1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][-a-zA-Z$._][-a-zA-Z$._0-9]*``'. Identifiers that require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves. The ``"\01"`` prefix
83 can be used on global variables to suppress mangling.
84 #. Unnamed values are represented as an unsigned numeric value with
85 their prefix. For example, ``%12``, ``@2``, ``%44``.
86 #. Constants, which are described in the section Constants_ below.
88 LLVM requires that values start with a prefix for two reasons: Compilers
89 don't need to worry about name clashes with reserved words, and the set
90 of reserved words may be expanded in the future without penalty.
91 Additionally, unnamed identifiers allow a compiler to quickly come up
92 with a temporary variable without having to avoid symbol table
95 Reserved words in LLVM are very similar to reserved words in other
96 languages. There are keywords for different opcodes ('``add``',
97 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
98 '``i32``', etc...), and others. These reserved words cannot conflict
99 with variable names, because none of them start with a prefix character
100 (``'%'`` or ``'@'``).
102 Here is an example of LLVM code to multiply the integer variable
109 %result = mul i32 %X, 8
111 After strength reduction:
115 %result = shl i32 %X, 3
121 %0 = add i32 %X, %X ; yields i32:%0
122 %1 = add i32 %0, %0 ; yields i32:%1
123 %result = add i32 %1, %1
125 This last way of multiplying ``%X`` by 8 illustrates several important
126 lexical features of LLVM:
128 #. Comments are delimited with a '``;``' and go until the end of line.
129 #. Unnamed temporaries are created when the result of a computation is
130 not assigned to a named value.
131 #. Unnamed temporaries are numbered sequentially (using a per-function
132 incrementing counter, starting with 0). Note that basic blocks and unnamed
133 function parameters are included in this numbering. For example, if the
134 entry basic block is not given a label name and all function parameters are
135 named, then it will get number 0.
137 It also shows a convention that we follow in this document. When
138 demonstrating instructions, we will follow an instruction with a comment
139 that defines the type and name of value produced.
147 LLVM programs are composed of ``Module``'s, each of which is a
148 translation unit of the input programs. Each module consists of
149 functions, global variables, and symbol table entries. Modules may be
150 combined together with the LLVM linker, which merges function (and
151 global variable) definitions, resolves forward declarations, and merges
152 symbol table entries. Here is an example of the "hello world" module:
156 ; Declare the string constant as a global constant.
157 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
159 ; External declaration of the puts function
160 declare i32 @puts(i8* nocapture) nounwind
162 ; Definition of main function
163 define i32 @main() { ; i32()*
164 ; Convert [13 x i8]* to i8 *...
165 %cast210 = getelementptr [13 x i8], [13 x i8]* @.str, i64 0, i64 0
167 ; Call puts function to write out the string to stdout.
168 call i32 @puts(i8* %cast210)
173 !0 = !{i32 42, null, !"string"}
176 This example is made up of a :ref:`global variable <globalvars>` named
177 "``.str``", an external declaration of the "``puts``" function, a
178 :ref:`function definition <functionstructure>` for "``main``" and
179 :ref:`named metadata <namedmetadatastructure>` "``foo``".
181 In general, a module is made up of a list of global values (where both
182 functions and global variables are global values). Global values are
183 represented by a pointer to a memory location (in this case, a pointer
184 to an array of char, and a pointer to a function), and have one of the
185 following :ref:`linkage types <linkage>`.
192 All Global Variables and Functions have one of the following types of
196 Global values with "``private``" linkage are only directly
197 accessible by objects in the current module. In particular, linking
198 code into a module with an private global value may cause the
199 private to be renamed as necessary to avoid collisions. Because the
200 symbol is private to the module, all references can be updated. This
201 doesn't show up in any symbol table in the object file.
203 Similar to private, but the value shows as a local symbol
204 (``STB_LOCAL`` in the case of ELF) in the object file. This
205 corresponds to the notion of the '``static``' keyword in C.
206 ``available_externally``
207 Globals with "``available_externally``" linkage are never emitted
208 into the object file corresponding to the LLVM module. They exist to
209 allow inlining and other optimizations to take place given knowledge
210 of the definition of the global, which is known to be somewhere
211 outside the module. Globals with ``available_externally`` linkage
212 are allowed to be discarded at will, and are otherwise the same as
213 ``linkonce_odr``. This linkage type is only allowed on definitions,
216 Globals with "``linkonce``" linkage are merged with other globals of
217 the same name when linkage occurs. This can be used to implement
218 some forms of inline functions, templates, or other code which must
219 be generated in each translation unit that uses it, but where the
220 body may be overridden with a more definitive definition later.
221 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
222 that ``linkonce`` linkage does not actually allow the optimizer to
223 inline the body of this function into callers because it doesn't
224 know if this definition of the function is the definitive definition
225 within the program or whether it will be overridden by a stronger
226 definition. To enable inlining and other optimizations, use
227 "``linkonce_odr``" linkage.
229 "``weak``" linkage has the same merging semantics as ``linkonce``
230 linkage, except that unreferenced globals with ``weak`` linkage may
231 not be discarded. This is used for globals that are declared "weak"
234 "``common``" linkage is most similar to "``weak``" linkage, but they
235 are used for tentative definitions in C, such as "``int X;``" at
236 global scope. Symbols with "``common``" linkage are merged in the
237 same way as ``weak symbols``, and they may not be deleted if
238 unreferenced. ``common`` symbols may not have an explicit section,
239 must have a zero initializer, and may not be marked
240 ':ref:`constant <globalvars>`'. Functions and aliases may not have
243 .. _linkage_appending:
246 "``appending``" linkage may only be applied to global variables of
247 pointer to array type. When two global variables with appending
248 linkage are linked together, the two global arrays are appended
249 together. This is the LLVM, typesafe, equivalent of having the
250 system linker append together "sections" with identical names when
253 The semantics of this linkage follow the ELF object file model: the
254 symbol is weak until linked, if not linked, the symbol becomes null
255 instead of being an undefined reference.
256 ``linkonce_odr``, ``weak_odr``
257 Some languages allow differing globals to be merged, such as two
258 functions with different semantics. Other languages, such as
259 ``C++``, ensure that only equivalent globals are ever merged (the
260 "one definition rule" --- "ODR"). Such languages can use the
261 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
262 global will only be merged with equivalent globals. These linkage
263 types are otherwise the same as their non-``odr`` versions.
265 If none of the above identifiers are used, the global is externally
266 visible, meaning that it participates in linkage and can be used to
267 resolve external symbol references.
269 It is illegal for a function *declaration* to have any linkage type
270 other than ``external`` or ``extern_weak``.
277 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
278 :ref:`invokes <i_invoke>` can all have an optional calling convention
279 specified for the call. The calling convention of any pair of dynamic
280 caller/callee must match, or the behavior of the program is undefined.
281 The following calling conventions are supported by LLVM, and more may be
284 "``ccc``" - The C calling convention
285 This calling convention (the default if no other calling convention
286 is specified) matches the target C calling conventions. This calling
287 convention supports varargs function calls and tolerates some
288 mismatch in the declared prototype and implemented declaration of
289 the function (as does normal C).
290 "``fastcc``" - The fast calling convention
291 This calling convention attempts to make calls as fast as possible
292 (e.g. by passing things in registers). This calling convention
293 allows the target to use whatever tricks it wants to produce fast
294 code for the target, without having to conform to an externally
295 specified ABI (Application Binary Interface). `Tail calls can only
296 be optimized when this, the GHC or the HiPE convention is
297 used. <CodeGenerator.html#id80>`_ This calling convention does not
298 support varargs and requires the prototype of all callees to exactly
299 match the prototype of the function definition.
300 "``coldcc``" - The cold calling convention
301 This calling convention attempts to make code in the caller as
302 efficient as possible under the assumption that the call is not
303 commonly executed. As such, these calls often preserve all registers
304 so that the call does not break any live ranges in the caller side.
305 This calling convention does not support varargs and requires the
306 prototype of all callees to exactly match the prototype of the
307 function definition. Furthermore the inliner doesn't consider such function
309 "``cc 10``" - GHC convention
310 This calling convention has been implemented specifically for use by
311 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
312 It passes everything in registers, going to extremes to achieve this
313 by disabling callee save registers. This calling convention should
314 not be used lightly but only for specific situations such as an
315 alternative to the *register pinning* performance technique often
316 used when implementing functional programming languages. At the
317 moment only X86 supports this convention and it has the following
320 - On *X86-32* only supports up to 4 bit type parameters. No
321 floating point types are supported.
322 - On *X86-64* only supports up to 10 bit type parameters and 6
323 floating point parameters.
325 This calling convention supports `tail call
326 optimization <CodeGenerator.html#id80>`_ but requires both the
327 caller and callee are using it.
328 "``cc 11``" - The HiPE calling convention
329 This calling convention has been implemented specifically for use by
330 the `High-Performance Erlang
331 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
332 native code compiler of the `Ericsson's Open Source Erlang/OTP
333 system <http://www.erlang.org/download.shtml>`_. It uses more
334 registers for argument passing than the ordinary C calling
335 convention and defines no callee-saved registers. The calling
336 convention properly supports `tail call
337 optimization <CodeGenerator.html#id80>`_ but requires that both the
338 caller and the callee use it. It uses a *register pinning*
339 mechanism, similar to GHC's convention, for keeping frequently
340 accessed runtime components pinned to specific hardware registers.
341 At the moment only X86 supports this convention (both 32 and 64
343 "``webkit_jscc``" - WebKit's JavaScript calling convention
344 This calling convention has been implemented for `WebKit FTL JIT
345 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
346 stack right to left (as cdecl does), and returns a value in the
347 platform's customary return register.
348 "``anyregcc``" - Dynamic calling convention for code patching
349 This is a special convention that supports patching an arbitrary code
350 sequence in place of a call site. This convention forces the call
351 arguments into registers but allows them to be dynamically
352 allocated. This can currently only be used with calls to
353 llvm.experimental.patchpoint because only this intrinsic records
354 the location of its arguments in a side table. See :doc:`StackMaps`.
355 "``preserve_mostcc``" - The `PreserveMost` calling convention
356 This calling convention attempts to make the code in the caller as
357 unintrusive as possible. This convention behaves identically to the `C`
358 calling convention on how arguments and return values are passed, but it
359 uses a different set of caller/callee-saved registers. This alleviates the
360 burden of saving and recovering a large register set before and after the
361 call in the caller. If the arguments are passed in callee-saved registers,
362 then they will be preserved by the callee across the call. This doesn't
363 apply for values returned in callee-saved registers.
365 - On X86-64 the callee preserves all general purpose registers, except for
366 R11. R11 can be used as a scratch register. Floating-point registers
367 (XMMs/YMMs) are not preserved and need to be saved by the caller.
369 The idea behind this convention is to support calls to runtime functions
370 that have a hot path and a cold path. The hot path is usually a small piece
371 of code that doesn't use many registers. The cold path might need to call out to
372 another function and therefore only needs to preserve the caller-saved
373 registers, which haven't already been saved by the caller. The
374 `PreserveMost` calling convention is very similar to the `cold` calling
375 convention in terms of caller/callee-saved registers, but they are used for
376 different types of function calls. `coldcc` is for function calls that are
377 rarely executed, whereas `preserve_mostcc` function calls are intended to be
378 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
379 doesn't prevent the inliner from inlining the function call.
381 This calling convention will be used by a future version of the ObjectiveC
382 runtime and should therefore still be considered experimental at this time.
383 Although this convention was created to optimize certain runtime calls to
384 the ObjectiveC runtime, it is not limited to this runtime and might be used
385 by other runtimes in the future too. The current implementation only
386 supports X86-64, but the intention is to support more architectures in the
388 "``preserve_allcc``" - The `PreserveAll` calling convention
389 This calling convention attempts to make the code in the caller even less
390 intrusive than the `PreserveMost` calling convention. This calling
391 convention also behaves identical to the `C` calling convention on how
392 arguments and return values are passed, but it uses a different set of
393 caller/callee-saved registers. This removes the burden of saving and
394 recovering a large register set before and after the call in the caller. If
395 the arguments are passed in callee-saved registers, then they will be
396 preserved by the callee across the call. This doesn't apply for values
397 returned in callee-saved registers.
399 - On X86-64 the callee preserves all general purpose registers, except for
400 R11. R11 can be used as a scratch register. Furthermore it also preserves
401 all floating-point registers (XMMs/YMMs).
403 The idea behind this convention is to support calls to runtime functions
404 that don't need to call out to any other functions.
406 This calling convention, like the `PreserveMost` calling convention, will be
407 used by a future version of the ObjectiveC runtime and should be considered
408 experimental at this time.
409 "``cc <n>``" - Numbered convention
410 Any calling convention may be specified by number, allowing
411 target-specific calling conventions to be used. Target specific
412 calling conventions start at 64.
414 More calling conventions can be added/defined on an as-needed basis, to
415 support Pascal conventions or any other well-known target-independent
418 .. _visibilitystyles:
423 All Global Variables and Functions have one of the following visibility
426 "``default``" - Default style
427 On targets that use the ELF object file format, default visibility
428 means that the declaration is visible to other modules and, in
429 shared libraries, means that the declared entity may be overridden.
430 On Darwin, default visibility means that the declaration is visible
431 to other modules. Default visibility corresponds to "external
432 linkage" in the language.
433 "``hidden``" - Hidden style
434 Two declarations of an object with hidden visibility refer to the
435 same object if they are in the same shared object. Usually, hidden
436 visibility indicates that the symbol will not be placed into the
437 dynamic symbol table, so no other module (executable or shared
438 library) can reference it directly.
439 "``protected``" - Protected style
440 On ELF, protected visibility indicates that the symbol will be
441 placed in the dynamic symbol table, but that references within the
442 defining module will bind to the local symbol. That is, the symbol
443 cannot be overridden by another module.
445 A symbol with ``internal`` or ``private`` linkage must have ``default``
453 All Global Variables, Functions and Aliases can have one of the following
457 "``dllimport``" causes the compiler to reference a function or variable via
458 a global pointer to a pointer that is set up by the DLL exporting the
459 symbol. On Microsoft Windows targets, the pointer name is formed by
460 combining ``__imp_`` and the function or variable name.
462 "``dllexport``" causes the compiler to provide a global pointer to a pointer
463 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
464 Microsoft Windows targets, the pointer name is formed by combining
465 ``__imp_`` and the function or variable name. Since this storage class
466 exists for defining a dll interface, the compiler, assembler and linker know
467 it is externally referenced and must refrain from deleting the symbol.
471 Thread Local Storage Models
472 ---------------------------
474 A variable may be defined as ``thread_local``, which means that it will
475 not be shared by threads (each thread will have a separated copy of the
476 variable). Not all targets support thread-local variables. Optionally, a
477 TLS model may be specified:
480 For variables that are only used within the current shared library.
482 For variables in modules that will not be loaded dynamically.
484 For variables defined in the executable and only used within it.
486 If no explicit model is given, the "general dynamic" model is used.
488 The models correspond to the ELF TLS models; see `ELF Handling For
489 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
490 more information on under which circumstances the different models may
491 be used. The target may choose a different TLS model if the specified
492 model is not supported, or if a better choice of model can be made.
494 A model can also be specified in a alias, but then it only governs how
495 the alias is accessed. It will not have any effect in the aliasee.
502 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
503 types <t_struct>`. Literal types are uniqued structurally, but identified types
504 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
505 to forward declare a type that is not yet available.
507 An example of a identified structure specification is:
511 %mytype = type { %mytype*, i32 }
513 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
514 literal types are uniqued in recent versions of LLVM.
521 Global variables define regions of memory allocated at compilation time
524 Global variable definitions must be initialized.
526 Global variables in other translation units can also be declared, in which
527 case they don't have an initializer.
529 Either global variable definitions or declarations may have an explicit section
530 to be placed in and may have an optional explicit alignment specified.
532 A variable may be defined as a global ``constant``, which indicates that
533 the contents of the variable will **never** be modified (enabling better
534 optimization, allowing the global data to be placed in the read-only
535 section of an executable, etc). Note that variables that need runtime
536 initialization cannot be marked ``constant`` as there is a store to the
539 LLVM explicitly allows *declarations* of global variables to be marked
540 constant, even if the final definition of the global is not. This
541 capability can be used to enable slightly better optimization of the
542 program, but requires the language definition to guarantee that
543 optimizations based on the 'constantness' are valid for the translation
544 units that do not include the definition.
546 As SSA values, global variables define pointer values that are in scope
547 (i.e. they dominate) all basic blocks in the program. Global variables
548 always define a pointer to their "content" type because they describe a
549 region of memory, and all memory objects in LLVM are accessed through
552 Global variables can be marked with ``unnamed_addr`` which indicates
553 that the address is not significant, only the content. Constants marked
554 like this can be merged with other constants if they have the same
555 initializer. Note that a constant with significant address *can* be
556 merged with a ``unnamed_addr`` constant, the result being a constant
557 whose address is significant.
559 A global variable may be declared to reside in a target-specific
560 numbered address space. For targets that support them, address spaces
561 may affect how optimizations are performed and/or what target
562 instructions are used to access the variable. The default address space
563 is zero. The address space qualifier must precede any other attributes.
565 LLVM allows an explicit section to be specified for globals. If the
566 target supports it, it will emit globals to the section specified.
567 Additionally, the global can placed in a comdat if the target has the necessary
570 By default, global initializers are optimized by assuming that global
571 variables defined within the module are not modified from their
572 initial values before the start of the global initializer. This is
573 true even for variables potentially accessible from outside the
574 module, including those with external linkage or appearing in
575 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
576 by marking the variable with ``externally_initialized``.
578 An explicit alignment may be specified for a global, which must be a
579 power of 2. If not present, or if the alignment is set to zero, the
580 alignment of the global is set by the target to whatever it feels
581 convenient. If an explicit alignment is specified, the global is forced
582 to have exactly that alignment. Targets and optimizers are not allowed
583 to over-align the global if the global has an assigned section. In this
584 case, the extra alignment could be observable: for example, code could
585 assume that the globals are densely packed in their section and try to
586 iterate over them as an array, alignment padding would break this
587 iteration. The maximum alignment is ``1 << 29``.
589 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
591 Variables and aliases can have a
592 :ref:`Thread Local Storage Model <tls_model>`.
596 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
597 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
598 <global | constant> <Type> [<InitializerConstant>]
599 [, section "name"] [, comdat [($name)]]
600 [, align <Alignment>]
602 For example, the following defines a global in a numbered address space
603 with an initializer, section, and alignment:
607 @G = addrspace(5) constant float 1.0, section "foo", align 4
609 The following example just declares a global variable
613 @G = external global i32
615 The following example defines a thread-local global with the
616 ``initialexec`` TLS model:
620 @G = thread_local(initialexec) global i32 0, align 4
622 .. _functionstructure:
627 LLVM function definitions consist of the "``define``" keyword, an
628 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
629 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
630 an optional :ref:`calling convention <callingconv>`,
631 an optional ``unnamed_addr`` attribute, a return type, an optional
632 :ref:`parameter attribute <paramattrs>` for the return type, a function
633 name, a (possibly empty) argument list (each with optional :ref:`parameter
634 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
635 an optional section, an optional alignment,
636 an optional :ref:`comdat <langref_comdats>`,
637 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
638 an optional :ref:`prologue <prologuedata>`, an opening
639 curly brace, a list of basic blocks, and a closing curly brace.
641 LLVM function declarations consist of the "``declare``" keyword, an
642 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
643 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
644 an optional :ref:`calling convention <callingconv>`,
645 an optional ``unnamed_addr`` attribute, a return type, an optional
646 :ref:`parameter attribute <paramattrs>` for the return type, a function
647 name, a possibly empty list of arguments, an optional alignment, an optional
648 :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
649 and an optional :ref:`prologue <prologuedata>`.
651 A function definition contains a list of basic blocks, forming the CFG (Control
652 Flow Graph) for the function. Each basic block may optionally start with a label
653 (giving the basic block a symbol table entry), contains a list of instructions,
654 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
655 function return). If an explicit label is not provided, a block is assigned an
656 implicit numbered label, using the next value from the same counter as used for
657 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
658 entry block does not have an explicit label, it will be assigned label "%0",
659 then the first unnamed temporary in that block will be "%1", etc.
661 The first basic block in a function is special in two ways: it is
662 immediately executed on entrance to the function, and it is not allowed
663 to have predecessor basic blocks (i.e. there can not be any branches to
664 the entry block of a function). Because the block can have no
665 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
667 LLVM allows an explicit section to be specified for functions. If the
668 target supports it, it will emit functions to the section specified.
669 Additionally, the function can be placed in a COMDAT.
671 An explicit alignment may be specified for a function. If not present,
672 or if the alignment is set to zero, the alignment of the function is set
673 by the target to whatever it feels convenient. If an explicit alignment
674 is specified, the function is forced to have at least that much
675 alignment. All alignments must be a power of 2.
677 If the ``unnamed_addr`` attribute is given, the address is known to not
678 be significant and two identical functions can be merged.
682 define [linkage] [visibility] [DLLStorageClass]
684 <ResultType> @<FunctionName> ([argument list])
685 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
686 [align N] [gc] [prefix Constant] [prologue Constant] { ... }
688 The argument list is a comma seperated sequence of arguments where each
689 argument is of the following form
693 <type> [parameter Attrs] [name]
701 Aliases, unlike function or variables, don't create any new data. They
702 are just a new symbol and metadata for an existing position.
704 Aliases have a name and an aliasee that is either a global value or a
707 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
708 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
709 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
713 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy> @<Aliasee>
715 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
716 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
717 might not correctly handle dropping a weak symbol that is aliased.
719 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
720 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
723 Since aliases are only a second name, some restrictions apply, of which
724 some can only be checked when producing an object file:
726 * The expression defining the aliasee must be computable at assembly
727 time. Since it is just a name, no relocations can be used.
729 * No alias in the expression can be weak as the possibility of the
730 intermediate alias being overridden cannot be represented in an
733 * No global value in the expression can be a declaration, since that
734 would require a relocation, which is not possible.
741 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
743 Comdats have a name which represents the COMDAT key. All global objects that
744 specify this key will only end up in the final object file if the linker chooses
745 that key over some other key. Aliases are placed in the same COMDAT that their
746 aliasee computes to, if any.
748 Comdats have a selection kind to provide input on how the linker should
749 choose between keys in two different object files.
753 $<Name> = comdat SelectionKind
755 The selection kind must be one of the following:
758 The linker may choose any COMDAT key, the choice is arbitrary.
760 The linker may choose any COMDAT key but the sections must contain the
763 The linker will choose the section containing the largest COMDAT key.
765 The linker requires that only section with this COMDAT key exist.
767 The linker may choose any COMDAT key but the sections must contain the
770 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
771 ``any`` as a selection kind.
773 Here is an example of a COMDAT group where a function will only be selected if
774 the COMDAT key's section is the largest:
778 $foo = comdat largest
779 @foo = global i32 2, comdat($foo)
781 define void @bar() comdat($foo) {
785 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
791 @foo = global i32 2, comdat
794 In a COFF object file, this will create a COMDAT section with selection kind
795 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
796 and another COMDAT section with selection kind
797 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
798 section and contains the contents of the ``@bar`` symbol.
800 There are some restrictions on the properties of the global object.
801 It, or an alias to it, must have the same name as the COMDAT group when
803 The contents and size of this object may be used during link-time to determine
804 which COMDAT groups get selected depending on the selection kind.
805 Because the name of the object must match the name of the COMDAT group, the
806 linkage of the global object must not be local; local symbols can get renamed
807 if a collision occurs in the symbol table.
809 The combined use of COMDATS and section attributes may yield surprising results.
816 @g1 = global i32 42, section "sec", comdat($foo)
817 @g2 = global i32 42, section "sec", comdat($bar)
819 From the object file perspective, this requires the creation of two sections
820 with the same name. This is necessary because both globals belong to different
821 COMDAT groups and COMDATs, at the object file level, are represented by
824 Note that certain IR constructs like global variables and functions may create
825 COMDATs in the object file in addition to any which are specified using COMDAT
826 IR. This arises, for example, when a global variable has linkonce_odr linkage.
828 .. _namedmetadatastructure:
833 Named metadata is a collection of metadata. :ref:`Metadata
834 nodes <metadata>` (but not metadata strings) are the only valid
835 operands for a named metadata.
839 ; Some unnamed metadata nodes, which are referenced by the named metadata.
844 !name = !{!0, !1, !2}
851 The return type and each parameter of a function type may have a set of
852 *parameter attributes* associated with them. Parameter attributes are
853 used to communicate additional information about the result or
854 parameters of a function. Parameter attributes are considered to be part
855 of the function, not of the function type, so functions with different
856 parameter attributes can have the same function type.
858 Parameter attributes are simple keywords that follow the type specified.
859 If multiple parameter attributes are needed, they are space separated.
864 declare i32 @printf(i8* noalias nocapture, ...)
865 declare i32 @atoi(i8 zeroext)
866 declare signext i8 @returns_signed_char()
868 Note that any attributes for the function result (``nounwind``,
869 ``readonly``) come immediately after the argument list.
871 Currently, only the following parameter attributes are defined:
874 This indicates to the code generator that the parameter or return
875 value should be zero-extended to the extent required by the target's
876 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
877 the caller (for a parameter) or the callee (for a return value).
879 This indicates to the code generator that the parameter or return
880 value should be sign-extended to the extent required by the target's
881 ABI (which is usually 32-bits) by the caller (for a parameter) or
882 the callee (for a return value).
884 This indicates that this parameter or return value should be treated
885 in a special target-dependent fashion during while emitting code for
886 a function call or return (usually, by putting it in a register as
887 opposed to memory, though some targets use it to distinguish between
888 two different kinds of registers). Use of this attribute is
891 This indicates that the pointer parameter should really be passed by
892 value to the function. The attribute implies that a hidden copy of
893 the pointee is made between the caller and the callee, so the callee
894 is unable to modify the value in the caller. This attribute is only
895 valid on LLVM pointer arguments. It is generally used to pass
896 structs and arrays by value, but is also valid on pointers to
897 scalars. The copy is considered to belong to the caller not the
898 callee (for example, ``readonly`` functions should not write to
899 ``byval`` parameters). This is not a valid attribute for return
902 The byval attribute also supports specifying an alignment with the
903 align attribute. It indicates the alignment of the stack slot to
904 form and the known alignment of the pointer specified to the call
905 site. If the alignment is not specified, then the code generator
906 makes a target-specific assumption.
912 The ``inalloca`` argument attribute allows the caller to take the
913 address of outgoing stack arguments. An ``inalloca`` argument must
914 be a pointer to stack memory produced by an ``alloca`` instruction.
915 The alloca, or argument allocation, must also be tagged with the
916 inalloca keyword. Only the last argument may have the ``inalloca``
917 attribute, and that argument is guaranteed to be passed in memory.
919 An argument allocation may be used by a call at most once because
920 the call may deallocate it. The ``inalloca`` attribute cannot be
921 used in conjunction with other attributes that affect argument
922 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
923 ``inalloca`` attribute also disables LLVM's implicit lowering of
924 large aggregate return values, which means that frontend authors
925 must lower them with ``sret`` pointers.
927 When the call site is reached, the argument allocation must have
928 been the most recent stack allocation that is still live, or the
929 results are undefined. It is possible to allocate additional stack
930 space after an argument allocation and before its call site, but it
931 must be cleared off with :ref:`llvm.stackrestore
934 See :doc:`InAlloca` for more information on how to use this
938 This indicates that the pointer parameter specifies the address of a
939 structure that is the return value of the function in the source
940 program. This pointer must be guaranteed by the caller to be valid:
941 loads and stores to the structure may be assumed by the callee
942 not to trap and to be properly aligned. This may only be applied to
943 the first parameter. This is not a valid attribute for return
947 This indicates that the pointer value may be assumed by the optimizer to
948 have the specified alignment.
950 Note that this attribute has additional semantics when combined with the
956 This indicates that objects accessed via pointer values
957 :ref:`based <pointeraliasing>` on the argument or return value are not also
958 accessed, during the execution of the function, via pointer values not
959 *based* on the argument or return value. The attribute on a return value
960 also has additional semantics described below. The caller shares the
961 responsibility with the callee for ensuring that these requirements are met.
962 For further details, please see the discussion of the NoAlias response in
963 :ref:`alias analysis <Must, May, or No>`.
965 Note that this definition of ``noalias`` is intentionally similar
966 to the definition of ``restrict`` in C99 for function arguments.
968 For function return values, C99's ``restrict`` is not meaningful,
969 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
970 attribute on return values are stronger than the semantics of the attribute
971 when used on function arguments. On function return values, the ``noalias``
972 attribute indicates that the function acts like a system memory allocation
973 function, returning a pointer to allocated storage disjoint from the
974 storage for any other object accessible to the caller.
977 This indicates that the callee does not make any copies of the
978 pointer that outlive the callee itself. This is not a valid
979 attribute for return values.
984 This indicates that the pointer parameter can be excised using the
985 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
986 attribute for return values and can only be applied to one parameter.
989 This indicates that the function always returns the argument as its return
990 value. This is an optimization hint to the code generator when generating
991 the caller, allowing tail call optimization and omission of register saves
992 and restores in some cases; it is not checked or enforced when generating
993 the callee. The parameter and the function return type must be valid
994 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
995 valid attribute for return values and can only be applied to one parameter.
998 This indicates that the parameter or return pointer is not null. This
999 attribute may only be applied to pointer typed parameters. This is not
1000 checked or enforced by LLVM, the caller must ensure that the pointer
1001 passed in is non-null, or the callee must ensure that the returned pointer
1004 ``dereferenceable(<n>)``
1005 This indicates that the parameter or return pointer is dereferenceable. This
1006 attribute may only be applied to pointer typed parameters. A pointer that
1007 is dereferenceable can be loaded from speculatively without a risk of
1008 trapping. The number of bytes known to be dereferenceable must be provided
1009 in parentheses. It is legal for the number of bytes to be less than the
1010 size of the pointee type. The ``nonnull`` attribute does not imply
1011 dereferenceability (consider a pointer to one element past the end of an
1012 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1013 ``addrspace(0)`` (which is the default address space).
1017 Garbage Collector Strategy Names
1018 --------------------------------
1020 Each function may specify a garbage collector strategy name, which is simply a
1023 .. code-block:: llvm
1025 define void @f() gc "name" { ... }
1027 The supported values of *name* includes those :ref:`built in to LLVM
1028 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1029 strategy will cause the compiler to alter its output in order to support the
1030 named garbage collection algorithm. Note that LLVM itself does not contain a
1031 garbage collector, this functionality is restricted to generating machine code
1032 which can interoperate with a collector provided externally.
