantics
1137 to the ``available_externally`` linkage in that the data may be used by the
1138 optimizers but will not be emitted in the object file.
1142 Personality Function
1143 --------------------
1145 The ``personality`` attribute permits functions to specify what function
1146 to use for exception handling.
1153 Attribute groups are groups of attributes that are referenced by objects within
1154 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1155 functions will use the same set of attributes. In the degenerative case of a
1156 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1157 group will capture the important command line flags used to build that file.
1159 An attribute group is a module-level object. To use an attribute group, an
1160 object references the attribute group's ID (e.g. ``#37``). An object may refer
1161 to more than one attribute group. In that situation, the attributes from the
1162 different groups are merged.
1164 Here is an example of attribute groups for a function that should always be
1165 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1167 .. code-block:: llvm
1169 ; Target-independent attributes:
1170 attributes #0 = { alwaysinline alignstack=4 }
1172 ; Target-dependent attributes:
1173 attributes #1 = { "no-sse" }
1175 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1176 define void @f() #0 #1 { ... }
1183 Function attributes are set to communicate additional information about
1184 a function. Function attributes are considered to be part of the
1185 function, not of the function type, so functions with different function
1186 attributes can have the same function type.
1188 Function attributes are simple keywords that follow the type specified.
1189 If multiple attributes are needed, they are space separated. For
1192 .. code-block:: llvm
1194 define void @f() noinline { ... }
1195 define void @f() alwaysinline { ... }
1196 define void @f() alwaysinline optsize { ... }
1197 define void @f() optsize { ... }
1200 This attribute indicates that, when emitting the prologue and
1201 epilogue, the backend should forcibly align the stack pointer.
1202 Specify the desired alignment, which must be a power of two, in
1205 This attribute indicates that the inliner should attempt to inline
1206 this function into callers whenever possible, ignoring any active
1207 inlining size threshold for this caller.
1209 This indicates that the callee function at a call site should be
1210 recognized as a built-in function, even though the function's declaration
1211 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1212 direct calls to functions that are declared with the ``nobuiltin``
1215 This attribute indicates that this function is rarely called. When
1216 computing edge weights, basic blocks post-dominated by a cold
1217 function call are also considered to be cold; and, thus, given low
1220 This attribute indicates that the callee is dependent on a convergent
1221 thread execution pattern under certain parallel execution models.
1222 Transformations that are execution model agnostic may not make the execution
1223 of a convergent operation control dependent on any additional values.
1225 This attribute indicates that the source code contained a hint that
1226 inlining this function is desirable (such as the "inline" keyword in
1227 C/C++). It is just a hint; it imposes no requirements on the
1230 This attribute indicates that the function should be added to a
1231 jump-instruction table at code-generation time, and that all address-taken
1232 references to this function should be replaced with a reference to the
1233 appropriate jump-instruction-table function pointer. Note that this creates
1234 a new pointer for the original function, which means that code that depends
1235 on function-pointer identity can break. So, any function annotated with
1236 ``jumptable`` must also be ``unnamed_addr``.
1238 This attribute suggests that optimization passes and code generator
1239 passes make choices that keep the code size of this function as small
1240 as possible and perform optimizations that may sacrifice runtime
1241 performance in order to minimize the size of the generated code.
1243 This attribute disables prologue / epilogue emission for the
1244 function. This can have very system-specific consequences.
1246 This indicates that the callee function at a call site is not recognized as
1247 a built-in function. LLVM will retain the original call and not replace it
1248 with equivalent code based on the semantics of the built-in function, unless
1249 the call site uses the ``builtin`` attribute. This is valid at call sites
1250 and on function declarations and definitions.
1252 This attribute indicates that calls to the function cannot be
1253 duplicated. A call to a ``noduplicate`` function may be moved
1254 within its parent function, but may not be duplicated within
1255 its parent function.
1257 A function containing a ``noduplicate`` call may still
1258 be an inlining candidate, provided that the call is not
1259 duplicated by inlining. That implies that the function has
1260 internal linkage and only has one call site, so the original
1261 call is dead after inlining.
1263 This attributes disables implicit floating point instructions.
1265 This attribute indicates that the inliner should never inline this
1266 function in any situation. This attribute may not be used together
1267 with the ``alwaysinline`` attribute.
1269 This attribute suppresses lazy symbol binding for the function. This
1270 may make calls to the function faster, at the cost of extra program
1271 startup time if the function is not called during program startup.
1273 This attribute indicates that the code generator should not use a
1274 red zone, even if the target-specific ABI normally permits it.
1276 This function attribute indicates that the function never returns
1277 normally. This produces undefined behavior at runtime if the
1278 function ever does dynamically return.
1280 This function attribute indicates that the function never raises an
1281 exception. If the function does raise an exception, its runtime
1282 behavior is undefined. However, functions marked nounwind may still
1283 trap or generate asynchronous exceptions. Exception handling schemes
1284 that are recognized by LLVM to handle asynchronous exceptions, such
1285 as SEH, will still provide their implementation defined semantics.
1287 This function attribute indicates that the function is not optimized
1288 by any optimization or code generator passes with the
1289 exception of interprocedural optimization passes.
1290 This attribute cannot be used together with the ``alwaysinline``
1291 attribute; this attribute is also incompatible
1292 with the ``minsize`` attribute and the ``optsize`` attribute.
1294 This attribute requires the ``noinline`` attribute to be specified on
1295 the function as well, so the function is never inlined into any caller.
1296 Only functions with the ``alwaysinline`` attribute are valid
1297 candidates for inlining into the body of this function.
1299 This attribute suggests that optimization passes and code generator
1300 passes make choices that keep the code size of this function low,
1301 and otherwise do optimizations specifically to reduce code size as
1302 long as they do not significantly impact runtime performance.
1304 On a function, this attribute indicates that the function computes its
1305 result (or decides to unwind an exception) based strictly on its arguments,
1306 without dereferencing any pointer arguments or otherwise accessing
1307 any mutable state (e.g. memory, control registers, etc) visible to
1308 caller functions. It does not write through any pointer arguments
1309 (including ``byval`` arguments) and never changes any state visible
1310 to callers. This means that it cannot unwind exceptions by calling
1311 the ``C++`` exception throwing methods.
1313 On an argument, this attribute indicates that the function does not
1314 dereference that pointer argument, even though it may read or write the
1315 memory that the pointer points to if accessed through other pointers.
1317 On a function, this attribute indicates that the function does not write
1318 through any pointer arguments (including ``byval`` arguments) or otherwise
1319 modify any state (e.g. memory, control registers, etc) visible to
1320 caller functions. It may dereference pointer arguments and read
1321 state that may be set in the caller. A readonly function always
1322 returns the same value (or unwinds an exception identically) when
1323 called with the same set of arguments and global state. It cannot
1324 unwind an exception by calling the ``C++`` exception throwing
1327 On an argument, this attribute indicates that the function does not write
1328 through this pointer argument, even though it may write to the memory that
1329 the pointer points to.
1331 This attribute indicates that the only memory accesses inside function are
1332 loads and stores from objects pointed to by its pointer-typed arguments,
1333 with arbitrary offsets. Or in other words, all memory operations in the
1334 function can refer to memory only using pointers based on its function
1336 Note that ``argmemonly`` can be used together with ``readonly`` attribute
1337 in order to specify that function reads only from its arguments.
1339 This attribute indicates that this function can return twice. The C
1340 ``setjmp`` is an example of such a function. The compiler disables
1341 some optimizations (like tail calls) in the caller of these
1344 This attribute indicates that
1345 `SafeStack <http://clang.llvm.org/docs/SafeStack.html>`_
1346 protection is enabled for this function.
1348 If a function that has a ``safestack`` attribute is inlined into a
1349 function that doesn't have a ``safestack`` attribute or which has an
1350 ``ssp``, ``sspstrong`` or ``sspreq`` attribute, then the resulting
1351 function will have a ``safestack`` attribute.
1352 ``sanitize_address``
1353 This attribute indicates that AddressSanitizer checks
1354 (dynamic address safety analysis) are enabled for this function.
1356 This attribute indicates that MemorySanitizer checks (dynamic detection
1357 of accesses to uninitialized memory) are enabled for this function.
1359 This attribute indicates that ThreadSanitizer checks
1360 (dynamic thread safety analysis) are enabled for this function.
1362 This attribute indicates that the function should emit a stack
1363 smashing protector. It is in the form of a "canary" --- a random value
1364 placed on the stack before the local variables that's checked upon
1365 return from the function to see if it has been overwritten. A
1366 heuristic is used to determine if a function needs stack protectors
1367 or not. The heuristic used will enable protectors for functions with:
1369 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1370 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1371 - Calls to alloca() with variable sizes or constant sizes greater than
1372 ``ssp-buffer-size``.
1374 Variables that are identified as requiring a protector will be arranged
1375 on the stack such that they are adjacent to the stack protector guard.
1377 If a function that has an ``ssp`` attribute is inlined into a
1378 function that doesn't have an ``ssp`` attribute, then the resulting
1379 function will have an ``ssp`` attribute.
1381 This attribute indicates that the function should *always* emit a
1382 stack smashing protector. This overrides the ``ssp`` function
1385 Variables that are identified as requiring a protector will be arranged
1386 on the stack such that they are adjacent to the stack protector guard.
1387 The specific layout rules are:
1389 #. Large arrays and structures containing large arrays
1390 (``>= ssp-buffer-size``) are closest to the stack protector.
1391 #. Small arrays and structures containing small arrays
1392 (``< ssp-buffer-size``) are 2nd closest to the protector.
1393 #. Variables that have had their address taken are 3rd closest to the
1396 If a function that has an ``sspreq`` attribute is inlined into a
1397 function that doesn't have an ``sspreq`` attribute or which has an
1398 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1399 an ``sspreq`` attribute.
1401 This attribute indicates that the function should emit a stack smashing
1402 protector. This attribute causes a strong heuristic to be used when
1403 determining if a function needs stack protectors. The strong heuristic
1404 will enable protectors for functions with:
1406 - Arrays of any size and type
1407 - Aggregates containing an array of any size and type.
1408 - Calls to alloca().
1409 - Local variables that have had their address taken.
1411 Variables that are identified as requiring a protector will be arranged
1412 on the stack such that they are adjacent to the stack protector guard.
1413 The specific layout rules are:
1415 #. Large arrays and structures containing large arrays
1416 (``>= ssp-buffer-size``) are closest to the stack protector.
1417 #. Small arrays and structures containing small arrays
1418 (``< ssp-buffer-size``) are 2nd closest to the protector.
1419 #. Variables that have had their address taken are 3rd closest to the
1422 This overrides the ``ssp`` function attribute.
1424 If a function that has an ``sspstrong`` attribute is inlined into a
1425 function that doesn't have an ``sspstrong`` attribute, then the
1426 resulting function will have an ``sspstrong`` attribute.
1428 This attribute indicates that the function will delegate to some other
1429 function with a tail call. The prototype of a thunk should not be used for
1430 optimization purposes. The caller is expected to cast the thunk prototype to
1431 match the thunk target prototype.
1433 This attribute indicates that the ABI being targeted requires that
1434 an unwind table entry be produced for this function even if we can
1435 show that no exceptions passes by it. This is normally the case for
1436 the ELF x86-64 abi, but it can be disabled for some compilation
1445 Note: operand bundles are a work in progress, and they should be
1446 considered experimental at this time.
1448 Operand bundles are tagged sets of SSA values that can be associated
1449 with certain LLVM instructions (currently only ``call`` s and
1450 ``invoke`` s). In a way they are like metadata, but dropping them is
1451 incorrect and will change program semantics.
1455 operand bundle set ::= '[' operand bundle ']'
1456 operand bundle ::= tag '(' [ bundle operand ] (, bundle operand )* ')'
1457 bundle operand ::= SSA value
1458 tag ::= string constant
1460 Operand bundles are **not** part of a function's signature, and a
1461 given function may be called from multiple places with different kinds
1462 of operand bundles. This reflects the fact that the operand bundles
1463 are conceptually a part of the ``call`` (or ``invoke``), not the
1464 callee being dispatched to.
1466 Operand bundles are a generic mechanism intended to support
1467 runtime-introspection-like functionality for managed languages. While
1468 the exact semantics of an operand bundle depend on the bundle tag,
1469 there are certain limitations to how much the presence of an operand
1470 bundle can influence the semantics of a program. These restrictions
1471 are described as the semantics of an "unknown" operand bundle. As
1472 long as the behavior of an operand bundle is describable within these
1473 restrictions, LLVM does not need to have special knowledge of the
1474 operand bundle to not miscompile programs containing it.
1476 - The bundle operands for an unknown operand bundle escape in unknown
1477 ways before control is transferred to the callee or invokee.
1478 - Calls and invokes with operand bundles have unknown read / write
1479 effect on the heap on entry and exit (even if the call target is
1480 ``readnone`` or ``readonly``).
1481 - An operand bundle at a call site cannot change the implementation
1482 of the called function. Inter-procedural optimizations work as
1483 usual as long as they take into account the first two properties.
1484 - The bundle operands for an unknown operand bundle escape in unknown
1485 ways before control is transferred to the callee or invokee.
1486 - Calls and invokes with operand bundles have unknown read / write
1487 effect on the heap on entry and exit (even if the call target is
1488 ``readnone`` or ``readonly``), unless they're overriden with
1489 callsite specific attributes.
1490 - An operand bundle at a call site cannot change the implementation
1491 of the called function. Inter-procedural optimizations work as
1492 usual as long as they take into account the first two properties.
1496 Module-Level Inline Assembly
1497 ----------------------------
1499 Modules may contain "module-level inline asm" blocks, which corresponds
1500 to the GCC "file scope inline asm" blocks. These blocks are internally
1501 concatenated by LLVM and treated as a single unit, but may be separated
1502 in the ``.ll`` file if desired. The syntax is very simple:
1504 .. code-block:: llvm
1506 module asm "inline asm code goes here"
1507 module asm "more can go here"
1509 The strings can contain any character by escaping non-printable
1510 characters. The escape sequence used is simply "\\xx" where "xx" is the
1511 two digit hex code for the number.
1513 Note that the assembly string *must* be parseable by LLVM's integrated assembler
1514 (unless it is disabled), even when emitting a ``.s`` file.
1516 .. _langref_datalayout:
1521 A module may specify a target specific data layout string that specifies
1522 how data is to be laid out in memory. The syntax for the data layout is
1525 .. code-block:: llvm
1527 target datalayout = "layout specification"
1529 The *layout specification* consists of a list of specifications
1530 separated by the minus sign character ('-'). Each specification starts
1531 with a letter and may include other information after the letter to
1532 define some aspect of the data layout. The specifications accepted are
1536 Specifies that the target lays out data in big-endian form. That is,
1537 the bits with the most significance have the lowest address
1540 Specifies that the target lays out data in little-endian form. That
1541 is, the bits with the least significance have the lowest address
1544 Specifies the natural alignment of the stack in bits. Alignment
1545 promotion of stack variables is limited to the natural stack
1546 alignment to avoid dynamic stack realignment. The stack alignment
1547 must be a multiple of 8-bits. If omitted, the natural stack
1548 alignment defaults to "unspecified", which does not prevent any
1549 alignment promotions.
1550 ``p[n]:<size>:<abi>:<pref>``
1551 This specifies the *size* of a pointer and its ``<abi>`` and
1552 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1553 bits. The address space, ``n``, is optional, and if not specified,
1554 denotes the default address space 0. The value of ``n`` must be
1555 in the range [1,2^23).
1556 ``i<size>:<abi>:<pref>``
1557 This specifies the alignment for an integer type of a given bit
1558 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1559 ``v<size>:<abi>:<pref>``
1560 This specifies the alignment for a vector type of a given bit
1562 ``f<size>:<abi>:<pref>``
1563 This specifies the alignment for a floating point type of a given bit
1564 ``<size>``. Only values of ``<size>`` that are supported by the target
1565 will work. 32 (float) and 64 (double) are supported on all targets; 80
1566 or 128 (different flavors of long double) are also supported on some
1569 This specifies the alignment for an object of aggregate type.
1571 If present, specifies that llvm names are mangled in the output. The
1574 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1575 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1576 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1577 symbols get a ``_`` prefix.
1578 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1579 functions also get a suffix based on the frame size.
1580 ``n<size1>:<size2>:<size3>...``
1581 This specifies a set of native integer widths for the target CPU in
1582 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1583 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1584 this set are considered to support most general arithmetic operations
1587 On every specification that takes a ``<abi>:<pref>``, specifying the
1588 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1589 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1591 When constructing the data layout for a given target, LLVM starts with a
1592 default set of specifications which are then (possibly) overridden by
1593 the specifications in the ``datalayout`` keyword. The default
1594 specifications are given in this list:
1596 - ``E`` - big endian
1597 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1598 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1599 same as the default address space.
1600 - ``S0`` - natural stack alignment is unspecified
1601 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1602 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1603 - ``i16:16:16`` - i16 is 16-bit aligned
1604 - ``i32:32:32`` - i32 is 32-bit aligned
1605 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1606 alignment of 64-bits
1607 - ``f16:16:16`` - half is 16-bit aligned
1608 - ``f32:32:32`` - float is 32-bit aligned
1609 - ``f64:64:64`` - double is 64-bit aligned
1610 - ``f128:128:128`` - quad is 128-bit aligned
1611 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1612 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1613 - ``a:0:64`` - aggregates are 64-bit aligned
1615 When LLVM is determining the alignment for a given type, it uses the
1618 #. If the type sought is an exact match for one of the specifications,
1619 that specification is used.
1620 #. If no match is found, and the type sought is an integer type, then
1621 the smallest integer type that is larger than the bitwidth of the
1622 sought type is used. If none of the specifications are larger than
1623 the bitwidth then the largest integer type is used. For example,
1624 given the default specifications above, the i7 type will use the
1625 alignment of i8 (next largest) while both i65 and i256 will use the
1626 alignment of i64 (largest specified).
1627 #. If no match is found, and the type sought is a vector type, then the
1628 largest vector type that is smaller than the sought vector type will
1629 be used as a fall back. This happens because <128 x double> can be
1630 implemented in terms of 64 <2 x double>, for example.
1632 The function of the data layout string may not be what you expect.
1633 Notably, this is not a specification from the frontend of what alignment
1634 the code generator should use.
1636 Instead, if specified, the target data layout is required to match what
1637 the ultimate *code generator* expects. This string is used by the
1638 mid-level optimizers to improve code, and this only works if it matches
1639 what the ultimate code generator uses. There is no way to generate IR
1640 that does not embed this target-specific detail into the IR. If you
1641 don't specify the string, the default specifications will be used to
1642 generate a Data Layout and the optimization phases will operate
1643 accordingly and introduce target specificity into the IR with respect to
1644 these default specifications.
1651 A module may specify a target triple string that describes the target
1652 host. The syntax for the target triple is simply:
1654 .. code-block:: llvm
1656 target triple = "x86_64-apple-macosx10.7.0"
1658 The *target triple* string consists of a series of identifiers delimited
1659 by the minus sign character ('-'). The canonical forms are:
1663 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1664 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1666 This information is passed along to the backend so that it generates
1667 code for the proper architecture. It's possible to override this on the
1668 command line with the ``-mtriple`` command line option.
1670 .. _pointeraliasing:
1672 Pointer Aliasing Rules
1673 ----------------------
1675 Any memory access must be done through a pointer value associated with
1676 an address range of the memory access, otherwise the behavior is
1677 undefined. Pointer values are associated with address ranges according
1678 to the following rules:
1680 - A pointer value is associated with the addresses associated with any
1681 value it is *based* on.
1682 - An address of a global variable is associated with the address range
1683 of the variable's storage.
1684 - The result value of an allocation instruction is associated with the
1685 address range of the allocated storage.
1686 - A null pointer in the default address-space is associated with no
1688 - An integer constant other than zero or a pointer value returned from
1689 a function not defined within LLVM may be associated with address
1690 ranges allocated through mechanisms other than those provided by
1691 LLVM. Such ranges shall not overlap with any ranges of addresses
1692 allocated by mechanisms provided by LLVM.
1694 A pointer value is *based* on another pointer value according to the
1697 - A pointer value formed from a ``getelementptr`` operation is *based*
1698 on the first value operand of the ``getelementptr``.
1699 - The result value of a ``bitcast`` is *based* on the operand of the
1701 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1702 values that contribute (directly or indirectly) to the computation of
1703 the pointer's value.
1704 - The "*based* on" relationship is transitive.
1706 Note that this definition of *"based"* is intentionally similar to the
1707 definition of *"based"* in C99, though it is slightly weaker.
1709 LLVM IR does not associate types with memory. The result type of a
1710 ``load`` merely indicates the size and alignment of the memory from
1711 which to load, as well as the interpretation of the value. The first
1712 operand type of a ``store`` similarly only indicates the size and
1713 alignment of the store.
1715 Consequently, type-based alias analysis, aka TBAA, aka
1716 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1717 :ref:`Metadata <metadata>` may be used to encode additional information
1718 which specialized optimization passes may use to implement type-based
1723 Volatile Memory Accesses
1724 ------------------------
1726 Certain memory accesses, such as :ref:`load <i_load>`'s,
1727 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1728 marked ``volatile``. The optimizers must not change the number of
1729 volatile operations or change their order of execution relative to other
1730 volatile operations. The optimizers *may* change the order of volatile
1731 operations relative to non-volatile operations. This is not Java's
1732 "volatile" and has no cross-thread synchronization behavior.
1734 IR-level volatile loads and stores cannot safely be optimized into
1735 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1736 flagged volatile. Likewise, the backend should never split or merge
1737 target-legal volatile load/store instructions.
1739 .. admonition:: Rationale
1741 Platforms may rely on volatile loads and stores of natively supported
1742 data width to be executed as single instruction. For example, in C
1743 this holds for an l-value of volatile primitive type with native
1744 hardware support, but not necessarily for aggregate types. The
1745 frontend upholds these expectations, which are intentionally
1746 unspecified in the IR. The rules above ensure that IR transformations
1747 do not violate the frontend's contract with the language.
1751 Memory Model for Concurrent Operations
1752 --------------------------------------
1754 The LLVM IR does not define any way to start parallel threads of
1755 execution or to register signal handlers. Nonetheless, there are
1756 platform-specific ways to create them, and we define LLVM IR's behavior
1757 in their presence. This model is inspired by the C++0x memory model.
1759 For a more informal introduction to this model, see the :doc:`Atomics`.
1761 We define a *happens-before* partial order as the least partial order
1764 - Is a superset of single-thread program order, and
1765 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1766 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1767 techniques, like pthread locks, thread creation, thread joining,
1768 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1769 Constraints <ordering>`).
1771 Note that program order does not introduce *happens-before* edges
1772 between a thread and signals executing inside that thread.
1774 Every (defined) read operation (load instructions, memcpy, atomic
1775 loads/read-modify-writes, etc.) R reads a series of bytes written by
1776 (defined) write operations (store instructions, atomic
1777 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1778 section, initialized globals are considered to have a write of the
1779 initializer which is atomic and happens before any other read or write
1780 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1781 may see any write to the same byte, except:
1783 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1784 write\ :sub:`2` happens before R\ :sub:`byte`, then
1785 R\ :sub:`byte` does not see write\ :sub:`1`.
1786 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1787 R\ :sub:`byte` does not see write\ :sub:`3`.
1789 Given that definition, R\ :sub:`byte` is defined as follows:
1791 - If R is volatile, the result is target-dependent. (Volatile is
1792 supposed to give guarantees which can support ``sig_atomic_t`` in
1793 C/C++, and may be used for accesses to addresses that do not behave
1794 like normal memory. It does not generally provide cross-thread
1796 - Otherwise, if there is no write to the same byte that happens before
1797 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1798 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1799 R\ :sub:`byte` returns the value written by that write.
1800 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1801 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1802 Memory Ordering Constraints <ordering>` section for additional
1803 constraints on how the choice is made.
1804 - Otherwise R\ :sub:`byte` returns ``undef``.
1806 R returns the value composed of the series of bytes it read. This
1807 implies that some bytes within the value may be ``undef`` **without**
1808 the entire value being ``undef``. Note that this only defines the
1809 semantics of the operation; it doesn't mean that targets will emit more
1810 than one instruction to read the series of bytes.
1812 Note that in cases where none of the atomic intrinsics are used, this
1813 model places only one restriction on IR transformations on top of what
1814 is required for single-threaded execution: introducing a store to a byte
1815 which might not otherwise be stored is not allowed in general.
1816 (Specifically, in the case where another thread might write to and read
1817 from an address, introducing a store can change a load that may see
1818 exactly one write into a load that may see multiple writes.)
1822 Atomic Memory Ordering Constraints
1823 ----------------------------------
1825 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1826 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1827 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1828 ordering parameters that determine which other atomic instructions on
1829 the same address they *synchronize with*. These semantics are borrowed
1830 from Java and C++0x, but are somewhat more colloquial. If these
1831 descriptions aren't precise enough, check those specs (see spec
1832 references in the :doc:`atomics guide <Atomics>`).
1833 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1834 differently since they don't take an address. See that instruction's
1835 documentation for details.
1837 For a simpler introduction to the ordering constraints, see the
1841 The set of values that can be read is governed by the happens-before
1842 partial order. A value cannot be read unless some operation wrote
1843 it. This is intended to provide a guarantee strong enough to model
1844 Java's non-volatile shared variables. This ordering cannot be
1845 specified for read-modify-write operations; it is not strong enough
1846 to make them atomic in any interesting way.
1848 In addition to the guarantees of ``unordered``, there is a single
1849 total order for modifications by ``monotonic`` operations on each
1850 address. All modification orders must be compatible with the
1851 happens-before order. There is no guarantee that the modification
1852 orders can be combined to a global total order for the whole program
1853 (and this often will not be possible). The read in an atomic
1854 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1855 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1856 order immediately before the value it writes. If one atomic read
1857 happens before another atomic read of the same address, the later
1858 read must see the same value or a later value in the address's
1859 modification order. This disallows reordering of ``monotonic`` (or
1860 stronger) operations on the same address. If an address is written
1861 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1862 read that address repeatedly, the other threads must eventually see
1863 the write. This corresponds to the C++0x/C1x
1864 ``memory_order_relaxed``.
1866 In addition to the guarantees of ``monotonic``, a
1867 *synchronizes-with* edge may be formed with a ``release`` operation.
1868 This is intended to model C++'s ``memory_order_acquire``.
1870 In addition to the guarantees of ``monotonic``, if this operation
1871 writes a value which is subsequently read by an ``acquire``
1872 operation, it *synchronizes-with* that operation. (This isn't a
1873 complete description; see the C++0x definition of a release
1874 sequence.) This corresponds to the C++0x/C1x
1875 ``memory_order_release``.
1876 ``acq_rel`` (acquire+release)
1877 Acts as both an ``acquire`` and ``release`` operation on its
1878 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1879 ``seq_cst`` (sequentially consistent)
1880 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1881 operation that only reads, ``release`` for an operation that only
1882 writes), there is a global total order on all
1883 sequentially-consistent operations on all addresses, which is
1884 consistent with the *happens-before* partial order and with the
1885 modification orders of all the affected addresses. Each
1886 sequentially-consistent read sees the last preceding write to the
1887 same address in this global order. This corresponds to the C++0x/C1x
1888 ``memory_order_seq_cst`` and Java volatile.
1892 If an atomic operation is marked ``singlethread``, it only *synchronizes
1893 with* or participates in modification and seq\_cst total orderings with
1894 other operations running in the same thread (for example, in signal
1902 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1903 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1904 :ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) have the following flags that can
1905 be set to enable otherwise unsafe floating point operations
1908 No NaNs - Allow optimizations to assume the arguments and result are not
1909 NaN. Such optimizations are required to retain defined behavior over
1910 NaNs, but the value of the result is undefined.
1913 No Infs - Allow optimizations to assume the arguments and result are not
1914 +/-Inf. Such optimizations are required to retain defined behavior over
1915 +/-Inf, but the value of the result is undefined.
1918 No Signed Zeros - Allow optimizations to treat the sign of a zero
1919 argument or result as insignificant.
1922 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1923 argument rather than perform division.
1926 Fast - Allow algebraically equivalent transformations that may
1927 dramatically change results in floating point (e.g. reassociate). This
1928 flag implies all the others.
1932 Use-list Order Directives
1933 -------------------------
1935 Use-list directives encode the in-memory order of each use-list, allowing the
1936 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1937 indexes that are assigned to the referenced value's uses. The referenced
1938 value's use-list is immediately sorted by these indexes.
1940 Use-list directives may appear at function scope or global scope. They are not
1941 instructions, and have no effect on the semantics of the IR. When they're at
1942 function scope, they must appear after the terminator of the final basic block.
1944 If basic blocks have their address taken via ``blockaddress()`` expressions,
1945 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1952 uselistorder <ty> <value>, { <order-indexes> }
1953 uselistorder_bb @function, %block { <order-indexes> }
1959 define void @foo(i32 %arg1, i32 %arg2) {
1961 ; ... instructions ...
1963 ; ... instructions ...
1965 ; At function scope.
1966 uselistorder i32 %arg1, { 1, 0, 2 }
1967 uselistorder label %bb, { 1, 0 }
1971 uselistorder i32* @global, { 1, 2, 0 }
1972 uselistorder i32 7, { 1, 0 }
1973 uselistorder i32 (i32) @bar, { 1, 0 }
1974 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1981 The LLVM type system is one of the most important features of the
1982 intermediate representation. Being typed enables a number of
1983 optimizations to be performed on the intermediate representation
1984 directly, without having to do extra analyses on the side before the
1985 transformation. A strong type system makes it easier to read the
1986 generated code and enables novel analyses and transformations that are
1987 not feasible to perform on normal three address code representations.
1997 The void type does not represent any value and has no size.
2015 The function type can be thought of as a function signature. It consists of a
2016 return type and a list of formal parameter types. The return type of a function
2017 type is a void type or first class type --- except for :ref:`label <t_label>`
2018 and :ref:`metadata <t_metadata>` types.
2024 <returntype> (<parameter list>)
2026 ...where '``<parameter list>``' is a comma-separated list of type
2027 specifiers. Optionally, the parameter list may include a type ``...``, which
2028 indicates that the function takes a variable number of arguments. Variable
2029 argument functions can access their arguments with the :ref:`variable argument
2030 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
2031 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
2035 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2036 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
2037 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2038 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
2039 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2040 | ``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. |
2041 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2042 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
2043 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2050 The :ref:`first class <t_firstclass>` types are perhaps the most important.
2051 Values of these types are the only ones which can be produced by
2059 These are the types that are valid in registers from CodeGen's perspective.
2068 The integer type is a very simple type that simply specifies an
2069 arbitrary bit width for the integer type desired. Any bit width from 1
2070 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2078 The number of bits the integer will occupy is specified by the ``N``
2084 +----------------+------------------------------------------------+
2085 | ``i1`` | a single-bit integer. |
2086 +----------------+------------------------------------------------+
2087 | ``i32`` | a 32-bit integer. |
2088 +----------------+------------------------------------------------+
2089 | ``i1942652`` | a really big integer of over 1 million bits. |
2090 +----------------+------------------------------------------------+
2094 Floating Point Types
2095 """"""""""""""""""""
2104 - 16-bit floating point value
2107 - 32-bit floating point value
2110 - 64-bit floating point value
2113 - 128-bit floating point value (112-bit mantissa)
2116 - 80-bit floating point value (X87)
2119 - 128-bit floating point value (two 64-bits)
2126 The x86_mmx type represents a value held in an MMX register on an x86
2127 machine. The operations allowed on it are quite limited: parameters and
2128 return values, load and store, and bitcast. User-specified MMX
2129 instructions are represented as intrinsic or asm calls with arguments
2130 and/or results of this type. There are no arrays, vectors or constants
2147 The pointer type is used to specify memory locations. Pointers are
2148 commonly used to reference objects in memory.
2150 Pointer types may have an optional address space attribute defining the
2151 numbered address space where the pointed-to object resides. The default
2152 address space is number zero. The semantics of non-zero address spaces
2153 are target-specific.
2155 Note that LLVM does not permit pointers to void (``void*``) nor does it
2156 permit pointers to labels (``label*``). Use ``i8*`` instead.
2166 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2167 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2168 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2169 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2170 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2171 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2172 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2181 A vector type is a simple derived type that represents a vector of
2182 elements. Vector types are used when multiple primitive data are
2183 operated in parallel using a single instruction (SIMD). A vector type
2184 requires a size (number of elements) and an underlying primitive data
2185 type. Vector types are considered :ref:`first class <t_firstclass>`.
