1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][-a-zA-Z$._][-a-zA-Z$._0-9]*``'. Identifiers that require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves. The ``"\01"`` prefix
83 can be used on global variables to suppress mangling.
84 #. Unnamed values are represented as an unsigned numeric value with
85 their prefix. For example, ``%12``, ``@2``, ``%44``.
86 #. Constants, which are described in the section Constants_ below.
88 LLVM requires that values start with a prefix for two reasons: Compilers
89 don't need to worry about name clashes with reserved words, and the set
90 of reserved words may be expanded in the future without penalty.
91 Additionally, unnamed identifiers allow a compiler to quickly come up
92 with a temporary variable without having to avoid symbol table
95 Reserved words in LLVM are very similar to reserved words in other
96 languages. There are keywords for different opcodes ('``add``',
97 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
98 '``i32``', etc...), and others. These reserved words cannot conflict
99 with variable names, because none of them start with a prefix character
100 (``'%'`` or ``'@'``).
102 Here is an example of LLVM code to multiply the integer variable
109 %result = mul i32 %X, 8
111 After strength reduction:
115 %result = shl i32 %X, 3
121 %0 = add i32 %X, %X ; yields i32:%0
122 %1 = add i32 %0, %0 ; yields i32:%1
123 %result = add i32 %1, %1
125 This last way of multiplying ``%X`` by 8 illustrates several important
126 lexical features of LLVM:
128 #. Comments are delimited with a '``;``' and go until the end of line.
129 #. Unnamed temporaries are created when the result of a computation is
130 not assigned to a named value.
131 #. Unnamed temporaries are numbered sequentially (using a per-function
132 incrementing counter, starting with 0). Note that basic blocks and unnamed
133 function parameters are included in this numbering. For example, if the
134 entry basic block is not given a label name and all function parameters are
135 named, then it will get number 0.
137 It also shows a convention that we follow in this document. When
138 demonstrating instructions, we will follow an instruction with a comment
139 that defines the type and name of value produced.
147 LLVM programs are composed of ``Module``'s, each of which is a
148 translation unit of the input programs. Each module consists of
149 functions, global variables, and symbol table entries. Modules may be
150 combined together with the LLVM linker, which merges function (and
151 global variable) definitions, resolves forward declarations, and merges
152 symbol table entries. Here is an example of the "hello world" module:
156 ; Declare the string constant as a global constant.
157 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
159 ; External declaration of the puts function
160 declare i32 @puts(i8* nocapture) nounwind
162 ; Definition of main function
163 define i32 @main() { ; i32()*
164 ; Convert [13 x i8]* to i8 *...
165 %cast210 = getelementptr [13 x i8], [13 x i8]* @.str, i64 0, i64 0
167 ; Call puts function to write out the string to stdout.
168 call i32 @puts(i8* %cast210)
173 !0 = !{i32 42, null, !"string"}
176 This example is made up of a :ref:`global variable <globalvars>` named
177 "``.str``", an external declaration of the "``puts``" function, a
178 :ref:`function definition <functionstructure>` for "``main``" and
179 :ref:`named metadata <namedmetadatastructure>` "``foo``".
181 In general, a module is made up of a list of global values (where both
182 functions and global variables are global values). Global values are
183 represented by a pointer to a memory location (in this case, a pointer
184 to an array of char, and a pointer to a function), and have one of the
185 following :ref:`linkage types <linkage>`.
192 All Global Variables and Functions have one of the following types of
196 Global values with "``private``" linkage are only directly
197 accessible by objects in the current module. In particular, linking
198 code into a module with an private global value may cause the
199 private to be renamed as necessary to avoid collisions. Because the
200 symbol is private to the module, all references can be updated. This
201 doesn't show up in any symbol table in the object file.
203 Similar to private, but the value shows as a local symbol
204 (``STB_LOCAL`` in the case of ELF) in the object file. This
205 corresponds to the notion of the '``static``' keyword in C.