1039 Prefix data is data associated with a function which the code
1040 generator will emit immediately before the function's entrypoint.
1041 The purpose of this feature is to allow frontends to associate
1042 language-specific runtime metadata with specific functions and make it
1043 available through the function pointer while still allowing the
1044 function pointer to be called.
1046 To access the data for a given function, a program may bitcast the
1047 function pointer to a pointer to the constant's type and dereference
1048 index -1. This implies that the IR symbol points just past the end of
1049 the prefix data. For instance, take the example of a function annotated
1050 with a single ``i32``,
1052 .. code-block:: llvm
1054 define void @f() prefix i32 123 { ... }
1056 The prefix data can be referenced as,
1058 .. code-block:: llvm
1060 %0 = bitcast void* () @f to i32*
1061 %a = getelementptr inbounds i32, i32* %0, i32 -1
1062 %b = load i32, i32* %a
1064 Prefix data is laid out as if it were an initializer for a global variable
1065 of the prefix data's type. The function will be placed such that the
1066 beginning of the prefix data is aligned. This means that if the size
1067 of the prefix data is not a multiple of the alignment size, the
1068 function's entrypoint will not be aligned. If alignment of the
1069 function's entrypoint is desired, padding must be added to the prefix
1072 A function may have prefix data but no body. This has similar semantics
1073 to the ``available_externally`` linkage in that the data may be used by the
1074 optimizers but will not be emitted in the object file.
1081 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1082 be inserted prior to the function body. This can be used for enabling
1083 function hot-patching and instrumentation.
1085 To maintain the semantics of ordinary function calls, the prologue data must
1086 have a particular format. Specifically, it must begin with a sequence of
1087 bytes which decode to a sequence of machine instructions, valid for the
1088 module's target, which transfer control to the point immediately succeeding
1089 the prologue data, without performing any other visible action. This allows
1090 the inliner and other passes to reason about the semantics of the function
1091 definition without needing to reason about the prologue data. Obviously this
1092 makes the format of the prologue data highly target dependent.
1094 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1095 which encodes the ``nop`` instruction:
1097 .. code-block:: llvm
1099 define void @f() prologue i8 144 { ... }
1101 Generally prologue data can be formed by encoding a relative branch instruction
1102 which skips the metadata, as in this example of valid prologue data for the
1103 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1105 .. code-block:: llvm
1107 %0 = type <{ i8, i8, i8* }>
1109 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1111 A function may have prologue data but no body. This has similar semantics
1112 to the ``available_externally`` linkage in that the data may be used by the
1113 optimizers but will not be emitted in the object file.
1120 Attribute groups are groups of attributes that are referenced by objects within
1121 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1122 functions will use the same set of attributes. In the degenerative case of a
1123 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1124 group will capture the important command line flags used to build that file.
1126 An attribute group is a module-level object. To use an attribute group, an
1127 object references the attribute group's ID (e.g. ``#37``). An object may refer
1128 to more than one attribute group. In that situation, the attributes from the
1129 different groups are merged.
1131 Here is an example of attribute groups for a function that should always be
1132 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1134 .. code-block:: llvm
1136 ; Target-independent attributes:
1137 attributes #0 = { alwaysinline alignstack=4 }
1139 ; Target-dependent attributes:
1140 attributes #1 = { "no-sse" }
1142 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1143 define void @f() #0 #1 { ... }
1150 Function attributes are set to communicate additional information about
1151 a function. Function attributes are considered to be part of the
1152 function, not of the function type, so functions with different function
1153 attributes can have the same function type.
1155 Function attributes are simple keywords that follow the type specified.
1156 If multiple attributes are needed, they are space separated. For
1159 .. code-block:: llvm
1161 define void @f() noinline { ... }
1162 define void @f() alwaysinline { ... }
1163 define void @f() alwaysinline optsize { ... }
1164 define void @f() optsize { ... }
1167 This attribute indicates that, when emitting the prologue and
1168 epilogue, the backend should forcibly align the stack pointer.
1169 Specify the desired alignment, which must be a power of two, in
1172 This attribute indicates that the inliner should attempt to inline
1173 this function into callers whenever possible, ignoring any active
1174 inlining size threshold for this caller.
1176 This indicates that the callee function at a call site should be
1177 recognized as a built-in function, even though the function's declaration
1178 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1179 direct calls to functions that are declared with the ``nobuiltin``
1182 This attribute indicates that this function is rarely called. When
1183 computing edge weights, basic blocks post-dominated by a cold
1184 function call are also considered to be cold; and, thus, given low
1187 This attribute indicates that the source code contained a hint that
1188 inlining this function is desirable (such as the "inline" keyword in
1189 C/C++). It is just a hint; it imposes no requirements on the
1192 This attribute indicates that the function should be added to a
1193 jump-instruction table at code-generation time, and that all address-taken
1194 references to this function should be replaced with a reference to the
1195 appropriate jump-instruction-table function pointer. Note that this creates
1196 a new pointer for the original function, which means that code that depends
1197 on function-pointer identity can break. So, any function annotated with
1198 ``jumptable`` must also be ``unnamed_addr``.
1200 This attribute suggests that optimization passes and code generator
1201 passes make choices that keep the code size of this function as small
1202 as possible and perform optimizations that may sacrifice runtime
1203 performance in order to minimize the size of the generated code.
1205 This attribute disables prologue / epilogue emission for the
1206 function. This can have very system-specific consequences.
1208 This indicates that the callee function at a call site is not recognized as
1209 a built-in function. LLVM will retain the original call and not replace it
1210 with equivalent code based on the semantics of the built-in function, unless
1211 the call site uses the ``builtin`` attribute. This is valid at call sites
1212 and on function declarations and definitions.
1214 This attribute indicates that calls to the function cannot be
1215 duplicated. A call to a ``noduplicate`` function may be moved
1216 within its parent function, but may not be duplicated within
1217 its parent function.
1219 A function containing a ``noduplicate`` call may still
1220 be an inlining candidate, provided that the call is not
1221 duplicated by inlining. That implies that the function has
1222 internal linkage and only has one call site, so the original
1223 call is dead after inlining.
1225 This attributes disables implicit floating point instructions.
1227 This attribute indicates that the inliner should never inline this
1228 function in any situation. This attribute may not be used together
1229 with the ``alwaysinline`` attribute.
1231 This attribute suppresses lazy symbol binding for the function. This
1232 may make calls to the function faster, at the cost of extra program
1233 startup time if the function is not called during program startup.
1235 This attribute indicates that the code generator should not use a
1236 red zone, even if the target-specific ABI normally permits it.
1238 This function attribute indicates that the function never returns
1239 normally. This produces undefined behavior at runtime if the
1240 function ever does dynamically return.
1242 This function attribute indicates that the function never raises an
1243 exception. If the function does raise an exception, its runtime
1244 behavior is undefined. However, functions marked nounwind may still
1245 trap or generate asynchronous exceptions. Exception handling schemes
1246 that are recognized by LLVM to handle asynchronous exceptions, such
1247 as SEH, will still provide their implementation defined semantics.
1249 This function attribute indicates that the function is not optimized
1250 by any optimization or code generator passes with the
1251 exception of interprocedural optimization passes.
1252 This attribute cannot be used together with the ``alwaysinline``
1253 attribute; this attribute is also incompatible
1254 with the ``minsize`` attribute and the ``optsize`` attribute.
1256 This attribute requires the ``noinline`` attribute to be specified on
1257 the function as well, so the function is never inlined into any caller.
1258 Only functions with the ``alwaysinline`` attribute are valid
1259 candidates for inlining into the body of this function.
1261 This attribute suggests that optimization passes and code generator
1262 passes make choices that keep the code size of this function low,
1263 and otherwise do optimizations specifically to reduce code size as
1264 long as they do not significantly impact runtime performance.
1266 On a function, this attribute indicates that the function computes its
1267 result (or decides to unwind an exception) based strictly on its arguments,
1268 without dereferencing any pointer arguments or otherwise accessing
1269 any mutable state (e.g. memory, control registers, etc) visible to
1270 caller functions. It does not write through any pointer arguments
1271 (including ``byval`` arguments) and never changes any state visible
1272 to callers. This means that it cannot unwind exceptions by calling
1273 the ``C++`` exception throwing methods.
1275 On an argument, this attribute indicates that the function does not
1276 dereference that pointer argument, even though it may read or write the
1277 memory that the pointer points to if accessed through other pointers.
1279 On a function, this attribute indicates that the function does not write
1280 through any pointer arguments (including ``byval`` arguments) or otherwise
1281 modify any state (e.g. memory, control registers, etc) visible to
1282 caller functions. It may dereference pointer arguments and read
1283 state that may be set in the caller. A readonly function always
1284 returns the same value (or unwinds an exception identically) when
1285 called with the same set of arguments and global state. It cannot
1286 unwind an exception by calling the ``C++`` exception throwing
1289 On an argument, this attribute indicates that the function does not write
1290 through this pointer argument, even though it may write to the memory that
1291 the pointer points to.
1293 This attribute indicates that this function can return twice. The C
1294 ``setjmp`` is an example of such a function. The compiler disables
1295 some optimizations (like tail calls) in the caller of these
1297 ``sanitize_address``
1298 This attribute indicates that AddressSanitizer checks
1299 (dynamic address safety analysis) are enabled for this function.
1301 This attribute indicates that MemorySanitizer checks (dynamic detection
1302 of accesses to uninitialized memory) are enabled for this function.
1304 This attribute indicates that ThreadSanitizer checks
1305 (dynamic thread safety analysis) are enabled for this function.
1307 This attribute indicates that the function should emit a stack
1308 smashing protector. It is in the form of a "canary" --- a random value
1309 placed on the stack before the local variables that's checked upon
1310 return from the function to see if it has been overwritten. A
1311 heuristic is used to determine if a function needs stack protectors
1312 or not. The heuristic used will enable protectors for functions with:
1314 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1315 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1316 - Calls to alloca() with variable sizes or constant sizes greater than
1317 ``ssp-buffer-size``.
1319 Variables that are identified as requiring a protector will be arranged
1320 on the stack such that they are adjacent to the stack protector guard.
1322 If a function that has an ``ssp`` attribute is inlined into a
1323 function that doesn't have an ``ssp`` attribute, then the resulting
1324 function will have an ``ssp`` attribute.
1326 This attribute indicates that the function should *always* emit a
1327 stack smashing protector. This overrides the ``ssp`` function
1330 Variables that are identified as requiring a protector will be arranged
1331 on the stack such that they are adjacent to the stack protector guard.
1332 The specific layout rules are:
1334 #. Large arrays and structures containing large arrays
1335 (``>= ssp-buffer-size``) are closest to the stack protector.
1336 #. Small arrays and structures containing small arrays
1337 (``< ssp-buffer-size``) are 2nd closest to the protector.
1338 #. Variables that have had their address taken are 3rd closest to the
1341 If a function that has an ``sspreq`` attribute is inlined into a
1342 function that doesn't have an ``sspreq`` attribute or which has an
1343 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1344 an ``sspreq`` attribute.
1346 This attribute indicates that the function should emit a stack smashing
1347 protector. This attribute causes a strong heuristic to be used when
1348 determining if a function needs stack protectors. The strong heuristic
1349 will enable protectors for functions with:
1351 - Arrays of any size and type
1352 - Aggregates containing an array of any size and type.
1353 - Calls to alloca().
1354 - Local variables that have had their address taken.
1356 Variables that are identified as requiring a protector will be arranged
1357 on the stack such that they are adjacent to the stack protector guard.
1358 The specific layout rules are:
1360 #. Large arrays and structures containing large arrays
1361 (``>= ssp-buffer-size``) are closest to the stack protector.
1362 #. Small arrays and structures containing small arrays
1363 (``< ssp-buffer-size``) are 2nd closest to the protector.
1364 #. Variables that have had their address taken are 3rd closest to the
1367 This overrides the ``ssp`` function attribute.
1369 If a function that has an ``sspstrong`` attribute is inlined into a
1370 function that doesn't have an ``sspstrong`` attribute, then the
1371 resulting function will have an ``sspstrong`` attribute.
1373 This attribute indicates that the function will delegate to some other
1374 function with a tail call. The prototype of a thunk should not be used for
1375 optimization purposes. The caller is expected to cast the thunk prototype to
1376 match the thunk target prototype.
1378 This attribute indicates that the ABI being targeted requires that
1379 an unwind table entry be produce for this function even if we can
1380 show that no exceptions passes by it. This is normally the case for
1381 the ELF x86-64 abi, but it can be disabled for some compilation
1386 Module-Level Inline Assembly
1387 ----------------------------
1389 Modules may contain "module-level inline asm" blocks, which corresponds
1390 to the GCC "file scope inline asm" blocks. These blocks are internally
1391 concatenated by LLVM and treated as a single unit, but may be separated
1392 in the ``.ll`` file if desired. The syntax is very simple:
1394 .. code-block:: llvm
1396 module asm "inline asm code goes here"
1397 module asm "more can go here"
1399 The strings can contain any character by escaping non-printable
1400 characters. The escape sequence used is simply "\\xx" where "xx" is the
1401 two digit hex code for the number.
1403 The inline asm code is simply printed to the machine code .s file when
1404 assembly code is generated.
1406 .. _langref_datalayout:
1411 A module may specify a target specific data layout string that specifies
1412 how data is to be laid out in memory. The syntax for the data layout is
1415 .. code-block:: llvm
1417 target datalayout = "layout specification"
1419 The *layout specification* consists of a list of specifications
1420 separated by the minus sign character ('-'). Each specification starts
1421 with a letter and may include other information after the letter to
1422 define some aspect of the data layout. The specifications accepted are
1426 Specifies that the target lays out data in big-endian form. That is,
1427 the bits with the most significance have the lowest address
1430 Specifies that the target lays out data in little-endian form. That
1431 is, the bits with the least significance have the lowest address
1434 Specifies the natural alignment of the stack in bits. Alignment
1435 promotion of stack variables is limited to the natural stack
1436 alignment to avoid dynamic stack realignment. The stack alignment
1437 must be a multiple of 8-bits. If omitted, the natural stack
1438 alignment defaults to "unspecified", which does not prevent any
1439 alignment promotions.
1440 ``p[n]:<size>:<abi>:<pref>``
1441 This specifies the *size* of a pointer and its ``<abi>`` and
1442 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1443 bits. The address space, ``n`` is optional, and if not specified,
1444 denotes the default address space 0. The value of ``n`` must be
1445 in the range [1,2^23).
1446 ``i<size>:<abi>:<pref>``
1447 This specifies the alignment for an integer type of a given bit
1448 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1449 ``v<size>:<abi>:<pref>``
1450 This specifies the alignment for a vector type of a given bit
1452 ``f<size>:<abi>:<pref>``
1453 This specifies the alignment for a floating point type of a given bit
1454 ``<size>``. Only values of ``<size>`` that are supported by the target
1455 will work. 32 (float) and 64 (double) are supported on all targets; 80
1456 or 128 (different flavors of long double) are also supported on some
1459 This specifies the alignment for an object of aggregate type.
1461 If present, specifies that llvm names are mangled in the output. The
1464 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1465 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1466 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1467 symbols get a ``_`` prefix.
1468 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1469 functions also get a suffix based on the frame size.
1470 ``n<size1>:<size2>:<size3>...``
1471 This specifies a set of native integer widths for the target CPU in
1472 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1473 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1474 this set are considered to support most general arithmetic operations
1477 On every specification that takes a ``<abi>:<pref>``, specifying the
1478 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1479 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1481 When constructing the data layout for a given target, LLVM starts with a
1482 default set of specifications which are then (possibly) overridden by
1483 the specifications in the ``datalayout`` keyword. The default
1484 specifications are given in this list:
1486 - ``E`` - big endian
1487 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1488 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1489 same as the default address space.
1490 - ``S0`` - natural stack alignment is unspecified
1491 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1492 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1493 - ``i16:16:16`` - i16 is 16-bit aligned
1494 - ``i32:32:32`` - i32 is 32-bit aligned
1495 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1496 alignment of 64-bits
1497 - ``f16:16:16`` - half is 16-bit aligned
1498 - ``f32:32:32`` - float is 32-bit aligned
1499 - ``f64:64:64`` - double is 64-bit aligned
1500 - ``f128:128:128`` - quad is 128-bit aligned
1501 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1502 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1503 - ``a:0:64`` - aggregates are 64-bit aligned
1505 When LLVM is determining the alignment for a given type, it uses the
1508 #. If the type sought is an exact match for one of the specifications,
1509 that specification is used.
1510 #. If no match is found, and the type sought is an integer type, then
1511 the smallest integer type that is larger than the bitwidth of the
1512 sought type is used. If none of the specifications are larger than
1513 the bitwidth then the largest integer type is used. For example,
1514 given the default specifications above, the i7 type will use the
1515 alignment of i8 (next largest) while both i65 and i256 will use the
1516 alignment of i64 (largest specified).
1517 #. If no match is found, and the type sought is a vector type, then the
1518 largest vector type that is smaller than the sought vector type will
1519 be used as a fall back. This happens because <128 x double> can be
1520 implemented in terms of 64 <2 x double>, for example.
1522 The function of the data layout string may not be what you expect.
1523 Notably, this is not a specification from the frontend of what alignment
1524 the code generator should use.
1526 Instead, if specified, the target data layout is required to match what
1527 the ultimate *code generator* expects. This string is used by the
1528 mid-level optimizers to improve code, and this only works if it matches
1529 what the ultimate code generator uses. There is no way to generate IR
1530 that does not embed this target-specific detail into the IR. If you
1531 don't specify the string, the default specifications will be used to
1532 generate a Data Layout and the optimization phases will operate
1533 accordingly and introduce target specificity into the IR with respect to
1534 these default specifications.
1541 A module may specify a target triple string that describes the target
1542 host. The syntax for the target triple is simply:
1544 .. code-block:: llvm
1546 target triple = "x86_64-apple-macosx10.7.0"
1548 The *target triple* string consists of a series of identifiers delimited
1549 by the minus sign character ('-'). The canonical forms are:
1553 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1554 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1556 This information is passed along to the backend so that it generates
1557 code for the proper architecture. It's possible to override this on the
1558 command line with the ``-mtriple`` command line option.
1560 .. _pointeraliasing:
1562 Pointer Aliasing Rules
1563 ----------------------
1565 Any memory access must be done through a pointer value associated with
1566 an address range of the memory access, otherwise the behavior is
1567 undefined. Pointer values are associated with address ranges according
1568 to the following rules:
1570 - A pointer value is associated with the addresses associated with any
1571 value it is *based* on.
1572 - An address of a global variable is associated with the address range
1573 of the variable's storage.
1574 - The result value of an allocation instruction is associated with the
1575 address range of the allocated storage.
1576 - A null pointer in the default address-space is associated with no
1578 - An integer constant other than zero or a pointer value returned from
1579 a function not defined within LLVM may be associated with address
1580 ranges allocated through mechanisms other than those provided by
1581 LLVM. Such ranges shall not overlap with any ranges of addresses
1582 allocated by mechanisms provided by LLVM.
1584 A pointer value is *based* on another pointer value according to the
1587 - A pointer value formed from a ``getelementptr`` operation is *based*
1588 on the first value operand of the ``getelementptr``.
1589 - The result value of a ``bitcast`` is *based* on the operand of the
1591 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1592 values that contribute (directly or indirectly) to the computation of
1593 the pointer's value.
1594 - The "*based* on" relationship is transitive.
1596 Note that this definition of *"based"* is intentionally similar to the
1597 definition of *"based"* in C99, though it is slightly weaker.
1599 LLVM IR does not associate types with memory. The result type of a
1600 ``load`` merely indicates the size and alignment of the memory from
1601 which to load, as well as the interpretation of the value. The first
1602 operand type of a ``store`` similarly only indicates the size and
1603 alignment of the store.
1605 Consequently, type-based alias analysis, aka TBAA, aka
1606 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1607 :ref:`Metadata <metadata>` may be used to encode additional information
1608 which specialized optimization passes may use to implement type-based
1613 Volatile Memory Accesses
1614 ------------------------
1616 Certain memory accesses, such as :ref:`load <i_load>`'s,
1617 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1618 marked ``volatile``. The optimizers must not change the number of
1619 volatile operations or change their order of execution relative to other
1620 volatile operations. The optimizers *may* change the order of volatile
1621 operations relative to non-volatile operations. This is not Java's
1622 "volatile" and has no cross-thread synchronization behavior.
1624 IR-level volatile loads and stores cannot safely be optimized into
1625 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1626 flagged volatile. Likewise, the backend should never split or merge
1627 target-legal volatile load/store instructions.
1629 .. admonition:: Rationale
1631 Platforms may rely on volatile loads and stores of natively supported
1632 data width to be executed as single instruction. For example, in C
1633 this holds for an l-value of volatile primitive type with native
1634 hardware support, but not necessarily for aggregate types. The
1635 frontend upholds these expectations, which are intentionally
1636 unspecified in the IR. The rules above ensure that IR transformation
1637 do not violate the frontend's contract with the language.
1641 Memory Model for Concurrent Operations
1642 --------------------------------------
1644 The LLVM IR does not define any way to start parallel threads of
1645 execution or to register signal handlers. Nonetheless, there are
1646 platform-specific ways to create them, and we define LLVM IR's behavior
1647 in their presence. This model is inspired by the C++0x memory model.
1649 For a more informal introduction to this model, see the :doc:`Atomics`.
1651 We define a *happens-before* partial order as the least partial order
1654 - Is a superset of single-thread program order, and
1655 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1656 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1657 techniques, like pthread locks, thread creation, thread joining,
1658 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1659 Constraints <ordering>`).
1661 Note that program order does not introduce *happens-before* edges
1662 between a thread and signals executing inside that thread.
1664 Every (defined) read operation (load instructions, memcpy, atomic
1665 loads/read-modify-writes, etc.) R reads a series of bytes written by
1666 (defined) write operations (store instructions, atomic
1667 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1668 section, initialized globals are considered to have a write of the
1669 initializer which is atomic and happens before any other read or write
1670 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1671 may see any write to the same byte, except:
1673 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1674 write\ :sub:`2` happens before R\ :sub:`byte`, then
1675 R\ :sub:`byte` does not see write\ :sub:`1`.
1676 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1677 R\ :sub:`byte` does not see write\ :sub:`3`.
1679 Given that definition, R\ :sub:`byte` is defined as follows:
1681 - If R is volatile, the result is target-dependent. (Volatile is
1682 supposed to give guarantees which can support ``sig_atomic_t`` in
1683 C/C++, and may be used for accesses to addresses that do not behave
1684 like normal memory. It does not generally provide cross-thread
1686 - Otherwise, if there is no write to the same byte that happens before
1687 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1688 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1689 R\ :sub:`byte` returns the value written by that write.
1690 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1691 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1692 Memory Ordering Constraints <ordering>` section for additional
1693 constraints on how the choice is made.
1694 - Otherwise R\ :sub:`byte` returns ``undef``.
1696 R returns the value composed of the series of bytes it read. This
1697 implies that some bytes within the value may be ``undef`` **without**
1698 the entire value being ``undef``. Note that this only defines the
1699 semantics of the operation; it doesn't mean that targets will emit more
1700 than one instruction to read the series of bytes.
1702 Note that in cases where none of the atomic intrinsics are used, this
1703 model places only one restriction on IR transformations on top of what
1704 is required for single-threaded execution: introducing a store to a byte
1705 which might not otherwise be stored is not allowed in general.
1706 (Specifically, in the case where another thread might write to and read
1707 from an address, introducing a store can change a load that may see
1708 exactly one write into a load that may see multiple writes.)
1712 Atomic Memory Ordering Constraints
1713 ----------------------------------
1715 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1716 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1717 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1718 ordering parameters that determine which other atomic instructions on
1719 the same address they *synchronize with*. These semantics are borrowed
1720 from Java and C++0x, but are somewhat more colloquial. If these
1721 descriptions aren't precise enough, check those specs (see spec
1722 references in the :doc:`atomics guide <Atomics>`).
1723 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1724 differently since they don't take an address. See that instruction's
1725 documentation for details.
1727 For a simpler introduction to the ordering constraints, see the
1731 The set of values that can be read is governed by the happens-before
1732 partial order. A value cannot be read unless some operation wrote
1733 it. This is intended to provide a guarantee strong enough to model
1734 Java's non-volatile shared variables. This ordering cannot be
1735 specified for read-modify-write operations; it is not strong enough
1736 to make them atomic in any interesting way.
1738 In addition to the guarantees of ``unordered``, there is a single
1739 total order for modifications by ``monotonic`` operations on each
1740 address. All modification orders must be compatible with the
1741 happens-before order. There is no guarantee that the modification
1742 orders can be combined to a global total order for the whole program
1743 (and this often will not be possible). The read in an atomic
1744 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1745 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1746 order immediately before the value it writes. If one atomic read
1747 happens before another atomic read of the same address, the later
1748 read must see the same value or a later value in the address's
1749 modification order. This disallows reordering of ``monotonic`` (or
1750 stronger) operations on the same address. If an address is written
1751 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1752 read that address repeatedly, the other threads must eventually see
1753 the write. This corresponds to the C++0x/C1x
1754 ``memory_order_relaxed``.
1756 In addition to the guarantees of ``monotonic``, a
1757 *synchronizes-with* edge may be formed with a ``release`` operation.
1758 This is intended to model C++'s ``memory_order_acquire``.
1760 In addition to the guarantees of ``monotonic``, if this operation
1761 writes a value which is subsequently read by an ``acquire``
1762 operation, it *synchronizes-with* that operation. (This isn't a
1763 complete description; see the C++0x definition of a release
1764 sequence.) This corresponds to the C++0x/C1x
1765 ``memory_order_release``.
1766 ``acq_rel`` (acquire+release)
1767 Acts as both an ``acquire`` and ``release`` operation on its
1768 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1769 ``seq_cst`` (sequentially consistent)
1770 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1771 operation that only reads, ``release`` for an operation that only
1772 writes), there is a global total order on all
1773 sequentially-consistent operations on all addresses, which is
1774 consistent with the *happens-before* partial order and with the
1775 modification orders of all the affected addresses. Each
1776 sequentially-consistent read sees the last preceding write to the
1777 same address in this global order. This corresponds to the C++0x/C1x
1778 ``memory_order_seq_cst`` and Java volatile.
1782 If an atomic operation is marked ``singlethread``, it only *synchronizes
1783 with* or participates in modification and seq\_cst total orderings with
1784 other operations running in the same thread (for example, in signal
1792 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1793 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1794 :ref:`frem <i_frem>`) have the following flags that can be set to enable
1795 otherwise unsafe floating point operations
1798 No NaNs - Allow optimizations to assume the arguments and result are not
1799 NaN. Such optimizations are required to retain defined behavior over
1800 NaNs, but the value of the result is undefined.
1803 No Infs - Allow optimizations to assume the arguments and result are not
1804 +/-Inf. Such optimizations are required to retain defined behavior over
1805 +/-Inf, but the value of the result is undefined.
1808 No Signed Zeros - Allow optimizations to treat the sign of a zero
1809 argument or result as insignificant.
1812 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1813 argument rather than perform division.
1816 Fast - Allow algebraically equivalent transformations that may
1817 dramatically change results in floating point (e.g. reassociate). This
1818 flag implies all the others.
1822 Use-list Order Directives
1823 -------------------------
1825 Use-list directives encode the in-memory order of each use-list, allowing the
1826 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1827 indexes that are assigned to the referenced value's uses. The referenced
1828 value's use-list is immediately sorted by these indexes.
1830 Use-list directives may appear at function scope or global scope. They are not
1831 instructions, and have no effect on the semantics of the IR. When they're at
1832 function scope, they must appear after the terminator of the final basic block.
1834 If basic blocks have their address taken via ``blockaddress()`` expressions,
1835 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1842 uselistorder <ty> <value>, { <order-indexes> }
1843 uselistorder_bb @function, %block { <order-indexes> }
1849 define void @foo(i32 %arg1, i32 %arg2) {
1851 ; ... instructions ...
1853 ; ... instructions ...
1855 ; At function scope.
1856 uselistorder i32 %arg1, { 1, 0, 2 }
1857 uselistorder label %bb, { 1, 0 }
1861 uselistorder i32* @global, { 1, 2, 0 }
1862 uselistorder i32 7, { 1, 0 }
1863 uselistorder i32 (i32) @bar, { 1, 0 }
1864 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1871 The LLVM type system is one of the most important features of the
1872 intermediate representation. Being typed enables a number of
1873 optimizations to be performed on the intermediate representation
1874 directly, without having to do extra analyses on the side before the
1875 transformation. A strong type system makes it easier to read the
1876 generated code and enables novel analyses and transformations that are
1877 not feasible to perform on normal three address code representations.
1887 The void type does not represent any value and has no size.
1905 The function type can be thought of as a function signature. It consists of a
1906 return type and a list of formal parameter types. The return type of a function
1907 type is a void type or first class type --- except for :ref:`label <t_label>`
1908 and :ref:`metadata <t_metadata>` types.
1914 <returntype> (<parameter list>)
1916 ...where '``<parameter list>``' is a comma-separated list of type
1917 specifiers. Optionally, the parameter list may include a type ``...``, which
1918 indicates that the function takes a variable number of arguments. Variable
1919 argument functions can access their arguments with the :ref:`variable argument
1920 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1921 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1925 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1926 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1927 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1928 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1929 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1930 | ``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. |
1931 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1932 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1933 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1940 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1941 Values of these types are the only ones which can be produced by
1949 These are the types that are valid in registers from CodeGen's perspective.
1958 The integer type is a very simple type that simply specifies an
1959 arbitrary bit width for the integer type desired. Any bit width from 1
1960 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1968 The number of bits the integer will occupy is specified by the ``N``
1974 +----------------+------------------------------------------------+
1975 | ``i1`` | a single-bit integer. |
1976 +----------------+------------------------------------------------+
1977 | ``i32`` | a 32-bit integer. |
1978 +----------------+------------------------------------------------+
1979 | ``i1942652`` | a really big integer of over 1 million bits. |
1980 +----------------+------------------------------------------------+
1984 Floating Point Types
1985 """"""""""""""""""""
1994 - 16-bit floating point value
1997 - 32-bit floating point value
2000 - 64-bit floating point value
2003 - 128-bit floating point value (112-bit mantissa)
2006 - 80-bit floating point value (X87)
2009 - 128-bit floating point value (two 64-bits)
2016 The x86_mmx type represents a value held in an MMX register on an x86
2017 machine. The operations allowed on it are quite limited: parameters and
2018 return values, load and store, and bitcast. User-specified MMX
2019 instructions are represented as intrinsic or asm calls with arguments
2020 and/or results of this type. There are no arrays, vectors or constants
2037 The pointer type is used to specify memory locations. Pointers are
2038 commonly used to reference objects in memory.
