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]* @.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 dynamcially
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 little
357 intrusive as possible. This calling convention behaves identical 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 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 variables 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 aliasaes 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 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 know 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 Alias 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 Names
1018 -----------------------
1020 Each function may specify a garbage collector name, which is simply a
1023 .. code-block:: llvm
1025 define void @f() gc "name" { ... }
1027 The compiler declares the supported values of *name*. Specifying a
1028 collector will cause the compiler to alter its output in order to
1029 support the named garbage collection algorithm.
1036 Prefix data is data associated with a function which the code
1037 generator will emit immediately before the function's entrypoint.
1038 The purpose of this feature is to allow frontends to associate
1039 language-specific runtime metadata with specific functions and make it
1040 available through the function pointer while still allowing the
1041 function pointer to be called.
1043 To access the data for a given function, a program may bitcast the
1044 function pointer to a pointer to the constant's type and dereference
1045 index -1. This implies that the IR symbol points just past the end of
1046 the prefix data. For instance, take the example of a function annotated
1047 with a single ``i32``,
1049 .. code-block:: llvm
1051 define void @f() prefix i32 123 { ... }
1053 The prefix data can be referenced as,
1055 .. code-block:: llvm
1057 %0 = bitcast *void () @f to *i32
1058 %a = getelementptr inbounds *i32 %0, i32 -1
1061 Prefix data is laid out as if it were an initializer for a global variable
1062 of the prefix data's type. The function will be placed such that the
1063 beginning of the prefix data is aligned. This means that if the size
1064 of the prefix data is not a multiple of the alignment size, the
1065 function's entrypoint will not be aligned. If alignment of the
1066 function's entrypoint is desired, padding must be added to the prefix
1069 A function may have prefix data but no body. This has similar semantics
1070 to the ``available_externally`` linkage in that the data may be used by the
1071 optimizers but will not be emitted in the object file.
1078 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1079 be inserted prior to the function body. This can be used for enabling
1080 function hot-patching and instrumentation.
1082 To maintain the semantics of ordinary function calls, the prologue data must
1083 have a particular format. Specifically, it must begin with a sequence of
1084 bytes which decode to a sequence of machine instructions, valid for the
1085 module's target, which transfer control to the point immediately succeeding
1086 the prologue data, without performing any other visible action. This allows
1087 the inliner and other passes to reason about the semantics of the function
1088 definition without needing to reason about the prologue data. Obviously this
1089 makes the format of the prologue data highly target dependent.
1091 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1092 which encodes the ``nop`` instruction:
1094 .. code-block:: llvm
1096 define void @f() prologue i8 144 { ... }
1098 Generally prologue data can be formed by encoding a relative branch instruction
1099 which skips the metadata, as in this example of valid prologue data for the
1100 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1102 .. code-block:: llvm
1104 %0 = type <{ i8, i8, i8* }>
1106 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1108 A function may have prologue data but no body. This has similar semantics
1109 to the ``available_externally`` linkage in that the data may be used by the
1110 optimizers but will not be emitted in the object file.
1117 Attribute groups are groups of attributes that are referenced by objects within
1118 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1119 functions will use the same set of attributes. In the degenerative case of a
1120 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1121 group will capture the important command line flags used to build that file.
1123 An attribute group is a module-level object. To use an attribute group, an
1124 object references the attribute group's ID (e.g. ``#37``). An object may refer
1125 to more than one attribute group. In that situation, the attributes from the
1126 different groups are merged.
1128 Here is an example of attribute groups for a function that should always be
1129 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1131 .. code-block:: llvm
1133 ; Target-independent attributes:
1134 attributes #0 = { alwaysinline alignstack=4 }
1136 ; Target-dependent attributes:
1137 attributes #1 = { "no-sse" }
1139 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1140 define void @f() #0 #1 { ... }
1147 Function attributes are set to communicate additional information about
1148 a function. Function attributes are considered to be part of the
1149 function, not of the function type, so functions with different function
1150 attributes can have the same function type.
1152 Function attributes are simple keywords that follow the type specified.
1153 If multiple attributes are needed, they are space separated. For
1156 .. code-block:: llvm
1158 define void @f() noinline { ... }
1159 define void @f() alwaysinline { ... }
1160 define void @f() alwaysinline optsize { ... }
1161 define void @f() optsize { ... }
1164 This attribute indicates that, when emitting the prologue and
1165 epilogue, the backend should forcibly align the stack pointer.
1166 Specify the desired alignment, which must be a power of two, in
1169 This attribute indicates that the inliner should attempt to inline
1170 this function into callers whenever possible, ignoring any active
1171 inlining size threshold for this caller.
1173 This indicates that the callee function at a call site should be
1174 recognized as a built-in function, even though the function's declaration
1175 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1176 direct calls to functions that are declared with the ``nobuiltin``
1179 This attribute indicates that this function is rarely called. When
1180 computing edge weights, basic blocks post-dominated by a cold
1181 function call are also considered to be cold; and, thus, given low
1184 This attribute indicates that the source code contained a hint that
1185 inlining this function is desirable (such as the "inline" keyword in
1186 C/C++). It is just a hint; it imposes no requirements on the
1189 This attribute indicates that the function should be added to a
1190 jump-instruction table at code-generation time, and that all address-taken
1191 references to this function should be replaced with a reference to the
1192 appropriate jump-instruction-table function pointer. Note that this creates
1193 a new pointer for the original function, which means that code that depends
1194 on function-pointer identity can break. So, any function annotated with
1195 ``jumptable`` must also be ``unnamed_addr``.
1197 This attribute suggests that optimization passes and code generator
1198 passes make choices that keep the code size of this function as small
1199 as possible and perform optimizations that may sacrifice runtime
1200 performance in order to minimize the size of the generated code.
1202 This attribute disables prologue / epilogue emission for the
1203 function. This can have very system-specific consequences.
1205 This indicates that the callee function at a call site is not recognized as
1206 a built-in function. LLVM will retain the original call and not replace it
1207 with equivalent code based on the semantics of the built-in function, unless
1208 the call site uses the ``builtin`` attribute. This is valid at call sites
1209 and on function declarations and definitions.
1211 This attribute indicates that calls to the function cannot be
1212 duplicated. A call to a ``noduplicate`` function may be moved
1213 within its parent function, but may not be duplicated within
1214 its parent function.
1216 A function containing a ``noduplicate`` call may still
1217 be an inlining candidate, provided that the call is not
1218 duplicated by inlining. That implies that the function has
1219 internal linkage and only has one call site, so the original
1220 call is dead after inlining.
1222 This attributes disables implicit floating point instructions.
1224 This attribute indicates that the inliner should never inline this
1225 function in any situation. This attribute may not be used together
1226 with the ``alwaysinline`` attribute.
1228 This attribute suppresses lazy symbol binding for the function. This
1229 may make calls to the function faster, at the cost of extra program
1230 startup time if the function is not called during program startup.
1232 This attribute indicates that the code generator should not use a
1233 red zone, even if the target-specific ABI normally permits it.
1235 This function attribute indicates that the function never returns
1236 normally. This produces undefined behavior at runtime if the
1237 function ever does dynamically return.
1239 This function attribute indicates that the function never returns
1240 with an unwind or exceptional control flow. If the function does
1241 unwind, its runtime behavior is undefined.
1243 This function attribute indicates that the function is not optimized
1244 by any optimization or code generator passes with the
1245 exception of interprocedural optimization passes.
1246 This attribute cannot be used together with the ``alwaysinline``
1247 attribute; this attribute is also incompatible
1248 with the ``minsize`` attribute and the ``optsize`` attribute.
1250 This attribute requires the ``noinline`` attribute to be specified on
1251 the function as well, so the function is never inlined into any caller.
1252 Only functions with the ``alwaysinline`` attribute are valid
1253 candidates for inlining into the body of this function.
1255 This attribute suggests that optimization passes and code generator
1256 passes make choices that keep the code size of this function low,
1257 and otherwise do optimizations specifically to reduce code size as
1258 long as they do not significantly impact runtime performance.
1260 On a function, this attribute indicates that the function computes its
1261 result (or decides to unwind an exception) based strictly on its arguments,
1262 without dereferencing any pointer arguments or otherwise accessing
1263 any mutable state (e.g. memory, control registers, etc) visible to
1264 caller functions. It does not write through any pointer arguments
1265 (including ``byval`` arguments) and never changes any state visible
1266 to callers. This means that it cannot unwind exceptions by calling
1267 the ``C++`` exception throwing methods.
1269 On an argument, this attribute indicates that the function does not
1270 dereference that pointer argument, even though it may read or write the
1271 memory that the pointer points to if accessed through other pointers.
1273 On a function, this attribute indicates that the function does not write
1274 through any pointer arguments (including ``byval`` arguments) or otherwise
1275 modify any state (e.g. memory, control registers, etc) visible to
1276 caller functions. It may dereference pointer arguments and read
1277 state that may be set in the caller. A readonly function always
1278 returns the same value (or unwinds an exception identically) when
1279 called with the same set of arguments and global state. It cannot
1280 unwind an exception by calling the ``C++`` exception throwing
1283 On an argument, this attribute indicates that the function does not write
1284 through this pointer argument, even though it may write to the memory that
1285 the pointer points to.
1287 This attribute indicates that this function can return twice. The C
1288 ``setjmp`` is an example of such a function. The compiler disables
1289 some optimizations (like tail calls) in the caller of these
1291 ``sanitize_address``
1292 This attribute indicates that AddressSanitizer checks
1293 (dynamic address safety analysis) are enabled for this function.
1295 This attribute indicates that MemorySanitizer checks (dynamic detection
1296 of accesses to uninitialized memory) are enabled for this function.
1298 This attribute indicates that ThreadSanitizer checks
1299 (dynamic thread safety analysis) are enabled for this function.
1301 This attribute indicates that the function should emit a stack
1302 smashing protector. It is in the form of a "canary" --- a random value
1303 placed on the stack before the local variables that's checked upon
1304 return from the function to see if it has been overwritten. A
1305 heuristic is used to determine if a function needs stack protectors
1306 or not. The heuristic used will enable protectors for functions with:
1308 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1309 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1310 - Calls to alloca() with variable sizes or constant sizes greater than
1311 ``ssp-buffer-size``.
1313 Variables that are identified as requiring a protector will be arranged
1314 on the stack such that they are adjacent to the stack protector guard.
1316 If a function that has an ``ssp`` attribute is inlined into a
1317 function that doesn't have an ``ssp`` attribute, then the resulting
1318 function will have an ``ssp`` attribute.
1320 This attribute indicates that the function should *always* emit a
1321 stack smashing protector. This overrides the ``ssp`` function
1324 Variables that are identified as requiring a protector will be arranged
1325 on the stack such that they are adjacent to the stack protector guard.
1326 The specific layout rules are:
1328 #. Large arrays and structures containing large arrays
1329 (``>= ssp-buffer-size``) are closest to the stack protector.
1330 #. Small arrays and structures containing small arrays
1331 (``< ssp-buffer-size``) are 2nd closest to the protector.
1332 #. Variables that have had their address taken are 3rd closest to the
1335 If a function that has an ``sspreq`` attribute is inlined into a
1336 function that doesn't have an ``sspreq`` attribute or which has an
1337 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1338 an ``sspreq`` attribute.
1340 This attribute indicates that the function should emit a stack smashing
1341 protector. This attribute causes a strong heuristic to be used when
1342 determining if a function needs stack protectors. The strong heuristic
1343 will enable protectors for functions with:
1345 - Arrays of any size and type
1346 - Aggregates containing an array of any size and type.
1347 - Calls to alloca().
1348 - Local variables that have had their address taken.
1350 Variables that are identified as requiring a protector will be arranged
1351 on the stack such that they are adjacent to the stack protector guard.
1352 The specific layout rules are:
1354 #. Large arrays and structures containing large arrays
1355 (``>= ssp-buffer-size``) are closest to the stack protector.
1356 #. Small arrays and structures containing small arrays
1357 (``< ssp-buffer-size``) are 2nd closest to the protector.
1358 #. Variables that have had their address taken are 3rd closest to the
1361 This overrides the ``ssp`` function attribute.
1363 If a function that has an ``sspstrong`` attribute is inlined into a
1364 function that doesn't have an ``sspstrong`` attribute, then the
1365 resulting function will have an ``sspstrong`` attribute.
1367 This attribute indicates that the ABI being targeted requires that
1368 an unwind table entry be produce for this function even if we can
1369 show that no exceptions passes by it. This is normally the case for
1370 the ELF x86-64 abi, but it can be disabled for some compilation
1375 Module-Level Inline Assembly
1376 ----------------------------
1378 Modules may contain "module-level inline asm" blocks, which corresponds
1379 to the GCC "file scope inline asm" blocks. These blocks are internally
1380 concatenated by LLVM and treated as a single unit, but may be separated
1381 in the ``.ll`` file if desired. The syntax is very simple:
1383 .. code-block:: llvm
1385 module asm "inline asm code goes here"
1386 module asm "more can go here"
1388 The strings can contain any character by escaping non-printable
1389 characters. The escape sequence used is simply "\\xx" where "xx" is the
1390 two digit hex code for the number.
1392 The inline asm code is simply printed to the machine code .s file when
1393 assembly code is generated.
1395 .. _langref_datalayout:
1400 A module may specify a target specific data layout string that specifies
1401 how data is to be laid out in memory. The syntax for the data layout is
1404 .. code-block:: llvm
1406 target datalayout = "layout specification"
1408 The *layout specification* consists of a list of specifications
1409 separated by the minus sign character ('-'). Each specification starts
1410 with a letter and may include other information after the letter to
1411 define some aspect of the data layout. The specifications accepted are
1415 Specifies that the target lays out data in big-endian form. That is,
1416 the bits with the most significance have the lowest address
1419 Specifies that the target lays out data in little-endian form. That
1420 is, the bits with the least significance have the lowest address
1423 Specifies the natural alignment of the stack in bits. Alignment
1424 promotion of stack variables is limited to the natural stack
1425 alignment to avoid dynamic stack realignment. The stack alignment
1426 must be a multiple of 8-bits. If omitted, the natural stack
1427 alignment defaults to "unspecified", which does not prevent any
1428 alignment promotions.
1429 ``p[n]:<size>:<abi>:<pref>``
1430 This specifies the *size* of a pointer and its ``<abi>`` and
1431 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1432 bits. The address space, ``n`` is optional, and if not specified,
1433 denotes the default address space 0. The value of ``n`` must be
1434 in the range [1,2^23).
1435 ``i<size>:<abi>:<pref>``
1436 This specifies the alignment for an integer type of a given bit
1437 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1438 ``v<size>:<abi>:<pref>``
1439 This specifies the alignment for a vector type of a given bit
1441 ``f<size>:<abi>:<pref>``
1442 This specifies the alignment for a floating point type of a given bit
1443 ``<size>``. Only values of ``<size>`` that are supported by the target
1444 will work. 32 (float) and 64 (double) are supported on all targets; 80
1445 or 128 (different flavors of long double) are also supported on some
1448 This specifies the alignment for an object of aggregate type.
1450 If present, specifies that llvm names are mangled in the output. The
1453 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1454 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1455 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1456 symbols get a ``_`` prefix.
1457 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1458 functions also get a suffix based on the frame size.
1459 ``n<size1>:<size2>:<size3>...``
1460 This specifies a set of native integer widths for the target CPU in
1461 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1462 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1463 this set are considered to support most general arithmetic operations
1466 On every specification that takes a ``<abi>:<pref>``, specifying the
1467 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1468 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1470 When constructing the data layout for a given target, LLVM starts with a
1471 default set of specifications which are then (possibly) overridden by
1472 the specifications in the ``datalayout`` keyword. The default
1473 specifications are given in this list:
1475 - ``E`` - big endian
1476 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1477 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1478 same as the default address space.
1479 - ``S0`` - natural stack alignment is unspecified
1480 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1481 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1482 - ``i16:16:16`` - i16 is 16-bit aligned
1483 - ``i32:32:32`` - i32 is 32-bit aligned
1484 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1485 alignment of 64-bits
1486 - ``f16:16:16`` - half is 16-bit aligned
1487 - ``f32:32:32`` - float is 32-bit aligned
1488 - ``f64:64:64`` - double is 64-bit aligned
1489 - ``f128:128:128`` - quad is 128-bit aligned
1490 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1491 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1492 - ``a:0:64`` - aggregates are 64-bit aligned
1494 When LLVM is determining the alignment for a given type, it uses the
1497 #. If the type sought is an exact match for one of the specifications,
1498 that specification is used.
1499 #. If no match is found, and the type sought is an integer type, then
1500 the smallest integer type that is larger than the bitwidth of the
1501 sought type is used. If none of the specifications are larger than
1502 the bitwidth then the largest integer type is used. For example,
1503 given the default specifications above, the i7 type will use the
1504 alignment of i8 (next largest) while both i65 and i256 will use the
1505 alignment of i64 (largest specified).
1506 #. If no match is found, and the type sought is a vector type, then the
1507 largest vector type that is smaller than the sought vector type will
1508 be used as a fall back. This happens because <128 x double> can be
1509 implemented in terms of 64 <2 x double>, for example.
1511 The function of the data layout string may not be what you expect.
1512 Notably, this is not a specification from the frontend of what alignment
1513 the code generator should use.
1515 Instead, if specified, the target data layout is required to match what
1516 the ultimate *code generator* expects. This string is used by the
1517 mid-level optimizers to improve code, and this only works if it matches
1518 what the ultimate code generator uses. If you would like to generate IR
1519 that does not embed this target-specific detail into the IR, then you
1520 don't have to specify the string. This will disable some optimizations
1521 that require precise layout information, but this also prevents those
1522 optimizations from introducing target specificity into the IR.
1529 A module may specify a target triple string that describes the target
1530 host. The syntax for the target triple is simply:
1532 .. code-block:: llvm
1534 target triple = "x86_64-apple-macosx10.7.0"
1536 The *target triple* string consists of a series of identifiers delimited
1537 by the minus sign character ('-'). The canonical forms are:
1541 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1542 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1544 This information is passed along to the backend so that it generates
1545 code for the proper architecture. It's possible to override this on the
1546 command line with the ``-mtriple`` command line option.
1548 .. _pointeraliasing:
1550 Pointer Aliasing Rules
1551 ----------------------
1553 Any memory access must be done through a pointer value associated with
1554 an address range of the memory access, otherwise the behavior is
1555 undefined. Pointer values are associated with address ranges according
1556 to the following rules:
1558 - A pointer value is associated with the addresses associated with any
1559 value it is *based* on.
1560 - An address of a global variable is associated with the address range
1561 of the variable's storage.
1562 - The result value of an allocation instruction is associated with the
1563 address range of the allocated storage.
1564 - A null pointer in the default address-space is associated with no
1566 - An integer constant other than zero or a pointer value returned from
1567 a function not defined within LLVM may be associated with address
1568 ranges allocated through mechanisms other than those provided by
1569 LLVM. Such ranges shall not overlap with any ranges of addresses
1570 allocated by mechanisms provided by LLVM.
1572 A pointer value is *based* on another pointer value according to the
1575 - A pointer value formed from a ``getelementptr`` operation is *based*
1576 on the first operand of the ``getelementptr``.
1577 - The result value of a ``bitcast`` is *based* on the operand of the
1579 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1580 values that contribute (directly or indirectly) to the computation of
1581 the pointer's value.
1582 - The "*based* on" relationship is transitive.
1584 Note that this definition of *"based"* is intentionally similar to the
1585 definition of *"based"* in C99, though it is slightly weaker.
1587 LLVM IR does not associate types with memory. The result type of a
1588 ``load`` merely indicates the size and alignment of the memory from
1589 which to load, as well as the interpretation of the value. The first
1590 operand type of a ``store`` similarly only indicates the size and
1591 alignment of the store.
1593 Consequently, type-based alias analysis, aka TBAA, aka
1594 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1595 :ref:`Metadata <metadata>` may be used to encode additional information
1596 which specialized optimization passes may use to implement type-based
1601 Volatile Memory Accesses
1602 ------------------------
1604 Certain memory accesses, such as :ref:`load <i_load>`'s,
1605 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1606 marked ``volatile``. The optimizers must not change the number of
1607 volatile operations or change their order of execution relative to other
1608 volatile operations. The optimizers *may* change the order of volatile
1609 operations relative to non-volatile operations. This is not Java's
1610 "volatile" and has no cross-thread synchronization behavior.
1612 IR-level volatile loads and stores cannot safely be optimized into
1613 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1614 flagged volatile. Likewise, the backend should never split or merge
1615 target-legal volatile load/store instructions.
1617 .. admonition:: Rationale
1619 Platforms may rely on volatile loads and stores of natively supported
1620 data width to be executed as single instruction. For example, in C
1621 this holds for an l-value of volatile primitive type with native
1622 hardware support, but not necessarily for aggregate types. The
1623 frontend upholds these expectations, which are intentionally
1624 unspecified in the IR. The rules above ensure that IR transformation
1625 do not violate the frontend's contract with the language.
1629 Memory Model for Concurrent Operations
1630 --------------------------------------
1632 The LLVM IR does not define any way to start parallel threads of
1633 execution or to register signal handlers. Nonetheless, there are
1634 platform-specific ways to create them, and we define LLVM IR's behavior
1635 in their presence. This model is inspired by the C++0x memory model.
1637 For a more informal introduction to this model, see the :doc:`Atomics`.
1639 We define a *happens-before* partial order as the least partial order
1642 - Is a superset of single-thread program order, and
1643 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1644 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1645 techniques, like pthread locks, thread creation, thread joining,
1646 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1647 Constraints <ordering>`).
1649 Note that program order does not introduce *happens-before* edges
1650 between a thread and signals executing inside that thread.
1652 Every (defined) read operation (load instructions, memcpy, atomic
1653 loads/read-modify-writes, etc.) R reads a series of bytes written by
1654 (defined) write operations (store instructions, atomic
1655 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1656 section, initialized globals are considered to have a write of the
1657 initializer which is atomic and happens before any other read or write
1658 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1659 may see any write to the same byte, except:
1661 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1662 write\ :sub:`2` happens before R\ :sub:`byte`, then
1663 R\ :sub:`byte` does not see write\ :sub:`1`.
1664 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1665 R\ :sub:`byte` does not see write\ :sub:`3`.
1667 Given that definition, R\ :sub:`byte` is defined as follows:
1669 - If R is volatile, the result is target-dependent. (Volatile is
1670 supposed to give guarantees which can support ``sig_atomic_t`` in
1671 C/C++, and may be used for accesses to addresses that do not behave
1672 like normal memory. It does not generally provide cross-thread
1674 - Otherwise, if there is no write to the same byte that happens before
1675 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1676 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1677 R\ :sub:`byte` returns the value written by that write.
1678 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1679 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1680 Memory Ordering Constraints <ordering>` section for additional
1681 constraints on how the choice is made.
1682 - Otherwise R\ :sub:`byte` returns ``undef``.
1684 R returns the value composed of the series of bytes it read. This
1685 implies that some bytes within the value may be ``undef`` **without**
1686 the entire value being ``undef``. Note that this only defines the
1687 semantics of the operation; it doesn't mean that targets will emit more
1688 than one instruction to read the series of bytes.
1690 Note that in cases where none of the atomic intrinsics are used, this
1691 model places only one restriction on IR transformations on top of what
1692 is required for single-threaded execution: introducing a store to a byte
1693 which might not otherwise be stored is not allowed in general.
1694 (Specifically, in the case where another thread might write to and read
1695 from an address, introducing a store can change a load that may see
1696 exactly one write into a load that may see multiple writes.)
1700 Atomic Memory Ordering Constraints
1701 ----------------------------------
1703 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1704 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1705 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1706 ordering parameters that determine which other atomic instructions on
1707 the same address they *synchronize with*. These semantics are borrowed
1708 from Java and C++0x, but are somewhat more colloquial. If these
1709 descriptions aren't precise enough, check those specs (see spec
1710 references in the :doc:`atomics guide <Atomics>`).
1711 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1712 differently since they don't take an address. See that instruction's
1713 documentation for details.
1715 For a simpler introduction to the ordering constraints, see the
1719 The set of values that can be read is governed by the happens-before
1720 partial order. A value cannot be read unless some operation wrote
1721 it. This is intended to provide a guarantee strong enough to model
1722 Java's non-volatile shared variables. This ordering cannot be
1723 specified for read-modify-write operations; it is not strong enough
1724 to make them atomic in any interesting way.
1726 In addition to the guarantees of ``unordered``, there is a single
1727 total order for modifications by ``monotonic`` operations on each
1728 address. All modification orders must be compatible with the
1729 happens-before order. There is no guarantee that the modification
1730 orders can be combined to a global total order for the whole program
1731 (and this often will not be possible). The read in an atomic
1732 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1733 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1734 order immediately before the value it writes. If one atomic read
1735 happens before another atomic read of the same address, the later
1736 read must see the same value or a later value in the address's
1737 modification order. This disallows reordering of ``monotonic`` (or
1738 stronger) operations on the same address. If an address is written
1739 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1740 read that address repeatedly, the other threads must eventually see
1741 the write. This corresponds to the C++0x/C1x
1742 ``memory_order_relaxed``.