206 ``available_externally``
207 Globals with "``available_externally``" linkage are never emitted
208 into the object file corresponding to the LLVM module. They exist to
209 allow inlining and other optimizations to take place given knowledge
210 of the definition of the global, which is known to be somewhere
211 outside the module. Globals with ``available_externally`` linkage
212 are allowed to be discarded at will, and are otherwise the same as
213 ``linkonce_odr``. This linkage type is only allowed on definitions,
216 Globals with "``linkonce``" linkage are merged with other globals of
217 the same name when linkage occurs. This can be used to implement
218 some forms of inline functions, templates, or other code which must
219 be generated in each translation unit that uses it, but where the
220 body may be overridden with a more definitive definition later.
221 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
222 that ``linkonce`` linkage does not actually allow the optimizer to
223 inline the body of this function into callers because it doesn't
224 know if this definition of the function is the definitive definition
225 within the program or whether it will be overridden by a stronger
226 definition. To enable inlining and other optimizations, use
227 "``linkonce_odr``" linkage.
229 "``weak``" linkage has the same merging semantics as ``linkonce``
230 linkage, except that unreferenced globals with ``weak`` linkage may
231 not be discarded. This is used for globals that are declared "weak"
234 "``common``" linkage is most similar to "``weak``" linkage, but they
235 are used for tentative definitions in C, such as "``int X;``" at
236 global scope. Symbols with "``common``" linkage are merged in the
237 same way as ``weak symbols``, and they may not be deleted if
238 unreferenced. ``common`` symbols may not have an explicit section,
239 must have a zero initializer, and may not be marked
240 ':ref:`constant <globalvars>`'. Functions and aliases may not have
243 .. _linkage_appending:
246 "``appending``" linkage may only be applied to global variables of
247 pointer to array type. When two global variables with appending
248 linkage are linked together, the two global arrays are appended
249 together. This is the LLVM, typesafe, equivalent of having the
250 system linker append together "sections" with identical names when
253 The semantics of this linkage follow the ELF object file model: the
254 symbol is weak until linked, if not linked, the symbol becomes null
255 instead of being an undefined reference.
256 ``linkonce_odr``, ``weak_odr``
257 Some languages allow differing globals to be merged, such as two
258 functions with different semantics. Other languages, such as
259 ``C++``, ensure that only equivalent globals are ever merged (the
260 "one definition rule" --- "ODR"). Such languages can use the
261 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
262 global will only be merged with equivalent globals. These linkage
263 types are otherwise the same as their non-``odr`` versions.
265 If none of the above identifiers are used, the global is externally
266 visible, meaning that it participates in linkage and can be used to
267 resolve external symbol references.
269 It is illegal for a function *declaration* to have any linkage type
270 other than ``external`` or ``extern_weak``.
277 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
278 :ref:`invokes <i_invoke>` can all have an optional calling convention
279 specified for the call. The calling convention of any pair of dynamic
280 caller/callee must match, or the behavior of the program is undefined.
281 The following calling conventions are supported by LLVM, and more may be
284 "``ccc``" - The C calling convention
285 This calling convention (the default if no other calling convention
286 is specified) matches the target C calling conventions. This calling
287 convention supports varargs function calls and tolerates some
288 mismatch in the declared prototype and implemented declaration of
289 the function (as does normal C).
290 "``fastcc``" - The fast calling convention
291 This calling convention attempts to make calls as fast as possible
292 (e.g. by passing things in registers). This calling convention
293 allows the target to use whatever tricks it wants to produce fast
294 code for the target, without having to conform to an externally
295 specified ABI (Application Binary Interface). `Tail calls can only
296 be optimized when this, the GHC or the HiPE convention is
297 used. <CodeGenerator.html#id80>`_ This calling convention does not
298 support varargs and requires the prototype of all callees to exactly
299 match the prototype of the function definition.
300 "``coldcc``" - The cold calling convention
301 This calling convention attempts to make code in the caller as
302 efficient as possible under the assumption that the call is not
303 commonly executed. As such, these calls often preserve all registers
304 so that the call does not break any live ranges in the caller side.