2191 < <# elements> x <elementtype> >
2193 The number of elements is a constant integer value larger than 0;
2194 elementtype may be any integer, floating point or pointer type. Vectors
2195 of size zero are not allowed.
2199 +-------------------+--------------------------------------------------+
2200 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2201 +-------------------+--------------------------------------------------+
2202 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2203 +-------------------+--------------------------------------------------+
2204 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2205 +-------------------+--------------------------------------------------+
2206 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2207 +-------------------+--------------------------------------------------+
2216 The label type represents code labels.
2231 The token type is used when a value is associated with an instruction
2232 but all uses of the value must not attempt to introspect or obscure it.
2233 As such, it is not appropriate to have a :ref:`phi <i_phi>` or
2234 :ref:`select <i_select>` of type token.
2251 The metadata type represents embedded metadata. No derived types may be
2252 created from metadata except for :ref:`function <t_function>` arguments.
2265 Aggregate Types are a subset of derived types that can contain multiple
2266 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2267 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2277 The array type is a very simple derived type that arranges elements
2278 sequentially in memory. The array type requires a size (number of
2279 elements) and an underlying data type.
2285 [<# elements> x <elementtype>]
2287 The number of elements is a constant integer value; ``elementtype`` may
2288 be any type with a size.
2292 +------------------+--------------------------------------+
2293 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2294 +------------------+--------------------------------------+
2295 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2296 +------------------+--------------------------------------+
2297 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2298 +------------------+--------------------------------------+
2300 Here are some examples of multidimensional arrays:
2302 +-----------------------------+----------------------------------------------------------+
2303 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2304 +-----------------------------+----------------------------------------------------------+
2305 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2306 +-----------------------------+----------------------------------------------------------+
2307 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2308 +-----------------------------+----------------------------------------------------------+
2310 There is no restriction on indexing beyond the end of the array implied
2311 by a static type (though there are restrictions on indexing beyond the
2312 bounds of an allocated object in some cases). This means that
2313 single-dimension 'variable sized array' addressing can be implemented in
2314 LLVM with a zero length array type. An implementation of 'pascal style
2315 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2325 The structure type is used to represent a collection of data members
2326 together in memory. The elements of a structure may be any type that has
2329 Structures in memory are accessed using '``load``' and '``store``' by
2330 getting a pointer to a field with the '``getelementptr``' instruction.
2331 Structures in registers are accessed using the '``extractvalue``' and
2332 '``insertvalue``' instructions.
2334 Structures may optionally be "packed" structures, which indicate that
2335 the alignment of the struct is one byte, and that there is no padding
2336 between the elements. In non-packed structs, padding between field types
2337 is inserted as defined by the DataLayout string in the module, which is
2338 required to match what the underlying code generator expects.
2340 Structures can either be "literal" or "identified". A literal structure
2341 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2342 identified types are always defined at the top level with a name.
2343 Literal types are uniqued by their contents and can never be recursive
2344 or opaque since there is no way to write one. Identified types can be
2345 recursive, can be opaqued, and are never uniqued.
2351 %T1 = type { <type list> } ; Identified normal struct type
2352 %T2 = type <{ <type list> }> ; Identified packed struct type
2356 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2357 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2358 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2359 | ``{ 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``. |
2360 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2361 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2362 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2366 Opaque Structure Types
2367 """"""""""""""""""""""
2371 Opaque structure types are used to represent named structure types that
2372 do not have a body specified. This corresponds (for example) to the C
2373 notion of a forward declared structure.
2384 +--------------+-------------------+
2385 | ``opaque`` | An opaque type. |
2386 +--------------+-------------------+
2393 LLVM has several different basic types of constants. This section
2394 describes them all and their syntax.
2399 **Boolean constants**
2400 The two strings '``true``' and '``false``' are both valid constants
2402 **Integer constants**
2403 Standard integers (such as '4') are constants of the
2404 :ref:`integer <t_integer>` type. Negative numbers may be used with
2406 **Floating point constants**
2407 Floating point constants use standard decimal notation (e.g.
2408 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2409 hexadecimal notation (see below). The assembler requires the exact
2410 decimal value of a floating-point constant. For example, the
2411 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2412 decimal in binary. Floating point constants must have a :ref:`floating
2413 point <t_floating>` type.
2414 **Null pointer constants**
2415 The identifier '``null``' is recognized as a null pointer constant
2416 and must be of :ref:`pointer type <t_pointer>`.
2418 The one non-intuitive notation for constants is the hexadecimal form of
2419 floating point constants. For example, the form
2420 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2421 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2422 constants are required (and the only time that they are generated by the
2423 disassembler) is when a floating point constant must be emitted but it
2424 cannot be represented as a decimal floating point number in a reasonable
2425 number of digits. For example, NaN's, infinities, and other special
2426 values are represented in their IEEE hexadecimal format so that assembly
2427 and disassembly do not cause any bits to change in the constants.
2429 When using the hexadecimal form, constants of types half, float, and
2430 double are represented using the 16-digit form shown above (which
2431 matches the IEEE754 representation for double); half and float values
2432 must, however, be exactly representable as IEEE 754 half and single
2433 precision, respectively. Hexadecimal format is always used for long
2434 double, and there are three forms of long double. The 80-bit format used
2435 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2436 128-bit format used by PowerPC (two adjacent doubles) is represented by
2437 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2438 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2439 will only work if they match the long double format on your target.
2440 The IEEE 16-bit format (half precision) is represented by ``0xH``
2441 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2442 (sign bit at the left).
2444 There are no constants of type x86_mmx.
2446 .. _complexconstants:
2451 Complex constants are a (potentially recursive) combination of simple
2452 constants and smaller complex constants.
2454 **Structure constants**
2455 Structure constants are represented with notation similar to
2456 structure type definitions (a comma separated list of elements,
2457 surrounded by braces (``{}``)). For example:
2458 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2459 "``@G = external global i32``". Structure constants must have
2460 :ref:`structure type <t_struct>`, and the number and types of elements
2461 must match those specified by the type.
2463 Array constants are represented with notation similar to array type
2464 definitions (a comma separated list of elements, surrounded by
2465 square brackets (``[]``)). For example:
2466 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2467 :ref:`array type <t_array>`, and the number and types of elements must
2468 match those specified by the type. As a special case, character array
2469 constants may also be represented as a double-quoted string using the ``c``
2470 prefix. For example: "``c"Hello World\0A\00"``".
2471 **Vector constants**
2472 Vector constants are represented with notation similar to vector
2473 type definitions (a comma separated list of elements, surrounded by
2474 less-than/greater-than's (``<>``)). For example:
2475 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2476 must have :ref:`vector type <t_vector>`, and the number and types of
2477 elements must match those specified by the type.
2478 **Zero initialization**
2479 The string '``zeroinitializer``' can be used to zero initialize a
2480 value to zero of *any* type, including scalar and
2481 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2482 having to print large zero initializers (e.g. for large arrays) and
2483 is always exactly equivalent to using explicit zero initializers.
2485 A metadata node is a constant tuple without types. For example:
2486 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2487 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2488 Unlike other typed constants that are meant to be interpreted as part of
2489 the instruction stream, metadata is a place to attach additional
2490 information such as debug info.
2492 Global Variable and Function Addresses
2493 --------------------------------------
2495 The addresses of :ref:`global variables <globalvars>` and
2496 :ref:`functions <functionstructure>` are always implicitly valid
2497 (link-time) constants. These constants are explicitly referenced when
2498 the :ref:`identifier for the global <identifiers>` is used and always have
2499 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2502 .. code-block:: llvm
2506 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2513 The string '``undef``' can be used anywhere a constant is expected, and
2514 indicates that the user of the value may receive an unspecified
2515 bit-pattern. Undefined values may be of any type (other than '``label``'
2516 or '``void``') and be used anywhere a constant is permitted.
2518 Undefined values are useful because they indicate to the compiler that
2519 the program is well defined no matter what value is used. This gives the
2520 compiler more freedom to optimize. Here are some examples of
2521 (potentially surprising) transformations that are valid (in pseudo IR):
2523 .. code-block:: llvm
2533 This is safe because all of the output bits are affected by the undef
2534 bits. Any output bit can have a zero or one depending on the input bits.
2536 .. code-block:: llvm
2547 These logical operations have bits that are not always affected by the
2548 input. For example, if ``%X`` has a zero bit, then the output of the
2549 '``and``' operation will always be a zero for that bit, no matter what
2550 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2551 optimize or assume that the result of the '``and``' is '``undef``'.
2552 However, it is safe to assume that all bits of the '``undef``' could be
2553 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2554 all the bits of the '``undef``' operand to the '``or``' could be set,
2555 allowing the '``or``' to be folded to -1.
2557 .. code-block:: llvm
2559 %A = select undef, %X, %Y
2560 %B = select undef, 42, %Y
2561 %C = select %X, %Y, undef
2571 This set of examples shows that undefined '``select``' (and conditional
2572 branch) conditions can go *either way*, but they have to come from one
2573 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2574 both known to have a clear low bit, then ``%A`` would have to have a
2575 cleared low bit. However, in the ``%C`` example, the optimizer is
2576 allowed to assume that the '``undef``' operand could be the same as
2577 ``%Y``, allowing the whole '``select``' to be eliminated.
2579 .. code-block:: llvm
2581 %A = xor undef, undef
2598 This example points out that two '``undef``' operands are not
2599 necessarily the same. This can be surprising to people (and also matches
2600 C semantics) where they assume that "``X^X``" is always zero, even if
2601 ``X`` is undefined. This isn't true for a number of reasons, but the
2602 short answer is that an '``undef``' "variable" can arbitrarily change
2603 its value over its "live range". This is true because the variable
2604 doesn't actually *have a live range*. Instead, the value is logically
2605 read from arbitrary registers that happen to be around when needed, so
2606 the value is not necessarily consistent over time. In fact, ``%A`` and
2607 ``%C`` need to have the same semantics or the core LLVM "replace all
2608 uses with" concept would not hold.
2610 .. code-block:: llvm
2618 These examples show the crucial difference between an *undefined value*
2619 and *undefined behavior*. An undefined value (like '``undef``') is
2620 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2621 operation can be constant folded to '``undef``', because the '``undef``'
2622 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2623 However, in the second example, we can make a more aggressive
2624 assumption: because the ``undef`` is allowed to be an arbitrary value,
2625 we are allowed to assume that it could be zero. Since a divide by zero
2626 has *undefined behavior*, we are allowed to assume that the operation
2627 does not execute at all. This allows us to delete the divide and all
2628 code after it. Because the undefined operation "can't happen", the
2629 optimizer can assume that it occurs in dead code.
2631 .. code-block:: llvm
2633 a: store undef -> %X
2634 b: store %X -> undef
2639 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2640 value can be assumed to not have any effect; we can assume that the
2641 value is overwritten with bits that happen to match what was already
2642 there. However, a store *to* an undefined location could clobber
2643 arbitrary memory, therefore, it has undefined behavior.
2650 Poison values are similar to :ref:`undef values <undefvalues>`, however
2651 they also represent the fact that an instruction or constant expression
2652 that cannot evoke side effects has nevertheless detected a condition
2653 that results in undefined behavior.
2655 There is currently no way of representing a poison value in the IR; they
2656 only exist when produced by operations such as :ref:`add <i_add>` with
2659 Poison value behavior is defined in terms of value *dependence*:
2661 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2662 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2663 their dynamic predecessor basic block.
2664 - Function arguments depend on the corresponding actual argument values
2665 in the dynamic callers of their functions.
2666 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2667 instructions that dynamically transfer control back to them.
2668 - :ref:`Invoke <i_invoke>` instructions depend on the
2669 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2670 call instructions that dynamically transfer control back to them.
2671 - Non-volatile loads and stores depend on the most recent stores to all
2672 of the referenced memory addresses, following the order in the IR
2673 (including loads and stores implied by intrinsics such as
2674 :ref:`@llvm.memcpy <int_memcpy>`.)
2675 - An instruction with externally visible side effects depends on the
2676 most recent preceding instruction with externally visible side
2677 effects, following the order in the IR. (This includes :ref:`volatile
2678 operations <volatile>`.)
2679 - An instruction *control-depends* on a :ref:`terminator
2680 instruction <terminators>` if the terminator instruction has
2681 multiple successors and the instruction is always executed when
2682 control transfers to one of the successors, and may not be executed
2683 when control is transferred to another.
2684 - Additionally, an instruction also *control-depends* on a terminator
2685 instruction if the set of instructions it otherwise depends on would
2686 be different if the terminator had transferred control to a different
2688 - Dependence is transitive.
2690 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2691 with the additional effect that any instruction that has a *dependence*
2692 on a poison value has undefined behavior.
2694 Here are some examples:
2696 .. code-block:: llvm
2699 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2700 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2701 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2702 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2704 store i32 %poison, i32* @g ; Poison value stored to memory.
2705 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2707 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2709 %narrowaddr = bitcast i32* @g to i16*
2710 %wideaddr = bitcast i32* @g to i64*
2711 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2712 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2714 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2715 br i1 %cmp, label %true, label %end ; Branch to either destination.
2718 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2719 ; it has undefined behavior.
2723 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2724 ; Both edges into this PHI are
2725 ; control-dependent on %cmp, so this
2726 ; always results in a poison value.
2728 store volatile i32 0, i32* @g ; This would depend on the store in %true
2729 ; if %cmp is true, or the store in %entry
2730 ; otherwise, so this is undefined behavior.
2732 br i1 %cmp, label %second_true, label %second_end
2733 ; The same branch again, but this time the
2734 ; true block doesn't have side effects.
2741 store volatile i32 0, i32* @g ; This time, the instruction always depends
2742 ; on the store in %end. Also, it is
2743 ; control-equivalent to %end, so this is
2744 ; well-defined (ignoring earlier undefined
2745 ; behavior in this example).
2749 Addresses of Basic Blocks
2750 -------------------------
2752 ``blockaddress(@function, %block)``
2754 The '``blockaddress``' constant computes the address of the specified
2755 basic block in the specified function, and always has an ``i8*`` type.
2756 Taking the address of the entry block is illegal.
2758 This value only has defined behavior when used as an operand to the
2759 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2760 against null. Pointer equality tests between labels addresses results in
2761 undefined behavior --- though, again, comparison against null is ok, and
2762 no label is equal to the null pointer. This may be passed around as an
2763 opaque pointer sized value as long as the bits are not inspected. This
2764 allows ``ptrtoint`` and arithmetic to be performed on these values so
2765 long as the original value is reconstituted before the ``indirectbr``
2768 Finally, some targets may provide defined semantics when using the value
2769 as the operand to an inline assembly, but that is target specific.
2773 Constant Expressions
2774 --------------------
2776 Constant expressions are used to allow expressions involving other
2777 constants to be used as constants. Constant expressions may be of any
2778 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2779 that does not have side effects (e.g. load and call are not supported).
2780 The following is the syntax for constant expressions:
2782 ``trunc (CST to TYPE)``
2783 Truncate a constant to another type. The bit size of CST must be
2784 larger than the bit size of TYPE. Both types must be integers.
2785 ``zext (CST to TYPE)``
2786 Zero extend a constant to another type. The bit size of CST must be
2787 smaller than the bit size of TYPE. Both types must be integers.
2788 ``sext (CST to TYPE)``
2789 Sign extend a constant to another type. The bit size of CST must be
2790 smaller than the bit size of TYPE. Both types must be integers.
2791 ``fptrunc (CST to TYPE)``
2792 Truncate a floating point constant to another floating point type.
2793 The size of CST must be larger than the size of TYPE. Both types
2794 must be floating point.
2795 ``fpext (CST to TYPE)``
2796 Floating point extend a constant to another type. The size of CST
2797 must be smaller or equal to the size of TYPE. Both types must be
2799 ``fptoui (CST to TYPE)``
2800 Convert a floating point constant to the corresponding unsigned
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 ``fptosi (CST to TYPE)``
2806 Convert a floating point constant to the corresponding signed
2807 integer constant. TYPE must be a scalar or vector integer type. CST
2808 must be of scalar or vector floating point type. Both CST and TYPE
2809 must be scalars, or vectors of the same number of elements. If the
2810 value won't fit in the integer type, the results are undefined.
2811 ``uitofp (CST to TYPE)``
2812 Convert an unsigned 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 ``sitofp (CST to TYPE)``
2818 Convert a signed integer constant to the corresponding floating
2819 point constant. TYPE must be a scalar or vector floating point type.
2820 CST must be of scalar or vector integer type. Both CST and TYPE must
2821 be scalars, or vectors of the same number of elements. If the value
2822 won't fit in the floating point type, the results are undefined.
2823 ``ptrtoint (CST to TYPE)``
2824 Convert a pointer typed constant to the corresponding integer
2825 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2826 pointer type. The ``CST`` value is zero extended, truncated, or
2827 unchanged to make it fit in ``TYPE``.
2828 ``inttoptr (CST to TYPE)``
2829 Convert an integer constant to a pointer constant. TYPE must be a
2830 pointer type. CST must be of integer type. The CST value is zero
2831 extended, truncated, or unchanged to make it fit in a pointer size.
2832 This one is *really* dangerous!
2833 ``bitcast (CST to TYPE)``
2834 Convert a constant, CST, to another TYPE. The constraints of the
2835 operands are the same as those for the :ref:`bitcast
2836 instruction <i_bitcast>`.
2837 ``addrspacecast (CST to TYPE)``
2838 Convert a constant pointer or constant vector of pointer, CST, to another
2839 TYPE in a different address space. The constraints of the operands are the
2840 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2841 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2842 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2843 constants. As with the :ref:`getelementptr <i_getelementptr>`
2844 instruction, the index list may have zero or more indexes, which are
2845 required to make sense for the type of "pointer to TY".
2846 ``select (COND, VAL1, VAL2)``
2847 Perform the :ref:`select operation <i_select>` on constants.
2848 ``icmp COND (VAL1, VAL2)``
2849 Performs the :ref:`icmp operation <i_icmp>` on constants.
2850 ``fcmp COND (VAL1, VAL2)``
2851 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2852 ``extractelement (VAL, IDX)``
2853 Perform the :ref:`extractelement operation <i_extractelement>` on
2855 ``insertelement (VAL, ELT, IDX)``
2856 Perform the :ref:`insertelement operation <i_insertelement>` on
2858 ``shufflevector (VEC1, VEC2, IDXMASK)``
2859 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2861 ``extractvalue (VAL, IDX0, IDX1, ...)``
2862 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2863 constants. The index list is interpreted in a similar manner as
2864 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2865 least one index value must be specified.
2866 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2867 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2868 The index list is interpreted in a similar manner as indices in a
2869 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2870 value must be specified.
2871 ``OPCODE (LHS, RHS)``
2872 Perform the specified operation of the LHS and RHS constants. OPCODE
2873 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2874 binary <bitwiseops>` operations. The constraints on operands are
2875 the same as those for the corresponding instruction (e.g. no bitwise
2876 operations on floating point values are allowed).
2883 Inline Assembler Expressions
2884 ----------------------------
2886 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2887 Inline Assembly <moduleasm>`) through the use of a special value. This value
2888 represents the inline assembler as a template string (containing the
2889 instructions to emit), a list of operand constraints (stored as a string), a
2890 flag that indicates whether or not the inline asm expression has side effects,
2891 and a flag indicating whether the function containing the asm needs to align its
2892 stack conservatively.
2894 The template string supports argument substitution of the operands using "``$``"
2895 followed by a number, to indicate substitution of the given register/memory
2896 location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
2897 be used, where ``MODIFIER`` is a target-specific annotation for how to print the
2898 operand (See :ref:`inline-asm-modifiers`).
2900 A literal "``$``" may be included by using "``$$``" in the template. To include
2901 other special characters into the output, the usual "``\XX``" escapes may be
2902 used, just as in other strings. Note that after template substitution, the
2903 resulting assembly string is parsed by LLVM's integrated assembler unless it is
2904 disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
2905 syntax known to LLVM.
2907 LLVM's support for inline asm is modeled closely on the requirements of Clang's
2908 GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
2909 modifier codes listed here are similar or identical to those in GCC's inline asm
2910 support. However, to be clear, the syntax of the template and constraint strings
2911 described here is *not* the same as the syntax accepted by GCC and Clang, and,
2912 while most constraint letters are passed through as-is by Clang, some get
2913 translated to other codes when converting from the C source to the LLVM
2916 An example inline assembler expression is:
2918 .. code-block:: llvm
2920 i32 (i32) asm "bswap $0", "=r,r"
2922 Inline assembler expressions may **only** be used as the callee operand
2923 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2924 Thus, typically we have:
2926 .. code-block:: llvm
2928 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2930 Inline asms with side effects not visible in the constraint list must be
2931 marked as having side effects. This is done through the use of the
2932 '``sideeffect``' keyword, like so:
2934 .. code-block:: llvm
2936 call void asm sideeffect "eieio", ""()
2938 In some cases inline asms will contain code that will not work unless
2939 the stack is aligned in some way, such as calls or SSE instructions on
2940 x86, yet will not contain code that does that alignment within the asm.
2941 The compiler should make conservative assumptions about what the asm
2942 might contain and should generate its usual stack alignment code in the
2943 prologue if the '``alignstack``' keyword is present:
2945 .. code-block:: llvm
2947 call void asm alignstack "eieio", ""()
2949 Inline asms also support using non-standard assembly dialects. The
2950 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2951 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2952 the only supported dialects. An example is:
2954 .. code-block:: llvm
2956 call void asm inteldialect "eieio", ""()
2958 If multiple keywords appear the '``sideeffect``' keyword must come
2959 first, the '``alignstack``' keyword second and the '``inteldialect``'
2962 Inline Asm Constraint String
2963 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2965 The constraint list is a comma-separated string, each element containing one or
2966 more constraint codes.
2968 For each element in the constraint list an appropriate register or memory
2969 operand will be chosen, and it will be made available to assembly template
2970 string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
2973 There are three different types of constraints, which are distinguished by a
2974 prefix symbol in front of the constraint code: Output, Input, and Clobber. The
2975 constraints must always be given in that order: outputs first, then inputs, then
2976 clobbers. They cannot be intermingled.
2978 There are also three different categories of constraint codes:
2980 - Register constraint. This is either a register class, or a fixed physical
2981 register. This kind of constraint will allocate a register, and if necessary,
2982 bitcast the argument or result to the appropriate type.
2983 - Memory constraint. This kind of constraint is for use with an instruction
2984 taking a memory operand. Different constraints allow for different addressing
2985 modes used by the target.
2986 - Immediate value constraint. This kind of constraint is for an integer or other
2987 immediate value which can be rendered directly into an instruction. The
2988 various target-specific constraints allow the selection of a value in the
2989 proper range for the instruction you wish to use it with.
2994 Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
2995 indicates that the assembly will write to this operand, and the operand will
2996 then be made available as a return value of the ``asm`` expression. Output
2997 constraints do not consume an argument from the call instruction. (Except, see
2998 below about indirect outputs).
3000 Normally, it is expected that no output locations are written to by the assembly
3001 expression until *all* of the inputs have been read. As such, LLVM may assign
3002 the same register to an output and an input. If this is not safe (e.g. if the
3003 assembly contains two instructions, where the first writes to one output, and
3004 the second reads an input and writes to a second output), then the "``&``"
3005 modifier must be used (e.g. "``=&r``") to specify that the output is an
3006 "early-clobber" output. Marking an ouput as "early-clobber" ensures that LLVM
3007 will not use the same register for any inputs (other than an input tied to this
3013 Input constraints do not have a prefix -- just the constraint codes. Each input
3014 constraint will consume one argument from the call instruction. It is not
3015 permitted for the asm to write to any input register or memory location (unless
3016 that input is tied to an output). Note also that multiple inputs may all be
3017 assigned to the same register, if LLVM can determine that they necessarily all
3018 contain the same value.
3020 Instead of providing a Constraint Code, input constraints may also "tie"
3021 themselves to an output constraint, by providing an integer as the constraint
3022 string. Tied inputs still consume an argument from the call instruction, and
3023 take up a position in the asm template numbering as is usual -- they will simply
3024 be constrained to always use the same register as the output they've been tied
3025 to. For example, a constraint string of "``=r,0``" says to assign a register for
3026 output, and use that register as an input as well (it being the 0'th
3029 It is permitted to tie an input to an "early-clobber" output. In that case, no
3030 *other* input may share the same register as the input tied to the early-clobber
3031 (even when the other input has the same value).
3033 You may only tie an input to an output which has a register constraint, not a
3034 memory constraint. Only a single input may be tied to an output.
3036 There is also an "interesting" feature which deserves a bit of explanation: if a
3037 register class constraint allocates a register which is too small for the value
3038 type operand provided as input, the input value will be split into multiple
3039 registers, and all of them passed to the inline asm.
3041 However, this feature is often not as useful as you might think.
3043 Firstly, the registers are *not* guaranteed to be consecutive. So, on those
3044 architectures that have instructions which operate on multiple consecutive
3045 instructions, this is not an appropriate way to support them. (e.g. the 32-bit
3046 SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
3047 hardware then loads into both the named register, and the next register. This
3048 feature of inline asm would not be useful to support that.)
3050 A few of the targets provide a template string modifier allowing explicit access
3051 to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
3052 ``D``). On such an architecture, you can actually access the second allocated
3053 register (yet, still, not any subsequent ones). But, in that case, you're still
3054 probably better off simply splitting the value into two separate operands, for
3055 clarity. (e.g. see the description of the ``A`` constraint on X86, which,
3056 despite existing only for use with this feature, is not really a good idea to
3059 Indirect inputs and outputs
3060 """""""""""""""""""""""""""
3062 Indirect output or input constraints can be specified by the "``*``" modifier
3063 (which goes after the "``=``" in case of an output). This indicates that the asm
3064 will write to or read from the contents of an *address* provided as an input
3065 argument. (Note that in this way, indirect outputs act more like an *input* than
3066 an output: just like an input, they consume an argument of the call expression,
3067 rather than producing a return value. An indirect output constraint is an
3068 "output" only in that the asm is expected to write to the contents of the input
3069 memory location, instead of just read from it).
3071 This is most typically used for memory constraint, e.g. "``=*m``", to pass the
3072 address of a variable as a value.
3074 It is also possible to use an indirect *register* constraint, but only on output
3075 (e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3076 value normally, and then, separately emit a store to the address provided as
3077 input, after the provided inline asm. (It's not clear what value this
3078 functionality provides, compared to writing the store explicitly after the asm
3079 statement, and it can only produce worse code, since it bypasses many
3080 optimization passes. I would recommend not using it.)
3086 A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3087 consume an input operand, nor generate an output. Clobbers cannot use any of the
3088 general constraint code letters -- they may use only explicit register
3089 constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3090 "``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3091 memory locations -- not only the memory pointed to by a declared indirect
3097 After a potential prefix comes constraint code, or codes.
3099 A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3100 followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3103 The one and two letter constraint codes are typically chosen to be the same as
3104 GCC's constraint codes.
3106 A single constraint may include one or more than constraint code in it, leaving
3107 it up to LLVM to choose which one to use. This is included mainly for
3108 compatibility with the translation of GCC inline asm coming from clang.
3110 There are two ways to specify alternatives, and either or both may be used in an
3111 inline asm constraint list:
3113 1) Append the codes to each other, making a constraint code set. E.g. "``im``"
3114 or "``{eax}m``". This means "choose any of the options in the set". The
3115 choice of constraint is made independently for each constraint in the
3118 2) Use "``|``" between constraint code sets, creating alternatives. Every
3119 constraint in the constraint list must have the same number of alternative
3120 sets. With this syntax, the same alternative in *all* of the items in the
3121 constraint list will be chosen together.
3123 Putting those together, you might have a two operand constraint string like
3124 ``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3125 operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3126 may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3128 However, the use of either of the alternatives features is *NOT* recommended, as
3129 LLVM is not able to make an intelligent choice about which one to use. (At the
3130 point it currently needs to choose, not enough information is available to do so
3131 in a smart way.) Thus, it simply tries to make a choice that's most likely to
3132 compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3133 always choose to use memory, not registers). And, if given multiple registers,
3134 or multiple register classes, it will simply choose the first one. (In fact, it
3135 doesn't currently even ensure explicitly specified physical registers are
3136 unique, so specifying multiple physical registers as alternatives, like
3137 ``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3140 Supported Constraint Code List
3141 """"""""""""""""""""""""""""""
3143 The constraint codes are, in general, expected to behave the same way they do in
3144 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3145 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3146 and GCC likely indicates a bug in LLVM.
3148 Some constraint codes are typically supported by all targets:
3150 - ``r``: A register in the target's general purpose register class.
3151 - ``m``: A memory address operand. It is target-specific what addressing modes
3152 are supported, typical examples are register, or register + register offset,
3153 or register + immediate offset (of some target-specific size).
3154 - ``i``: An integer constant (of target-specific width). Allows either a simple
3155 immediate, or a relocatable value.
3156 - ``n``: An integer constant -- *not* including relocatable values.
3157 - ``s``: An integer constant, but allowing *only* relocatable values.
3158 - ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3159 useful to pass a label for an asm branch or call.
3161 .. FIXME: but that surely isn't actually okay to jump out of an asm
3162 block without telling llvm about the control transfer???)
3164 - ``{register-name}``: Requires exactly the named physical register.
3166 Other constraints are target-specific:
3170 - ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3171 - ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3172 i.e. 0 to 4095 with optional shift by 12.
3173 - ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3174 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3175 - ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3176 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3177 - ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3178 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3179 - ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3180 32-bit register. This is a superset of ``K``: in addition to the bitmask
3181 immediate, also allows immediate integers which can be loaded with a single
3182 ``MOVZ`` or ``MOVL`` instruction.
3183 - ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3184 64-bit register. This is a superset of ``L``.
3185 - ``Q``: Memory address operand must be in a single register (no
3186 offsets). (However, LLVM currently does this for the ``m`` constraint as
3188 - ``r``: A 32 or 64-bit integer register (W* or X*).
3189 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3190 - ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3194 - ``r``: A 32 or 64-bit integer register.
3195 - ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3196 - ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3201 - ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3202 operand. Treated the same as operand ``m``, at the moment.
3204 ARM and ARM's Thumb2 mode:
3206 - ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3207 - ``I``: An immediate integer valid for a data-processing instruction.
3208 - ``J``: An immediate integer between -4095 and 4095.
3209 - ``K``: An immediate integer whose bitwise inverse is valid for a
3210 data-processing instruction. (Can be used with template modifier "``B``" to
3211 print the inverted value).
3212 - ``L``: An immediate integer whose negation is valid for a data-processing
3213 instruction. (Can be used with template modifier "``n``" to print the negated
3215 - ``M``: A power of two or a integer between 0 and 32.
3216 - ``N``: Invalid immediate constraint.
3217 - ``O``: Invalid immediate constraint.
3218 - ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3219 - ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3221 - ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3223 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3224 ``d0-d31``, or ``q0-q15``.
3225 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3226 ``d0-d7``, or ``q0-q3``.
3227 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3232 - ``I``: An immediate integer between 0 and 255.
3233 - ``J``: An immediate integer between -255 and -1.
3234 - ``K``: An immediate integer between 0 and 255, with optional left-shift by
3236 - ``L``: An immediate integer between -7 and 7.
3237 - ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3238 - ``N``: An immediate integer between 0 and 31.
3239 - ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3240 - ``r``: A low 32-bit GPR register (``r0-r7``).
3241 - ``l``: A low 32-bit GPR register (``r0-r7``).
3242 - ``h``: A high GPR register (``r0-r7``).
3243 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3244 ``d0-d31``, or ``q0-q15``.
3245 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3246 ``d0-d7``, or ``q0-q3``.
3247 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3253 - ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3255 - ``r``: A 32 or 64-bit register.
3259 - ``r``: An 8 or 16-bit register.
3263 - ``I``: An immediate signed 16-bit integer.
3264 - ``J``: An immediate integer zero.
3265 - ``K``: An immediate unsigned 16-bit integer.
3266 - ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3267 - ``N``: An immediate integer between -65535 and -1.
3268 - ``O``: An immediate signed 15-bit integer.
3269 - ``P``: An immediate integer between 1 and 65535.
3270 - ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3271 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3272 - ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3273 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3275 - ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3276 ``sc`` instruction on the given subtarget (details vary).