1744 In addition to the guarantees of ``monotonic``, a
1745 *synchronizes-with* edge may be formed with a ``release`` operation.
1746 This is intended to model C++'s ``memory_order_acquire``.
1748 In addition to the guarantees of ``monotonic``, if this operation
1749 writes a value which is subsequently read by an ``acquire``
1750 operation, it *synchronizes-with* that operation. (This isn't a
1751 complete description; see the C++0x definition of a release
1752 sequence.) This corresponds to the C++0x/C1x
1753 ``memory_order_release``.
1754 ``acq_rel`` (acquire+release)
1755 Acts as both an ``acquire`` and ``release`` operation on its
1756 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1757 ``seq_cst`` (sequentially consistent)
1758 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1759 operation that only reads, ``release`` for an operation that only
1760 writes), there is a global total order on all
1761 sequentially-consistent operations on all addresses, which is
1762 consistent with the *happens-before* partial order and with the
1763 modification orders of all the affected addresses. Each
1764 sequentially-consistent read sees the last preceding write to the
1765 same address in this global order. This corresponds to the C++0x/C1x
1766 ``memory_order_seq_cst`` and Java volatile.
1770 If an atomic operation is marked ``singlethread``, it only *synchronizes
1771 with* or participates in modification and seq\_cst total orderings with
1772 other operations running in the same thread (for example, in signal
1780 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1781 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1782 :ref:`frem <i_frem>`) have the following flags that can set to enable
1783 otherwise unsafe floating point operations
1786 No NaNs - Allow optimizations to assume the arguments and result are not
1787 NaN. Such optimizations are required to retain defined behavior over
1788 NaNs, but the value of the result is undefined.
1791 No Infs - Allow optimizations to assume the arguments and result are not
1792 +/-Inf. Such optimizations are required to retain defined behavior over
1793 +/-Inf, but the value of the result is undefined.
1796 No Signed Zeros - Allow optimizations to treat the sign of a zero
1797 argument or result as insignificant.
1800 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1801 argument rather than perform division.
1804 Fast - Allow algebraically equivalent transformations that may
1805 dramatically change results in floating point (e.g. reassociate). This
1806 flag implies all the others.
1810 Use-list Order Directives
1811 -------------------------
1813 Use-list directives encode the in-memory order of each use-list, allowing the
1814 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1815 indexes that are assigned to the referenced value's uses. The referenced
1816 value's use-list is immediately sorted by these indexes.
1818 Use-list directives may appear at function scope or global scope. They are not
1819 instructions, and have no effect on the semantics of the IR. When they're at
1820 function scope, they must appear after the terminator of the final basic block.
1822 If basic blocks have their address taken via ``blockaddress()`` expressions,
1823 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1830 uselistorder <ty> <value>, { <order-indexes> }
1831 uselistorder_bb @function, %block { <order-indexes> }
1837 define void @foo(i32 %arg1, i32 %arg2) {
1839 ; ... instructions ...
1841 ; ... instructions ...
1843 ; At function scope.
1844 uselistorder i32 %arg1, { 1, 0, 2 }
1845 uselistorder label %bb, { 1, 0 }
1849 uselistorder i32* @global, { 1, 2, 0 }
1850 uselistorder i32 7, { 1, 0 }
1851 uselistorder i32 (i32) @bar, { 1, 0 }
1852 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1859 The LLVM type system is one of the most important features of the
1860 intermediate representation. Being typed enables a number of
1861 optimizations to be performed on the intermediate representation
1862 directly, without having to do extra analyses on the side before the
1863 transformation. A strong type system makes it easier to read the
1864 generated code and enables novel analyses and transformations that are
1865 not feasible to perform on normal three address code representations.
1875 The void type does not represent any value and has no size.
1893 The function type can be thought of as a function signature. It consists of a
1894 return type and a list of formal parameter types. The return type of a function
1895 type is a void type or first class type --- except for :ref:`label <t_label>`
1896 and :ref:`metadata <t_metadata>` types.
1902 <returntype> (<parameter list>)
1904 ...where '``<parameter list>``' is a comma-separated list of type
1905 specifiers. Optionally, the parameter list may include a type ``...``, which
1906 indicates that the function takes a variable number of arguments. Variable
1907 argument functions can access their arguments with the :ref:`variable argument
1908 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1909 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1913 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1914 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1915 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1916 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1917 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1918 | ``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. |
1919 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1920 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1921 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1928 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1929 Values of these types are the only ones which can be produced by
1937 These are the types that are valid in registers from CodeGen's perspective.
1946 The integer type is a very simple type that simply specifies an
1947 arbitrary bit width for the integer type desired. Any bit width from 1
1948 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1956 The number of bits the integer will occupy is specified by the ``N``
1962 +----------------+------------------------------------------------+
1963 | ``i1`` | a single-bit integer. |
1964 +----------------+------------------------------------------------+
1965 | ``i32`` | a 32-bit integer. |
1966 +----------------+------------------------------------------------+
1967 | ``i1942652`` | a really big integer of over 1 million bits. |
1968 +----------------+------------------------------------------------+
1972 Floating Point Types
1973 """"""""""""""""""""
1982 - 16-bit floating point value
1985 - 32-bit floating point value
1988 - 64-bit floating point value
1991 - 128-bit floating point value (112-bit mantissa)
1994 - 80-bit floating point value (X87)
1997 - 128-bit floating point value (two 64-bits)
2004 The x86_mmx type represents a value held in an MMX register on an x86
2005 machine. The operations allowed on it are quite limited: parameters and
2006 return values, load and store, and bitcast. User-specified MMX
2007 instructions are represented as intrinsic or asm calls with arguments
2008 and/or results of this type. There are no arrays, vectors or constants
2025 The pointer type is used to specify memory locations. Pointers are
2026 commonly used to reference objects in memory.
2028 Pointer types may have an optional address space attribute defining the
2029 numbered address space where the pointed-to object resides. The default
2030 address space is number zero. The semantics of non-zero address spaces
2031 are target-specific.
2033 Note that LLVM does not permit pointers to void (``void*``) nor does it
2034 permit pointers to labels (``label*``). Use ``i8*`` instead.
2044 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2045 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2046 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2047 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2048 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2049 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2050 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2059 A vector type is a simple derived type that represents a vector of
2060 elements. Vector types are used when multiple primitive data are
2061 operated in parallel using a single instruction (SIMD). A vector type
2062 requires a size (number of elements) and an underlying primitive data
2063 type. Vector types are considered :ref:`first class <t_firstclass>`.
2069 < <# elements> x <elementtype> >
2071 The number of elements is a constant integer value larger than 0;
2072 elementtype may be any integer, floating point or pointer type. Vectors
2073 of size zero are not allowed.
2077 +-------------------+--------------------------------------------------+
2078 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2079 +-------------------+--------------------------------------------------+
2080 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2081 +-------------------+--------------------------------------------------+
2082 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2083 +-------------------+--------------------------------------------------+
2084 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2085 +-------------------+--------------------------------------------------+
2094 The label type represents code labels.
2109 The metadata type represents embedded metadata. No derived types may be
2110 created from metadata except for :ref:`function <t_function>` arguments.
2123 Aggregate Types are a subset of derived types that can contain multiple
2124 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2125 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2135 The array type is a very simple derived type that arranges elements
2136 sequentially in memory. The array type requires a size (number of
2137 elements) and an underlying data type.
2143 [<# elements> x <elementtype>]
2145 The number of elements is a constant integer value; ``elementtype`` may
2146 be any type with a size.
2150 +------------------+--------------------------------------+
2151 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2152 +------------------+--------------------------------------+
2153 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2154 +------------------+--------------------------------------+
2155 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2156 +------------------+--------------------------------------+
2158 Here are some examples of multidimensional arrays:
2160 +-----------------------------+----------------------------------------------------------+
2161 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2162 +-----------------------------+----------------------------------------------------------+
2163 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2164 +-----------------------------+----------------------------------------------------------+
2165 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2166 +-----------------------------+----------------------------------------------------------+
2168 There is no restriction on indexing beyond the end of the array implied
2169 by a static type (though there are restrictions on indexing beyond the
2170 bounds of an allocated object in some cases). This means that
2171 single-dimension 'variable sized array' addressing can be implemented in
2172 LLVM with a zero length array type. An implementation of 'pascal style
2173 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2183 The structure type is used to represent a collection of data members
2184 together in memory. The elements of a structure may be any type that has
2187 Structures in memory are accessed using '``load``' and '``store``' by
2188 getting a pointer to a field with the '``getelementptr``' instruction.
2189 Structures in registers are accessed using the '``extractvalue``' and
2190 '``insertvalue``' instructions.
2192 Structures may optionally be "packed" structures, which indicate that
2193 the alignment of the struct is one byte, and that there is no padding
2194 between the elements. In non-packed structs, padding between field types
2195 is inserted as defined by the DataLayout string in the module, which is
2196 required to match what the underlying code generator expects.
2198 Structures can either be "literal" or "identified". A literal structure
2199 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2200 identified types are always defined at the top level with a name.
2201 Literal types are uniqued by their contents and can never be recursive
2202 or opaque since there is no way to write one. Identified types can be
2203 recursive, can be opaqued, and are never uniqued.
2209 %T1 = type { <type list> } ; Identified normal struct type
2210 %T2 = type <{ <type list> }> ; Identified packed struct type
2214 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2215 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2216 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2217 | ``{ 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``. |
2218 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2219 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2220 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2224 Opaque Structure Types
2225 """"""""""""""""""""""
2229 Opaque structure types are used to represent named structure types that
2230 do not have a body specified. This corresponds (for example) to the C
2231 notion of a forward declared structure.
2242 +--------------+-------------------+
2243 | ``opaque`` | An opaque type. |
2244 +--------------+-------------------+
2251 LLVM has several different basic types of constants. This section
2252 describes them all and their syntax.
2257 **Boolean constants**
2258 The two strings '``true``' and '``false``' are both valid constants
2260 **Integer constants**
2261 Standard integers (such as '4') are constants of the
2262 :ref:`integer <t_integer>` type. Negative numbers may be used with
2264 **Floating point constants**
2265 Floating point constants use standard decimal notation (e.g.
2266 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2267 hexadecimal notation (see below). The assembler requires the exact
2268 decimal value of a floating-point constant. For example, the
2269 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2270 decimal in binary. Floating point constants must have a :ref:`floating
2271 point <t_floating>` type.
2272 **Null pointer constants**
2273 The identifier '``null``' is recognized as a null pointer constant
2274 and must be of :ref:`pointer type <t_pointer>`.
2276 The one non-intuitive notation for constants is the hexadecimal form of
2277 floating point constants. For example, the form
2278 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2279 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2280 constants are required (and the only time that they are generated by the
2281 disassembler) is when a floating point constant must be emitted but it
2282 cannot be represented as a decimal floating point number in a reasonable
2283 number of digits. For example, NaN's, infinities, and other special
2284 values are represented in their IEEE hexadecimal format so that assembly
2285 and disassembly do not cause any bits to change in the constants.
2287 When using the hexadecimal form, constants of types half, float, and
2288 double are represented using the 16-digit form shown above (which
2289 matches the IEEE754 representation for double); half and float values
2290 must, however, be exactly representable as IEEE 754 half and single
2291 precision, respectively. Hexadecimal format is always used for long
2292 double, and there are three forms of long double. The 80-bit format used
2293 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2294 128-bit format used by PowerPC (two adjacent doubles) is represented by
2295 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2296 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2297 will only work if they match the long double format on your target.
2298 The IEEE 16-bit format (half precision) is represented by ``0xH``
2299 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2300 (sign bit at the left).
2302 There are no constants of type x86_mmx.
2304 .. _complexconstants:
2309 Complex constants are a (potentially recursive) combination of simple
2310 constants and smaller complex constants.
2312 **Structure constants**
2313 Structure constants are represented with notation similar to
2314 structure type definitions (a comma separated list of elements,
2315 surrounded by braces (``{}``)). For example:
2316 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2317 "``@G = external global i32``". Structure constants must have
2318 :ref:`structure type <t_struct>`, and the number and types of elements
2319 must match those specified by the type.
2321 Array constants are represented with notation similar to array type
2322 definitions (a comma separated list of elements, surrounded by
2323 square brackets (``[]``)). For example:
2324 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2325 :ref:`array type <t_array>`, and the number and types of elements must
2326 match those specified by the type. As a special case, character array
2327 constants may also be represented as a double-quoted string using the ``c``
2328 prefix. For example: "``c"Hello World\0A\00"``".
2329 **Vector constants**
2330 Vector constants are represented with notation similar to vector
2331 type definitions (a comma separated list of elements, surrounded by
2332 less-than/greater-than's (``<>``)). For example:
2333 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2334 must have :ref:`vector type <t_vector>`, and the number and types of
2335 elements must match those specified by the type.
2336 **Zero initialization**
2337 The string '``zeroinitializer``' can be used to zero initialize a
2338 value to zero of *any* type, including scalar and
2339 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2340 having to print large zero initializers (e.g. for large arrays) and
2341 is always exactly equivalent to using explicit zero initializers.
2343 A metadata node is a constant tuple without types. For example:
2344 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2345 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2346 Unlike other typed constants that are meant to be interpreted as part of
2347 the instruction stream, metadata is a place to attach additional
2348 information such as debug info.
2350 Global Variable and Function Addresses
2351 --------------------------------------
2353 The addresses of :ref:`global variables <globalvars>` and
2354 :ref:`functions <functionstructure>` are always implicitly valid
2355 (link-time) constants. These constants are explicitly referenced when
2356 the :ref:`identifier for the global <identifiers>` is used and always have
2357 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2360 .. code-block:: llvm
2364 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2371 The string '``undef``' can be used anywhere a constant is expected, and
2372 indicates that the user of the value may receive an unspecified
2373 bit-pattern. Undefined values may be of any type (other than '``label``'
2374 or '``void``') and be used anywhere a constant is permitted.
2376 Undefined values are useful because they indicate to the compiler that
2377 the program is well defined no matter what value is used. This gives the
2378 compiler more freedom to optimize. Here are some examples of
2379 (potentially surprising) transformations that are valid (in pseudo IR):
2381 .. code-block:: llvm
2391 This is safe because all of the output bits are affected by the undef
2392 bits. Any output bit can have a zero or one depending on the input bits.
2394 .. code-block:: llvm
2405 These logical operations have bits that are not always affected by the
2406 input. For example, if ``%X`` has a zero bit, then the output of the
2407 '``and``' operation will always be a zero for that bit, no matter what
2408 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2409 optimize or assume that the result of the '``and``' is '``undef``'.
2410 However, it is safe to assume that all bits of the '``undef``' could be
2411 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2412 all the bits of the '``undef``' operand to the '``or``' could be set,
2413 allowing the '``or``' to be folded to -1.
2415 .. code-block:: llvm
2417 %A = select undef, %X, %Y
2418 %B = select undef, 42, %Y
2419 %C = select %X, %Y, undef
2429 This set of examples shows that undefined '``select``' (and conditional
2430 branch) conditions can go *either way*, but they have to come from one
2431 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2432 both known to have a clear low bit, then ``%A`` would have to have a
2433 cleared low bit. However, in the ``%C`` example, the optimizer is
2434 allowed to assume that the '``undef``' operand could be the same as
2435 ``%Y``, allowing the whole '``select``' to be eliminated.
2437 .. code-block:: llvm
2439 %A = xor undef, undef
2456 This example points out that two '``undef``' operands are not
2457 necessarily the same. This can be surprising to people (and also matches
2458 C semantics) where they assume that "``X^X``" is always zero, even if
2459 ``X`` is undefined. This isn't true for a number of reasons, but the
2460 short answer is that an '``undef``' "variable" can arbitrarily change
2461 its value over its "live range". This is true because the variable
2462 doesn't actually *have a live range*. Instead, the value is logically
2463 read from arbitrary registers that happen to be around when needed, so
2464 the value is not necessarily consistent over time. In fact, ``%A`` and
2465 ``%C`` need to have the same semantics or the core LLVM "replace all
2466 uses with" concept would not hold.
2468 .. code-block:: llvm
2476 These examples show the crucial difference between an *undefined value*
2477 and *undefined behavior*. An undefined value (like '``undef``') is
2478 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2479 operation can be constant folded to '``undef``', because the '``undef``'
2480 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2481 However, in the second example, we can make a more aggressive
2482 assumption: because the ``undef`` is allowed to be an arbitrary value,
2483 we are allowed to assume that it could be zero. Since a divide by zero
2484 has *undefined behavior*, we are allowed to assume that the operation
2485 does not execute at all. This allows us to delete the divide and all
2486 code after it. Because the undefined operation "can't happen", the
2487 optimizer can assume that it occurs in dead code.
2489 .. code-block:: llvm
2491 a: store undef -> %X
2492 b: store %X -> undef
2497 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2498 value can be assumed to not have any effect; we can assume that the
2499 value is overwritten with bits that happen to match what was already
2500 there. However, a store *to* an undefined location could clobber
2501 arbitrary memory, therefore, it has undefined behavior.
2508 Poison values are similar to :ref:`undef values <undefvalues>`, however
2509 they also represent the fact that an instruction or constant expression
2510 that cannot evoke side effects has nevertheless detected a condition
2511 that results in undefined behavior.
2513 There is currently no way of representing a poison value in the IR; they
2514 only exist when produced by operations such as :ref:`add <i_add>` with
2517 Poison value behavior is defined in terms of value *dependence*:
2519 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2520 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2521 their dynamic predecessor basic block.
2522 - Function arguments depend on the corresponding actual argument values
2523 in the dynamic callers of their functions.
2524 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2525 instructions that dynamically transfer control back to them.
2526 - :ref:`Invoke <i_invoke>` instructions depend on the
2527 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2528 call instructions that dynamically transfer control back to them.
2529 - Non-volatile loads and stores depend on the most recent stores to all
2530 of the referenced memory addresses, following the order in the IR
2531 (including loads and stores implied by intrinsics such as
2532 :ref:`@llvm.memcpy <int_memcpy>`.)
2533 - An instruction with externally visible side effects depends on the
2534 most recent preceding instruction with externally visible side
2535 effects, following the order in the IR. (This includes :ref:`volatile
2536 operations <volatile>`.)
2537 - An instruction *control-depends* on a :ref:`terminator
2538 instruction <terminators>` if the terminator instruction has
2539 multiple successors and the instruction is always executed when
2540 control transfers to one of the successors, and may not be executed
2541 when control is transferred to another.
2542 - Additionally, an instruction also *control-depends* on a terminator
2543 instruction if the set of instructions it otherwise depends on would
2544 be different if the terminator had transferred control to a different
2546 - Dependence is transitive.
2548 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2549 with the additional effect that any instruction that has a *dependence*
2550 on a poison value has undefined behavior.
2552 Here are some examples:
2554 .. code-block:: llvm
2557 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2558 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2559 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2560 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2562 store i32 %poison, i32* @g ; Poison value stored to memory.
2563 %poison2 = load i32* @g ; Poison value loaded back from memory.
2565 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2567 %narrowaddr = bitcast i32* @g to i16*
2568 %wideaddr = bitcast i32* @g to i64*
2569 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2570 %poison4 = load i64* %wideaddr ; Returns a poison value.
2572 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2573 br i1 %cmp, label %true, label %end ; Branch to either destination.
2576 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2577 ; it has undefined behavior.
2581 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2582 ; Both edges into this PHI are
2583 ; control-dependent on %cmp, so this
2584 ; always results in a poison value.
2586 store volatile i32 0, i32* @g ; This would depend on the store in %true
2587 ; if %cmp is true, or the store in %entry
2588 ; otherwise, so this is undefined behavior.
2590 br i1 %cmp, label %second_true, label %second_end
2591 ; The same branch again, but this time the
2592 ; true block doesn't have side effects.
2599 store volatile i32 0, i32* @g ; This time, the instruction always depends
2600 ; on the store in %end. Also, it is
2601 ; control-equivalent to %end, so this is
2602 ; well-defined (ignoring earlier undefined
2603 ; behavior in this example).
2607 Addresses of Basic Blocks
2608 -------------------------
2610 ``blockaddress(@function, %block)``
2612 The '``blockaddress``' constant computes the address of the specified
2613 basic block in the specified function, and always has an ``i8*`` type.
2614 Taking the address of the entry block is illegal.
2616 This value only has defined behavior when used as an operand to the
2617 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2618 against null. Pointer equality tests between labels addresses results in
2619 undefined behavior --- though, again, comparison against null is ok, and
2620 no label is equal to the null pointer. This may be passed around as an
2621 opaque pointer sized value as long as the bits are not inspected. This
2622 allows ``ptrtoint`` and arithmetic to be performed on these values so
2623 long as the original value is reconstituted before the ``indirectbr``
2626 Finally, some targets may provide defined semantics when using the value
2627 as the operand to an inline assembly, but that is target specific.
2631 Constant Expressions
2632 --------------------
2634 Constant expressions are used to allow expressions involving other
2635 constants to be used as constants. Constant expressions may be of any
2636 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2637 that does not have side effects (e.g. load and call are not supported).
2638 The following is the syntax for constant expressions:
2640 ``trunc (CST to TYPE)``
2641 Truncate a constant to another type. The bit size of CST must be
2642 larger than the bit size of TYPE. Both types must be integers.
2643 ``zext (CST to TYPE)``
2644 Zero extend a constant to another type. The bit size of CST must be
2645 smaller than the bit size of TYPE. Both types must be integers.
2646 ``sext (CST to TYPE)``
2647 Sign extend a constant to another type. The bit size of CST must be
2648 smaller than the bit size of TYPE. Both types must be integers.
2649 ``fptrunc (CST to TYPE)``
2650 Truncate a floating point constant to another floating point type.
2651 The size of CST must be larger than the size of TYPE. Both types
2652 must be floating point.
2653 ``fpext (CST to TYPE)``
2654 Floating point extend a constant to another type. The size of CST
2655 must be smaller or equal to the size of TYPE. Both types must be
2657 ``fptoui (CST to TYPE)``
2658 Convert a floating point constant to the corresponding unsigned
2659 integer constant. TYPE must be a scalar or vector integer type. CST
2660 must be of scalar or vector floating point type. Both CST and TYPE
2661 must be scalars, or vectors of the same number of elements. If the
2662 value won't fit in the integer type, the results are undefined.
2663 ``fptosi (CST to TYPE)``
2664 Convert a floating point constant to the corresponding signed
2665 integer constant. TYPE must be a scalar or vector integer type. CST
2666 must be of scalar or vector floating point type. Both CST and TYPE
2667 must be scalars, or vectors of the same number of elements. If the
2668 value won't fit in the integer type, the results are undefined.
2669 ``uitofp (CST to TYPE)``
2670 Convert an unsigned integer constant to the corresponding floating
2671 point constant. TYPE must be a scalar or vector floating point type.
2672 CST must be of scalar or vector integer type. Both CST and TYPE must
2673 be scalars, or vectors of the same number of elements. If the value
2674 won't fit in the floating point type, the results are undefined.
2675 ``sitofp (CST to TYPE)``
2676 Convert a signed integer constant to the corresponding floating
2677 point constant. TYPE must be a scalar or vector floating point type.
2678 CST must be of scalar or vector integer type. Both CST and TYPE must
2679 be scalars, or vectors of the same number of elements. If the value
2680 won't fit in the floating point type, the results are undefined.
2681 ``ptrtoint (CST to TYPE)``
2682 Convert a pointer typed constant to the corresponding integer
2683 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2684 pointer type. The ``CST`` value is zero extended, truncated, or
2685 unchanged to make it fit in ``TYPE``.
2686 ``inttoptr (CST to TYPE)``
2687 Convert an integer constant to a pointer constant. TYPE must be a
2688 pointer type. CST must be of integer type. The CST value is zero
2689 extended, truncated, or unchanged to make it fit in a pointer size.
2690 This one is *really* dangerous!
2691 ``bitcast (CST to TYPE)``
2692 Convert a constant, CST, to another TYPE. The constraints of the
2693 operands are the same as those for the :ref:`bitcast
2694 instruction <i_bitcast>`.
2695 ``addrspacecast (CST to TYPE)``
2696 Convert a constant pointer or constant vector of pointer, CST, to another
2697 TYPE in a different address space. The constraints of the operands are the
2698 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2699 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2700 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2701 constants. As with the :ref:`getelementptr <i_getelementptr>`
2702 instruction, the index list may have zero or more indexes, which are
2703 required to make sense for the type of "CSTPTR".