2040 Pointer types may have an optional address space attribute defining the
2041 numbered address space where the pointed-to object resides. The default
2042 address space is number zero. The semantics of non-zero address spaces
2043 are target-specific.
2045 Note that LLVM does not permit pointers to void (``void*``) nor does it
2046 permit pointers to labels (``label*``). Use ``i8*`` instead.
2056 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2057 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2058 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2059 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2060 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2061 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2062 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2071 A vector type is a simple derived type that represents a vector of
2072 elements. Vector types are used when multiple primitive data are
2073 operated in parallel using a single instruction (SIMD). A vector type
2074 requires a size (number of elements) and an underlying primitive data
2075 type. Vector types are considered :ref:`first class <t_firstclass>`.
2081 < <# elements> x <elementtype> >
2083 The number of elements is a constant integer value larger than 0;
2084 elementtype may be any integer, floating point or pointer type. Vectors
2085 of size zero are not allowed.
2089 +-------------------+--------------------------------------------------+
2090 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2091 +-------------------+--------------------------------------------------+
2092 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2093 +-------------------+--------------------------------------------------+
2094 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2095 +-------------------+--------------------------------------------------+
2096 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2097 +-------------------+--------------------------------------------------+
2106 The label type represents code labels.
2121 The metadata type represents embedded metadata. No derived types may be
2122 created from metadata except for :ref:`function <t_function>` arguments.
2135 Aggregate Types are a subset of derived types that can contain multiple
2136 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2137 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2147 The array type is a very simple derived type that arranges elements
2148 sequentially in memory. The array type requires a size (number of
2149 elements) and an underlying data type.
2155 [<# elements> x <elementtype>]
2157 The number of elements is a constant integer value; ``elementtype`` may
2158 be any type with a size.
2162 +------------------+--------------------------------------+
2163 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2164 +------------------+--------------------------------------+
2165 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2166 +------------------+--------------------------------------+
2167 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2168 +------------------+--------------------------------------+
2170 Here are some examples of multidimensional arrays:
2172 +-----------------------------+----------------------------------------------------------+
2173 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2174 +-----------------------------+----------------------------------------------------------+
2175 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2176 +-----------------------------+----------------------------------------------------------+
2177 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2178 +-----------------------------+----------------------------------------------------------+
2180 There is no restriction on indexing beyond the end of the array implied
2181 by a static type (though there are restrictions on indexing beyond the
2182 bounds of an allocated object in some cases). This means that
2183 single-dimension 'variable sized array' addressing can be implemented in
2184 LLVM with a zero length array type. An implementation of 'pascal style
2185 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2195 The structure type is used to represent a collection of data members
2196 together in memory. The elements of a structure may be any type that has
2199 Structures in memory are accessed using '``load``' and '``store``' by
2200 getting a pointer to a field with the '``getelementptr``' instruction.
2201 Structures in registers are accessed using the '``extractvalue``' and
2202 '``insertvalue``' instructions.
2204 Structures may optionally be "packed" structures, which indicate that
2205 the alignment of the struct is one byte, and that there is no padding
2206 between the elements. In non-packed structs, padding between field types
2207 is inserted as defined by the DataLayout string in the module, which is
2208 required to match what the underlying code generator expects.
2210 Structures can either be "literal" or "identified". A literal structure
2211 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2212 identified types are always defined at the top level with a name.
2213 Literal types are uniqued by their contents and can never be recursive
2214 or opaque since there is no way to write one. Identified types can be
2215 recursive, can be opaqued, and are never uniqued.
2221 %T1 = type { <type list> } ; Identified normal struct type
2222 %T2 = type <{ <type list> }> ; Identified packed struct type
2226 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2227 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2228 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2229 | ``{ 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``. |
2230 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2231 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2232 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2236 Opaque Structure Types
2237 """"""""""""""""""""""
2241 Opaque structure types are used to represent named structure types that
2242 do not have a body specified. This corresponds (for example) to the C
2243 notion of a forward declared structure.
2254 +--------------+-------------------+
2255 | ``opaque`` | An opaque type. |
2256 +--------------+-------------------+
2263 LLVM has several different basic types of constants. This section
2264 describes them all and their syntax.
2269 **Boolean constants**
2270 The two strings '``true``' and '``false``' are both valid constants
2272 **Integer constants**
2273 Standard integers (such as '4') are constants of the
2274 :ref:`integer <t_integer>` type. Negative numbers may be used with
2276 **Floating point constants**
2277 Floating point constants use standard decimal notation (e.g.
2278 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2279 hexadecimal notation (see below). The assembler requires the exact
2280 decimal value of a floating-point constant. For example, the
2281 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2282 decimal in binary. Floating point constants must have a :ref:`floating
2283 point <t_floating>` type.
2284 **Null pointer constants**
2285 The identifier '``null``' is recognized as a null pointer constant
2286 and must be of :ref:`pointer type <t_pointer>`.
2288 The one non-intuitive notation for constants is the hexadecimal form of
2289 floating point constants. For example, the form
2290 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2291 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2292 constants are required (and the only time that they are generated by the
2293 disassembler) is when a floating point constant must be emitted but it
2294 cannot be represented as a decimal floating point number in a reasonable
2295 number of digits. For example, NaN's, infinities, and other special
2296 values are represented in their IEEE hexadecimal format so that assembly
2297 and disassembly do not cause any bits to change in the constants.
2299 When using the hexadecimal form, constants of types half, float, and
2300 double are represented using the 16-digit form shown above (which
2301 matches the IEEE754 representation for double); half and float values
2302 must, however, be exactly representable as IEEE 754 half and single
2303 precision, respectively. Hexadecimal format is always used for long
2304 double, and there are three forms of long double. The 80-bit format used
2305 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2306 128-bit format used by PowerPC (two adjacent doubles) is represented by
2307 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2308 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2309 will only work if they match the long double format on your target.
2310 The IEEE 16-bit format (half precision) is represented by ``0xH``
2311 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2312 (sign bit at the left).
2314 There are no constants of type x86_mmx.
2316 .. _complexconstants:
2321 Complex constants are a (potentially recursive) combination of simple
2322 constants and smaller complex constants.
2324 **Structure constants**
2325 Structure constants are represented with notation similar to
2326 structure type definitions (a comma separated list of elements,
2327 surrounded by braces (``{}``)). For example:
2328 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2329 "``@G = external global i32``". Structure constants must have
2330 :ref:`structure type <t_struct>`, and the number and types of elements
2331 must match those specified by the type.
2333 Array constants are represented with notation similar to array type
2334 definitions (a comma separated list of elements, surrounded by
2335 square brackets (``[]``)). For example:
2336 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2337 :ref:`array type <t_array>`, and the number and types of elements must
2338 match those specified by the type. As a special case, character array
2339 constants may also be represented as a double-quoted string using the ``c``
2340 prefix. For example: "``c"Hello World\0A\00"``".
2341 **Vector constants**
2342 Vector constants are represented with notation similar to vector
2343 type definitions (a comma separated list of elements, surrounded by
2344 less-than/greater-than's (``<>``)). For example:
2345 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2346 must have :ref:`vector type <t_vector>`, and the number and types of
2347 elements must match those specified by the type.
2348 **Zero initialization**
2349 The string '``zeroinitializer``' can be used to zero initialize a
2350 value to zero of *any* type, including scalar and
2351 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2352 having to print large zero initializers (e.g. for large arrays) and
2353 is always exactly equivalent to using explicit zero initializers.
2355 A metadata node is a constant tuple without types. For example:
2356 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2357 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2358 Unlike other typed constants that are meant to be interpreted as part of
2359 the instruction stream, metadata is a place to attach additional
2360 information such as debug info.
2362 Global Variable and Function Addresses
2363 --------------------------------------
2365 The addresses of :ref:`global variables <globalvars>` and
2366 :ref:`functions <functionstructure>` are always implicitly valid
2367 (link-time) constants. These constants are explicitly referenced when
2368 the :ref:`identifier for the global <identifiers>` is used and always have
2369 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2372 .. code-block:: llvm
2376 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2383 The string '``undef``' can be used anywhere a constant is expected, and
2384 indicates that the user of the value may receive an unspecified
2385 bit-pattern. Undefined values may be of any type (other than '``label``'
2386 or '``void``') and be used anywhere a constant is permitted.
2388 Undefined values are useful because they indicate to the compiler that
2389 the program is well defined no matter what value is used. This gives the
2390 compiler more freedom to optimize. Here are some examples of
2391 (potentially surprising) transformations that are valid (in pseudo IR):
2393 .. code-block:: llvm
2403 This is safe because all of the output bits are affected by the undef
2404 bits. Any output bit can have a zero or one depending on the input bits.
2406 .. code-block:: llvm
2417 These logical operations have bits that are not always affected by the
2418 input. For example, if ``%X`` has a zero bit, then the output of the
2419 '``and``' operation will always be a zero for that bit, no matter what
2420 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2421 optimize or assume that the result of the '``and``' is '``undef``'.
2422 However, it is safe to assume that all bits of the '``undef``' could be
2423 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2424 all the bits of the '``undef``' operand to the '``or``' could be set,
2425 allowing the '``or``' to be folded to -1.
2427 .. code-block:: llvm
2429 %A = select undef, %X, %Y
2430 %B = select undef, 42, %Y
2431 %C = select %X, %Y, undef
2441 This set of examples shows that undefined '``select``' (and conditional
2442 branch) conditions can go *either way*, but they have to come from one
2443 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2444 both known to have a clear low bit, then ``%A`` would have to have a
2445 cleared low bit. However, in the ``%C`` example, the optimizer is
2446 allowed to assume that the '``undef``' operand could be the same as
2447 ``%Y``, allowing the whole '``select``' to be eliminated.
2449 .. code-block:: llvm
2451 %A = xor undef, undef
2468 This example points out that two '``undef``' operands are not
2469 necessarily the same. This can be surprising to people (and also matches
2470 C semantics) where they assume that "``X^X``" is always zero, even if
2471 ``X`` is undefined. This isn't true for a number of reasons, but the
2472 short answer is that an '``undef``' "variable" can arbitrarily change
2473 its value over its "live range". This is true because the variable
2474 doesn't actually *have a live range*. Instead, the value is logically
2475 read from arbitrary registers that happen to be around when needed, so
2476 the value is not necessarily consistent over time. In fact, ``%A`` and
2477 ``%C`` need to have the same semantics or the core LLVM "replace all
2478 uses with" concept would not hold.
2480 .. code-block:: llvm
2488 These examples show the crucial difference between an *undefined value*
2489 and *undefined behavior*. An undefined value (like '``undef``') is
2490 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2491 operation can be constant folded to '``undef``', because the '``undef``'
2492 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2493 However, in the second example, we can make a more aggressive
2494 assumption: because the ``undef`` is allowed to be an arbitrary value,
2495 we are allowed to assume that it could be zero. Since a divide by zero
2496 has *undefined behavior*, we are allowed to assume that the operation
2497 does not execute at all. This allows us to delete the divide and all
2498 code after it. Because the undefined operation "can't happen", the
2499 optimizer can assume that it occurs in dead code.
2501 .. code-block:: llvm
2503 a: store undef -> %X
2504 b: store %X -> undef
2509 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2510 value can be assumed to not have any effect; we can assume that the
2511 value is overwritten with bits that happen to match what was already
2512 there. However, a store *to* an undefined location could clobber
2513 arbitrary memory, therefore, it has undefined behavior.
2520 Poison values are similar to :ref:`undef values <undefvalues>`, however
2521 they also represent the fact that an instruction or constant expression
2522 that cannot evoke side effects has nevertheless detected a condition
2523 that results in undefined behavior.
2525 There is currently no way of representing a poison value in the IR; they
2526 only exist when produced by operations such as :ref:`add <i_add>` with
2529 Poison value behavior is defined in terms of value *dependence*:
2531 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2532 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2533 their dynamic predecessor basic block.
2534 - Function arguments depend on the corresponding actual argument values
2535 in the dynamic callers of their functions.
2536 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2537 instructions that dynamically transfer control back to them.
2538 - :ref:`Invoke <i_invoke>` instructions depend on the
2539 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2540 call instructions that dynamically transfer control back to them.
2541 - Non-volatile loads and stores depend on the most recent stores to all
2542 of the referenced memory addresses, following the order in the IR
2543 (including loads and stores implied by intrinsics such as
2544 :ref:`@llvm.memcpy <int_memcpy>`.)
2545 - An instruction with externally visible side effects depends on the
2546 most recent preceding instruction with externally visible side
2547 effects, following the order in the IR. (This includes :ref:`volatile
2548 operations <volatile>`.)
2549 - An instruction *control-depends* on a :ref:`terminator
2550 instruction <terminators>` if the terminator instruction has
2551 multiple successors and the instruction is always executed when
2552 control transfers to one of the successors, and may not be executed
2553 when control is transferred to another.
2554 - Additionally, an instruction also *control-depends* on a terminator
2555 instruction if the set of instructions it otherwise depends on would
2556 be different if the terminator had transferred control to a different
2558 - Dependence is transitive.
2560 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2561 with the additional effect that any instruction that has a *dependence*
2562 on a poison value has undefined behavior.
2564 Here are some examples:
2566 .. code-block:: llvm
2569 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2570 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2571 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2572 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2574 store i32 %poison, i32* @g ; Poison value stored to memory.
2575 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2577 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2579 %narrowaddr = bitcast i32* @g to i16*
2580 %wideaddr = bitcast i32* @g to i64*
2581 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2582 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2584 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2585 br i1 %cmp, label %true, label %end ; Branch to either destination.
2588 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2589 ; it has undefined behavior.
2593 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2594 ; Both edges into this PHI are
2595 ; control-dependent on %cmp, so this
2596 ; always results in a poison value.
2598 store volatile i32 0, i32* @g ; This would depend on the store in %true
2599 ; if %cmp is true, or the store in %entry
2600 ; otherwise, so this is undefined behavior.
2602 br i1 %cmp, label %second_true, label %second_end
2603 ; The same branch again, but this time the
2604 ; true block doesn't have side effects.
2611 store volatile i32 0, i32* @g ; This time, the instruction always depends
2612 ; on the store in %end. Also, it is
2613 ; control-equivalent to %end, so this is
2614 ; well-defined (ignoring earlier undefined
2615 ; behavior in this example).
2619 Addresses of Basic Blocks
2620 -------------------------
2622 ``blockaddress(@function, %block)``
2624 The '``blockaddress``' constant computes the address of the specified
2625 basic block in the specified function, and always has an ``i8*`` type.
2626 Taking the address of the entry block is illegal.
2628 This value only has defined behavior when used as an operand to the
2629 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2630 against null. Pointer equality tests between labels addresses results in
2631 undefined behavior --- though, again, comparison against null is ok, and
2632 no label is equal to the null pointer. This may be passed around as an
2633 opaque pointer sized value as long as the bits are not inspected. This
2634 allows ``ptrtoint`` and arithmetic to be performed on these values so
2635 long as the original value is reconstituted before the ``indirectbr``
2638 Finally, some targets may provide defined semantics when using the value
2639 as the operand to an inline assembly, but that is target specific.
2643 Constant Expressions
2644 --------------------
2646 Constant expressions are used to allow expressions involving other
2647 constants to be used as constants. Constant expressions may be of any
2648 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2649 that does not have side effects (e.g. load and call are not supported).
2650 The following is the syntax for constant expressions:
2652 ``trunc (CST to TYPE)``
2653 Truncate a constant to another type. The bit size of CST must be
2654 larger than the bit size of TYPE. Both types must be integers.
2655 ``zext (CST to TYPE)``
2656 Zero extend a constant to another type. The bit size of CST must be
2657 smaller than the bit size of TYPE. Both types must be integers.
2658 ``sext (CST to TYPE)``
2659 Sign extend a constant to another type. The bit size of CST must be
2660 smaller than the bit size of TYPE. Both types must be integers.
2661 ``fptrunc (CST to TYPE)``
2662 Truncate a floating point constant to another floating point type.
2663 The size of CST must be larger than the size of TYPE. Both types
2664 must be floating point.
2665 ``fpext (CST to TYPE)``
2666 Floating point extend a constant to another type. The size of CST
2667 must be smaller or equal to the size of TYPE. Both types must be
2669 ``fptoui (CST to TYPE)``
2670 Convert a floating point constant to the corresponding unsigned
2671 integer constant. TYPE must be a scalar or vector integer type. CST
2672 must be of scalar or vector floating point type. Both CST and TYPE
2673 must be scalars, or vectors of the same number of elements. If the
2674 value won't fit in the integer type, the results are undefined.
2675 ``fptosi (CST to TYPE)``
2676 Convert a floating point constant to the corresponding signed
2677 integer constant. TYPE must be a scalar or vector integer type. CST
2678 must be of scalar or vector floating point type. Both CST and TYPE
2679 must be scalars, or vectors of the same number of elements. If the
2680 value won't fit in the integer type, the results are undefined.
2681 ``uitofp (CST to TYPE)``
2682 Convert an unsigned integer constant to the corresponding floating
2683 point constant. TYPE must be a scalar or vector floating point type.
2684 CST must be of scalar or vector integer type. Both CST and TYPE must
2685 be scalars, or vectors of the same number of elements. If the value
2686 won't fit in the floating point type, the results are undefined.
2687 ``sitofp (CST to TYPE)``
2688 Convert a signed integer constant to the corresponding floating
2689 point constant. TYPE must be a scalar or vector floating point type.
2690 CST must be of scalar or vector integer type. Both CST and TYPE must
2691 be scalars, or vectors of the same number of elements. If the value
2692 won't fit in the floating point type, the results are undefined.
2693 ``ptrtoint (CST to TYPE)``
2694 Convert a pointer typed constant to the corresponding integer
2695 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2696 pointer type. The ``CST`` value is zero extended, truncated, or
2697 unchanged to make it fit in ``TYPE``.
2698 ``inttoptr (CST to TYPE)``
2699 Convert an integer constant to a pointer constant. TYPE must be a
2700 pointer type. CST must be of integer type. The CST value is zero
2701 extended, truncated, or unchanged to make it fit in a pointer size.
2702 This one is *really* dangerous!
2703 ``bitcast (CST to TYPE)``
2704 Convert a constant, CST, to another TYPE. The constraints of the
2705 operands are the same as those for the :ref:`bitcast
2706 instruction <i_bitcast>`.
2707 ``addrspacecast (CST to TYPE)``
2708 Convert a constant pointer or constant vector of pointer, CST, to another
2709 TYPE in a different address space. The constraints of the operands are the
2710 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2711 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2712 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2713 constants. As with the :ref:`getelementptr <i_getelementptr>`
2714 instruction, the index list may have zero or more indexes, which are
2715 required to make sense for the type of "pointer to TY".
2716 ``select (COND, VAL1, VAL2)``
2717 Perform the :ref:`select operation <i_select>` on constants.
2718 ``icmp COND (VAL1, VAL2)``
2719 Performs the :ref:`icmp operation <i_icmp>` on constants.
2720 ``fcmp COND (VAL1, VAL2)``
2721 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2722 ``extractelement (VAL, IDX)``
2723 Perform the :ref:`extractelement operation <i_extractelement>` on
2725 ``insertelement (VAL, ELT, IDX)``
2726 Perform the :ref:`insertelement operation <i_insertelement>` on
2728 ``shufflevector (VEC1, VEC2, IDXMASK)``
2729 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2731 ``extractvalue (VAL, IDX0, IDX1, ...)``
2732 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2733 constants. The index list is interpreted in a similar manner as
2734 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2735 least one index value must be specified.
2736 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2737 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2738 The index list is interpreted in a similar manner as indices in a
2739 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2740 value must be specified.
2741 ``OPCODE (LHS, RHS)``
2742 Perform the specified operation of the LHS and RHS constants. OPCODE
2743 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2744 binary <bitwiseops>` operations. The constraints on operands are
2745 the same as those for the corresponding instruction (e.g. no bitwise
2746 operations on floating point values are allowed).
2753 Inline Assembler Expressions
2754 ----------------------------
2756 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2757 Inline Assembly <moduleasm>`) through the use of a special value. This
2758 value represents the inline assembler as a string (containing the
2759 instructions to emit), a list of operand constraints (stored as a
2760 string), a flag that indicates whether or not the inline asm expression
2761 has side effects, and a flag indicating whether the function containing
2762 the asm needs to align its stack conservatively. An example inline
2763 assembler expression is:
2765 .. code-block:: llvm
2767 i32 (i32) asm "bswap $0", "=r,r"
2769 Inline assembler expressions may **only** be used as the callee operand
2770 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2771 Thus, typically we have:
2773 .. code-block:: llvm
2775 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2777 Inline asms with side effects not visible in the constraint list must be
2778 marked as having side effects. This is done through the use of the
2779 '``sideeffect``' keyword, like so:
2781 .. code-block:: llvm
2783 call void asm sideeffect "eieio", ""()
2785 In some cases inline asms will contain code that will not work unless
2786 the stack is aligned in some way, such as calls or SSE instructions on
2787 x86, yet will not contain code that does that alignment within the asm.
2788 The compiler should make conservative assumptions about what the asm
2789 might contain and should generate its usual stack alignment code in the
2790 prologue if the '``alignstack``' keyword is present:
2792 .. code-block:: llvm
2794 call void asm alignstack "eieio", ""()
2796 Inline asms also support using non-standard assembly dialects. The
2797 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2798 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2799 the only supported dialects. An example is:
2801 .. code-block:: llvm
2803 call void asm inteldialect "eieio", ""()
2805 If multiple keywords appear the '``sideeffect``' keyword must come
2806 first, the '``alignstack``' keyword second and the '``inteldialect``'
2812 The call instructions that wrap inline asm nodes may have a
2813 "``!srcloc``" MDNode attached to it that contains a list of constant
2814 integers. If present, the code generator will use the integer as the
2815 location cookie value when report errors through the ``LLVMContext``
2816 error reporting mechanisms. This allows a front-end to correlate backend
2817 errors that occur with inline asm back to the source code that produced
2820 .. code-block:: llvm
2822 call void asm sideeffect "something bad", ""(), !srcloc !42
2824 !42 = !{ i32 1234567 }
2826 It is up to the front-end to make sense of the magic numbers it places
2827 in the IR. If the MDNode contains multiple constants, the code generator
2828 will use the one that corresponds to the line of the asm that the error
2836 LLVM IR allows metadata to be attached to instructions in the program
2837 that can convey extra information about the code to the optimizers and
2838 code generator. One example application of metadata is source-level
2839 debug information. There are two metadata primitives: strings and nodes.
2841 Metadata does not have a type, and is not a value. If referenced from a
2842 ``call`` instruction, it uses the ``metadata`` type.
2844 All metadata are identified in syntax by a exclamation point ('``!``').
2846 .. _metadata-string:
2848 Metadata Nodes and Metadata Strings
2849 -----------------------------------
2851 A metadata string is a string surrounded by double quotes. It can
2852 contain any character by escaping non-printable characters with
2853 "``\xx``" where "``xx``" is the two digit hex code. For example:
2856 Metadata nodes are represented with notation similar to structure
2857 constants (a comma separated list of elements, surrounded by braces and
2858 preceded by an exclamation point). Metadata nodes can have any values as
2859 their operand. For example:
2861 .. code-block:: llvm
2863 !{ !"test\00", i32 10}
2865 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
2867 .. code-block:: llvm
2869 !0 = distinct !{!"test\00", i32 10}
2871 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
2872 content. They can also occur when transformations cause uniquing collisions
2873 when metadata operands change.
2875 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2876 metadata nodes, which can be looked up in the module symbol table. For
2879 .. code-block:: llvm
2883 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2884 function is using two metadata arguments:
2886 .. code-block:: llvm
2888 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2890 Metadata can be attached with an instruction. Here metadata ``!21`` is
2891 attached to the ``add`` instruction using the ``!dbg`` identifier:
2893 .. code-block:: llvm
2895 %indvar.next = add i64 %indvar, 1, !dbg !21
2897 More information about specific metadata nodes recognized by the
2898 optimizers and code generator is found below.
2900 .. _specialized-metadata:
2902 Specialized Metadata Nodes
2903 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2905 Specialized metadata nodes are custom data structures in metadata (as opposed
2906 to generic tuples). Their fields are labelled, and can be specified in any
2909 These aren't inherently debug info centric, but currently all the specialized
2910 metadata nodes are related to debug info.
2917 ``MDCompileUnit`` nodes represent a compile unit. The ``enums:``,
2918 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
2919 tuples containing the debug info to be emitted along with the compile unit,
2920 regardless of code optimizations (some nodes are only emitted if there are
2921 references to them from instructions).
2923 .. code-block:: llvm
2925 !0 = !MDCompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
2926 isOptimized: true, flags: "-O2", runtimeVersion: 2,
2927 splitDebugFilename: "abc.debug", emissionKind: 1,
2928 enums: !2, retainedTypes: !3, subprograms: !4,
2929 globals: !5, imports: !6)
2931 Compile unit descriptors provide the root scope for objects declared in a
2932 specific compilation unit. File descriptors are defined using this scope.
2933 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
2934 keep track of subprograms, global variables, type information, and imported
2935 entities (declarations and namespaces).
2942 ``MDFile`` nodes represent files. The ``filename:`` can include slashes.
2944 .. code-block:: llvm
2946 !0 = !MDFile(filename: "path/to/file", directory: "/path/to/dir")
2948 Files are sometimes used in ``scope:`` fields, and are the only valid target
2949 for ``file:`` fields.
2956 ``MDBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
2957 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
2959 .. code-block:: llvm
2961 !0 = !MDBasicType(name: "unsigned char", size: 8, align: 8,
2962 encoding: DW_ATE_unsigned_char)
2963 !1 = !MDBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
2965 The ``encoding:`` describes the details of the type. Usually it's one of the
2968 .. code-block:: llvm
2974 DW_ATE_signed_char = 6
2976 DW_ATE_unsigned_char = 8
2978 .. _MDSubroutineType:
2983 ``MDSubroutineType`` nodes represent subroutine types. Their ``types:`` field
2984 refers to a tuple; the first operand is the return type, while the rest are the
2985 types of the formal arguments in order. If the first operand is ``null``, that
2986 represents a function with no return value (such as ``void foo() {}`` in C++).
2988 .. code-block:: llvm
2990 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
2991 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
2992 !2 = !MDSubroutineType(types: !{null, !0, !1}) ; void (int, char)
2999 ``MDDerivedType`` nodes represent types derived from other types, such as
3002 .. code-block:: llvm
3004 !0 = !MDBasicType(name: "unsigned char", size: 8, align: 8,
3005 encoding: DW_ATE_unsigned_char)
3006 !1 = !MDDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3009 The following ``tag:`` values are valid:
3011 .. code-block:: llvm
3013 DW_TAG_formal_parameter = 5
3015 DW_TAG_pointer_type = 15
3016 DW_TAG_reference_type = 16
3018 DW_TAG_ptr_to_member_type = 31
3019 DW_TAG_const_type = 38
3020 DW_TAG_volatile_type = 53
3021 DW_TAG_restrict_type = 55
3023 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3024 <MDCompositeType>` or :ref:`subprogram <MDSubprogram>`. The type of the member
3025 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3026 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3027 argument of a subprogram.
3029 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3031 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3032 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3035 Note that the ``void *`` type is expressed as a type derived from NULL.
3037 .. _MDCompositeType:
3042 ``MDCompositeType`` nodes represent types composed of other types, like
3043 structures and unions. ``elements:`` points to a tuple of the composed types.
3045 If the source language supports ODR, the ``identifier:`` field gives the unique
3046 identifier used for type merging between modules. When specified, other types
3047 can refer to composite types indirectly via a :ref:`metadata string
3048 <metadata-string>` that matches their identifier.
3050 .. code-block:: llvm
3052 !0 = !MDEnumerator(name: "SixKind", value: 7)
3053 !1 = !MDEnumerator(name: "SevenKind", value: 7)
3054 !2 = !MDEnumerator(name: "NegEightKind", value: -8)
3055 !3 = !MDCompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3056 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3057 elements: !{!0, !1, !2})
3059 The following ``tag:`` values are valid:
3061 .. code-block:: llvm
3063 DW_TAG_array_type = 1
3064 DW_TAG_class_type = 2
3065 DW_TAG_enumeration_type = 4
3066 DW_TAG_structure_type = 19
3067 DW_TAG_union_type = 23
3068 DW_TAG_subroutine_type = 21
3069 DW_TAG_inheritance = 28
3072 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3073 descriptors <MDSubrange>`, each representing the range of subscripts at that
3074 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3075 array type is a native packed vector.
3077 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3078 descriptors <MDEnumerator>`, each representing the definition of an enumeration
3079 value for the set. All enumeration type descriptors are collected in the
3080 ``enums:`` field of the :ref:`compile unit <MDCompileUnit>`.
3082 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3083 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3084 <MDDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3091 ``MDSubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3092 :ref:`MDCompositeType`. ``count: -1`` indicates an empty array.
3094 .. code-block:: llvm
3096 !0 = !MDSubrange(count: 5, lowerBound: 0) ; array counting from 0
3097 !1 = !MDSubrange(count: 5, lowerBound: 1) ; array counting from 1
3098 !2 = !MDSubrange(count: -1) ; empty array.
3105 ``MDEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3106 variants of :ref:`MDCompositeType`.
3108 .. code-block:: llvm
3110 !0 = !MDEnumerator(name: "SixKind", value: 7)
3111 !1 = !MDEnumerator(name: "SevenKind", value: 7)
3112 !2 = !MDEnumerator(name: "NegEightKind", value: -8)
3114 MDTemplateTypeParameter
3115 """""""""""""""""""""""
3117 ``MDTemplateTypeParameter`` nodes represent type parameters to generic source
3118 language constructs. They are used (optionally) in :ref:`MDCompositeType` and
3119 :ref:`MDSubprogram` ``templateParams:`` fields.
3121 .. code-block:: llvm
3123 !0 = !MDTemplateTypeParameter(name: "Ty", type: !1)
3125 MDTemplateValueParameter
3126 """"""""""""""""""""""""
3128 ``MDTemplateValueParameter`` nodes represent value parameters to generic source
3129 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3130 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3131 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3132 :ref:`MDCompositeType` and :ref:`MDSubprogram` ``templateParams:`` fields.
3134 .. code-block:: llvm
3136 !0 = !MDTemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3141 ``MDNamespace`` nodes represent namespaces in the source language.
3143 .. code-block:: llvm
3145 !0 = !MDNamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3150 ``MDGlobalVariable`` nodes represent global variables in the source language.