3277 - ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3278 - ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
3279 (``W0-W31``). In the case of MSA registers, it is recommended to use the ``w``
3280 argument modifier for compatibility with GCC.
3281 - ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3283 - ``l``: The ``lo`` register, 32 or 64-bit.
3288 - ``b``: A 1-bit integer register.
3289 - ``c`` or ``h``: A 16-bit integer register.
3290 - ``r``: A 32-bit integer register.
3291 - ``l`` or ``N``: A 64-bit integer register.
3292 - ``f``: A 32-bit float register.
3293 - ``d``: A 64-bit float register.
3298 - ``I``: An immediate signed 16-bit integer.
3299 - ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3300 - ``K``: An immediate unsigned 16-bit integer.
3301 - ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3302 - ``M``: An immediate integer greater than 31.
3303 - ``N``: An immediate integer that is an exact power of 2.
3304 - ``O``: The immediate integer constant 0.
3305 - ``P``: An immediate integer constant whose negation is a signed 16-bit
3307 - ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3308 treated the same as ``m``.
3309 - ``r``: A 32 or 64-bit integer register.
3310 - ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3312 - ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3313 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3314 - ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3315 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3316 altivec vector register (``V0-V31``).
3318 .. FIXME: is this a bug that v accepts QPX registers? I think this
3319 is supposed to only use the altivec vector registers?
3321 - ``y``: Condition register (``CR0-CR7``).
3322 - ``wc``: An individual CR bit in a CR register.
3323 - ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3324 register set (overlapping both the floating-point and vector register files).
3325 - ``ws``: A 32 or 64-bit floating point register, from the full VSX register
3330 - ``I``: An immediate 13-bit signed integer.
3331 - ``r``: A 32-bit integer register.
3335 - ``I``: An immediate unsigned 8-bit integer.
3336 - ``J``: An immediate unsigned 12-bit integer.
3337 - ``K``: An immediate signed 16-bit integer.
3338 - ``L``: An immediate signed 20-bit integer.
3339 - ``M``: An immediate integer 0x7fffffff.
3340 - ``Q``, ``R``, ``S``, ``T``: A memory address operand, treated the same as
3341 ``m``, at the moment.
3342 - ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3343 - ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3344 address context evaluates as zero).
3345 - ``h``: A 32-bit value in the high part of a 64bit data register
3347 - ``f``: A 32, 64, or 128-bit floating point register.
3351 - ``I``: An immediate integer between 0 and 31.
3352 - ``J``: An immediate integer between 0 and 64.
3353 - ``K``: An immediate signed 8-bit integer.
3354 - ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3356 - ``M``: An immediate integer between 0 and 3.
3357 - ``N``: An immediate unsigned 8-bit integer.
3358 - ``O``: An immediate integer between 0 and 127.
3359 - ``e``: An immediate 32-bit signed integer.
3360 - ``Z``: An immediate 32-bit unsigned integer.
3361 - ``o``, ``v``: Treated the same as ``m``, at the moment.
3362 - ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3363 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3364 registers, and on X86-64, it is all of the integer registers.
3365 - ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3366 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3367 - ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3368 - ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3369 existed since i386, and can be accessed without the REX prefix.
3370 - ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3371 - ``y``: A 64-bit MMX register, if MMX is enabled.
3372 - ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3373 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3374 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3375 512-bit vector operand in an AVX512 register, Otherwise, an error.
3376 - ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3377 - ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3378 32-bit mode, a 64-bit integer operand will get split into two registers). It
3379 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3380 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3381 you're better off splitting it yourself, before passing it to the asm
3386 - ``r``: A 32-bit integer register.
3389 .. _inline-asm-modifiers:
3391 Asm template argument modifiers
3392 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3394 In the asm template string, modifiers can be used on the operand reference, like
3397 The modifiers are, in general, expected to behave the same way they do in
3398 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3399 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3400 and GCC likely indicates a bug in LLVM.
3404 - ``c``: Print an immediate integer constant unadorned, without
3405 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3406 - ``n``: Negate and print immediate integer constant unadorned, without the
3407 target-specific immediate punctuation (e.g. no ``$`` prefix).
3408 - ``l``: Print as an unadorned label, without the target-specific label
3409 punctuation (e.g. no ``$`` prefix).
3413 - ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3414 instead of ``x30``, print ``w30``.
3415 - ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3416 - ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3417 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3426 - ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3430 - ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3431 as ``d4[1]`` instead of ``s9``)
3432 - ``B``: Bitwise invert and print an immediate integer constant without ``#``
3434 - ``L``: Print the low 16-bits of an immediate integer constant.
3435 - ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3436 register operands subsequent to the specified one (!), so use carefully.
3437 - ``Q``: Print the low-order register of a register-pair, or the low-order
3438 register of a two-register operand.
3439 - ``R``: Print the high-order register of a register-pair, or the high-order
3440 register of a two-register operand.
3441 - ``H``: Print the second register of a register-pair. (On a big-endian system,
3442 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3445 .. FIXME: H doesn't currently support printing the second register
3446 of a two-register operand.
3448 - ``e``: Print the low doubleword register of a NEON quad register.
3449 - ``f``: Print the high doubleword register of a NEON quad register.
3450 - ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3455 - ``L``: Print the second register of a two-register operand. Requires that it
3456 has been allocated consecutively to the first.
3458 .. FIXME: why is it restricted to consecutive ones? And there's
3459 nothing that ensures that happens, is there?
3461 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3462 nothing. Used to print 'addi' vs 'add' instructions.
3466 No additional modifiers.
3470 - ``X``: Print an immediate integer as hexadecimal
3471 - ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3472 - ``d``: Print an immediate integer as decimal.
3473 - ``m``: Subtract one and print an immediate integer as decimal.
3474 - ``z``: Print $0 if an immediate zero, otherwise print normally.
3475 - ``L``: Print the low-order register of a two-register operand, or prints the
3476 address of the low-order word of a double-word memory operand.
3478 .. FIXME: L seems to be missing memory operand support.
3480 - ``M``: Print the high-order register of a two-register operand, or prints the
3481 address of the high-order word of a double-word memory operand.
3483 .. FIXME: M seems to be missing memory operand support.
3485 - ``D``: Print the second register of a two-register operand, or prints the
3486 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3487 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3489 - ``w``: No effect. Provided for compatibility with GCC which requires this
3490 modifier in order to print MSA registers (``W0-W31``) with the ``f``
3499 - ``L``: Print the second register of a two-register operand. Requires that it
3500 has been allocated consecutively to the first.
3502 .. FIXME: why is it restricted to consecutive ones? And there's
3503 nothing that ensures that happens, is there?
3505 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3506 nothing. Used to print 'addi' vs 'add' instructions.
3507 - ``y``: For a memory operand, prints formatter for a two-register X-form
3508 instruction. (Currently always prints ``r0,OPERAND``).
3509 - ``U``: Prints 'u' if the memory operand is an update form, and nothing
3510 otherwise. (NOTE: LLVM does not support update form, so this will currently
3511 always print nothing)
3512 - ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
3513 not support indexed form, so this will currently always print nothing)
3521 SystemZ implements only ``n``, and does *not* support any of the other
3522 target-independent modifiers.
3526 - ``c``: Print an unadorned integer or symbol name. (The latter is
3527 target-specific behavior for this typically target-independent modifier).
3528 - ``A``: Print a register name with a '``*``' before it.
3529 - ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
3531 - ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
3533 - ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
3535 - ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
3537 - ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
3538 available, otherwise the 32-bit register name; do nothing on a memory operand.
3539 - ``n``: Negate and print an unadorned integer, or, for operands other than an
3540 immediate integer (e.g. a relocatable symbol expression), print a '-' before
3541 the operand. (The behavior for relocatable symbol expressions is a
3542 target-specific behavior for this typically target-independent modifier)
3543 - ``H``: Print a memory reference with additional offset +8.
3544 - ``P``: Print a memory reference or operand for use as the argument of a call
3545 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
3549 No additional modifiers.
3555 The call instructions that wrap inline asm nodes may have a
3556 "``!srcloc``" MDNode attached to it that contains a list of constant
3557 integers. If present, the code generator will use the integer as the
3558 location cookie value when report errors through the ``LLVMContext``
3559 error reporting mechanisms. This allows a front-end to correlate backend
3560 errors that occur with inline asm back to the source code that produced
3563 .. code-block:: llvm
3565 call void asm sideeffect "something bad", ""(), !srcloc !42
3567 !42 = !{ i32 1234567 }
3569 It is up to the front-end to make sense of the magic numbers it places
3570 in the IR. If the MDNode contains multiple constants, the code generator
3571 will use the one that corresponds to the line of the asm that the error
3579 LLVM IR allows metadata to be attached to instructions in the program
3580 that can convey extra information about the code to the optimizers and
3581 code generator. One example application of metadata is source-level
3582 debug information. There are two metadata primitives: strings and nodes.
3584 Metadata does not have a type, and is not a value. If referenced from a
3585 ``call`` instruction, it uses the ``metadata`` type.
3587 All metadata are identified in syntax by a exclamation point ('``!``').
3589 .. _metadata-string:
3591 Metadata Nodes and Metadata Strings
3592 -----------------------------------
3594 A metadata string is a string surrounded by double quotes. It can
3595 contain any character by escaping non-printable characters with
3596 "``\xx``" where "``xx``" is the two digit hex code. For example:
3599 Metadata nodes are represented with notation similar to structure
3600 constants (a comma separated list of elements, surrounded by braces and
3601 preceded by an exclamation point). Metadata nodes can have any values as
3602 their operand. For example:
3604 .. code-block:: llvm
3606 !{ !"test\00", i32 10}
3608 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
3610 .. code-block:: llvm
3612 !0 = distinct !{!"test\00", i32 10}
3614 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
3615 content. They can also occur when transformations cause uniquing collisions
3616 when metadata operands change.
3618 A :ref:`named metadata <namedmetadatastructure>` is a collection of
3619 metadata nodes, which can be looked up in the module symbol table. For
3622 .. code-block:: llvm
3626 Metadata can be used as function arguments. Here ``llvm.dbg.value``
3627 function is using two metadata arguments:
3629 .. code-block:: llvm
3631 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3633 Metadata can be attached with an instruction. Here metadata ``!21`` is
3634 attached to the ``add`` instruction using the ``!dbg`` identifier:
3636 .. code-block:: llvm
3638 %indvar.next = add i64 %indvar, 1, !dbg !21
3640 More information about specific metadata nodes recognized by the
3641 optimizers and code generator is found below.
3643 .. _specialized-metadata:
3645 Specialized Metadata Nodes
3646 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3648 Specialized metadata nodes are custom data structures in metadata (as opposed
3649 to generic tuples). Their fields are labelled, and can be specified in any
3652 These aren't inherently debug info centric, but currently all the specialized
3653 metadata nodes are related to debug info.
3660 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
3661 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
3662 tuples containing the debug info to be emitted along with the compile unit,
3663 regardless of code optimizations (some nodes are only emitted if there are
3664 references to them from instructions).
3666 .. code-block:: llvm
3668 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
3669 isOptimized: true, flags: "-O2", runtimeVersion: 2,
3670 splitDebugFilename: "abc.debug", emissionKind: 1,
3671 enums: !2, retainedTypes: !3, subprograms: !4,
3672 globals: !5, imports: !6)
3674 Compile unit descriptors provide the root scope for objects declared in a
3675 specific compilation unit. File descriptors are defined using this scope.
3676 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
3677 keep track of subprograms, global variables, type information, and imported
3678 entities (declarations and namespaces).
3685 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
3687 .. code-block:: llvm
3689 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
3691 Files are sometimes used in ``scope:`` fields, and are the only valid target
3692 for ``file:`` fields.
3699 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
3700 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
3702 .. code-block:: llvm
3704 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3705 encoding: DW_ATE_unsigned_char)
3706 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
3708 The ``encoding:`` describes the details of the type. Usually it's one of the
3711 .. code-block:: llvm
3717 DW_ATE_signed_char = 6
3719 DW_ATE_unsigned_char = 8
3721 .. _DISubroutineType:
3726 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3727 refers to a tuple; the first operand is the return type, while the rest are the
3728 types of the formal arguments in order. If the first operand is ``null``, that
3729 represents a function with no return value (such as ``void foo() {}`` in C++).
3731 .. code-block:: llvm
3733 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3734 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3735 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3742 ``DIDerivedType`` nodes represent types derived from other types, such as
3745 .. code-block:: llvm
3747 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3748 encoding: DW_ATE_unsigned_char)
3749 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3752 The following ``tag:`` values are valid:
3754 .. code-block:: llvm
3756 DW_TAG_formal_parameter = 5
3758 DW_TAG_pointer_type = 15
3759 DW_TAG_reference_type = 16
3761 DW_TAG_ptr_to_member_type = 31
3762 DW_TAG_const_type = 38
3763 DW_TAG_volatile_type = 53
3764 DW_TAG_restrict_type = 55
3766 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3767 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3768 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3769 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3770 argument of a subprogram.
3772 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3774 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3775 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3778 Note that the ``void *`` type is expressed as a type derived from NULL.
3780 .. _DICompositeType:
3785 ``DICompositeType`` nodes represent types composed of other types, like
3786 structures and unions. ``elements:`` points to a tuple of the composed types.
3788 If the source language supports ODR, the ``identifier:`` field gives the unique
3789 identifier used for type merging between modules. When specified, other types
3790 can refer to composite types indirectly via a :ref:`metadata string
3791 <metadata-string>` that matches their identifier.
3793 .. code-block:: llvm
3795 !0 = !DIEnumerator(name: "SixKind", value: 7)
3796 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3797 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3798 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3799 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3800 elements: !{!0, !1, !2})
3802 The following ``tag:`` values are valid:
3804 .. code-block:: llvm
3806 DW_TAG_array_type = 1
3807 DW_TAG_class_type = 2
3808 DW_TAG_enumeration_type = 4
3809 DW_TAG_structure_type = 19
3810 DW_TAG_union_type = 23
3811 DW_TAG_subroutine_type = 21
3812 DW_TAG_inheritance = 28
3815 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3816 descriptors <DISubrange>`, each representing the range of subscripts at that
3817 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3818 array type is a native packed vector.
3820 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3821 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3822 value for the set. All enumeration type descriptors are collected in the
3823 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3825 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3826 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3827 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3834 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3835 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3837 .. code-block:: llvm
3839 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3840 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3841 !2 = !DISubrange(count: -1) ; empty array.
3848 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3849 variants of :ref:`DICompositeType`.
3851 .. code-block:: llvm
3853 !0 = !DIEnumerator(name: "SixKind", value: 7)
3854 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3855 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3857 DITemplateTypeParameter
3858 """""""""""""""""""""""
3860 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3861 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3862 :ref:`DISubprogram` ``templateParams:`` fields.
3864 .. code-block:: llvm
3866 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3868 DITemplateValueParameter
3869 """"""""""""""""""""""""
3871 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3872 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3873 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3874 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3875 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3877 .. code-block:: llvm
3879 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3884 ``DINamespace`` nodes represent namespaces in the source language.
3886 .. code-block:: llvm
3888 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3893 ``DIGlobalVariable`` nodes represent global variables in the source language.
3895 .. code-block:: llvm
3897 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3898 file: !2, line: 7, type: !3, isLocal: true,
3899 isDefinition: false, variable: i32* @foo,
3902 All global variables should be referenced by the `globals:` field of a
3903 :ref:`compile unit <DICompileUnit>`.
3910 ``DISubprogram`` nodes represent functions from the source language. The
3911 ``variables:`` field points at :ref:`variables <DILocalVariable>` that must be
3912 retained, even if their IR counterparts are optimized out of the IR. The
3913 ``type:`` field must point at an :ref:`DISubroutineType`.
3915 .. code-block:: llvm
3917 !0 = !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3918 file: !2, line: 7, type: !3, isLocal: true,
3919 isDefinition: false, scopeLine: 8, containingType: !4,
3920 virtuality: DW_VIRTUALITY_pure_virtual, virtualIndex: 10,
3921 flags: DIFlagPrototyped, isOptimized: true,
3922 function: void ()* @_Z3foov,
3923 templateParams: !5, declaration: !6, variables: !7)
3930 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3931 <DISubprogram>`. The line number and column numbers are used to distinguish
3932 two lexical blocks at same depth. They are valid targets for ``scope:``
3935 .. code-block:: llvm
3937 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3939 Usually lexical blocks are ``distinct`` to prevent node merging based on
3942 .. _DILexicalBlockFile:
3947 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3948 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
3949 indicate textual inclusion, or the ``discriminator:`` field can be used to
3950 discriminate between control flow within a single block in the source language.
3952 .. code-block:: llvm
3954 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3955 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3956 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3963 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
3964 mandatory, and points at an :ref:`DILexicalBlockFile`, an
3965 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
3967 .. code-block:: llvm
3969 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3971 .. _DILocalVariable:
3976 ``DILocalVariable`` nodes represent local variables in the source language. If
3977 the ``arg:`` field is set to non-zero, then this variable is a subprogram
3978 parameter, and it will be included in the ``variables:`` field of its
3979 :ref:`DISubprogram`.
3981 .. code-block:: llvm
3983 !0 = !DILocalVariable(name: "this", arg: 1, scope: !3, file: !2, line: 7,
3984 type: !3, flags: DIFlagArtificial)
3985 !1 = !DILocalVariable(name: "x", arg: 2, scope: !4, file: !2, line: 7,
3987 !2 = !DILocalVariable(name: "y", scope: !5, file: !2, line: 7, type: !3)
3992 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
3993 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
3994 describe how the referenced LLVM variable relates to the source language
3997 The current supported vocabulary is limited:
3999 - ``DW_OP_deref`` dereferences the working expression.
4000 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
4001 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
4002 here, respectively) of the variable piece from the working expression.
4004 .. code-block:: llvm
4006 !0 = !DIExpression(DW_OP_deref)
4007 !1 = !DIExpression(DW_OP_plus, 3)
4008 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
4009 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
4014 ``DIObjCProperty`` nodes represent Objective-C property nodes.
4016 .. code-block:: llvm
4018 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
4019 getter: "getFoo", attributes: 7, type: !2)
4024 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
4027 .. code-block:: llvm
4029 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
4030 entity: !1, line: 7)
4035 In LLVM IR, memory does not have types, so LLVM's own type system is not
4036 suitable for doing TBAA. Instead, metadata is added to the IR to
4037 describe a type system of a higher level language. This can be used to
4038 implement typical C/C++ TBAA, but it can also be used to implement
4039 custom alias analysis behavior for other languages.
4041 The current metadata format is very simple. TBAA metadata nodes have up
4042 to three fields, e.g.:
4044 .. code-block:: llvm
4046 !0 = !{ !"an example type tree" }
4047 !1 = !{ !"int", !0 }
4048 !2 = !{ !"float", !0 }
4049 !3 = !{ !"const float", !2, i64 1 }
4051 The first field is an identity field. It can be any value, usually a
4052 metadata string, which uniquely identifies the type. The most important
4053 name in the tree is the name of the root node. Two trees with different
4054 root node names are entirely disjoint, even if they have leaves with
4057 The second field identifies the type's parent node in the tree, or is
4058 null or omitted for a root node. A type is considered to alias all of
4059 its descendants and all of its ancestors in the tree. Also, a type is
4060 considered to alias all types in other trees, so that bitcode produced
4061 from multiple front-ends is handled conservatively.
4063 If the third field is present, it's an integer which if equal to 1
4064 indicates that the type is "constant" (meaning
4065 ``pointsToConstantMemory`` should return true; see `other useful
4066 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
4068 '``tbaa.struct``' Metadata
4069 ^^^^^^^^^^^^^^^^^^^^^^^^^^
4071 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
4072 aggregate assignment operations in C and similar languages, however it
4073 is defined to copy a contiguous region of memory, which is more than
4074 strictly necessary for aggregate types which contain holes due to
4075 padding. Also, it doesn't contain any TBAA information about the fields
4078 ``!tbaa.struct`` metadata can describe which memory subregions in a
4079 memcpy are padding and what the TBAA tags of the struct are.
4081 The current metadata format is very simple. ``!tbaa.struct`` metadata
4082 nodes are a list of operands which are in conceptual groups of three.
4083 For each group of three, the first operand gives the byte offset of a
4084 field in bytes, the second gives its size in bytes, and the third gives
4087 .. code-block:: llvm
4089 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
4091 This describes a struct with two fields. The first is at offset 0 bytes
4092 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4093 and has size 4 bytes and has tbaa tag !2.
4095 Note that the fields need not be contiguous. In this example, there is a
4096 4 byte gap between the two fields. This gap represents padding which
4097 does not carry useful data and need not be preserved.
4099 '``noalias``' and '``alias.scope``' Metadata
4100 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4102 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4103 noalias memory-access sets. This means that some collection of memory access
4104 instructions (loads, stores, memory-accessing calls, etc.) that carry
4105 ``noalias`` metadata can specifically be specified not to alias with some other
4106 collection of memory access instructions that carry ``alias.scope`` metadata.
4107 Each type of metadata specifies a list of scopes where each scope has an id and
4108 a domain. When evaluating an aliasing query, if for some domain, the set
4109 of scopes with that domain in one instruction's ``alias.scope`` list is a
4110 subset of (or equal to) the set of scopes for that domain in another
4111 instruction's ``noalias`` list, then the two memory accesses are assumed not to
4114 The metadata identifying each domain is itself a list containing one or two
4115 entries. The first entry is the name of the domain. Note that if the name is a
4116 string then it can be combined across functions and translation units. A
4117 self-reference can be used to create globally unique domain names. A
4118 descriptive string may optionally be provided as a second list entry.
4120 The metadata identifying each scope is also itself a list containing two or
4121 three entries. The first entry is the name of the scope. Note that if the name
4122 is a string then it can be combined across functions and translation units. A
4123 self-reference can be used to create globally unique scope names. A metadata
4124 reference to the scope's domain is the second entry. A descriptive string may
4125 optionally be provided as a third list entry.
4129 .. code-block:: llvm
4131 ; Two scope domains:
4135 ; Some scopes in these domains:
4141 !5 = !{!4} ; A list containing only scope !4
4145 ; These two instructions don't alias:
4146 %0 = load float, float* %c, align 4, !alias.scope !5
4147 store float %0, float* %arrayidx.i, align 4, !noalias !5
4149 ; These two instructions also don't alias (for domain !1, the set of scopes
4150 ; in the !alias.scope equals that in the !noalias list):
4151 %2 = load float, float* %c, align 4, !alias.scope !5
4152 store float %2, float* %arrayidx.i2, align 4, !noalias !6
4154 ; These two instructions may alias (for domain !0, the set of scopes in
4155 ; the !noalias list is not a superset of, or equal to, the scopes in the
4156 ; !alias.scope list):
4157 %2 = load float, float* %c, align 4, !alias.scope !6
4158 store float %0, float* %arrayidx.i, align 4, !noalias !7
4160 '``fpmath``' Metadata
4161 ^^^^^^^^^^^^^^^^^^^^^
4163 ``fpmath`` metadata may be attached to any instruction of floating point
4164 type. It can be used to express the maximum acceptable error in the
4165 result of that instruction, in ULPs, thus potentially allowing the
4166 compiler to use a more efficient but less accurate method of computing
4167 it. ULP is defined as follows:
4169 If ``x`` is a real number that lies between two finite consecutive
4170 floating-point numbers ``a`` and ``b``, without being equal to one
4171 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4172 distance between the two non-equal finite floating-point numbers
4173 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4175 The metadata node shall consist of a single positive floating point
4176 number representing the maximum relative error, for example:
4178 .. code-block:: llvm
4180 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
4184 '``range``' Metadata
4185 ^^^^^^^^^^^^^^^^^^^^
4187 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4188 integer types. It expresses the possible ranges the loaded value or the value
4189 returned by the called function at this call site is in. The ranges are
4190 represented with a flattened list of integers. The loaded value or the value
4191 returned is known to be in the union of the ranges defined by each consecutive
4192 pair. Each pair has the following properties:
4194 - The type must match the type loaded by the instruction.
4195 - The pair ``a,b`` represents the range ``[a,b)``.
4196 - Both ``a`` and ``b`` are constants.
4197 - The range is allowed to wrap.
4198 - The range should not represent the full or empty set. That is,
4201 In addition, the pairs must be in signed order of the lower bound and
4202 they must be non-contiguous.
4206 .. code-block:: llvm
4208 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4209 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
4210 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4211 %d = invoke i8 @bar() to label %cont
4212 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
4214 !0 = !{ i8 0, i8 2 }
4215 !1 = !{ i8 255, i8 2 }
4216 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4217 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
4219 '``unpredictable``' Metadata
4220 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4222 ``unpredictable`` metadata may be attached to any branch or switch
4223 instruction. It can be used to express the unpredictability of control
4224 flow. Similar to the llvm.expect intrinsic, it may be used to alter
4225 optimizations related to compare and branch instructions. The metadata
4226 is treated as a boolean value; if it exists, it signals that the branch
4227 or switch that it is attached to is completely unpredictable.
4232 It is sometimes useful to attach information to loop constructs. Currently,
4233 loop metadata is implemented as metadata attached to the branch instruction
4234 in the loop latch block. This type of metadata refer to a metadata node that is
4235 guaranteed to be separate for each loop. The loop identifier metadata is
4236 specified with the name ``llvm.loop``.
4238 The loop identifier metadata is implemented using a metadata that refers to
4239 itself to avoid merging it with any other identifier metadata, e.g.,
4240 during module linkage or function inlining. That is, each loop should refer
4241 to their own identification metadata even if they reside in separate functions.
4242 The following example contains loop identifier metadata for two separate loop
4245 .. code-block:: llvm
4250 The loop identifier metadata can be used to specify additional
4251 per-loop metadata. Any operands after the first operand can be treated
4252 as user-defined metadata. For example the ``llvm.loop.unroll.count``
4253 suggests an unroll factor to the loop unroller:
4255 .. code-block:: llvm
4257 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
4260 !1 = !{!"llvm.loop.unroll.count", i32 4}
4262 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
4263 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4265 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
4266 used to control per-loop vectorization and interleaving parameters such as
4267 vectorization width and interleave count. These metadata should be used in
4268 conjunction with ``llvm.loop`` loop identification metadata. The
4269 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
4270 optimization hints and the optimizer will only interleave and vectorize loops if
4271 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
4272 which contains information about loop-carried memory dependencies can be helpful
4273 in determining the safety of these transformations.
4275 '``llvm.loop.interleave.count``' Metadata
4276 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4278 This metadata suggests an interleave count to the loop interleaver.
4279 The first operand is the string ``llvm.loop.interleave.count`` and the
4280 second operand is an integer specifying the interleave count. For
4283 .. code-block:: llvm
4285 !0 = !{!"llvm.loop.interleave.count", i32 4}
4287 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
4288 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
4289 then the interleave count will be determined automatically.
4291 '``llvm.loop.vectorize.enable``' Metadata
4292 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4294 This metadata selectively enables or disables vectorization for the loop. The
4295 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
4296 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
4297 0 disables vectorization:
4299 .. code-block:: llvm
4301 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
4302 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
4304 '``llvm.loop.vectorize.width``' Metadata
4305 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4307 This metadata sets the target width of the vectorizer. The first
4308 operand is the string ``llvm.loop.vectorize.width`` and the second
4309 operand is an integer specifying the width. For example:
4311 .. code-block:: llvm
4313 !0 = !{!"llvm.loop.vectorize.width", i32 4}
4315 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
4316 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
4317 0 or if the loop does not have this metadata the width will be
4318 determined automatically.
4320 '``llvm.loop.unroll``'
4321 ^^^^^^^^^^^^^^^^^^^^^^
4323 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
4324 optimization hints such as the unroll factor. ``llvm.loop.unroll``
4325 metadata should be used in conjunction with ``llvm.loop`` loop
4326 identification metadata. The ``llvm.loop.unroll`` metadata are only
4327 optimization hints and the unrolling will only be performed if the
4328 optimizer believes it is safe to do so.
4330 '``llvm.loop.unroll.count``' Metadata
4331 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4333 This metadata suggests an unroll factor to the loop unroller. The
4334 first operand is the string ``llvm.loop.unroll.count`` and the second
4335 operand is a positive integer specifying the unroll factor. For
4338 .. code-block:: llvm
4340 !0 = !{!"llvm.loop.unroll.count", i32 4}
4342 If the trip count of the loop is less than the unroll count the loop
4343 will be partially unrolled.
4345 '``llvm.loop.unroll.disable``' Metadata
4346 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4348 This metadata disables loop unrolling. The metadata has a single operand
4349 which is the string ``llvm.loop.unroll.disable``. For example:
4351 .. code-block:: llvm
4353 !0 = !{!"llvm.loop.unroll.disable"}
4355 '``llvm.loop.unroll.runtime.disable``' Metadata
4356 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4358 This metadata disables runtime loop unrolling. The metadata has a single
4359 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
4361 .. code-block:: llvm
4363 !0 = !{!"llvm.loop.unroll.runtime.disable"}
4365 '``llvm.loop.unroll.enable``' Metadata
4366 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4368 This metadata suggests that the loop should be fully unrolled if the trip count
4369 is known at compile time and partially unrolled if the trip count is not known
4370 at compile time. The metadata has a single operand which is the string
4371 ``llvm.loop.unroll.enable``. For example:
4373 .. code-block:: llvm
4375 !0 = !{!"llvm.loop.unroll.enable"}
4377 '``llvm.loop.unroll.full``' Metadata
4378 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4380 This metadata suggests that the loop should be unrolled fully. The
4381 metadata has a single operand which is the string ``llvm.loop.unroll.full``.
4384 .. code-block:: llvm
4386 !0 = !{!"llvm.loop.unroll.full"}
4391 Metadata types used to annotate memory accesses with information helpful
4392 for optimizations are prefixed with ``llvm.mem``.
4394 '``llvm.mem.parallel_loop_access``' Metadata
4395 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4397 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
4398 or metadata containing a list of loop identifiers for nested loops.
4399 The metadata is attached to memory accessing instructions and denotes that
4400 no loop carried memory dependence exist between it and other instructions denoted
4401 with the same loop identifier.
4403 Precisely, given two instructions ``m1`` and ``m2`` that both have the
4404 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
4405 set of loops associated with that metadata, respectively, then there is no loop
4406 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
4409 As a special case, if all memory accessing instructions in a loop have
4410 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
4411 loop has no loop carried memory dependences and is considered to be a parallel
4414 Note that if not all memory access instructions have such metadata referring to
4415 the loop, then the loop is considered not being trivially parallel. Additional
4416 memory dependence analysis is required to make that determination. As a fail
4417 safe mechanism, this causes loops that were originally parallel to be considered
4418 sequential (if optimization passes that are unaware of the parallel semantics
4419 insert new memory instructions into the loop body).
4421 Example of a loop that is considered parallel due to its correct use of
4422 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
4423 metadata types that refer to the same loop identifier metadata.
4425 .. code-block:: llvm
4429 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
4431 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4433 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
4439 It is also possible to have nested parallel loops. In that case the
4440 memory accesses refer to a list of loop identifier metadata nodes instead of
4441 the loop identifier metadata node directly:
4443 .. code-block:: llvm
4447 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
4449 br label %inner.for.body
4453 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4455 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
4457 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
4461 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
4463 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
4465 outer.for.end: ; preds = %for.body
4467 !0 = !{!1, !2} ; a list of loop identifiers
4468 !1 = !{!1} ; an identifier for the inner loop
4469 !2 = !{!2} ; an identifier for the outer loop
4474 The ``llvm.bitsets`` global metadata is used to implement
4475 :doc:`bitsets <BitSets>`.
4477 '``invariant.group``' Metadata
4478 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4480 The ``invariant.group`` metadata may be attached to ``load``/``store`` instructions.
4481 The existence of the ``invariant.group`` metadata on the instruction tells
4482 the optimizer that every ``load`` and ``store`` to the same pointer operand
4483 within the same invariant group can be assumed to load or store the same
4484 value (but see the ``llvm.invariant.group.barrier`` intrinsic which affects
4485 when two pointers are considered the same).