2704 ``select (COND, VAL1, VAL2)``
2705 Perform the :ref:`select operation <i_select>` on constants.
2706 ``icmp COND (VAL1, VAL2)``
2707 Performs the :ref:`icmp operation <i_icmp>` on constants.
2708 ``fcmp COND (VAL1, VAL2)``
2709 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2710 ``extractelement (VAL, IDX)``
2711 Perform the :ref:`extractelement operation <i_extractelement>` on
2713 ``insertelement (VAL, ELT, IDX)``
2714 Perform the :ref:`insertelement operation <i_insertelement>` on
2716 ``shufflevector (VEC1, VEC2, IDXMASK)``
2717 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2719 ``extractvalue (VAL, IDX0, IDX1, ...)``
2720 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2721 constants. The index list is interpreted in a similar manner as
2722 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2723 least one index value must be specified.
2724 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2725 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2726 The index list is interpreted in a similar manner as indices in a
2727 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2728 value must be specified.
2729 ``OPCODE (LHS, RHS)``
2730 Perform the specified operation of the LHS and RHS constants. OPCODE
2731 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2732 binary <bitwiseops>` operations. The constraints on operands are
2733 the same as those for the corresponding instruction (e.g. no bitwise
2734 operations on floating point values are allowed).
2741 Inline Assembler Expressions
2742 ----------------------------
2744 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2745 Inline Assembly <moduleasm>`) through the use of a special value. This
2746 value represents the inline assembler as a string (containing the
2747 instructions to emit), a list of operand constraints (stored as a
2748 string), a flag that indicates whether or not the inline asm expression
2749 has side effects, and a flag indicating whether the function containing
2750 the asm needs to align its stack conservatively. An example inline
2751 assembler expression is:
2753 .. code-block:: llvm
2755 i32 (i32) asm "bswap $0", "=r,r"
2757 Inline assembler expressions may **only** be used as the callee operand
2758 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2759 Thus, typically we have:
2761 .. code-block:: llvm
2763 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2765 Inline asms with side effects not visible in the constraint list must be
2766 marked as having side effects. This is done through the use of the
2767 '``sideeffect``' keyword, like so:
2769 .. code-block:: llvm
2771 call void asm sideeffect "eieio", ""()
2773 In some cases inline asms will contain code that will not work unless
2774 the stack is aligned in some way, such as calls or SSE instructions on
2775 x86, yet will not contain code that does that alignment within the asm.
2776 The compiler should make conservative assumptions about what the asm
2777 might contain and should generate its usual stack alignment code in the
2778 prologue if the '``alignstack``' keyword is present:
2780 .. code-block:: llvm
2782 call void asm alignstack "eieio", ""()
2784 Inline asms also support using non-standard assembly dialects. The
2785 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2786 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2787 the only supported dialects. An example is:
2789 .. code-block:: llvm
2791 call void asm inteldialect "eieio", ""()
2793 If multiple keywords appear the '``sideeffect``' keyword must come
2794 first, the '``alignstack``' keyword second and the '``inteldialect``'
2800 The call instructions that wrap inline asm nodes may have a
2801 "``!srcloc``" MDNode attached to it that contains a list of constant
2802 integers. If present, the code generator will use the integer as the
2803 location cookie value when report errors through the ``LLVMContext``
2804 error reporting mechanisms. This allows a front-end to correlate backend
2805 errors that occur with inline asm back to the source code that produced
2808 .. code-block:: llvm
2810 call void asm sideeffect "something bad", ""(), !srcloc !42
2812 !42 = !{ i32 1234567 }
2814 It is up to the front-end to make sense of the magic numbers it places
2815 in the IR. If the MDNode contains multiple constants, the code generator
2816 will use the one that corresponds to the line of the asm that the error
2824 LLVM IR allows metadata to be attached to instructions in the program
2825 that can convey extra information about the code to the optimizers and
2826 code generator. One example application of metadata is source-level
2827 debug information. There are two metadata primitives: strings and nodes.
2829 Metadata does not have a type, and is not a value. If referenced from a
2830 ``call`` instruction, it uses the ``metadata`` type.
2832 All metadata are identified in syntax by a exclamation point ('``!``').
2834 Metadata Nodes and Metadata Strings
2835 -----------------------------------
2837 A metadata string is a string surrounded by double quotes. It can
2838 contain any character by escaping non-printable characters with
2839 "``\xx``" where "``xx``" is the two digit hex code. For example:
2842 Metadata nodes are represented with notation similar to structure
2843 constants (a comma separated list of elements, surrounded by braces and
2844 preceded by an exclamation point). Metadata nodes can have any values as
2845 their operand. For example:
2847 .. code-block:: llvm
2849 !{ !"test\00", i32 10}
2851 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2852 metadata nodes, which can be looked up in the module symbol table. For
2855 .. code-block:: llvm
2859 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2860 function is using two metadata arguments:
2862 .. code-block:: llvm
2864 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2866 Metadata can be attached with an instruction. Here metadata ``!21`` is
2867 attached to the ``add`` instruction using the ``!dbg`` identifier:
2869 .. code-block:: llvm
2871 %indvar.next = add i64 %indvar, 1, !dbg !21
2873 More information about specific metadata nodes recognized by the
2874 optimizers and code generator is found below.
2879 In LLVM IR, memory does not have types, so LLVM's own type system is not
2880 suitable for doing TBAA. Instead, metadata is added to the IR to
2881 describe a type system of a higher level language. This can be used to
2882 implement typical C/C++ TBAA, but it can also be used to implement
2883 custom alias analysis behavior for other languages.
2885 The current metadata format is very simple. TBAA metadata nodes have up
2886 to three fields, e.g.:
2888 .. code-block:: llvm
2890 !0 = !{ !"an example type tree" }
2891 !1 = !{ !"int", !0 }
2892 !2 = !{ !"float", !0 }
2893 !3 = !{ !"const float", !2, i64 1 }
2895 The first field is an identity field. It can be any value, usually a
2896 metadata string, which uniquely identifies the type. The most important
2897 name in the tree is the name of the root node. Two trees with different
2898 root node names are entirely disjoint, even if they have leaves with
2901 The second field identifies the type's parent node in the tree, or is
2902 null or omitted for a root node. A type is considered to alias all of
2903 its descendants and all of its ancestors in the tree. Also, a type is
2904 considered to alias all types in other trees, so that bitcode produced
2905 from multiple front-ends is handled conservatively.
2907 If the third field is present, it's an integer which if equal to 1
2908 indicates that the type is "constant" (meaning
2909 ``pointsToConstantMemory`` should return true; see `other useful
2910 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2912 '``tbaa.struct``' Metadata
2913 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2915 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2916 aggregate assignment operations in C and similar languages, however it
2917 is defined to copy a contiguous region of memory, which is more than
2918 strictly necessary for aggregate types which contain holes due to
2919 padding. Also, it doesn't contain any TBAA information about the fields
2922 ``!tbaa.struct`` metadata can describe which memory subregions in a
2923 memcpy are padding and what the TBAA tags of the struct are.
2925 The current metadata format is very simple. ``!tbaa.struct`` metadata
2926 nodes are a list of operands which are in conceptual groups of three.
2927 For each group of three, the first operand gives the byte offset of a
2928 field in bytes, the second gives its size in bytes, and the third gives
2931 .. code-block:: llvm
2933 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
2935 This describes a struct with two fields. The first is at offset 0 bytes
2936 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2937 and has size 4 bytes and has tbaa tag !2.
2939 Note that the fields need not be contiguous. In this example, there is a
2940 4 byte gap between the two fields. This gap represents padding which
2941 does not carry useful data and need not be preserved.
2943 '``noalias``' and '``alias.scope``' Metadata
2944 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2946 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
2947 noalias memory-access sets. This means that some collection of memory access
2948 instructions (loads, stores, memory-accessing calls, etc.) that carry
2949 ``noalias`` metadata can specifically be specified not to alias with some other
2950 collection of memory access instructions that carry ``alias.scope`` metadata.
2951 Each type of metadata specifies a list of scopes where each scope has an id and
2952 a domain. When evaluating an aliasing query, if for some some domain, the set
2953 of scopes with that domain in one instruction's ``alias.scope`` list is a
2954 subset of (or qual to) the set of scopes for that domain in another
2955 instruction's ``noalias`` list, then the two memory accesses are assumed not to
2958 The metadata identifying each domain is itself a list containing one or two
2959 entries. The first entry is the name of the domain. Note that if the name is a
2960 string then it can be combined accross functions and translation units. A
2961 self-reference can be used to create globally unique domain names. A
2962 descriptive string may optionally be provided as a second list entry.
2964 The metadata identifying each scope is also itself a list containing two or
2965 three entries. The first entry is the name of the scope. Note that if the name
2966 is a string then it can be combined accross functions and translation units. A
2967 self-reference can be used to create globally unique scope names. A metadata
2968 reference to the scope's domain is the second entry. A descriptive string may
2969 optionally be provided as a third list entry.
2973 .. code-block:: llvm
2975 ; Two scope domains:
2979 ; Some scopes in these domains:
2985 !5 = !{!4} ; A list containing only scope !4
2989 ; These two instructions don't alias:
2990 %0 = load float* %c, align 4, !alias.scope !5
2991 store float %0, float* %arrayidx.i, align 4, !noalias !5
2993 ; These two instructions also don't alias (for domain !1, the set of scopes
2994 ; in the !alias.scope equals that in the !noalias list):
2995 %2 = load float* %c, align 4, !alias.scope !5
2996 store float %2, float* %arrayidx.i2, align 4, !noalias !6
2998 ; These two instructions don't alias (for domain !0, the set of scopes in
2999 ; the !noalias list is not a superset of, or equal to, the scopes in the
3000 ; !alias.scope list):
3001 %2 = load float* %c, align 4, !alias.scope !6
3002 store float %0, float* %arrayidx.i, align 4, !noalias !7
3004 '``fpmath``' Metadata
3005 ^^^^^^^^^^^^^^^^^^^^^
3007 ``fpmath`` metadata may be attached to any instruction of floating point
3008 type. It can be used to express the maximum acceptable error in the
3009 result of that instruction, in ULPs, thus potentially allowing the
3010 compiler to use a more efficient but less accurate method of computing
3011 it. ULP is defined as follows:
3013 If ``x`` is a real number that lies between two finite consecutive
3014 floating-point numbers ``a`` and ``b``, without being equal to one
3015 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
3016 distance between the two non-equal finite floating-point numbers
3017 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
3019 The metadata node shall consist of a single positive floating point
3020 number representing the maximum relative error, for example:
3022 .. code-block:: llvm
3024 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3026 '``range``' Metadata
3027 ^^^^^^^^^^^^^^^^^^^^
3029 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
3030 integer types. It expresses the possible ranges the loaded value or the value
3031 returned by the called function at this call site is in. The ranges are
3032 represented with a flattened list of integers. The loaded value or the value
3033 returned is known to be in the union of the ranges defined by each consecutive
3034 pair. Each pair has the following properties:
3036 - The type must match the type loaded by the instruction.
3037 - The pair ``a,b`` represents the range ``[a,b)``.
3038 - Both ``a`` and ``b`` are constants.
3039 - The range is allowed to wrap.
3040 - The range should not represent the full or empty set. That is,
3043 In addition, the pairs must be in signed order of the lower bound and
3044 they must be non-contiguous.
3048 .. code-block:: llvm
3050 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3051 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3052 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
3053 %d = invoke i8 @bar() to label %cont
3054 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
3056 !0 = !{ i8 0, i8 2 }
3057 !1 = !{ i8 255, i8 2 }
3058 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
3059 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
3064 It is sometimes useful to attach information to loop constructs. Currently,
3065 loop metadata is implemented as metadata attached to the branch instruction
3066 in the loop latch block. This type of metadata refer to a metadata node that is
3067 guaranteed to be separate for each loop. The loop identifier metadata is
3068 specified with the name ``llvm.loop``.
3070 The loop identifier metadata is implemented using a metadata that refers to
3071 itself to avoid merging it with any other identifier metadata, e.g.,
3072 during module linkage or function inlining. That is, each loop should refer
3073 to their own identification metadata even if they reside in separate functions.
3074 The following example contains loop identifier metadata for two separate loop
3077 .. code-block:: llvm
3082 The loop identifier metadata can be used to specify additional
3083 per-loop metadata. Any operands after the first operand can be treated
3084 as user-defined metadata. For example the ``llvm.loop.unroll.count``
3085 suggests an unroll factor to the loop unroller:
3087 .. code-block:: llvm
3089 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
3092 !1 = !{!"llvm.loop.unroll.count", i32 4}
3094 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
3095 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3097 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
3098 used to control per-loop vectorization and interleaving parameters such as
3099 vectorization width and interleave count. These metadata should be used in
3100 conjunction with ``llvm.loop`` loop identification metadata. The
3101 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
3102 optimization hints and the optimizer will only interleave and vectorize loops if
3103 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
3104 which contains information about loop-carried memory dependencies can be helpful
3105 in determining the safety of these transformations.
3107 '``llvm.loop.interleave.count``' Metadata
3108 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3110 This metadata suggests an interleave count to the loop interleaver.
3111 The first operand is the string ``llvm.loop.interleave.count`` and the
3112 second operand is an integer specifying the interleave count. For
3115 .. code-block:: llvm
3117 !0 = !{!"llvm.loop.interleave.count", i32 4}
3119 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
3120 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
3121 then the interleave count will be determined automatically.
3123 '``llvm.loop.vectorize.enable``' Metadata
3124 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3126 This metadata selectively enables or disables vectorization for the loop. The
3127 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
3128 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
3129 0 disables vectorization:
3131 .. code-block:: llvm
3133 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
3134 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
3136 '``llvm.loop.vectorize.width``' Metadata
3137 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3139 This metadata sets the target width of the vectorizer. The first
3140 operand is the string ``llvm.loop.vectorize.width`` and the second
3141 operand is an integer specifying the width. For example:
3143 .. code-block:: llvm
3145 !0 = !{!"llvm.loop.vectorize.width", i32 4}
3147 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
3148 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
3149 0 or if the loop does not have this metadata the width will be
3150 determined automatically.
3152 '``llvm.loop.unroll``'
3153 ^^^^^^^^^^^^^^^^^^^^^^
3155 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
3156 optimization hints such as the unroll factor. ``llvm.loop.unroll``
3157 metadata should be used in conjunction with ``llvm.loop`` loop
3158 identification metadata. The ``llvm.loop.unroll`` metadata are only
3159 optimization hints and the unrolling will only be performed if the
3160 optimizer believes it is safe to do so.
3162 '``llvm.loop.unroll.count``' Metadata
3163 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3165 This metadata suggests an unroll factor to the loop unroller. The
3166 first operand is the string ``llvm.loop.unroll.count`` and the second
3167 operand is a positive integer specifying the unroll factor. For
3170 .. code-block:: llvm
3172 !0 = !{!"llvm.loop.unroll.count", i32 4}
3174 If the trip count of the loop is less than the unroll count the loop
3175 will be partially unrolled.
3177 '``llvm.loop.unroll.disable``' Metadata
3178 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3180 This metadata either disables loop unrolling. The metadata has a single operand
3181 which is the string ``llvm.loop.unroll.disable``. For example:
3183 .. code-block:: llvm
3185 !0 = !{!"llvm.loop.unroll.disable"}
3187 '``llvm.loop.unroll.full``' Metadata
3188 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3190 This metadata either suggests that the loop should be unrolled fully. The
3191 metadata has a single operand which is the string ``llvm.loop.unroll.disable``.
3194 .. code-block:: llvm
3196 !0 = !{!"llvm.loop.unroll.full"}
3201 Metadata types used to annotate memory accesses with information helpful
3202 for optimizations are prefixed with ``llvm.mem``.
3204 '``llvm.mem.parallel_loop_access``' Metadata
3205 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3207 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
3208 or metadata containing a list of loop identifiers for nested loops.
3209 The metadata is attached to memory accessing instructions and denotes that
3210 no loop carried memory dependence exist between it and other instructions denoted
3211 with the same loop identifier.
3213 Precisely, given two instructions ``m1`` and ``m2`` that both have the
3214 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
3215 set of loops associated with that metadata, respectively, then there is no loop
3216 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
3219 As a special case, if all memory accessing instructions in a loop have
3220 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
3221 loop has no loop carried memory dependences and is considered to be a parallel
3224 Note that if not all memory access instructions have such metadata referring to
3225 the loop, then the loop is considered not being trivially parallel. Additional
3226 memory dependence analysis is required to make that determination. As a fail
3227 safe mechanism, this causes loops that were originally parallel to be considered
3228 sequential (if optimization passes that are unaware of the parallel semantics
3229 insert new memory instructions into the loop body).
3231 Example of a loop that is considered parallel due to its correct use of
3232 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
3233 metadata types that refer to the same loop identifier metadata.
3235 .. code-block:: llvm
3239 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
3241 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3243 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
3249 It is also possible to have nested parallel loops. In that case the
3250 memory accesses refer to a list of loop identifier metadata nodes instead of
3251 the loop identifier metadata node directly:
3253 .. code-block:: llvm
3257 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
3259 br label %inner.for.body
3263 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3265 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
3267 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
3271 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
3273 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
3275 outer.for.end: ; preds = %for.body
3277 !0 = !{!1, !2} ; a list of loop identifiers
3278 !1 = !{!1} ; an identifier for the inner loop
3279 !2 = !{!2} ; an identifier for the outer loop
3281 Module Flags Metadata
3282 =====================
3284 Information about the module as a whole is difficult to convey to LLVM's
3285 subsystems. The LLVM IR isn't sufficient to transmit this information.
3286 The ``llvm.module.flags`` named metadata exists in order to facilitate
3287 this. These flags are in the form of key / value pairs --- much like a
3288 dictionary --- making it easy for any subsystem who cares about a flag to
3291 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
3292 Each triplet has the following form:
3294 - The first element is a *behavior* flag, which specifies the behavior
3295 when two (or more) modules are merged together, and it encounters two
3296 (or more) metadata with the same ID. The supported behaviors are
3298 - The second element is a metadata string that is a unique ID for the
3299 metadata. Each module may only have one flag entry for each unique ID (not
3300 including entries with the **Require** behavior).
3301 - The third element is the value of the flag.
3303 When two (or more) modules are merged together, the resulting
3304 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
3305 each unique metadata ID string, there will be exactly one entry in the merged
3306 modules ``llvm.module.flags`` metadata table, and the value for that entry will
3307 be determined by the merge behavior flag, as described below. The only exception
3308 is that entries with the *Require* behavior are always preserved.
3310 The following behaviors are supported:
3321 Emits an error if two values disagree, otherwise the resulting value
3322 is that of the operands.
3326 Emits a warning if two values disagree. The result value will be the
3327 operand for the flag from the first module being linked.
3331 Adds a requirement that another module flag be present and have a
3332 specified value after linking is performed. The value must be a
3333 metadata pair, where the first element of the pair is the ID of the
3334 module flag to be restricted, and the second element of the pair is
3335 the value the module flag should be restricted to. This behavior can
3336 be used to restrict the allowable results (via triggering of an
3337 error) of linking IDs with the **Override** behavior.
3341 Uses the specified value, regardless of the behavior or value of the
3342 other module. If both modules specify **Override**, but the values
3343 differ, an error will be emitted.
3347 Appends the two values, which are required to be metadata nodes.
3351 Appends the two values, which are required to be metadata
3352 nodes. However, duplicate entries in the second list are dropped
3353 during the append operation.
3355 It is an error for a particular unique flag ID to have multiple behaviors,
3356 except in the case of **Require** (which adds restrictions on another metadata
3357 value) or **Override**.
3359 An example of module flags:
3361 .. code-block:: llvm
3363 !0 = !{ i32 1, !"foo", i32 1 }
3364 !1 = !{ i32 4, !"bar", i32 37 }
3365 !2 = !{ i32 2, !"qux", i32 42 }
3366 !3 = !{ i32 3, !"qux",
3371 !llvm.module.flags = !{ !0, !1, !2, !3 }
3373 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3374 if two or more ``!"foo"`` flags are seen is to emit an error if their
3375 values are not equal.
3377 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3378 behavior if two or more ``!"bar"`` flags are seen is to use the value
3381 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3382 behavior if two or more ``!"qux"`` flags are seen is to emit a
3383 warning if their values are not equal.
3385 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3391 The behavior is to emit an error if the ``llvm.module.flags`` does not
3392 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3395 Objective-C Garbage Collection Module Flags Metadata
3396 ----------------------------------------------------
3398 On the Mach-O platform, Objective-C stores metadata about garbage
3399 collection in a special section called "image info". The metadata
3400 consists of a version number and a bitmask specifying what types of
3401 garbage collection are supported (if any) by the file. If two or more
3402 modules are linked together their garbage collection metadata needs to
3403 be merged rather than appended together.
3405 The Objective-C garbage collection module flags metadata consists of the
3406 following key-value pairs:
3415 * - ``Objective-C Version``
3416 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3418 * - ``Objective-C Image Info Version``
3419 - **[Required]** --- The version of the image info section. Currently
3422 * - ``Objective-C Image Info Section``
3423 - **[Required]** --- The section to place the metadata. Valid values are
3424 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3425 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3426 Objective-C ABI version 2.
3428 * - ``Objective-C Garbage Collection``
3429 - **[Required]** --- Specifies whether garbage collection is supported or
3430 not. Valid values are 0, for no garbage collection, and 2, for garbage
3431 collection supported.
3433 * - ``Objective-C GC Only``
3434 - **[Optional]** --- Specifies that only garbage collection is supported.
3435 If present, its value must be 6. This flag requires that the
3436 ``Objective-C Garbage Collection`` flag have the value 2.
3438 Some important flag interactions:
3440 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3441 merged with a module with ``Objective-C Garbage Collection`` set to
3442 2, then the resulting module has the
3443 ``Objective-C Garbage Collection`` flag set to 0.
3444 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3445 merged with a module with ``Objective-C GC Only`` set to 6.
3447 Automatic Linker Flags Module Flags Metadata
3448 --------------------------------------------
3450 Some targets support embedding flags to the linker inside individual object
3451 files. Typically this is used in conjunction with language extensions which
3452 allow source files to explicitly declare the libraries they depend on, and have
3453 these automatically be transmitted to the linker via object files.
3455 These flags are encoded in the IR using metadata in the module flags section,
3456 using the ``Linker Options`` key. The merge behavior for this flag is required
3457 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3458 node which should be a list of other metadata nodes, each of which should be a
3459 list of metadata strings defining linker options.
3461 For example, the following metadata section specifies two separate sets of
3462 linker options, presumably to link against ``libz`` and the ``Cocoa``
3465 !0 = !{ i32 6, !"Linker Options",
3468 !{ !"-framework", !"Cocoa" } } }
3469 !llvm.module.flags = !{ !0 }
3471 The metadata encoding as lists of lists of options, as opposed to a collapsed
3472 list of options, is chosen so that the IR encoding can use multiple option
3473 strings to specify e.g., a single library, while still having that specifier be
3474 preserved as an atomic element that can be recognized by a target specific
3475 assembly writer or object file emitter.
3477 Each individual option is required to be either a valid option for the target's
3478 linker, or an option that is reserved by the target specific assembly writer or
3479 object file emitter. No other aspect of these options is defined by the IR.
3481 C type width Module Flags Metadata
3482 ----------------------------------
3484 The ARM backend emits a section into each generated object file describing the
3485 options that it was compiled with (in a compiler-independent way) to prevent
3486 linking incompatible objects, and to allow automatic library selection. Some
3487 of these options are not visible at the IR level, namely wchar_t width and enum
3490 To pass this information to the backend, these options are encoded in module
3491 flags metadata, using the following key-value pairs:
3501 - * 0 --- sizeof(wchar_t) == 4
3502 * 1 --- sizeof(wchar_t) == 2
3505 - * 0 --- Enums are at least as large as an ``int``.
3506 * 1 --- Enums are stored in the smallest integer type which can
3507 represent all of its values.
3509 For example, the following metadata section specifies that the module was
3510 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
3511 enum is the smallest type which can represent all of its values::
3513 !llvm.module.flags = !{!0, !1}
3514 !0 = !{i32 1, !"short_wchar", i32 1}
3515 !1 = !{i32 1, !"short_enum", i32 0}
3517 .. _intrinsicglobalvariables:
3519 Intrinsic Global Variables
3520 ==========================
3522 LLVM has a number of "magic" global variables that contain data that
3523 affect code generation or other IR semantics. These are documented here.
3524 All globals of this sort should have a section specified as
3525 "``llvm.metadata``". This section and all globals that start with
3526 "``llvm.``" are reserved for use by LLVM.
3530 The '``llvm.used``' Global Variable
3531 -----------------------------------
3533 The ``@llvm.used`` global is an array which has
3534 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3535 pointers to named global variables, functions and aliases which may optionally
3536 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3539 .. code-block:: llvm
3544 @llvm.used = appending global [2 x i8*] [
3546 i8* bitcast (i32* @Y to i8*)
3547 ], section "llvm.metadata"
3549 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3550 and linker are required to treat the symbol as if there is a reference to the
3551 symbol that it cannot see (which is why they have to be named). For example, if
3552 a variable has internal linkage and no references other than that from the
3553 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3554 references from inline asms and other things the compiler cannot "see", and
3555 corresponds to "``attribute((used))``" in GNU C.