3152 .. code-block:: llvm
3154 !0 = !MDGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3155 file: !2, line: 7, type: !3, isLocal: true,
3156 isDefinition: false, variable: i32* @foo,
3159 All global variables should be referenced by the `globals:` field of a
3160 :ref:`compile unit <MDCompileUnit>`.
3167 ``MDSubprogram`` nodes represent functions from the source language. The
3168 ``variables:`` field points at :ref:`variables <MDLocalVariable>` that must be
3169 retained, even if their IR counterparts are optimized out of the IR. The
3170 ``type:`` field must point at an :ref:`MDSubroutineType`.
3172 .. code-block:: llvm
3174 !0 = !MDSubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3175 file: !2, line: 7, type: !3, isLocal: true,
3176 isDefinition: false, scopeLine: 8, containingType: !4,
3177 virtuality: DW_VIRTUALITY_pure_virtual, virtualIndex: 10,
3178 flags: DIFlagPrototyped, isOptimized: true,
3179 function: void ()* @_Z3foov,
3180 templateParams: !5, declaration: !6, variables: !7)
3187 ``MDLexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3188 <MDSubprogram>`. The line number and column numbers are used to dinstinguish
3189 two lexical blocks at same depth. They are valid targets for ``scope:``
3192 .. code-block:: llvm
3194 !0 = distinct !MDLexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3196 Usually lexical blocks are ``distinct`` to prevent node merging based on
3199 .. _MDLexicalBlockFile:
3204 ``MDLexicalBlockFile`` nodes are used to discriminate between sections of a
3205 :ref:`lexical block <MDLexicalBlock>`. The ``file:`` field can be changed to
3206 indicate textual inclusion, or the ``discriminator:`` field can be used to
3207 discriminate between control flow within a single block in the source language.
3209 .. code-block:: llvm
3211 !0 = !MDLexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3212 !1 = !MDLexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3213 !2 = !MDLexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3218 ``MDLocation`` nodes represent source debug locations. The ``scope:`` field is
3219 mandatory, and points at an :ref:`MDLexicalBlockFile`, an
3220 :ref:`MDLexicalBlock`, or an :ref:`MDSubprogram`.
3222 .. code-block:: llvm
3224 !0 = !MDLocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3226 .. _MDLocalVariable:
3231 ``MDLocalVariable`` nodes represent local variables in the source language.
3232 Instead of ``DW_TAG_variable``, they use LLVM-specific fake tags to
3233 discriminate between local variables (``DW_TAG_auto_variable``) and subprogram
3234 arguments (``DW_TAG_arg_variable``). In the latter case, the ``arg:`` field
3235 specifies the argument position, and this variable will be included in the
3236 ``variables:`` field of its :ref:`MDSubprogram`.
3238 .. code-block:: llvm
3240 !0 = !MDLocalVariable(tag: DW_TAG_arg_variable, name: "this", arg: 0,
3241 scope: !3, file: !2, line: 7, type: !3,
3242 flags: DIFlagArtificial)
3243 !1 = !MDLocalVariable(tag: DW_TAG_arg_variable, name: "x", arg: 1,
3244 scope: !4, file: !2, line: 7, type: !3)
3245 !1 = !MDLocalVariable(tag: DW_TAG_auto_variable, name: "y",
3246 scope: !5, file: !2, line: 7, type: !3)
3251 ``MDExpression`` nodes represent DWARF expression sequences. They are used in
3252 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
3253 describe how the referenced LLVM variable relates to the source language
3256 The current supported vocabulary is limited:
3258 - ``DW_OP_deref`` dereferences the working expression.
3259 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
3260 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
3261 here, respectively) of the variable piece from the working expression.
3263 .. code-block:: llvm
3265 !0 = !MDExpression(DW_OP_deref)
3266 !1 = !MDExpression(DW_OP_plus, 3)
3267 !2 = !MDExpression(DW_OP_bit_piece, 3, 7)
3268 !3 = !MDExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
3273 ``MDObjCProperty`` nodes represent Objective-C property nodes.
3275 .. code-block:: llvm
3277 !3 = !MDObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
3278 getter: "getFoo", attributes: 7, type: !2)
3283 ``MDImportedEntity`` nodes represent entities (such as modules) imported into a
3286 .. code-block:: llvm
3288 !2 = !MDImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
3289 entity: !1, line: 7)
3294 In LLVM IR, memory does not have types, so LLVM's own type system is not
3295 suitable for doing TBAA. Instead, metadata is added to the IR to
3296 describe a type system of a higher level language. This can be used to
3297 implement typical C/C++ TBAA, but it can also be used to implement
3298 custom alias analysis behavior for other languages.
3300 The current metadata format is very simple. TBAA metadata nodes have up
3301 to three fields, e.g.:
3303 .. code-block:: llvm
3305 !0 = !{ !"an example type tree" }
3306 !1 = !{ !"int", !0 }
3307 !2 = !{ !"float", !0 }
3308 !3 = !{ !"const float", !2, i64 1 }
3310 The first field is an identity field. It can be any value, usually a
3311 metadata string, which uniquely identifies the type. The most important
3312 name in the tree is the name of the root node. Two trees with different
3313 root node names are entirely disjoint, even if they have leaves with
3316 The second field identifies the type's parent node in the tree, or is
3317 null or omitted for a root node. A type is considered to alias all of
3318 its descendants and all of its ancestors in the tree. Also, a type is
3319 considered to alias all types in other trees, so that bitcode produced
3320 from multiple front-ends is handled conservatively.
3322 If the third field is present, it's an integer which if equal to 1
3323 indicates that the type is "constant" (meaning
3324 ``pointsToConstantMemory`` should return true; see `other useful
3325 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
3327 '``tbaa.struct``' Metadata
3328 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3330 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
3331 aggregate assignment operations in C and similar languages, however it
3332 is defined to copy a contiguous region of memory, which is more than
3333 strictly necessary for aggregate types which contain holes due to
3334 padding. Also, it doesn't contain any TBAA information about the fields
3337 ``!tbaa.struct`` metadata can describe which memory subregions in a
3338 memcpy are padding and what the TBAA tags of the struct are.
3340 The current metadata format is very simple. ``!tbaa.struct`` metadata
3341 nodes are a list of operands which are in conceptual groups of three.
3342 For each group of three, the first operand gives the byte offset of a
3343 field in bytes, the second gives its size in bytes, and the third gives
3346 .. code-block:: llvm
3348 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
3350 This describes a struct with two fields. The first is at offset 0 bytes
3351 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
3352 and has size 4 bytes and has tbaa tag !2.
3354 Note that the fields need not be contiguous. In this example, there is a
3355 4 byte gap between the two fields. This gap represents padding which
3356 does not carry useful data and need not be preserved.
3358 '``noalias``' and '``alias.scope``' Metadata
3359 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3361 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
3362 noalias memory-access sets. This means that some collection of memory access
3363 instructions (loads, stores, memory-accessing calls, etc.) that carry
3364 ``noalias`` metadata can specifically be specified not to alias with some other
3365 collection of memory access instructions that carry ``alias.scope`` metadata.
3366 Each type of metadata specifies a list of scopes where each scope has an id and
3367 a domain. When evaluating an aliasing query, if for some domain, the set
3368 of scopes with that domain in one instruction's ``alias.scope`` list is a
3369 subset of (or equal to) the set of scopes for that domain in another
3370 instruction's ``noalias`` list, then the two memory accesses are assumed not to
3373 The metadata identifying each domain is itself a list containing one or two
3374 entries. The first entry is the name of the domain. Note that if the name is a
3375 string then it can be combined accross functions and translation units. A
3376 self-reference can be used to create globally unique domain names. A
3377 descriptive string may optionally be provided as a second list entry.
3379 The metadata identifying each scope is also itself a list containing two or
3380 three entries. The first entry is the name of the scope. Note that if the name
3381 is a string then it can be combined accross functions and translation units. A
3382 self-reference can be used to create globally unique scope names. A metadata
3383 reference to the scope's domain is the second entry. A descriptive string may
3384 optionally be provided as a third list entry.
3388 .. code-block:: llvm
3390 ; Two scope domains:
3394 ; Some scopes in these domains:
3400 !5 = !{!4} ; A list containing only scope !4
3404 ; These two instructions don't alias:
3405 %0 = load float, float* %c, align 4, !alias.scope !5
3406 store float %0, float* %arrayidx.i, align 4, !noalias !5
3408 ; These two instructions also don't alias (for domain !1, the set of scopes
3409 ; in the !alias.scope equals that in the !noalias list):
3410 %2 = load float, float* %c, align 4, !alias.scope !5
3411 store float %2, float* %arrayidx.i2, align 4, !noalias !6
3413 ; These two instructions don't alias (for domain !0, the set of scopes in
3414 ; the !noalias list is not a superset of, or equal to, the scopes in the
3415 ; !alias.scope list):
3416 %2 = load float, float* %c, align 4, !alias.scope !6
3417 store float %0, float* %arrayidx.i, align 4, !noalias !7
3419 '``fpmath``' Metadata
3420 ^^^^^^^^^^^^^^^^^^^^^
3422 ``fpmath`` metadata may be attached to any instruction of floating point
3423 type. It can be used to express the maximum acceptable error in the
3424 result of that instruction, in ULPs, thus potentially allowing the
3425 compiler to use a more efficient but less accurate method of computing
3426 it. ULP is defined as follows:
3428 If ``x`` is a real number that lies between two finite consecutive
3429 floating-point numbers ``a`` and ``b``, without being equal to one
3430 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
3431 distance between the two non-equal finite floating-point numbers
3432 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
3434 The metadata node shall consist of a single positive floating point
3435 number representing the maximum relative error, for example:
3437 .. code-block:: llvm
3439 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3443 '``range``' Metadata
3444 ^^^^^^^^^^^^^^^^^^^^
3446 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
3447 integer types. It expresses the possible ranges the loaded value or the value
3448 returned by the called function at this call site is in. The ranges are
3449 represented with a flattened list of integers. The loaded value or the value
3450 returned is known to be in the union of the ranges defined by each consecutive
3451 pair. Each pair has the following properties:
3453 - The type must match the type loaded by the instruction.
3454 - The pair ``a,b`` represents the range ``[a,b)``.
3455 - Both ``a`` and ``b`` are constants.
3456 - The range is allowed to wrap.
3457 - The range should not represent the full or empty set. That is,
3460 In addition, the pairs must be in signed order of the lower bound and
3461 they must be non-contiguous.
3465 .. code-block:: llvm
3467 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
3468 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3469 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
3470 %d = invoke i8 @bar() to label %cont
3471 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
3473 !0 = !{ i8 0, i8 2 }
3474 !1 = !{ i8 255, i8 2 }
3475 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
3476 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
3481 It is sometimes useful to attach information to loop constructs. Currently,
3482 loop metadata is implemented as metadata attached to the branch instruction
3483 in the loop latch block. This type of metadata refer to a metadata node that is
3484 guaranteed to be separate for each loop. The loop identifier metadata is
3485 specified with the name ``llvm.loop``.
3487 The loop identifier metadata is implemented using a metadata that refers to
3488 itself to avoid merging it with any other identifier metadata, e.g.,
3489 during module linkage or function inlining. That is, each loop should refer
3490 to their own identification metadata even if they reside in separate functions.
3491 The following example contains loop identifier metadata for two separate loop
3494 .. code-block:: llvm
3499 The loop identifier metadata can be used to specify additional
3500 per-loop metadata. Any operands after the first operand can be treated
3501 as user-defined metadata. For example the ``llvm.loop.unroll.count``
3502 suggests an unroll factor to the loop unroller:
3504 .. code-block:: llvm
3506 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
3509 !1 = !{!"llvm.loop.unroll.count", i32 4}
3511 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
3512 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3514 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
3515 used to control per-loop vectorization and interleaving parameters such as
3516 vectorization width and interleave count. These metadata should be used in
3517 conjunction with ``llvm.loop`` loop identification metadata. The
3518 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
3519 optimization hints and the optimizer will only interleave and vectorize loops if
3520 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
3521 which contains information about loop-carried memory dependencies can be helpful
3522 in determining the safety of these transformations.
3524 '``llvm.loop.interleave.count``' Metadata
3525 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3527 This metadata suggests an interleave count to the loop interleaver.
3528 The first operand is the string ``llvm.loop.interleave.count`` and the
3529 second operand is an integer specifying the interleave count. For
3532 .. code-block:: llvm
3534 !0 = !{!"llvm.loop.interleave.count", i32 4}
3536 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
3537 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
3538 then the interleave count will be determined automatically.
3540 '``llvm.loop.vectorize.enable``' Metadata
3541 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3543 This metadata selectively enables or disables vectorization for the loop. The
3544 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
3545 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
3546 0 disables vectorization:
3548 .. code-block:: llvm
3550 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
3551 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
3553 '``llvm.loop.vectorize.width``' Metadata
3554 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3556 This metadata sets the target width of the vectorizer. The first
3557 operand is the string ``llvm.loop.vectorize.width`` and the second
3558 operand is an integer specifying the width. For example:
3560 .. code-block:: llvm
3562 !0 = !{!"llvm.loop.vectorize.width", i32 4}
3564 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
3565 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
3566 0 or if the loop does not have this metadata the width will be
3567 determined automatically.
3569 '``llvm.loop.unroll``'
3570 ^^^^^^^^^^^^^^^^^^^^^^
3572 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
3573 optimization hints such as the unroll factor. ``llvm.loop.unroll``
3574 metadata should be used in conjunction with ``llvm.loop`` loop
3575 identification metadata. The ``llvm.loop.unroll`` metadata are only
3576 optimization hints and the unrolling will only be performed if the
3577 optimizer believes it is safe to do so.
3579 '``llvm.loop.unroll.count``' Metadata
3580 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3582 This metadata suggests an unroll factor to the loop unroller. The
3583 first operand is the string ``llvm.loop.unroll.count`` and the second
3584 operand is a positive integer specifying the unroll factor. For
3587 .. code-block:: llvm
3589 !0 = !{!"llvm.loop.unroll.count", i32 4}
3591 If the trip count of the loop is less than the unroll count the loop
3592 will be partially unrolled.
3594 '``llvm.loop.unroll.disable``' Metadata
3595 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3597 This metadata either disables loop unrolling. The metadata has a single operand
3598 which is the string ``llvm.loop.unroll.disable``. For example:
3600 .. code-block:: llvm
3602 !0 = !{!"llvm.loop.unroll.disable"}
3604 '``llvm.loop.unroll.runtime.disable``' Metadata
3605 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3607 This metadata either disables runtime loop unrolling. The metadata has a single
3608 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
3610 .. code-block:: llvm
3612 !0 = !{!"llvm.loop.unroll.runtime.disable"}
3614 '``llvm.loop.unroll.full``' Metadata
3615 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3617 This metadata either suggests that the loop should be unrolled fully. The
3618 metadata has a single operand which is the string ``llvm.loop.unroll.disable``.
3621 .. code-block:: llvm
3623 !0 = !{!"llvm.loop.unroll.full"}
3628 Metadata types used to annotate memory accesses with information helpful
3629 for optimizations are prefixed with ``llvm.mem``.
3631 '``llvm.mem.parallel_loop_access``' Metadata
3632 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3634 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
3635 or metadata containing a list of loop identifiers for nested loops.
3636 The metadata is attached to memory accessing instructions and denotes that
3637 no loop carried memory dependence exist between it and other instructions denoted
3638 with the same loop identifier.
3640 Precisely, given two instructions ``m1`` and ``m2`` that both have the
3641 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
3642 set of loops associated with that metadata, respectively, then there is no loop
3643 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
3646 As a special case, if all memory accessing instructions in a loop have
3647 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
3648 loop has no loop carried memory dependences and is considered to be a parallel
3651 Note that if not all memory access instructions have such metadata referring to
3652 the loop, then the loop is considered not being trivially parallel. Additional
3653 memory dependence analysis is required to make that determination. As a fail
3654 safe mechanism, this causes loops that were originally parallel to be considered
3655 sequential (if optimization passes that are unaware of the parallel semantics
3656 insert new memory instructions into the loop body).
3658 Example of a loop that is considered parallel due to its correct use of
3659 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
3660 metadata types that refer to the same loop identifier metadata.
3662 .. code-block:: llvm
3666 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
3668 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3670 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
3676 It is also possible to have nested parallel loops. In that case the
3677 memory accesses refer to a list of loop identifier metadata nodes instead of
3678 the loop identifier metadata node directly:
3680 .. code-block:: llvm
3684 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
3686 br label %inner.for.body
3690 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3692 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
3694 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
3698 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
3700 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
3702 outer.for.end: ; preds = %for.body
3704 !0 = !{!1, !2} ; a list of loop identifiers
3705 !1 = !{!1} ; an identifier for the inner loop
3706 !2 = !{!2} ; an identifier for the outer loop
3711 The ``llvm.bitsets`` global metadata is used to implement
3712 :doc:`bitsets <BitSets>`.
3714 Module Flags Metadata
3715 =====================
3717 Information about the module as a whole is difficult to convey to LLVM's
3718 subsystems. The LLVM IR isn't sufficient to transmit this information.
3719 The ``llvm.module.flags`` named metadata exists in order to facilitate
3720 this. These flags are in the form of key / value pairs --- much like a
3721 dictionary --- making it easy for any subsystem who cares about a flag to
3724 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
3725 Each triplet has the following form:
3727 - The first element is a *behavior* flag, which specifies the behavior
3728 when two (or more) modules are merged together, and it encounters two
3729 (or more) metadata with the same ID. The supported behaviors are
3731 - The second element is a metadata string that is a unique ID for the
3732 metadata. Each module may only have one flag entry for each unique ID (not
3733 including entries with the **Require** behavior).
3734 - The third element is the value of the flag.
3736 When two (or more) modules are merged together, the resulting
3737 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
3738 each unique metadata ID string, there will be exactly one entry in the merged
3739 modules ``llvm.module.flags`` metadata table, and the value for that entry will
3740 be determined by the merge behavior flag, as described below. The only exception
3741 is that entries with the *Require* behavior are always preserved.
3743 The following behaviors are supported:
3754 Emits an error if two values disagree, otherwise the resulting value
3755 is that of the operands.
3759 Emits a warning if two values disagree. The result value will be the
3760 operand for the flag from the first module being linked.
3764 Adds a requirement that another module flag be present and have a
3765 specified value after linking is performed. The value must be a
3766 metadata pair, where the first element of the pair is the ID of the
3767 module flag to be restricted, and the second element of the pair is
3768 the value the module flag should be restricted to. This behavior can
3769 be used to restrict the allowable results (via triggering of an
3770 error) of linking IDs with the **Override** behavior.
3774 Uses the specified value, regardless of the behavior or value of the
3775 other module. If both modules specify **Override**, but the values
3776 differ, an error will be emitted.
3780 Appends the two values, which are required to be metadata nodes.
3784 Appends the two values, which are required to be metadata
3785 nodes. However, duplicate entries in the second list are dropped
3786 during the append operation.
3788 It is an error for a particular unique flag ID to have multiple behaviors,
3789 except in the case of **Require** (which adds restrictions on another metadata
3790 value) or **Override**.
3792 An example of module flags:
3794 .. code-block:: llvm
3796 !0 = !{ i32 1, !"foo", i32 1 }
3797 !1 = !{ i32 4, !"bar", i32 37 }
3798 !2 = !{ i32 2, !"qux", i32 42 }
3799 !3 = !{ i32 3, !"qux",
3804 !llvm.module.flags = !{ !0, !1, !2, !3 }
3806 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3807 if two or more ``!"foo"`` flags are seen is to emit an error if their
3808 values are not equal.
3810 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3811 behavior if two or more ``!"bar"`` flags are seen is to use the value
3814 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3815 behavior if two or more ``!"qux"`` flags are seen is to emit a
3816 warning if their values are not equal.
3818 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3824 The behavior is to emit an error if the ``llvm.module.flags`` does not
3825 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3828 Objective-C Garbage Collection Module Flags Metadata
3829 ----------------------------------------------------
3831 On the Mach-O platform, Objective-C stores metadata about garbage
3832 collection in a special section called "image info". The metadata
3833 consists of a version number and a bitmask specifying what types of
3834 garbage collection are supported (if any) by the file. If two or more
3835 modules are linked together their garbage collection metadata needs to
3836 be merged rather than appended together.
3838 The Objective-C garbage collection module flags metadata consists of the
3839 following key-value pairs:
3848 * - ``Objective-C Version``
3849 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3851 * - ``Objective-C Image Info Version``
3852 - **[Required]** --- The version of the image info section. Currently
3855 * - ``Objective-C Image Info Section``
3856 - **[Required]** --- The section to place the metadata. Valid values are
3857 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3858 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3859 Objective-C ABI version 2.
3861 * - ``Objective-C Garbage Collection``
3862 - **[Required]** --- Specifies whether garbage collection is supported or
3863 not. Valid values are 0, for no garbage collection, and 2, for garbage
3864 collection supported.
3866 * - ``Objective-C GC Only``
3867 - **[Optional]** --- Specifies that only garbage collection is supported.
3868 If present, its value must be 6. This flag requires that the
3869 ``Objective-C Garbage Collection`` flag have the value 2.
3871 Some important flag interactions:
3873 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3874 merged with a module with ``Objective-C Garbage Collection`` set to
3875 2, then the resulting module has the
3876 ``Objective-C Garbage Collection`` flag set to 0.
3877 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3878 merged with a module with ``Objective-C GC Only`` set to 6.
3880 Automatic Linker Flags Module Flags Metadata
3881 --------------------------------------------
3883 Some targets support embedding flags to the linker inside individual object
3884 files. Typically this is used in conjunction with language extensions which
3885 allow source files to explicitly declare the libraries they depend on, and have
3886 these automatically be transmitted to the linker via object files.
3888 These flags are encoded in the IR using metadata in the module flags section,
3889 using the ``Linker Options`` key. The merge behavior for this flag is required
3890 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3891 node which should be a list of other metadata nodes, each of which should be a
3892 list of metadata strings defining linker options.
3894 For example, the following metadata section specifies two separate sets of
3895 linker options, presumably to link against ``libz`` and the ``Cocoa``
3898 !0 = !{ i32 6, !"Linker Options",
3901 !{ !"-framework", !"Cocoa" } } }
3902 !llvm.module.flags = !{ !0 }
3904 The metadata encoding as lists of lists of options, as opposed to a collapsed
3905 list of options, is chosen so that the IR encoding can use multiple option
3906 strings to specify e.g., a single library, while still having that specifier be
3907 preserved as an atomic element that can be recognized by a target specific
3908 assembly writer or object file emitter.
3910 Each individual option is required to be either a valid option for the target's
3911 linker, or an option that is reserved by the target specific assembly writer or
3912 object file emitter. No other aspect of these options is defined by the IR.
3914 C type width Module Flags Metadata
3915 ----------------------------------
3917 The ARM backend emits a section into each generated object file describing the
3918 options that it was compiled with (in a compiler-independent way) to prevent
3919 linking incompatible objects, and to allow automatic library selection. Some
3920 of these options are not visible at the IR level, namely wchar_t width and enum
3923 To pass this information to the backend, these options are encoded in module
3924 flags metadata, using the following key-value pairs:
3934 - * 0 --- sizeof(wchar_t) == 4
3935 * 1 --- sizeof(wchar_t) == 2
3938 - * 0 --- Enums are at least as large as an ``int``.
3939 * 1 --- Enums are stored in the smallest integer type which can
3940 represent all of its values.
3942 For example, the following metadata section specifies that the module was
3943 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
3944 enum is the smallest type which can represent all of its values::
3946 !llvm.module.flags = !{!0, !1}
3947 !0 = !{i32 1, !"short_wchar", i32 1}
3948 !1 = !{i32 1, !"short_enum", i32 0}
3950 .. _intrinsicglobalvariables:
3952 Intrinsic Global Variables
3953 ==========================
3955 LLVM has a number of "magic" global variables that contain data that
3956 affect code generation or other IR semantics. These are documented here.
3957 All globals of this sort should have a section specified as
3958 "``llvm.metadata``". This section and all globals that start with
3959 "``llvm.``" are reserved for use by LLVM.
3963 The '``llvm.used``' Global Variable
3964 -----------------------------------
3966 The ``@llvm.used`` global is an array which has
3967 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3968 pointers to named global variables, functions and aliases which may optionally
3969 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3972 .. code-block:: llvm
3977 @llvm.used = appending global [2 x i8*] [
3979 i8* bitcast (i32* @Y to i8*)
3980 ], section "llvm.metadata"
3982 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3983 and linker are required to treat the symbol as if there is a reference to the
3984 symbol that it cannot see (which is why they have to be named). For example, if
3985 a variable has internal linkage and no references other than that from the
3986 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3987 references from inline asms and other things the compiler cannot "see", and
3988 corresponds to "``attribute((used))``" in GNU C.
3990 On some targets, the code generator must emit a directive to the
3991 assembler or object file to prevent the assembler and linker from
3992 molesting the symbol.
3994 .. _gv_llvmcompilerused:
3996 The '``llvm.compiler.used``' Global Variable
3997 --------------------------------------------
3999 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4000 directive, except that it only prevents the compiler from touching the
4001 symbol. On targets that support it, this allows an intelligent linker to
4002 optimize references to the symbol without being impeded as it would be
4005 This is a rare construct that should only be used in rare circumstances,
4006 and should not be exposed to source languages.
4008 .. _gv_llvmglobalctors:
4010 The '``llvm.global_ctors``' Global Variable
4011 -------------------------------------------
4013 .. code-block:: llvm
4015 %0 = type { i32, void ()*, i8* }
4016 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4018 The ``@llvm.global_ctors`` array contains a list of constructor
4019 functions, priorities, and an optional associated global or function.
4020 The functions referenced by this array will be called in ascending order
4021 of priority (i.e. lowest first) when the module is loaded. The order of
4022 functions with the same priority is not defined.
4024 If the third field is present, non-null, and points to a global variable
4025 or function, the initializer function will only run if the associated
4026 data from the current module is not discarded.
4028 .. _llvmglobaldtors:
4030 The '``llvm.global_dtors``' Global Variable
4031 -------------------------------------------
4033 .. code-block:: llvm
4035 %0 = type { i32, void ()*, i8* }
4036 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4038 The ``@llvm.global_dtors`` array contains a list of destructor
4039 functions, priorities, and an optional associated global or function.
4040 The functions referenced by this array will be called in descending
4041 order of priority (i.e. highest first) when the module is unloaded. The
4042 order of functions with the same priority is not defined.
4044 If the third field is present, non-null, and points to a global variable
4045 or function, the destructor function will only run if the associated
4046 data from the current module is not discarded.
4048 Instruction Reference
4049 =====================
4051 The LLVM instruction set consists of several different classifications
4052 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4053 instructions <binaryops>`, :ref:`bitwise binary
4054 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4055 :ref:`other instructions <otherops>`.
4059 Terminator Instructions
4060 -----------------------
4062 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4063 program ends with a "Terminator" instruction, which indicates which
4064 block should be executed after the current block is finished. These
4065 terminator instructions typically yield a '``void``' value: they produce
4066 control flow, not values (the one exception being the
4067 ':ref:`invoke <i_invoke>`' instruction).
4069 The terminator instructions are: ':ref:`ret <i_ret>`',
4070 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4071 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4072 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
4076 '``ret``' Instruction
4077 ^^^^^^^^^^^^^^^^^^^^^
4084 ret <type> <value> ; Return a value from a non-void function
4085 ret void ; Return from void function
4090 The '``ret``' instruction is used to return control flow (and optionally
4091 a value) from a function back to the caller.
4093 There are two forms of the '``ret``' instruction: one that returns a
4094 value and then causes control flow, and one that just causes control
4100 The '``ret``' instruction optionally accepts a single argument, the
4101 return value. The type of the return value must be a ':ref:`first
4102 class <t_firstclass>`' type.
4104 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4105 return type and contains a '``ret``' instruction with no return value or
4106 a return value with a type that does not match its type, or if it has a
4107 void return type and contains a '``ret``' instruction with a return
4113 When the '``ret``' instruction is executed, control flow returns back to
4114 the calling function's context. If the caller is a
4115 ":ref:`call <i_call>`" instruction, execution continues at the
4116 instruction after the call. If the caller was an
4117 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4118 beginning of the "normal" destination block. If the instruction returns
4119 a value, that value shall set the call or invoke instruction's return
4125 .. code-block:: llvm
4127 ret i32 5 ; Return an integer value of 5
4128 ret void ; Return from a void function
4129 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4133 '``br``' Instruction
4134 ^^^^^^^^^^^^^^^^^^^^
4141 br i1 <cond>, label <iftrue>, label <iffalse>
4142 br label <dest> ; Unconditional branch
4147 The '``br``' instruction is used to cause control flow to transfer to a
4148 different basic block in the current function. There are two forms of
4149 this instruction, corresponding to a conditional branch and an
4150 unconditional branch.
4155 The conditional branch form of the '``br``' instruction takes a single
4156 '``i1``' value and two '``label``' values. The unconditional form of the
4157 '``br``' instruction takes a single '``label``' value as a target.
4162 Upon execution of a conditional '``br``' instruction, the '``i1``'
4163 argument is evaluated. If the value is ``true``, control flows to the
4164 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4165 to the '``iffalse``' ``label`` argument.
4170 .. code-block:: llvm
4173 %cond = icmp eq i32 %a, %b
4174 br i1 %cond, label %IfEqual, label %IfUnequal
4182 '``switch``' Instruction
4183 ^^^^^^^^^^^^^^^^^^^^^^^^
4190 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
4195 The '``switch``' instruction is used to transfer control flow to one of
4196 several different places. It is a generalization of the '``br``'
4197 instruction, allowing a branch to occur to one of many possible
4203 The '``switch``' instruction uses three parameters: an integer
4204 comparison value '``value``', a default '``label``' destination, and an
4205 array of pairs of comparison value constants and '``label``'s. The table
4206 is not allowed to contain duplicate constant entries.
4211 The ``switch`` instruction specifies a table of values and destinations.
4212 When the '``switch``' instruction is executed, this table is searched
4213 for the given value. If the value is found, control flow is transferred
4214 to the corresponding destination; otherwise, control flow is transferred
4215 to the default destination.
4220 Depending on properties of the target machine and the particular
4221 ``switch`` instruction, this instruction may be code generated in
4222 different ways. For example, it could be generated as a series of
4223 chained conditional branches or with a lookup table.
4228 .. code-block:: llvm
4230 ; Emulate a conditional br instruction
4231 %Val = zext i1 %value to i32
4232 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
4234 ; Emulate an unconditional br instruction
4235 switch i32 0, label %dest [ ]
4237 ; Implement a jump table:
4238 switch i32 %val, label %otherwise [ i32 0, label %onzero
4240 i32 2, label %ontwo ]
4244 '``indirectbr``' Instruction
4245 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4252 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
4257 The '``indirectbr``' instruction implements an indirect branch to a
4258 label within the current function, whose address is specified by
4259 "``address``". Address must be derived from a
4260 :ref:`blockaddress <blockaddress>` constant.