4489 .. code-block:: llvm
4491 @unknownPtr = external global i8
4494 store i8 42, i8* %ptr, !invariant.group !0
4495 call void @foo(i8* %ptr)
4497 %a = load i8, i8* %ptr, !invariant.group !0 ; Can assume that value under %ptr didn't change
4498 call void @foo(i8* %ptr)
4499 %b = load i8, i8* %ptr, !invariant.group !1 ; Can't assume anything, because group changed
4501 %newPtr = call i8* @getPointer(i8* %ptr)
4502 %c = load i8, i8* %newPtr, !invariant.group !0 ; Can't assume anything, because we only have information about %ptr
4504 %unknownValue = load i8, i8* @unknownPtr
4505 store i8 %unknownValue, i8* %ptr, !invariant.group !0 ; Can assume that %unknownValue == 42
4507 call void @foo(i8* %ptr)
4508 %newPtr2 = call i8* @llvm.invariant.group.barrier(i8* %ptr)
4509 %d = load i8, i8* %newPtr2, !invariant.group !0 ; Can't step through invariant.group.barrier to get value of %ptr
4512 declare void @foo(i8*)
4513 declare i8* @getPointer(i8*)
4514 declare i8* @llvm.invariant.group.barrier(i8*)
4516 !0 = !{!"magic ptr"}
4517 !1 = !{!"other ptr"}
4521 Module Flags Metadata
4522 =====================
4524 Information about the module as a whole is difficult to convey to LLVM's
4525 subsystems. The LLVM IR isn't sufficient to transmit this information.
4526 The ``llvm.module.flags`` named metadata exists in order to facilitate
4527 this. These flags are in the form of key / value pairs --- much like a
4528 dictionary --- making it easy for any subsystem who cares about a flag to
4531 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
4532 Each triplet has the following form:
4534 - The first element is a *behavior* flag, which specifies the behavior
4535 when two (or more) modules are merged together, and it encounters two
4536 (or more) metadata with the same ID. The supported behaviors are
4538 - The second element is a metadata string that is a unique ID for the
4539 metadata. Each module may only have one flag entry for each unique ID (not
4540 including entries with the **Require** behavior).
4541 - The third element is the value of the flag.
4543 When two (or more) modules are merged together, the resulting
4544 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
4545 each unique metadata ID string, there will be exactly one entry in the merged
4546 modules ``llvm.module.flags`` metadata table, and the value for that entry will
4547 be determined by the merge behavior flag, as described below. The only exception
4548 is that entries with the *Require* behavior are always preserved.
4550 The following behaviors are supported:
4561 Emits an error if two values disagree, otherwise the resulting value
4562 is that of the operands.
4566 Emits a warning if two values disagree. The result value will be the
4567 operand for the flag from the first module being linked.
4571 Adds a requirement that another module flag be present and have a
4572 specified value after linking is performed. The value must be a
4573 metadata pair, where the first element of the pair is the ID of the
4574 module flag to be restricted, and the second element of the pair is
4575 the value the module flag should be restricted to. This behavior can
4576 be used to restrict the allowable results (via triggering of an
4577 error) of linking IDs with the **Override** behavior.
4581 Uses the specified value, regardless of the behavior or value of the
4582 other module. If both modules specify **Override**, but the values
4583 differ, an error will be emitted.
4587 Appends the two values, which are required to be metadata nodes.
4591 Appends the two values, which are required to be metadata
4592 nodes. However, duplicate entries in the second list are dropped
4593 during the append operation.
4595 It is an error for a particular unique flag ID to have multiple behaviors,
4596 except in the case of **Require** (which adds restrictions on another metadata
4597 value) or **Override**.
4599 An example of module flags:
4601 .. code-block:: llvm
4603 !0 = !{ i32 1, !"foo", i32 1 }
4604 !1 = !{ i32 4, !"bar", i32 37 }
4605 !2 = !{ i32 2, !"qux", i32 42 }
4606 !3 = !{ i32 3, !"qux",
4611 !llvm.module.flags = !{ !0, !1, !2, !3 }
4613 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
4614 if two or more ``!"foo"`` flags are seen is to emit an error if their
4615 values are not equal.
4617 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
4618 behavior if two or more ``!"bar"`` flags are seen is to use the value
4621 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
4622 behavior if two or more ``!"qux"`` flags are seen is to emit a
4623 warning if their values are not equal.
4625 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
4631 The behavior is to emit an error if the ``llvm.module.flags`` does not
4632 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
4635 Objective-C Garbage Collection Module Flags Metadata
4636 ----------------------------------------------------
4638 On the Mach-O platform, Objective-C stores metadata about garbage
4639 collection in a special section called "image info". The metadata
4640 consists of a version number and a bitmask specifying what types of
4641 garbage collection are supported (if any) by the file. If two or more
4642 modules are linked together their garbage collection metadata needs to
4643 be merged rather than appended together.
4645 The Objective-C garbage collection module flags metadata consists of the
4646 following key-value pairs:
4655 * - ``Objective-C Version``
4656 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
4658 * - ``Objective-C Image Info Version``
4659 - **[Required]** --- The version of the image info section. Currently
4662 * - ``Objective-C Image Info Section``
4663 - **[Required]** --- The section to place the metadata. Valid values are
4664 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
4665 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
4666 Objective-C ABI version 2.
4668 * - ``Objective-C Garbage Collection``
4669 - **[Required]** --- Specifies whether garbage collection is supported or
4670 not. Valid values are 0, for no garbage collection, and 2, for garbage
4671 collection supported.
4673 * - ``Objective-C GC Only``
4674 - **[Optional]** --- Specifies that only garbage collection is supported.
4675 If present, its value must be 6. This flag requires that the
4676 ``Objective-C Garbage Collection`` flag have the value 2.
4678 Some important flag interactions:
4680 - If a module with ``Objective-C Garbage Collection`` set to 0 is
4681 merged with a module with ``Objective-C Garbage Collection`` set to
4682 2, then the resulting module has the
4683 ``Objective-C Garbage Collection`` flag set to 0.
4684 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
4685 merged with a module with ``Objective-C GC Only`` set to 6.
4687 Automatic Linker Flags Module Flags Metadata
4688 --------------------------------------------
4690 Some targets support embedding flags to the linker inside individual object
4691 files. Typically this is used in conjunction with language extensions which
4692 allow source files to explicitly declare the libraries they depend on, and have
4693 these automatically be transmitted to the linker via object files.
4695 These flags are encoded in the IR using metadata in the module flags section,
4696 using the ``Linker Options`` key. The merge behavior for this flag is required
4697 to be ``AppendUnique``, and the value for the key is expected to be a metadata
4698 node which should be a list of other metadata nodes, each of which should be a
4699 list of metadata strings defining linker options.
4701 For example, the following metadata section specifies two separate sets of
4702 linker options, presumably to link against ``libz`` and the ``Cocoa``
4705 !0 = !{ i32 6, !"Linker Options",
4708 !{ !"-framework", !"Cocoa" } } }
4709 !llvm.module.flags = !{ !0 }
4711 The metadata encoding as lists of lists of options, as opposed to a collapsed
4712 list of options, is chosen so that the IR encoding can use multiple option
4713 strings to specify e.g., a single library, while still having that specifier be
4714 preserved as an atomic element that can be recognized by a target specific
4715 assembly writer or object file emitter.
4717 Each individual option is required to be either a valid option for the target's
4718 linker, or an option that is reserved by the target specific assembly writer or
4719 object file emitter. No other aspect of these options is defined by the IR.
4721 C type width Module Flags Metadata
4722 ----------------------------------
4724 The ARM backend emits a section into each generated object file describing the
4725 options that it was compiled with (in a compiler-independent way) to prevent
4726 linking incompatible objects, and to allow automatic library selection. Some
4727 of these options are not visible at the IR level, namely wchar_t width and enum
4730 To pass this information to the backend, these options are encoded in module
4731 flags metadata, using the following key-value pairs:
4741 - * 0 --- sizeof(wchar_t) == 4
4742 * 1 --- sizeof(wchar_t) == 2
4745 - * 0 --- Enums are at least as large as an ``int``.
4746 * 1 --- Enums are stored in the smallest integer type which can
4747 represent all of its values.
4749 For example, the following metadata section specifies that the module was
4750 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
4751 enum is the smallest type which can represent all of its values::
4753 !llvm.module.flags = !{!0, !1}
4754 !0 = !{i32 1, !"short_wchar", i32 1}
4755 !1 = !{i32 1, !"short_enum", i32 0}
4757 .. _intrinsicglobalvariables:
4759 Intrinsic Global Variables
4760 ==========================
4762 LLVM has a number of "magic" global variables that contain data that
4763 affect code generation or other IR semantics. These are documented here.
4764 All globals of this sort should have a section specified as
4765 "``llvm.metadata``". This section and all globals that start with
4766 "``llvm.``" are reserved for use by LLVM.
4770 The '``llvm.used``' Global Variable
4771 -----------------------------------
4773 The ``@llvm.used`` global is an array which has
4774 :ref:`appending linkage <linkage_appending>`. This array contains a list of
4775 pointers to named global variables, functions and aliases which may optionally
4776 have a pointer cast formed of bitcast or getelementptr. For example, a legal
4779 .. code-block:: llvm
4784 @llvm.used = appending global [2 x i8*] [
4786 i8* bitcast (i32* @Y to i8*)
4787 ], section "llvm.metadata"
4789 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4790 and linker are required to treat the symbol as if there is a reference to the
4791 symbol that it cannot see (which is why they have to be named). For example, if
4792 a variable has internal linkage and no references other than that from the
4793 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4794 references from inline asms and other things the compiler cannot "see", and
4795 corresponds to "``attribute((used))``" in GNU C.
4797 On some targets, the code generator must emit a directive to the
4798 assembler or object file to prevent the assembler and linker from
4799 molesting the symbol.
4801 .. _gv_llvmcompilerused:
4803 The '``llvm.compiler.used``' Global Variable
4804 --------------------------------------------
4806 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4807 directive, except that it only prevents the compiler from touching the
4808 symbol. On targets that support it, this allows an intelligent linker to
4809 optimize references to the symbol without being impeded as it would be
4812 This is a rare construct that should only be used in rare circumstances,
4813 and should not be exposed to source languages.
4815 .. _gv_llvmglobalctors:
4817 The '``llvm.global_ctors``' Global Variable
4818 -------------------------------------------
4820 .. code-block:: llvm
4822 %0 = type { i32, void ()*, i8* }
4823 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4825 The ``@llvm.global_ctors`` array contains a list of constructor
4826 functions, priorities, and an optional associated global or function.
4827 The functions referenced by this array will be called in ascending order
4828 of priority (i.e. lowest first) when the module is loaded. The order of
4829 functions with the same priority is not defined.
4831 If the third field is present, non-null, and points to a global variable
4832 or function, the initializer function will only run if the associated
4833 data from the current module is not discarded.
4835 .. _llvmglobaldtors:
4837 The '``llvm.global_dtors``' Global Variable
4838 -------------------------------------------
4840 .. code-block:: llvm
4842 %0 = type { i32, void ()*, i8* }
4843 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4845 The ``@llvm.global_dtors`` array contains a list of destructor
4846 functions, priorities, and an optional associated global or function.
4847 The functions referenced by this array will be called in descending
4848 order of priority (i.e. highest first) when the module is unloaded. The
4849 order of functions with the same priority is not defined.
4851 If the third field is present, non-null, and points to a global variable
4852 or function, the destructor function will only run if the associated
4853 data from the current module is not discarded.
4855 Instruction Reference
4856 =====================
4858 The LLVM instruction set consists of several different classifications
4859 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4860 instructions <binaryops>`, :ref:`bitwise binary
4861 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4862 :ref:`other instructions <otherops>`.
4866 Terminator Instructions
4867 -----------------------
4869 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4870 program ends with a "Terminator" instruction, which indicates which
4871 block should be executed after the current block is finished. These
4872 terminator instructions typically yield a '``void``' value: they produce
4873 control flow, not values (the one exception being the
4874 ':ref:`invoke <i_invoke>`' instruction).
4876 The terminator instructions are: ':ref:`ret <i_ret>`',
4877 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4878 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4879 ':ref:`resume <i_resume>`', ':ref:`catchpad <i_catchpad>`',
4880 ':ref:`catchendpad <i_catchendpad>`',
4881 ':ref:`catchret <i_catchret>`',
4882 ':ref:`cleanupendpad <i_cleanupendpad>`',
4883 ':ref:`cleanupret <i_cleanupret>`',
4884 ':ref:`terminatepad <i_terminatepad>`',
4885 and ':ref:`unreachable <i_unreachable>`'.
4889 '``ret``' Instruction
4890 ^^^^^^^^^^^^^^^^^^^^^
4897 ret <type> <value> ; Return a value from a non-void function
4898 ret void ; Return from void function
4903 The '``ret``' instruction is used to return control flow (and optionally
4904 a value) from a function back to the caller.
4906 There are two forms of the '``ret``' instruction: one that returns a
4907 value and then causes control flow, and one that just causes control
4913 The '``ret``' instruction optionally accepts a single argument, the
4914 return value. The type of the return value must be a ':ref:`first
4915 class <t_firstclass>`' type.
4917 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4918 return type and contains a '``ret``' instruction with no return value or
4919 a return value with a type that does not match its type, or if it has a
4920 void return type and contains a '``ret``' instruction with a return
4926 When the '``ret``' instruction is executed, control flow returns back to
4927 the calling function's context. If the caller is a
4928 ":ref:`call <i_call>`" instruction, execution continues at the
4929 instruction after the call. If the caller was an
4930 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4931 beginning of the "normal" destination block. If the instruction returns
4932 a value, that value shall set the call or invoke instruction's return
4938 .. code-block:: llvm
4940 ret i32 5 ; Return an integer value of 5
4941 ret void ; Return from a void function
4942 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4946 '``br``' Instruction
4947 ^^^^^^^^^^^^^^^^^^^^
4954 br i1 <cond>, label <iftrue>, label <iffalse>
4955 br label <dest> ; Unconditional branch
4960 The '``br``' instruction is used to cause control flow to transfer to a
4961 different basic block in the current function. There are two forms of
4962 this instruction, corresponding to a conditional branch and an
4963 unconditional branch.
4968 The conditional branch form of the '``br``' instruction takes a single
4969 '``i1``' value and two '``label``' values. The unconditional form of the
4970 '``br``' instruction takes a single '``label``' value as a target.
4975 Upon execution of a conditional '``br``' instruction, the '``i1``'
4976 argument is evaluated. If the value is ``true``, control flows to the
4977 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4978 to the '``iffalse``' ``label`` argument.
4983 .. code-block:: llvm
4986 %cond = icmp eq i32 %a, %b
4987 br i1 %cond, label %IfEqual, label %IfUnequal
4995 '``switch``' Instruction
4996 ^^^^^^^^^^^^^^^^^^^^^^^^
5003 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
5008 The '``switch``' instruction is used to transfer control flow to one of
5009 several different places. It is a generalization of the '``br``'
5010 instruction, allowing a branch to occur to one of many possible
5016 The '``switch``' instruction uses three parameters: an integer
5017 comparison value '``value``', a default '``label``' destination, and an
5018 array of pairs of comparison value constants and '``label``'s. The table
5019 is not allowed to contain duplicate constant entries.
5024 The ``switch`` instruction specifies a table of values and destinations.
5025 When the '``switch``' instruction is executed, this table is searched
5026 for the given value. If the value is found, control flow is transferred
5027 to the corresponding destination; otherwise, control flow is transferred
5028 to the default destination.
5033 Depending on properties of the target machine and the particular
5034 ``switch`` instruction, this instruction may be code generated in
5035 different ways. For example, it could be generated as a series of
5036 chained conditional branches or with a lookup table.
5041 .. code-block:: llvm
5043 ; Emulate a conditional br instruction
5044 %Val = zext i1 %value to i32
5045 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
5047 ; Emulate an unconditional br instruction
5048 switch i32 0, label %dest [ ]
5050 ; Implement a jump table:
5051 switch i32 %val, label %otherwise [ i32 0, label %onzero
5053 i32 2, label %ontwo ]
5057 '``indirectbr``' Instruction
5058 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5065 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
5070 The '``indirectbr``' instruction implements an indirect branch to a
5071 label within the current function, whose address is specified by
5072 "``address``". Address must be derived from a
5073 :ref:`blockaddress <blockaddress>` constant.
5078 The '``address``' argument is the address of the label to jump to. The
5079 rest of the arguments indicate the full set of possible destinations
5080 that the address may point to. Blocks are allowed to occur multiple
5081 times in the destination list, though this isn't particularly useful.
5083 This destination list is required so that dataflow analysis has an
5084 accurate understanding of the CFG.
5089 Control transfers to the block specified in the address argument. All
5090 possible destination blocks must be listed in the label list, otherwise
5091 this instruction has undefined behavior. This implies that jumps to
5092 labels defined in other functions have undefined behavior as well.
5097 This is typically implemented with a jump through a register.
5102 .. code-block:: llvm
5104 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
5108 '``invoke``' Instruction
5109 ^^^^^^^^^^^^^^^^^^^^^^^^
5116 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
5117 [operand bundles] to label <normal label> unwind label <exception label>
5122 The '``invoke``' instruction causes control to transfer to a specified
5123 function, with the possibility of control flow transfer to either the
5124 '``normal``' label or the '``exception``' label. If the callee function
5125 returns with the "``ret``" instruction, control flow will return to the
5126 "normal" label. If the callee (or any indirect callees) returns via the
5127 ":ref:`resume <i_resume>`" instruction or other exception handling
5128 mechanism, control is interrupted and continued at the dynamically
5129 nearest "exception" label.
5131 The '``exception``' label is a `landing
5132 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
5133 '``exception``' label is required to have the
5134 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
5135 information about the behavior of the program after unwinding happens,
5136 as its first non-PHI instruction. The restrictions on the
5137 "``landingpad``" instruction's tightly couples it to the "``invoke``"
5138 instruction, so that the important information contained within the
5139 "``landingpad``" instruction can't be lost through normal code motion.
5144 This instruction requires several arguments:
5146 #. The optional "cconv" marker indicates which :ref:`calling
5147 convention <callingconv>` the call should use. If none is
5148 specified, the call defaults to using C calling conventions.
5149 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5150 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5152 #. '``ptr to function ty``': shall be the signature of the pointer to
5153 function value being invoked. In most cases, this is a direct
5154 function invocation, but indirect ``invoke``'s are just as possible,
5155 branching off an arbitrary pointer to function value.
5156 #. '``function ptr val``': An LLVM value containing a pointer to a
5157 function to be invoked.
5158 #. '``function args``': argument list whose types match the function
5159 signature argument types and parameter attributes. All arguments must
5160 be of :ref:`first class <t_firstclass>` type. If the function signature
5161 indicates the function accepts a variable number of arguments, the
5162 extra arguments can be specified.
5163 #. '``normal label``': the label reached when the called function
5164 executes a '``ret``' instruction.
5165 #. '``exception label``': the label reached when a callee returns via
5166 the :ref:`resume <i_resume>` instruction or other exception handling
5168 #. The optional :ref:`function attributes <fnattrs>` list. Only
5169 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5170 attributes are valid here.
5171 #. The optional :ref:`operand bundles <opbundles>` list.
5176 This instruction is designed to operate as a standard '``call``'
5177 instruction in most regards. The primary difference is that it
5178 establishes an association with a label, which is used by the runtime
5179 library to unwind the stack.
5181 This instruction is used in languages with destructors to ensure that
5182 proper cleanup is performed in the case of either a ``longjmp`` or a
5183 thrown exception. Additionally, this is important for implementation of
5184 '``catch``' clauses in high-level languages that support them.
5186 For the purposes of the SSA form, the definition of the value returned
5187 by the '``invoke``' instruction is deemed to occur on the edge from the
5188 current block to the "normal" label. If the callee unwinds then no
5189 return value is available.
5194 .. code-block:: llvm
5196 %retval = invoke i32 @Test(i32 15) to label %Continue
5197 unwind label %TestCleanup ; i32:retval set
5198 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
5199 unwind label %TestCleanup ; i32:retval set
5203 '``resume``' Instruction
5204 ^^^^^^^^^^^^^^^^^^^^^^^^
5211 resume <type> <value>
5216 The '``resume``' instruction is a terminator instruction that has no
5222 The '``resume``' instruction requires one argument, which must have the
5223 same type as the result of any '``landingpad``' instruction in the same
5229 The '``resume``' instruction resumes propagation of an existing
5230 (in-flight) exception whose unwinding was interrupted with a
5231 :ref:`landingpad <i_landingpad>` instruction.
5236 .. code-block:: llvm
5238 resume { i8*, i32 } %exn
5242 '``catchpad``' Instruction
5243 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5250 <resultval> = catchpad [<args>*]
5251 to label <normal label> unwind label <exception label>
5256 The '``catchpad``' instruction is used by `LLVM's exception handling
5257 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5258 is a catch block --- one where a personality routine attempts to transfer
5259 control to catch an exception.
5260 The ``args`` correspond to whatever information the personality
5261 routine requires to know if this is an appropriate place to catch the
5262 exception. Control is transfered to the ``exception`` label if the
5263 ``catchpad`` is not an appropriate handler for the in-flight exception.
5264 The ``normal`` label should contain the code found in the ``catch``
5265 portion of a ``try``/``catch`` sequence. The ``resultval`` has the type
5266 :ref:`token <t_token>` and is used to match the ``catchpad`` to
5267 corresponding :ref:`catchrets <i_catchret>`.
5272 The instruction takes a list of arbitrary values which are interpreted
5273 by the :ref:`personality function <personalityfn>`.
5275 The ``catchpad`` must be provided a ``normal`` label to transfer control
5276 to if the ``catchpad`` matches the exception and an ``exception``
5277 label to transfer control to if it doesn't.
5282 When the call stack is being unwound due to an exception being thrown,
5283 the exception is compared against the ``args``. If it doesn't match,
5284 then control is transfered to the ``exception`` basic block.
5285 As with calling conventions, how the personality function results are
5286 represented in LLVM IR is target specific.
5288 The ``catchpad`` instruction has several restrictions:
5290 - A catch block is a basic block which is the unwind destination of
5291 an exceptional instruction.
5292 - A catch block must have a '``catchpad``' instruction as its
5293 first non-PHI instruction.
5294 - A catch block's ``exception`` edge must refer to a catch block or a
5296 - There can be only one '``catchpad``' instruction within the
5298 - A basic block that is not a catch block may not include a
5299 '``catchpad``' instruction.
5300 - A catch block which has another catch block as a predecessor may not have
5301 any other predecessors.
5302 - It is undefined behavior for control to transfer from a ``catchpad`` to a
5303 ``ret`` without first executing a ``catchret`` that consumes the
5304 ``catchpad`` or unwinding through its ``catchendpad``.
5305 - It is undefined behavior for control to transfer from a ``catchpad`` to
5306 itself without first executing a ``catchret`` that consumes the
5307 ``catchpad`` or unwinding through its ``catchendpad``.
5312 .. code-block:: llvm
5314 ;; A catch block which can catch an integer.
5315 %tok = catchpad [i8** @_ZTIi]
5316 to label %int.handler unwind label %terminate
5320 '``catchendpad``' Instruction
5321 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5328 catchendpad unwind label <nextaction>
5329 catchendpad unwind to caller
5334 The '``catchendpad``' instruction is used by `LLVM's exception handling
5335 system <ExceptionHandling.html#overview>`_ to communicate to the
5336 :ref:`personality function <personalityfn>` which invokes are associated
5337 with a chain of :ref:`catchpad <i_catchpad>` instructions; propagating an
5338 exception out of a catch handler is represented by unwinding through its
5339 ``catchendpad``. Unwinding to the outer scope when a chain of catch handlers
5340 do not handle an exception is also represented by unwinding through their
5343 The ``nextaction`` label indicates where control should transfer to if
5344 none of the ``catchpad`` instructions are suitable for catching the
5345 in-flight exception.
5347 If a ``nextaction`` label is not present, the instruction unwinds out of
5348 its parent function. The
5349 :ref:`personality function <personalityfn>` will continue processing
5350 exception handling actions in the caller.
5355 The instruction optionally takes a label, ``nextaction``, indicating
5356 where control should transfer to if none of the preceding
5357 ``catchpad`` instructions are suitable for the in-flight exception.
5362 When the call stack is being unwound due to an exception being thrown
5363 and none of the constituent ``catchpad`` instructions match, then
5364 control is transfered to ``nextaction`` if it is present. If it is not
5365 present, control is transfered to the caller.
5367 The ``catchendpad`` instruction has several restrictions:
5369 - A catch-end block is a basic block which is the unwind destination of
5370 an exceptional instruction.
5371 - A catch-end block must have a '``catchendpad``' instruction as its
5372 first non-PHI instruction.
5373 - There can be only one '``catchendpad``' instruction within the
5375 - A basic block that is not a catch-end block may not include a
5376 '``catchendpad``' instruction.
5377 - Exactly one catch block may unwind to a ``catchendpad``.
5378 - It is undefined behavior to execute a ``catchendpad`` if none of the
5379 '``catchpad``'s chained to it have been executed.
5380 - It is undefined behavior to execute a ``catchendpad`` twice without an
5381 intervening execution of one or more of the '``catchpad``'s chained to it.
5382 - It is undefined behavior to execute a ``catchendpad`` if, after the most
5383 recent execution of the normal successor edge of any ``catchpad`` chained
5384 to it, some ``catchret`` consuming that ``catchpad`` has already been
5386 - It is undefined behavior to execute a ``catchendpad`` if, after the most
5387 recent execution of the normal successor edge of any ``catchpad`` chained
5388 to it, any other ``catchpad`` or ``cleanuppad`` has been executed but has
5389 not had a corresponding
5390 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5395 .. code-block:: llvm
5397 catchendpad unwind label %terminate
5398 catchendpad unwind to caller
5402 '``catchret``' Instruction
5403 ^^^^^^^^^^^^^^^^^^^^^^^^^^
5410 catchret <value> to label <normal>
5415 The '``catchret``' instruction is a terminator instruction that has a
5422 The first argument to a '``catchret``' indicates which ``catchpad`` it
5423 exits. It must be a :ref:`catchpad <i_catchpad>`.
5424 The second argument to a '``catchret``' specifies where control will
5430 The '``catchret``' instruction ends the existing (in-flight) exception
5431 whose unwinding was interrupted with a
5432 :ref:`catchpad <i_catchpad>` instruction.
5433 The :ref:`personality function <personalityfn>` gets a chance to execute
5434 arbitrary code to, for example, run a C++ destructor.
5435 Control then transfers to ``normal``.
5436 It may be passed an optional, personality specific, value.
5438 It is undefined behavior to execute a ``catchret`` whose ``catchpad`` has
5441 It is undefined behavior to execute a ``catchret`` if, after the most recent
5442 execution of its ``catchpad``, some ``catchret`` or ``catchendpad`` linked
5443 to the same ``catchpad`` has already been executed.
5445 It is undefined behavior to execute a ``catchret`` if, after the most recent
5446 execution of its ``catchpad``, any other ``catchpad`` or ``cleanuppad`` has
5447 been executed but has not had a corresponding
5448 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5453 .. code-block:: llvm
5455 catchret %catch label %continue
5457 .. _i_cleanupendpad:
5459 '``cleanupendpad``' Instruction
5460 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5467 cleanupendpad <value> unwind label <nextaction>
5468 cleanupendpad <value> unwind to caller
5473 The '``cleanupendpad``' instruction is used by `LLVM's exception handling
5474 system <ExceptionHandling.html#overview>`_ to communicate to the
5475 :ref:`personality function <personalityfn>` which invokes are associated
5476 with a :ref:`cleanuppad <i_cleanuppad>` instructions; propagating an exception
5477 out of a cleanup is represented by unwinding through its ``cleanupendpad``.
5479 The ``nextaction`` label indicates where control should unwind to next, in the
5480 event that a cleanup is exited by means of an(other) exception being raised.
5482 If a ``nextaction`` label is not present, the instruction unwinds out of
5483 its parent function. The
5484 :ref:`personality function <personalityfn>` will continue processing
5485 exception handling actions in the caller.
5490 The '``cleanupendpad``' instruction requires one argument, which indicates
5491 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5492 It also has an optional successor, ``nextaction``, indicating where control
5498 When and exception propagates to a ``cleanupendpad``, control is transfered to
5499 ``nextaction`` if it is present. If it is not present, control is transfered to
5502 The ``cleanupendpad`` instruction has several restrictions:
5504 - A cleanup-end block is a basic block which is the unwind destination of
5505 an exceptional instruction.
5506 - A cleanup-end block must have a '``cleanupendpad``' instruction as its
5507 first non-PHI instruction.
5508 - There can be only one '``cleanupendpad``' instruction within the
5510 - A basic block that is not a cleanup-end block may not include a
5511 '``cleanupendpad``' instruction.
5512 - It is undefined behavior to execute a ``cleanupendpad`` whose ``cleanuppad``
5513 has not been executed.
5514 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5515 recent execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5516 consuming the same ``cleanuppad`` has already been executed.
5517 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5518 recent execution of its ``cleanuppad``, any other ``cleanuppad`` or
5519 ``catchpad`` has been executed but has not had a corresponding
5520 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5525 .. code-block:: llvm
5527 cleanupendpad %cleanup unwind label %terminate
5528 cleanupendpad %cleanup unwind to caller
5532 '``cleanupret``' Instruction
5533 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5540 cleanupret <value> unwind label <continue>
5541 cleanupret <value> unwind to caller
5546 The '``cleanupret``' instruction is a terminator instruction that has
5547 an optional successor.
5553 The '``cleanupret``' instruction requires one argument, which indicates
5554 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5555 It also has an optional successor, ``continue``.
5560 The '``cleanupret``' instruction indicates to the
5561 :ref:`personality function <personalityfn>` that one
5562 :ref:`cleanuppad <i_cleanuppad>` it transferred control to has ended.
5563 It transfers control to ``continue`` or unwinds out of the function.
5565 It is undefined behavior to execute a ``cleanupret`` whose ``cleanuppad`` has
5568 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5569 execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5570 consuming the same ``cleanuppad`` has already been executed.
5572 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5573 execution of its ``cleanuppad``, any other ``cleanuppad`` or ``catchpad`` has
5574 been executed but has not had a corresponding
5575 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5580 .. code-block:: llvm
5582 cleanupret %cleanup unwind to caller
5583 cleanupret %cleanup unwind label %continue
5587 '``terminatepad``' Instruction
5588 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5595 terminatepad [<args>*] unwind label <exception label>
5596 terminatepad [<args>*] unwind to caller
5601 The '``terminatepad``' instruction is used by `LLVM's exception handling
5602 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5603 is a terminate block --- one where a personality routine may decide to
5604 terminate the program.
5605 The ``args`` correspond to whatever information the personality
5606 routine requires to know if this is an appropriate place to terminate the
5607 program. Control is transferred to the ``exception`` label if the
5608 personality routine decides not to terminate the program for the
5609 in-flight exception.
5614 The instruction takes a list of arbitrary values which are interpreted
5615 by the :ref:`personality function <personalityfn>`.
5617 The ``terminatepad`` may be given an ``exception`` label to
5618 transfer control to if the in-flight exception matches the ``args``.
5623 When the call stack is being unwound due to an exception being thrown,
5624 the exception is compared against the ``args``. If it matches,
5625 then control is transfered to the ``exception`` basic block. Otherwise,
5626 the program is terminated via personality-specific means. Typically,
5627 the first argument to ``terminatepad`` specifies what function the
5628 personality should defer to in order to terminate the program.
5630 The ``terminatepad`` instruction has several restrictions:
5632 - A terminate block is a basic block which is the unwind destination of
5633 an exceptional instruction.
5634 - A terminate block must have a '``terminatepad``' instruction as its
5635 first non-PHI instruction.
5636 - There can be only one '``terminatepad``' instruction within the
5638 - A basic block that is not a terminate block may not include a
5639 '``terminatepad``' instruction.
5644 .. code-block:: llvm
5646 ;; A terminate block which only permits integers.
5647 terminatepad [i8** @_ZTIi] unwind label %continue
5651 '``unreachable``' Instruction
5652 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5664 The '``unreachable``' instruction has no defined semantics. This
5665 instruction is used to inform the optimizer that a particular portion of
5666 the code is not reachable. This can be used to indicate that the code
5667 after a no-return function cannot be reached, and other facts.
5672 The '``unreachable``' instruction has no defined semantics.
5679 Binary operators are used to do most of the computation in a program.
5680 They require two operands of the same type, execute an operation on
5681 them, and produce a single value. The operands might represent multiple
5682 data, as is the case with the :ref:`vector <t_vector>` data type. The
5683 result value has the same type as its operands.