3557 On some targets, the code generator must emit a directive to the
3558 assembler or object file to prevent the assembler and linker from
3559 molesting the symbol.
3561 .. _gv_llvmcompilerused:
3563 The '``llvm.compiler.used``' Global Variable
3564 --------------------------------------------
3566 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3567 directive, except that it only prevents the compiler from touching the
3568 symbol. On targets that support it, this allows an intelligent linker to
3569 optimize references to the symbol without being impeded as it would be
3572 This is a rare construct that should only be used in rare circumstances,
3573 and should not be exposed to source languages.
3575 .. _gv_llvmglobalctors:
3577 The '``llvm.global_ctors``' Global Variable
3578 -------------------------------------------
3580 .. code-block:: llvm
3582 %0 = type { i32, void ()*, i8* }
3583 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
3585 The ``@llvm.global_ctors`` array contains a list of constructor
3586 functions, priorities, and an optional associated global or function.
3587 The functions referenced by this array will be called in ascending order
3588 of priority (i.e. lowest first) when the module is loaded. The order of
3589 functions with the same priority is not defined.
3591 If the third field is present, non-null, and points to a global variable
3592 or function, the initializer function will only run if the associated
3593 data from the current module is not discarded.
3595 .. _llvmglobaldtors:
3597 The '``llvm.global_dtors``' Global Variable
3598 -------------------------------------------
3600 .. code-block:: llvm
3602 %0 = type { i32, void ()*, i8* }
3603 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
3605 The ``@llvm.global_dtors`` array contains a list of destructor
3606 functions, priorities, and an optional associated global or function.
3607 The functions referenced by this array will be called in descending
3608 order of priority (i.e. highest first) when the module is unloaded. The
3609 order of functions with the same priority is not defined.
3611 If the third field is present, non-null, and points to a global variable
3612 or function, the destructor function will only run if the associated
3613 data from the current module is not discarded.
3615 Instruction Reference
3616 =====================
3618 The LLVM instruction set consists of several different classifications
3619 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3620 instructions <binaryops>`, :ref:`bitwise binary
3621 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3622 :ref:`other instructions <otherops>`.
3626 Terminator Instructions
3627 -----------------------
3629 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3630 program ends with a "Terminator" instruction, which indicates which
3631 block should be executed after the current block is finished. These
3632 terminator instructions typically yield a '``void``' value: they produce
3633 control flow, not values (the one exception being the
3634 ':ref:`invoke <i_invoke>`' instruction).
3636 The terminator instructions are: ':ref:`ret <i_ret>`',
3637 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3638 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3639 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3643 '``ret``' Instruction
3644 ^^^^^^^^^^^^^^^^^^^^^
3651 ret <type> <value> ; Return a value from a non-void function
3652 ret void ; Return from void function
3657 The '``ret``' instruction is used to return control flow (and optionally
3658 a value) from a function back to the caller.
3660 There are two forms of the '``ret``' instruction: one that returns a
3661 value and then causes control flow, and one that just causes control
3667 The '``ret``' instruction optionally accepts a single argument, the
3668 return value. The type of the return value must be a ':ref:`first
3669 class <t_firstclass>`' type.
3671 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3672 return type and contains a '``ret``' instruction with no return value or
3673 a return value with a type that does not match its type, or if it has a
3674 void return type and contains a '``ret``' instruction with a return
3680 When the '``ret``' instruction is executed, control flow returns back to
3681 the calling function's context. If the caller is a
3682 ":ref:`call <i_call>`" instruction, execution continues at the
3683 instruction after the call. If the caller was an
3684 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3685 beginning of the "normal" destination block. If the instruction returns
3686 a value, that value shall set the call or invoke instruction's return
3692 .. code-block:: llvm
3694 ret i32 5 ; Return an integer value of 5
3695 ret void ; Return from a void function
3696 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3700 '``br``' Instruction
3701 ^^^^^^^^^^^^^^^^^^^^
3708 br i1 <cond>, label <iftrue>, label <iffalse>
3709 br label <dest> ; Unconditional branch
3714 The '``br``' instruction is used to cause control flow to transfer to a
3715 different basic block in the current function. There are two forms of
3716 this instruction, corresponding to a conditional branch and an
3717 unconditional branch.
3722 The conditional branch form of the '``br``' instruction takes a single
3723 '``i1``' value and two '``label``' values. The unconditional form of the
3724 '``br``' instruction takes a single '``label``' value as a target.
3729 Upon execution of a conditional '``br``' instruction, the '``i1``'
3730 argument is evaluated. If the value is ``true``, control flows to the
3731 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3732 to the '``iffalse``' ``label`` argument.
3737 .. code-block:: llvm
3740 %cond = icmp eq i32 %a, %b
3741 br i1 %cond, label %IfEqual, label %IfUnequal
3749 '``switch``' Instruction
3750 ^^^^^^^^^^^^^^^^^^^^^^^^
3757 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3762 The '``switch``' instruction is used to transfer control flow to one of
3763 several different places. It is a generalization of the '``br``'
3764 instruction, allowing a branch to occur to one of many possible
3770 The '``switch``' instruction uses three parameters: an integer
3771 comparison value '``value``', a default '``label``' destination, and an
3772 array of pairs of comparison value constants and '``label``'s. The table
3773 is not allowed to contain duplicate constant entries.
3778 The ``switch`` instruction specifies a table of values and destinations.
3779 When the '``switch``' instruction is executed, this table is searched
3780 for the given value. If the value is found, control flow is transferred
3781 to the corresponding destination; otherwise, control flow is transferred
3782 to the default destination.
3787 Depending on properties of the target machine and the particular
3788 ``switch`` instruction, this instruction may be code generated in
3789 different ways. For example, it could be generated as a series of
3790 chained conditional branches or with a lookup table.
3795 .. code-block:: llvm
3797 ; Emulate a conditional br instruction
3798 %Val = zext i1 %value to i32
3799 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3801 ; Emulate an unconditional br instruction
3802 switch i32 0, label %dest [ ]
3804 ; Implement a jump table:
3805 switch i32 %val, label %otherwise [ i32 0, label %onzero
3807 i32 2, label %ontwo ]
3811 '``indirectbr``' Instruction
3812 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3819 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3824 The '``indirectbr``' instruction implements an indirect branch to a
3825 label within the current function, whose address is specified by
3826 "``address``". Address must be derived from a
3827 :ref:`blockaddress <blockaddress>` constant.
3832 The '``address``' argument is the address of the label to jump to. The
3833 rest of the arguments indicate the full set of possible destinations
3834 that the address may point to. Blocks are allowed to occur multiple
3835 times in the destination list, though this isn't particularly useful.
3837 This destination list is required so that dataflow analysis has an
3838 accurate understanding of the CFG.
3843 Control transfers to the block specified in the address argument. All
3844 possible destination blocks must be listed in the label list, otherwise
3845 this instruction has undefined behavior. This implies that jumps to
3846 labels defined in other functions have undefined behavior as well.
3851 This is typically implemented with a jump through a register.
3856 .. code-block:: llvm
3858 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3862 '``invoke``' Instruction
3863 ^^^^^^^^^^^^^^^^^^^^^^^^
3870 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3871 to label <normal label> unwind label <exception label>
3876 The '``invoke``' instruction causes control to transfer to a specified
3877 function, with the possibility of control flow transfer to either the
3878 '``normal``' label or the '``exception``' label. If the callee function
3879 returns with the "``ret``" instruction, control flow will return to the
3880 "normal" label. If the callee (or any indirect callees) returns via the
3881 ":ref:`resume <i_resume>`" instruction or other exception handling
3882 mechanism, control is interrupted and continued at the dynamically
3883 nearest "exception" label.
3885 The '``exception``' label is a `landing
3886 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3887 '``exception``' label is required to have the
3888 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3889 information about the behavior of the program after unwinding happens,
3890 as its first non-PHI instruction. The restrictions on the
3891 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3892 instruction, so that the important information contained within the
3893 "``landingpad``" instruction can't be lost through normal code motion.
3898 This instruction requires several arguments:
3900 #. The optional "cconv" marker indicates which :ref:`calling
3901 convention <callingconv>` the call should use. If none is
3902 specified, the call defaults to using C calling conventions.
3903 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3904 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3906 #. '``ptr to function ty``': shall be the signature of the pointer to
3907 function value being invoked. In most cases, this is a direct
3908 function invocation, but indirect ``invoke``'s are just as possible,
3909 branching off an arbitrary pointer to function value.
3910 #. '``function ptr val``': An LLVM value containing a pointer to a
3911 function to be invoked.
3912 #. '``function args``': argument list whose types match the function
3913 signature argument types and parameter attributes. All arguments must
3914 be of :ref:`first class <t_firstclass>` type. If the function signature
3915 indicates the function accepts a variable number of arguments, the
3916 extra arguments can be specified.
3917 #. '``normal label``': the label reached when the called function
3918 executes a '``ret``' instruction.
3919 #. '``exception label``': the label reached when a callee returns via
3920 the :ref:`resume <i_resume>` instruction or other exception handling
3922 #. The optional :ref:`function attributes <fnattrs>` list. Only
3923 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3924 attributes are valid here.
3929 This instruction is designed to operate as a standard '``call``'
3930 instruction in most regards. The primary difference is that it
3931 establishes an association with a label, which is used by the runtime
3932 library to unwind the stack.
3934 This instruction is used in languages with destructors to ensure that
3935 proper cleanup is performed in the case of either a ``longjmp`` or a
3936 thrown exception. Additionally, this is important for implementation of
3937 '``catch``' clauses in high-level languages that support them.
3939 For the purposes of the SSA form, the definition of the value returned
3940 by the '``invoke``' instruction is deemed to occur on the edge from the
3941 current block to the "normal" label. If the callee unwinds then no
3942 return value is available.
3947 .. code-block:: llvm
3949 %retval = invoke i32 @Test(i32 15) to label %Continue
3950 unwind label %TestCleanup ; i32:retval set
3951 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3952 unwind label %TestCleanup ; i32:retval set
3956 '``resume``' Instruction
3957 ^^^^^^^^^^^^^^^^^^^^^^^^
3964 resume <type> <value>
3969 The '``resume``' instruction is a terminator instruction that has no
3975 The '``resume``' instruction requires one argument, which must have the
3976 same type as the result of any '``landingpad``' instruction in the same
3982 The '``resume``' instruction resumes propagation of an existing
3983 (in-flight) exception whose unwinding was interrupted with a
3984 :ref:`landingpad <i_landingpad>` instruction.
3989 .. code-block:: llvm
3991 resume { i8*, i32 } %exn
3995 '``unreachable``' Instruction
3996 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4008 The '``unreachable``' instruction has no defined semantics. This
4009 instruction is used to inform the optimizer that a particular portion of
4010 the code is not reachable. This can be used to indicate that the code
4011 after a no-return function cannot be reached, and other facts.
4016 The '``unreachable``' instruction has no defined semantics.
4023 Binary operators are used to do most of the computation in a program.
4024 They require two operands of the same type, execute an operation on
4025 them, and produce a single value. The operands might represent multiple
4026 data, as is the case with the :ref:`vector <t_vector>` data type. The
4027 result value has the same type as its operands.
4029 There are several different binary operators:
4033 '``add``' Instruction
4034 ^^^^^^^^^^^^^^^^^^^^^
4041 <result> = add <ty> <op1>, <op2> ; yields ty:result
4042 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
4043 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
4044 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
4049 The '``add``' instruction returns the sum of its two operands.
4054 The two arguments to the '``add``' instruction must be
4055 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4056 arguments must have identical types.
4061 The value produced is the integer sum of the two operands.
4063 If the sum has unsigned overflow, the result returned is the
4064 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4067 Because LLVM integers use a two's complement representation, this
4068 instruction is appropriate for both signed and unsigned integers.
4070 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4071 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4072 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
4073 unsigned and/or signed overflow, respectively, occurs.
4078 .. code-block:: llvm
4080 <result> = add i32 4, %var ; yields i32:result = 4 + %var
4084 '``fadd``' Instruction
4085 ^^^^^^^^^^^^^^^^^^^^^^
4092 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4097 The '``fadd``' instruction returns the sum of its two operands.
4102 The two arguments to the '``fadd``' instruction must be :ref:`floating
4103 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4104 Both arguments must have identical types.
4109 The value produced is the floating point sum of the two operands. This
4110 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
4111 which are optimization hints to enable otherwise unsafe floating point
4117 .. code-block:: llvm
4119 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
4121 '``sub``' Instruction
4122 ^^^^^^^^^^^^^^^^^^^^^
4129 <result> = sub <ty> <op1>, <op2> ; yields ty:result
4130 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
4131 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
4132 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
4137 The '``sub``' instruction returns the difference of its two operands.
4139 Note that the '``sub``' instruction is used to represent the '``neg``'
4140 instruction present in most other intermediate representations.
4145 The two arguments to the '``sub``' instruction must be
4146 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4147 arguments must have identical types.
4152 The value produced is the integer difference of the two operands.
4154 If the difference has unsigned overflow, the result returned is the
4155 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4158 Because LLVM integers use a two's complement representation, this
4159 instruction is appropriate for both signed and unsigned integers.
4161 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4162 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4163 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
4164 unsigned and/or signed overflow, respectively, occurs.
4169 .. code-block:: llvm
4171 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
4172 <result> = sub i32 0, %val ; yields i32:result = -%var
4176 '``fsub``' Instruction
4177 ^^^^^^^^^^^^^^^^^^^^^^
4184 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4189 The '``fsub``' instruction returns the difference of its two operands.
4191 Note that the '``fsub``' instruction is used to represent the '``fneg``'
4192 instruction present in most other intermediate representations.
4197 The two arguments to the '``fsub``' instruction must be :ref:`floating
4198 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4199 Both arguments must have identical types.
4204 The value produced is the floating point difference of the two operands.
4205 This instruction can also take any number of :ref:`fast-math
4206 flags <fastmath>`, which are optimization hints to enable otherwise
4207 unsafe floating point optimizations:
4212 .. code-block:: llvm
4214 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
4215 <result> = fsub float -0.0, %val ; yields float:result = -%var
4217 '``mul``' Instruction
4218 ^^^^^^^^^^^^^^^^^^^^^
4225 <result> = mul <ty> <op1>, <op2> ; yields ty:result
4226 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
4227 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
4228 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
4233 The '``mul``' instruction returns the product of its two operands.
4238 The two arguments to the '``mul``' instruction must be
4239 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4240 arguments must have identical types.
4245 The value produced is the integer product of the two operands.
4247 If the result of the multiplication has unsigned overflow, the result
4248 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
4249 bit width of the result.
4251 Because LLVM integers use a two's complement representation, and the
4252 result is the same width as the operands, this instruction returns the
4253 correct result for both signed and unsigned integers. If a full product
4254 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
4255 sign-extended or zero-extended as appropriate to the width of the full
4258 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4259 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4260 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
4261 unsigned and/or signed overflow, respectively, occurs.
4266 .. code-block:: llvm
4268 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
4272 '``fmul``' Instruction
4273 ^^^^^^^^^^^^^^^^^^^^^^
4280 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4285 The '``fmul``' instruction returns the product of its two operands.
4290 The two arguments to the '``fmul``' instruction must be :ref:`floating
4291 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4292 Both arguments must have identical types.
4297 The value produced is the floating point product of the two operands.
4298 This instruction can also take any number of :ref:`fast-math
4299 flags <fastmath>`, which are optimization hints to enable otherwise
4300 unsafe floating point optimizations:
4305 .. code-block:: llvm
4307 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
4309 '``udiv``' Instruction
4310 ^^^^^^^^^^^^^^^^^^^^^^
4317 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
4318 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
4323 The '``udiv``' instruction returns the quotient of its two operands.
4328 The two arguments to the '``udiv``' instruction must be
4329 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4330 arguments must have identical types.
4335 The value produced is the unsigned integer quotient of the two operands.
4337 Note that unsigned integer division and signed integer division are
4338 distinct operations; for signed integer division, use '``sdiv``'.
4340 Division by zero leads to undefined behavior.
4342 If the ``exact`` keyword is present, the result value of the ``udiv`` is
4343 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
4344 such, "((a udiv exact b) mul b) == a").
4349 .. code-block:: llvm
4351 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
4353 '``sdiv``' Instruction
4354 ^^^^^^^^^^^^^^^^^^^^^^
4361 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
4362 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
4367 The '``sdiv``' instruction returns the quotient of its two operands.
4372 The two arguments to the '``sdiv``' instruction must be
4373 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4374 arguments must have identical types.
4379 The value produced is the signed integer quotient of the two operands
4380 rounded towards zero.
4382 Note that signed integer division and unsigned integer division are
4383 distinct operations; for unsigned integer division, use '``udiv``'.
4385 Division by zero leads to undefined behavior. Overflow also leads to
4386 undefined behavior; this is a rare case, but can occur, for example, by
4387 doing a 32-bit division of -2147483648 by -1.
4389 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4390 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4395 .. code-block:: llvm
4397 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
4401 '``fdiv``' Instruction
4402 ^^^^^^^^^^^^^^^^^^^^^^
4409 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4414 The '``fdiv``' instruction returns the quotient of its two operands.
4419 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4420 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4421 Both arguments must have identical types.
4426 The value produced is the floating point quotient of the two operands.
4427 This instruction can also take any number of :ref:`fast-math
4428 flags <fastmath>`, which are optimization hints to enable otherwise
4429 unsafe floating point optimizations:
4434 .. code-block:: llvm
4436 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
4438 '``urem``' Instruction
4439 ^^^^^^^^^^^^^^^^^^^^^^
4446 <result> = urem <ty> <op1>, <op2> ; yields ty:result
4451 The '``urem``' instruction returns the remainder from the unsigned
4452 division of its two arguments.
4457 The two arguments to the '``urem``' instruction must be
4458 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4459 arguments must have identical types.
4464 This instruction returns the unsigned integer *remainder* of a division.
4465 This instruction always performs an unsigned division to get the
4468 Note that unsigned integer remainder and signed integer remainder are
4469 distinct operations; for signed integer remainder, use '``srem``'.
4471 Taking the remainder of a division by zero leads to undefined behavior.
4476 .. code-block:: llvm
4478 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
4480 '``srem``' Instruction
4481 ^^^^^^^^^^^^^^^^^^^^^^
4488 <result> = srem <ty> <op1>, <op2> ; yields ty:result
4493 The '``srem``' instruction returns the remainder from the signed
4494 division of its two operands. This instruction can also take
4495 :ref:`vector <t_vector>` versions of the values in which case the elements
4501 The two arguments to the '``srem``' instruction must be
4502 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4503 arguments must have identical types.
4508 This instruction returns the *remainder* of a division (where the result
4509 is either zero or has the same sign as the dividend, ``op1``), not the
4510 *modulo* operator (where the result is either zero or has the same sign
4511 as the divisor, ``op2``) of a value. For more information about the
4512 difference, see `The Math
4513 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4514 table of how this is implemented in various languages, please see
4516 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4518 Note that signed integer remainder and unsigned integer remainder are
4519 distinct operations; for unsigned integer remainder, use '``urem``'.
4521 Taking the remainder of a division by zero leads to undefined behavior.
4522 Overflow also leads to undefined behavior; this is a rare case, but can
4523 occur, for example, by taking the remainder of a 32-bit division of
4524 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4525 rule lets srem be implemented using instructions that return both the
4526 result of the division and the remainder.)
4531 .. code-block:: llvm
4533 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
4537 '``frem``' Instruction
4538 ^^^^^^^^^^^^^^^^^^^^^^
4545 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4550 The '``frem``' instruction returns the remainder from the division of
4556 The two arguments to the '``frem``' instruction must be :ref:`floating
4557 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4558 Both arguments must have identical types.
4563 This instruction returns the *remainder* of a division. The remainder
4564 has the same sign as the dividend. This instruction can also take any
4565 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4566 to enable otherwise unsafe floating point optimizations:
4571 .. code-block:: llvm
4573 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
4577 Bitwise Binary Operations
4578 -------------------------
4580 Bitwise binary operators are used to do various forms of bit-twiddling
4581 in a program. They are generally very efficient instructions and can
4582 commonly be strength reduced from other instructions. They require two
4583 operands of the same type, execute an operation on them, and produce a
4584 single value. The resulting value is the same type as its operands.
4586 '``shl``' Instruction
4587 ^^^^^^^^^^^^^^^^^^^^^
4594 <result> = shl <ty> <op1>, <op2> ; yields ty:result
4595 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
4596 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
4597 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
4602 The '``shl``' instruction returns the first operand shifted to the left
4603 a specified number of bits.
4608 Both arguments to the '``shl``' instruction must be the same
4609 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4610 '``op2``' is treated as an unsigned value.
4615 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4616 where ``n`` is the width of the result. If ``op2`` is (statically or
4617 dynamically) negative or equal to or larger than the number of bits in
4618 ``op1``, the result is undefined. If the arguments are vectors, each
4619 vector element of ``op1`` is shifted by the corresponding shift amount
4622 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4623 value <poisonvalues>` if it shifts out any non-zero bits. If the
4624 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4625 value <poisonvalues>` if it shifts out any bits that disagree with the
4626 resultant sign bit. As such, NUW/NSW have the same semantics as they
4627 would if the shift were expressed as a mul instruction with the same
4628 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4633 .. code-block:: llvm
4635 <result> = shl i32 4, %var ; yields i32: 4 << %var
4636 <result> = shl i32 4, 2 ; yields i32: 16
4637 <result> = shl i32 1, 10 ; yields i32: 1024
4638 <result> = shl i32 1, 32 ; undefined
4639 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4641 '``lshr``' Instruction
4642 ^^^^^^^^^^^^^^^^^^^^^^
4649 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
4650 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
4655 The '``lshr``' instruction (logical shift right) returns the first
4656 operand shifted to the right a specified number of bits with zero fill.
4661 Both arguments to the '``lshr``' instruction must be the same
4662 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4663 '``op2``' is treated as an unsigned value.
4668 This instruction always performs a logical shift right operation. The
4669 most significant bits of the result will be filled with zero bits after
4670 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4671 than the number of bits in ``op1``, the result is undefined. If the
4672 arguments are vectors, each vector element of ``op1`` is shifted by the
4673 corresponding shift amount in ``op2``.
4675 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4676 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4682 .. code-block:: llvm
4684 <result> = lshr i32 4, 1 ; yields i32:result = 2
4685 <result> = lshr i32 4, 2 ; yields i32:result = 1
4686 <result> = lshr i8 4, 3 ; yields i8:result = 0
4687 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
4688 <result> = lshr i32 1, 32 ; undefined
4689 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4691 '``ashr``' Instruction
4692 ^^^^^^^^^^^^^^^^^^^^^^
4699 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
4700 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
4705 The '``ashr``' instruction (arithmetic shift right) returns the first
4706 operand shifted to the right a specified number of bits with sign
4712 Both arguments to the '``ashr``' instruction must be the same
4713 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4714 '``op2``' is treated as an unsigned value.
4719 This instruction always performs an arithmetic shift right operation,
4720 The most significant bits of the result will be filled with the sign bit
4721 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4722 than the number of bits in ``op1``, the result is undefined. If the
4723 arguments are vectors, each vector element of ``op1`` is shifted by the
4724 corresponding shift amount in ``op2``.
4726 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4727 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4733 .. code-block:: llvm
4735 <result> = ashr i32 4, 1 ; yields i32:result = 2
4736 <result> = ashr i32 4, 2 ; yields i32:result = 1
4737 <result> = ashr i8 4, 3 ; yields i8:result = 0
4738 <result> = ashr i8 -2, 1 ; yields i8:result = -1
4739 <result> = ashr i32 1, 32 ; undefined
4740 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4742 '``and``' Instruction
4743 ^^^^^^^^^^^^^^^^^^^^^
4750 <result> = and <ty> <op1>, <op2> ; yields ty:result
4755 The '``and``' instruction returns the bitwise logical and of its two
4761 The two arguments to the '``and``' 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 truth table used for the '``and``' instruction is:
4785 .. code-block:: llvm
4787 <result> = and i32 4, %var ; yields i32:result = 4 & %var
4788 <result> = and i32 15, 40 ; yields i32:result = 8
4789 <result> = and i32 4, 8 ; yields i32:result = 0
4791 '``or``' Instruction
4792 ^^^^^^^^^^^^^^^^^^^^
4799 <result> = or <ty> <op1>, <op2> ; yields ty:result
4804 The '``or``' instruction returns the bitwise logical inclusive or of its
4810 The two arguments to the '``or``' instruction must be
4811 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4812 arguments must have identical types.