4265 The '``address``' argument is the address of the label to jump to. The
4266 rest of the arguments indicate the full set of possible destinations
4267 that the address may point to. Blocks are allowed to occur multiple
4268 times in the destination list, though this isn't particularly useful.
4270 This destination list is required so that dataflow analysis has an
4271 accurate understanding of the CFG.
4276 Control transfers to the block specified in the address argument. All
4277 possible destination blocks must be listed in the label list, otherwise
4278 this instruction has undefined behavior. This implies that jumps to
4279 labels defined in other functions have undefined behavior as well.
4284 This is typically implemented with a jump through a register.
4289 .. code-block:: llvm
4291 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
4295 '``invoke``' Instruction
4296 ^^^^^^^^^^^^^^^^^^^^^^^^
4303 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
4304 to label <normal label> unwind label <exception label>
4309 The '``invoke``' instruction causes control to transfer to a specified
4310 function, with the possibility of control flow transfer to either the
4311 '``normal``' label or the '``exception``' label. If the callee function
4312 returns with the "``ret``" instruction, control flow will return to the
4313 "normal" label. If the callee (or any indirect callees) returns via the
4314 ":ref:`resume <i_resume>`" instruction or other exception handling
4315 mechanism, control is interrupted and continued at the dynamically
4316 nearest "exception" label.
4318 The '``exception``' label is a `landing
4319 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
4320 '``exception``' label is required to have the
4321 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
4322 information about the behavior of the program after unwinding happens,
4323 as its first non-PHI instruction. The restrictions on the
4324 "``landingpad``" instruction's tightly couples it to the "``invoke``"
4325 instruction, so that the important information contained within the
4326 "``landingpad``" instruction can't be lost through normal code motion.
4331 This instruction requires several arguments:
4333 #. The optional "cconv" marker indicates which :ref:`calling
4334 convention <callingconv>` the call should use. If none is
4335 specified, the call defaults to using C calling conventions.
4336 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
4337 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
4339 #. '``ptr to function ty``': shall be the signature of the pointer to
4340 function value being invoked. In most cases, this is a direct
4341 function invocation, but indirect ``invoke``'s are just as possible,
4342 branching off an arbitrary pointer to function value.
4343 #. '``function ptr val``': An LLVM value containing a pointer to a
4344 function to be invoked.
4345 #. '``function args``': argument list whose types match the function
4346 signature argument types and parameter attributes. All arguments must
4347 be of :ref:`first class <t_firstclass>` type. If the function signature
4348 indicates the function accepts a variable number of arguments, the
4349 extra arguments can be specified.
4350 #. '``normal label``': the label reached when the called function
4351 executes a '``ret``' instruction.
4352 #. '``exception label``': the label reached when a callee returns via
4353 the :ref:`resume <i_resume>` instruction or other exception handling
4355 #. The optional :ref:`function attributes <fnattrs>` list. Only
4356 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
4357 attributes are valid here.
4362 This instruction is designed to operate as a standard '``call``'
4363 instruction in most regards. The primary difference is that it
4364 establishes an association with a label, which is used by the runtime
4365 library to unwind the stack.
4367 This instruction is used in languages with destructors to ensure that
4368 proper cleanup is performed in the case of either a ``longjmp`` or a
4369 thrown exception. Additionally, this is important for implementation of
4370 '``catch``' clauses in high-level languages that support them.
4372 For the purposes of the SSA form, the definition of the value returned
4373 by the '``invoke``' instruction is deemed to occur on the edge from the
4374 current block to the "normal" label. If the callee unwinds then no
4375 return value is available.
4380 .. code-block:: llvm
4382 %retval = invoke i32 @Test(i32 15) to label %Continue
4383 unwind label %TestCleanup ; i32:retval set
4384 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
4385 unwind label %TestCleanup ; i32:retval set
4389 '``resume``' Instruction
4390 ^^^^^^^^^^^^^^^^^^^^^^^^
4397 resume <type> <value>
4402 The '``resume``' instruction is a terminator instruction that has no
4408 The '``resume``' instruction requires one argument, which must have the
4409 same type as the result of any '``landingpad``' instruction in the same
4415 The '``resume``' instruction resumes propagation of an existing
4416 (in-flight) exception whose unwinding was interrupted with a
4417 :ref:`landingpad <i_landingpad>` instruction.
4422 .. code-block:: llvm
4424 resume { i8*, i32 } %exn
4428 '``unreachable``' Instruction
4429 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4441 The '``unreachable``' instruction has no defined semantics. This
4442 instruction is used to inform the optimizer that a particular portion of
4443 the code is not reachable. This can be used to indicate that the code
4444 after a no-return function cannot be reached, and other facts.
4449 The '``unreachable``' instruction has no defined semantics.
4456 Binary operators are used to do most of the computation in a program.
4457 They require two operands of the same type, execute an operation on
4458 them, and produce a single value. The operands might represent multiple
4459 data, as is the case with the :ref:`vector <t_vector>` data type. The
4460 result value has the same type as its operands.
4462 There are several different binary operators:
4466 '``add``' Instruction
4467 ^^^^^^^^^^^^^^^^^^^^^
4474 <result> = add <ty> <op1>, <op2> ; yields ty:result
4475 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
4476 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
4477 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
4482 The '``add``' instruction returns the sum of its two operands.
4487 The two arguments to the '``add``' instruction must be
4488 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4489 arguments must have identical types.
4494 The value produced is the integer sum of the two operands.
4496 If the sum has unsigned overflow, the result returned is the
4497 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4500 Because LLVM integers use a two's complement representation, this
4501 instruction is appropriate for both signed and unsigned integers.
4503 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4504 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4505 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
4506 unsigned and/or signed overflow, respectively, occurs.
4511 .. code-block:: llvm
4513 <result> = add i32 4, %var ; yields i32:result = 4 + %var
4517 '``fadd``' Instruction
4518 ^^^^^^^^^^^^^^^^^^^^^^
4525 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4530 The '``fadd``' instruction returns the sum of its two operands.
4535 The two arguments to the '``fadd``' instruction must be :ref:`floating
4536 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4537 Both arguments must have identical types.
4542 The value produced is the floating point sum of the two operands. This
4543 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
4544 which are optimization hints to enable otherwise unsafe floating point
4550 .. code-block:: llvm
4552 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
4554 '``sub``' Instruction
4555 ^^^^^^^^^^^^^^^^^^^^^
4562 <result> = sub <ty> <op1>, <op2> ; yields ty:result
4563 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
4564 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
4565 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
4570 The '``sub``' instruction returns the difference of its two operands.
4572 Note that the '``sub``' instruction is used to represent the '``neg``'
4573 instruction present in most other intermediate representations.
4578 The two arguments to the '``sub``' instruction must be
4579 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4580 arguments must have identical types.
4585 The value produced is the integer difference of the two operands.
4587 If the difference has unsigned overflow, the result returned is the
4588 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4591 Because LLVM integers use a two's complement representation, this
4592 instruction is appropriate for both signed and unsigned integers.
4594 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4595 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4596 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
4597 unsigned and/or signed overflow, respectively, occurs.
4602 .. code-block:: llvm
4604 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
4605 <result> = sub i32 0, %val ; yields i32:result = -%var
4609 '``fsub``' Instruction
4610 ^^^^^^^^^^^^^^^^^^^^^^
4617 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4622 The '``fsub``' instruction returns the difference of its two operands.
4624 Note that the '``fsub``' instruction is used to represent the '``fneg``'
4625 instruction present in most other intermediate representations.
4630 The two arguments to the '``fsub``' instruction must be :ref:`floating
4631 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4632 Both arguments must have identical types.
4637 The value produced is the floating point difference of the two operands.
4638 This instruction can also take any number of :ref:`fast-math
4639 flags <fastmath>`, which are optimization hints to enable otherwise
4640 unsafe floating point optimizations:
4645 .. code-block:: llvm
4647 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
4648 <result> = fsub float -0.0, %val ; yields float:result = -%var
4650 '``mul``' Instruction
4651 ^^^^^^^^^^^^^^^^^^^^^
4658 <result> = mul <ty> <op1>, <op2> ; yields ty:result
4659 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
4660 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
4661 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
4666 The '``mul``' instruction returns the product of its two operands.
4671 The two arguments to the '``mul``' instruction must be
4672 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4673 arguments must have identical types.
4678 The value produced is the integer product of the two operands.
4680 If the result of the multiplication has unsigned overflow, the result
4681 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
4682 bit width of the result.
4684 Because LLVM integers use a two's complement representation, and the
4685 result is the same width as the operands, this instruction returns the
4686 correct result for both signed and unsigned integers. If a full product
4687 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
4688 sign-extended or zero-extended as appropriate to the width of the full
4691 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4692 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4693 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
4694 unsigned and/or signed overflow, respectively, occurs.
4699 .. code-block:: llvm
4701 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
4705 '``fmul``' Instruction
4706 ^^^^^^^^^^^^^^^^^^^^^^
4713 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4718 The '``fmul``' instruction returns the product of its two operands.
4723 The two arguments to the '``fmul``' instruction must be :ref:`floating
4724 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4725 Both arguments must have identical types.
4730 The value produced is the floating point product of the two operands.
4731 This instruction can also take any number of :ref:`fast-math
4732 flags <fastmath>`, which are optimization hints to enable otherwise
4733 unsafe floating point optimizations:
4738 .. code-block:: llvm
4740 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
4742 '``udiv``' Instruction
4743 ^^^^^^^^^^^^^^^^^^^^^^
4750 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
4751 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
4756 The '``udiv``' instruction returns the quotient of its two operands.
4761 The two arguments to the '``udiv``' instruction must be
4762 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4763 arguments must have identical types.
4768 The value produced is the unsigned integer quotient of the two operands.
4770 Note that unsigned integer division and signed integer division are
4771 distinct operations; for signed integer division, use '``sdiv``'.
4773 Division by zero leads to undefined behavior.
4775 If the ``exact`` keyword is present, the result value of the ``udiv`` is
4776 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
4777 such, "((a udiv exact b) mul b) == a").
4782 .. code-block:: llvm
4784 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
4786 '``sdiv``' Instruction
4787 ^^^^^^^^^^^^^^^^^^^^^^
4794 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
4795 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
4800 The '``sdiv``' instruction returns the quotient of its two operands.
4805 The two arguments to the '``sdiv``' instruction must be
4806 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4807 arguments must have identical types.
4812 The value produced is the signed integer quotient of the two operands
4813 rounded towards zero.
4815 Note that signed integer division and unsigned integer division are
4816 distinct operations; for unsigned integer division, use '``udiv``'.
4818 Division by zero leads to undefined behavior. Overflow also leads to
4819 undefined behavior; this is a rare case, but can occur, for example, by
4820 doing a 32-bit division of -2147483648 by -1.
4822 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4823 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4828 .. code-block:: llvm
4830 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
4834 '``fdiv``' Instruction
4835 ^^^^^^^^^^^^^^^^^^^^^^
4842 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4847 The '``fdiv``' instruction returns the quotient of its two operands.
4852 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4853 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4854 Both arguments must have identical types.
4859 The value produced is the floating point quotient of the two operands.
4860 This instruction can also take any number of :ref:`fast-math
4861 flags <fastmath>`, which are optimization hints to enable otherwise
4862 unsafe floating point optimizations:
4867 .. code-block:: llvm
4869 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
4871 '``urem``' Instruction
4872 ^^^^^^^^^^^^^^^^^^^^^^
4879 <result> = urem <ty> <op1>, <op2> ; yields ty:result
4884 The '``urem``' instruction returns the remainder from the unsigned
4885 division of its two arguments.
4890 The two arguments to the '``urem``' instruction must be
4891 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4892 arguments must have identical types.
4897 This instruction returns the unsigned integer *remainder* of a division.
4898 This instruction always performs an unsigned division to get the
4901 Note that unsigned integer remainder and signed integer remainder are
4902 distinct operations; for signed integer remainder, use '``srem``'.
4904 Taking the remainder of a division by zero leads to undefined behavior.
4909 .. code-block:: llvm
4911 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
4913 '``srem``' Instruction
4914 ^^^^^^^^^^^^^^^^^^^^^^
4921 <result> = srem <ty> <op1>, <op2> ; yields ty:result
4926 The '``srem``' instruction returns the remainder from the signed
4927 division of its two operands. This instruction can also take
4928 :ref:`vector <t_vector>` versions of the values in which case the elements
4934 The two arguments to the '``srem``' instruction must be
4935 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4936 arguments must have identical types.
4941 This instruction returns the *remainder* of a division (where the result
4942 is either zero or has the same sign as the dividend, ``op1``), not the
4943 *modulo* operator (where the result is either zero or has the same sign
4944 as the divisor, ``op2``) of a value. For more information about the
4945 difference, see `The Math
4946 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4947 table of how this is implemented in various languages, please see
4949 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4951 Note that signed integer remainder and unsigned integer remainder are
4952 distinct operations; for unsigned integer remainder, use '``urem``'.
4954 Taking the remainder of a division by zero leads to undefined behavior.
4955 Overflow also leads to undefined behavior; this is a rare case, but can
4956 occur, for example, by taking the remainder of a 32-bit division of
4957 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4958 rule lets srem be implemented using instructions that return both the
4959 result of the division and the remainder.)
4964 .. code-block:: llvm
4966 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
4970 '``frem``' Instruction
4971 ^^^^^^^^^^^^^^^^^^^^^^
4978 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4983 The '``frem``' instruction returns the remainder from the division of
4989 The two arguments to the '``frem``' instruction must be :ref:`floating
4990 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4991 Both arguments must have identical types.
4996 This instruction returns the *remainder* of a division. The remainder
4997 has the same sign as the dividend. This instruction can also take any
4998 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4999 to enable otherwise unsafe floating point optimizations:
5004 .. code-block:: llvm
5006 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
5010 Bitwise Binary Operations
5011 -------------------------
5013 Bitwise binary operators are used to do various forms of bit-twiddling
5014 in a program. They are generally very efficient instructions and can
5015 commonly be strength reduced from other instructions. They require two
5016 operands of the same type, execute an operation on them, and produce a
5017 single value. The resulting value is the same type as its operands.
5019 '``shl``' Instruction
5020 ^^^^^^^^^^^^^^^^^^^^^
5027 <result> = shl <ty> <op1>, <op2> ; yields ty:result
5028 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
5029 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
5030 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
5035 The '``shl``' instruction returns the first operand shifted to the left
5036 a specified number of bits.
5041 Both arguments to the '``shl``' instruction must be the same
5042 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5043 '``op2``' is treated as an unsigned value.
5048 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
5049 where ``n`` is the width of the result. If ``op2`` is (statically or
5050 dynamically) negative or equal to or larger than the number of bits in
5051 ``op1``, the result is undefined. If the arguments are vectors, each
5052 vector element of ``op1`` is shifted by the corresponding shift amount
5055 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
5056 value <poisonvalues>` if it shifts out any non-zero bits. If the
5057 ``nsw`` keyword is present, then the shift produces a :ref:`poison
5058 value <poisonvalues>` if it shifts out any bits that disagree with the
5059 resultant sign bit. As such, NUW/NSW have the same semantics as they
5060 would if the shift were expressed as a mul instruction with the same
5061 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
5066 .. code-block:: llvm
5068 <result> = shl i32 4, %var ; yields i32: 4 << %var
5069 <result> = shl i32 4, 2 ; yields i32: 16
5070 <result> = shl i32 1, 10 ; yields i32: 1024
5071 <result> = shl i32 1, 32 ; undefined
5072 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
5074 '``lshr``' Instruction
5075 ^^^^^^^^^^^^^^^^^^^^^^
5082 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
5083 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
5088 The '``lshr``' instruction (logical shift right) returns the first
5089 operand shifted to the right a specified number of bits with zero fill.
5094 Both arguments to the '``lshr``' instruction must be the same
5095 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5096 '``op2``' is treated as an unsigned value.
5101 This instruction always performs a logical shift right operation. The
5102 most significant bits of the result will be filled with zero bits after
5103 the shift. If ``op2`` is (statically or dynamically) equal to or larger
5104 than the number of bits in ``op1``, the result is undefined. If the
5105 arguments are vectors, each vector element of ``op1`` is shifted by the
5106 corresponding shift amount in ``op2``.
5108 If the ``exact`` keyword is present, the result value of the ``lshr`` is
5109 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5115 .. code-block:: llvm
5117 <result> = lshr i32 4, 1 ; yields i32:result = 2
5118 <result> = lshr i32 4, 2 ; yields i32:result = 1
5119 <result> = lshr i8 4, 3 ; yields i8:result = 0
5120 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
5121 <result> = lshr i32 1, 32 ; undefined
5122 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
5124 '``ashr``' Instruction
5125 ^^^^^^^^^^^^^^^^^^^^^^
5132 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
5133 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
5138 The '``ashr``' instruction (arithmetic shift right) returns the first
5139 operand shifted to the right a specified number of bits with sign
5145 Both arguments to the '``ashr``' instruction must be the same
5146 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5147 '``op2``' is treated as an unsigned value.
5152 This instruction always performs an arithmetic shift right operation,
5153 The most significant bits of the result will be filled with the sign bit
5154 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
5155 than the number of bits in ``op1``, the result is undefined. If the
5156 arguments are vectors, each vector element of ``op1`` is shifted by the
5157 corresponding shift amount in ``op2``.
5159 If the ``exact`` keyword is present, the result value of the ``ashr`` is
5160 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5166 .. code-block:: llvm
5168 <result> = ashr i32 4, 1 ; yields i32:result = 2
5169 <result> = ashr i32 4, 2 ; yields i32:result = 1
5170 <result> = ashr i8 4, 3 ; yields i8:result = 0
5171 <result> = ashr i8 -2, 1 ; yields i8:result = -1
5172 <result> = ashr i32 1, 32 ; undefined
5173 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
5175 '``and``' Instruction
5176 ^^^^^^^^^^^^^^^^^^^^^
5183 <result> = and <ty> <op1>, <op2> ; yields ty:result
5188 The '``and``' instruction returns the bitwise logical and of its two
5194 The two arguments to the '``and``' instruction must be
5195 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5196 arguments must have identical types.
5201 The truth table used for the '``and``' instruction is:
5218 .. code-block:: llvm
5220 <result> = and i32 4, %var ; yields i32:result = 4 & %var
5221 <result> = and i32 15, 40 ; yields i32:result = 8
5222 <result> = and i32 4, 8 ; yields i32:result = 0
5224 '``or``' Instruction
5225 ^^^^^^^^^^^^^^^^^^^^
5232 <result> = or <ty> <op1>, <op2> ; yields ty:result
5237 The '``or``' instruction returns the bitwise logical inclusive or of its
5243 The two arguments to the '``or``' instruction must be
5244 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5245 arguments must have identical types.
5250 The truth table used for the '``or``' instruction is:
5269 <result> = or i32 4, %var ; yields i32:result = 4 | %var
5270 <result> = or i32 15, 40 ; yields i32:result = 47
5271 <result> = or i32 4, 8 ; yields i32:result = 12
5273 '``xor``' Instruction
5274 ^^^^^^^^^^^^^^^^^^^^^
5281 <result> = xor <ty> <op1>, <op2> ; yields ty:result
5286 The '``xor``' instruction returns the bitwise logical exclusive or of
5287 its two operands. The ``xor`` is used to implement the "one's
5288 complement" operation, which is the "~" operator in C.
5293 The two arguments to the '``xor``' instruction must be
5294 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5295 arguments must have identical types.
5300 The truth table used for the '``xor``' instruction is:
5317 .. code-block:: llvm
5319 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
5320 <result> = xor i32 15, 40 ; yields i32:result = 39
5321 <result> = xor i32 4, 8 ; yields i32:result = 12
5322 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
5327 LLVM supports several instructions to represent vector operations in a
5328 target-independent manner. These instructions cover the element-access
5329 and vector-specific operations needed to process vectors effectively.
5330 While LLVM does directly support these vector operations, many
5331 sophisticated algorithms will want to use target-specific intrinsics to
5332 take full advantage of a specific target.
5334 .. _i_extractelement:
5336 '``extractelement``' Instruction
5337 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5344 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
5349 The '``extractelement``' instruction extracts a single scalar element
5350 from a vector at a specified index.
5355 The first operand of an '``extractelement``' instruction is a value of
5356 :ref:`vector <t_vector>` type. The second operand is an index indicating
5357 the position from which to extract the element. The index may be a
5358 variable of any integer type.
5363 The result is a scalar of the same type as the element type of ``val``.
5364 Its value is the value at position ``idx`` of ``val``. If ``idx``
5365 exceeds the length of ``val``, the results are undefined.
5370 .. code-block:: llvm
5372 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
5374 .. _i_insertelement:
5376 '``insertelement``' Instruction
5377 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5384 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
5389 The '``insertelement``' instruction inserts a scalar element into a
5390 vector at a specified index.
5395 The first operand of an '``insertelement``' instruction is a value of
5396 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
5397 type must equal the element type of the first operand. The third operand
5398 is an index indicating the position at which to insert the value. The
5399 index may be a variable of any integer type.
5404 The result is a vector of the same type as ``val``. Its element values
5405 are those of ``val`` except at position ``idx``, where it gets the value
5406 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
5412 .. code-block:: llvm
5414 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
5416 .. _i_shufflevector:
5418 '``shufflevector``' Instruction
5419 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5426 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
5431 The '``shufflevector``' instruction constructs a permutation of elements
5432 from two input vectors, returning a vector with the same element type as
5433 the input and length that is the same as the shuffle mask.
5438 The first two operands of a '``shufflevector``' instruction are vectors
5439 with the same type. The third argument is a shuffle mask whose element
5440 type is always 'i32'. The result of the instruction is a vector whose
5441 length is the same as the shuffle mask and whose element type is the
5442 same as the element type of the first two operands.
5444 The shuffle mask operand is required to be a constant vector with either
5445 constant integer or undef values.
5450 The elements of the two input vectors are numbered from left to right
5451 across both of the vectors. The shuffle mask operand specifies, for each
5452 element of the result vector, which element of the two input vectors the
5453 result element gets. The element selector may be undef (meaning "don't
5454 care") and the second operand may be undef if performing a shuffle from
5460 .. code-block:: llvm
5462 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5463 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
5464 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
5465 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
5466 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
5467 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
5468 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5469 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
5471 Aggregate Operations
5472 --------------------
5474 LLVM supports several instructions for working with
5475 :ref:`aggregate <t_aggregate>` values.
5479 '``extractvalue``' Instruction
5480 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5487 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
5492 The '``extractvalue``' instruction extracts the value of a member field
5493 from an :ref:`aggregate <t_aggregate>` value.
5498 The first operand of an '``extractvalue``' instruction is a value of
5499 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
5500 constant indices to specify which value to extract in a similar manner
5501 as indices in a '``getelementptr``' instruction.
5503 The major differences to ``getelementptr`` indexing are:
5505 - Since the value being indexed is not a pointer, the first index is
5506 omitted and assumed to be zero.
5507 - At least one index must be specified.
5508 - Not only struct indices but also array indices must be in bounds.
5513 The result is the value at the position in the aggregate specified by
5519 .. code-block:: llvm
5521 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
5525 '``insertvalue``' Instruction
5526 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5533 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
5538 The '``insertvalue``' instruction inserts a value into a member field in
5539 an :ref:`aggregate <t_aggregate>` value.
5544 The first operand of an '``insertvalue``' instruction is a value of
5545 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
5546 a first-class value to insert. The following operands are constant
5547 indices indicating the position at which to insert the value in a
5548 similar manner as indices in a '``extractvalue``' instruction. The value
5549 to insert must have the same type as the value identified by the
5555 The result is an aggregate of the same type as ``val``. Its value is
5556 that of ``val`` except that the value at the position specified by the
5557 indices is that of ``elt``.
5562 .. code-block:: llvm
5564 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
5565 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
5566 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
5570 Memory Access and Addressing Operations
5571 ---------------------------------------
5573 A key design point of an SSA-based representation is how it represents
5574 memory. In LLVM, no memory locations are in SSA form, which makes things
5575 very simple. This section describes how to read, write, and allocate
5580 '``alloca``' Instruction
5581 ^^^^^^^^^^^^^^^^^^^^^^^^
5588 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
5593 The '``alloca``' instruction allocates memory on the stack frame of the
5594 currently executing function, to be automatically released when this
5595 function returns to its caller. The object is always allocated in the
5596 generic address space (address space zero).
5601 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
5602 bytes of memory on the runtime stack, returning a pointer of the
5603 appropriate type to the program. If "NumElements" is specified, it is
5604 the number of elements allocated, otherwise "NumElements" is defaulted
5605 to be one. If a constant alignment is specified, the value result of the
5606 allocation is guaranteed to be aligned to at least that boundary. The
5607 alignment may not be greater than ``1 << 29``. If not specified, or if
5608 zero, the target can choose to align the allocation on any convenient
5609 boundary compatible with the type.
5611 '``type``' may be any sized type.
5616 Memory is allocated; a pointer is returned. The operation is undefined
5617 if there is insufficient stack space for the allocation. '``alloca``'d
5618 memory is automatically released when the function returns. The
5619 '``alloca``' instruction is commonly used to represent automatic
5620 variables that must have an address available. When the function returns
5621 (either with the ``ret`` or ``resume`` instructions), the memory is
5622 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5623 The order in which memory is allocated (ie., which way the stack grows)
5629 .. code-block:: llvm
5631 %ptr = alloca i32 ; yields i32*:ptr
5632 %ptr = alloca i32, i32 4 ; yields i32*:ptr
5633 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
5634 %ptr = alloca i32, align 1024 ; yields i32*:ptr
5638 '``load``' Instruction
5639 ^^^^^^^^^^^^^^^^^^^^^^
5646 <result> = load [volatile] <ty>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !nonnull !<index>]
5647 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5648 !<index> = !{ i32 1 }
5653 The '``load``' instruction is used to read from memory.
5658 The argument to the ``load`` instruction specifies the memory address
5659 from which to load. The type specified must be a :ref:`first
5660 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
5661 then the optimizer is not allowed to modify the number or order of
5662 execution of this ``load`` with other :ref:`volatile
5663 operations <volatile>`.
5665 If the ``load`` is marked as ``atomic``, it takes an extra
5666 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5667 ``release`` and ``acq_rel`` orderings are not valid on ``load``
5668 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5669 when they may see multiple atomic stores. The type of the pointee must
5670 be an integer type whose bit width is a power of two greater than or
5671 equal to eight and less than or equal to a target-specific size limit.
5672 ``align`` must be explicitly specified on atomic loads, and the load has
5673 undefined behavior if the alignment is not set to a value which is at
5674 least the size in bytes of the pointee. ``!nontemporal`` does not have
5675 any defined semantics for atomic loads.
5677 The optional constant ``align`` argument specifies the alignment of the
5678 operation (that is, the alignment of the memory address). A value of 0
5679 or an omitted ``align`` argument means that the operation has the ABI
5680 alignment for the target. It is the responsibility of the code emitter
5681 to ensure that the alignment information is correct. Overestimating the
5682 alignment results in undefined behavior. Underestimating the alignment
5683 may produce less efficient code. An alignment of 1 is always safe. The
5684 maximum possible alignment is ``1 << 29``.
5686 The optional ``!nontemporal`` metadata must reference a single
5687 metadata name ``<index>`` corresponding to a metadata node with one
5688 ``i32`` entry of value 1. The existence of the ``!nontemporal``
5689 metadata on the instruction tells the optimizer and code generator
5690 that this load is not expected to be reused in the cache. The code
5691 generator may select special instructions to save cache bandwidth, such
5692 as the ``MOVNT`` instruction on x86.
5694 The optional ``!invariant.load`` metadata must reference a single
5695 metadata name ``<index>`` corresponding to a metadata node with no
5696 entries. The existence of the ``!invariant.load`` metadata on the
5697 instruction tells the optimizer and code generator that the address
5698 operand to this load points to memory which can be assumed unchanged.
5699 Being invariant does not imply that a location is dereferenceable,
5700 but it does imply that once the location is known dereferenceable
5701 its value is henceforth unchanging.
5703 The optional ``!nonnull`` metadata must reference a single
5704 metadata name ``<index>`` corresponding to a metadata node with no
5705 entries. The existence of the ``!nonnull`` metadata on the
5706 instruction tells the optimizer that the value loaded is known to
5707 never be null. This is analogous to the ''nonnull'' attribute
5708 on parameters and return values. This metadata can only be applied
5709 to loads of a pointer type.
5714 The location of memory pointed to is loaded. If the value being loaded
5715 is of scalar type then the number of bytes read does not exceed the
5716 minimum number of bytes needed to hold all bits of the type. For
5717 example, loading an ``i24`` reads at most three bytes. When loading a
5718 value of a type like ``i20`` with a size that is not an integral number
5719 of bytes, the result is undefined if the value was not originally
5720 written using a store of the same type.
5725 .. code-block:: llvm
5727 %ptr = alloca i32 ; yields i32*:ptr
5728 store i32 3, i32* %ptr ; yields void
5729 %val = load i32, i32* %ptr ; yields i32:val = i32 3
5733 '``store``' Instruction
5734 ^^^^^^^^^^^^^^^^^^^^^^^
5741 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
5742 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
5747 The '``store``' instruction is used to write to memory.
5752 There are two arguments to the ``store`` instruction: a value to store
5753 and an address at which to store it. The type of the ``<pointer>``
5754 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
5755 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
5756 then the optimizer is not allowed to modify the number or order of
5757 execution of this ``store`` with other :ref:`volatile
5758 operations <volatile>`.
5760 If the ``store`` is marked as ``atomic``, it takes an extra
5761 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5762 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
5763 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5764 when they may see multiple atomic stores. The type of the pointee must
5765 be an integer type whose bit width is a power of two greater than or
5766 equal to eight and less than or equal to a target-specific size limit.
5767 ``align`` must be explicitly specified on atomic stores, and the store
5768 has undefined behavior if the alignment is not set to a value which is
5769 at least the size in bytes of the pointee. ``!nontemporal`` does not
5770 have any defined semantics for atomic stores.
5772 The optional constant ``align`` argument specifies the alignment of the
5773 operation (that is, the alignment of the memory address). A value of 0
5774 or an omitted ``align`` argument means that the operation has the ABI
5775 alignment for the target. It is the responsibility of the code emitter
5776 to ensure that the alignment information is correct. Overestimating the
5777 alignment results in undefined behavior. Underestimating the
5778 alignment may produce less efficient code. An alignment of 1 is always
5779 safe. The maximum possible alignment is ``1 << 29``.