5685 There are several different binary operators:
5689 '``add``' Instruction
5690 ^^^^^^^^^^^^^^^^^^^^^
5697 <result> = add <ty> <op1>, <op2> ; yields ty:result
5698 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
5699 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
5700 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
5705 The '``add``' instruction returns the sum of its two operands.
5710 The two arguments to the '``add``' instruction must be
5711 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5712 arguments must have identical types.
5717 The value produced is the integer sum of the two operands.
5719 If the sum has unsigned overflow, the result returned is the
5720 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5723 Because LLVM integers use a two's complement representation, this
5724 instruction is appropriate for both signed and unsigned integers.
5726 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5727 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5728 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
5729 unsigned and/or signed overflow, respectively, occurs.
5734 .. code-block:: llvm
5736 <result> = add i32 4, %var ; yields i32:result = 4 + %var
5740 '``fadd``' Instruction
5741 ^^^^^^^^^^^^^^^^^^^^^^
5748 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5753 The '``fadd``' instruction returns the sum of its two operands.
5758 The two arguments to the '``fadd``' instruction must be :ref:`floating
5759 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5760 Both arguments must have identical types.
5765 The value produced is the floating point sum of the two operands. This
5766 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
5767 which are optimization hints to enable otherwise unsafe floating point
5773 .. code-block:: llvm
5775 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
5777 '``sub``' Instruction
5778 ^^^^^^^^^^^^^^^^^^^^^
5785 <result> = sub <ty> <op1>, <op2> ; yields ty:result
5786 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
5787 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
5788 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
5793 The '``sub``' instruction returns the difference of its two operands.
5795 Note that the '``sub``' instruction is used to represent the '``neg``'
5796 instruction present in most other intermediate representations.
5801 The two arguments to the '``sub``' instruction must be
5802 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5803 arguments must have identical types.
5808 The value produced is the integer difference of the two operands.
5810 If the difference has unsigned overflow, the result returned is the
5811 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5814 Because LLVM integers use a two's complement representation, this
5815 instruction is appropriate for both signed and unsigned integers.
5817 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5818 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5819 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
5820 unsigned and/or signed overflow, respectively, occurs.
5825 .. code-block:: llvm
5827 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
5828 <result> = sub i32 0, %val ; yields i32:result = -%var
5832 '``fsub``' Instruction
5833 ^^^^^^^^^^^^^^^^^^^^^^
5840 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5845 The '``fsub``' instruction returns the difference of its two operands.
5847 Note that the '``fsub``' instruction is used to represent the '``fneg``'
5848 instruction present in most other intermediate representations.
5853 The two arguments to the '``fsub``' instruction must be :ref:`floating
5854 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5855 Both arguments must have identical types.
5860 The value produced is the floating point difference of the two operands.
5861 This instruction can also take any number of :ref:`fast-math
5862 flags <fastmath>`, which are optimization hints to enable otherwise
5863 unsafe floating point optimizations:
5868 .. code-block:: llvm
5870 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
5871 <result> = fsub float -0.0, %val ; yields float:result = -%var
5873 '``mul``' Instruction
5874 ^^^^^^^^^^^^^^^^^^^^^
5881 <result> = mul <ty> <op1>, <op2> ; yields ty:result
5882 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
5883 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
5884 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
5889 The '``mul``' instruction returns the product of its two operands.
5894 The two arguments to the '``mul``' instruction must be
5895 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5896 arguments must have identical types.
5901 The value produced is the integer product of the two operands.
5903 If the result of the multiplication has unsigned overflow, the result
5904 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
5905 bit width of the result.
5907 Because LLVM integers use a two's complement representation, and the
5908 result is the same width as the operands, this instruction returns the
5909 correct result for both signed and unsigned integers. If a full product
5910 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
5911 sign-extended or zero-extended as appropriate to the width of the full
5914 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5915 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5916 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
5917 unsigned and/or signed overflow, respectively, occurs.
5922 .. code-block:: llvm
5924 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
5928 '``fmul``' Instruction
5929 ^^^^^^^^^^^^^^^^^^^^^^
5936 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5941 The '``fmul``' instruction returns the product of its two operands.
5946 The two arguments to the '``fmul``' instruction must be :ref:`floating
5947 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5948 Both arguments must have identical types.
5953 The value produced is the floating point product of the two operands.
5954 This instruction can also take any number of :ref:`fast-math
5955 flags <fastmath>`, which are optimization hints to enable otherwise
5956 unsafe floating point optimizations:
5961 .. code-block:: llvm
5963 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
5965 '``udiv``' Instruction
5966 ^^^^^^^^^^^^^^^^^^^^^^
5973 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
5974 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
5979 The '``udiv``' instruction returns the quotient of its two operands.
5984 The two arguments to the '``udiv``' instruction must be
5985 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5986 arguments must have identical types.
5991 The value produced is the unsigned integer quotient of the two operands.
5993 Note that unsigned integer division and signed integer division are
5994 distinct operations; for signed integer division, use '``sdiv``'.
5996 Division by zero leads to undefined behavior.
5998 If the ``exact`` keyword is present, the result value of the ``udiv`` is
5999 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
6000 such, "((a udiv exact b) mul b) == a").
6005 .. code-block:: llvm
6007 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
6009 '``sdiv``' Instruction
6010 ^^^^^^^^^^^^^^^^^^^^^^
6017 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
6018 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
6023 The '``sdiv``' instruction returns the quotient of its two operands.
6028 The two arguments to the '``sdiv``' instruction must be
6029 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6030 arguments must have identical types.
6035 The value produced is the signed integer quotient of the two operands
6036 rounded towards zero.
6038 Note that signed integer division and unsigned integer division are
6039 distinct operations; for unsigned integer division, use '``udiv``'.
6041 Division by zero leads to undefined behavior. Overflow also leads to
6042 undefined behavior; this is a rare case, but can occur, for example, by
6043 doing a 32-bit division of -2147483648 by -1.
6045 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
6046 a :ref:`poison value <poisonvalues>` if the result would be rounded.
6051 .. code-block:: llvm
6053 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
6057 '``fdiv``' Instruction
6058 ^^^^^^^^^^^^^^^^^^^^^^
6065 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6070 The '``fdiv``' instruction returns the quotient of its two operands.
6075 The two arguments to the '``fdiv``' instruction must be :ref:`floating
6076 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6077 Both arguments must have identical types.
6082 The value produced is the floating point quotient of the two operands.
6083 This instruction can also take any number of :ref:`fast-math
6084 flags <fastmath>`, which are optimization hints to enable otherwise
6085 unsafe floating point optimizations:
6090 .. code-block:: llvm
6092 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
6094 '``urem``' Instruction
6095 ^^^^^^^^^^^^^^^^^^^^^^
6102 <result> = urem <ty> <op1>, <op2> ; yields ty:result
6107 The '``urem``' instruction returns the remainder from the unsigned
6108 division of its two arguments.
6113 The two arguments to the '``urem``' instruction must be
6114 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6115 arguments must have identical types.
6120 This instruction returns the unsigned integer *remainder* of a division.
6121 This instruction always performs an unsigned division to get the
6124 Note that unsigned integer remainder and signed integer remainder are
6125 distinct operations; for signed integer remainder, use '``srem``'.
6127 Taking the remainder of a division by zero leads to undefined behavior.
6132 .. code-block:: llvm
6134 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
6136 '``srem``' Instruction
6137 ^^^^^^^^^^^^^^^^^^^^^^
6144 <result> = srem <ty> <op1>, <op2> ; yields ty:result
6149 The '``srem``' instruction returns the remainder from the signed
6150 division of its two operands. This instruction can also take
6151 :ref:`vector <t_vector>` versions of the values in which case the elements
6157 The two arguments to the '``srem``' instruction must be
6158 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6159 arguments must have identical types.
6164 This instruction returns the *remainder* of a division (where the result
6165 is either zero or has the same sign as the dividend, ``op1``), not the
6166 *modulo* operator (where the result is either zero or has the same sign
6167 as the divisor, ``op2``) of a value. For more information about the
6168 difference, see `The Math
6169 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
6170 table of how this is implemented in various languages, please see
6172 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
6174 Note that signed integer remainder and unsigned integer remainder are
6175 distinct operations; for unsigned integer remainder, use '``urem``'.
6177 Taking the remainder of a division by zero leads to undefined behavior.
6178 Overflow also leads to undefined behavior; this is a rare case, but can
6179 occur, for example, by taking the remainder of a 32-bit division of
6180 -2147483648 by -1. (The remainder doesn't actually overflow, but this
6181 rule lets srem be implemented using instructions that return both the
6182 result of the division and the remainder.)
6187 .. code-block:: llvm
6189 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
6193 '``frem``' Instruction
6194 ^^^^^^^^^^^^^^^^^^^^^^
6201 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6206 The '``frem``' instruction returns the remainder from the division of
6212 The two arguments to the '``frem``' instruction must be :ref:`floating
6213 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6214 Both arguments must have identical types.
6219 This instruction returns the *remainder* of a division. The remainder
6220 has the same sign as the dividend. This instruction can also take any
6221 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
6222 to enable otherwise unsafe floating point optimizations:
6227 .. code-block:: llvm
6229 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
6233 Bitwise Binary Operations
6234 -------------------------
6236 Bitwise binary operators are used to do various forms of bit-twiddling
6237 in a program. They are generally very efficient instructions and can
6238 commonly be strength reduced from other instructions. They require two
6239 operands of the same type, execute an operation on them, and produce a
6240 single value. The resulting value is the same type as its operands.
6242 '``shl``' Instruction
6243 ^^^^^^^^^^^^^^^^^^^^^
6250 <result> = shl <ty> <op1>, <op2> ; yields ty:result
6251 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
6252 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
6253 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
6258 The '``shl``' instruction returns the first operand shifted to the left
6259 a specified number of bits.
6264 Both arguments to the '``shl``' instruction must be the same
6265 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6266 '``op2``' is treated as an unsigned value.
6271 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
6272 where ``n`` is the width of the result. If ``op2`` is (statically or
6273 dynamically) equal to or larger than the number of bits in
6274 ``op1``, the result is undefined. If the arguments are vectors, each
6275 vector element of ``op1`` is shifted by the corresponding shift amount
6278 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
6279 value <poisonvalues>` if it shifts out any non-zero bits. If the
6280 ``nsw`` keyword is present, then the shift produces a :ref:`poison
6281 value <poisonvalues>` if it shifts out any bits that disagree with the
6282 resultant sign bit. As such, NUW/NSW have the same semantics as they
6283 would if the shift were expressed as a mul instruction with the same
6284 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
6289 .. code-block:: llvm
6291 <result> = shl i32 4, %var ; yields i32: 4 << %var
6292 <result> = shl i32 4, 2 ; yields i32: 16
6293 <result> = shl i32 1, 10 ; yields i32: 1024
6294 <result> = shl i32 1, 32 ; undefined
6295 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
6297 '``lshr``' Instruction
6298 ^^^^^^^^^^^^^^^^^^^^^^
6305 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
6306 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
6311 The '``lshr``' instruction (logical shift right) returns the first
6312 operand shifted to the right a specified number of bits with zero fill.
6317 Both arguments to the '``lshr``' instruction must be the same
6318 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6319 '``op2``' is treated as an unsigned value.
6324 This instruction always performs a logical shift right operation. The
6325 most significant bits of the result will be filled with zero bits after
6326 the shift. If ``op2`` is (statically or dynamically) equal to or larger
6327 than the number of bits in ``op1``, the result is undefined. If the
6328 arguments are vectors, each vector element of ``op1`` is shifted by the
6329 corresponding shift amount in ``op2``.
6331 If the ``exact`` keyword is present, the result value of the ``lshr`` is
6332 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6338 .. code-block:: llvm
6340 <result> = lshr i32 4, 1 ; yields i32:result = 2
6341 <result> = lshr i32 4, 2 ; yields i32:result = 1
6342 <result> = lshr i8 4, 3 ; yields i8:result = 0
6343 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
6344 <result> = lshr i32 1, 32 ; undefined
6345 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
6347 '``ashr``' Instruction
6348 ^^^^^^^^^^^^^^^^^^^^^^
6355 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
6356 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
6361 The '``ashr``' instruction (arithmetic shift right) returns the first
6362 operand shifted to the right a specified number of bits with sign
6368 Both arguments to the '``ashr``' instruction must be the same
6369 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6370 '``op2``' is treated as an unsigned value.
6375 This instruction always performs an arithmetic shift right operation,
6376 The most significant bits of the result will be filled with the sign bit
6377 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
6378 than the number of bits in ``op1``, the result is undefined. If the
6379 arguments are vectors, each vector element of ``op1`` is shifted by the
6380 corresponding shift amount in ``op2``.
6382 If the ``exact`` keyword is present, the result value of the ``ashr`` is
6383 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6389 .. code-block:: llvm
6391 <result> = ashr i32 4, 1 ; yields i32:result = 2
6392 <result> = ashr i32 4, 2 ; yields i32:result = 1
6393 <result> = ashr i8 4, 3 ; yields i8:result = 0
6394 <result> = ashr i8 -2, 1 ; yields i8:result = -1
6395 <result> = ashr i32 1, 32 ; undefined
6396 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
6398 '``and``' Instruction
6399 ^^^^^^^^^^^^^^^^^^^^^
6406 <result> = and <ty> <op1>, <op2> ; yields ty:result
6411 The '``and``' instruction returns the bitwise logical and of its two
6417 The two arguments to the '``and``' instruction must be
6418 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6419 arguments must have identical types.
6424 The truth table used for the '``and``' instruction is:
6441 .. code-block:: llvm
6443 <result> = and i32 4, %var ; yields i32:result = 4 & %var
6444 <result> = and i32 15, 40 ; yields i32:result = 8
6445 <result> = and i32 4, 8 ; yields i32:result = 0
6447 '``or``' Instruction
6448 ^^^^^^^^^^^^^^^^^^^^
6455 <result> = or <ty> <op1>, <op2> ; yields ty:result
6460 The '``or``' instruction returns the bitwise logical inclusive or of its
6466 The two arguments to the '``or``' instruction must be
6467 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6468 arguments must have identical types.
6473 The truth table used for the '``or``' instruction is:
6492 <result> = or i32 4, %var ; yields i32:result = 4 | %var
6493 <result> = or i32 15, 40 ; yields i32:result = 47
6494 <result> = or i32 4, 8 ; yields i32:result = 12
6496 '``xor``' Instruction
6497 ^^^^^^^^^^^^^^^^^^^^^
6504 <result> = xor <ty> <op1>, <op2> ; yields ty:result
6509 The '``xor``' instruction returns the bitwise logical exclusive or of
6510 its two operands. The ``xor`` is used to implement the "one's
6511 complement" operation, which is the "~" operator in C.
6516 The two arguments to the '``xor``' instruction must be
6517 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6518 arguments must have identical types.
6523 The truth table used for the '``xor``' instruction is:
6540 .. code-block:: llvm
6542 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
6543 <result> = xor i32 15, 40 ; yields i32:result = 39
6544 <result> = xor i32 4, 8 ; yields i32:result = 12
6545 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
6550 LLVM supports several instructions to represent vector operations in a
6551 target-independent manner. These instructions cover the element-access
6552 and vector-specific operations needed to process vectors effectively.
6553 While LLVM does directly support these vector operations, many
6554 sophisticated algorithms will want to use target-specific intrinsics to
6555 take full advantage of a specific target.
6557 .. _i_extractelement:
6559 '``extractelement``' Instruction
6560 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6567 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
6572 The '``extractelement``' instruction extracts a single scalar element
6573 from a vector at a specified index.
6578 The first operand of an '``extractelement``' instruction is a value of
6579 :ref:`vector <t_vector>` type. The second operand is an index indicating
6580 the position from which to extract the element. The index may be a
6581 variable of any integer type.
6586 The result is a scalar of the same type as the element type of ``val``.
6587 Its value is the value at position ``idx`` of ``val``. If ``idx``
6588 exceeds the length of ``val``, the results are undefined.
6593 .. code-block:: llvm
6595 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
6597 .. _i_insertelement:
6599 '``insertelement``' Instruction
6600 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6607 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
6612 The '``insertelement``' instruction inserts a scalar element into a
6613 vector at a specified index.
6618 The first operand of an '``insertelement``' instruction is a value of
6619 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
6620 type must equal the element type of the first operand. The third operand
6621 is an index indicating the position at which to insert the value. The
6622 index may be a variable of any integer type.
6627 The result is a vector of the same type as ``val``. Its element values
6628 are those of ``val`` except at position ``idx``, where it gets the value
6629 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
6635 .. code-block:: llvm
6637 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
6639 .. _i_shufflevector:
6641 '``shufflevector``' Instruction
6642 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6649 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
6654 The '``shufflevector``' instruction constructs a permutation of elements
6655 from two input vectors, returning a vector with the same element type as
6656 the input and length that is the same as the shuffle mask.
6661 The first two operands of a '``shufflevector``' instruction are vectors
6662 with the same type. The third argument is a shuffle mask whose element
6663 type is always 'i32'. The result of the instruction is a vector whose
6664 length is the same as the shuffle mask and whose element type is the
6665 same as the element type of the first two operands.
6667 The shuffle mask operand is required to be a constant vector with either
6668 constant integer or undef values.
6673 The elements of the two input vectors are numbered from left to right
6674 across both of the vectors. The shuffle mask operand specifies, for each
6675 element of the result vector, which element of the two input vectors the
6676 result element gets. The element selector may be undef (meaning "don't
6677 care") and the second operand may be undef if performing a shuffle from
6683 .. code-block:: llvm
6685 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6686 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
6687 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
6688 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
6689 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
6690 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
6691 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6692 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
6694 Aggregate Operations
6695 --------------------
6697 LLVM supports several instructions for working with
6698 :ref:`aggregate <t_aggregate>` values.
6702 '``extractvalue``' Instruction
6703 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6710 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
6715 The '``extractvalue``' instruction extracts the value of a member field
6716 from an :ref:`aggregate <t_aggregate>` value.
6721 The first operand of an '``extractvalue``' instruction is a value of
6722 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The other operands are
6723 constant indices to specify which value to extract in a similar manner
6724 as indices in a '``getelementptr``' instruction.
6726 The major differences to ``getelementptr`` indexing are:
6728 - Since the value being indexed is not a pointer, the first index is
6729 omitted and assumed to be zero.
6730 - At least one index must be specified.
6731 - Not only struct indices but also array indices must be in bounds.
6736 The result is the value at the position in the aggregate specified by
6742 .. code-block:: llvm
6744 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
6748 '``insertvalue``' Instruction
6749 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6756 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
6761 The '``insertvalue``' instruction inserts a value into a member field in
6762 an :ref:`aggregate <t_aggregate>` value.
6767 The first operand of an '``insertvalue``' instruction is a value of
6768 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
6769 a first-class value to insert. The following operands are constant
6770 indices indicating the position at which to insert the value in a
6771 similar manner as indices in a '``extractvalue``' instruction. The value
6772 to insert must have the same type as the value identified by the
6778 The result is an aggregate of the same type as ``val``. Its value is
6779 that of ``val`` except that the value at the position specified by the
6780 indices is that of ``elt``.
6785 .. code-block:: llvm
6787 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
6788 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
6789 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
6793 Memory Access and Addressing Operations
6794 ---------------------------------------
6796 A key design point of an SSA-based representation is how it represents
6797 memory. In LLVM, no memory locations are in SSA form, which makes things
6798 very simple. This section describes how to read, write, and allocate
6803 '``alloca``' Instruction
6804 ^^^^^^^^^^^^^^^^^^^^^^^^
6811 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
6816 The '``alloca``' instruction allocates memory on the stack frame of the
6817 currently executing function, to be automatically released when this
6818 function returns to its caller. The object is always allocated in the
6819 generic address space (address space zero).
6824 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
6825 bytes of memory on the runtime stack, returning a pointer of the
6826 appropriate type to the program. If "NumElements" is specified, it is
6827 the number of elements allocated, otherwise "NumElements" is defaulted
6828 to be one. If a constant alignment is specified, the value result of the
6829 allocation is guaranteed to be aligned to at least that boundary. The
6830 alignment may not be greater than ``1 << 29``. If not specified, or if
6831 zero, the target can choose to align the allocation on any convenient
6832 boundary compatible with the type.
6834 '``type``' may be any sized type.
6839 Memory is allocated; a pointer is returned. The operation is undefined
6840 if there is insufficient stack space for the allocation. '``alloca``'d
6841 memory is automatically released when the function returns. The
6842 '``alloca``' instruction is commonly used to represent automatic
6843 variables that must have an address available. When the function returns
6844 (either with the ``ret`` or ``resume`` instructions), the memory is
6845 reclaimed. Allocating zero bytes is legal, but the result is undefined.
6846 The order in which memory is allocated (ie., which way the stack grows)
6852 .. code-block:: llvm
6854 %ptr = alloca i32 ; yields i32*:ptr
6855 %ptr = alloca i32, i32 4 ; yields i32*:ptr
6856 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
6857 %ptr = alloca i32, align 1024 ; yields i32*:ptr
6861 '``load``' Instruction
6862 ^^^^^^^^^^^^^^^^^^^^^^
6869 <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>]
6870 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>]
6871 !<index> = !{ i32 1 }
6872 !<deref_bytes_node> = !{i64 <dereferenceable_bytes>}
6873 !<align_node> = !{ i64 <value_alignment> }
6878 The '``load``' instruction is used to read from memory.
6883 The argument to the ``load`` instruction specifies the memory address
6884 from which to load. The type specified must be a :ref:`first
6885 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
6886 then the optimizer is not allowed to modify the number or order of
6887 execution of this ``load`` with other :ref:`volatile
6888 operations <volatile>`.
6890 If the ``load`` is marked as ``atomic``, it takes an extra
6891 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6892 ``release`` and ``acq_rel`` orderings are not valid on ``load``
6893 instructions. Atomic loads produce :ref:`defined <memmodel>` results
6894 when they may see multiple atomic stores. The type of the pointee must
6895 be an integer type whose bit width is a power of two greater than or
6896 equal to eight and less than or equal to a target-specific size limit.
6897 ``align`` must be explicitly specified on atomic loads, and the load has
6898 undefined behavior if the alignment is not set to a value which is at
6899 least the size in bytes of the pointee. ``!nontemporal`` does not have
6900 any defined semantics for atomic loads.
6902 The optional constant ``align`` argument specifies the alignment of the
6903 operation (that is, the alignment of the memory address). A value of 0
6904 or an omitted ``align`` argument means that the operation has the ABI
6905 alignment for the target. It is the responsibility of the code emitter
6906 to ensure that the alignment information is correct. Overestimating the
6907 alignment results in undefined behavior. Underestimating the alignment
6908 may produce less efficient code. An alignment of 1 is always safe. The
6909 maximum possible alignment is ``1 << 29``.
6911 The optional ``!nontemporal`` metadata must reference a single
6912 metadata name ``<index>`` corresponding to a metadata node with one
6913 ``i32`` entry of value 1. The existence of the ``!nontemporal``
6914 metadata on the instruction tells the optimizer and code generator
6915 that this load is not expected to be reused in the cache. The code
6916 generator may select special instructions to save cache bandwidth, such
6917 as the ``MOVNT`` instruction on x86.
6919 The optional ``!invariant.load`` metadata must reference a single
6920 metadata name ``<index>`` corresponding to a metadata node with no
6921 entries. The existence of the ``!invariant.load`` metadata on the
6922 instruction tells the optimizer and code generator that the address
6923 operand to this load points to memory which can be assumed unchanged.
6924 Being invariant does not imply that a location is dereferenceable,
6925 but it does imply that once the location is known dereferenceable
6926 its value is henceforth unchanging.
6928 The optional ``!invariant.group`` metadata must reference a single metadata name
6929 ``<index>`` corresponding to a metadata node. See ``invariant.group`` metadata.
6931 The optional ``!nonnull`` metadata must reference a single
6932 metadata name ``<index>`` corresponding to a metadata node with no
6933 entries. The existence of the ``!nonnull`` metadata on the
6934 instruction tells the optimizer that the value loaded is known to
6935 never be null. This is analogous to the ``nonnull`` attribute
6936 on parameters and return values. This metadata can only be applied
6937 to loads of a pointer type.
6939 The optional ``!dereferenceable`` metadata must reference a single metadata
6940 name ``<deref_bytes_node>`` corresponding to a metadata node with one ``i64``
6941 entry. The existence of the ``!dereferenceable`` metadata on the instruction
6942 tells the optimizer that the value loaded is known to be dereferenceable.
6943 The number of bytes known to be dereferenceable is specified by the integer
6944 value in the metadata node. This is analogous to the ''dereferenceable''
6945 attribute on parameters and return values. This metadata can only be applied
6946 to loads of a pointer type.
6948 The optional ``!dereferenceable_or_null`` metadata must reference a single
6949 metadata name ``<deref_bytes_node>`` corresponding to a metadata node with one
6950 ``i64`` entry. The existence of the ``!dereferenceable_or_null`` metadata on the
6951 instruction tells the optimizer that the value loaded is known to be either
6952 dereferenceable or null.
6953 The number of bytes known to be dereferenceable is specified by the integer
6954 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
6955 attribute on parameters and return values. This metadata can only be applied
6956 to loads of a pointer type.
6958 The optional ``!align`` metadata must reference a single metadata name
6959 ``<align_node>`` corresponding to a metadata node with one ``i64`` entry.
6960 The existence of the ``!align`` metadata on the instruction tells the
6961 optimizer that the value loaded is known to be aligned to a boundary specified
6962 by the integer value in the metadata node. The alignment must be a power of 2.
6963 This is analogous to the ''align'' attribute on parameters and return values.
6964 This metadata can only be applied to loads of a pointer type.
6969 The location of memory pointed to is loaded. If the value being loaded
6970 is of scalar type then the number of bytes read does not exceed the
6971 minimum number of bytes needed to hold all bits of the type. For
6972 example, loading an ``i24`` reads at most three bytes. When loading a
6973 value of a type like ``i20`` with a size that is not an integral number
6974 of bytes, the result is undefined if the value was not originally
6975 written using a store of the same type.
6980 .. code-block:: llvm
6982 %ptr = alloca i32 ; yields i32*:ptr
6983 store i32 3, i32* %ptr ; yields void
6984 %val = load i32, i32* %ptr ; yields i32:val = i32 3
6988 '``store``' Instruction
6989 ^^^^^^^^^^^^^^^^^^^^^^^
6996 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.group !<index>] ; yields void
6997 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>] ; yields void
7002 The '``store``' instruction is used to write to memory.
7007 There are two arguments to the ``store`` instruction: a value to store
7008 and an address at which to store it. The type of the ``<pointer>``
7009 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
7010 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
7011 then the optimizer is not allowed to modify the number or order of
7012 execution of this ``store`` with other :ref:`volatile
7013 operations <volatile>`.
7015 If the ``store`` is marked as ``atomic``, it takes an extra
7016 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
7017 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
7018 instructions. Atomic loads produce :ref:`defined <memmodel>` results
7019 when they may see multiple atomic stores. The type of the pointee must
7020 be an integer type whose bit width is a power of two greater than or
7021 equal to eight and less than or equal to a target-specific size limit.
7022 ``align`` must be explicitly specified on atomic stores, and the store
7023 has undefined behavior if the alignment is not set to a value which is
7024 at least the size in bytes of the pointee. ``!nontemporal`` does not
7025 have any defined semantics for atomic stores.
7027 The optional constant ``align`` argument specifies the alignment of the
7028 operation (that is, the alignment of the memory address). A value of 0
7029 or an omitted ``align`` argument means that the operation has the ABI
7030 alignment for the target. It is the responsibility of the code emitter
7031 to ensure that the alignment information is correct. Overestimating the
7032 alignment results in undefined behavior. Underestimating the
7033 alignment may produce less efficient code. An alignment of 1 is always
7034 safe. The maximum possible alignment is ``1 << 29``.
7036 The optional ``!nontemporal`` metadata must reference a single metadata
7037 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
7038 value 1. The existence of the ``!nontemporal`` metadata on the instruction
7039 tells the optimizer and code generator that this load is not expected to
7040 be reused in the cache. The code generator may select special
7041 instructions to save cache bandwidth, such as the MOVNT instruction on
7044 The optional ``!invariant.group`` metadata must reference a
7045 single metadata name ``<index>``. See ``invariant.group`` metadata.
7050 The contents of memory are updated to contain ``<value>`` at the
7051 location specified by the ``<pointer>`` operand. If ``<value>`` is
7052 of scalar type then the number of bytes written does not exceed the
7053 minimum number of bytes needed to hold all bits of the type. For
7054 example, storing an ``i24`` writes at most three bytes. When writing a
7055 value of a type like ``i20`` with a size that is not an integral number
7056 of bytes, it is unspecified what happens to the extra bits that do not
7057 belong to the type, but they will typically be overwritten.
7062 .. code-block:: llvm
7064 %ptr = alloca i32 ; yields i32*:ptr
7065 store i32 3, i32* %ptr ; yields void
7066 %val = load i32, i32* %ptr ; yields i32:val = i32 3
7070 '``fence``' Instruction
7071 ^^^^^^^^^^^^^^^^^^^^^^^
7078 fence [singlethread] <ordering> ; yields void
7083 The '``fence``' instruction is used to introduce happens-before edges
7089 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
7090 defines what *synchronizes-with* edges they add. They can only be given
7091 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
7096 A fence A which has (at least) ``release`` ordering semantics
7097 *synchronizes with* a fence B with (at least) ``acquire`` ordering
7098 semantics if and only if there exist atomic operations X and Y, both
7099 operating on some atomic object M, such that A is sequenced before X, X
7100 modifies M (either directly or through some side effect of a sequence
7101 headed by X), Y is sequenced before B, and Y observes M. This provides a
7102 *happens-before* dependency between A and B. Rather than an explicit
7103 ``fence``, one (but not both) of the atomic operations X or Y might
7104 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
7105 still *synchronize-with* the explicit ``fence`` and establish the
7106 *happens-before* edge.
7108 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
7109 ``acquire`` and ``release`` semantics specified above, participates in
7110 the global program order of other ``seq_cst`` operations and/or fences.
7112 The optional ":ref:`singlethread <singlethread>`" argument specifies
7113 that the fence only synchronizes with other fences in the same thread.
7114 (This is useful for interacting with signal handlers.)
7119 .. code-block:: llvm
7121 fence acquire ; yields void
7122 fence singlethread seq_cst ; yields void
7126 '``cmpxchg``' Instruction
7127 ^^^^^^^^^^^^^^^^^^^^^^^^^
7134 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
7139 The '``cmpxchg``' instruction is used to atomically modify memory. It
7140 loads a value in memory and compares it to a given value. If they are
7141 equal, it tries to store a new value into the memory.
7146 There are three arguments to the '``cmpxchg``' instruction: an address
7147 to operate on, a value to compare to the value currently be at that
7148 address, and a new value to place at that address if the compared values
7149 are equal. The type of '<cmp>' must be an integer type whose bit width
7150 is a power of two greater than or equal to eight and less than or equal
7151 to a target-specific size limit. '<cmp>' and '<new>' must have the same
7152 type, and the type of '<pointer>' must be a pointer to that type. If the
7153 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
7154 to modify the number or order of execution of this ``cmpxchg`` with
7155 other :ref:`volatile operations <volatile>`.
7157 The success and failure :ref:`ordering <ordering>` arguments specify how this
7158 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
7159 must be at least ``monotonic``, the ordering constraint on failure must be no
7160 stronger than that on success, and the failure ordering cannot be either
7161 ``release`` or ``acq_rel``.
7163 The optional "``singlethread``" argument declares that the ``cmpxchg``
7164 is only atomic with respect to code (usually signal handlers) running in
7165 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
7166 respect to all other code in the system.
7168 The pointer passed into cmpxchg must have alignment greater than or
7169 equal to the size in memory of the operand.
7174 The contents of memory at the location specified by the '``<pointer>``' operand
7175 is read and compared to '``<cmp>``'; if the read value is the equal, the
7176 '``<new>``' is written. The original value at the location is returned, together
7177 with a flag indicating success (true) or failure (false).
7179 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
7180 permitted: the operation may not write ``<new>`` even if the comparison
7183 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
7184 if the value loaded equals ``cmp``.
7186 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
7187 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
7188 load with an ordering parameter determined the second ordering parameter.