4817 The truth table used for the '``or``' instruction is:
4836 <result> = or i32 4, %var ; yields i32:result = 4 | %var
4837 <result> = or i32 15, 40 ; yields i32:result = 47
4838 <result> = or i32 4, 8 ; yields i32:result = 12
4840 '``xor``' Instruction
4841 ^^^^^^^^^^^^^^^^^^^^^
4848 <result> = xor <ty> <op1>, <op2> ; yields ty:result
4853 The '``xor``' instruction returns the bitwise logical exclusive or of
4854 its two operands. The ``xor`` is used to implement the "one's
4855 complement" operation, which is the "~" operator in C.
4860 The two arguments to the '``xor``' instruction must be
4861 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4862 arguments must have identical types.
4867 The truth table used for the '``xor``' instruction is:
4884 .. code-block:: llvm
4886 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
4887 <result> = xor i32 15, 40 ; yields i32:result = 39
4888 <result> = xor i32 4, 8 ; yields i32:result = 12
4889 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
4894 LLVM supports several instructions to represent vector operations in a
4895 target-independent manner. These instructions cover the element-access
4896 and vector-specific operations needed to process vectors effectively.
4897 While LLVM does directly support these vector operations, many
4898 sophisticated algorithms will want to use target-specific intrinsics to
4899 take full advantage of a specific target.
4901 .. _i_extractelement:
4903 '``extractelement``' Instruction
4904 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4911 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
4916 The '``extractelement``' instruction extracts a single scalar element
4917 from a vector at a specified index.
4922 The first operand of an '``extractelement``' instruction is a value of
4923 :ref:`vector <t_vector>` type. The second operand is an index indicating
4924 the position from which to extract the element. The index may be a
4925 variable of any integer type.
4930 The result is a scalar of the same type as the element type of ``val``.
4931 Its value is the value at position ``idx`` of ``val``. If ``idx``
4932 exceeds the length of ``val``, the results are undefined.
4937 .. code-block:: llvm
4939 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4941 .. _i_insertelement:
4943 '``insertelement``' Instruction
4944 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4951 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
4956 The '``insertelement``' instruction inserts a scalar element into a
4957 vector at a specified index.
4962 The first operand of an '``insertelement``' instruction is a value of
4963 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4964 type must equal the element type of the first operand. The third operand
4965 is an index indicating the position at which to insert the value. The
4966 index may be a variable of any integer type.
4971 The result is a vector of the same type as ``val``. Its element values
4972 are those of ``val`` except at position ``idx``, where it gets the value
4973 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4979 .. code-block:: llvm
4981 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4983 .. _i_shufflevector:
4985 '``shufflevector``' Instruction
4986 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4993 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4998 The '``shufflevector``' instruction constructs a permutation of elements
4999 from two input vectors, returning a vector with the same element type as
5000 the input and length that is the same as the shuffle mask.
5005 The first two operands of a '``shufflevector``' instruction are vectors
5006 with the same type. The third argument is a shuffle mask whose element
5007 type is always 'i32'. The result of the instruction is a vector whose
5008 length is the same as the shuffle mask and whose element type is the
5009 same as the element type of the first two operands.
5011 The shuffle mask operand is required to be a constant vector with either
5012 constant integer or undef values.
5017 The elements of the two input vectors are numbered from left to right
5018 across both of the vectors. The shuffle mask operand specifies, for each
5019 element of the result vector, which element of the two input vectors the
5020 result element gets. The element selector may be undef (meaning "don't
5021 care") and the second operand may be undef if performing a shuffle from
5027 .. code-block:: llvm
5029 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5030 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
5031 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
5032 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
5033 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
5034 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
5035 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5036 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
5038 Aggregate Operations
5039 --------------------
5041 LLVM supports several instructions for working with
5042 :ref:`aggregate <t_aggregate>` values.
5046 '``extractvalue``' Instruction
5047 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5054 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
5059 The '``extractvalue``' instruction extracts the value of a member field
5060 from an :ref:`aggregate <t_aggregate>` value.
5065 The first operand of an '``extractvalue``' instruction is a value of
5066 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
5067 constant indices to specify which value to extract in a similar manner
5068 as indices in a '``getelementptr``' instruction.
5070 The major differences to ``getelementptr`` indexing are:
5072 - Since the value being indexed is not a pointer, the first index is
5073 omitted and assumed to be zero.
5074 - At least one index must be specified.
5075 - Not only struct indices but also array indices must be in bounds.
5080 The result is the value at the position in the aggregate specified by
5086 .. code-block:: llvm
5088 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
5092 '``insertvalue``' Instruction
5093 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5100 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
5105 The '``insertvalue``' instruction inserts a value into a member field in
5106 an :ref:`aggregate <t_aggregate>` value.
5111 The first operand of an '``insertvalue``' instruction is a value of
5112 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
5113 a first-class value to insert. The following operands are constant
5114 indices indicating the position at which to insert the value in a
5115 similar manner as indices in a '``extractvalue``' instruction. The value
5116 to insert must have the same type as the value identified by the
5122 The result is an aggregate of the same type as ``val``. Its value is
5123 that of ``val`` except that the value at the position specified by the
5124 indices is that of ``elt``.
5129 .. code-block:: llvm
5131 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
5132 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
5133 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
5137 Memory Access and Addressing Operations
5138 ---------------------------------------
5140 A key design point of an SSA-based representation is how it represents
5141 memory. In LLVM, no memory locations are in SSA form, which makes things
5142 very simple. This section describes how to read, write, and allocate
5147 '``alloca``' Instruction
5148 ^^^^^^^^^^^^^^^^^^^^^^^^
5155 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
5160 The '``alloca``' instruction allocates memory on the stack frame of the
5161 currently executing function, to be automatically released when this
5162 function returns to its caller. The object is always allocated in the
5163 generic address space (address space zero).
5168 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
5169 bytes of memory on the runtime stack, returning a pointer of the
5170 appropriate type to the program. If "NumElements" is specified, it is
5171 the number of elements allocated, otherwise "NumElements" is defaulted
5172 to be one. If a constant alignment is specified, the value result of the
5173 allocation is guaranteed to be aligned to at least that boundary. The
5174 alignment may not be greater than ``1 << 29``. If not specified, or if
5175 zero, the target can choose to align the allocation on any convenient
5176 boundary compatible with the type.
5178 '``type``' may be any sized type.
5183 Memory is allocated; a pointer is returned. The operation is undefined
5184 if there is insufficient stack space for the allocation. '``alloca``'d
5185 memory is automatically released when the function returns. The
5186 '``alloca``' instruction is commonly used to represent automatic
5187 variables that must have an address available. When the function returns
5188 (either with the ``ret`` or ``resume`` instructions), the memory is
5189 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5190 The order in which memory is allocated (ie., which way the stack grows)
5196 .. code-block:: llvm
5198 %ptr = alloca i32 ; yields i32*:ptr
5199 %ptr = alloca i32, i32 4 ; yields i32*:ptr
5200 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
5201 %ptr = alloca i32, align 1024 ; yields i32*:ptr
5205 '``load``' Instruction
5206 ^^^^^^^^^^^^^^^^^^^^^^
5213 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !nonnull !<index>]
5214 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5215 !<index> = !{ i32 1 }
5220 The '``load``' instruction is used to read from memory.
5225 The argument to the ``load`` instruction specifies the memory address
5226 from which to load. The pointer must point to a :ref:`first
5227 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
5228 then the optimizer is not allowed to modify the number or order of
5229 execution of this ``load`` with other :ref:`volatile
5230 operations <volatile>`.
5232 If the ``load`` is marked as ``atomic``, it takes an extra
5233 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5234 ``release`` and ``acq_rel`` orderings are not valid on ``load``
5235 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5236 when they may see multiple atomic stores. The type of the pointee must
5237 be an integer type whose bit width is a power of two greater than or
5238 equal to eight and less than or equal to a target-specific size limit.
5239 ``align`` must be explicitly specified on atomic loads, and the load has
5240 undefined behavior if the alignment is not set to a value which is at
5241 least the size in bytes of the pointee. ``!nontemporal`` does not have
5242 any defined semantics for atomic loads.
5244 The optional constant ``align`` argument specifies the alignment of the
5245 operation (that is, the alignment of the memory address). A value of 0
5246 or an omitted ``align`` argument means that the operation has the ABI
5247 alignment for the target. It is the responsibility of the code emitter
5248 to ensure that the alignment information is correct. Overestimating the
5249 alignment results in undefined behavior. Underestimating the alignment
5250 may produce less efficient code. An alignment of 1 is always safe. The
5251 maximum possible alignment is ``1 << 29``.
5253 The optional ``!nontemporal`` metadata must reference a single
5254 metadata name ``<index>`` corresponding to a metadata node with one
5255 ``i32`` entry of value 1. The existence of the ``!nontemporal``
5256 metadata on the instruction tells the optimizer and code generator
5257 that this load is not expected to be reused in the cache. The code
5258 generator may select special instructions to save cache bandwidth, such
5259 as the ``MOVNT`` instruction on x86.
5261 The optional ``!invariant.load`` metadata must reference a single
5262 metadata name ``<index>`` corresponding to a metadata node with no
5263 entries. The existence of the ``!invariant.load`` metadata on the
5264 instruction tells the optimizer and code generator that the address
5265 operand to this load points to memory which can be assumed unchanged.
5266 Being invariant does not imply that a location is dereferenceable,
5267 but it does imply that once the location is known dereferenceable
5268 its value is henceforth unchanging.
5270 The optional ``!nonnull`` metadata must reference a single
5271 metadata name ``<index>`` corresponding to a metadata node with no
5272 entries. The existence of the ``!nonnull`` metadata on the
5273 instruction tells the optimizer that the value loaded is known to
5274 never be null. This is analogous to the ''nonnull'' attribute
5275 on parameters and return values. This metadata can only be applied
5276 to loads of a pointer type.
5281 The location of memory pointed to is loaded. If the value being loaded
5282 is of scalar type then the number of bytes read does not exceed the
5283 minimum number of bytes needed to hold all bits of the type. For
5284 example, loading an ``i24`` reads at most three bytes. When loading a
5285 value of a type like ``i20`` with a size that is not an integral number
5286 of bytes, the result is undefined if the value was not originally
5287 written using a store of the same type.
5292 .. code-block:: llvm
5294 %ptr = alloca i32 ; yields i32*:ptr
5295 store i32 3, i32* %ptr ; yields void
5296 %val = load i32* %ptr ; yields i32:val = i32 3
5300 '``store``' Instruction
5301 ^^^^^^^^^^^^^^^^^^^^^^^
5308 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
5309 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
5314 The '``store``' instruction is used to write to memory.
5319 There are two arguments to the ``store`` instruction: a value to store
5320 and an address at which to store it. The type of the ``<pointer>``
5321 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
5322 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
5323 then the optimizer is not allowed to modify the number or order of
5324 execution of this ``store`` with other :ref:`volatile
5325 operations <volatile>`.
5327 If the ``store`` is marked as ``atomic``, it takes an extra
5328 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5329 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
5330 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5331 when they may see multiple atomic stores. The type of the pointee must
5332 be an integer type whose bit width is a power of two greater than or
5333 equal to eight and less than or equal to a target-specific size limit.
5334 ``align`` must be explicitly specified on atomic stores, and the store
5335 has undefined behavior if the alignment is not set to a value which is
5336 at least the size in bytes of the pointee. ``!nontemporal`` does not
5337 have any defined semantics for atomic stores.
5339 The optional constant ``align`` argument specifies the alignment of the
5340 operation (that is, the alignment of the memory address). A value of 0
5341 or an omitted ``align`` argument means that the operation has the ABI
5342 alignment for the target. It is the responsibility of the code emitter
5343 to ensure that the alignment information is correct. Overestimating the
5344 alignment results in undefined behavior. Underestimating the
5345 alignment may produce less efficient code. An alignment of 1 is always
5346 safe. The maximum possible alignment is ``1 << 29``.
5348 The optional ``!nontemporal`` metadata must reference a single metadata
5349 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
5350 value 1. The existence of the ``!nontemporal`` metadata on the instruction
5351 tells the optimizer and code generator that this load is not expected to
5352 be reused in the cache. The code generator may select special
5353 instructions to save cache bandwidth, such as the MOVNT instruction on
5359 The contents of memory are updated to contain ``<value>`` at the
5360 location specified by the ``<pointer>`` operand. If ``<value>`` is
5361 of scalar type then the number of bytes written does not exceed the
5362 minimum number of bytes needed to hold all bits of the type. For
5363 example, storing an ``i24`` writes at most three bytes. When writing a
5364 value of a type like ``i20`` with a size that is not an integral number
5365 of bytes, it is unspecified what happens to the extra bits that do not
5366 belong to the type, but they will typically be overwritten.
5371 .. code-block:: llvm
5373 %ptr = alloca i32 ; yields i32*:ptr
5374 store i32 3, i32* %ptr ; yields void
5375 %val = load i32* %ptr ; yields i32:val = i32 3
5379 '``fence``' Instruction
5380 ^^^^^^^^^^^^^^^^^^^^^^^
5387 fence [singlethread] <ordering> ; yields void
5392 The '``fence``' instruction is used to introduce happens-before edges
5398 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5399 defines what *synchronizes-with* edges they add. They can only be given
5400 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5405 A fence A which has (at least) ``release`` ordering semantics
5406 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5407 semantics if and only if there exist atomic operations X and Y, both
5408 operating on some atomic object M, such that A is sequenced before X, X
5409 modifies M (either directly or through some side effect of a sequence
5410 headed by X), Y is sequenced before B, and Y observes M. This provides a
5411 *happens-before* dependency between A and B. Rather than an explicit
5412 ``fence``, one (but not both) of the atomic operations X or Y might
5413 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5414 still *synchronize-with* the explicit ``fence`` and establish the
5415 *happens-before* edge.
5417 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5418 ``acquire`` and ``release`` semantics specified above, participates in
5419 the global program order of other ``seq_cst`` operations and/or fences.
5421 The optional ":ref:`singlethread <singlethread>`" argument specifies
5422 that the fence only synchronizes with other fences in the same thread.
5423 (This is useful for interacting with signal handlers.)
5428 .. code-block:: llvm
5430 fence acquire ; yields void
5431 fence singlethread seq_cst ; yields void
5435 '``cmpxchg``' Instruction
5436 ^^^^^^^^^^^^^^^^^^^^^^^^^
5443 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
5448 The '``cmpxchg``' instruction is used to atomically modify memory. It
5449 loads a value in memory and compares it to a given value. If they are
5450 equal, it tries to store a new value into the memory.
5455 There are three arguments to the '``cmpxchg``' instruction: an address
5456 to operate on, a value to compare to the value currently be at that
5457 address, and a new value to place at that address if the compared values
5458 are equal. The type of '<cmp>' must be an integer type whose bit width
5459 is a power of two greater than or equal to eight and less than or equal
5460 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5461 type, and the type of '<pointer>' must be a pointer to that type. If the
5462 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5463 to modify the number or order of execution of this ``cmpxchg`` with
5464 other :ref:`volatile operations <volatile>`.
5466 The success and failure :ref:`ordering <ordering>` arguments specify how this
5467 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
5468 must be at least ``monotonic``, the ordering constraint on failure must be no
5469 stronger than that on success, and the failure ordering cannot be either
5470 ``release`` or ``acq_rel``.
5472 The optional "``singlethread``" argument declares that the ``cmpxchg``
5473 is only atomic with respect to code (usually signal handlers) running in
5474 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5475 respect to all other code in the system.
5477 The pointer passed into cmpxchg must have alignment greater than or
5478 equal to the size in memory of the operand.
5483 The contents of memory at the location specified by the '``<pointer>``' operand
5484 is read and compared to '``<cmp>``'; if the read value is the equal, the
5485 '``<new>``' is written. The original value at the location is returned, together
5486 with a flag indicating success (true) or failure (false).
5488 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
5489 permitted: the operation may not write ``<new>`` even if the comparison
5492 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
5493 if the value loaded equals ``cmp``.
5495 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5496 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5497 load with an ordering parameter determined the second ordering parameter.
5502 .. code-block:: llvm
5505 %orig = atomic load i32* %ptr unordered ; yields i32
5509 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5510 %squared = mul i32 %cmp, %cmp
5511 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
5512 %value_loaded = extractvalue { i32, i1 } %val_success, 0
5513 %success = extractvalue { i32, i1 } %val_success, 1
5514 br i1 %success, label %done, label %loop
5521 '``atomicrmw``' Instruction
5522 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5529 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
5534 The '``atomicrmw``' instruction is used to atomically modify memory.
5539 There are three arguments to the '``atomicrmw``' instruction: an
5540 operation to apply, an address whose value to modify, an argument to the
5541 operation. The operation must be one of the following keywords:
5555 The type of '<value>' must be an integer type whose bit width is a power
5556 of two greater than or equal to eight and less than or equal to a
5557 target-specific size limit. The type of the '``<pointer>``' operand must
5558 be a pointer to that type. If the ``atomicrmw`` is marked as
5559 ``volatile``, then the optimizer is not allowed to modify the number or
5560 order of execution of this ``atomicrmw`` with other :ref:`volatile
5561 operations <volatile>`.
5566 The contents of memory at the location specified by the '``<pointer>``'
5567 operand are atomically read, modified, and written back. The original
5568 value at the location is returned. The modification is specified by the
5571 - xchg: ``*ptr = val``
5572 - add: ``*ptr = *ptr + val``
5573 - sub: ``*ptr = *ptr - val``
5574 - and: ``*ptr = *ptr & val``
5575 - nand: ``*ptr = ~(*ptr & val)``
5576 - or: ``*ptr = *ptr | val``
5577 - xor: ``*ptr = *ptr ^ val``
5578 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5579 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5580 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5582 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5588 .. code-block:: llvm
5590 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
5592 .. _i_getelementptr:
5594 '``getelementptr``' Instruction
5595 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5602 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5603 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5604 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5609 The '``getelementptr``' instruction is used to get the address of a
5610 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5611 address calculation only and does not access memory.
5616 The first argument is always a pointer or a vector of pointers, and
5617 forms the basis of the calculation. The remaining arguments are indices
5618 that indicate which of the elements of the aggregate object are indexed.
5619 The interpretation of each index is dependent on the type being indexed
5620 into. The first index always indexes the pointer value given as the
5621 first argument, the second index indexes a value of the type pointed to
5622 (not necessarily the value directly pointed to, since the first index
5623 can be non-zero), etc. The first type indexed into must be a pointer
5624 value, subsequent types can be arrays, vectors, and structs. Note that
5625 subsequent types being indexed into can never be pointers, since that
5626 would require loading the pointer before continuing calculation.
5628 The type of each index argument depends on the type it is indexing into.
5629 When indexing into a (optionally packed) structure, only ``i32`` integer
5630 **constants** are allowed (when using a vector of indices they must all
5631 be the **same** ``i32`` integer constant). When indexing into an array,
5632 pointer or vector, integers of any width are allowed, and they are not
5633 required to be constant. These integers are treated as signed values
5636 For example, let's consider a C code fragment and how it gets compiled
5652 int *foo(struct ST *s) {
5653 return &s[1].Z.B[5][13];
5656 The LLVM code generated by Clang is:
5658 .. code-block:: llvm
5660 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5661 %struct.ST = type { i32, double, %struct.RT }
5663 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5665 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5672 In the example above, the first index is indexing into the
5673 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5674 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5675 indexes into the third element of the structure, yielding a
5676 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5677 structure. The third index indexes into the second element of the
5678 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5679 dimensions of the array are subscripted into, yielding an '``i32``'
5680 type. The '``getelementptr``' instruction returns a pointer to this
5681 element, thus computing a value of '``i32*``' type.
5683 Note that it is perfectly legal to index partially through a structure,
5684 returning a pointer to an inner element. Because of this, the LLVM code
5685 for the given testcase is equivalent to:
5687 .. code-block:: llvm
5689 define i32* @foo(%struct.ST* %s) {
5690 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5691 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5692 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5693 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5694 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5698 If the ``inbounds`` keyword is present, the result value of the
5699 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5700 pointer is not an *in bounds* address of an allocated object, or if any
5701 of the addresses that would be formed by successive addition of the
5702 offsets implied by the indices to the base address with infinitely
5703 precise signed arithmetic are not an *in bounds* address of that
5704 allocated object. The *in bounds* addresses for an allocated object are
5705 all the addresses that point into the object, plus the address one byte
5706 past the end. In cases where the base is a vector of pointers the
5707 ``inbounds`` keyword applies to each of the computations element-wise.
5709 If the ``inbounds`` keyword is not present, the offsets are added to the
5710 base address with silently-wrapping two's complement arithmetic. If the
5711 offsets have a different width from the pointer, they are sign-extended
5712 or truncated to the width of the pointer. The result value of the
5713 ``getelementptr`` may be outside the object pointed to by the base
5714 pointer. The result value may not necessarily be used to access memory
5715 though, even if it happens to point into allocated storage. See the
5716 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5719 The getelementptr instruction is often confusing. For some more insight
5720 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5725 .. code-block:: llvm
5727 ; yields [12 x i8]*:aptr
5728 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5730 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5732 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5734 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5736 In cases where the pointer argument is a vector of pointers, each index
5737 must be a vector with the same number of elements. For example:
5739 .. code-block:: llvm
5741 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5743 Conversion Operations
5744 ---------------------
5746 The instructions in this category are the conversion instructions
5747 (casting) which all take a single operand and a type. They perform
5748 various bit conversions on the operand.
5750 '``trunc .. to``' Instruction
5751 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5758 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5763 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5768 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5769 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5770 of the same number of integers. The bit size of the ``value`` must be
5771 larger than the bit size of the destination type, ``ty2``. Equal sized
5772 types are not allowed.
5777 The '``trunc``' instruction truncates the high order bits in ``value``
5778 and converts the remaining bits to ``ty2``. Since the source size must
5779 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5780 It will always truncate bits.
5785 .. code-block:: llvm
5787 %X = trunc i32 257 to i8 ; yields i8:1
5788 %Y = trunc i32 123 to i1 ; yields i1:true
5789 %Z = trunc i32 122 to i1 ; yields i1:false
5790 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5792 '``zext .. to``' Instruction
5793 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5800 <result> = zext <ty> <value> to <ty2> ; yields ty2
5805 The '``zext``' instruction zero extends its operand to type ``ty2``.
5810 The '``zext``' instruction takes a value to cast, and a type to cast it
5811 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5812 the same number of integers. The bit size of the ``value`` must be
5813 smaller than the bit size of the destination type, ``ty2``.
5818 The ``zext`` fills the high order bits of the ``value`` with zero bits
5819 until it reaches the size of the destination type, ``ty2``.
5821 When zero extending from i1, the result will always be either 0 or 1.
5826 .. code-block:: llvm
5828 %X = zext i32 257 to i64 ; yields i64:257
5829 %Y = zext i1 true to i32 ; yields i32:1
5830 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5832 '``sext .. to``' Instruction
5833 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5840 <result> = sext <ty> <value> to <ty2> ; yields ty2
5845 The '``sext``' sign extends ``value`` to the type ``ty2``.
5850 The '``sext``' instruction takes a value to cast, and a type to cast it
5851 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5852 the same number of integers. The bit size of the ``value`` must be
5853 smaller than the bit size of the destination type, ``ty2``.
5858 The '``sext``' instruction performs a sign extension by copying the sign
5859 bit (highest order bit) of the ``value`` until it reaches the bit size
5860 of the type ``ty2``.
5862 When sign extending from i1, the extension always results in -1 or 0.
5867 .. code-block:: llvm
5869 %X = sext i8 -1 to i16 ; yields i16 :65535
5870 %Y = sext i1 true to i32 ; yields i32:-1
5871 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5873 '``fptrunc .. to``' Instruction
5874 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5881 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5886 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5891 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5892 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5893 The size of ``value`` must be larger than the size of ``ty2``. This
5894 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5899 The '``fptrunc``' instruction truncates a ``value`` from a larger
5900 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5901 point <t_floating>` type. If the value cannot fit within the
5902 destination type, ``ty2``, then the results are undefined.
5907 .. code-block:: llvm
5909 %X = fptrunc double 123.0 to float ; yields float:123.0
5910 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5912 '``fpext .. to``' Instruction
5913 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5920 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5925 The '``fpext``' extends a floating point ``value`` to a larger floating
5931 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5932 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5933 to. The source type must be smaller than the destination type.
5938 The '``fpext``' instruction extends the ``value`` from a smaller
5939 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5940 point <t_floating>` type. The ``fpext`` cannot be used to make a
5941 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5942 *no-op cast* for a floating point cast.
5947 .. code-block:: llvm
5949 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5950 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5952 '``fptoui .. to``' Instruction
5953 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5960 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5965 The '``fptoui``' converts a floating point ``value`` to its unsigned
5966 integer equivalent of type ``ty2``.