5781 The optional ``!nontemporal`` metadata must reference a single metadata
5782 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
5783 value 1. The existence of the ``!nontemporal`` metadata on the instruction
5784 tells the optimizer and code generator that this load is not expected to
5785 be reused in the cache. The code generator may select special
5786 instructions to save cache bandwidth, such as the MOVNT instruction on
5792 The contents of memory are updated to contain ``<value>`` at the
5793 location specified by the ``<pointer>`` operand. If ``<value>`` is
5794 of scalar type then the number of bytes written does not exceed the
5795 minimum number of bytes needed to hold all bits of the type. For
5796 example, storing an ``i24`` writes at most three bytes. When writing a
5797 value of a type like ``i20`` with a size that is not an integral number
5798 of bytes, it is unspecified what happens to the extra bits that do not
5799 belong to the type, but they will typically be overwritten.
5804 .. code-block:: llvm
5806 %ptr = alloca i32 ; yields i32*:ptr
5807 store i32 3, i32* %ptr ; yields void
5808 %val = load i32* %ptr ; yields i32:val = i32 3
5812 '``fence``' Instruction
5813 ^^^^^^^^^^^^^^^^^^^^^^^
5820 fence [singlethread] <ordering> ; yields void
5825 The '``fence``' instruction is used to introduce happens-before edges
5831 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5832 defines what *synchronizes-with* edges they add. They can only be given
5833 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5838 A fence A which has (at least) ``release`` ordering semantics
5839 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5840 semantics if and only if there exist atomic operations X and Y, both
5841 operating on some atomic object M, such that A is sequenced before X, X
5842 modifies M (either directly or through some side effect of a sequence
5843 headed by X), Y is sequenced before B, and Y observes M. This provides a
5844 *happens-before* dependency between A and B. Rather than an explicit
5845 ``fence``, one (but not both) of the atomic operations X or Y might
5846 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5847 still *synchronize-with* the explicit ``fence`` and establish the
5848 *happens-before* edge.
5850 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5851 ``acquire`` and ``release`` semantics specified above, participates in
5852 the global program order of other ``seq_cst`` operations and/or fences.
5854 The optional ":ref:`singlethread <singlethread>`" argument specifies
5855 that the fence only synchronizes with other fences in the same thread.
5856 (This is useful for interacting with signal handlers.)
5861 .. code-block:: llvm
5863 fence acquire ; yields void
5864 fence singlethread seq_cst ; yields void
5868 '``cmpxchg``' Instruction
5869 ^^^^^^^^^^^^^^^^^^^^^^^^^
5876 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
5881 The '``cmpxchg``' instruction is used to atomically modify memory. It
5882 loads a value in memory and compares it to a given value. If they are
5883 equal, it tries to store a new value into the memory.
5888 There are three arguments to the '``cmpxchg``' instruction: an address
5889 to operate on, a value to compare to the value currently be at that
5890 address, and a new value to place at that address if the compared values
5891 are equal. The type of '<cmp>' must be an integer type whose bit width
5892 is a power of two greater than or equal to eight and less than or equal
5893 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5894 type, and the type of '<pointer>' must be a pointer to that type. If the
5895 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5896 to modify the number or order of execution of this ``cmpxchg`` with
5897 other :ref:`volatile operations <volatile>`.
5899 The success and failure :ref:`ordering <ordering>` arguments specify how this
5900 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
5901 must be at least ``monotonic``, the ordering constraint on failure must be no
5902 stronger than that on success, and the failure ordering cannot be either
5903 ``release`` or ``acq_rel``.
5905 The optional "``singlethread``" argument declares that the ``cmpxchg``
5906 is only atomic with respect to code (usually signal handlers) running in
5907 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5908 respect to all other code in the system.
5910 The pointer passed into cmpxchg must have alignment greater than or
5911 equal to the size in memory of the operand.
5916 The contents of memory at the location specified by the '``<pointer>``' operand
5917 is read and compared to '``<cmp>``'; if the read value is the equal, the
5918 '``<new>``' is written. The original value at the location is returned, together
5919 with a flag indicating success (true) or failure (false).
5921 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
5922 permitted: the operation may not write ``<new>`` even if the comparison
5925 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
5926 if the value loaded equals ``cmp``.
5928 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5929 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5930 load with an ordering parameter determined the second ordering parameter.
5935 .. code-block:: llvm
5938 %orig = atomic load i32, i32* %ptr unordered ; yields i32
5942 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5943 %squared = mul i32 %cmp, %cmp
5944 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
5945 %value_loaded = extractvalue { i32, i1 } %val_success, 0
5946 %success = extractvalue { i32, i1 } %val_success, 1
5947 br i1 %success, label %done, label %loop
5954 '``atomicrmw``' Instruction
5955 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5962 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
5967 The '``atomicrmw``' instruction is used to atomically modify memory.
5972 There are three arguments to the '``atomicrmw``' instruction: an
5973 operation to apply, an address whose value to modify, an argument to the
5974 operation. The operation must be one of the following keywords:
5988 The type of '<value>' must be an integer type whose bit width is a power
5989 of two greater than or equal to eight and less than or equal to a
5990 target-specific size limit. The type of the '``<pointer>``' operand must
5991 be a pointer to that type. If the ``atomicrmw`` is marked as
5992 ``volatile``, then the optimizer is not allowed to modify the number or
5993 order of execution of this ``atomicrmw`` with other :ref:`volatile
5994 operations <volatile>`.
5999 The contents of memory at the location specified by the '``<pointer>``'
6000 operand are atomically read, modified, and written back. The original
6001 value at the location is returned. The modification is specified by the
6004 - xchg: ``*ptr = val``
6005 - add: ``*ptr = *ptr + val``
6006 - sub: ``*ptr = *ptr - val``
6007 - and: ``*ptr = *ptr & val``
6008 - nand: ``*ptr = ~(*ptr & val)``
6009 - or: ``*ptr = *ptr | val``
6010 - xor: ``*ptr = *ptr ^ val``
6011 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
6012 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
6013 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
6015 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
6021 .. code-block:: llvm
6023 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
6025 .. _i_getelementptr:
6027 '``getelementptr``' Instruction
6028 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6035 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
6036 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
6037 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
6042 The '``getelementptr``' instruction is used to get the address of a
6043 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
6044 address calculation only and does not access memory.
6049 The first argument is always a type used as the basis for the calculations.
6050 The second argument is always a pointer or a vector of pointers, and is the
6051 base address to start from. The remaining arguments are indices
6052 that indicate which of the elements of the aggregate object are indexed.
6053 The interpretation of each index is dependent on the type being indexed
6054 into. The first index always indexes the pointer value given as the
6055 first argument, the second index indexes a value of the type pointed to
6056 (not necessarily the value directly pointed to, since the first index
6057 can be non-zero), etc. The first type indexed into must be a pointer
6058 value, subsequent types can be arrays, vectors, and structs. Note that
6059 subsequent types being indexed into can never be pointers, since that
6060 would require loading the pointer before continuing calculation.
6062 The type of each index argument depends on the type it is indexing into.
6063 When indexing into a (optionally packed) structure, only ``i32`` integer
6064 **constants** are allowed (when using a vector of indices they must all
6065 be the **same** ``i32`` integer constant). When indexing into an array,
6066 pointer or vector, integers of any width are allowed, and they are not
6067 required to be constant. These integers are treated as signed values
6070 For example, let's consider a C code fragment and how it gets compiled
6086 int *foo(struct ST *s) {
6087 return &s[1].Z.B[5][13];
6090 The LLVM code generated by Clang is:
6092 .. code-block:: llvm
6094 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
6095 %struct.ST = type { i32, double, %struct.RT }
6097 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
6099 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
6106 In the example above, the first index is indexing into the
6107 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
6108 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
6109 indexes into the third element of the structure, yielding a
6110 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
6111 structure. The third index indexes into the second element of the
6112 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
6113 dimensions of the array are subscripted into, yielding an '``i32``'
6114 type. The '``getelementptr``' instruction returns a pointer to this
6115 element, thus computing a value of '``i32*``' type.
6117 Note that it is perfectly legal to index partially through a structure,
6118 returning a pointer to an inner element. Because of this, the LLVM code
6119 for the given testcase is equivalent to:
6121 .. code-block:: llvm
6123 define i32* @foo(%struct.ST* %s) {
6124 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
6125 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
6126 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
6127 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
6128 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
6132 If the ``inbounds`` keyword is present, the result value of the
6133 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
6134 pointer is not an *in bounds* address of an allocated object, or if any
6135 of the addresses that would be formed by successive addition of the
6136 offsets implied by the indices to the base address with infinitely
6137 precise signed arithmetic are not an *in bounds* address of that
6138 allocated object. The *in bounds* addresses for an allocated object are
6139 all the addresses that point into the object, plus the address one byte
6140 past the end. In cases where the base is a vector of pointers the
6141 ``inbounds`` keyword applies to each of the computations element-wise.
6143 If the ``inbounds`` keyword is not present, the offsets are added to the
6144 base address with silently-wrapping two's complement arithmetic. If the
6145 offsets have a different width from the pointer, they are sign-extended
6146 or truncated to the width of the pointer. The result value of the
6147 ``getelementptr`` may be outside the object pointed to by the base
6148 pointer. The result value may not necessarily be used to access memory
6149 though, even if it happens to point into allocated storage. See the
6150 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
6153 The getelementptr instruction is often confusing. For some more insight
6154 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
6159 .. code-block:: llvm
6161 ; yields [12 x i8]*:aptr
6162 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
6164 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
6166 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
6168 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
6170 In cases where the pointer argument is a vector of pointers, each index
6171 must be a vector with the same number of elements. For example:
6173 .. code-block:: llvm
6175 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets,
6177 Conversion Operations
6178 ---------------------
6180 The instructions in this category are the conversion instructions
6181 (casting) which all take a single operand and a type. They perform
6182 various bit conversions on the operand.
6184 '``trunc .. to``' Instruction
6185 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6192 <result> = trunc <ty> <value> to <ty2> ; yields ty2
6197 The '``trunc``' instruction truncates its operand to the type ``ty2``.
6202 The '``trunc``' instruction takes a value to trunc, and a type to trunc
6203 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
6204 of the same number of integers. The bit size of the ``value`` must be
6205 larger than the bit size of the destination type, ``ty2``. Equal sized
6206 types are not allowed.
6211 The '``trunc``' instruction truncates the high order bits in ``value``
6212 and converts the remaining bits to ``ty2``. Since the source size must
6213 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
6214 It will always truncate bits.
6219 .. code-block:: llvm
6221 %X = trunc i32 257 to i8 ; yields i8:1
6222 %Y = trunc i32 123 to i1 ; yields i1:true
6223 %Z = trunc i32 122 to i1 ; yields i1:false
6224 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
6226 '``zext .. to``' Instruction
6227 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6234 <result> = zext <ty> <value> to <ty2> ; yields ty2
6239 The '``zext``' instruction zero extends its operand to type ``ty2``.
6244 The '``zext``' instruction takes a value to cast, and a type to cast it
6245 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
6246 the same number of integers. The bit size of the ``value`` must be
6247 smaller than the bit size of the destination type, ``ty2``.
6252 The ``zext`` fills the high order bits of the ``value`` with zero bits
6253 until it reaches the size of the destination type, ``ty2``.
6255 When zero extending from i1, the result will always be either 0 or 1.
6260 .. code-block:: llvm
6262 %X = zext i32 257 to i64 ; yields i64:257
6263 %Y = zext i1 true to i32 ; yields i32:1
6264 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
6266 '``sext .. to``' Instruction
6267 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6274 <result> = sext <ty> <value> to <ty2> ; yields ty2
6279 The '``sext``' sign extends ``value`` to the type ``ty2``.
6284 The '``sext``' instruction takes a value to cast, and a type to cast it
6285 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
6286 the same number of integers. The bit size of the ``value`` must be
6287 smaller than the bit size of the destination type, ``ty2``.
6292 The '``sext``' instruction performs a sign extension by copying the sign
6293 bit (highest order bit) of the ``value`` until it reaches the bit size
6294 of the type ``ty2``.
6296 When sign extending from i1, the extension always results in -1 or 0.
6301 .. code-block:: llvm
6303 %X = sext i8 -1 to i16 ; yields i16 :65535
6304 %Y = sext i1 true to i32 ; yields i32:-1
6305 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
6307 '``fptrunc .. to``' Instruction
6308 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6315 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
6320 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
6325 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
6326 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
6327 The size of ``value`` must be larger than the size of ``ty2``. This
6328 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
6333 The '``fptrunc``' instruction truncates a ``value`` from a larger
6334 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
6335 point <t_floating>` type. If the value cannot fit within the
6336 destination type, ``ty2``, then the results are undefined.
6341 .. code-block:: llvm
6343 %X = fptrunc double 123.0 to float ; yields float:123.0
6344 %Y = fptrunc double 1.0E+300 to float ; yields undefined
6346 '``fpext .. to``' Instruction
6347 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6354 <result> = fpext <ty> <value> to <ty2> ; yields ty2
6359 The '``fpext``' extends a floating point ``value`` to a larger floating
6365 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
6366 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
6367 to. The source type must be smaller than the destination type.
6372 The '``fpext``' instruction extends the ``value`` from a smaller
6373 :ref:`floating point <t_floating>` type to a larger :ref:`floating
6374 point <t_floating>` type. The ``fpext`` cannot be used to make a
6375 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
6376 *no-op cast* for a floating point cast.
6381 .. code-block:: llvm
6383 %X = fpext float 3.125 to double ; yields double:3.125000e+00
6384 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
6386 '``fptoui .. to``' Instruction
6387 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6394 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
6399 The '``fptoui``' converts a floating point ``value`` to its unsigned
6400 integer equivalent of type ``ty2``.
6405 The '``fptoui``' instruction takes a value to cast, which must be a
6406 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6407 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6408 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6409 type with the same number of elements as ``ty``
6414 The '``fptoui``' instruction converts its :ref:`floating
6415 point <t_floating>` operand into the nearest (rounding towards zero)
6416 unsigned integer value. If the value cannot fit in ``ty2``, the results
6422 .. code-block:: llvm
6424 %X = fptoui double 123.0 to i32 ; yields i32:123
6425 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
6426 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
6428 '``fptosi .. to``' Instruction
6429 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6436 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
6441 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
6442 ``value`` to type ``ty2``.
6447 The '``fptosi``' instruction takes a value to cast, which must be a
6448 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6449 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6450 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6451 type with the same number of elements as ``ty``
6456 The '``fptosi``' instruction converts its :ref:`floating
6457 point <t_floating>` operand into the nearest (rounding towards zero)
6458 signed integer value. If the value cannot fit in ``ty2``, the results
6464 .. code-block:: llvm
6466 %X = fptosi double -123.0 to i32 ; yields i32:-123
6467 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
6468 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
6470 '``uitofp .. to``' Instruction
6471 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6478 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
6483 The '``uitofp``' instruction regards ``value`` as an unsigned integer
6484 and converts that value to the ``ty2`` type.
6489 The '``uitofp``' instruction takes a value to cast, which must be a
6490 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6491 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6492 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6493 type with the same number of elements as ``ty``
6498 The '``uitofp``' instruction interprets its operand as an unsigned
6499 integer quantity and converts it to the corresponding floating point
6500 value. If the value cannot fit in the floating point value, the results
6506 .. code-block:: llvm
6508 %X = uitofp i32 257 to float ; yields float:257.0
6509 %Y = uitofp i8 -1 to double ; yields double:255.0
6511 '``sitofp .. to``' Instruction
6512 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6519 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
6524 The '``sitofp``' instruction regards ``value`` as a signed integer and
6525 converts that value to the ``ty2`` type.
6530 The '``sitofp``' instruction takes a value to cast, which must be a
6531 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6532 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6533 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6534 type with the same number of elements as ``ty``
6539 The '``sitofp``' instruction interprets its operand as a signed integer
6540 quantity and converts it to the corresponding floating point value. If
6541 the value cannot fit in the floating point value, the results are
6547 .. code-block:: llvm
6549 %X = sitofp i32 257 to float ; yields float:257.0
6550 %Y = sitofp i8 -1 to double ; yields double:-1.0
6554 '``ptrtoint .. to``' Instruction
6555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6562 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
6567 The '``ptrtoint``' instruction converts the pointer or a vector of
6568 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
6573 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
6574 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
6575 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
6576 a vector of integers type.
6581 The '``ptrtoint``' instruction converts ``value`` to integer type
6582 ``ty2`` by interpreting the pointer value as an integer and either
6583 truncating or zero extending that value to the size of the integer type.
6584 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
6585 ``value`` is larger than ``ty2`` then a truncation is done. If they are
6586 the same size, then nothing is done (*no-op cast*) other than a type
6592 .. code-block:: llvm
6594 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
6595 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
6596 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
6600 '``inttoptr .. to``' Instruction
6601 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6608 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
6613 The '``inttoptr``' instruction converts an integer ``value`` to a
6614 pointer type, ``ty2``.
6619 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
6620 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
6626 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
6627 applying either a zero extension or a truncation depending on the size
6628 of the integer ``value``. If ``value`` is larger than the size of a
6629 pointer then a truncation is done. If ``value`` is smaller than the size
6630 of a pointer then a zero extension is done. If they are the same size,
6631 nothing is done (*no-op cast*).
6636 .. code-block:: llvm
6638 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
6639 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
6640 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
6641 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
6645 '``bitcast .. to``' Instruction
6646 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6653 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
6658 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
6664 The '``bitcast``' instruction takes a value to cast, which must be a
6665 non-aggregate first class value, and a type to cast it to, which must
6666 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
6667 bit sizes of ``value`` and the destination type, ``ty2``, must be
6668 identical. If the source type is a pointer, the destination type must
6669 also be a pointer of the same size. This instruction supports bitwise
6670 conversion of vectors to integers and to vectors of other types (as
6671 long as they have the same size).
6676 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
6677 is always a *no-op cast* because no bits change with this
6678 conversion. The conversion is done as if the ``value`` had been stored
6679 to memory and read back as type ``ty2``. Pointer (or vector of
6680 pointers) types may only be converted to other pointer (or vector of
6681 pointers) types with the same address space through this instruction.
6682 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
6683 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
6688 .. code-block:: llvm
6690 %X = bitcast i8 255 to i8 ; yields i8 :-1
6691 %Y = bitcast i32* %x to sint* ; yields sint*:%x
6692 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
6693 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
6695 .. _i_addrspacecast:
6697 '``addrspacecast .. to``' Instruction
6698 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6705 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
6710 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
6711 address space ``n`` to type ``pty2`` in address space ``m``.
6716 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
6717 to cast and a pointer type to cast it to, which must have a different
6723 The '``addrspacecast``' instruction converts the pointer value
6724 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
6725 value modification, depending on the target and the address space
6726 pair. Pointer conversions within the same address space must be
6727 performed with the ``bitcast`` instruction. Note that if the address space
6728 conversion is legal then both result and operand refer to the same memory
6734 .. code-block:: llvm
6736 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
6737 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
6738 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
6745 The instructions in this category are the "miscellaneous" instructions,
6746 which defy better classification.
6750 '``icmp``' Instruction
6751 ^^^^^^^^^^^^^^^^^^^^^^
6758 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6763 The '``icmp``' instruction returns a boolean value or a vector of
6764 boolean values based on comparison of its two integer, integer vector,
6765 pointer, or pointer vector operands.
6770 The '``icmp``' instruction takes three operands. The first operand is
6771 the condition code indicating the kind of comparison to perform. It is
6772 not a value, just a keyword. The possible condition code are:
6775 #. ``ne``: not equal
6776 #. ``ugt``: unsigned greater than
6777 #. ``uge``: unsigned greater or equal
6778 #. ``ult``: unsigned less than
6779 #. ``ule``: unsigned less or equal
6780 #. ``sgt``: signed greater than
6781 #. ``sge``: signed greater or equal
6782 #. ``slt``: signed less than
6783 #. ``sle``: signed less or equal
6785 The remaining two arguments must be :ref:`integer <t_integer>` or
6786 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
6787 must also be identical types.
6792 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
6793 code given as ``cond``. The comparison performed always yields either an
6794 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
6796 #. ``eq``: yields ``true`` if the operands are equal, ``false``
6797 otherwise. No sign interpretation is necessary or performed.
6798 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
6799 otherwise. No sign interpretation is necessary or performed.
6800 #. ``ugt``: interprets the operands as unsigned values and yields
6801 ``true`` if ``op1`` is greater than ``op2``.
6802 #. ``uge``: interprets the operands as unsigned values and yields
6803 ``true`` if ``op1`` is greater than or equal to ``op2``.
6804 #. ``ult``: interprets the operands as unsigned values and yields
6805 ``true`` if ``op1`` is less than ``op2``.
6806 #. ``ule``: interprets the operands as unsigned values and yields
6807 ``true`` if ``op1`` is less than or equal to ``op2``.
6808 #. ``sgt``: interprets the operands as signed values and yields ``true``
6809 if ``op1`` is greater than ``op2``.
6810 #. ``sge``: interprets the operands as signed values and yields ``true``
6811 if ``op1`` is greater than or equal to ``op2``.
6812 #. ``slt``: interprets the operands as signed values and yields ``true``
6813 if ``op1`` is less than ``op2``.
6814 #. ``sle``: interprets the operands as signed values and yields ``true``
6815 if ``op1`` is less than or equal to ``op2``.
6817 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
6818 are compared as if they were integers.
6820 If the operands are integer vectors, then they are compared element by
6821 element. The result is an ``i1`` vector with the same number of elements
6822 as the values being compared. Otherwise, the result is an ``i1``.
6827 .. code-block:: llvm
6829 <result> = icmp eq i32 4, 5 ; yields: result=false
6830 <result> = icmp ne float* %X, %X ; yields: result=false
6831 <result> = icmp ult i16 4, 5 ; yields: result=true
6832 <result> = icmp sgt i16 4, 5 ; yields: result=false
6833 <result> = icmp ule i16 -4, 5 ; yields: result=false
6834 <result> = icmp sge i16 4, 5 ; yields: result=false
6836 Note that the code generator does not yet support vector types with the
6837 ``icmp`` instruction.
6841 '``fcmp``' Instruction
6842 ^^^^^^^^^^^^^^^^^^^^^^
6849 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6854 The '``fcmp``' instruction returns a boolean value or vector of boolean
6855 values based on comparison of its operands.
6857 If the operands are floating point scalars, then the result type is a
6858 boolean (:ref:`i1 <t_integer>`).
6860 If the operands are floating point vectors, then the result type is a
6861 vector of boolean with the same number of elements as the operands being
6867 The '``fcmp``' instruction takes three operands. The first operand is
6868 the condition code indicating the kind of comparison to perform. It is
6869 not a value, just a keyword. The possible condition code are:
6871 #. ``false``: no comparison, always returns false
6872 #. ``oeq``: ordered and equal
6873 #. ``ogt``: ordered and greater than
6874 #. ``oge``: ordered and greater than or equal
6875 #. ``olt``: ordered and less than
6876 #. ``ole``: ordered and less than or equal
6877 #. ``one``: ordered and not equal
6878 #. ``ord``: ordered (no nans)
6879 #. ``ueq``: unordered or equal
6880 #. ``ugt``: unordered or greater than
6881 #. ``uge``: unordered or greater than or equal
6882 #. ``ult``: unordered or less than
6883 #. ``ule``: unordered or less than or equal
6884 #. ``une``: unordered or not equal
6885 #. ``uno``: unordered (either nans)
6886 #. ``true``: no comparison, always returns true
6888 *Ordered* means that neither operand is a QNAN while *unordered* means
6889 that either operand may be a QNAN.
6891 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6892 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6893 type. They must have identical types.
6898 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6899 condition code given as ``cond``. If the operands are vectors, then the
6900 vectors are compared element by element. Each comparison performed
6901 always yields an :ref:`i1 <t_integer>` result, as follows:
6903 #. ``false``: always yields ``false``, regardless of operands.
6904 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6905 is equal to ``op2``.
6906 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6907 is greater than ``op2``.
6908 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6909 is greater than or equal to ``op2``.
6910 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6911 is less than ``op2``.
6912 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6913 is less than or equal to ``op2``.
6914 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6915 is not equal to ``op2``.
6916 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6917 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6919 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6920 greater than ``op2``.
6921 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6922 greater than or equal to ``op2``.
6923 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6925 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6926 less than or equal to ``op2``.
6927 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6928 not equal to ``op2``.
6929 #. ``uno``: yields ``true`` if either operand is a QNAN.
6930 #. ``true``: always yields ``true``, regardless of operands.
6935 .. code-block:: llvm
6937 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6938 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6939 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6940 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6942 Note that the code generator does not yet support vector types with the
6943 ``fcmp`` instruction.
6947 '``phi``' Instruction
6948 ^^^^^^^^^^^^^^^^^^^^^
6955 <result> = phi <ty> [ <val0>, <label0>], ...
6960 The '``phi``' instruction is used to implement the φ node in the SSA
6961 graph representing the function.
6966 The type of the incoming values is specified with the first type field.
6967 After this, the '``phi``' instruction takes a list of pairs as
6968 arguments, with one pair for each predecessor basic block of the current
6969 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6970 the value arguments to the PHI node. Only labels may be used as the
6973 There must be no non-phi instructions between the start of a basic block
6974 and the PHI instructions: i.e. PHI instructions must be first in a basic
6977 For the purposes of the SSA form, the use of each incoming value is
6978 deemed to occur on the edge from the corresponding predecessor block to
6979 the current block (but after any definition of an '``invoke``'
6980 instruction's return value on the same edge).
6985 At runtime, the '``phi``' instruction logically takes on the value
6986 specified by the pair corresponding to the predecessor basic block that
6987 executed just prior to the current block.
6992 .. code-block:: llvm
6994 Loop: ; Infinite loop that counts from 0 on up...
6995 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6996 %nextindvar = add i32 %indvar, 1
7001 '``select``' Instruction
7002 ^^^^^^^^^^^^^^^^^^^^^^^^
7009 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
7011 selty is either i1 or {<N x i1>}
7016 The '``select``' instruction is used to choose one value based on a
7017 condition, without IR-level branching.
7022 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
7023 values indicating the condition, and two values of the same :ref:`first
7024 class <t_firstclass>` type.
7029 If the condition is an i1 and it evaluates to 1, the instruction returns
7030 the first value argument; otherwise, it returns the second value
7033 If the condition is a vector of i1, then the value arguments must be
7034 vectors of the same size, and the selection is done element by element.
7036 If the condition is an i1 and the value arguments are vectors of the
7037 same size, then an entire vector is selected.
7042 .. code-block:: llvm
7044 %X = select i1 true, i8 17, i8 42 ; yields i8:17
7048 '``call``' Instruction
7049 ^^^^^^^^^^^^^^^^^^^^^^
7056 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
7061 The '``call``' instruction represents a simple function call.
7066 This instruction requires several arguments:
7068 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
7069 should perform tail call optimization. The ``tail`` marker is a hint that
7070 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
7071 means that the call must be tail call optimized in order for the program to
7072 be correct. The ``musttail`` marker provides these guarantees:
7074 #. The call will not cause unbounded stack growth if it is part of a
7075 recursive cycle in the call graph.
7076 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
7079 Both markers imply that the callee does not access allocas or varargs from
7080 the caller. Calls marked ``musttail`` must obey the following additional
7083 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
7084 or a pointer bitcast followed by a ret instruction.
7085 - The ret instruction must return the (possibly bitcasted) value
7086 produced by the call or void.
7087 - The caller and callee prototypes must match. Pointer types of
7088 parameters or return types may differ in pointee type, but not
7090 - The calling conventions of the caller and callee must match.
7091 - All ABI-impacting function attributes, such as sret, byval, inreg,
7092 returned, and inalloca, must match.
7093 - The callee must be varargs iff the caller is varargs. Bitcasting a
7094 non-varargs function to the appropriate varargs type is legal so
7095 long as the non-varargs prefixes obey the other rules.
7097 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
7098 the following conditions are met:
7100 - Caller and callee both have the calling convention ``fastcc``.
7101 - The call is in tail position (ret immediately follows call and ret
7102 uses value of call or is void).
7103 - Option ``-tailcallopt`` is enabled, or
7104 ``llvm::GuaranteedTailCallOpt`` is ``true``.
7105 - `Platform-specific constraints are
7106 met. <CodeGenerator.html#tailcallopt>`_
7108 #. The optional "cconv" marker indicates which :ref:`calling
7109 convention <callingconv>` the call should use. If none is
7110 specified, the call defaults to using C calling conventions. The
7111 calling convention of the call must match the calling convention of
7112 the target function, or else the behavior is undefined.
7113 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
7114 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
7116 #. '``ty``': the type of the call instruction itself which is also the
7117 type of the return value. Functions that return no value are marked
7119 #. '``fnty``': shall be the signature of the pointer to function value
7120 being invoked. The argument types must match the types implied by
7121 this signature. This type can be omitted if the function is not
7122 varargs and if the function type does not return a pointer to a
7124 #. '``fnptrval``': An LLVM value containing a pointer to a function to
7125 be invoked. In most cases, this is a direct function invocation, but
7126 indirect ``call``'s are just as possible, calling an arbitrary pointer
7128 #. '``function args``': argument list whose types match the function
7129 signature argument types and parameter attributes. All arguments must
7130 be of :ref:`first class <t_firstclass>` type. If the function signature
7131 indicates the function accepts a variable number of arguments, the
7132 extra arguments can be specified.
7133 #. The optional :ref:`function attributes <fnattrs>` list. Only
7134 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
7135 attributes are valid here.
7140 The '``call``' instruction is used to cause control flow to transfer to
7141 a specified function, with its incoming arguments bound to the specified
7142 values. Upon a '``ret``' instruction in the called function, control
7143 flow continues with the instruction after the function call, and the
7144 return value of the function is bound to the result argument.
7149 .. code-block:: llvm
7151 %retval = call i32 @test(i32 %argc)
7152 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
7153 %X = tail call i32 @foo() ; yields i32
7154 %Y = tail call fastcc i32 @foo() ; yields i32
7155 call void %foo(i8 97 signext)
7157 %struct.A = type { i32, i8 }
7158 %r = call %struct.A @foo() ; yields { i32, i8 }
7159 %gr = extractvalue %struct.A %r, 0 ; yields i32
7160 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
7161 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
7162 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
7164 llvm treats calls to some functions with names and arguments that match
7165 the standard C99 library as being the C99 library functions, and may
7166 perform optimizations or generate code for them under that assumption.
7167 This is something we'd like to change in the future to provide better
7168 support for freestanding environments and non-C-based languages.