7193 .. code-block:: llvm
7196 %orig = atomic load i32, i32* %ptr unordered ; yields i32
7200 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
7201 %squared = mul i32 %cmp, %cmp
7202 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
7203 %value_loaded = extractvalue { i32, i1 } %val_success, 0
7204 %success = extractvalue { i32, i1 } %val_success, 1
7205 br i1 %success, label %done, label %loop
7212 '``atomicrmw``' Instruction
7213 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7220 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
7225 The '``atomicrmw``' instruction is used to atomically modify memory.
7230 There are three arguments to the '``atomicrmw``' instruction: an
7231 operation to apply, an address whose value to modify, an argument to the
7232 operation. The operation must be one of the following keywords:
7246 The type of '<value>' must be an integer type whose bit width is a power
7247 of two greater than or equal to eight and less than or equal to a
7248 target-specific size limit. The type of the '``<pointer>``' operand must
7249 be a pointer to that type. If the ``atomicrmw`` is marked as
7250 ``volatile``, then the optimizer is not allowed to modify the number or
7251 order of execution of this ``atomicrmw`` with other :ref:`volatile
7252 operations <volatile>`.
7257 The contents of memory at the location specified by the '``<pointer>``'
7258 operand are atomically read, modified, and written back. The original
7259 value at the location is returned. The modification is specified by the
7262 - xchg: ``*ptr = val``
7263 - add: ``*ptr = *ptr + val``
7264 - sub: ``*ptr = *ptr - val``
7265 - and: ``*ptr = *ptr & val``
7266 - nand: ``*ptr = ~(*ptr & val)``
7267 - or: ``*ptr = *ptr | val``
7268 - xor: ``*ptr = *ptr ^ val``
7269 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
7270 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
7271 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
7273 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
7279 .. code-block:: llvm
7281 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
7283 .. _i_getelementptr:
7285 '``getelementptr``' Instruction
7286 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7293 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7294 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7295 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
7300 The '``getelementptr``' instruction is used to get the address of a
7301 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
7302 address calculation only and does not access memory. The instruction can also
7303 be used to calculate a vector of such addresses.
7308 The first argument is always a type used as the basis for the calculations.
7309 The second argument is always a pointer or a vector of pointers, and is the
7310 base address to start from. The remaining arguments are indices
7311 that indicate which of the elements of the aggregate object are indexed.
7312 The interpretation of each index is dependent on the type being indexed
7313 into. The first index always indexes the pointer value given as the
7314 first argument, the second index indexes a value of the type pointed to
7315 (not necessarily the value directly pointed to, since the first index
7316 can be non-zero), etc. The first type indexed into must be a pointer
7317 value, subsequent types can be arrays, vectors, and structs. Note that
7318 subsequent types being indexed into can never be pointers, since that
7319 would require loading the pointer before continuing calculation.
7321 The type of each index argument depends on the type it is indexing into.
7322 When indexing into a (optionally packed) structure, only ``i32`` integer
7323 **constants** are allowed (when using a vector of indices they must all
7324 be the **same** ``i32`` integer constant). When indexing into an array,
7325 pointer or vector, integers of any width are allowed, and they are not
7326 required to be constant. These integers are treated as signed values
7329 For example, let's consider a C code fragment and how it gets compiled
7345 int *foo(struct ST *s) {
7346 return &s[1].Z.B[5][13];
7349 The LLVM code generated by Clang is:
7351 .. code-block:: llvm
7353 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
7354 %struct.ST = type { i32, double, %struct.RT }
7356 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
7358 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
7365 In the example above, the first index is indexing into the
7366 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
7367 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
7368 indexes into the third element of the structure, yielding a
7369 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
7370 structure. The third index indexes into the second element of the
7371 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
7372 dimensions of the array are subscripted into, yielding an '``i32``'
7373 type. The '``getelementptr``' instruction returns a pointer to this
7374 element, thus computing a value of '``i32*``' type.
7376 Note that it is perfectly legal to index partially through a structure,
7377 returning a pointer to an inner element. Because of this, the LLVM code
7378 for the given testcase is equivalent to:
7380 .. code-block:: llvm
7382 define i32* @foo(%struct.ST* %s) {
7383 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
7384 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
7385 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
7386 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
7387 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
7391 If the ``inbounds`` keyword is present, the result value of the
7392 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
7393 pointer is not an *in bounds* address of an allocated object, or if any
7394 of the addresses that would be formed by successive addition of the
7395 offsets implied by the indices to the base address with infinitely
7396 precise signed arithmetic are not an *in bounds* address of that
7397 allocated object. The *in bounds* addresses for an allocated object are
7398 all the addresses that point into the object, plus the address one byte
7399 past the end. In cases where the base is a vector of pointers the
7400 ``inbounds`` keyword applies to each of the computations element-wise.
7402 If the ``inbounds`` keyword is not present, the offsets are added to the
7403 base address with silently-wrapping two's complement arithmetic. If the
7404 offsets have a different width from the pointer, they are sign-extended
7405 or truncated to the width of the pointer. The result value of the
7406 ``getelementptr`` may be outside the object pointed to by the base
7407 pointer. The result value may not necessarily be used to access memory
7408 though, even if it happens to point into allocated storage. See the
7409 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
7412 The getelementptr instruction is often confusing. For some more insight
7413 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
7418 .. code-block:: llvm
7420 ; yields [12 x i8]*:aptr
7421 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
7423 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
7425 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
7427 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
7432 The ``getelementptr`` returns a vector of pointers, instead of a single address,
7433 when one or more of its arguments is a vector. In such cases, all vector
7434 arguments should have the same number of elements, and every scalar argument
7435 will be effectively broadcast into a vector during address calculation.
7437 .. code-block:: llvm
7439 ; All arguments are vectors:
7440 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
7441 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
7443 ; Add the same scalar offset to each pointer of a vector:
7444 ; A[i] = ptrs[i] + offset*sizeof(i8)
7445 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
7447 ; Add distinct offsets to the same pointer:
7448 ; A[i] = ptr + offsets[i]*sizeof(i8)
7449 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
7451 ; In all cases described above the type of the result is <4 x i8*>
7453 The two following instructions are equivalent:
7455 .. code-block:: llvm
7457 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7458 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
7459 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
7461 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
7463 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7464 i32 2, i32 1, <4 x i32> %ind4, i64 13
7466 Let's look at the C code, where the vector version of ``getelementptr``
7471 // Let's assume that we vectorize the following loop:
7472 double *A, B; int *C;
7473 for (int i = 0; i < size; ++i) {
7477 .. code-block:: llvm
7479 ; get pointers for 8 elements from array B
7480 %ptrs = getelementptr double, double* %B, <8 x i32> %C
7481 ; load 8 elements from array B into A
7482 %A = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs,
7483 i32 8, <8 x i1> %mask, <8 x double> %passthru)
7485 Conversion Operations
7486 ---------------------
7488 The instructions in this category are the conversion instructions
7489 (casting) which all take a single operand and a type. They perform
7490 various bit conversions on the operand.
7492 '``trunc .. to``' Instruction
7493 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7500 <result> = trunc <ty> <value> to <ty2> ; yields ty2
7505 The '``trunc``' instruction truncates its operand to the type ``ty2``.
7510 The '``trunc``' instruction takes a value to trunc, and a type to trunc
7511 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
7512 of the same number of integers. The bit size of the ``value`` must be
7513 larger than the bit size of the destination type, ``ty2``. Equal sized
7514 types are not allowed.
7519 The '``trunc``' instruction truncates the high order bits in ``value``
7520 and converts the remaining bits to ``ty2``. Since the source size must
7521 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
7522 It will always truncate bits.
7527 .. code-block:: llvm
7529 %X = trunc i32 257 to i8 ; yields i8:1
7530 %Y = trunc i32 123 to i1 ; yields i1:true
7531 %Z = trunc i32 122 to i1 ; yields i1:false
7532 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
7534 '``zext .. to``' Instruction
7535 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7542 <result> = zext <ty> <value> to <ty2> ; yields ty2
7547 The '``zext``' instruction zero extends its operand to type ``ty2``.
7552 The '``zext``' instruction takes a value to cast, and a type to cast it
7553 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7554 the same number of integers. The bit size of the ``value`` must be
7555 smaller than the bit size of the destination type, ``ty2``.
7560 The ``zext`` fills the high order bits of the ``value`` with zero bits
7561 until it reaches the size of the destination type, ``ty2``.
7563 When zero extending from i1, the result will always be either 0 or 1.
7568 .. code-block:: llvm
7570 %X = zext i32 257 to i64 ; yields i64:257
7571 %Y = zext i1 true to i32 ; yields i32:1
7572 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7574 '``sext .. to``' Instruction
7575 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7582 <result> = sext <ty> <value> to <ty2> ; yields ty2
7587 The '``sext``' sign extends ``value`` to the type ``ty2``.
7592 The '``sext``' instruction takes a value to cast, and a type to cast it
7593 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7594 the same number of integers. The bit size of the ``value`` must be
7595 smaller than the bit size of the destination type, ``ty2``.
7600 The '``sext``' instruction performs a sign extension by copying the sign
7601 bit (highest order bit) of the ``value`` until it reaches the bit size
7602 of the type ``ty2``.
7604 When sign extending from i1, the extension always results in -1 or 0.
7609 .. code-block:: llvm
7611 %X = sext i8 -1 to i16 ; yields i16 :65535
7612 %Y = sext i1 true to i32 ; yields i32:-1
7613 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7615 '``fptrunc .. to``' Instruction
7616 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7623 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
7628 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
7633 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
7634 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
7635 The size of ``value`` must be larger than the size of ``ty2``. This
7636 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
7641 The '``fptrunc``' instruction casts a ``value`` from a larger
7642 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
7643 point <t_floating>` type. If the value cannot fit (i.e. overflows) within the
7644 destination type, ``ty2``, then the results are undefined. If the cast produces
7645 an inexact result, how rounding is performed (e.g. truncation, also known as
7646 round to zero) is undefined.
7651 .. code-block:: llvm
7653 %X = fptrunc double 123.0 to float ; yields float:123.0
7654 %Y = fptrunc double 1.0E+300 to float ; yields undefined
7656 '``fpext .. to``' Instruction
7657 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7664 <result> = fpext <ty> <value> to <ty2> ; yields ty2
7669 The '``fpext``' extends a floating point ``value`` to a larger floating
7675 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
7676 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
7677 to. The source type must be smaller than the destination type.
7682 The '``fpext``' instruction extends the ``value`` from a smaller
7683 :ref:`floating point <t_floating>` type to a larger :ref:`floating
7684 point <t_floating>` type. The ``fpext`` cannot be used to make a
7685 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
7686 *no-op cast* for a floating point cast.
7691 .. code-block:: llvm
7693 %X = fpext float 3.125 to double ; yields double:3.125000e+00
7694 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
7696 '``fptoui .. to``' Instruction
7697 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7704 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
7709 The '``fptoui``' converts a floating point ``value`` to its unsigned
7710 integer equivalent of type ``ty2``.
7715 The '``fptoui``' instruction takes a value to cast, which must be a
7716 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7717 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7718 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7719 type with the same number of elements as ``ty``
7724 The '``fptoui``' instruction converts its :ref:`floating
7725 point <t_floating>` operand into the nearest (rounding towards zero)
7726 unsigned integer value. If the value cannot fit in ``ty2``, the results
7732 .. code-block:: llvm
7734 %X = fptoui double 123.0 to i32 ; yields i32:123
7735 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
7736 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
7738 '``fptosi .. to``' Instruction
7739 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7746 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
7751 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
7752 ``value`` to type ``ty2``.
7757 The '``fptosi``' instruction takes a value to cast, which must be a
7758 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7759 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7760 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7761 type with the same number of elements as ``ty``
7766 The '``fptosi``' instruction converts its :ref:`floating
7767 point <t_floating>` operand into the nearest (rounding towards zero)
7768 signed integer value. If the value cannot fit in ``ty2``, the results
7774 .. code-block:: llvm
7776 %X = fptosi double -123.0 to i32 ; yields i32:-123
7777 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
7778 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
7780 '``uitofp .. to``' Instruction
7781 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7788 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
7793 The '``uitofp``' instruction regards ``value`` as an unsigned integer
7794 and converts that value to the ``ty2`` type.
7799 The '``uitofp``' instruction takes a value to cast, which must be a
7800 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7801 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7802 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7803 type with the same number of elements as ``ty``
7808 The '``uitofp``' instruction interprets its operand as an unsigned
7809 integer quantity and converts it to the corresponding floating point
7810 value. If the value cannot fit in the floating point value, the results
7816 .. code-block:: llvm
7818 %X = uitofp i32 257 to float ; yields float:257.0
7819 %Y = uitofp i8 -1 to double ; yields double:255.0
7821 '``sitofp .. to``' Instruction
7822 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7829 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
7834 The '``sitofp``' instruction regards ``value`` as a signed integer and
7835 converts that value to the ``ty2`` type.
7840 The '``sitofp``' instruction takes a value to cast, which must be a
7841 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7842 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7843 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7844 type with the same number of elements as ``ty``
7849 The '``sitofp``' instruction interprets its operand as a signed integer
7850 quantity and converts it to the corresponding floating point value. If
7851 the value cannot fit in the floating point value, the results are
7857 .. code-block:: llvm
7859 %X = sitofp i32 257 to float ; yields float:257.0
7860 %Y = sitofp i8 -1 to double ; yields double:-1.0
7864 '``ptrtoint .. to``' Instruction
7865 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7872 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
7877 The '``ptrtoint``' instruction converts the pointer or a vector of
7878 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
7883 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
7884 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
7885 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
7886 a vector of integers type.
7891 The '``ptrtoint``' instruction converts ``value`` to integer type
7892 ``ty2`` by interpreting the pointer value as an integer and either
7893 truncating or zero extending that value to the size of the integer type.
7894 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
7895 ``value`` is larger than ``ty2`` then a truncation is done. If they are
7896 the same size, then nothing is done (*no-op cast*) other than a type
7902 .. code-block:: llvm
7904 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
7905 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
7906 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
7910 '``inttoptr .. to``' Instruction
7911 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7918 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
7923 The '``inttoptr``' instruction converts an integer ``value`` to a
7924 pointer type, ``ty2``.
7929 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
7930 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
7936 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
7937 applying either a zero extension or a truncation depending on the size
7938 of the integer ``value``. If ``value`` is larger than the size of a
7939 pointer then a truncation is done. If ``value`` is smaller than the size
7940 of a pointer then a zero extension is done. If they are the same size,
7941 nothing is done (*no-op cast*).
7946 .. code-block:: llvm
7948 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
7949 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
7950 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
7951 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
7955 '``bitcast .. to``' Instruction
7956 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7963 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
7968 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
7974 The '``bitcast``' instruction takes a value to cast, which must be a
7975 non-aggregate first class value, and a type to cast it to, which must
7976 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
7977 bit sizes of ``value`` and the destination type, ``ty2``, must be
7978 identical. If the source type is a pointer, the destination type must
7979 also be a pointer of the same size. This instruction supports bitwise
7980 conversion of vectors to integers and to vectors of other types (as
7981 long as they have the same size).
7986 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
7987 is always a *no-op cast* because no bits change with this
7988 conversion. The conversion is done as if the ``value`` had been stored
7989 to memory and read back as type ``ty2``. Pointer (or vector of
7990 pointers) types may only be converted to other pointer (or vector of
7991 pointers) types with the same address space through this instruction.
7992 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
7993 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
7998 .. code-block:: llvm
8000 %X = bitcast i8 255 to i8 ; yields i8 :-1
8001 %Y = bitcast i32* %x to sint* ; yields sint*:%x
8002 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
8003 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
8005 .. _i_addrspacecast:
8007 '``addrspacecast .. to``' Instruction
8008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8015 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
8020 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
8021 address space ``n`` to type ``pty2`` in address space ``m``.
8026 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
8027 to cast and a pointer type to cast it to, which must have a different
8033 The '``addrspacecast``' instruction converts the pointer value
8034 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
8035 value modification, depending on the target and the address space
8036 pair. Pointer conversions within the same address space must be
8037 performed with the ``bitcast`` instruction. Note that if the address space
8038 conversion is legal then both result and operand refer to the same memory
8044 .. code-block:: llvm
8046 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
8047 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
8048 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
8055 The instructions in this category are the "miscellaneous" instructions,
8056 which defy better classification.
8060 '``icmp``' Instruction
8061 ^^^^^^^^^^^^^^^^^^^^^^
8068 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8073 The '``icmp``' instruction returns a boolean value or a vector of
8074 boolean values based on comparison of its two integer, integer vector,
8075 pointer, or pointer vector operands.
8080 The '``icmp``' instruction takes three operands. The first operand is
8081 the condition code indicating the kind of comparison to perform. It is
8082 not a value, just a keyword. The possible condition code are:
8085 #. ``ne``: not equal
8086 #. ``ugt``: unsigned greater than
8087 #. ``uge``: unsigned greater or equal
8088 #. ``ult``: unsigned less than
8089 #. ``ule``: unsigned less or equal
8090 #. ``sgt``: signed greater than
8091 #. ``sge``: signed greater or equal
8092 #. ``slt``: signed less than
8093 #. ``sle``: signed less or equal
8095 The remaining two arguments must be :ref:`integer <t_integer>` or
8096 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
8097 must also be identical types.
8102 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
8103 code given as ``cond``. The comparison performed always yields either an
8104 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
8106 #. ``eq``: yields ``true`` if the operands are equal, ``false``
8107 otherwise. No sign interpretation is necessary or performed.
8108 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
8109 otherwise. No sign interpretation is necessary or performed.
8110 #. ``ugt``: interprets the operands as unsigned values and yields
8111 ``true`` if ``op1`` is greater than ``op2``.
8112 #. ``uge``: interprets the operands as unsigned values and yields
8113 ``true`` if ``op1`` is greater than or equal to ``op2``.
8114 #. ``ult``: interprets the operands as unsigned values and yields
8115 ``true`` if ``op1`` is less than ``op2``.
8116 #. ``ule``: interprets the operands as unsigned values and yields
8117 ``true`` if ``op1`` is less than or equal to ``op2``.
8118 #. ``sgt``: interprets the operands as signed values and yields ``true``
8119 if ``op1`` is greater than ``op2``.
8120 #. ``sge``: interprets the operands as signed values and yields ``true``
8121 if ``op1`` is greater than or equal to ``op2``.
8122 #. ``slt``: interprets the operands as signed values and yields ``true``
8123 if ``op1`` is less than ``op2``.
8124 #. ``sle``: interprets the operands as signed values and yields ``true``
8125 if ``op1`` is less than or equal to ``op2``.
8127 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
8128 are compared as if they were integers.
8130 If the operands are integer vectors, then they are compared element by
8131 element. The result is an ``i1`` vector with the same number of elements
8132 as the values being compared. Otherwise, the result is an ``i1``.
8137 .. code-block:: llvm
8139 <result> = icmp eq i32 4, 5 ; yields: result=false
8140 <result> = icmp ne float* %X, %X ; yields: result=false
8141 <result> = icmp ult i16 4, 5 ; yields: result=true
8142 <result> = icmp sgt i16 4, 5 ; yields: result=false
8143 <result> = icmp ule i16 -4, 5 ; yields: result=false
8144 <result> = icmp sge i16 4, 5 ; yields: result=false
8146 Note that the code generator does not yet support vector types with the
8147 ``icmp`` instruction.
8151 '``fcmp``' Instruction
8152 ^^^^^^^^^^^^^^^^^^^^^^
8159 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8164 The '``fcmp``' instruction returns a boolean value or vector of boolean
8165 values based on comparison of its operands.
8167 If the operands are floating point scalars, then the result type is a
8168 boolean (:ref:`i1 <t_integer>`).
8170 If the operands are floating point vectors, then the result type is a
8171 vector of boolean with the same number of elements as the operands being
8177 The '``fcmp``' instruction takes three operands. The first operand is
8178 the condition code indicating the kind of comparison to perform. It is
8179 not a value, just a keyword. The possible condition code are:
8181 #. ``false``: no comparison, always returns false
8182 #. ``oeq``: ordered and equal
8183 #. ``ogt``: ordered and greater than
8184 #. ``oge``: ordered and greater than or equal
8185 #. ``olt``: ordered and less than
8186 #. ``ole``: ordered and less than or equal
8187 #. ``one``: ordered and not equal
8188 #. ``ord``: ordered (no nans)
8189 #. ``ueq``: unordered or equal
8190 #. ``ugt``: unordered or greater than
8191 #. ``uge``: unordered or greater than or equal
8192 #. ``ult``: unordered or less than
8193 #. ``ule``: unordered or less than or equal
8194 #. ``une``: unordered or not equal
8195 #. ``uno``: unordered (either nans)
8196 #. ``true``: no comparison, always returns true
8198 *Ordered* means that neither operand is a QNAN while *unordered* means
8199 that either operand may be a QNAN.
8201 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
8202 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
8203 type. They must have identical types.
8208 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
8209 condition code given as ``cond``. If the operands are vectors, then the
8210 vectors are compared element by element. Each comparison performed
8211 always yields an :ref:`i1 <t_integer>` result, as follows:
8213 #. ``false``: always yields ``false``, regardless of operands.
8214 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
8215 is equal to ``op2``.
8216 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
8217 is greater than ``op2``.
8218 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
8219 is greater than or equal to ``op2``.
8220 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
8221 is less than ``op2``.
8222 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
8223 is less than or equal to ``op2``.
8224 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
8225 is not equal to ``op2``.
8226 #. ``ord``: yields ``true`` if both operands are not a QNAN.
8227 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
8229 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
8230 greater than ``op2``.
8231 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
8232 greater than or equal to ``op2``.
8233 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
8235 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
8236 less than or equal to ``op2``.
8237 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
8238 not equal to ``op2``.
8239 #. ``uno``: yields ``true`` if either operand is a QNAN.
8240 #. ``true``: always yields ``true``, regardless of operands.
8242 The ``fcmp`` instruction can also optionally take any number of
8243 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8244 otherwise unsafe floating point optimizations.
8246 Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
8247 only flags that have any effect on its semantics are those that allow
8248 assumptions to be made about the values of input arguments; namely
8249 ``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
8254 .. code-block:: llvm
8256 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
8257 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
8258 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
8259 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
8261 Note that the code generator does not yet support vector types with the
8262 ``fcmp`` instruction.
8266 '``phi``' Instruction
8267 ^^^^^^^^^^^^^^^^^^^^^
8274 <result> = phi <ty> [ <val0>, <label0>], ...
8279 The '``phi``' instruction is used to implement the φ node in the SSA
8280 graph representing the function.
8285 The type of the incoming values is specified with the first type field.
8286 After this, the '``phi``' instruction takes a list of pairs as
8287 arguments, with one pair for each predecessor basic block of the current
8288 block. Only values of :ref:`first class <t_firstclass>` type may be used as
8289 the value arguments to the PHI node. Only labels may be used as the
8292 There must be no non-phi instructions between the start of a basic block
8293 and the PHI instructions: i.e. PHI instructions must be first in a basic
8296 For the purposes of the SSA form, the use of each incoming value is
8297 deemed to occur on the edge from the corresponding predecessor block to
8298 the current block (but after any definition of an '``invoke``'
8299 instruction's return value on the same edge).
8304 At runtime, the '``phi``' instruction logically takes on the value
8305 specified by the pair corresponding to the predecessor basic block that
8306 executed just prior to the current block.
8311 .. code-block:: llvm
8313 Loop: ; Infinite loop that counts from 0 on up...
8314 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
8315 %nextindvar = add i32 %indvar, 1
8320 '``select``' Instruction
8321 ^^^^^^^^^^^^^^^^^^^^^^^^
8328 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
8330 selty is either i1 or {<N x i1>}
8335 The '``select``' instruction is used to choose one value based on a
8336 condition, without IR-level branching.
8341 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
8342 values indicating the condition, and two values of the same :ref:`first
8343 class <t_firstclass>` type.
8348 If the condition is an i1 and it evaluates to 1, the instruction returns
8349 the first value argument; otherwise, it returns the second value
8352 If the condition is a vector of i1, then the value arguments must be
8353 vectors of the same size, and the selection is done element by element.
8355 If the condition is an i1 and the value arguments are vectors of the
8356 same size, then an entire vector is selected.
8361 .. code-block:: llvm
8363 %X = select i1 true, i8 17, i8 42 ; yields i8:17
8367 '``call``' Instruction
8368 ^^^^^^^^^^^^^^^^^^^^^^
8375 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
8381 The '``call``' instruction represents a simple function call.
8386 This instruction requires several arguments:
8388 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
8389 should perform tail call optimization. The ``tail`` marker is a hint that
8390 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
8391 means that the call must be tail call optimized in order for the program to
8392 be correct. The ``musttail`` marker provides these guarantees:
8394 #. The call will not cause unbounded stack growth if it is part of a
8395 recursive cycle in the call graph.
8396 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
8399 Both markers imply that the callee does not access allocas or varargs from
8400 the caller. Calls marked ``musttail`` must obey the following additional
8403 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
8404 or a pointer bitcast followed by a ret instruction.
8405 - The ret instruction must return the (possibly bitcasted) value
8406 produced by the call or void.
8407 - The caller and callee prototypes must match. Pointer types of
8408 parameters or return types may differ in pointee type, but not
8410 - The calling conventions of the caller and callee must match.
8411 - All ABI-impacting function attributes, such as sret, byval, inreg,
8412 returned, and inalloca, must match.
8413 - The callee must be varargs iff the caller is varargs. Bitcasting a
8414 non-varargs function to the appropriate varargs type is legal so
8415 long as the non-varargs prefixes obey the other rules.
8417 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
8418 the following conditions are met:
8420 - Caller and callee both have the calling convention ``fastcc``.
8421 - The call is in tail position (ret immediately follows call and ret
8422 uses value of call or is void).
8423 - Option ``-tailcallopt`` is enabled, or
8424 ``llvm::GuaranteedTailCallOpt`` is ``true``.
8425 - `Platform-specific constraints are
8426 met. <CodeGenerator.html#tailcallopt>`_
8428 #. The optional "cconv" marker indicates which :ref:`calling
8429 convention <callingconv>` the call should use. If none is
8430 specified, the call defaults to using C calling conventions. The
8431 calling convention of the call must match the calling convention of
8432 the target function, or else the behavior is undefined.
8433 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
8434 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
8436 #. '``ty``': the type of the call instruction itself which is also the
8437 type of the return value. Functions that return no value are marked
8439 #. '``fnty``': shall be the signature of the pointer to function value
8440 being invoked. The argument types must match the types implied by
8441 this signature. This type can be omitted if the function is not
8442 varargs and if the function type does not return a pointer to a
8444 #. '``fnptrval``': An LLVM value containing a pointer to a function to
8445 be invoked. In most cases, this is a direct function invocation, but
8446 indirect ``call``'s are just as possible, calling an arbitrary pointer
8448 #. '``function args``': argument list whose types match the function
8449 signature argument types and parameter attributes. All arguments must
8450 be of :ref:`first class <t_firstclass>` type. If the function signature
8451 indicates the function accepts a variable number of arguments, the
8452 extra arguments can be specified.
8453 #. The optional :ref:`function attributes <fnattrs>` list. Only
8454 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
8455 attributes are valid here.
8456 #. The optional :ref:`operand bundles <opbundles>` list.
8461 The '``call``' instruction is used to cause control flow to transfer to
8462 a specified function, with its incoming arguments bound to the specified
8463 values. Upon a '``ret``' instruction in the called function, control
8464 flow continues with the instruction after the function call, and the
8465 return value of the function is bound to the result argument.
8470 .. code-block:: llvm
8472 %retval = call i32 @test(i32 %argc)
8473 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
8474 %X = tail call i32 @foo() ; yields i32
8475 %Y = tail call fastcc i32 @foo() ; yields i32
8476 call void %foo(i8 97 signext)
8478 %struct.A = type { i32, i8 }
8479 %r = call %struct.A @foo() ; yields { i32, i8 }
8480 %gr = extractvalue %struct.A %r, 0 ; yields i32
8481 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
8482 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
8483 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
8485 llvm treats calls to some functions with names and arguments that match
8486 the standard C99 library as being the C99 library functions, and may
8487 perform optimizations or generate code for them under that assumption.
8488 This is something we'd like to change in the future to provide better
8489 support for freestanding environments and non-C-based languages.
8493 '``va_arg``' Instruction
8494 ^^^^^^^^^^^^^^^^^^^^^^^^
8501 <resultval> = va_arg <va_list*> <arglist>, <argty>
8506 The '``va_arg``' instruction is used to access arguments passed through
8507 the "variable argument" area of a function call. It is used to implement
8508 the ``va_arg`` macro in C.
8513 This instruction takes a ``va_list*`` value and the type of the
8514 argument. It returns a value of the specified argument type and
8515 increments the ``va_list`` to point to the next argument. The actual
8516 type of ``va_list`` is target specific.
8521 The '``va_arg``' instruction loads an argument of the specified type
8522 from the specified ``va_list`` and causes the ``va_list`` to point to
8523 the next argument. For more information, see the variable argument
8524 handling :ref:`Intrinsic Functions <int_varargs>`.
8526 It is legal for this instruction to be called in a function which does
8527 not take a variable number of arguments, for example, the ``vfprintf``
8530 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
8531 function <intrinsics>` because it takes a type as an argument.
8536 See the :ref:`variable argument processing <int_varargs>` section.
8538 Note that the code generator does not yet fully support va\_arg on many
8539 targets. Also, it does not currently support va\_arg with aggregate
8540 types on any target.
8544 '``landingpad``' Instruction
8545 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8552 <resultval> = landingpad <resultty> <clause>+
8553 <resultval> = landingpad <resultty> cleanup <clause>*
8555 <clause> := catch <type> <value>
8556 <clause> := filter <array constant type> <array constant>
8561 The '``landingpad``' instruction is used by `LLVM's exception handling
8562 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8563 is a landing pad --- one where the exception lands, and corresponds to the
8564 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
8565 defines values supplied by the :ref:`personality function <personalityfn>` upon
8566 re-entry to the function. The ``resultval`` has the type ``resultty``.
8572 ``cleanup`` flag indicates that the landing pad block is a cleanup.
8574 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
8575 contains the global variable representing the "type" that may be caught
8576 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
8577 clause takes an array constant as its argument. Use
8578 "``[0 x i8**] undef``" for a filter which cannot throw. The
8579 '``landingpad``' instruction must contain *at least* one ``clause`` or
8580 the ``cleanup`` flag.
8585 The '``landingpad``' instruction defines the values which are set by the
8586 :ref:`personality function <personalityfn>` upon re-entry to the function, and
8587 therefore the "result type" of the ``landingpad`` instruction. As with
8588 calling conventions, how the personality function results are
8589 represented in LLVM IR is target specific.
8591 The clauses are applied in order from top to bottom. If two
8592 ``landingpad`` instructions are merged together through inlining, the
8593 clauses from the calling function are appended to the list of clauses.
8594 When the call stack is being unwound due to an exception being thrown,
8595 the exception is compared against each ``clause`` in turn. If it doesn't
8596 match any of the clauses, and the ``cleanup`` flag is not set, then
8597 unwinding continues further up the call stack.
8599 The ``landingpad`` instruction has several restrictions:
8601 - A landing pad block is a basic block which is the unwind destination
8602 of an '``invoke``' instruction.
8603 - A landing pad block must have a '``landingpad``' instruction as its
8604 first non-PHI instruction.
8605 - There can be only one '``landingpad``' instruction within the landing
8607 - A basic block that is not a landing pad block may not include a
8608 '``landingpad``' instruction.
8613 .. code-block:: llvm
8615 ;; A landing pad which can catch an integer.
8616 %res = landingpad { i8*, i32 }
8618 ;; A landing pad that is a cleanup.
8619 %res = landingpad { i8*, i32 }
8621 ;; A landing pad which can catch an integer and can only throw a double.
8622 %res = landingpad { i8*, i32 }
8624 filter [1 x i8**] [@_ZTId]
8628 '``cleanuppad``' Instruction
8629 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8636 <resultval> = cleanuppad [<args>*]
8641 The '``cleanuppad``' instruction is used by `LLVM's exception handling
8642 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8643 is a cleanup block --- one where a personality routine attempts to
8644 transfer control to run cleanup actions.
8645 The ``args`` correspond to whatever additional
8646 information the :ref:`personality function <personalityfn>` requires to
8647 execute the cleanup.
8648 The ``resultval`` has the type :ref:`token <t_token>` and is used to
8649 match the ``cleanuppad`` to corresponding :ref:`cleanuprets <i_cleanupret>`
8650 and :ref:`cleanupendpads <i_cleanupendpad>`.