5971 The '``fptoui``' instruction takes a value to cast, which must be a
5972 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5973 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5974 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5975 type with the same number of elements as ``ty``
5980 The '``fptoui``' instruction converts its :ref:`floating
5981 point <t_floating>` operand into the nearest (rounding towards zero)
5982 unsigned integer value. If the value cannot fit in ``ty2``, the results
5988 .. code-block:: llvm
5990 %X = fptoui double 123.0 to i32 ; yields i32:123
5991 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5992 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5994 '``fptosi .. to``' Instruction
5995 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6002 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
6007 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
6008 ``value`` to type ``ty2``.
6013 The '``fptosi``' instruction takes a value to cast, which must be a
6014 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6015 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6016 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6017 type with the same number of elements as ``ty``
6022 The '``fptosi``' instruction converts its :ref:`floating
6023 point <t_floating>` operand into the nearest (rounding towards zero)
6024 signed integer value. If the value cannot fit in ``ty2``, the results
6030 .. code-block:: llvm
6032 %X = fptosi double -123.0 to i32 ; yields i32:-123
6033 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
6034 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
6036 '``uitofp .. to``' Instruction
6037 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6044 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
6049 The '``uitofp``' instruction regards ``value`` as an unsigned integer
6050 and converts that value to the ``ty2`` type.
6055 The '``uitofp``' instruction takes a value to cast, which must be a
6056 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6057 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6058 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6059 type with the same number of elements as ``ty``
6064 The '``uitofp``' instruction interprets its operand as an unsigned
6065 integer quantity and converts it to the corresponding floating point
6066 value. If the value cannot fit in the floating point value, the results
6072 .. code-block:: llvm
6074 %X = uitofp i32 257 to float ; yields float:257.0
6075 %Y = uitofp i8 -1 to double ; yields double:255.0
6077 '``sitofp .. to``' Instruction
6078 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6085 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
6090 The '``sitofp``' instruction regards ``value`` as a signed integer and
6091 converts that value to the ``ty2`` type.
6096 The '``sitofp``' instruction takes a value to cast, which must be a
6097 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6098 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6099 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6100 type with the same number of elements as ``ty``
6105 The '``sitofp``' instruction interprets its operand as a signed integer
6106 quantity and converts it to the corresponding floating point value. If
6107 the value cannot fit in the floating point value, the results are
6113 .. code-block:: llvm
6115 %X = sitofp i32 257 to float ; yields float:257.0
6116 %Y = sitofp i8 -1 to double ; yields double:-1.0
6120 '``ptrtoint .. to``' Instruction
6121 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6128 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
6133 The '``ptrtoint``' instruction converts the pointer or a vector of
6134 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
6139 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
6140 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
6141 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
6142 a vector of integers type.
6147 The '``ptrtoint``' instruction converts ``value`` to integer type
6148 ``ty2`` by interpreting the pointer value as an integer and either
6149 truncating or zero extending that value to the size of the integer type.
6150 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
6151 ``value`` is larger than ``ty2`` then a truncation is done. If they are
6152 the same size, then nothing is done (*no-op cast*) other than a type
6158 .. code-block:: llvm
6160 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
6161 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
6162 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
6166 '``inttoptr .. to``' Instruction
6167 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6174 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
6179 The '``inttoptr``' instruction converts an integer ``value`` to a
6180 pointer type, ``ty2``.
6185 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
6186 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
6192 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
6193 applying either a zero extension or a truncation depending on the size
6194 of the integer ``value``. If ``value`` is larger than the size of a
6195 pointer then a truncation is done. If ``value`` is smaller than the size
6196 of a pointer then a zero extension is done. If they are the same size,
6197 nothing is done (*no-op cast*).
6202 .. code-block:: llvm
6204 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
6205 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
6206 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
6207 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
6211 '``bitcast .. to``' Instruction
6212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6219 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
6224 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
6230 The '``bitcast``' instruction takes a value to cast, which must be a
6231 non-aggregate first class value, and a type to cast it to, which must
6232 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
6233 bit sizes of ``value`` and the destination type, ``ty2``, must be
6234 identical. If the source type is a pointer, the destination type must
6235 also be a pointer of the same size. This instruction supports bitwise
6236 conversion of vectors to integers and to vectors of other types (as
6237 long as they have the same size).
6242 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
6243 is always a *no-op cast* because no bits change with this
6244 conversion. The conversion is done as if the ``value`` had been stored
6245 to memory and read back as type ``ty2``. Pointer (or vector of
6246 pointers) types may only be converted to other pointer (or vector of
6247 pointers) types with the same address space through this instruction.
6248 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
6249 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
6254 .. code-block:: llvm
6256 %X = bitcast i8 255 to i8 ; yields i8 :-1
6257 %Y = bitcast i32* %x to sint* ; yields sint*:%x
6258 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
6259 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
6261 .. _i_addrspacecast:
6263 '``addrspacecast .. to``' Instruction
6264 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6271 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
6276 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
6277 address space ``n`` to type ``pty2`` in address space ``m``.
6282 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
6283 to cast and a pointer type to cast it to, which must have a different
6289 The '``addrspacecast``' instruction converts the pointer value
6290 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
6291 value modification, depending on the target and the address space
6292 pair. Pointer conversions within the same address space must be
6293 performed with the ``bitcast`` instruction. Note that if the address space
6294 conversion is legal then both result and operand refer to the same memory
6300 .. code-block:: llvm
6302 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
6303 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
6304 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
6311 The instructions in this category are the "miscellaneous" instructions,
6312 which defy better classification.
6316 '``icmp``' Instruction
6317 ^^^^^^^^^^^^^^^^^^^^^^
6324 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6329 The '``icmp``' instruction returns a boolean value or a vector of
6330 boolean values based on comparison of its two integer, integer vector,
6331 pointer, or pointer vector operands.
6336 The '``icmp``' instruction takes three operands. The first operand is
6337 the condition code indicating the kind of comparison to perform. It is
6338 not a value, just a keyword. The possible condition code are:
6341 #. ``ne``: not equal
6342 #. ``ugt``: unsigned greater than
6343 #. ``uge``: unsigned greater or equal
6344 #. ``ult``: unsigned less than
6345 #. ``ule``: unsigned less or equal
6346 #. ``sgt``: signed greater than
6347 #. ``sge``: signed greater or equal
6348 #. ``slt``: signed less than
6349 #. ``sle``: signed less or equal
6351 The remaining two arguments must be :ref:`integer <t_integer>` or
6352 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
6353 must also be identical types.
6358 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
6359 code given as ``cond``. The comparison performed always yields either an
6360 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
6362 #. ``eq``: yields ``true`` if the operands are equal, ``false``
6363 otherwise. No sign interpretation is necessary or performed.
6364 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
6365 otherwise. No sign interpretation is necessary or performed.
6366 #. ``ugt``: interprets the operands as unsigned values and yields
6367 ``true`` if ``op1`` is greater than ``op2``.
6368 #. ``uge``: interprets the operands as unsigned values and yields
6369 ``true`` if ``op1`` is greater than or equal to ``op2``.
6370 #. ``ult``: interprets the operands as unsigned values and yields
6371 ``true`` if ``op1`` is less than ``op2``.
6372 #. ``ule``: interprets the operands as unsigned values and yields
6373 ``true`` if ``op1`` is less than or equal to ``op2``.
6374 #. ``sgt``: interprets the operands as signed values and yields ``true``
6375 if ``op1`` is greater than ``op2``.
6376 #. ``sge``: interprets the operands as signed values and yields ``true``
6377 if ``op1`` is greater than or equal to ``op2``.
6378 #. ``slt``: interprets the operands as signed values and yields ``true``
6379 if ``op1`` is less than ``op2``.
6380 #. ``sle``: interprets the operands as signed values and yields ``true``
6381 if ``op1`` is less than or equal to ``op2``.
6383 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
6384 are compared as if they were integers.
6386 If the operands are integer vectors, then they are compared element by
6387 element. The result is an ``i1`` vector with the same number of elements
6388 as the values being compared. Otherwise, the result is an ``i1``.
6393 .. code-block:: llvm
6395 <result> = icmp eq i32 4, 5 ; yields: result=false
6396 <result> = icmp ne float* %X, %X ; yields: result=false
6397 <result> = icmp ult i16 4, 5 ; yields: result=true
6398 <result> = icmp sgt i16 4, 5 ; yields: result=false
6399 <result> = icmp ule i16 -4, 5 ; yields: result=false
6400 <result> = icmp sge i16 4, 5 ; yields: result=false
6402 Note that the code generator does not yet support vector types with the
6403 ``icmp`` instruction.
6407 '``fcmp``' Instruction
6408 ^^^^^^^^^^^^^^^^^^^^^^
6415 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6420 The '``fcmp``' instruction returns a boolean value or vector of boolean
6421 values based on comparison of its operands.
6423 If the operands are floating point scalars, then the result type is a
6424 boolean (:ref:`i1 <t_integer>`).
6426 If the operands are floating point vectors, then the result type is a
6427 vector of boolean with the same number of elements as the operands being
6433 The '``fcmp``' instruction takes three operands. The first operand is
6434 the condition code indicating the kind of comparison to perform. It is
6435 not a value, just a keyword. The possible condition code are:
6437 #. ``false``: no comparison, always returns false
6438 #. ``oeq``: ordered and equal
6439 #. ``ogt``: ordered and greater than
6440 #. ``oge``: ordered and greater than or equal
6441 #. ``olt``: ordered and less than
6442 #. ``ole``: ordered and less than or equal
6443 #. ``one``: ordered and not equal
6444 #. ``ord``: ordered (no nans)
6445 #. ``ueq``: unordered or equal
6446 #. ``ugt``: unordered or greater than
6447 #. ``uge``: unordered or greater than or equal
6448 #. ``ult``: unordered or less than
6449 #. ``ule``: unordered or less than or equal
6450 #. ``une``: unordered or not equal
6451 #. ``uno``: unordered (either nans)
6452 #. ``true``: no comparison, always returns true
6454 *Ordered* means that neither operand is a QNAN while *unordered* means
6455 that either operand may be a QNAN.
6457 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6458 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6459 type. They must have identical types.
6464 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6465 condition code given as ``cond``. If the operands are vectors, then the
6466 vectors are compared element by element. Each comparison performed
6467 always yields an :ref:`i1 <t_integer>` result, as follows:
6469 #. ``false``: always yields ``false``, regardless of operands.
6470 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6471 is equal to ``op2``.
6472 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6473 is greater than ``op2``.
6474 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6475 is greater than or equal to ``op2``.
6476 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6477 is less than ``op2``.
6478 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6479 is less than or equal to ``op2``.
6480 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6481 is not equal to ``op2``.
6482 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6483 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6485 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6486 greater than ``op2``.
6487 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6488 greater than or equal to ``op2``.
6489 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6491 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6492 less than or equal to ``op2``.
6493 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6494 not equal to ``op2``.
6495 #. ``uno``: yields ``true`` if either operand is a QNAN.
6496 #. ``true``: always yields ``true``, regardless of operands.
6501 .. code-block:: llvm
6503 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6504 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6505 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6506 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6508 Note that the code generator does not yet support vector types with the
6509 ``fcmp`` instruction.
6513 '``phi``' Instruction
6514 ^^^^^^^^^^^^^^^^^^^^^
6521 <result> = phi <ty> [ <val0>, <label0>], ...
6526 The '``phi``' instruction is used to implement the φ node in the SSA
6527 graph representing the function.
6532 The type of the incoming values is specified with the first type field.
6533 After this, the '``phi``' instruction takes a list of pairs as
6534 arguments, with one pair for each predecessor basic block of the current
6535 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6536 the value arguments to the PHI node. Only labels may be used as the
6539 There must be no non-phi instructions between the start of a basic block
6540 and the PHI instructions: i.e. PHI instructions must be first in a basic
6543 For the purposes of the SSA form, the use of each incoming value is
6544 deemed to occur on the edge from the corresponding predecessor block to
6545 the current block (but after any definition of an '``invoke``'
6546 instruction's return value on the same edge).
6551 At runtime, the '``phi``' instruction logically takes on the value
6552 specified by the pair corresponding to the predecessor basic block that
6553 executed just prior to the current block.
6558 .. code-block:: llvm
6560 Loop: ; Infinite loop that counts from 0 on up...
6561 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6562 %nextindvar = add i32 %indvar, 1
6567 '``select``' Instruction
6568 ^^^^^^^^^^^^^^^^^^^^^^^^
6575 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6577 selty is either i1 or {<N x i1>}
6582 The '``select``' instruction is used to choose one value based on a
6583 condition, without IR-level branching.
6588 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6589 values indicating the condition, and two values of the same :ref:`first
6590 class <t_firstclass>` type. If the val1/val2 are vectors and the
6591 condition is a scalar, then entire vectors are selected, not individual
6597 If the condition is an i1 and it evaluates to 1, the instruction returns
6598 the first value argument; otherwise, it returns the second value
6601 If the condition is a vector of i1, then the value arguments must be
6602 vectors of the same size, and the selection is done element by element.
6607 .. code-block:: llvm
6609 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6613 '``call``' Instruction
6614 ^^^^^^^^^^^^^^^^^^^^^^
6621 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6626 The '``call``' instruction represents a simple function call.
6631 This instruction requires several arguments:
6633 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
6634 should perform tail call optimization. The ``tail`` marker is a hint that
6635 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
6636 means that the call must be tail call optimized in order for the program to
6637 be correct. The ``musttail`` marker provides these guarantees:
6639 #. The call will not cause unbounded stack growth if it is part of a
6640 recursive cycle in the call graph.
6641 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
6644 Both markers imply that the callee does not access allocas or varargs from
6645 the caller. Calls marked ``musttail`` must obey the following additional
6648 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
6649 or a pointer bitcast followed by a ret instruction.
6650 - The ret instruction must return the (possibly bitcasted) value
6651 produced by the call or void.
6652 - The caller and callee prototypes must match. Pointer types of
6653 parameters or return types may differ in pointee type, but not
6655 - The calling conventions of the caller and callee must match.
6656 - All ABI-impacting function attributes, such as sret, byval, inreg,
6657 returned, and inalloca, must match.
6658 - The callee must be varargs iff the caller is varargs. Bitcasting a
6659 non-varargs function to the appropriate varargs type is legal so
6660 long as the non-varargs prefixes obey the other rules.
6662 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
6663 the following conditions are met:
6665 - Caller and callee both have the calling convention ``fastcc``.
6666 - The call is in tail position (ret immediately follows call and ret
6667 uses value of call or is void).
6668 - Option ``-tailcallopt`` is enabled, or
6669 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6670 - `Platform-specific constraints are
6671 met. <CodeGenerator.html#tailcallopt>`_
6673 #. The optional "cconv" marker indicates which :ref:`calling
6674 convention <callingconv>` the call should use. If none is
6675 specified, the call defaults to using C calling conventions. The
6676 calling convention of the call must match the calling convention of
6677 the target function, or else the behavior is undefined.
6678 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6679 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6681 #. '``ty``': the type of the call instruction itself which is also the
6682 type of the return value. Functions that return no value are marked
6684 #. '``fnty``': shall be the signature of the pointer to function value
6685 being invoked. The argument types must match the types implied by
6686 this signature. This type can be omitted if the function is not
6687 varargs and if the function type does not return a pointer to a
6689 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6690 be invoked. In most cases, this is a direct function invocation, but
6691 indirect ``call``'s are just as possible, calling an arbitrary pointer
6693 #. '``function args``': argument list whose types match the function
6694 signature argument types and parameter attributes. All arguments must
6695 be of :ref:`first class <t_firstclass>` type. If the function signature
6696 indicates the function accepts a variable number of arguments, the
6697 extra arguments can be specified.
6698 #. The optional :ref:`function attributes <fnattrs>` list. Only
6699 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6700 attributes are valid here.
6705 The '``call``' instruction is used to cause control flow to transfer to
6706 a specified function, with its incoming arguments bound to the specified
6707 values. Upon a '``ret``' instruction in the called function, control
6708 flow continues with the instruction after the function call, and the
6709 return value of the function is bound to the result argument.
6714 .. code-block:: llvm
6716 %retval = call i32 @test(i32 %argc)
6717 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6718 %X = tail call i32 @foo() ; yields i32
6719 %Y = tail call fastcc i32 @foo() ; yields i32
6720 call void %foo(i8 97 signext)
6722 %struct.A = type { i32, i8 }
6723 %r = call %struct.A @foo() ; yields { i32, i8 }
6724 %gr = extractvalue %struct.A %r, 0 ; yields i32
6725 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6726 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6727 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6729 llvm treats calls to some functions with names and arguments that match
6730 the standard C99 library as being the C99 library functions, and may
6731 perform optimizations or generate code for them under that assumption.
6732 This is something we'd like to change in the future to provide better
6733 support for freestanding environments and non-C-based languages.
6737 '``va_arg``' Instruction
6738 ^^^^^^^^^^^^^^^^^^^^^^^^
6745 <resultval> = va_arg <va_list*> <arglist>, <argty>
6750 The '``va_arg``' instruction is used to access arguments passed through
6751 the "variable argument" area of a function call. It is used to implement
6752 the ``va_arg`` macro in C.
6757 This instruction takes a ``va_list*`` value and the type of the
6758 argument. It returns a value of the specified argument type and
6759 increments the ``va_list`` to point to the next argument. The actual
6760 type of ``va_list`` is target specific.
6765 The '``va_arg``' instruction loads an argument of the specified type
6766 from the specified ``va_list`` and causes the ``va_list`` to point to
6767 the next argument. For more information, see the variable argument
6768 handling :ref:`Intrinsic Functions <int_varargs>`.
6770 It is legal for this instruction to be called in a function which does
6771 not take a variable number of arguments, for example, the ``vfprintf``
6774 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6775 function <intrinsics>` because it takes a type as an argument.
6780 See the :ref:`variable argument processing <int_varargs>` section.
6782 Note that the code generator does not yet fully support va\_arg on many
6783 targets. Also, it does not currently support va\_arg with aggregate
6784 types on any target.
6788 '``landingpad``' Instruction
6789 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6796 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6797 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6799 <clause> := catch <type> <value>
6800 <clause> := filter <array constant type> <array constant>
6805 The '``landingpad``' instruction is used by `LLVM's exception handling
6806 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6807 is a landing pad --- one where the exception lands, and corresponds to the
6808 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6809 defines values supplied by the personality function (``pers_fn``) upon
6810 re-entry to the function. The ``resultval`` has the type ``resultty``.
6815 This instruction takes a ``pers_fn`` value. This is the personality
6816 function associated with the unwinding mechanism. The optional
6817 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6819 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6820 contains the global variable representing the "type" that may be caught
6821 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6822 clause takes an array constant as its argument. Use
6823 "``[0 x i8**] undef``" for a filter which cannot throw. The
6824 '``landingpad``' instruction must contain *at least* one ``clause`` or
6825 the ``cleanup`` flag.
6830 The '``landingpad``' instruction defines the values which are set by the
6831 personality function (``pers_fn``) upon re-entry to the function, and
6832 therefore the "result type" of the ``landingpad`` instruction. As with
6833 calling conventions, how the personality function results are
6834 represented in LLVM IR is target specific.
6836 The clauses are applied in order from top to bottom. If two
6837 ``landingpad`` instructions are merged together through inlining, the
6838 clauses from the calling function are appended to the list of clauses.
6839 When the call stack is being unwound due to an exception being thrown,
6840 the exception is compared against each ``clause`` in turn. If it doesn't
6841 match any of the clauses, and the ``cleanup`` flag is not set, then
6842 unwinding continues further up the call stack.
6844 The ``landingpad`` instruction has several restrictions:
6846 - A landing pad block is a basic block which is the unwind destination
6847 of an '``invoke``' instruction.
6848 - A landing pad block must have a '``landingpad``' instruction as its
6849 first non-PHI instruction.
6850 - There can be only one '``landingpad``' instruction within the landing
6852 - A basic block that is not a landing pad block may not include a
6853 '``landingpad``' instruction.
6854 - All '``landingpad``' instructions in a function must have the same
6855 personality function.
6860 .. code-block:: llvm
6862 ;; A landing pad which can catch an integer.
6863 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6865 ;; A landing pad that is a cleanup.
6866 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6868 ;; A landing pad which can catch an integer and can only throw a double.
6869 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6871 filter [1 x i8**] [@_ZTId]
6878 LLVM supports the notion of an "intrinsic function". These functions
6879 have well known names and semantics and are required to follow certain
6880 restrictions. Overall, these intrinsics represent an extension mechanism
6881 for the LLVM language that does not require changing all of the
6882 transformations in LLVM when adding to the language (or the bitcode
6883 reader/writer, the parser, etc...).
6885 Intrinsic function names must all start with an "``llvm.``" prefix. This
6886 prefix is reserved in LLVM for intrinsic names; thus, function names may
6887 not begin with this prefix. Intrinsic functions must always be external
6888 functions: you cannot define the body of intrinsic functions. Intrinsic
6889 functions may only be used in call or invoke instructions: it is illegal
6890 to take the address of an intrinsic function. Additionally, because
6891 intrinsic functions are part of the LLVM language, it is required if any
6892 are added that they be documented here.
6894 Some intrinsic functions can be overloaded, i.e., the intrinsic
6895 represents a family of functions that perform the same operation but on
6896 different data types. Because LLVM can represent over 8 million
6897 different integer types, overloading is used commonly to allow an
6898 intrinsic function to operate on any integer type. One or more of the
6899 argument types or the result type can be overloaded to accept any
6900 integer type. Argument types may also be defined as exactly matching a
6901 previous argument's type or the result type. This allows an intrinsic
6902 function which accepts multiple arguments, but needs all of them to be
6903 of the same type, to only be overloaded with respect to a single
6904 argument or the result.
6906 Overloaded intrinsics will have the names of its overloaded argument
6907 types encoded into its function name, each preceded by a period. Only
6908 those types which are overloaded result in a name suffix. Arguments
6909 whose type is matched against another type do not. For example, the
6910 ``llvm.ctpop`` function can take an integer of any width and returns an
6911 integer of exactly the same integer width. This leads to a family of
6912 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6913 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6914 overloaded, and only one type suffix is required. Because the argument's
6915 type is matched against the return type, it does not require its own
6918 To learn how to add an intrinsic function, please see the `Extending
6919 LLVM Guide <ExtendingLLVM.html>`_.
6923 Variable Argument Handling Intrinsics
6924 -------------------------------------
6926 Variable argument support is defined in LLVM with the
6927 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6928 functions. These functions are related to the similarly named macros
6929 defined in the ``<stdarg.h>`` header file.
6931 All of these functions operate on arguments that use a target-specific
6932 value type "``va_list``". The LLVM assembly language reference manual
6933 does not define what this type is, so all transformations should be
6934 prepared to handle these functions regardless of the type used.
6936 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6937 variable argument handling intrinsic functions are used.
6939 .. code-block:: llvm
6941 ; This struct is different for every platform. For most platforms,
6942 ; it is merely an i8*.
6943 %struct.va_list = type { i8* }
6945 ; For Unix x86_64 platforms, va_list is the following struct:
6946 ; %struct.va_list = type { i32, i32, i8*, i8* }
6948 define i32 @test(i32 %X, ...) {
6949 ; Initialize variable argument processing
6950 %ap = alloca %struct.va_list
6951 %ap2 = bitcast %struct.va_list* %ap to i8*
6952 call void @llvm.va_start(i8* %ap2)
6954 ; Read a single integer argument
6955 %tmp = va_arg i8* %ap2, i32
6957 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6959 %aq2 = bitcast i8** %aq to i8*
6960 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6961 call void @llvm.va_end(i8* %aq2)
6963 ; Stop processing of arguments.
6964 call void @llvm.va_end(i8* %ap2)
6968 declare void @llvm.va_start(i8*)
6969 declare void @llvm.va_copy(i8*, i8*)
6970 declare void @llvm.va_end(i8*)
6974 '``llvm.va_start``' Intrinsic
6975 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6982 declare void @llvm.va_start(i8* <arglist>)
6987 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6988 subsequent use by ``va_arg``.
6993 The argument is a pointer to a ``va_list`` element to initialize.
6998 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6999 available in C. In a target-dependent way, it initializes the
7000 ``va_list`` element to which the argument points, so that the next call
7001 to ``va_arg`` will produce the first variable argument passed to the
7002 function. Unlike the C ``va_start`` macro, this intrinsic does not need
7003 to know the last argument of the function as the compiler can figure
7006 '``llvm.va_end``' Intrinsic
7007 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7014 declare void @llvm.va_end(i8* <arglist>)
7019 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
7020 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
7025 The argument is a pointer to a ``va_list`` to destroy.
7030 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
7031 available in C. In a target-dependent way, it destroys the ``va_list``
7032 element to which the argument points. Calls to
7033 :ref:`llvm.va_start <int_va_start>` and
7034 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
7039 '``llvm.va_copy``' Intrinsic
7040 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7047 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
7052 The '``llvm.va_copy``' intrinsic copies the current argument position
7053 from the source argument list to the destination argument list.