7172 '``va_arg``' Instruction
7173 ^^^^^^^^^^^^^^^^^^^^^^^^
7180 <resultval> = va_arg <va_list*> <arglist>, <argty>
7185 The '``va_arg``' instruction is used to access arguments passed through
7186 the "variable argument" area of a function call. It is used to implement
7187 the ``va_arg`` macro in C.
7192 This instruction takes a ``va_list*`` value and the type of the
7193 argument. It returns a value of the specified argument type and
7194 increments the ``va_list`` to point to the next argument. The actual
7195 type of ``va_list`` is target specific.
7200 The '``va_arg``' instruction loads an argument of the specified type
7201 from the specified ``va_list`` and causes the ``va_list`` to point to
7202 the next argument. For more information, see the variable argument
7203 handling :ref:`Intrinsic Functions <int_varargs>`.
7205 It is legal for this instruction to be called in a function which does
7206 not take a variable number of arguments, for example, the ``vfprintf``
7209 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
7210 function <intrinsics>` because it takes a type as an argument.
7215 See the :ref:`variable argument processing <int_varargs>` section.
7217 Note that the code generator does not yet fully support va\_arg on many
7218 targets. Also, it does not currently support va\_arg with aggregate
7219 types on any target.
7223 '``landingpad``' Instruction
7224 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7231 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
7232 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
7234 <clause> := catch <type> <value>
7235 <clause> := filter <array constant type> <array constant>
7240 The '``landingpad``' instruction is used by `LLVM's exception handling
7241 system <ExceptionHandling.html#overview>`_ to specify that a basic block
7242 is a landing pad --- one where the exception lands, and corresponds to the
7243 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
7244 defines values supplied by the personality function (``pers_fn``) upon
7245 re-entry to the function. The ``resultval`` has the type ``resultty``.
7250 This instruction takes a ``pers_fn`` value. This is the personality
7251 function associated with the unwinding mechanism. The optional
7252 ``cleanup`` flag indicates that the landing pad block is a cleanup.
7254 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
7255 contains the global variable representing the "type" that may be caught
7256 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
7257 clause takes an array constant as its argument. Use
7258 "``[0 x i8**] undef``" for a filter which cannot throw. The
7259 '``landingpad``' instruction must contain *at least* one ``clause`` or
7260 the ``cleanup`` flag.
7265 The '``landingpad``' instruction defines the values which are set by the
7266 personality function (``pers_fn``) upon re-entry to the function, and
7267 therefore the "result type" of the ``landingpad`` instruction. As with
7268 calling conventions, how the personality function results are
7269 represented in LLVM IR is target specific.
7271 The clauses are applied in order from top to bottom. If two
7272 ``landingpad`` instructions are merged together through inlining, the
7273 clauses from the calling function are appended to the list of clauses.
7274 When the call stack is being unwound due to an exception being thrown,
7275 the exception is compared against each ``clause`` in turn. If it doesn't
7276 match any of the clauses, and the ``cleanup`` flag is not set, then
7277 unwinding continues further up the call stack.
7279 The ``landingpad`` instruction has several restrictions:
7281 - A landing pad block is a basic block which is the unwind destination
7282 of an '``invoke``' instruction.
7283 - A landing pad block must have a '``landingpad``' instruction as its
7284 first non-PHI instruction.
7285 - There can be only one '``landingpad``' instruction within the landing
7287 - A basic block that is not a landing pad block may not include a
7288 '``landingpad``' instruction.
7289 - All '``landingpad``' instructions in a function must have the same
7290 personality function.
7295 .. code-block:: llvm
7297 ;; A landing pad which can catch an integer.
7298 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7300 ;; A landing pad that is a cleanup.
7301 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7303 ;; A landing pad which can catch an integer and can only throw a double.
7304 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7306 filter [1 x i8**] [@_ZTId]
7313 LLVM supports the notion of an "intrinsic function". These functions
7314 have well known names and semantics and are required to follow certain
7315 restrictions. Overall, these intrinsics represent an extension mechanism
7316 for the LLVM language that does not require changing all of the
7317 transformations in LLVM when adding to the language (or the bitcode
7318 reader/writer, the parser, etc...).
7320 Intrinsic function names must all start with an "``llvm.``" prefix. This
7321 prefix is reserved in LLVM for intrinsic names; thus, function names may
7322 not begin with this prefix. Intrinsic functions must always be external
7323 functions: you cannot define the body of intrinsic functions. Intrinsic
7324 functions may only be used in call or invoke instructions: it is illegal
7325 to take the address of an intrinsic function. Additionally, because
7326 intrinsic functions are part of the LLVM language, it is required if any
7327 are added that they be documented here.
7329 Some intrinsic functions can be overloaded, i.e., the intrinsic
7330 represents a family of functions that perform the same operation but on
7331 different data types. Because LLVM can represent over 8 million
7332 different integer types, overloading is used commonly to allow an
7333 intrinsic function to operate on any integer type. One or more of the
7334 argument types or the result type can be overloaded to accept any
7335 integer type. Argument types may also be defined as exactly matching a
7336 previous argument's type or the result type. This allows an intrinsic
7337 function which accepts multiple arguments, but needs all of them to be
7338 of the same type, to only be overloaded with respect to a single
7339 argument or the result.
7341 Overloaded intrinsics will have the names of its overloaded argument
7342 types encoded into its function name, each preceded by a period. Only
7343 those types which are overloaded result in a name suffix. Arguments
7344 whose type is matched against another type do not. For example, the
7345 ``llvm.ctpop`` function can take an integer of any width and returns an
7346 integer of exactly the same integer width. This leads to a family of
7347 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
7348 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
7349 overloaded, and only one type suffix is required. Because the argument's
7350 type is matched against the return type, it does not require its own
7353 To learn how to add an intrinsic function, please see the `Extending
7354 LLVM Guide <ExtendingLLVM.html>`_.
7358 Variable Argument Handling Intrinsics
7359 -------------------------------------
7361 Variable argument support is defined in LLVM with the
7362 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
7363 functions. These functions are related to the similarly named macros
7364 defined in the ``<stdarg.h>`` header file.
7366 All of these functions operate on arguments that use a target-specific
7367 value type "``va_list``". The LLVM assembly language reference manual
7368 does not define what this type is, so all transformations should be
7369 prepared to handle these functions regardless of the type used.
7371 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
7372 variable argument handling intrinsic functions are used.
7374 .. code-block:: llvm
7376 ; This struct is different for every platform. For most platforms,
7377 ; it is merely an i8*.
7378 %struct.va_list = type { i8* }
7380 ; For Unix x86_64 platforms, va_list is the following struct:
7381 ; %struct.va_list = type { i32, i32, i8*, i8* }
7383 define i32 @test(i32 %X, ...) {
7384 ; Initialize variable argument processing
7385 %ap = alloca %struct.va_list
7386 %ap2 = bitcast %struct.va_list* %ap to i8*
7387 call void @llvm.va_start(i8* %ap2)
7389 ; Read a single integer argument
7390 %tmp = va_arg i8* %ap2, i32
7392 ; Demonstrate usage of llvm.va_copy and llvm.va_end
7394 %aq2 = bitcast i8** %aq to i8*
7395 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
7396 call void @llvm.va_end(i8* %aq2)
7398 ; Stop processing of arguments.
7399 call void @llvm.va_end(i8* %ap2)
7403 declare void @llvm.va_start(i8*)
7404 declare void @llvm.va_copy(i8*, i8*)
7405 declare void @llvm.va_end(i8*)
7409 '``llvm.va_start``' Intrinsic
7410 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7417 declare void @llvm.va_start(i8* <arglist>)
7422 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
7423 subsequent use by ``va_arg``.
7428 The argument is a pointer to a ``va_list`` element to initialize.
7433 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
7434 available in C. In a target-dependent way, it initializes the
7435 ``va_list`` element to which the argument points, so that the next call
7436 to ``va_arg`` will produce the first variable argument passed to the
7437 function. Unlike the C ``va_start`` macro, this intrinsic does not need
7438 to know the last argument of the function as the compiler can figure
7441 '``llvm.va_end``' Intrinsic
7442 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7449 declare void @llvm.va_end(i8* <arglist>)
7454 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
7455 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
7460 The argument is a pointer to a ``va_list`` to destroy.
7465 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
7466 available in C. In a target-dependent way, it destroys the ``va_list``
7467 element to which the argument points. Calls to
7468 :ref:`llvm.va_start <int_va_start>` and
7469 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
7474 '``llvm.va_copy``' Intrinsic
7475 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7482 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
7487 The '``llvm.va_copy``' intrinsic copies the current argument position
7488 from the source argument list to the destination argument list.
7493 The first argument is a pointer to a ``va_list`` element to initialize.
7494 The second argument is a pointer to a ``va_list`` element to copy from.
7499 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
7500 available in C. In a target-dependent way, it copies the source
7501 ``va_list`` element into the destination ``va_list`` element. This
7502 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
7503 arbitrarily complex and require, for example, memory allocation.
7505 Accurate Garbage Collection Intrinsics
7506 --------------------------------------
7508 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
7509 (GC) requires the frontend to generate code containing appropriate intrinsic
7510 calls and select an appropriate GC strategy which knows how to lower these
7511 intrinsics in a manner which is appropriate for the target collector.
7513 These intrinsics allow identification of :ref:`GC roots on the
7514 stack <int_gcroot>`, as well as garbage collector implementations that
7515 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
7516 Frontends for type-safe garbage collected languages should generate
7517 these intrinsics to make use of the LLVM garbage collectors. For more
7518 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
7520 Experimental Statepoint Intrinsics
7521 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7523 LLVM provides an second experimental set of intrinsics for describing garbage
7524 collection safepoints in compiled code. These intrinsics are an alternative
7525 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
7526 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
7527 differences in approach are covered in the `Garbage Collection with LLVM
7528 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
7529 described in :doc:`Statepoints`.
7533 '``llvm.gcroot``' Intrinsic
7534 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7541 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
7546 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
7547 the code generator, and allows some metadata to be associated with it.
7552 The first argument specifies the address of a stack object that contains
7553 the root pointer. The second pointer (which must be either a constant or
7554 a global value address) contains the meta-data to be associated with the
7560 At runtime, a call to this intrinsic stores a null pointer into the
7561 "ptrloc" location. At compile-time, the code generator generates
7562 information to allow the runtime to find the pointer at GC safe points.
7563 The '``llvm.gcroot``' intrinsic may only be used in a function which
7564 :ref:`specifies a GC algorithm <gc>`.
7568 '``llvm.gcread``' Intrinsic
7569 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7576 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
7581 The '``llvm.gcread``' intrinsic identifies reads of references from heap
7582 locations, allowing garbage collector implementations that require read
7588 The second argument is the address to read from, which should be an
7589 address allocated from the garbage collector. The first object is a
7590 pointer to the start of the referenced object, if needed by the language
7591 runtime (otherwise null).
7596 The '``llvm.gcread``' intrinsic has the same semantics as a load
7597 instruction, but may be replaced with substantially more complex code by
7598 the garbage collector runtime, as needed. The '``llvm.gcread``'
7599 intrinsic may only be used in a function which :ref:`specifies a GC
7604 '``llvm.gcwrite``' Intrinsic
7605 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7612 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
7617 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
7618 locations, allowing garbage collector implementations that require write
7619 barriers (such as generational or reference counting collectors).
7624 The first argument is the reference to store, the second is the start of
7625 the object to store it to, and the third is the address of the field of
7626 Obj to store to. If the runtime does not require a pointer to the
7627 object, Obj may be null.
7632 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
7633 instruction, but may be replaced with substantially more complex code by
7634 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
7635 intrinsic may only be used in a function which :ref:`specifies a GC
7638 Code Generator Intrinsics
7639 -------------------------
7641 These intrinsics are provided by LLVM to expose special features that
7642 may only be implemented with code generator support.
7644 '``llvm.returnaddress``' Intrinsic
7645 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7652 declare i8 *@llvm.returnaddress(i32 <level>)
7657 The '``llvm.returnaddress``' intrinsic attempts to compute a
7658 target-specific value indicating the return address of the current
7659 function or one of its callers.
7664 The argument to this intrinsic indicates which function to return the
7665 address for. Zero indicates the calling function, one indicates its
7666 caller, etc. The argument is **required** to be a constant integer
7672 The '``llvm.returnaddress``' intrinsic either returns a pointer
7673 indicating the return address of the specified call frame, or zero if it
7674 cannot be identified. The value returned by this intrinsic is likely to
7675 be incorrect or 0 for arguments other than zero, so it should only be
7676 used for debugging purposes.
7678 Note that calling this intrinsic does not prevent function inlining or
7679 other aggressive transformations, so the value returned may not be that
7680 of the obvious source-language caller.
7682 '``llvm.frameaddress``' Intrinsic
7683 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7690 declare i8* @llvm.frameaddress(i32 <level>)
7695 The '``llvm.frameaddress``' intrinsic attempts to return the
7696 target-specific frame pointer value for the specified stack frame.
7701 The argument to this intrinsic indicates which function to return the
7702 frame pointer for. Zero indicates the calling function, one indicates
7703 its caller, etc. The argument is **required** to be a constant integer
7709 The '``llvm.frameaddress``' intrinsic either returns a pointer
7710 indicating the frame address of the specified call frame, or zero if it
7711 cannot be identified. The value returned by this intrinsic is likely to
7712 be incorrect or 0 for arguments other than zero, so it should only be
7713 used for debugging purposes.
7715 Note that calling this intrinsic does not prevent function inlining or
7716 other aggressive transformations, so the value returned may not be that
7717 of the obvious source-language caller.
7719 '``llvm.frameescape``' and '``llvm.framerecover``' Intrinsics
7720 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7727 declare void @llvm.frameescape(...)
7728 declare i8* @llvm.framerecover(i8* %func, i8* %fp, i32 %idx)
7733 The '``llvm.frameescape``' intrinsic escapes offsets of a collection of static
7734 allocas, and the '``llvm.framerecover``' intrinsic applies those offsets to a
7735 live frame pointer to recover the address of the allocation. The offset is
7736 computed during frame layout of the caller of ``llvm.frameescape``.
7741 All arguments to '``llvm.frameescape``' must be pointers to static allocas or
7742 casts of static allocas. Each function can only call '``llvm.frameescape``'
7743 once, and it can only do so from the entry block.
7745 The ``func`` argument to '``llvm.framerecover``' must be a constant
7746 bitcasted pointer to a function defined in the current module. The code
7747 generator cannot determine the frame allocation offset of functions defined in
7750 The ``fp`` argument to '``llvm.framerecover``' must be a frame
7751 pointer of a call frame that is currently live. The return value of
7752 '``llvm.frameaddress``' is one way to produce such a value, but most platforms
7753 also expose the frame pointer through stack unwinding mechanisms.
7755 The ``idx`` argument to '``llvm.framerecover``' indicates which alloca passed to
7756 '``llvm.frameescape``' to recover. It is zero-indexed.
7761 These intrinsics allow a group of functions to access one stack memory
7762 allocation in an ancestor stack frame. The memory returned from
7763 '``llvm.frameallocate``' may be allocated prior to stack realignment, so the
7764 memory is only aligned to the ABI-required stack alignment. Each function may
7765 only call '``llvm.frameallocate``' one or zero times from the function entry
7766 block. The frame allocation intrinsic inhibits inlining, as any frame
7767 allocations in the inlined function frame are likely to be at a different
7768 offset from the one used by '``llvm.framerecover``' called with the
7771 .. _int_read_register:
7772 .. _int_write_register:
7774 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
7775 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7782 declare i32 @llvm.read_register.i32(metadata)
7783 declare i64 @llvm.read_register.i64(metadata)
7784 declare void @llvm.write_register.i32(metadata, i32 @value)
7785 declare void @llvm.write_register.i64(metadata, i64 @value)
7791 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
7792 provides access to the named register. The register must be valid on
7793 the architecture being compiled to. The type needs to be compatible
7794 with the register being read.
7799 The '``llvm.read_register``' intrinsic returns the current value of the
7800 register, where possible. The '``llvm.write_register``' intrinsic sets
7801 the current value of the register, where possible.
7803 This is useful to implement named register global variables that need
7804 to always be mapped to a specific register, as is common practice on
7805 bare-metal programs including OS kernels.
7807 The compiler doesn't check for register availability or use of the used
7808 register in surrounding code, including inline assembly. Because of that,
7809 allocatable registers are not supported.
7811 Warning: So far it only works with the stack pointer on selected
7812 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
7813 work is needed to support other registers and even more so, allocatable
7818 '``llvm.stacksave``' Intrinsic
7819 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7826 declare i8* @llvm.stacksave()
7831 The '``llvm.stacksave``' intrinsic is used to remember the current state
7832 of the function stack, for use with
7833 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
7834 implementing language features like scoped automatic variable sized
7840 This intrinsic returns a opaque pointer value that can be passed to
7841 :ref:`llvm.stackrestore <int_stackrestore>`. When an
7842 ``llvm.stackrestore`` intrinsic is executed with a value saved from
7843 ``llvm.stacksave``, it effectively restores the state of the stack to
7844 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
7845 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
7846 were allocated after the ``llvm.stacksave`` was executed.
7848 .. _int_stackrestore:
7850 '``llvm.stackrestore``' Intrinsic
7851 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7858 declare void @llvm.stackrestore(i8* %ptr)
7863 The '``llvm.stackrestore``' intrinsic is used to restore the state of
7864 the function stack to the state it was in when the corresponding
7865 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
7866 useful for implementing language features like scoped automatic variable
7867 sized arrays in C99.
7872 See the description for :ref:`llvm.stacksave <int_stacksave>`.
7874 '``llvm.prefetch``' Intrinsic
7875 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7882 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7887 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
7888 insert a prefetch instruction if supported; otherwise, it is a noop.
7889 Prefetches have no effect on the behavior of the program but can change
7890 its performance characteristics.
7895 ``address`` is the address to be prefetched, ``rw`` is the specifier
7896 determining if the fetch should be for a read (0) or write (1), and
7897 ``locality`` is a temporal locality specifier ranging from (0) - no
7898 locality, to (3) - extremely local keep in cache. The ``cache type``
7899 specifies whether the prefetch is performed on the data (1) or
7900 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
7901 arguments must be constant integers.
7906 This intrinsic does not modify the behavior of the program. In
7907 particular, prefetches cannot trap and do not produce a value. On
7908 targets that support this intrinsic, the prefetch can provide hints to
7909 the processor cache for better performance.
7911 '``llvm.pcmarker``' Intrinsic
7912 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7919 declare void @llvm.pcmarker(i32 <id>)
7924 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7925 Counter (PC) in a region of code to simulators and other tools. The
7926 method is target specific, but it is expected that the marker will use
7927 exported symbols to transmit the PC of the marker. The marker makes no
7928 guarantees that it will remain with any specific instruction after
7929 optimizations. It is possible that the presence of a marker will inhibit
7930 optimizations. The intended use is to be inserted after optimizations to
7931 allow correlations of simulation runs.
7936 ``id`` is a numerical id identifying the marker.
7941 This intrinsic does not modify the behavior of the program. Backends
7942 that do not support this intrinsic may ignore it.
7944 '``llvm.readcyclecounter``' Intrinsic
7945 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7952 declare i64 @llvm.readcyclecounter()
7957 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
7958 counter register (or similar low latency, high accuracy clocks) on those
7959 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7960 should map to RPCC. As the backing counters overflow quickly (on the
7961 order of 9 seconds on alpha), this should only be used for small
7967 When directly supported, reading the cycle counter should not modify any
7968 memory. Implementations are allowed to either return a application
7969 specific value or a system wide value. On backends without support, this
7970 is lowered to a constant 0.
7972 Note that runtime support may be conditional on the privilege-level code is
7973 running at and the host platform.
7975 '``llvm.clear_cache``' Intrinsic
7976 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7983 declare void @llvm.clear_cache(i8*, i8*)
7988 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
7989 in the specified range to the execution unit of the processor. On
7990 targets with non-unified instruction and data cache, the implementation
7991 flushes the instruction cache.
7996 On platforms with coherent instruction and data caches (e.g. x86), this
7997 intrinsic is a nop. On platforms with non-coherent instruction and data
7998 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
7999 instructions or a system call, if cache flushing requires special
8002 The default behavior is to emit a call to ``__clear_cache`` from the run
8005 This instrinsic does *not* empty the instruction pipeline. Modifications
8006 of the current function are outside the scope of the intrinsic.
8008 '``llvm.instrprof_increment``' Intrinsic
8009 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8016 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
8017 i32 <num-counters>, i32 <index>)
8022 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
8023 frontend for use with instrumentation based profiling. These will be
8024 lowered by the ``-instrprof`` pass to generate execution counts of a
8030 The first argument is a pointer to a global variable containing the
8031 name of the entity being instrumented. This should generally be the
8032 (mangled) function name for a set of counters.
8034 The second argument is a hash value that can be used by the consumer
8035 of the profile data to detect changes to the instrumented source, and
8036 the third is the number of counters associated with ``name``. It is an
8037 error if ``hash`` or ``num-counters`` differ between two instances of
8038 ``instrprof_increment`` that refer to the same name.
8040 The last argument refers to which of the counters for ``name`` should
8041 be incremented. It should be a value between 0 and ``num-counters``.
8046 This intrinsic represents an increment of a profiling counter. It will
8047 cause the ``-instrprof`` pass to generate the appropriate data
8048 structures and the code to increment the appropriate value, in a
8049 format that can be written out by a compiler runtime and consumed via
8050 the ``llvm-profdata`` tool.
8052 Standard C Library Intrinsics
8053 -----------------------------
8055 LLVM provides intrinsics for a few important standard C library
8056 functions. These intrinsics allow source-language front-ends to pass
8057 information about the alignment of the pointer arguments to the code
8058 generator, providing opportunity for more efficient code generation.
8062 '``llvm.memcpy``' Intrinsic
8063 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8068 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
8069 integer bit width and for different address spaces. Not all targets
8070 support all bit widths however.
8074 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
8075 i32 <len>, i32 <align>, i1 <isvolatile>)
8076 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
8077 i64 <len>, i32 <align>, i1 <isvolatile>)
8082 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
8083 source location to the destination location.
8085 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
8086 intrinsics do not return a value, takes extra alignment/isvolatile
8087 arguments and the pointers can be in specified address spaces.
8092 The first argument is a pointer to the destination, the second is a
8093 pointer to the source. The third argument is an integer argument
8094 specifying the number of bytes to copy, the fourth argument is the
8095 alignment of the source and destination locations, and the fifth is a
8096 boolean indicating a volatile access.
8098 If the call to this intrinsic has an alignment value that is not 0 or 1,
8099 then the caller guarantees that both the source and destination pointers
8100 are aligned to that boundary.
8102 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
8103 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
8104 very cleanly specified and it is unwise to depend on it.
8109 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
8110 source location to the destination location, which are not allowed to
8111 overlap. It copies "len" bytes of memory over. If the argument is known
8112 to be aligned to some boundary, this can be specified as the fourth
8113 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
8115 '``llvm.memmove``' Intrinsic
8116 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8121 This is an overloaded intrinsic. You can use llvm.memmove on any integer
8122 bit width and for different address space. Not all targets support all
8127 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
8128 i32 <len>, i32 <align>, i1 <isvolatile>)
8129 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
8130 i64 <len>, i32 <align>, i1 <isvolatile>)
8135 The '``llvm.memmove.*``' intrinsics move a block of memory from the
8136 source location to the destination location. It is similar to the
8137 '``llvm.memcpy``' intrinsic but allows the two memory locations to
8140 Note that, unlike the standard libc function, the ``llvm.memmove.*``
8141 intrinsics do not return a value, takes extra alignment/isvolatile
8142 arguments and the pointers can be in specified address spaces.
8147 The first argument is a pointer to the destination, the second is a
8148 pointer to the source. The third argument is an integer argument
8149 specifying the number of bytes to copy, the fourth argument is the
8150 alignment of the source and destination locations, and the fifth is a
8151 boolean indicating a volatile access.
8153 If the call to this intrinsic has an alignment value that is not 0 or 1,
8154 then the caller guarantees that the source and destination pointers are
8155 aligned to that boundary.
8157 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
8158 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
8159 not very cleanly specified and it is unwise to depend on it.
8164 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
8165 source location to the destination location, which may overlap. It
8166 copies "len" bytes of memory over. If the argument is known to be
8167 aligned to some boundary, this can be specified as the fourth argument,
8168 otherwise it should be set to 0 or 1 (both meaning no alignment).
8170 '``llvm.memset.*``' Intrinsics
8171 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8176 This is an overloaded intrinsic. You can use llvm.memset on any integer
8177 bit width and for different address spaces. However, not all targets
8178 support all bit widths.
8182 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
8183 i32 <len>, i32 <align>, i1 <isvolatile>)
8184 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
8185 i64 <len>, i32 <align>, i1 <isvolatile>)
8190 The '``llvm.memset.*``' intrinsics fill a block of memory with a
8191 particular byte value.
8193 Note that, unlike the standard libc function, the ``llvm.memset``
8194 intrinsic does not return a value and takes extra alignment/volatile
8195 arguments. Also, the destination can be in an arbitrary address space.
8200 The first argument is a pointer to the destination to fill, the second
8201 is the byte value with which to fill it, the third argument is an
8202 integer argument specifying the number of bytes to fill, and the fourth
8203 argument is the known alignment of the destination location.
8205 If the call to this intrinsic has an alignment value that is not 0 or 1,
8206 then the caller guarantees that the destination pointer is aligned to
8209 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
8210 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
8211 very cleanly specified and it is unwise to depend on it.
8216 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
8217 at the destination location. If the argument is known to be aligned to
8218 some boundary, this can be specified as the fourth argument, otherwise
8219 it should be set to 0 or 1 (both meaning no alignment).
8221 '``llvm.sqrt.*``' Intrinsic
8222 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8227 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
8228 floating point or vector of floating point type. Not all targets support
8233 declare float @llvm.sqrt.f32(float %Val)
8234 declare double @llvm.sqrt.f64(double %Val)
8235 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
8236 declare fp128 @llvm.sqrt.f128(fp128 %Val)
8237 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
8242 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
8243 returning the same value as the libm '``sqrt``' functions would. Unlike
8244 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
8245 negative numbers other than -0.0 (which allows for better optimization,
8246 because there is no need to worry about errno being set).
8247 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
8252 The argument and return value are floating point numbers of the same
8258 This function returns the sqrt of the specified operand if it is a
8259 nonnegative floating point number.
8261 '``llvm.powi.*``' Intrinsic
8262 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8267 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
8268 floating point or vector of floating point type. Not all targets support
8273 declare float @llvm.powi.f32(float %Val, i32 %power)
8274 declare double @llvm.powi.f64(double %Val, i32 %power)
8275 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
8276 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
8277 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
8282 The '``llvm.powi.*``' intrinsics return the first operand raised to the
8283 specified (positive or negative) power. The order of evaluation of
8284 multiplications is not defined. When a vector of floating point type is
8285 used, the second argument remains a scalar integer value.
8290 The second argument is an integer power, and the first is a value to
8291 raise to that power.
8296 This function returns the first value raised to the second power with an
8297 unspecified sequence of rounding operations.
8299 '``llvm.sin.*``' Intrinsic
8300 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8305 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
8306 floating point or vector of floating point type. Not all targets support
8311 declare float @llvm.sin.f32(float %Val)
8312 declare double @llvm.sin.f64(double %Val)
8313 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
8314 declare fp128 @llvm.sin.f128(fp128 %Val)
8315 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
8320 The '``llvm.sin.*``' intrinsics return the sine of the operand.
8325 The argument and return value are floating point numbers of the same
8331 This function returns the sine of the specified operand, returning the
8332 same values as the libm ``sin`` functions would, and handles error
8333 conditions in the same way.
8335 '``llvm.cos.*``' Intrinsic
8336 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8341 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
8342 floating point or vector of floating point type. Not all targets support
8347 declare float @llvm.cos.f32(float %Val)
8348 declare double @llvm.cos.f64(double %Val)
8349 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
8350 declare fp128 @llvm.cos.f128(fp128 %Val)
8351 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
8356 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
8361 The argument and return value are floating point numbers of the same
8367 This function returns the cosine of the specified operand, returning the
8368 same values as the libm ``cos`` functions would, and handles error
8369 conditions in the same way.
8371 '``llvm.pow.*``' Intrinsic
8372 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8377 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
8378 floating point or vector of floating point type. Not all targets support
8383 declare float @llvm.pow.f32(float %Val, float %Power)
8384 declare double @llvm.pow.f64(double %Val, double %Power)
8385 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
8386 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
8387 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
8392 The '``llvm.pow.*``' intrinsics return the first operand raised to the
8393 specified (positive or negative) power.
8398 The second argument is a floating point power, and the first is a value
8399 to raise to that power.
8404 This function returns the first value raised to the second power,
8405 returning the same values as the libm ``pow`` functions would, and
8406 handles error conditions in the same way.
8408 '``llvm.exp.*``' Intrinsic
8409 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8414 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
8415 floating point or vector of floating point type. Not all targets support
8420 declare float @llvm.exp.f32(float %Val)
8421 declare double @llvm.exp.f64(double %Val)
8422 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
8423 declare fp128 @llvm.exp.f128(fp128 %Val)
8424 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
8429 The '``llvm.exp.*``' intrinsics perform the exp function.
8434 The argument and return value are floating point numbers of the same
8440 This function returns the same values as the libm ``exp`` functions
8441 would, and handles error conditions in the same way.
8443 '``llvm.exp2.*``' Intrinsic
8444 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8449 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
8450 floating point or vector of floating point type. Not all targets support
8455 declare float @llvm.exp2.f32(float %Val)
8456 declare double @llvm.exp2.f64(double %Val)
8457 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
8458 declare fp128 @llvm.exp2.f128(fp128 %Val)
8459 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
8464 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
8469 The argument and return value are floating point numbers of the same
8475 This function returns the same values as the libm ``exp2`` functions
8476 would, and handles error conditions in the same way.
8478 '``llvm.log.*``' Intrinsic
8479 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8484 This is an overloaded intrinsic. You can use ``llvm.log`` on any
8485 floating point or vector of floating point type. Not all targets support
8490 declare float @llvm.log.f32(float %Val)
8491 declare double @llvm.log.f64(double %Val)
8492 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
8493 declare fp128 @llvm.log.f128(fp128 %Val)
8494 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
8499 The '``llvm.log.*``' intrinsics perform the log function.
8504 The argument and return value are floating point numbers of the same
8510 This function returns the same values as the libm ``log`` functions
8511 would, and handles error conditions in the same way.