8655 The instruction takes a list of arbitrary values which are interpreted
8656 by the :ref:`personality function <personalityfn>`.
8661 When the call stack is being unwound due to an exception being thrown,
8662 the :ref:`personality function <personalityfn>` transfers control to the
8663 ``cleanuppad`` with the aid of the personality-specific arguments.
8664 As with calling conventions, how the personality function results are
8665 represented in LLVM IR is target specific.
8667 The ``cleanuppad`` instruction has several restrictions:
8669 - A cleanup block is a basic block which is the unwind destination of
8670 an exceptional instruction.
8671 - A cleanup block must have a '``cleanuppad``' instruction as its
8672 first non-PHI instruction.
8673 - There can be only one '``cleanuppad``' instruction within the
8675 - A basic block that is not a cleanup block may not include a
8676 '``cleanuppad``' instruction.
8677 - All '``cleanupret``'s and '``cleanupendpad``'s which consume a ``cleanuppad``
8678 must have the same exceptional successor.
8679 - It is undefined behavior for control to transfer from a ``cleanuppad`` to a
8680 ``ret`` without first executing a ``cleanupret`` or ``cleanupendpad`` that
8681 consumes the ``cleanuppad``.
8682 - It is undefined behavior for control to transfer from a ``cleanuppad`` to
8683 itself without first executing a ``cleanupret`` or ``cleanupendpad`` that
8684 consumes the ``cleanuppad``.
8689 .. code-block:: llvm
8691 %tok = cleanuppad []
8698 LLVM supports the notion of an "intrinsic function". These functions
8699 have well known names and semantics and are required to follow certain
8700 restrictions. Overall, these intrinsics represent an extension mechanism
8701 for the LLVM language that does not require changing all of the
8702 transformations in LLVM when adding to the language (or the bitcode
8703 reader/writer, the parser, etc...).
8705 Intrinsic function names must all start with an "``llvm.``" prefix. This
8706 prefix is reserved in LLVM for intrinsic names; thus, function names may
8707 not begin with this prefix. Intrinsic functions must always be external
8708 functions: you cannot define the body of intrinsic functions. Intrinsic
8709 functions may only be used in call or invoke instructions: it is illegal
8710 to take the address of an intrinsic function. Additionally, because
8711 intrinsic functions are part of the LLVM language, it is required if any
8712 are added that they be documented here.
8714 Some intrinsic functions can be overloaded, i.e., the intrinsic
8715 represents a family of functions that perform the same operation but on
8716 different data types. Because LLVM can represent over 8 million
8717 different integer types, overloading is used commonly to allow an
8718 intrinsic function to operate on any integer type. One or more of the
8719 argument types or the result type can be overloaded to accept any
8720 integer type. Argument types may also be defined as exactly matching a
8721 previous argument's type or the result type. This allows an intrinsic
8722 function which accepts multiple arguments, but needs all of them to be
8723 of the same type, to only be overloaded with respect to a single
8724 argument or the result.
8726 Overloaded intrinsics will have the names of its overloaded argument
8727 types encoded into its function name, each preceded by a period. Only
8728 those types which are overloaded result in a name suffix. Arguments
8729 whose type is matched against another type do not. For example, the
8730 ``llvm.ctpop`` function can take an integer of any width and returns an
8731 integer of exactly the same integer width. This leads to a family of
8732 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
8733 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
8734 overloaded, and only one type suffix is required. Because the argument's
8735 type is matched against the return type, it does not require its own
8738 To learn how to add an intrinsic function, please see the `Extending
8739 LLVM Guide <ExtendingLLVM.html>`_.
8743 Variable Argument Handling Intrinsics
8744 -------------------------------------
8746 Variable argument support is defined in LLVM with the
8747 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
8748 functions. These functions are related to the similarly named macros
8749 defined in the ``<stdarg.h>`` header file.
8751 All of these functions operate on arguments that use a target-specific
8752 value type "``va_list``". The LLVM assembly language reference manual
8753 does not define what this type is, so all transformations should be
8754 prepared to handle these functions regardless of the type used.
8756 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
8757 variable argument handling intrinsic functions are used.
8759 .. code-block:: llvm
8761 ; This struct is different for every platform. For most platforms,
8762 ; it is merely an i8*.
8763 %struct.va_list = type { i8* }
8765 ; For Unix x86_64 platforms, va_list is the following struct:
8766 ; %struct.va_list = type { i32, i32, i8*, i8* }
8768 define i32 @test(i32 %X, ...) {
8769 ; Initialize variable argument processing
8770 %ap = alloca %struct.va_list
8771 %ap2 = bitcast %struct.va_list* %ap to i8*
8772 call void @llvm.va_start(i8* %ap2)
8774 ; Read a single integer argument
8775 %tmp = va_arg i8* %ap2, i32
8777 ; Demonstrate usage of llvm.va_copy and llvm.va_end
8779 %aq2 = bitcast i8** %aq to i8*
8780 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
8781 call void @llvm.va_end(i8* %aq2)
8783 ; Stop processing of arguments.
8784 call void @llvm.va_end(i8* %ap2)
8788 declare void @llvm.va_start(i8*)
8789 declare void @llvm.va_copy(i8*, i8*)
8790 declare void @llvm.va_end(i8*)
8794 '``llvm.va_start``' Intrinsic
8795 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8802 declare void @llvm.va_start(i8* <arglist>)
8807 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
8808 subsequent use by ``va_arg``.
8813 The argument is a pointer to a ``va_list`` element to initialize.
8818 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
8819 available in C. In a target-dependent way, it initializes the
8820 ``va_list`` element to which the argument points, so that the next call
8821 to ``va_arg`` will produce the first variable argument passed to the
8822 function. Unlike the C ``va_start`` macro, this intrinsic does not need
8823 to know the last argument of the function as the compiler can figure
8826 '``llvm.va_end``' Intrinsic
8827 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8834 declare void @llvm.va_end(i8* <arglist>)
8839 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
8840 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
8845 The argument is a pointer to a ``va_list`` to destroy.
8850 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
8851 available in C. In a target-dependent way, it destroys the ``va_list``
8852 element to which the argument points. Calls to
8853 :ref:`llvm.va_start <int_va_start>` and
8854 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
8859 '``llvm.va_copy``' Intrinsic
8860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8867 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
8872 The '``llvm.va_copy``' intrinsic copies the current argument position
8873 from the source argument list to the destination argument list.
8878 The first argument is a pointer to a ``va_list`` element to initialize.
8879 The second argument is a pointer to a ``va_list`` element to copy from.
8884 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
8885 available in C. In a target-dependent way, it copies the source
8886 ``va_list`` element into the destination ``va_list`` element. This
8887 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
8888 arbitrarily complex and require, for example, memory allocation.
8890 Accurate Garbage Collection Intrinsics
8891 --------------------------------------
8893 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
8894 (GC) requires the frontend to generate code containing appropriate intrinsic
8895 calls and select an appropriate GC strategy which knows how to lower these
8896 intrinsics in a manner which is appropriate for the target collector.
8898 These intrinsics allow identification of :ref:`GC roots on the
8899 stack <int_gcroot>`, as well as garbage collector implementations that
8900 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
8901 Frontends for type-safe garbage collected languages should generate
8902 these intrinsics to make use of the LLVM garbage collectors. For more
8903 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
8905 Experimental Statepoint Intrinsics
8906 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8908 LLVM provides an second experimental set of intrinsics for describing garbage
8909 collection safepoints in compiled code. These intrinsics are an alternative
8910 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
8911 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
8912 differences in approach are covered in the `Garbage Collection with LLVM
8913 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
8914 described in :doc:`Statepoints`.
8918 '``llvm.gcroot``' Intrinsic
8919 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8926 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
8931 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
8932 the code generator, and allows some metadata to be associated with it.
8937 The first argument specifies the address of a stack object that contains
8938 the root pointer. The second pointer (which must be either a constant or
8939 a global value address) contains the meta-data to be associated with the
8945 At runtime, a call to this intrinsic stores a null pointer into the
8946 "ptrloc" location. At compile-time, the code generator generates
8947 information to allow the runtime to find the pointer at GC safe points.
8948 The '``llvm.gcroot``' intrinsic may only be used in a function which
8949 :ref:`specifies a GC algorithm <gc>`.
8953 '``llvm.gcread``' Intrinsic
8954 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8961 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
8966 The '``llvm.gcread``' intrinsic identifies reads of references from heap
8967 locations, allowing garbage collector implementations that require read
8973 The second argument is the address to read from, which should be an
8974 address allocated from the garbage collector. The first object is a
8975 pointer to the start of the referenced object, if needed by the language
8976 runtime (otherwise null).
8981 The '``llvm.gcread``' intrinsic has the same semantics as a load
8982 instruction, but may be replaced with substantially more complex code by
8983 the garbage collector runtime, as needed. The '``llvm.gcread``'
8984 intrinsic may only be used in a function which :ref:`specifies a GC
8989 '``llvm.gcwrite``' Intrinsic
8990 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8997 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
9002 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
9003 locations, allowing garbage collector implementations that require write
9004 barriers (such as generational or reference counting collectors).
9009 The first argument is the reference to store, the second is the start of
9010 the object to store it to, and the third is the address of the field of
9011 Obj to store to. If the runtime does not require a pointer to the
9012 object, Obj may be null.
9017 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
9018 instruction, but may be replaced with substantially more complex code by
9019 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
9020 intrinsic may only be used in a function which :ref:`specifies a GC
9023 Code Generator Intrinsics
9024 -------------------------
9026 These intrinsics are provided by LLVM to expose special features that
9027 may only be implemented with code generator support.
9029 '``llvm.returnaddress``' Intrinsic
9030 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9037 declare i8 *@llvm.returnaddress(i32 <level>)
9042 The '``llvm.returnaddress``' intrinsic attempts to compute a
9043 target-specific value indicating the return address of the current
9044 function or one of its callers.
9049 The argument to this intrinsic indicates which function to return the
9050 address for. Zero indicates the calling function, one indicates its
9051 caller, etc. The argument is **required** to be a constant integer
9057 The '``llvm.returnaddress``' intrinsic either returns a pointer
9058 indicating the return address of the specified call frame, or zero if it
9059 cannot be identified. The value returned by this intrinsic is likely to
9060 be incorrect or 0 for arguments other than zero, so it should only be
9061 used for debugging purposes.
9063 Note that calling this intrinsic does not prevent function inlining or
9064 other aggressive transformations, so the value returned may not be that
9065 of the obvious source-language caller.
9067 '``llvm.frameaddress``' Intrinsic
9068 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9075 declare i8* @llvm.frameaddress(i32 <level>)
9080 The '``llvm.frameaddress``' intrinsic attempts to return the
9081 target-specific frame pointer value for the specified stack frame.
9086 The argument to this intrinsic indicates which function to return the
9087 frame pointer for. Zero indicates the calling function, one indicates
9088 its caller, etc. The argument is **required** to be a constant integer
9094 The '``llvm.frameaddress``' intrinsic either returns a pointer
9095 indicating the frame address of the specified call frame, or zero if it
9096 cannot be identified. The value returned by this intrinsic is likely to
9097 be incorrect or 0 for arguments other than zero, so it should only be
9098 used for debugging purposes.
9100 Note that calling this intrinsic does not prevent function inlining or
9101 other aggressive transformations, so the value returned may not be that
9102 of the obvious source-language caller.
9104 '``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
9105 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9112 declare void @llvm.localescape(...)
9113 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
9118 The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
9119 allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
9120 live frame pointer to recover the address of the allocation. The offset is
9121 computed during frame layout of the caller of ``llvm.localescape``.
9126 All arguments to '``llvm.localescape``' must be pointers to static allocas or
9127 casts of static allocas. Each function can only call '``llvm.localescape``'
9128 once, and it can only do so from the entry block.
9130 The ``func`` argument to '``llvm.localrecover``' must be a constant
9131 bitcasted pointer to a function defined in the current module. The code
9132 generator cannot determine the frame allocation offset of functions defined in
9135 The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
9136 call frame that is currently live. The return value of '``llvm.localaddress``'
9137 is one way to produce such a value, but various runtimes also expose a suitable
9138 pointer in platform-specific ways.
9140 The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
9141 '``llvm.localescape``' to recover. It is zero-indexed.
9146 These intrinsics allow a group of functions to share access to a set of local
9147 stack allocations of a one parent function. The parent function may call the
9148 '``llvm.localescape``' intrinsic once from the function entry block, and the
9149 child functions can use '``llvm.localrecover``' to access the escaped allocas.
9150 The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
9151 the escaped allocas are allocated, which would break attempts to use
9152 '``llvm.localrecover``'.
9154 .. _int_read_register:
9155 .. _int_write_register:
9157 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
9158 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9165 declare i32 @llvm.read_register.i32(metadata)
9166 declare i64 @llvm.read_register.i64(metadata)
9167 declare void @llvm.write_register.i32(metadata, i32 @value)
9168 declare void @llvm.write_register.i64(metadata, i64 @value)
9174 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
9175 provides access to the named register. The register must be valid on
9176 the architecture being compiled to. The type needs to be compatible
9177 with the register being read.
9182 The '``llvm.read_register``' intrinsic returns the current value of the
9183 register, where possible. The '``llvm.write_register``' intrinsic sets
9184 the current value of the register, where possible.
9186 This is useful to implement named register global variables that need
9187 to always be mapped to a specific register, as is common practice on
9188 bare-metal programs including OS kernels.
9190 The compiler doesn't check for register availability or use of the used
9191 register in surrounding code, including inline assembly. Because of that,
9192 allocatable registers are not supported.
9194 Warning: So far it only works with the stack pointer on selected
9195 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
9196 work is needed to support other registers and even more so, allocatable
9201 '``llvm.stacksave``' Intrinsic
9202 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9209 declare i8* @llvm.stacksave()
9214 The '``llvm.stacksave``' intrinsic is used to remember the current state
9215 of the function stack, for use with
9216 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
9217 implementing language features like scoped automatic variable sized
9223 This intrinsic returns a opaque pointer value that can be passed to
9224 :ref:`llvm.stackrestore <int_stackrestore>`. When an
9225 ``llvm.stackrestore`` intrinsic is executed with a value saved from
9226 ``llvm.stacksave``, it effectively restores the state of the stack to
9227 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
9228 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
9229 were allocated after the ``llvm.stacksave`` was executed.
9231 .. _int_stackrestore:
9233 '``llvm.stackrestore``' Intrinsic
9234 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9241 declare void @llvm.stackrestore(i8* %ptr)
9246 The '``llvm.stackrestore``' intrinsic is used to restore the state of
9247 the function stack to the state it was in when the corresponding
9248 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
9249 useful for implementing language features like scoped automatic variable
9250 sized arrays in C99.
9255 See the description for :ref:`llvm.stacksave <int_stacksave>`.
9257 '``llvm.prefetch``' Intrinsic
9258 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9265 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
9270 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
9271 insert a prefetch instruction if supported; otherwise, it is a noop.
9272 Prefetches have no effect on the behavior of the program but can change
9273 its performance characteristics.
9278 ``address`` is the address to be prefetched, ``rw`` is the specifier
9279 determining if the fetch should be for a read (0) or write (1), and
9280 ``locality`` is a temporal locality specifier ranging from (0) - no
9281 locality, to (3) - extremely local keep in cache. The ``cache type``
9282 specifies whether the prefetch is performed on the data (1) or
9283 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
9284 arguments must be constant integers.
9289 This intrinsic does not modify the behavior of the program. In
9290 particular, prefetches cannot trap and do not produce a value. On
9291 targets that support this intrinsic, the prefetch can provide hints to
9292 the processor cache for better performance.
9294 '``llvm.pcmarker``' Intrinsic
9295 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9302 declare void @llvm.pcmarker(i32 <id>)
9307 The '``llvm.pcmarker``' intrinsic is a method to export a Program
9308 Counter (PC) in a region of code to simulators and other tools. The
9309 method is target specific, but it is expected that the marker will use
9310 exported symbols to transmit the PC of the marker. The marker makes no
9311 guarantees that it will remain with any specific instruction after
9312 optimizations. It is possible that the presence of a marker will inhibit
9313 optimizations. The intended use is to be inserted after optimizations to
9314 allow correlations of simulation runs.
9319 ``id`` is a numerical id identifying the marker.
9324 This intrinsic does not modify the behavior of the program. Backends
9325 that do not support this intrinsic may ignore it.
9327 '``llvm.readcyclecounter``' Intrinsic
9328 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9335 declare i64 @llvm.readcyclecounter()
9340 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
9341 counter register (or similar low latency, high accuracy clocks) on those
9342 targets that support it. On X86, it should map to RDTSC. On Alpha, it
9343 should map to RPCC. As the backing counters overflow quickly (on the
9344 order of 9 seconds on alpha), this should only be used for small
9350 When directly supported, reading the cycle counter should not modify any
9351 memory. Implementations are allowed to either return a application
9352 specific value or a system wide value. On backends without support, this
9353 is lowered to a constant 0.
9355 Note that runtime support may be conditional on the privilege-level code is
9356 running at and the host platform.
9358 '``llvm.clear_cache``' Intrinsic
9359 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9366 declare void @llvm.clear_cache(i8*, i8*)
9371 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
9372 in the specified range to the execution unit of the processor. On
9373 targets with non-unified instruction and data cache, the implementation
9374 flushes the instruction cache.
9379 On platforms with coherent instruction and data caches (e.g. x86), this
9380 intrinsic is a nop. On platforms with non-coherent instruction and data
9381 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
9382 instructions or a system call, if cache flushing requires special
9385 The default behavior is to emit a call to ``__clear_cache`` from the run
9388 This instrinsic does *not* empty the instruction pipeline. Modifications
9389 of the current function are outside the scope of the intrinsic.
9391 '``llvm.instrprof_increment``' Intrinsic
9392 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9399 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
9400 i32 <num-counters>, i32 <index>)
9405 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
9406 frontend for use with instrumentation based profiling. These will be
9407 lowered by the ``-instrprof`` pass to generate execution counts of a
9413 The first argument is a pointer to a global variable containing the
9414 name of the entity being instrumented. This should generally be the
9415 (mangled) function name for a set of counters.
9417 The second argument is a hash value that can be used by the consumer
9418 of the profile data to detect changes to the instrumented source, and
9419 the third is the number of counters associated with ``name``. It is an
9420 error if ``hash`` or ``num-counters`` differ between two instances of
9421 ``instrprof_increment`` that refer to the same name.
9423 The last argument refers to which of the counters for ``name`` should
9424 be incremented. It should be a value between 0 and ``num-counters``.
9429 This intrinsic represents an increment of a profiling counter. It will
9430 cause the ``-instrprof`` pass to generate the appropriate data
9431 structures and the code to increment the appropriate value, in a
9432 format that can be written out by a compiler runtime and consumed via
9433 the ``llvm-profdata`` tool.
9435 Standard C Library Intrinsics
9436 -----------------------------
9438 LLVM provides intrinsics for a few important standard C library
9439 functions. These intrinsics allow source-language front-ends to pass
9440 information about the alignment of the pointer arguments to the code
9441 generator, providing opportunity for more efficient code generation.
9445 '``llvm.memcpy``' Intrinsic
9446 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9451 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
9452 integer bit width and for different address spaces. Not all targets
9453 support all bit widths however.
9457 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9458 i32 <len>, i32 <align>, i1 <isvolatile>)
9459 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9460 i64 <len>, i32 <align>, i1 <isvolatile>)
9465 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9466 source location to the destination location.
9468 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
9469 intrinsics do not return a value, takes extra alignment/isvolatile
9470 arguments and the pointers can be in specified address spaces.
9475 The first argument is a pointer to the destination, the second is a
9476 pointer to the source. The third argument is an integer argument
9477 specifying the number of bytes to copy, the fourth argument is the
9478 alignment of the source and destination locations, and the fifth is a
9479 boolean indicating a volatile access.
9481 If the call to this intrinsic has an alignment value that is not 0 or 1,
9482 then the caller guarantees that both the source and destination pointers
9483 are aligned to that boundary.
9485 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
9486 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9487 very cleanly specified and it is unwise to depend on it.
9492 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9493 source location to the destination location, which are not allowed to
9494 overlap. It copies "len" bytes of memory over. If the argument is known
9495 to be aligned to some boundary, this can be specified as the fourth
9496 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
9498 '``llvm.memmove``' Intrinsic
9499 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9504 This is an overloaded intrinsic. You can use llvm.memmove on any integer
9505 bit width and for different address space. Not all targets support all
9510 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9511 i32 <len>, i32 <align>, i1 <isvolatile>)
9512 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9513 i64 <len>, i32 <align>, i1 <isvolatile>)
9518 The '``llvm.memmove.*``' intrinsics move a block of memory from the
9519 source location to the destination location. It is similar to the
9520 '``llvm.memcpy``' intrinsic but allows the two memory locations to
9523 Note that, unlike the standard libc function, the ``llvm.memmove.*``
9524 intrinsics do not return a value, takes extra alignment/isvolatile
9525 arguments and the pointers can be in specified address spaces.
9530 The first argument is a pointer to the destination, the second is a
9531 pointer to the source. The third argument is an integer argument
9532 specifying the number of bytes to copy, the fourth argument is the
9533 alignment of the source and destination locations, and the fifth is a
9534 boolean indicating a volatile access.
9536 If the call to this intrinsic has an alignment value that is not 0 or 1,
9537 then the caller guarantees that the source and destination pointers are
9538 aligned to that boundary.
9540 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
9541 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
9542 not very cleanly specified and it is unwise to depend on it.
9547 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
9548 source location to the destination location, which may overlap. It
9549 copies "len" bytes of memory over. If the argument is known to be
9550 aligned to some boundary, this can be specified as the fourth argument,
9551 otherwise it should be set to 0 or 1 (both meaning no alignment).
9553 '``llvm.memset.*``' Intrinsics
9554 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9559 This is an overloaded intrinsic. You can use llvm.memset on any integer
9560 bit width and for different address spaces. However, not all targets
9561 support all bit widths.
9565 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
9566 i32 <len>, i32 <align>, i1 <isvolatile>)
9567 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
9568 i64 <len>, i32 <align>, i1 <isvolatile>)
9573 The '``llvm.memset.*``' intrinsics fill a block of memory with a
9574 particular byte value.
9576 Note that, unlike the standard libc function, the ``llvm.memset``
9577 intrinsic does not return a value and takes extra alignment/volatile
9578 arguments. Also, the destination can be in an arbitrary address space.
9583 The first argument is a pointer to the destination to fill, the second
9584 is the byte value with which to fill it, the third argument is an
9585 integer argument specifying the number of bytes to fill, and the fourth
9586 argument is the known alignment of the destination location.
9588 If the call to this intrinsic has an alignment value that is not 0 or 1,
9589 then the caller guarantees that the destination pointer is aligned to
9592 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
9593 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9594 very cleanly specified and it is unwise to depend on it.
9599 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
9600 at the destination location. If the argument is known to be aligned to
9601 some boundary, this can be specified as the fourth argument, otherwise
9602 it should be set to 0 or 1 (both meaning no alignment).
9604 '``llvm.sqrt.*``' Intrinsic
9605 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9610 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
9611 floating point or vector of floating point type. Not all targets support
9616 declare float @llvm.sqrt.f32(float %Val)
9617 declare double @llvm.sqrt.f64(double %Val)
9618 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
9619 declare fp128 @llvm.sqrt.f128(fp128 %Val)
9620 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
9625 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
9626 returning the same value as the libm '``sqrt``' functions would. Unlike
9627 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
9628 negative numbers other than -0.0 (which allows for better optimization,
9629 because there is no need to worry about errno being set).
9630 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
9635 The argument and return value are floating point numbers of the same
9641 This function returns the sqrt of the specified operand if it is a
9642 nonnegative floating point number.
9644 '``llvm.powi.*``' Intrinsic
9645 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9650 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
9651 floating point or vector of floating point type. Not all targets support
9656 declare float @llvm.powi.f32(float %Val, i32 %power)
9657 declare double @llvm.powi.f64(double %Val, i32 %power)
9658 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
9659 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
9660 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
9665 The '``llvm.powi.*``' intrinsics return the first operand raised to the
9666 specified (positive or negative) power. The order of evaluation of
9667 multiplications is not defined. When a vector of floating point type is
9668 used, the second argument remains a scalar integer value.
9673 The second argument is an integer power, and the first is a value to
9674 raise to that power.
9679 This function returns the first value raised to the second power with an
9680 unspecified sequence of rounding operations.
9682 '``llvm.sin.*``' Intrinsic
9683 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9688 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
9689 floating point or vector of floating point type. Not all targets support
9694 declare float @llvm.sin.f32(float %Val)
9695 declare double @llvm.sin.f64(double %Val)
9696 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
9697 declare fp128 @llvm.sin.f128(fp128 %Val)
9698 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
9703 The '``llvm.sin.*``' intrinsics return the sine of the operand.
9708 The argument and return value are floating point numbers of the same
9714 This function returns the sine of the specified operand, returning the
9715 same values as the libm ``sin`` functions would, and handles error
9716 conditions in the same way.
9718 '``llvm.cos.*``' Intrinsic
9719 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9724 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
9725 floating point or vector of floating point type. Not all targets support
9730 declare float @llvm.cos.f32(float %Val)
9731 declare double @llvm.cos.f64(double %Val)
9732 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
9733 declare fp128 @llvm.cos.f128(fp128 %Val)
9734 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
9739 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
9744 The argument and return value are floating point numbers of the same
9750 This function returns the cosine of the specified operand, returning the
9751 same values as the libm ``cos`` functions would, and handles error
9752 conditions in the same way.
9754 '``llvm.pow.*``' Intrinsic
9755 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9760 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
9761 floating point or vector of floating point type. Not all targets support
9766 declare float @llvm.pow.f32(float %Val, float %Power)
9767 declare double @llvm.pow.f64(double %Val, double %Power)
9768 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
9769 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
9770 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
9775 The '``llvm.pow.*``' intrinsics return the first operand raised to the
9776 specified (positive or negative) power.
9781 The second argument is a floating point power, and the first is a value
9782 to raise to that power.
9787 This function returns the first value raised to the second power,
9788 returning the same values as the libm ``pow`` functions would, and
9789 handles error conditions in the same way.
9791 '``llvm.exp.*``' Intrinsic
9792 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9797 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
9798 floating point or vector of floating point type. Not all targets support
9803 declare float @llvm.exp.f32(float %Val)
9804 declare double @llvm.exp.f64(double %Val)
9805 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
9806 declare fp128 @llvm.exp.f128(fp128 %Val)
9807 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
9812 The '``llvm.exp.*``' intrinsics perform the exp function.
9817 The argument and return value are floating point numbers of the same
9823 This function returns the same values as the libm ``exp`` functions
9824 would, and handles error conditions in the same way.
9826 '``llvm.exp2.*``' Intrinsic
9827 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9832 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
9833 floating point or vector of floating point type. Not all targets support
9838 declare float @llvm.exp2.f32(float %Val)
9839 declare double @llvm.exp2.f64(double %Val)
9840 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
9841 declare fp128 @llvm.exp2.f128(fp128 %Val)
9842 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
9847 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
9852 The argument and return value are floating point numbers of the same
9858 This function returns the same values as the libm ``exp2`` functions
9859 would, and handles error conditions in the same way.
9861 '``llvm.log.*``' Intrinsic
9862 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9867 This is an overloaded intrinsic. You can use ``llvm.log`` on any
9868 floating point or vector of floating point type. Not all targets support
9873 declare float @llvm.log.f32(float %Val)
9874 declare double @llvm.log.f64(double %Val)
9875 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
9876 declare fp128 @llvm.log.f128(fp128 %Val)
9877 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
9882 The '``llvm.log.*``' intrinsics perform the log function.
9887 The argument and return value are floating point numbers of the same
9893 This function returns the same values as the libm ``log`` functions
9894 would, and handles error conditions in the same way.
9896 '``llvm.log10.*``' Intrinsic
9897 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9902 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
9903 floating point or vector of floating point type. Not all targets support
9908 declare float @llvm.log10.f32(float %Val)
9909 declare double @llvm.log10.f64(double %Val)
9910 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
9911 declare fp128 @llvm.log10.f128(fp128 %Val)
9912 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
9917 The '``llvm.log10.*``' intrinsics perform the log10 function.
9922 The argument and return value are floating point numbers of the same
9928 This function returns the same values as the libm ``log10`` functions
9929 would, and handles error conditions in the same way.
9931 '``llvm.log2.*``' Intrinsic
9932 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9937 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
9938 floating point or vector of floating point type. Not all targets support
9943 declare float @llvm.log2.f32(float %Val)
9944 declare double @llvm.log2.f64(double %Val)
9945 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
9946 declare fp128 @llvm.log2.f128(fp128 %Val)
9947 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
9952 The '``llvm.log2.*``' intrinsics perform the log2 function.
9957 The argument and return value are floating point numbers of the same
9963 This function returns the same values as the libm ``log2`` functions
9964 would, and handles error conditions in the same way.
9966 '``llvm.fma.*``' Intrinsic
9967 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9972 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
9973 floating point or vector of floating point type. Not all targets support
9978 declare float @llvm.fma.f32(float %a, float %b, float %c)
9979 declare double @llvm.fma.f64(double %a, double %b, double %c)
9980 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
9981 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
9982 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
9987 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
9993 The argument and return value are floating point numbers of the same
9999 This function returns the same values as the libm ``fma`` functions
10000 would, and does not set errno.
10002 '``llvm.fabs.*``' Intrinsic
10003 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10008 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
10009 floating point or vector of floating point type. Not all targets support
10014 declare float @llvm.fabs.f32(float %Val)
10015 declare double @llvm.fabs.f64(double %Val)
10016 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
10017 declare fp128 @llvm.fabs.f128(fp128 %Val)
10018 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
10023 The '``llvm.fabs.*``' intrinsics return the absolute value of the
10029 The argument and return value are floating point numbers of the same
10035 This function returns the same values as the libm ``fabs`` functions
10036 would, and handles error conditions in the same way.
10038 '``llvm.minnum.*``' Intrinsic
10039 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10044 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
10045 floating point or vector of floating point type. Not all targets support
10050 declare float @llvm.minnum.f32(float %Val0, float %Val1)
10051 declare double @llvm.minnum.f64(double %Val0, double %Val1)
10052 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10053 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
10054 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10059 The '``llvm.minnum.*``' intrinsics return the minimum of the two
10066 The arguments and return value are floating point numbers of the same
10072 Follows the IEEE-754 semantics for minNum, which also match for libm's
10075 If either operand is a NaN, returns the other non-NaN operand. Returns
10076 NaN only if both operands are NaN. If the operands compare equal,
10077 returns a value that compares equal to both operands. This means that
10078 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10080 '``llvm.maxnum.*``' Intrinsic
10081 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10086 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
10087 floating point or vector of floating point type. Not all targets support
10092 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
10093 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
10094 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10095 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
10096 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10101 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
10108 The arguments and return value are floating point numbers of the same
10113 Follows the IEEE-754 semantics for maxNum, which also match for libm's
10116 If either operand is a NaN, returns the other non-NaN operand. Returns
10117 NaN only if both operands are NaN. If the operands compare equal,
10118 returns a value that compares equal to both operands. This means that
10119 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10121 '``llvm.copysign.*``' Intrinsic
10122 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10127 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
10128 floating point or vector of floating point type. Not all targets support
10133 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
10134 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
10135 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
10136 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
10137 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
10142 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
10143 first operand and the sign of the second operand.
10148 The arguments and return value are floating point numbers of the same
10154 This function returns the same values as the libm ``copysign``
10155 functions would, and handles error conditions in the same way.
10157 '``llvm.floor.*``' Intrinsic
10158 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10163 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
10164 floating point or vector of floating point type. Not all targets support
10169 declare float @llvm.floor.f32(float %Val)
10170 declare double @llvm.floor.f64(double %Val)
10171 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
10172 declare fp128 @llvm.floor.f128(fp128 %Val)
10173 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
10178 The '``llvm.floor.*``' intrinsics return the floor of the operand.
10183 The argument and return value are floating point numbers of the same
10189 This function returns the same values as the libm ``floor`` functions
10190 would, and handles error conditions in the same way.