7058 The first argument is a pointer to a ``va_list`` element to initialize.
7059 The second argument is a pointer to a ``va_list`` element to copy from.
7064 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
7065 available in C. In a target-dependent way, it copies the source
7066 ``va_list`` element into the destination ``va_list`` element. This
7067 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
7068 arbitrarily complex and require, for example, memory allocation.
7070 Accurate Garbage Collection Intrinsics
7071 --------------------------------------
7073 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
7074 (GC) requires the implementation and generation of these intrinsics.
7075 These intrinsics allow identification of :ref:`GC roots on the
7076 stack <int_gcroot>`, as well as garbage collector implementations that
7077 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
7078 Front-ends for type-safe garbage collected languages should generate
7079 these intrinsics to make use of the LLVM garbage collectors. For more
7080 details, see `Accurate Garbage Collection with
7081 LLVM <GarbageCollection.html>`_.
7083 The garbage collection intrinsics only operate on objects in the generic
7084 address space (address space zero).
7088 '``llvm.gcroot``' Intrinsic
7089 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7096 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
7101 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
7102 the code generator, and allows some metadata to be associated with it.
7107 The first argument specifies the address of a stack object that contains
7108 the root pointer. The second pointer (which must be either a constant or
7109 a global value address) contains the meta-data to be associated with the
7115 At runtime, a call to this intrinsic stores a null pointer into the
7116 "ptrloc" location. At compile-time, the code generator generates
7117 information to allow the runtime to find the pointer at GC safe points.
7118 The '``llvm.gcroot``' intrinsic may only be used in a function which
7119 :ref:`specifies a GC algorithm <gc>`.
7123 '``llvm.gcread``' Intrinsic
7124 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7131 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
7136 The '``llvm.gcread``' intrinsic identifies reads of references from heap
7137 locations, allowing garbage collector implementations that require read
7143 The second argument is the address to read from, which should be an
7144 address allocated from the garbage collector. The first object is a
7145 pointer to the start of the referenced object, if needed by the language
7146 runtime (otherwise null).
7151 The '``llvm.gcread``' intrinsic has the same semantics as a load
7152 instruction, but may be replaced with substantially more complex code by
7153 the garbage collector runtime, as needed. The '``llvm.gcread``'
7154 intrinsic may only be used in a function which :ref:`specifies a GC
7159 '``llvm.gcwrite``' Intrinsic
7160 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7167 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
7172 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
7173 locations, allowing garbage collector implementations that require write
7174 barriers (such as generational or reference counting collectors).
7179 The first argument is the reference to store, the second is the start of
7180 the object to store it to, and the third is the address of the field of
7181 Obj to store to. If the runtime does not require a pointer to the
7182 object, Obj may be null.
7187 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
7188 instruction, but may be replaced with substantially more complex code by
7189 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
7190 intrinsic may only be used in a function which :ref:`specifies a GC
7193 Code Generator Intrinsics
7194 -------------------------
7196 These intrinsics are provided by LLVM to expose special features that
7197 may only be implemented with code generator support.
7199 '``llvm.returnaddress``' Intrinsic
7200 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7207 declare i8 *@llvm.returnaddress(i32 <level>)
7212 The '``llvm.returnaddress``' intrinsic attempts to compute a
7213 target-specific value indicating the return address of the current
7214 function or one of its callers.
7219 The argument to this intrinsic indicates which function to return the
7220 address for. Zero indicates the calling function, one indicates its
7221 caller, etc. The argument is **required** to be a constant integer
7227 The '``llvm.returnaddress``' intrinsic either returns a pointer
7228 indicating the return address of the specified call frame, or zero if it
7229 cannot be identified. The value returned by this intrinsic is likely to
7230 be incorrect or 0 for arguments other than zero, so it should only be
7231 used for debugging purposes.
7233 Note that calling this intrinsic does not prevent function inlining or
7234 other aggressive transformations, so the value returned may not be that
7235 of the obvious source-language caller.
7237 '``llvm.frameaddress``' Intrinsic
7238 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7245 declare i8* @llvm.frameaddress(i32 <level>)
7250 The '``llvm.frameaddress``' intrinsic attempts to return the
7251 target-specific frame pointer value for the specified stack frame.
7256 The argument to this intrinsic indicates which function to return the
7257 frame pointer for. Zero indicates the calling function, one indicates
7258 its caller, etc. The argument is **required** to be a constant integer
7264 The '``llvm.frameaddress``' intrinsic either returns a pointer
7265 indicating the frame address of the specified call frame, or zero if it
7266 cannot be identified. The value returned by this intrinsic is likely to
7267 be incorrect or 0 for arguments other than zero, so it should only be
7268 used for debugging purposes.
7270 Note that calling this intrinsic does not prevent function inlining or
7271 other aggressive transformations, so the value returned may not be that
7272 of the obvious source-language caller.
7274 .. _int_read_register:
7275 .. _int_write_register:
7277 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
7278 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7285 declare i32 @llvm.read_register.i32(metadata)
7286 declare i64 @llvm.read_register.i64(metadata)
7287 declare void @llvm.write_register.i32(metadata, i32 @value)
7288 declare void @llvm.write_register.i64(metadata, i64 @value)
7294 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
7295 provides access to the named register. The register must be valid on
7296 the architecture being compiled to. The type needs to be compatible
7297 with the register being read.
7302 The '``llvm.read_register``' intrinsic returns the current value of the
7303 register, where possible. The '``llvm.write_register``' intrinsic sets
7304 the current value of the register, where possible.
7306 This is useful to implement named register global variables that need
7307 to always be mapped to a specific register, as is common practice on
7308 bare-metal programs including OS kernels.
7310 The compiler doesn't check for register availability or use of the used
7311 register in surrounding code, including inline assembly. Because of that,
7312 allocatable registers are not supported.
7314 Warning: So far it only works with the stack pointer on selected
7315 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
7316 work is needed to support other registers and even more so, allocatable
7321 '``llvm.stacksave``' Intrinsic
7322 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7329 declare i8* @llvm.stacksave()
7334 The '``llvm.stacksave``' intrinsic is used to remember the current state
7335 of the function stack, for use with
7336 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
7337 implementing language features like scoped automatic variable sized
7343 This intrinsic returns a opaque pointer value that can be passed to
7344 :ref:`llvm.stackrestore <int_stackrestore>`. When an
7345 ``llvm.stackrestore`` intrinsic is executed with a value saved from
7346 ``llvm.stacksave``, it effectively restores the state of the stack to
7347 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
7348 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
7349 were allocated after the ``llvm.stacksave`` was executed.
7351 .. _int_stackrestore:
7353 '``llvm.stackrestore``' Intrinsic
7354 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7361 declare void @llvm.stackrestore(i8* %ptr)
7366 The '``llvm.stackrestore``' intrinsic is used to restore the state of
7367 the function stack to the state it was in when the corresponding
7368 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
7369 useful for implementing language features like scoped automatic variable
7370 sized arrays in C99.
7375 See the description for :ref:`llvm.stacksave <int_stacksave>`.
7377 '``llvm.prefetch``' Intrinsic
7378 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7385 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7390 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
7391 insert a prefetch instruction if supported; otherwise, it is a noop.
7392 Prefetches have no effect on the behavior of the program but can change
7393 its performance characteristics.
7398 ``address`` is the address to be prefetched, ``rw`` is the specifier
7399 determining if the fetch should be for a read (0) or write (1), and
7400 ``locality`` is a temporal locality specifier ranging from (0) - no
7401 locality, to (3) - extremely local keep in cache. The ``cache type``
7402 specifies whether the prefetch is performed on the data (1) or
7403 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
7404 arguments must be constant integers.
7409 This intrinsic does not modify the behavior of the program. In
7410 particular, prefetches cannot trap and do not produce a value. On
7411 targets that support this intrinsic, the prefetch can provide hints to
7412 the processor cache for better performance.
7414 '``llvm.pcmarker``' Intrinsic
7415 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7422 declare void @llvm.pcmarker(i32 <id>)
7427 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7428 Counter (PC) in a region of code to simulators and other tools. The
7429 method is target specific, but it is expected that the marker will use
7430 exported symbols to transmit the PC of the marker. The marker makes no
7431 guarantees that it will remain with any specific instruction after
7432 optimizations. It is possible that the presence of a marker will inhibit
7433 optimizations. The intended use is to be inserted after optimizations to
7434 allow correlations of simulation runs.
7439 ``id`` is a numerical id identifying the marker.
7444 This intrinsic does not modify the behavior of the program. Backends
7445 that do not support this intrinsic may ignore it.
7447 '``llvm.readcyclecounter``' Intrinsic
7448 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7455 declare i64 @llvm.readcyclecounter()
7460 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
7461 counter register (or similar low latency, high accuracy clocks) on those
7462 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7463 should map to RPCC. As the backing counters overflow quickly (on the
7464 order of 9 seconds on alpha), this should only be used for small
7470 When directly supported, reading the cycle counter should not modify any
7471 memory. Implementations are allowed to either return a application
7472 specific value or a system wide value. On backends without support, this
7473 is lowered to a constant 0.
7475 Note that runtime support may be conditional on the privilege-level code is
7476 running at and the host platform.
7478 '``llvm.clear_cache``' Intrinsic
7479 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7486 declare void @llvm.clear_cache(i8*, i8*)
7491 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
7492 in the specified range to the execution unit of the processor. On
7493 targets with non-unified instruction and data cache, the implementation
7494 flushes the instruction cache.
7499 On platforms with coherent instruction and data caches (e.g. x86), this
7500 intrinsic is a nop. On platforms with non-coherent instruction and data
7501 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
7502 instructions or a system call, if cache flushing requires special
7505 The default behavior is to emit a call to ``__clear_cache`` from the run
7508 This instrinsic does *not* empty the instruction pipeline. Modifications
7509 of the current function are outside the scope of the intrinsic.
7511 '``llvm.instrprof_increment``' Intrinsic
7512 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7519 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
7520 i32 <num-counters>, i32 <index>)
7525 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
7526 frontend for use with instrumentation based profiling. These will be
7527 lowered by the ``-instrprof`` pass to generate execution counts of a
7533 The first argument is a pointer to a global variable containing the
7534 name of the entity being instrumented. This should generally be the
7535 (mangled) function name for a set of counters.
7537 The second argument is a hash value that can be used by the consumer
7538 of the profile data to detect changes to the instrumented source, and
7539 the third is the number of counters associated with ``name``. It is an
7540 error if ``hash`` or ``num-counters`` differ between two instances of
7541 ``instrprof_increment`` that refer to the same name.
7543 The last argument refers to which of the counters for ``name`` should
7544 be incremented. It should be a value between 0 and ``num-counters``.
7549 This intrinsic represents an increment of a profiling counter. It will
7550 cause the ``-instrprof`` pass to generate the appropriate data
7551 structures and the code to increment the appropriate value, in a
7552 format that can be written out by a compiler runtime and consumed via
7553 the ``llvm-profdata`` tool.
7555 Standard C Library Intrinsics
7556 -----------------------------
7558 LLVM provides intrinsics for a few important standard C library
7559 functions. These intrinsics allow source-language front-ends to pass
7560 information about the alignment of the pointer arguments to the code
7561 generator, providing opportunity for more efficient code generation.
7565 '``llvm.memcpy``' Intrinsic
7566 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7571 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
7572 integer bit width and for different address spaces. Not all targets
7573 support all bit widths however.
7577 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7578 i32 <len>, i32 <align>, i1 <isvolatile>)
7579 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7580 i64 <len>, i32 <align>, i1 <isvolatile>)
7585 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7586 source location to the destination location.
7588 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
7589 intrinsics do not return a value, takes extra alignment/isvolatile
7590 arguments and the pointers can be in specified address spaces.
7595 The first argument is a pointer to the destination, the second is a
7596 pointer to the source. The third argument is an integer argument
7597 specifying the number of bytes to copy, the fourth argument is the
7598 alignment of the source and destination locations, and the fifth is a
7599 boolean indicating a volatile access.
7601 If the call to this intrinsic has an alignment value that is not 0 or 1,
7602 then the caller guarantees that both the source and destination pointers
7603 are aligned to that boundary.
7605 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
7606 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7607 very cleanly specified and it is unwise to depend on it.
7612 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7613 source location to the destination location, which are not allowed to
7614 overlap. It copies "len" bytes of memory over. If the argument is known
7615 to be aligned to some boundary, this can be specified as the fourth
7616 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
7618 '``llvm.memmove``' Intrinsic
7619 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7624 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7625 bit width and for different address space. Not all targets support all
7630 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7631 i32 <len>, i32 <align>, i1 <isvolatile>)
7632 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7633 i64 <len>, i32 <align>, i1 <isvolatile>)
7638 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7639 source location to the destination location. It is similar to the
7640 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7643 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7644 intrinsics do not return a value, takes extra alignment/isvolatile
7645 arguments and the pointers can be in specified address spaces.
7650 The first argument is a pointer to the destination, the second is a
7651 pointer to the source. The third argument is an integer argument
7652 specifying the number of bytes to copy, the fourth argument is the
7653 alignment of the source and destination locations, and the fifth is a
7654 boolean indicating a volatile access.
7656 If the call to this intrinsic has an alignment value that is not 0 or 1,
7657 then the caller guarantees that the source and destination pointers are
7658 aligned to that boundary.
7660 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7661 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7662 not very cleanly specified and it is unwise to depend on it.
7667 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7668 source location to the destination location, which may overlap. It
7669 copies "len" bytes of memory over. If the argument is known to be
7670 aligned to some boundary, this can be specified as the fourth argument,
7671 otherwise it should be set to 0 or 1 (both meaning no alignment).
7673 '``llvm.memset.*``' Intrinsics
7674 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7679 This is an overloaded intrinsic. You can use llvm.memset on any integer
7680 bit width and for different address spaces. However, not all targets
7681 support all bit widths.
7685 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7686 i32 <len>, i32 <align>, i1 <isvolatile>)
7687 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7688 i64 <len>, i32 <align>, i1 <isvolatile>)
7693 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7694 particular byte value.
7696 Note that, unlike the standard libc function, the ``llvm.memset``
7697 intrinsic does not return a value and takes extra alignment/volatile
7698 arguments. Also, the destination can be in an arbitrary address space.
7703 The first argument is a pointer to the destination to fill, the second
7704 is the byte value with which to fill it, the third argument is an
7705 integer argument specifying the number of bytes to fill, and the fourth
7706 argument is the known alignment of the destination location.
7708 If the call to this intrinsic has an alignment value that is not 0 or 1,
7709 then the caller guarantees that the destination pointer is aligned to
7712 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7713 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7714 very cleanly specified and it is unwise to depend on it.
7719 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7720 at the destination location. If the argument is known to be aligned to
7721 some boundary, this can be specified as the fourth argument, otherwise
7722 it should be set to 0 or 1 (both meaning no alignment).
7724 '``llvm.sqrt.*``' Intrinsic
7725 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7730 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7731 floating point or vector of floating point type. Not all targets support
7736 declare float @llvm.sqrt.f32(float %Val)
7737 declare double @llvm.sqrt.f64(double %Val)
7738 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7739 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7740 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7745 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7746 returning the same value as the libm '``sqrt``' functions would. Unlike
7747 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7748 negative numbers other than -0.0 (which allows for better optimization,
7749 because there is no need to worry about errno being set).
7750 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7755 The argument and return value are floating point numbers of the same
7761 This function returns the sqrt of the specified operand if it is a
7762 nonnegative floating point number.
7764 '``llvm.powi.*``' Intrinsic
7765 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7770 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7771 floating point or vector of floating point type. Not all targets support
7776 declare float @llvm.powi.f32(float %Val, i32 %power)
7777 declare double @llvm.powi.f64(double %Val, i32 %power)
7778 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7779 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7780 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7785 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7786 specified (positive or negative) power. The order of evaluation of
7787 multiplications is not defined. When a vector of floating point type is
7788 used, the second argument remains a scalar integer value.
7793 The second argument is an integer power, and the first is a value to
7794 raise to that power.
7799 This function returns the first value raised to the second power with an
7800 unspecified sequence of rounding operations.
7802 '``llvm.sin.*``' Intrinsic
7803 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7808 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7809 floating point or vector of floating point type. Not all targets support
7814 declare float @llvm.sin.f32(float %Val)
7815 declare double @llvm.sin.f64(double %Val)
7816 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7817 declare fp128 @llvm.sin.f128(fp128 %Val)
7818 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7823 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7828 The argument and return value are floating point numbers of the same
7834 This function returns the sine of the specified operand, returning the
7835 same values as the libm ``sin`` functions would, and handles error
7836 conditions in the same way.
7838 '``llvm.cos.*``' Intrinsic
7839 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7844 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7845 floating point or vector of floating point type. Not all targets support
7850 declare float @llvm.cos.f32(float %Val)
7851 declare double @llvm.cos.f64(double %Val)
7852 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7853 declare fp128 @llvm.cos.f128(fp128 %Val)
7854 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7859 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7864 The argument and return value are floating point numbers of the same
7870 This function returns the cosine of the specified operand, returning the
7871 same values as the libm ``cos`` functions would, and handles error
7872 conditions in the same way.
7874 '``llvm.pow.*``' Intrinsic
7875 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7880 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7881 floating point or vector of floating point type. Not all targets support
7886 declare float @llvm.pow.f32(float %Val, float %Power)
7887 declare double @llvm.pow.f64(double %Val, double %Power)
7888 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7889 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7890 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7895 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7896 specified (positive or negative) power.
7901 The second argument is a floating point power, and the first is a value
7902 to raise to that power.
7907 This function returns the first value raised to the second power,
7908 returning the same values as the libm ``pow`` functions would, and
7909 handles error conditions in the same way.
7911 '``llvm.exp.*``' Intrinsic
7912 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7917 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7918 floating point or vector of floating point type. Not all targets support
7923 declare float @llvm.exp.f32(float %Val)
7924 declare double @llvm.exp.f64(double %Val)
7925 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7926 declare fp128 @llvm.exp.f128(fp128 %Val)
7927 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7932 The '``llvm.exp.*``' intrinsics perform the exp function.
7937 The argument and return value are floating point numbers of the same
7943 This function returns the same values as the libm ``exp`` functions
7944 would, and handles error conditions in the same way.
7946 '``llvm.exp2.*``' Intrinsic
7947 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7952 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7953 floating point or vector of floating point type. Not all targets support
7958 declare float @llvm.exp2.f32(float %Val)
7959 declare double @llvm.exp2.f64(double %Val)
7960 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7961 declare fp128 @llvm.exp2.f128(fp128 %Val)
7962 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7967 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7972 The argument and return value are floating point numbers of the same
7978 This function returns the same values as the libm ``exp2`` functions
7979 would, and handles error conditions in the same way.
7981 '``llvm.log.*``' Intrinsic
7982 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7987 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7988 floating point or vector of floating point type. Not all targets support
7993 declare float @llvm.log.f32(float %Val)
7994 declare double @llvm.log.f64(double %Val)
7995 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7996 declare fp128 @llvm.log.f128(fp128 %Val)
7997 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
8002 The '``llvm.log.*``' intrinsics perform the log function.
8007 The argument and return value are floating point numbers of the same
8013 This function returns the same values as the libm ``log`` functions
8014 would, and handles error conditions in the same way.
8016 '``llvm.log10.*``' Intrinsic
8017 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8022 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
8023 floating point or vector of floating point type. Not all targets support
8028 declare float @llvm.log10.f32(float %Val)
8029 declare double @llvm.log10.f64(double %Val)
8030 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
8031 declare fp128 @llvm.log10.f128(fp128 %Val)
8032 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
8037 The '``llvm.log10.*``' intrinsics perform the log10 function.
8042 The argument and return value are floating point numbers of the same
8048 This function returns the same values as the libm ``log10`` functions
8049 would, and handles error conditions in the same way.
8051 '``llvm.log2.*``' Intrinsic
8052 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8057 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
8058 floating point or vector of floating point type. Not all targets support
8063 declare float @llvm.log2.f32(float %Val)
8064 declare double @llvm.log2.f64(double %Val)
8065 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
8066 declare fp128 @llvm.log2.f128(fp128 %Val)
8067 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
8072 The '``llvm.log2.*``' intrinsics perform the log2 function.
8077 The argument and return value are floating point numbers of the same
8083 This function returns the same values as the libm ``log2`` functions
8084 would, and handles error conditions in the same way.
8086 '``llvm.fma.*``' Intrinsic
8087 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8092 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
8093 floating point or vector of floating point type. Not all targets support
8098 declare float @llvm.fma.f32(float %a, float %b, float %c)
8099 declare double @llvm.fma.f64(double %a, double %b, double %c)
8100 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
8101 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
8102 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
8107 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
8113 The argument and return value are floating point numbers of the same
8119 This function returns the same values as the libm ``fma`` functions
8120 would, and does not set errno.
8122 '``llvm.fabs.*``' Intrinsic
8123 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8128 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
8129 floating point or vector of floating point type. Not all targets support
8134 declare float @llvm.fabs.f32(float %Val)
8135 declare double @llvm.fabs.f64(double %Val)
8136 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
8137 declare fp128 @llvm.fabs.f128(fp128 %Val)
8138 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
8143 The '``llvm.fabs.*``' intrinsics return the absolute value of the
8149 The argument and return value are floating point numbers of the same
8155 This function returns the same values as the libm ``fabs`` functions
8156 would, and handles error conditions in the same way.
8158 '``llvm.minnum.*``' Intrinsic
8159 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8164 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
8165 floating point or vector of floating point type. Not all targets support
8170 declare float @llvm.minnum.f32(float %Val0, float %Val1)
8171 declare double @llvm.minnum.f64(double %Val0, double %Val1)
8172 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8173 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
8174 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8179 The '``llvm.minnum.*``' intrinsics return the minimum of the two
8186 The arguments and return value are floating point numbers of the same
8192 Follows the IEEE-754 semantics for minNum, which also match for libm's
8195 If either operand is a NaN, returns the other non-NaN operand. Returns
8196 NaN only if both operands are NaN. If the operands compare equal,
8197 returns a value that compares equal to both operands. This means that
8198 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8200 '``llvm.maxnum.*``' Intrinsic
8201 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8206 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
8207 floating point or vector of floating point type. Not all targets support
8212 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
8213 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
8214 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8215 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
8216 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8221 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
8228 The arguments and return value are floating point numbers of the same
8233 Follows the IEEE-754 semantics for maxNum, which also match for libm's
8236 If either operand is a NaN, returns the other non-NaN operand. Returns
8237 NaN only if both operands are NaN. If the operands compare equal,
8238 returns a value that compares equal to both operands. This means that
8239 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8241 '``llvm.copysign.*``' Intrinsic
8242 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8247 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
8248 floating point or vector of floating point type. Not all targets support
8253 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
8254 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
8255 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
8256 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
8257 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
8262 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
8263 first operand and the sign of the second operand.
8268 The arguments and return value are floating point numbers of the same
8274 This function returns the same values as the libm ``copysign``
8275 functions would, and handles error conditions in the same way.
8277 '``llvm.floor.*``' Intrinsic
8278 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8283 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
8284 floating point or vector of floating point type. Not all targets support
8289 declare float @llvm.floor.f32(float %Val)
8290 declare double @llvm.floor.f64(double %Val)
8291 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
8292 declare fp128 @llvm.floor.f128(fp128 %Val)
8293 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
8298 The '``llvm.floor.*``' intrinsics return the floor of the operand.
8303 The argument and return value are floating point numbers of the same
8309 This function returns the same values as the libm ``floor`` functions
8310 would, and handles error conditions in the same way.
8312 '``llvm.ceil.*``' Intrinsic
8313 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8318 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
8319 floating point or vector of floating point type. Not all targets support
8324 declare float @llvm.ceil.f32(float %Val)
8325 declare double @llvm.ceil.f64(double %Val)
8326 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
8327 declare fp128 @llvm.ceil.f128(fp128 %Val)
8328 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
8333 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
8338 The argument and return value are floating point numbers of the same
8344 This function returns the same values as the libm ``ceil`` functions
8345 would, and handles error conditions in the same way.
8347 '``llvm.trunc.*``' Intrinsic
8348 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8353 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
8354 floating point or vector of floating point type. Not all targets support
8359 declare float @llvm.trunc.f32(float %Val)
8360 declare double @llvm.trunc.f64(double %Val)
8361 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
8362 declare fp128 @llvm.trunc.f128(fp128 %Val)
8363 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
8368 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
8369 nearest integer not larger in magnitude than the operand.
8374 The argument and return value are floating point numbers of the same
8380 This function returns the same values as the libm ``trunc`` functions
8381 would, and handles error conditions in the same way.
8383 '``llvm.rint.*``' Intrinsic
8384 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8389 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
8390 floating point or vector of floating point type. Not all targets support
8395 declare float @llvm.rint.f32(float %Val)
8396 declare double @llvm.rint.f64(double %Val)
8397 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
8398 declare fp128 @llvm.rint.f128(fp128 %Val)
8399 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
8404 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
8405 nearest integer. It may raise an inexact floating-point exception if the
8406 operand isn't an integer.