8513 '``llvm.log10.*``' Intrinsic
8514 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8519 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
8520 floating point or vector of floating point type. Not all targets support
8525 declare float @llvm.log10.f32(float %Val)
8526 declare double @llvm.log10.f64(double %Val)
8527 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
8528 declare fp128 @llvm.log10.f128(fp128 %Val)
8529 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
8534 The '``llvm.log10.*``' intrinsics perform the log10 function.
8539 The argument and return value are floating point numbers of the same
8545 This function returns the same values as the libm ``log10`` functions
8546 would, and handles error conditions in the same way.
8548 '``llvm.log2.*``' Intrinsic
8549 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8554 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
8555 floating point or vector of floating point type. Not all targets support
8560 declare float @llvm.log2.f32(float %Val)
8561 declare double @llvm.log2.f64(double %Val)
8562 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
8563 declare fp128 @llvm.log2.f128(fp128 %Val)
8564 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
8569 The '``llvm.log2.*``' intrinsics perform the log2 function.
8574 The argument and return value are floating point numbers of the same
8580 This function returns the same values as the libm ``log2`` functions
8581 would, and handles error conditions in the same way.
8583 '``llvm.fma.*``' Intrinsic
8584 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8589 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
8590 floating point or vector of floating point type. Not all targets support
8595 declare float @llvm.fma.f32(float %a, float %b, float %c)
8596 declare double @llvm.fma.f64(double %a, double %b, double %c)
8597 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
8598 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
8599 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
8604 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
8610 The argument and return value are floating point numbers of the same
8616 This function returns the same values as the libm ``fma`` functions
8617 would, and does not set errno.
8619 '``llvm.fabs.*``' Intrinsic
8620 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8625 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
8626 floating point or vector of floating point type. Not all targets support
8631 declare float @llvm.fabs.f32(float %Val)
8632 declare double @llvm.fabs.f64(double %Val)
8633 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
8634 declare fp128 @llvm.fabs.f128(fp128 %Val)
8635 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
8640 The '``llvm.fabs.*``' intrinsics return the absolute value of the
8646 The argument and return value are floating point numbers of the same
8652 This function returns the same values as the libm ``fabs`` functions
8653 would, and handles error conditions in the same way.
8655 '``llvm.minnum.*``' Intrinsic
8656 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8661 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
8662 floating point or vector of floating point type. Not all targets support
8667 declare float @llvm.minnum.f32(float %Val0, float %Val1)
8668 declare double @llvm.minnum.f64(double %Val0, double %Val1)
8669 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8670 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
8671 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8676 The '``llvm.minnum.*``' intrinsics return the minimum of the two
8683 The arguments and return value are floating point numbers of the same
8689 Follows the IEEE-754 semantics for minNum, which also match for libm's
8692 If either operand is a NaN, returns the other non-NaN operand. Returns
8693 NaN only if both operands are NaN. If the operands compare equal,
8694 returns a value that compares equal to both operands. This means that
8695 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8697 '``llvm.maxnum.*``' Intrinsic
8698 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8703 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
8704 floating point or vector of floating point type. Not all targets support
8709 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
8710 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
8711 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8712 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
8713 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8718 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
8725 The arguments and return value are floating point numbers of the same
8730 Follows the IEEE-754 semantics for maxNum, which also match for libm's
8733 If either operand is a NaN, returns the other non-NaN operand. Returns
8734 NaN only if both operands are NaN. If the operands compare equal,
8735 returns a value that compares equal to both operands. This means that
8736 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8738 '``llvm.copysign.*``' Intrinsic
8739 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8744 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
8745 floating point or vector of floating point type. Not all targets support
8750 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
8751 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
8752 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
8753 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
8754 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
8759 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
8760 first operand and the sign of the second operand.
8765 The arguments and return value are floating point numbers of the same
8771 This function returns the same values as the libm ``copysign``
8772 functions would, and handles error conditions in the same way.
8774 '``llvm.floor.*``' Intrinsic
8775 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8780 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
8781 floating point or vector of floating point type. Not all targets support
8786 declare float @llvm.floor.f32(float %Val)
8787 declare double @llvm.floor.f64(double %Val)
8788 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
8789 declare fp128 @llvm.floor.f128(fp128 %Val)
8790 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
8795 The '``llvm.floor.*``' intrinsics return the floor of the operand.
8800 The argument and return value are floating point numbers of the same
8806 This function returns the same values as the libm ``floor`` functions
8807 would, and handles error conditions in the same way.
8809 '``llvm.ceil.*``' Intrinsic
8810 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8815 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
8816 floating point or vector of floating point type. Not all targets support
8821 declare float @llvm.ceil.f32(float %Val)
8822 declare double @llvm.ceil.f64(double %Val)
8823 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
8824 declare fp128 @llvm.ceil.f128(fp128 %Val)
8825 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
8830 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
8835 The argument and return value are floating point numbers of the same
8841 This function returns the same values as the libm ``ceil`` functions
8842 would, and handles error conditions in the same way.
8844 '``llvm.trunc.*``' Intrinsic
8845 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8850 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
8851 floating point or vector of floating point type. Not all targets support
8856 declare float @llvm.trunc.f32(float %Val)
8857 declare double @llvm.trunc.f64(double %Val)
8858 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
8859 declare fp128 @llvm.trunc.f128(fp128 %Val)
8860 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
8865 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
8866 nearest integer not larger in magnitude than the operand.
8871 The argument and return value are floating point numbers of the same
8877 This function returns the same values as the libm ``trunc`` functions
8878 would, and handles error conditions in the same way.
8880 '``llvm.rint.*``' Intrinsic
8881 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8886 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
8887 floating point or vector of floating point type. Not all targets support
8892 declare float @llvm.rint.f32(float %Val)
8893 declare double @llvm.rint.f64(double %Val)
8894 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
8895 declare fp128 @llvm.rint.f128(fp128 %Val)
8896 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
8901 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
8902 nearest integer. It may raise an inexact floating-point exception if the
8903 operand isn't an integer.
8908 The argument and return value are floating point numbers of the same
8914 This function returns the same values as the libm ``rint`` functions
8915 would, and handles error conditions in the same way.
8917 '``llvm.nearbyint.*``' Intrinsic
8918 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8923 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
8924 floating point or vector of floating point type. Not all targets support
8929 declare float @llvm.nearbyint.f32(float %Val)
8930 declare double @llvm.nearbyint.f64(double %Val)
8931 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
8932 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
8933 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
8938 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
8944 The argument and return value are floating point numbers of the same
8950 This function returns the same values as the libm ``nearbyint``
8951 functions would, and handles error conditions in the same way.
8953 '``llvm.round.*``' Intrinsic
8954 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8959 This is an overloaded intrinsic. You can use ``llvm.round`` on any
8960 floating point or vector of floating point type. Not all targets support
8965 declare float @llvm.round.f32(float %Val)
8966 declare double @llvm.round.f64(double %Val)
8967 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
8968 declare fp128 @llvm.round.f128(fp128 %Val)
8969 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
8974 The '``llvm.round.*``' intrinsics returns the operand rounded to the
8980 The argument and return value are floating point numbers of the same
8986 This function returns the same values as the libm ``round``
8987 functions would, and handles error conditions in the same way.
8989 Bit Manipulation Intrinsics
8990 ---------------------------
8992 LLVM provides intrinsics for a few important bit manipulation
8993 operations. These allow efficient code generation for some algorithms.
8995 '``llvm.bswap.*``' Intrinsics
8996 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9001 This is an overloaded intrinsic function. You can use bswap on any
9002 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
9006 declare i16 @llvm.bswap.i16(i16 <id>)
9007 declare i32 @llvm.bswap.i32(i32 <id>)
9008 declare i64 @llvm.bswap.i64(i64 <id>)
9013 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
9014 values with an even number of bytes (positive multiple of 16 bits).
9015 These are useful for performing operations on data that is not in the
9016 target's native byte order.
9021 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
9022 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
9023 intrinsic returns an i32 value that has the four bytes of the input i32
9024 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
9025 returned i32 will have its bytes in 3, 2, 1, 0 order. The
9026 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
9027 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
9030 '``llvm.ctpop.*``' Intrinsic
9031 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9036 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
9037 bit width, or on any vector with integer elements. Not all targets
9038 support all bit widths or vector types, however.
9042 declare i8 @llvm.ctpop.i8(i8 <src>)
9043 declare i16 @llvm.ctpop.i16(i16 <src>)
9044 declare i32 @llvm.ctpop.i32(i32 <src>)
9045 declare i64 @llvm.ctpop.i64(i64 <src>)
9046 declare i256 @llvm.ctpop.i256(i256 <src>)
9047 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
9052 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
9058 The only argument is the value to be counted. The argument may be of any
9059 integer type, or a vector with integer elements. The return type must
9060 match the argument type.
9065 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
9066 each element of a vector.
9068 '``llvm.ctlz.*``' Intrinsic
9069 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9074 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
9075 integer bit width, or any vector whose elements are integers. Not all
9076 targets support all bit widths or vector types, however.
9080 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
9081 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
9082 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
9083 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
9084 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
9085 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
9090 The '``llvm.ctlz``' family of intrinsic functions counts the number of
9091 leading zeros in a variable.
9096 The first argument is the value to be counted. This argument may be of
9097 any integer type, or a vector with integer element type. The return
9098 type must match the first argument type.
9100 The second argument must be a constant and is a flag to indicate whether
9101 the intrinsic should ensure that a zero as the first argument produces a
9102 defined result. Historically some architectures did not provide a
9103 defined result for zero values as efficiently, and many algorithms are
9104 now predicated on avoiding zero-value inputs.
9109 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
9110 zeros in a variable, or within each element of the vector. If
9111 ``src == 0`` then the result is the size in bits of the type of ``src``
9112 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9113 ``llvm.ctlz(i32 2) = 30``.
9115 '``llvm.cttz.*``' Intrinsic
9116 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9121 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
9122 integer bit width, or any vector of integer elements. Not all targets
9123 support all bit widths or vector types, however.
9127 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
9128 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
9129 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
9130 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
9131 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
9132 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
9137 The '``llvm.cttz``' family of intrinsic functions counts the number of
9143 The first argument is the value to be counted. This argument may be of
9144 any integer type, or a vector with integer element type. The return
9145 type must match the first argument type.
9147 The second argument must be a constant and is a flag to indicate whether
9148 the intrinsic should ensure that a zero as the first argument produces a
9149 defined result. Historically some architectures did not provide a
9150 defined result for zero values as efficiently, and many algorithms are
9151 now predicated on avoiding zero-value inputs.
9156 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
9157 zeros in a variable, or within each element of a vector. If ``src == 0``
9158 then the result is the size in bits of the type of ``src`` if
9159 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9160 ``llvm.cttz(2) = 1``.
9164 Arithmetic with Overflow Intrinsics
9165 -----------------------------------
9167 LLVM provides intrinsics for some arithmetic with overflow operations.
9169 '``llvm.sadd.with.overflow.*``' Intrinsics
9170 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9175 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
9176 on any integer bit width.
9180 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
9181 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9182 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
9187 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9188 a signed addition of the two arguments, and indicate whether an overflow
9189 occurred during the signed summation.
9194 The arguments (%a and %b) and the first element of the result structure
9195 may be of integer types of any bit width, but they must have the same
9196 bit width. The second element of the result structure must be of type
9197 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9203 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9204 a signed addition of the two variables. They return a structure --- the
9205 first element of which is the signed summation, and the second element
9206 of which is a bit specifying if the signed summation resulted in an
9212 .. code-block:: llvm
9214 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9215 %sum = extractvalue {i32, i1} %res, 0
9216 %obit = extractvalue {i32, i1} %res, 1
9217 br i1 %obit, label %overflow, label %normal
9219 '``llvm.uadd.with.overflow.*``' Intrinsics
9220 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9225 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
9226 on any integer bit width.
9230 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
9231 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
9232 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
9237 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9238 an unsigned addition of the two arguments, and indicate whether a carry
9239 occurred during the unsigned summation.
9244 The arguments (%a and %b) and the first element of the result structure
9245 may be of integer types of any bit width, but they must have the same
9246 bit width. The second element of the result structure must be of type
9247 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9253 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9254 an unsigned addition of the two arguments. They return a structure --- the
9255 first element of which is the sum, and the second element of which is a
9256 bit specifying if the unsigned summation resulted in a carry.
9261 .. code-block:: llvm
9263 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
9264 %sum = extractvalue {i32, i1} %res, 0
9265 %obit = extractvalue {i32, i1} %res, 1
9266 br i1 %obit, label %carry, label %normal
9268 '``llvm.ssub.with.overflow.*``' Intrinsics
9269 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9274 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
9275 on any integer bit width.
9279 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
9280 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
9281 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
9286 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
9287 a signed subtraction of the two arguments, and indicate whether an
9288 overflow occurred during the signed subtraction.
9293 The arguments (%a and %b) and the first element of the result structure
9294 may be of integer types of any bit width, but they must have the same
9295 bit width. The second element of the result structure must be of type
9296 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9302 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
9303 a signed subtraction of the two arguments. They return a structure --- the
9304 first element of which is the subtraction, and the second element of
9305 which is a bit specifying if the signed subtraction resulted in an
9311 .. code-block:: llvm
9313 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
9314 %sum = extractvalue {i32, i1} %res, 0
9315 %obit = extractvalue {i32, i1} %res, 1
9316 br i1 %obit, label %overflow, label %normal
9318 '``llvm.usub.with.overflow.*``' Intrinsics
9319 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9324 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
9325 on any integer bit width.
9329 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
9330 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
9331 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
9336 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
9337 an unsigned subtraction of the two arguments, and indicate whether an
9338 overflow occurred during the unsigned subtraction.
9343 The arguments (%a and %b) and the first element of the result structure
9344 may be of integer types of any bit width, but they must have the same
9345 bit width. The second element of the result structure must be of type
9346 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9352 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
9353 an unsigned subtraction of the two arguments. They return a structure ---
9354 the first element of which is the subtraction, and the second element of
9355 which is a bit specifying if the unsigned subtraction resulted in an
9361 .. code-block:: llvm
9363 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
9364 %sum = extractvalue {i32, i1} %res, 0
9365 %obit = extractvalue {i32, i1} %res, 1
9366 br i1 %obit, label %overflow, label %normal
9368 '``llvm.smul.with.overflow.*``' Intrinsics
9369 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9374 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
9375 on any integer bit width.
9379 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
9380 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
9381 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
9386 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
9387 a signed multiplication of the two arguments, and indicate whether an
9388 overflow occurred during the signed multiplication.
9393 The arguments (%a and %b) and the first element of the result structure
9394 may be of integer types of any bit width, but they must have the same
9395 bit width. The second element of the result structure must be of type
9396 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9402 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
9403 a signed multiplication of the two arguments. They return a structure ---
9404 the first element of which is the multiplication, and the second element
9405 of which is a bit specifying if the signed multiplication resulted in an
9411 .. code-block:: llvm
9413 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
9414 %sum = extractvalue {i32, i1} %res, 0
9415 %obit = extractvalue {i32, i1} %res, 1
9416 br i1 %obit, label %overflow, label %normal
9418 '``llvm.umul.with.overflow.*``' Intrinsics
9419 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9424 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
9425 on any integer bit width.
9429 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
9430 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9431 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
9436 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9437 a unsigned multiplication of the two arguments, and indicate whether an
9438 overflow occurred during the unsigned multiplication.
9443 The arguments (%a and %b) and the first element of the result structure
9444 may be of integer types of any bit width, but they must have the same
9445 bit width. The second element of the result structure must be of type
9446 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9452 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9453 an unsigned multiplication of the two arguments. They return a structure ---
9454 the first element of which is the multiplication, and the second
9455 element of which is a bit specifying if the unsigned multiplication
9456 resulted in an overflow.
9461 .. code-block:: llvm
9463 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9464 %sum = extractvalue {i32, i1} %res, 0
9465 %obit = extractvalue {i32, i1} %res, 1
9466 br i1 %obit, label %overflow, label %normal
9468 Specialised Arithmetic Intrinsics
9469 ---------------------------------
9471 '``llvm.fmuladd.*``' Intrinsic
9472 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9479 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
9480 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
9485 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
9486 expressions that can be fused if the code generator determines that (a) the
9487 target instruction set has support for a fused operation, and (b) that the
9488 fused operation is more efficient than the equivalent, separate pair of mul
9489 and add instructions.
9494 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
9495 multiplicands, a and b, and an addend c.
9504 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
9506 is equivalent to the expression a \* b + c, except that rounding will
9507 not be performed between the multiplication and addition steps if the
9508 code generator fuses the operations. Fusion is not guaranteed, even if
9509 the target platform supports it. If a fused multiply-add is required the
9510 corresponding llvm.fma.\* intrinsic function should be used
9511 instead. This never sets errno, just as '``llvm.fma.*``'.
9516 .. code-block:: llvm
9518 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
9520 Half Precision Floating Point Intrinsics
9521 ----------------------------------------
9523 For most target platforms, half precision floating point is a
9524 storage-only format. This means that it is a dense encoding (in memory)
9525 but does not support computation in the format.
9527 This means that code must first load the half-precision floating point
9528 value as an i16, then convert it to float with
9529 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
9530 then be performed on the float value (including extending to double
9531 etc). To store the value back to memory, it is first converted to float
9532 if needed, then converted to i16 with
9533 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
9536 .. _int_convert_to_fp16:
9538 '``llvm.convert.to.fp16``' Intrinsic
9539 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9546 declare i16 @llvm.convert.to.fp16.f32(float %a)
9547 declare i16 @llvm.convert.to.fp16.f64(double %a)
9552 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9553 conventional floating point type to half precision floating point format.
9558 The intrinsic function contains single argument - the value to be
9564 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9565 conventional floating point format to half precision floating point format. The
9566 return value is an ``i16`` which contains the converted number.
9571 .. code-block:: llvm
9573 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
9574 store i16 %res, i16* @x, align 2
9576 .. _int_convert_from_fp16:
9578 '``llvm.convert.from.fp16``' Intrinsic
9579 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9586 declare float @llvm.convert.from.fp16.f32(i16 %a)
9587 declare double @llvm.convert.from.fp16.f64(i16 %a)
9592 The '``llvm.convert.from.fp16``' intrinsic function performs a
9593 conversion from half precision floating point format to single precision
9594 floating point format.
9599 The intrinsic function contains single argument - the value to be
9605 The '``llvm.convert.from.fp16``' intrinsic function performs a
9606 conversion from half single precision floating point format to single
9607 precision floating point format. The input half-float value is
9608 represented by an ``i16`` value.
9613 .. code-block:: llvm
9615 %a = load i16, i16* @x, align 2
9616 %res = call float @llvm.convert.from.fp16(i16 %a)
9623 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
9624 prefix), are described in the `LLVM Source Level
9625 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
9628 Exception Handling Intrinsics
9629 -----------------------------
9631 The LLVM exception handling intrinsics (which all start with
9632 ``llvm.eh.`` prefix), are described in the `LLVM Exception
9633 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
9637 Trampoline Intrinsics
9638 ---------------------
9640 These intrinsics make it possible to excise one parameter, marked with
9641 the :ref:`nest <nest>` attribute, from a function. The result is a
9642 callable function pointer lacking the nest parameter - the caller does
9643 not need to provide a value for it. Instead, the value to use is stored
9644 in advance in a "trampoline", a block of memory usually allocated on the
9645 stack, which also contains code to splice the nest value into the
9646 argument list. This is used to implement the GCC nested function address
9649 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
9650 then the resulting function pointer has signature ``i32 (i32, i32)*``.
9651 It can be created as follows:
9653 .. code-block:: llvm
9655 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
9656 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
9657 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
9658 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
9659 %fp = bitcast i8* %p to i32 (i32, i32)*
9661 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
9662 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
9666 '``llvm.init.trampoline``' Intrinsic
9667 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9674 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
9679 This fills the memory pointed to by ``tramp`` with executable code,
9680 turning it into a trampoline.
9685 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
9686 pointers. The ``tramp`` argument must point to a sufficiently large and
9687 sufficiently aligned block of memory; this memory is written to by the
9688 intrinsic. Note that the size and the alignment are target-specific -
9689 LLVM currently provides no portable way of determining them, so a
9690 front-end that generates this intrinsic needs to have some
9691 target-specific knowledge. The ``func`` argument must hold a function
9692 bitcast to an ``i8*``.
9697 The block of memory pointed to by ``tramp`` is filled with target
9698 dependent code, turning it into a function. Then ``tramp`` needs to be
9699 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
9700 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
9701 function's signature is the same as that of ``func`` with any arguments
9702 marked with the ``nest`` attribute removed. At most one such ``nest``
9703 argument is allowed, and it must be of pointer type. Calling the new
9704 function is equivalent to calling ``func`` with the same argument list,
9705 but with ``nval`` used for the missing ``nest`` argument. If, after
9706 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
9707 modified, then the effect of any later call to the returned function
9708 pointer is undefined.
9712 '``llvm.adjust.trampoline``' Intrinsic
9713 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9720 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
9725 This performs any required machine-specific adjustment to the address of
9726 a trampoline (passed as ``tramp``).
9731 ``tramp`` must point to a block of memory which already has trampoline
9732 code filled in by a previous call to
9733 :ref:`llvm.init.trampoline <int_it>`.
9738 On some architectures the address of the code to be executed needs to be
9739 different than the address where the trampoline is actually stored. This
9740 intrinsic returns the executable address corresponding to ``tramp``
9741 after performing the required machine specific adjustments. The pointer
9742 returned can then be :ref:`bitcast and executed <int_trampoline>`.
9744 Masked Vector Load and Store Intrinsics
9745 ---------------------------------------
9747 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.
9751 '``llvm.masked.load.*``' Intrinsics
9752 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9756 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
9760 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
9761 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
9766 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 in the passthru operand.
9772 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 'i1' 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 passthru operand are the same vector types.
9778 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.
9779 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.
9784 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
9786 ;; The result of the two following instructions is identical aside from potential memory access exception
9787 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
9788 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
9792 '``llvm.masked.store.*``' Intrinsics
9793 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9797 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
9801 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
9802 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
9807 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.
9812 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.
9818 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.
9819 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.
9823 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
9825 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
9826 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
9827 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
9828 store <16 x float> %res, <16 x float>* %ptr, align 4
9834 This class of intrinsics provides information about the lifetime of
9835 memory objects and ranges where variables are immutable.
9839 '``llvm.lifetime.start``' Intrinsic
9840 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9847 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
9852 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
9858 The first argument is a constant integer representing the size of the
9859 object, or -1 if it is variable sized. The second argument is a pointer
9865 This intrinsic indicates that before this point in the code, the value
9866 of the memory pointed to by ``ptr`` is dead. This means that it is known
9867 to never be used and has an undefined value. A load from the pointer
9868 that precedes this intrinsic can be replaced with ``'undef'``.
9872 '``llvm.lifetime.end``' Intrinsic
9873 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9880 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
9885 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
9891 The first argument is a constant integer representing the size of the
9892 object, or -1 if it is variable sized. The second argument is a pointer
9898 This intrinsic indicates that after this point in the code, the value of
9899 the memory pointed to by ``ptr`` is dead. This means that it is known to
9900 never be used and has an undefined value. Any stores into the memory
9901 object following this intrinsic may be removed as dead.
9903 '``llvm.invariant.start``' Intrinsic
9904 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9911 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
9916 The '``llvm.invariant.start``' intrinsic specifies that the contents of
9917 a memory object will not change.
9922 The first argument is a constant integer representing the size of the
9923 object, or -1 if it is variable sized. The second argument is a pointer
9929 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
9930 the return value, the referenced memory location is constant and
9933 '``llvm.invariant.end``' Intrinsic
9934 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9941 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
9946 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
9947 memory object are mutable.
9952 The first argument is the matching ``llvm.invariant.start`` intrinsic.
9953 The second argument is a constant integer representing the size of the
9954 object, or -1 if it is variable sized and the third argument is a
9955 pointer to the object.
9960 This intrinsic indicates that the memory is mutable again.
9965 This class of intrinsics is designed to be generic and has no specific
9968 '``llvm.var.annotation``' Intrinsic
9969 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9976 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9981 The '``llvm.var.annotation``' intrinsic.
9986 The first argument is a pointer to a value, the second is a pointer to a
9987 global string, the third is a pointer to a global string which is the
9988 source file name, and the last argument is the line number.
9993 This intrinsic allows annotation of local variables with arbitrary
9994 strings. This can be useful for special purpose optimizations that want
9995 to look for these annotations. These have no other defined use; they are
9996 ignored by code generation and optimization.
9998 '``llvm.ptr.annotation.*``' Intrinsic
9999 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10004 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
10005 pointer to an integer of any width. *NOTE* you must specify an address space for
10006 the pointer. The identifier for the default address space is the integer
10011 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
10012 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
10013 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
10014 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
10015 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
10020 The '``llvm.ptr.annotation``' intrinsic.
10025 The first argument is a pointer to an integer value of arbitrary bitwidth
10026 (result of some expression), the second is a pointer to a global string, the
10027 third is a pointer to a global string which is the source file name, and the
10028 last argument is the line number. It returns the value of the first argument.
10033 This intrinsic allows annotation of a pointer to an integer with arbitrary
10034 strings. This can be useful for special purpose optimizations that want to look
10035 for these annotations. These have no other defined use; they are ignored by code
10036 generation and optimization.
10038 '``llvm.annotation.*``' Intrinsic
10039 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10044 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
10045 any integer bit width.
10049 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
10050 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
10051 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
10052 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
10053 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
10058 The '``llvm.annotation``' intrinsic.
10063 The first argument is an integer value (result of some expression), the
10064 second is a pointer to a global string, the third is a pointer to a
10065 global string which is the source file name, and the last argument is
10066 the line number. It returns the value of the first argument.
10071 This intrinsic allows annotations to be put on arbitrary expressions
10072 with arbitrary strings. This can be useful for special purpose
10073 optimizations that want to look for these annotations. These have no
10074 other defined use; they are ignored by code generation and optimization.
10076 '``llvm.trap``' Intrinsic
10077 ^^^^^^^^^^^^^^^^^^^^^^^^^
10084 declare void @llvm.trap() noreturn nounwind
10089 The '``llvm.trap``' intrinsic.
10099 This intrinsic is lowered to the target dependent trap instruction. If
10100 the target does not have a trap instruction, this intrinsic will be
10101 lowered to a call of the ``abort()`` function.
10103 '``llvm.debugtrap``' Intrinsic
10104 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10111 declare void @llvm.debugtrap() nounwind
10116 The '``llvm.debugtrap``' intrinsic.
10126 This intrinsic is lowered to code which is intended to cause an
10127 execution trap with the intention of requesting the attention of a
10130 '``llvm.stackprotector``' Intrinsic
10131 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10138 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
10143 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
10144 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
10145 is placed on the stack before local variables.
10150 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
10151 The first argument is the value loaded from the stack guard
10152 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
10153 enough space to hold the value of the guard.
10158 This intrinsic causes the prologue/epilogue inserter to force the position of
10159 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
10160 to ensure that if a local variable on the stack is overwritten, it will destroy
10161 the value of the guard. When the function exits, the guard on the stack is
10162 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
10163 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
10164 calling the ``__stack_chk_fail()`` function.
10166 '``llvm.stackprotectorcheck``' Intrinsic
10167 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10174 declare void @llvm.stackprotectorcheck(i8** <guard>)
10179 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
10180 created stack protector and if they are not equal calls the
10181 ``__stack_chk_fail()`` function.
10186 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
10187 the variable ``@__stack_chk_guard``.
10192 This intrinsic is provided to perform the stack protector check by comparing
10193 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
10194 values do not match call the ``__stack_chk_fail()`` function.
10196 The reason to provide this as an IR level intrinsic instead of implementing it
10197 via other IR operations is that in order to perform this operation at the IR
10198 level without an intrinsic, one would need to create additional basic blocks to
10199 handle the success/failure cases. This makes it difficult to stop the stack
10200 protector check from disrupting sibling tail calls in Codegen. With this
10201 intrinsic, we are able to generate the stack protector basic blocks late in
10202 codegen after the tail call decision has occurred.
10204 '``llvm.objectsize``' Intrinsic
10205 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10212 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
10213 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
10218 The ``llvm.objectsize`` intrinsic is designed to provide information to
10219 the optimizers to determine at compile time whether a) an operation
10220 (like memcpy) will overflow a buffer that corresponds to an object, or
10221 b) that a runtime check for overflow isn't necessary. An object in this
10222 context means an allocation of a specific class, structure, array, or
10228 The ``llvm.objectsize`` intrinsic takes two arguments. The first
10229 argument is a pointer to or into the ``object``. The second argument is
10230 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
10231 or -1 (if false) when the object size is unknown. The second argument
10232 only accepts constants.
10237 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
10238 the size of the object concerned. If the size cannot be determined at
10239 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
10240 on the ``min`` argument).
10242 '``llvm.expect``' Intrinsic
10243 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10248 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
10253 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
10254 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
10255 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
10260 The ``llvm.expect`` intrinsic provides information about expected (the
10261 most probable) value of ``val``, which can be used by optimizers.
10266 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
10267 a value. The second argument is an expected value, this needs to be a
10268 constant value, variables are not allowed.
10273 This intrinsic is lowered to the ``val``.
10275 '``llvm.assume``' Intrinsic
10276 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10283 declare void @llvm.assume(i1 %cond)
10288 The ``llvm.assume`` allows the optimizer to assume that the provided
10289 condition is true. This information can then be used in simplifying other parts
10295 The condition which the optimizer may assume is always true.
10300 The intrinsic allows the optimizer to assume that the provided condition is
10301 always true whenever the control flow reaches the intrinsic call. No code is
10302 generated for this intrinsic, and instructions that contribute only to the
10303 provided condition are not used for code generation. If the condition is
10304 violated during execution, the behavior is undefined.
10306 Note that the optimizer might limit the transformations performed on values
10307 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
10308 only used to form the intrinsic's input argument. This might prove undesirable
10309 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
10310 sufficient overall improvement in code quality. For this reason,
10311 ``llvm.assume`` should not be used to document basic mathematical invariants
10312 that the optimizer can otherwise deduce or facts that are of little use to the
10317 '``llvm.bitset.test``' Intrinsic
10318 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10325 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
10331 The first argument is a pointer to be tested. The second argument is a
10332 metadata string containing the name of a :doc:`bitset <BitSets>`.
10337 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
10338 member of the given bitset.
10340 '``llvm.donothing``' Intrinsic
10341 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10348 declare void @llvm.donothing() nounwind readnone
10353 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
10354 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
10355 with an invoke instruction.
10365 This intrinsic does nothing, and it's removed by optimizers and ignored
10368 Stack Map Intrinsics
10369 --------------------
10371 LLVM provides experimental intrinsics to support runtime patching
10372 mechanisms commonly desired in dynamic language JITs. These intrinsics
10373 are described in :doc:`StackMaps`.