10192 '``llvm.ceil.*``' Intrinsic
10193 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10198 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
10199 floating point or vector of floating point type. Not all targets support
10204 declare float @llvm.ceil.f32(float %Val)
10205 declare double @llvm.ceil.f64(double %Val)
10206 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
10207 declare fp128 @llvm.ceil.f128(fp128 %Val)
10208 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
10213 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
10218 The argument and return value are floating point numbers of the same
10224 This function returns the same values as the libm ``ceil`` functions
10225 would, and handles error conditions in the same way.
10227 '``llvm.trunc.*``' Intrinsic
10228 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10233 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
10234 floating point or vector of floating point type. Not all targets support
10239 declare float @llvm.trunc.f32(float %Val)
10240 declare double @llvm.trunc.f64(double %Val)
10241 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
10242 declare fp128 @llvm.trunc.f128(fp128 %Val)
10243 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
10248 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
10249 nearest integer not larger in magnitude than the operand.
10254 The argument and return value are floating point numbers of the same
10260 This function returns the same values as the libm ``trunc`` functions
10261 would, and handles error conditions in the same way.
10263 '``llvm.rint.*``' Intrinsic
10264 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10269 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
10270 floating point or vector of floating point type. Not all targets support
10275 declare float @llvm.rint.f32(float %Val)
10276 declare double @llvm.rint.f64(double %Val)
10277 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
10278 declare fp128 @llvm.rint.f128(fp128 %Val)
10279 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
10284 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
10285 nearest integer. It may raise an inexact floating-point exception if the
10286 operand isn't an integer.
10291 The argument and return value are floating point numbers of the same
10297 This function returns the same values as the libm ``rint`` functions
10298 would, and handles error conditions in the same way.
10300 '``llvm.nearbyint.*``' Intrinsic
10301 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10306 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
10307 floating point or vector of floating point type. Not all targets support
10312 declare float @llvm.nearbyint.f32(float %Val)
10313 declare double @llvm.nearbyint.f64(double %Val)
10314 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
10315 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
10316 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
10321 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
10327 The argument and return value are floating point numbers of the same
10333 This function returns the same values as the libm ``nearbyint``
10334 functions would, and handles error conditions in the same way.
10336 '``llvm.round.*``' Intrinsic
10337 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10342 This is an overloaded intrinsic. You can use ``llvm.round`` on any
10343 floating point or vector of floating point type. Not all targets support
10348 declare float @llvm.round.f32(float %Val)
10349 declare double @llvm.round.f64(double %Val)
10350 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
10351 declare fp128 @llvm.round.f128(fp128 %Val)
10352 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
10357 The '``llvm.round.*``' intrinsics returns the operand rounded to the
10363 The argument and return value are floating point numbers of the same
10369 This function returns the same values as the libm ``round``
10370 functions would, and handles error conditions in the same way.
10372 Bit Manipulation Intrinsics
10373 ---------------------------
10375 LLVM provides intrinsics for a few important bit manipulation
10376 operations. These allow efficient code generation for some algorithms.
10378 '``llvm.bswap.*``' Intrinsics
10379 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10384 This is an overloaded intrinsic function. You can use bswap on any
10385 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
10389 declare i16 @llvm.bswap.i16(i16 <id>)
10390 declare i32 @llvm.bswap.i32(i32 <id>)
10391 declare i64 @llvm.bswap.i64(i64 <id>)
10396 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
10397 values with an even number of bytes (positive multiple of 16 bits).
10398 These are useful for performing operations on data that is not in the
10399 target's native byte order.
10404 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
10405 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
10406 intrinsic returns an i32 value that has the four bytes of the input i32
10407 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
10408 returned i32 will have its bytes in 3, 2, 1, 0 order. The
10409 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
10410 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
10413 '``llvm.ctpop.*``' Intrinsic
10414 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10419 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
10420 bit width, or on any vector with integer elements. Not all targets
10421 support all bit widths or vector types, however.
10425 declare i8 @llvm.ctpop.i8(i8 <src>)
10426 declare i16 @llvm.ctpop.i16(i16 <src>)
10427 declare i32 @llvm.ctpop.i32(i32 <src>)
10428 declare i64 @llvm.ctpop.i64(i64 <src>)
10429 declare i256 @llvm.ctpop.i256(i256 <src>)
10430 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
10435 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
10441 The only argument is the value to be counted. The argument may be of any
10442 integer type, or a vector with integer elements. The return type must
10443 match the argument type.
10448 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
10449 each element of a vector.
10451 '``llvm.ctlz.*``' Intrinsic
10452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10457 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
10458 integer bit width, or any vector whose elements are integers. Not all
10459 targets support all bit widths or vector types, however.
10463 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
10464 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
10465 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
10466 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
10467 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
10468 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10473 The '``llvm.ctlz``' family of intrinsic functions counts the number of
10474 leading zeros in a variable.
10479 The first argument is the value to be counted. This argument may be of
10480 any integer type, or a vector with integer element type. The return
10481 type must match the first argument type.
10483 The second argument must be a constant and is a flag to indicate whether
10484 the intrinsic should ensure that a zero as the first argument produces a
10485 defined result. Historically some architectures did not provide a
10486 defined result for zero values as efficiently, and many algorithms are
10487 now predicated on avoiding zero-value inputs.
10492 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
10493 zeros in a variable, or within each element of the vector. If
10494 ``src == 0`` then the result is the size in bits of the type of ``src``
10495 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10496 ``llvm.ctlz(i32 2) = 30``.
10498 '``llvm.cttz.*``' Intrinsic
10499 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10504 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
10505 integer bit width, or any vector of integer elements. Not all targets
10506 support all bit widths or vector types, however.
10510 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
10511 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
10512 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
10513 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
10514 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
10515 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10520 The '``llvm.cttz``' family of intrinsic functions counts the number of
10526 The first argument is the value to be counted. This argument may be of
10527 any integer type, or a vector with integer element type. The return
10528 type must match the first argument type.
10530 The second argument must be a constant and is a flag to indicate whether
10531 the intrinsic should ensure that a zero as the first argument produces a
10532 defined result. Historically some architectures did not provide a
10533 defined result for zero values as efficiently, and many algorithms are
10534 now predicated on avoiding zero-value inputs.
10539 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
10540 zeros in a variable, or within each element of a vector. If ``src == 0``
10541 then the result is the size in bits of the type of ``src`` if
10542 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10543 ``llvm.cttz(2) = 1``.
10547 Arithmetic with Overflow Intrinsics
10548 -----------------------------------
10550 LLVM provides intrinsics for some arithmetic with overflow operations.
10552 '``llvm.sadd.with.overflow.*``' Intrinsics
10553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10558 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
10559 on any integer bit width.
10563 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
10564 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10565 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
10570 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10571 a signed addition of the two arguments, and indicate whether an overflow
10572 occurred during the signed summation.
10577 The arguments (%a and %b) and the first element of the result structure
10578 may be of integer types of any bit width, but they must have the same
10579 bit width. The second element of the result structure must be of type
10580 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10586 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10587 a signed addition of the two variables. They return a structure --- the
10588 first element of which is the signed summation, and the second element
10589 of which is a bit specifying if the signed summation resulted in an
10595 .. code-block:: llvm
10597 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10598 %sum = extractvalue {i32, i1} %res, 0
10599 %obit = extractvalue {i32, i1} %res, 1
10600 br i1 %obit, label %overflow, label %normal
10602 '``llvm.uadd.with.overflow.*``' Intrinsics
10603 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10608 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
10609 on any integer bit width.
10613 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
10614 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10615 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
10620 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10621 an unsigned addition of the two arguments, and indicate whether a carry
10622 occurred during the unsigned summation.
10627 The arguments (%a and %b) and the first element of the result structure
10628 may be of integer types of any bit width, but they must have the same
10629 bit width. The second element of the result structure must be of type
10630 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10636 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10637 an unsigned addition of the two arguments. They return a structure --- the
10638 first element of which is the sum, and the second element of which is a
10639 bit specifying if the unsigned summation resulted in a carry.
10644 .. code-block:: llvm
10646 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10647 %sum = extractvalue {i32, i1} %res, 0
10648 %obit = extractvalue {i32, i1} %res, 1
10649 br i1 %obit, label %carry, label %normal
10651 '``llvm.ssub.with.overflow.*``' Intrinsics
10652 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10657 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
10658 on any integer bit width.
10662 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
10663 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10664 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
10669 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10670 a signed subtraction of the two arguments, and indicate whether an
10671 overflow occurred during the signed subtraction.
10676 The arguments (%a and %b) and the first element of the result structure
10677 may be of integer types of any bit width, but they must have the same
10678 bit width. The second element of the result structure must be of type
10679 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10685 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10686 a signed subtraction of the two arguments. They return a structure --- the
10687 first element of which is the subtraction, and the second element of
10688 which is a bit specifying if the signed subtraction resulted in an
10694 .. code-block:: llvm
10696 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10697 %sum = extractvalue {i32, i1} %res, 0
10698 %obit = extractvalue {i32, i1} %res, 1
10699 br i1 %obit, label %overflow, label %normal
10701 '``llvm.usub.with.overflow.*``' Intrinsics
10702 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10707 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
10708 on any integer bit width.
10712 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
10713 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10714 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
10719 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10720 an unsigned subtraction of the two arguments, and indicate whether an
10721 overflow occurred during the unsigned subtraction.
10726 The arguments (%a and %b) and the first element of the result structure
10727 may be of integer types of any bit width, but they must have the same
10728 bit width. The second element of the result structure must be of type
10729 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10735 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10736 an unsigned subtraction of the two arguments. They return a structure ---
10737 the first element of which is the subtraction, and the second element of
10738 which is a bit specifying if the unsigned subtraction resulted in an
10744 .. code-block:: llvm
10746 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10747 %sum = extractvalue {i32, i1} %res, 0
10748 %obit = extractvalue {i32, i1} %res, 1
10749 br i1 %obit, label %overflow, label %normal
10751 '``llvm.smul.with.overflow.*``' Intrinsics
10752 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10757 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
10758 on any integer bit width.
10762 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
10763 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10764 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
10769 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10770 a signed multiplication of the two arguments, and indicate whether an
10771 overflow occurred during the signed multiplication.
10776 The arguments (%a and %b) and the first element of the result structure
10777 may be of integer types of any bit width, but they must have the same
10778 bit width. The second element of the result structure must be of type
10779 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10785 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10786 a signed multiplication of the two arguments. They return a structure ---
10787 the first element of which is the multiplication, and the second element
10788 of which is a bit specifying if the signed multiplication resulted in an
10794 .. code-block:: llvm
10796 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10797 %sum = extractvalue {i32, i1} %res, 0
10798 %obit = extractvalue {i32, i1} %res, 1
10799 br i1 %obit, label %overflow, label %normal
10801 '``llvm.umul.with.overflow.*``' Intrinsics
10802 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10807 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
10808 on any integer bit width.
10812 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
10813 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10814 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
10819 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10820 a unsigned multiplication of the two arguments, and indicate whether an
10821 overflow occurred during the unsigned multiplication.
10826 The arguments (%a and %b) and the first element of the result structure
10827 may be of integer types of any bit width, but they must have the same
10828 bit width. The second element of the result structure must be of type
10829 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10835 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10836 an unsigned multiplication of the two arguments. They return a structure ---
10837 the first element of which is the multiplication, and the second
10838 element of which is a bit specifying if the unsigned multiplication
10839 resulted in an overflow.
10844 .. code-block:: llvm
10846 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10847 %sum = extractvalue {i32, i1} %res, 0
10848 %obit = extractvalue {i32, i1} %res, 1
10849 br i1 %obit, label %overflow, label %normal
10851 Specialised Arithmetic Intrinsics
10852 ---------------------------------
10854 '``llvm.canonicalize.*``' Intrinsic
10855 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10862 declare float @llvm.canonicalize.f32(float %a)
10863 declare double @llvm.canonicalize.f64(double %b)
10868 The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
10869 encoding of a floating point number. This canonicalization is useful for
10870 implementing certain numeric primitives such as frexp. The canonical encoding is
10871 defined by IEEE-754-2008 to be:
10875 2.1.8 canonical encoding: The preferred encoding of a floating-point
10876 representation in a format. Applied to declets, significands of finite
10877 numbers, infinities, and NaNs, especially in decimal formats.
10879 This operation can also be considered equivalent to the IEEE-754-2008
10880 conversion of a floating-point value to the same format. NaNs are handled
10881 according to section 6.2.
10883 Examples of non-canonical encodings:
10885 - x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
10886 converted to a canonical representation per hardware-specific protocol.
10887 - Many normal decimal floating point numbers have non-canonical alternative
10889 - Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
10890 These are treated as non-canonical encodings of zero and with be flushed to
10891 a zero of the same sign by this operation.
10893 Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
10894 default exception handling must signal an invalid exception, and produce a
10897 This function should always be implementable as multiplication by 1.0, provided
10898 that the compiler does not constant fold the operation. Likewise, division by
10899 1.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
10900 -0.0 is also sufficient provided that the rounding mode is not -Infinity.
10902 ``@llvm.canonicalize`` must preserve the equality relation. That is:
10904 - ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
10905 - ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
10908 Additionally, the sign of zero must be conserved:
10909 ``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
10911 The payload bits of a NaN must be conserved, with two exceptions.
10912 First, environments which use only a single canonical representation of NaN
10913 must perform said canonicalization. Second, SNaNs must be quieted per the
10916 The canonicalization operation may be optimized away if:
10918 - The input is known to be canonical. For example, it was produced by a
10919 floating-point operation that is required by the standard to be canonical.
10920 - The result is consumed only by (or fused with) other floating-point
10921 operations. That is, the bits of the floating point value are not examined.
10923 '``llvm.fmuladd.*``' Intrinsic
10924 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10931 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
10932 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
10937 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
10938 expressions that can be fused if the code generator determines that (a) the
10939 target instruction set has support for a fused operation, and (b) that the
10940 fused operation is more efficient than the equivalent, separate pair of mul
10941 and add instructions.
10946 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
10947 multiplicands, a and b, and an addend c.
10956 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
10958 is equivalent to the expression a \* b + c, except that rounding will
10959 not be performed between the multiplication and addition steps if the
10960 code generator fuses the operations. Fusion is not guaranteed, even if
10961 the target platform supports it. If a fused multiply-add is required the
10962 corresponding llvm.fma.\* intrinsic function should be used
10963 instead. This never sets errno, just as '``llvm.fma.*``'.
10968 .. code-block:: llvm
10970 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
10973 '``llvm.uabsdiff.*``' and '``llvm.sabsdiff.*``' Intrinsics
10974 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10978 This is an overloaded intrinsic. The loaded data is a vector of any integer bit width.
10980 .. code-block:: llvm
10982 declare <4 x integer> @llvm.uabsdiff.v4i32(<4 x integer> %a, <4 x integer> %b)
10988 The ``llvm.uabsdiff`` intrinsic returns a vector result of the absolute difference
10989 of the two operands, treating them both as unsigned integers. The intermediate
10990 calculations are computed using infinitely precise unsigned arithmetic. The final
10991 result will be truncated to the given type.
10993 The ``llvm.sabsdiff`` intrinsic returns a vector result of the absolute difference of
10994 the two operands, treating them both as signed integers. If the result overflows, the
10995 behavior is undefined.
10999 These intrinsics are primarily used during the code generation stage of compilation.
11000 They are generated by compiler passes such as the Loop and SLP vectorizers. It is not
11001 recommended for users to create them manually.
11006 Both intrinsics take two integer of the same bitwidth.
11013 call <4 x i32> @llvm.uabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11017 %1 = zext <4 x i32> %a to <4 x i64>
11018 %2 = zext <4 x i32> %b to <4 x i64>
11019 %sub = sub <4 x i64> %1, %2
11020 %trunc = trunc <4 x i64> to <4 x i32>
11022 and the expression::
11024 call <4 x i32> @llvm.sabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11028 %sub = sub nsw <4 x i32> %a, %b
11029 %ispos = icmp sge <4 x i32> %sub, zeroinitializer
11030 %neg = sub nsw <4 x i32> zeroinitializer, %sub
11031 %1 = select <4 x i1> %ispos, <4 x i32> %sub, <4 x i32> %neg
11034 Half Precision Floating Point Intrinsics
11035 ----------------------------------------
11037 For most target platforms, half precision floating point is a
11038 storage-only format. This means that it is a dense encoding (in memory)
11039 but does not support computation in the format.
11041 This means that code must first load the half-precision floating point
11042 value as an i16, then convert it to float with
11043 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
11044 then be performed on the float value (including extending to double
11045 etc). To store the value back to memory, it is first converted to float
11046 if needed, then converted to i16 with
11047 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
11050 .. _int_convert_to_fp16:
11052 '``llvm.convert.to.fp16``' Intrinsic
11053 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11060 declare i16 @llvm.convert.to.fp16.f32(float %a)
11061 declare i16 @llvm.convert.to.fp16.f64(double %a)
11066 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11067 conventional floating point type to half precision floating point format.
11072 The intrinsic function contains single argument - the value to be
11078 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11079 conventional floating point format to half precision floating point format. The
11080 return value is an ``i16`` which contains the converted number.
11085 .. code-block:: llvm
11087 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
11088 store i16 %res, i16* @x, align 2
11090 .. _int_convert_from_fp16:
11092 '``llvm.convert.from.fp16``' Intrinsic
11093 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11100 declare float @llvm.convert.from.fp16.f32(i16 %a)
11101 declare double @llvm.convert.from.fp16.f64(i16 %a)
11106 The '``llvm.convert.from.fp16``' intrinsic function performs a
11107 conversion from half precision floating point format to single precision
11108 floating point format.
11113 The intrinsic function contains single argument - the value to be
11119 The '``llvm.convert.from.fp16``' intrinsic function performs a
11120 conversion from half single precision floating point format to single
11121 precision floating point format. The input half-float value is
11122 represented by an ``i16`` value.
11127 .. code-block:: llvm
11129 %a = load i16, i16* @x, align 2
11130 %res = call float @llvm.convert.from.fp16(i16 %a)
11132 .. _dbg_intrinsics:
11134 Debugger Intrinsics
11135 -------------------
11137 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
11138 prefix), are described in the `LLVM Source Level
11139 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
11142 Exception Handling Intrinsics
11143 -----------------------------
11145 The LLVM exception handling intrinsics (which all start with
11146 ``llvm.eh.`` prefix), are described in the `LLVM Exception
11147 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
11149 .. _int_trampoline:
11151 Trampoline Intrinsics
11152 ---------------------
11154 These intrinsics make it possible to excise one parameter, marked with
11155 the :ref:`nest <nest>` attribute, from a function. The result is a
11156 callable function pointer lacking the nest parameter - the caller does
11157 not need to provide a value for it. Instead, the value to use is stored
11158 in advance in a "trampoline", a block of memory usually allocated on the
11159 stack, which also contains code to splice the nest value into the
11160 argument list. This is used to implement the GCC nested function address
11163 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
11164 then the resulting function pointer has signature ``i32 (i32, i32)*``.
11165 It can be created as follows:
11167 .. code-block:: llvm
11169 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
11170 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
11171 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
11172 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
11173 %fp = bitcast i8* %p to i32 (i32, i32)*
11175 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
11176 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
11180 '``llvm.init.trampoline``' Intrinsic
11181 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11188 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
11193 This fills the memory pointed to by ``tramp`` with executable code,
11194 turning it into a trampoline.
11199 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
11200 pointers. The ``tramp`` argument must point to a sufficiently large and
11201 sufficiently aligned block of memory; this memory is written to by the
11202 intrinsic. Note that the size and the alignment are target-specific -
11203 LLVM currently provides no portable way of determining them, so a
11204 front-end that generates this intrinsic needs to have some
11205 target-specific knowledge. The ``func`` argument must hold a function
11206 bitcast to an ``i8*``.
11211 The block of memory pointed to by ``tramp`` is filled with target
11212 dependent code, turning it into a function. Then ``tramp`` needs to be
11213 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
11214 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
11215 function's signature is the same as that of ``func`` with any arguments
11216 marked with the ``nest`` attribute removed. At most one such ``nest``
11217 argument is allowed, and it must be of pointer type. Calling the new
11218 function is equivalent to calling ``func`` with the same argument list,
11219 but with ``nval`` used for the missing ``nest`` argument. If, after
11220 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
11221 modified, then the effect of any later call to the returned function
11222 pointer is undefined.
11226 '``llvm.adjust.trampoline``' Intrinsic
11227 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11234 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
11239 This performs any required machine-specific adjustment to the address of
11240 a trampoline (passed as ``tramp``).
11245 ``tramp`` must point to a block of memory which already has trampoline
11246 code filled in by a previous call to
11247 :ref:`llvm.init.trampoline <int_it>`.
11252 On some architectures the address of the code to be executed needs to be
11253 different than the address where the trampoline is actually stored. This
11254 intrinsic returns the executable address corresponding to ``tramp``
11255 after performing the required machine specific adjustments. The pointer
11256 returned can then be :ref:`bitcast and executed <int_trampoline>`.
11258 .. _int_mload_mstore:
11260 Masked Vector Load and Store Intrinsics
11261 ---------------------------------------
11263 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.
11267 '``llvm.masked.load.*``' Intrinsics
11268 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11272 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
11276 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11277 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11282 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.
11288 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.
11294 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.
11295 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.
11300 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
11302 ;; The result of the two following instructions is identical aside from potential memory access exception
11303 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
11304 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
11308 '``llvm.masked.store.*``' Intrinsics
11309 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11313 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
11317 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
11318 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
11323 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.
11328 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.
11334 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.
11335 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.
11339 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
11341 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
11342 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
11343 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
11344 store <16 x float> %res, <16 x float>* %ptr, align 4
11347 Masked Vector Gather and Scatter Intrinsics
11348 -------------------------------------------
11350 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.
11354 '``llvm.masked.gather.*``' Intrinsics
11355 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11359 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.
11363 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11364 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11369 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.
11375 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.
11381 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.
11382 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.
11387 %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>)
11389 ;; The gather with all-true mask is equivalent to the following instruction sequence
11390 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
11391 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
11392 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
11393 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
11395 %val0 = load double, double* %ptr0, align 8
11396 %val1 = load double, double* %ptr1, align 8
11397 %val2 = load double, double* %ptr2, align 8
11398 %val3 = load double, double* %ptr3, align 8
11400 %vec0 = insertelement <4 x double>undef, %val0, 0
11401 %vec01 = insertelement <4 x double>%vec0, %val1, 1
11402 %vec012 = insertelement <4 x double>%vec01, %val2, 2
11403 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
11407 '``llvm.masked.scatter.*``' Intrinsics
11408 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11412 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.
11416 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
11417 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
11422 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.
11427 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.
11433 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.
11437 ;; This instruction unconditionaly stores data vector in multiple addresses
11438 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
11440 ;; It is equivalent to a list of scalar stores
11441 %val0 = extractelement <8 x i32> %value, i32 0
11442 %val1 = extractelement <8 x i32> %value, i32 1
11444 %val7 = extractelement <8 x i32> %value, i32 7
11445 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
11446 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
11448 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
11449 ;; Note: the order of the following stores is important when they overlap:
11450 store i32 %val0, i32* %ptr0, align 4
11451 store i32 %val1, i32* %ptr1, align 4
11453 store i32 %val7, i32* %ptr7, align 4
11459 This class of intrinsics provides information about the lifetime of
11460 memory objects and ranges where variables are immutable.
11464 '``llvm.lifetime.start``' Intrinsic
11465 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11472 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
11477 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
11483 The first argument is a constant integer representing the size of the
11484 object, or -1 if it is variable sized. The second argument is a pointer
11490 This intrinsic indicates that before this point in the code, the value
11491 of the memory pointed to by ``ptr`` is dead. This means that it is known
11492 to never be used and has an undefined value. A load from the pointer
11493 that precedes this intrinsic can be replaced with ``'undef'``.
11497 '``llvm.lifetime.end``' Intrinsic
11498 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11505 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
11510 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
11516 The first argument is a constant integer representing the size of the
11517 object, or -1 if it is variable sized. The second argument is a pointer
11523 This intrinsic indicates that after this point in the code, the value of
11524 the memory pointed to by ``ptr`` is dead. This means that it is known to
11525 never be used and has an undefined value. Any stores into the memory
11526 object following this intrinsic may be removed as dead.
11528 '``llvm.invariant.start``' Intrinsic
11529 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11536 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
11541 The '``llvm.invariant.start``' intrinsic specifies that the contents of
11542 a memory object will not change.
11547 The first argument is a constant integer representing the size of the
11548 object, or -1 if it is variable sized. The second argument is a pointer
11554 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
11555 the return value, the referenced memory location is constant and
11558 '``llvm.invariant.end``' Intrinsic
11559 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11566 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
11571 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
11572 memory object are mutable.
11577 The first argument is the matching ``llvm.invariant.start`` intrinsic.
11578 The second argument is a constant integer representing the size of the
11579 object, or -1 if it is variable sized and the third argument is a
11580 pointer to the object.
11585 This intrinsic indicates that the memory is mutable again.
11587 '``llvm.invariant.group.barrier``' Intrinsic
11588 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11595 declare i8* @llvm.invariant.group.barrier(i8* <ptr>)
11600 The '``llvm.invariant.group.barrier``' intrinsic can be used when an invariant
11601 established by invariant.group metadata no longer holds, to obtain a new pointer
11602 value that does not carry the invariant information.
11608 The ``llvm.invariant.group.barrier`` takes only one argument, which is
11609 the pointer to the memory for which the ``invariant.group`` no longer holds.
11614 Returns another pointer that aliases its argument but which is considered different
11615 for the purposes of ``load``/``store`` ``invariant.group`` metadata.
11620 This class of intrinsics is designed to be generic and has no specific
11623 '``llvm.var.annotation``' Intrinsic
11624 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11631 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11636 The '``llvm.var.annotation``' intrinsic.
11641 The first argument is a pointer to a value, the second is a pointer to a
11642 global string, the third is a pointer to a global string which is the
11643 source file name, and the last argument is the line number.
11648 This intrinsic allows annotation of local variables with arbitrary
11649 strings. This can be useful for special purpose optimizations that want
11650 to look for these annotations. These have no other defined use; they are
11651 ignored by code generation and optimization.
11653 '``llvm.ptr.annotation.*``' Intrinsic
11654 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11659 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
11660 pointer to an integer of any width. *NOTE* you must specify an address space for
11661 the pointer. The identifier for the default address space is the integer
11666 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11667 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
11668 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
11669 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
11670 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
11675 The '``llvm.ptr.annotation``' intrinsic.
11680 The first argument is a pointer to an integer value of arbitrary bitwidth
11681 (result of some expression), the second is a pointer to a global string, the
11682 third is a pointer to a global string which is the source file name, and the
11683 last argument is the line number. It returns the value of the first argument.
11688 This intrinsic allows annotation of a pointer to an integer with arbitrary
11689 strings. This can be useful for special purpose optimizations that want to look
11690 for these annotations. These have no other defined use; they are ignored by code
11691 generation and optimization.
11693 '``llvm.annotation.*``' Intrinsic
11694 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11699 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
11700 any integer bit width.
11704 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
11705 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
11706 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
11707 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
11708 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
11713 The '``llvm.annotation``' intrinsic.
11718 The first argument is an integer value (result of some expression), the
11719 second is a pointer to a global string, the third is a pointer to a
11720 global string which is the source file name, and the last argument is
11721 the line number. It returns the value of the first argument.
11726 This intrinsic allows annotations to be put on arbitrary expressions
11727 with arbitrary strings. This can be useful for special purpose
11728 optimizations that want to look for these annotations. These have no
11729 other defined use; they are ignored by code generation and optimization.
11731 '``llvm.trap``' Intrinsic
11732 ^^^^^^^^^^^^^^^^^^^^^^^^^
11739 declare void @llvm.trap() noreturn nounwind
11744 The '``llvm.trap``' intrinsic.
11754 This intrinsic is lowered to the target dependent trap instruction. If
11755 the target does not have a trap instruction, this intrinsic will be
11756 lowered to a call of the ``abort()`` function.
11758 '``llvm.debugtrap``' Intrinsic
11759 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11766 declare void @llvm.debugtrap() nounwind
11771 The '``llvm.debugtrap``' intrinsic.
11781 This intrinsic is lowered to code which is intended to cause an
11782 execution trap with the intention of requesting the attention of a
11785 '``llvm.stackprotector``' Intrinsic
11786 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11793 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
11798 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
11799 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
11800 is placed on the stack before local variables.
11805 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
11806 The first argument is the value loaded from the stack guard
11807 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
11808 enough space to hold the value of the guard.
11813 This intrinsic causes the prologue/epilogue inserter to force the position of
11814 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
11815 to ensure that if a local variable on the stack is overwritten, it will destroy
11816 the value of the guard. When the function exits, the guard on the stack is
11817 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
11818 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
11819 calling the ``__stack_chk_fail()`` function.
11821 '``llvm.stackprotectorcheck``' Intrinsic
11822 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11829 declare void @llvm.stackprotectorcheck(i8** <guard>)
11834 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
11835 created stack protector and if they are not equal calls the
11836 ``__stack_chk_fail()`` function.
11841 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
11842 the variable ``@__stack_chk_guard``.
11847 This intrinsic is provided to perform the stack protector check by comparing
11848 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
11849 values do not match call the ``__stack_chk_fail()`` function.
11851 The reason to provide this as an IR level intrinsic instead of implementing it
11852 via other IR operations is that in order to perform this operation at the IR
11853 level without an intrinsic, one would need to create additional basic blocks to
11854 handle the success/failure cases. This makes it difficult to stop the stack
11855 protector check from disrupting sibling tail calls in Codegen. With this
11856 intrinsic, we are able to generate the stack protector basic blocks late in
11857 codegen after the tail call decision has occurred.
11859 '``llvm.objectsize``' Intrinsic
11860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11867 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
11868 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
11873 The ``llvm.objectsize`` intrinsic is designed to provide information to
11874 the optimizers to determine at compile time whether a) an operation
11875 (like memcpy) will overflow a buffer that corresponds to an object, or
11876 b) that a runtime check for overflow isn't necessary. An object in this
11877 context means an allocation of a specific class, structure, array, or
11883 The ``llvm.objectsize`` intrinsic takes two arguments. The first
11884 argument is a pointer to or into the ``object``. The second argument is
11885 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
11886 or -1 (if false) when the object size is unknown. The second argument
11887 only accepts constants.
11892 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
11893 the size of the object concerned. If the size cannot be determined at
11894 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
11895 on the ``min`` argument).
11897 '``llvm.expect``' Intrinsic
11898 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11903 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
11908 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
11909 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
11910 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
11915 The ``llvm.expect`` intrinsic provides information about expected (the
11916 most probable) value of ``val``, which can be used by optimizers.
11921 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
11922 a value. The second argument is an expected value, this needs to be a
11923 constant value, variables are not allowed.
11928 This intrinsic is lowered to the ``val``.
11932 '``llvm.assume``' Intrinsic
11933 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11940 declare void @llvm.assume(i1 %cond)
11945 The ``llvm.assume`` allows the optimizer to assume that the provided
11946 condition is true. This information can then be used in simplifying other parts
11952 The condition which the optimizer may assume is always true.
11957 The intrinsic allows the optimizer to assume that the provided condition is
11958 always true whenever the control flow reaches the intrinsic call. No code is
11959 generated for this intrinsic, and instructions that contribute only to the
11960 provided condition are not used for code generation. If the condition is
11961 violated during execution, the behavior is undefined.
11963 Note that the optimizer might limit the transformations performed on values
11964 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
11965 only used to form the intrinsic's input argument. This might prove undesirable
11966 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
11967 sufficient overall improvement in code quality. For this reason,
11968 ``llvm.assume`` should not be used to document basic mathematical invariants
11969 that the optimizer can otherwise deduce or facts that are of little use to the
11974 '``llvm.bitset.test``' Intrinsic
11975 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11982 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
11988 The first argument is a pointer to be tested. The second argument is a
11989 metadata object representing an identifier for a :doc:`bitset <BitSets>`.
11994 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
11995 member of the given bitset.
11997 '``llvm.donothing``' Intrinsic
11998 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12005 declare void @llvm.donothing() nounwind readnone
12010 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
12011 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
12012 with an invoke instruction.
12022 This intrinsic does nothing, and it's removed by optimizers and ignored
12025 Stack Map Intrinsics
12026 --------------------
12028 LLVM provides experimental intrinsics to support runtime patching
12029 mechanisms commonly desired in dynamic language JITs. These intrinsics
12030 are described in :doc:`StackMaps`.