8411 The argument and return value are floating point numbers of the same
8417 This function returns the same values as the libm ``rint`` functions
8418 would, and handles error conditions in the same way.
8420 '``llvm.nearbyint.*``' Intrinsic
8421 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8426 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
8427 floating point or vector of floating point type. Not all targets support
8432 declare float @llvm.nearbyint.f32(float %Val)
8433 declare double @llvm.nearbyint.f64(double %Val)
8434 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
8435 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
8436 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
8441 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
8447 The argument and return value are floating point numbers of the same
8453 This function returns the same values as the libm ``nearbyint``
8454 functions would, and handles error conditions in the same way.
8456 '``llvm.round.*``' Intrinsic
8457 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8462 This is an overloaded intrinsic. You can use ``llvm.round`` on any
8463 floating point or vector of floating point type. Not all targets support
8468 declare float @llvm.round.f32(float %Val)
8469 declare double @llvm.round.f64(double %Val)
8470 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
8471 declare fp128 @llvm.round.f128(fp128 %Val)
8472 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
8477 The '``llvm.round.*``' intrinsics returns the operand rounded to the
8483 The argument and return value are floating point numbers of the same
8489 This function returns the same values as the libm ``round``
8490 functions would, and handles error conditions in the same way.
8492 Bit Manipulation Intrinsics
8493 ---------------------------
8495 LLVM provides intrinsics for a few important bit manipulation
8496 operations. These allow efficient code generation for some algorithms.
8498 '``llvm.bswap.*``' Intrinsics
8499 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8504 This is an overloaded intrinsic function. You can use bswap on any
8505 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
8509 declare i16 @llvm.bswap.i16(i16 <id>)
8510 declare i32 @llvm.bswap.i32(i32 <id>)
8511 declare i64 @llvm.bswap.i64(i64 <id>)
8516 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
8517 values with an even number of bytes (positive multiple of 16 bits).
8518 These are useful for performing operations on data that is not in the
8519 target's native byte order.
8524 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
8525 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
8526 intrinsic returns an i32 value that has the four bytes of the input i32
8527 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
8528 returned i32 will have its bytes in 3, 2, 1, 0 order. The
8529 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
8530 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
8533 '``llvm.ctpop.*``' Intrinsic
8534 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8539 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
8540 bit width, or on any vector with integer elements. Not all targets
8541 support all bit widths or vector types, however.
8545 declare i8 @llvm.ctpop.i8(i8 <src>)
8546 declare i16 @llvm.ctpop.i16(i16 <src>)
8547 declare i32 @llvm.ctpop.i32(i32 <src>)
8548 declare i64 @llvm.ctpop.i64(i64 <src>)
8549 declare i256 @llvm.ctpop.i256(i256 <src>)
8550 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
8555 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
8561 The only argument is the value to be counted. The argument may be of any
8562 integer type, or a vector with integer elements. The return type must
8563 match the argument type.
8568 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
8569 each element of a vector.
8571 '``llvm.ctlz.*``' Intrinsic
8572 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8577 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
8578 integer bit width, or any vector whose elements are integers. Not all
8579 targets support all bit widths or vector types, however.
8583 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
8584 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
8585 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
8586 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
8587 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
8588 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8593 The '``llvm.ctlz``' family of intrinsic functions counts the number of
8594 leading zeros in a variable.
8599 The first argument is the value to be counted. This argument may be of
8600 any integer type, or a vector with integer element type. The return
8601 type must match the first argument type.
8603 The second argument must be a constant and is a flag to indicate whether
8604 the intrinsic should ensure that a zero as the first argument produces a
8605 defined result. Historically some architectures did not provide a
8606 defined result for zero values as efficiently, and many algorithms are
8607 now predicated on avoiding zero-value inputs.
8612 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
8613 zeros in a variable, or within each element of the vector. If
8614 ``src == 0`` then the result is the size in bits of the type of ``src``
8615 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8616 ``llvm.ctlz(i32 2) = 30``.
8618 '``llvm.cttz.*``' Intrinsic
8619 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8624 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
8625 integer bit width, or any vector of integer elements. Not all targets
8626 support all bit widths or vector types, however.
8630 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
8631 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
8632 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
8633 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
8634 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
8635 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8640 The '``llvm.cttz``' family of intrinsic functions counts the number of
8646 The first argument is the value to be counted. This argument may be of
8647 any integer type, or a vector with integer element type. The return
8648 type must match the first argument type.
8650 The second argument must be a constant and is a flag to indicate whether
8651 the intrinsic should ensure that a zero as the first argument produces a
8652 defined result. Historically some architectures did not provide a
8653 defined result for zero values as efficiently, and many algorithms are
8654 now predicated on avoiding zero-value inputs.
8659 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
8660 zeros in a variable, or within each element of a vector. If ``src == 0``
8661 then the result is the size in bits of the type of ``src`` if
8662 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8663 ``llvm.cttz(2) = 1``.
8665 Arithmetic with Overflow Intrinsics
8666 -----------------------------------
8668 LLVM provides intrinsics for some arithmetic with overflow operations.
8670 '``llvm.sadd.with.overflow.*``' Intrinsics
8671 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8676 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
8677 on any integer bit width.
8681 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8682 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8683 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8688 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8689 a signed addition of the two arguments, and indicate whether an overflow
8690 occurred during the signed summation.
8695 The arguments (%a and %b) and the first element of the result structure
8696 may be of integer types of any bit width, but they must have the same
8697 bit width. The second element of the result structure must be of type
8698 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8704 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8705 a signed addition of the two variables. They return a structure --- the
8706 first element of which is the signed summation, and the second element
8707 of which is a bit specifying if the signed summation resulted in an
8713 .. code-block:: llvm
8715 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8716 %sum = extractvalue {i32, i1} %res, 0
8717 %obit = extractvalue {i32, i1} %res, 1
8718 br i1 %obit, label %overflow, label %normal
8720 '``llvm.uadd.with.overflow.*``' Intrinsics
8721 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8726 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8727 on any integer bit width.
8731 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8732 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8733 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8738 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8739 an unsigned addition of the two arguments, and indicate whether a carry
8740 occurred during the unsigned summation.
8745 The arguments (%a and %b) and the first element of the result structure
8746 may be of integer types of any bit width, but they must have the same
8747 bit width. The second element of the result structure must be of type
8748 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8754 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8755 an unsigned addition of the two arguments. They return a structure --- the
8756 first element of which is the sum, and the second element of which is a
8757 bit specifying if the unsigned summation resulted in a carry.
8762 .. code-block:: llvm
8764 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8765 %sum = extractvalue {i32, i1} %res, 0
8766 %obit = extractvalue {i32, i1} %res, 1
8767 br i1 %obit, label %carry, label %normal
8769 '``llvm.ssub.with.overflow.*``' Intrinsics
8770 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8775 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8776 on any integer bit width.
8780 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8781 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8782 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8787 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8788 a signed subtraction of the two arguments, and indicate whether an
8789 overflow occurred during the signed subtraction.
8794 The arguments (%a and %b) and the first element of the result structure
8795 may be of integer types of any bit width, but they must have the same
8796 bit width. The second element of the result structure must be of type
8797 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8803 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8804 a signed subtraction of the two arguments. They return a structure --- the
8805 first element of which is the subtraction, and the second element of
8806 which is a bit specifying if the signed subtraction resulted in an
8812 .. code-block:: llvm
8814 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8815 %sum = extractvalue {i32, i1} %res, 0
8816 %obit = extractvalue {i32, i1} %res, 1
8817 br i1 %obit, label %overflow, label %normal
8819 '``llvm.usub.with.overflow.*``' Intrinsics
8820 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8825 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8826 on any integer bit width.
8830 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8831 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8832 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8837 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8838 an unsigned subtraction of the two arguments, and indicate whether an
8839 overflow occurred during the unsigned subtraction.
8844 The arguments (%a and %b) and the first element of the result structure
8845 may be of integer types of any bit width, but they must have the same
8846 bit width. The second element of the result structure must be of type
8847 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8853 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8854 an unsigned subtraction of the two arguments. They return a structure ---
8855 the first element of which is the subtraction, and the second element of
8856 which is a bit specifying if the unsigned subtraction resulted in an
8862 .. code-block:: llvm
8864 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8865 %sum = extractvalue {i32, i1} %res, 0
8866 %obit = extractvalue {i32, i1} %res, 1
8867 br i1 %obit, label %overflow, label %normal
8869 '``llvm.smul.with.overflow.*``' Intrinsics
8870 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8875 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8876 on any integer bit width.
8880 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8881 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8882 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8887 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8888 a signed multiplication of the two arguments, and indicate whether an
8889 overflow occurred during the signed multiplication.
8894 The arguments (%a and %b) and the first element of the result structure
8895 may be of integer types of any bit width, but they must have the same
8896 bit width. The second element of the result structure must be of type
8897 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8903 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8904 a signed multiplication of the two arguments. They return a structure ---
8905 the first element of which is the multiplication, and the second element
8906 of which is a bit specifying if the signed multiplication resulted in an
8912 .. code-block:: llvm
8914 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8915 %sum = extractvalue {i32, i1} %res, 0
8916 %obit = extractvalue {i32, i1} %res, 1
8917 br i1 %obit, label %overflow, label %normal
8919 '``llvm.umul.with.overflow.*``' Intrinsics
8920 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8925 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8926 on any integer bit width.
8930 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8931 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8932 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8937 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8938 a unsigned multiplication of the two arguments, and indicate whether an
8939 overflow occurred during the unsigned multiplication.
8944 The arguments (%a and %b) and the first element of the result structure
8945 may be of integer types of any bit width, but they must have the same
8946 bit width. The second element of the result structure must be of type
8947 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8953 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8954 an unsigned multiplication of the two arguments. They return a structure ---
8955 the first element of which is the multiplication, and the second
8956 element of which is a bit specifying if the unsigned multiplication
8957 resulted in an overflow.
8962 .. code-block:: llvm
8964 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8965 %sum = extractvalue {i32, i1} %res, 0
8966 %obit = extractvalue {i32, i1} %res, 1
8967 br i1 %obit, label %overflow, label %normal
8969 Specialised Arithmetic Intrinsics
8970 ---------------------------------
8972 '``llvm.fmuladd.*``' Intrinsic
8973 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8980 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8981 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8986 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8987 expressions that can be fused if the code generator determines that (a) the
8988 target instruction set has support for a fused operation, and (b) that the
8989 fused operation is more efficient than the equivalent, separate pair of mul
8990 and add instructions.
8995 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8996 multiplicands, a and b, and an addend c.
9005 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
9007 is equivalent to the expression a \* b + c, except that rounding will
9008 not be performed between the multiplication and addition steps if the
9009 code generator fuses the operations. Fusion is not guaranteed, even if
9010 the target platform supports it. If a fused multiply-add is required the
9011 corresponding llvm.fma.\* intrinsic function should be used
9012 instead. This never sets errno, just as '``llvm.fma.*``'.
9017 .. code-block:: llvm
9019 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
9021 Half Precision Floating Point Intrinsics
9022 ----------------------------------------
9024 For most target platforms, half precision floating point is a
9025 storage-only format. This means that it is a dense encoding (in memory)
9026 but does not support computation in the format.
9028 This means that code must first load the half-precision floating point
9029 value as an i16, then convert it to float with
9030 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
9031 then be performed on the float value (including extending to double
9032 etc). To store the value back to memory, it is first converted to float
9033 if needed, then converted to i16 with
9034 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
9037 .. _int_convert_to_fp16:
9039 '``llvm.convert.to.fp16``' Intrinsic
9040 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9047 declare i16 @llvm.convert.to.fp16.f32(float %a)
9048 declare i16 @llvm.convert.to.fp16.f64(double %a)
9053 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9054 conventional floating point type to half precision floating point format.
9059 The intrinsic function contains single argument - the value to be
9065 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9066 conventional floating point format to half precision floating point format. The
9067 return value is an ``i16`` which contains the converted number.
9072 .. code-block:: llvm
9074 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
9075 store i16 %res, i16* @x, align 2
9077 .. _int_convert_from_fp16:
9079 '``llvm.convert.from.fp16``' Intrinsic
9080 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9087 declare float @llvm.convert.from.fp16.f32(i16 %a)
9088 declare double @llvm.convert.from.fp16.f64(i16 %a)
9093 The '``llvm.convert.from.fp16``' intrinsic function performs a
9094 conversion from half precision floating point format to single precision
9095 floating point format.
9100 The intrinsic function contains single argument - the value to be
9106 The '``llvm.convert.from.fp16``' intrinsic function performs a
9107 conversion from half single precision floating point format to single
9108 precision floating point format. The input half-float value is
9109 represented by an ``i16`` value.
9114 .. code-block:: llvm
9116 %a = load i16* @x, align 2
9117 %res = call float @llvm.convert.from.fp16(i16 %a)
9122 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
9123 prefix), are described in the `LLVM Source Level
9124 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
9127 Exception Handling Intrinsics
9128 -----------------------------
9130 The LLVM exception handling intrinsics (which all start with
9131 ``llvm.eh.`` prefix), are described in the `LLVM Exception
9132 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
9136 Trampoline Intrinsics
9137 ---------------------
9139 These intrinsics make it possible to excise one parameter, marked with
9140 the :ref:`nest <nest>` attribute, from a function. The result is a
9141 callable function pointer lacking the nest parameter - the caller does
9142 not need to provide a value for it. Instead, the value to use is stored
9143 in advance in a "trampoline", a block of memory usually allocated on the
9144 stack, which also contains code to splice the nest value into the
9145 argument list. This is used to implement the GCC nested function address
9148 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
9149 then the resulting function pointer has signature ``i32 (i32, i32)*``.
9150 It can be created as follows:
9152 .. code-block:: llvm
9154 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
9155 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
9156 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
9157 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
9158 %fp = bitcast i8* %p to i32 (i32, i32)*
9160 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
9161 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
9165 '``llvm.init.trampoline``' Intrinsic
9166 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9173 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
9178 This fills the memory pointed to by ``tramp`` with executable code,
9179 turning it into a trampoline.
9184 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
9185 pointers. The ``tramp`` argument must point to a sufficiently large and
9186 sufficiently aligned block of memory; this memory is written to by the
9187 intrinsic. Note that the size and the alignment are target-specific -
9188 LLVM currently provides no portable way of determining them, so a
9189 front-end that generates this intrinsic needs to have some
9190 target-specific knowledge. The ``func`` argument must hold a function
9191 bitcast to an ``i8*``.
9196 The block of memory pointed to by ``tramp`` is filled with target
9197 dependent code, turning it into a function. Then ``tramp`` needs to be
9198 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
9199 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
9200 function's signature is the same as that of ``func`` with any arguments
9201 marked with the ``nest`` attribute removed. At most one such ``nest``
9202 argument is allowed, and it must be of pointer type. Calling the new
9203 function is equivalent to calling ``func`` with the same argument list,
9204 but with ``nval`` used for the missing ``nest`` argument. If, after
9205 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
9206 modified, then the effect of any later call to the returned function
9207 pointer is undefined.
9211 '``llvm.adjust.trampoline``' Intrinsic
9212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9219 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
9224 This performs any required machine-specific adjustment to the address of
9225 a trampoline (passed as ``tramp``).
9230 ``tramp`` must point to a block of memory which already has trampoline
9231 code filled in by a previous call to
9232 :ref:`llvm.init.trampoline <int_it>`.
9237 On some architectures the address of the code to be executed needs to be
9238 different than the address where the trampoline is actually stored. This
9239 intrinsic returns the executable address corresponding to ``tramp``
9240 after performing the required machine specific adjustments. The pointer
9241 returned can then be :ref:`bitcast and executed <int_trampoline>`.
9243 Masked Vector Load and Store Intrinsics
9244 ---------------------------------------
9246 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.
9250 '``llvm.masked.load.*``' Intrinsics
9251 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9255 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
9259 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
9260 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
9265 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.
9271 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.
9277 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.
9278 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.
9283 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
9285 ;; The result of the two following instructions is identical aside from potential memory access exception
9286 %loadlal = load <16 x float>* %ptr, align 4
9287 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
9291 '``llvm.masked.store.*``' Intrinsics
9292 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9296 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
9300 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
9301 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
9306 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.
9311 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.
9317 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.
9318 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.
9322 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
9324 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
9325 %oldval = load <16 x float>* %ptr, align 4
9326 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
9327 store <16 x float> %res, <16 x float>* %ptr, align 4
9333 This class of intrinsics provides information about the lifetime of
9334 memory objects and ranges where variables are immutable.
9338 '``llvm.lifetime.start``' Intrinsic
9339 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9346 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
9351 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
9357 The first argument is a constant integer representing the size of the
9358 object, or -1 if it is variable sized. The second argument is a pointer
9364 This intrinsic indicates that before this point in the code, the value
9365 of the memory pointed to by ``ptr`` is dead. This means that it is known
9366 to never be used and has an undefined value. A load from the pointer
9367 that precedes this intrinsic can be replaced with ``'undef'``.
9371 '``llvm.lifetime.end``' Intrinsic
9372 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9379 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
9384 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
9390 The first argument is a constant integer representing the size of the
9391 object, or -1 if it is variable sized. The second argument is a pointer
9397 This intrinsic indicates that after this point in the code, the value of
9398 the memory pointed to by ``ptr`` is dead. This means that it is known to
9399 never be used and has an undefined value. Any stores into the memory
9400 object following this intrinsic may be removed as dead.
9402 '``llvm.invariant.start``' Intrinsic
9403 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9410 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
9415 The '``llvm.invariant.start``' intrinsic specifies that the contents of
9416 a memory object will not change.
9421 The first argument is a constant integer representing the size of the
9422 object, or -1 if it is variable sized. The second argument is a pointer
9428 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
9429 the return value, the referenced memory location is constant and
9432 '``llvm.invariant.end``' Intrinsic
9433 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9440 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
9445 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
9446 memory object are mutable.
9451 The first argument is the matching ``llvm.invariant.start`` intrinsic.
9452 The second argument is a constant integer representing the size of the
9453 object, or -1 if it is variable sized and the third argument is a
9454 pointer to the object.
9459 This intrinsic indicates that the memory is mutable again.
9464 This class of intrinsics is designed to be generic and has no specific
9467 '``llvm.var.annotation``' Intrinsic
9468 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9475 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9480 The '``llvm.var.annotation``' intrinsic.
9485 The first argument is a pointer to a value, the second is a pointer to a
9486 global string, the third is a pointer to a global string which is the
9487 source file name, and the last argument is the line number.
9492 This intrinsic allows annotation of local variables with arbitrary
9493 strings. This can be useful for special purpose optimizations that want
9494 to look for these annotations. These have no other defined use; they are
9495 ignored by code generation and optimization.
9497 '``llvm.ptr.annotation.*``' Intrinsic
9498 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9503 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
9504 pointer to an integer of any width. *NOTE* you must specify an address space for
9505 the pointer. The identifier for the default address space is the integer
9510 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9511 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
9512 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
9513 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
9514 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
9519 The '``llvm.ptr.annotation``' intrinsic.
9524 The first argument is a pointer to an integer value of arbitrary bitwidth
9525 (result of some expression), the second is a pointer to a global string, the
9526 third is a pointer to a global string which is the source file name, and the
9527 last argument is the line number. It returns the value of the first argument.
9532 This intrinsic allows annotation of a pointer to an integer with arbitrary
9533 strings. This can be useful for special purpose optimizations that want to look
9534 for these annotations. These have no other defined use; they are ignored by code
9535 generation and optimization.
9537 '``llvm.annotation.*``' Intrinsic
9538 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9543 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
9544 any integer bit width.
9548 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
9549 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
9550 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
9551 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
9552 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
9557 The '``llvm.annotation``' intrinsic.
9562 The first argument is an integer value (result of some expression), the
9563 second is a pointer to a global string, the third is a pointer to a
9564 global string which is the source file name, and the last argument is
9565 the line number. It returns the value of the first argument.
9570 This intrinsic allows annotations to be put on arbitrary expressions
9571 with arbitrary strings. This can be useful for special purpose
9572 optimizations that want to look for these annotations. These have no
9573 other defined use; they are ignored by code generation and optimization.
9575 '``llvm.trap``' Intrinsic
9576 ^^^^^^^^^^^^^^^^^^^^^^^^^
9583 declare void @llvm.trap() noreturn nounwind
9588 The '``llvm.trap``' intrinsic.
9598 This intrinsic is lowered to the target dependent trap instruction. If
9599 the target does not have a trap instruction, this intrinsic will be
9600 lowered to a call of the ``abort()`` function.
9602 '``llvm.debugtrap``' Intrinsic
9603 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9610 declare void @llvm.debugtrap() nounwind
9615 The '``llvm.debugtrap``' intrinsic.
9625 This intrinsic is lowered to code which is intended to cause an
9626 execution trap with the intention of requesting the attention of a
9629 '``llvm.stackprotector``' Intrinsic
9630 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9637 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
9642 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
9643 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
9644 is placed on the stack before local variables.
9649 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
9650 The first argument is the value loaded from the stack guard
9651 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
9652 enough space to hold the value of the guard.
9657 This intrinsic causes the prologue/epilogue inserter to force the position of
9658 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
9659 to ensure that if a local variable on the stack is overwritten, it will destroy
9660 the value of the guard. When the function exits, the guard on the stack is
9661 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
9662 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
9663 calling the ``__stack_chk_fail()`` function.
9665 '``llvm.stackprotectorcheck``' Intrinsic
9666 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9673 declare void @llvm.stackprotectorcheck(i8** <guard>)
9678 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
9679 created stack protector and if they are not equal calls the
9680 ``__stack_chk_fail()`` function.
9685 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
9686 the variable ``@__stack_chk_guard``.
9691 This intrinsic is provided to perform the stack protector check by comparing
9692 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
9693 values do not match call the ``__stack_chk_fail()`` function.
9695 The reason to provide this as an IR level intrinsic instead of implementing it
9696 via other IR operations is that in order to perform this operation at the IR
9697 level without an intrinsic, one would need to create additional basic blocks to
9698 handle the success/failure cases. This makes it difficult to stop the stack
9699 protector check from disrupting sibling tail calls in Codegen. With this
9700 intrinsic, we are able to generate the stack protector basic blocks late in
9701 codegen after the tail call decision has occurred.
9703 '``llvm.objectsize``' Intrinsic
9704 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9711 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
9712 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
9717 The ``llvm.objectsize`` intrinsic is designed to provide information to
9718 the optimizers to determine at compile time whether a) an operation
9719 (like memcpy) will overflow a buffer that corresponds to an object, or
9720 b) that a runtime check for overflow isn't necessary. An object in this
9721 context means an allocation of a specific class, structure, array, or
9727 The ``llvm.objectsize`` intrinsic takes two arguments. The first
9728 argument is a pointer to or into the ``object``. The second argument is
9729 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
9730 or -1 (if false) when the object size is unknown. The second argument
9731 only accepts constants.
9736 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
9737 the size of the object concerned. If the size cannot be determined at
9738 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
9739 on the ``min`` argument).
9741 '``llvm.expect``' Intrinsic
9742 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9747 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
9752 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
9753 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9754 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9759 The ``llvm.expect`` intrinsic provides information about expected (the
9760 most probable) value of ``val``, which can be used by optimizers.
9765 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
9766 a value. The second argument is an expected value, this needs to be a
9767 constant value, variables are not allowed.
9772 This intrinsic is lowered to the ``val``.
9774 '``llvm.assume``' Intrinsic
9775 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9782 declare void @llvm.assume(i1 %cond)
9787 The ``llvm.assume`` allows the optimizer to assume that the provided
9788 condition is true. This information can then be used in simplifying other parts
9794 The condition which the optimizer may assume is always true.
9799 The intrinsic allows the optimizer to assume that the provided condition is
9800 always true whenever the control flow reaches the intrinsic call. No code is
9801 generated for this intrinsic, and instructions that contribute only to the
9802 provided condition are not used for code generation. If the condition is
9803 violated during execution, the behavior is undefined.
9805 Please note that optimizer might limit the transformations performed on values
9806 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
9807 only used to form the intrinsic's input argument. This might prove undesirable
9808 if the extra information provided by the ``llvm.assume`` intrinsic does cause
9809 sufficient overall improvement in code quality. For this reason,
9810 ``llvm.assume`` should not be used to document basic mathematical invariants
9811 that the optimizer can otherwise deduce or facts that are of little use to the
9814 '``llvm.donothing``' Intrinsic
9815 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9822 declare void @llvm.donothing() nounwind readnone
9827 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
9828 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
9829 with an invoke instruction.
9839 This intrinsic does nothing, and it's removed by optimizers and ignored
9842 Stack Map Intrinsics
9843 --------------------
9845 LLVM provides experimental intrinsics to support runtime patching
9846 mechanisms commonly desired in dynamic language JITs. These intrinsics
9847 are described in :doc:`StackMaps`.