305 This calling convention does not support varargs and requires the
306 prototype of all callees to exactly match the prototype of the
307 function definition. Furthermore the inliner doesn't consider such function
309 "``cc 10``" - GHC convention
310 This calling convention has been implemented specifically for use by
311 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
312 It passes everything in registers, going to extremes to achieve this
313 by disabling callee save registers. This calling convention should
314 not be used lightly but only for specific situations such as an
315 alternative to the *register pinning* performance technique often
316 used when implementing functional programming languages. At the
317 moment only X86 supports this convention and it has the following
320 - On *X86-32* only supports up to 4 bit type parameters. No
321 floating point types are supported.
322 - On *X86-64* only supports up to 10 bit type parameters and 6
323 floating point parameters.
325 This calling convention supports `tail call
326 optimization <CodeGenerator.html#id80>`_ but requires both the
327 caller and callee are using it.
328 "``cc 11``" - The HiPE calling convention
329 This calling convention has been implemented specifically for use by
330 the `High-Performance Erlang
331 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
332 native code compiler of the `Ericsson's Open Source Erlang/OTP
333 system <http://www.erlang.org/download.shtml>`_. It uses more
334 registers for argument passing than the ordinary C calling
335 convention and defines no callee-saved registers. The calling
336 convention properly supports `tail call
337 optimization <CodeGenerator.html#id80>`_ but requires that both the
338 caller and the callee use it. It uses a *register pinning*
339 mechanism, similar to GHC's convention, for keeping frequently
340 accessed runtime components pinned to specific hardware registers.
341 At the moment only X86 supports this convention (both 32 and 64
343 "``webkit_jscc``" - WebKit's JavaScript calling convention
344 This calling convention has been implemented for `WebKit FTL JIT
345 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
346 stack right to left (as cdecl does), and returns a value in the
347 platform's customary return register.
348 "``anyregcc``" - Dynamic calling convention for code patching
349 This is a special convention that supports patching an arbitrary code
350 sequence in place of a call site. This convention forces the call
351 arguments into registers but allows them to be dynamically
352 allocated. This can currently only be used with calls to
353 llvm.experimental.patchpoint because only this intrinsic records
354 the location of its arguments in a side table. See :doc:`StackMaps`.
355 "``preserve_mostcc``" - The `PreserveMost` calling convention
356 This calling convention attempts to make the code in the caller as
357 unintrusive as possible. This convention behaves identically to the `C`
358 calling convention on how arguments and return values are passed, but it
359 uses a different set of caller/callee-saved registers. This alleviates the
360 burden of saving and recovering a large register set before and after the
361 call in the caller. If the arguments are passed in callee-saved registers,
362 then they will be preserved by the callee across the call. This doesn't
363 apply for values returned in callee-saved registers.
365 - On X86-64 the callee preserves all general purpose registers, except for
366 R11. R11 can be used as a scratch register. Floating-point registers
367 (XMMs/YMMs) are not preserved and need to be saved by the caller.
369 The idea behind this convention is to support calls to runtime functions
370 that have a hot path and a cold path. The hot path is usually a small piece
371 of code that doesn't use many registers. The cold path might need to call out to
372 another function and therefore only needs to preserve the caller-saved
373 registers, which haven't already been saved by the caller. The
374 `PreserveMost` calling convention is very similar to the `cold` calling
375 convention in terms of caller/callee-saved registers, but they are used for
376 different types of function calls. `coldcc` is for function calls that are
377 rarely executed, whereas `preserve_mostcc` function calls are intended to be
378 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
379 doesn't prevent the inliner from inlining the function call.
381 This calling convention will be used by a future version of the ObjectiveC
382 runtime and should therefore still be considered experimental at this time.
383 Although this convention was created to optimize certain runtime calls to
384 the ObjectiveC runtime, it is not limited to this runtime and might be used
385 by other runtimes in the future too. The current implementation only
386 supports X86-64, but the intention is to support more architectures in the
388 "``preserve_allcc``" - The `PreserveAll` calling convention
389 This calling convention attempts to make the code in the caller even less
390 intrusive than the `PreserveMost` calling convention. This calling
391 convention also behaves identical to the `C` calling convention on how
392 arguments and return values are passed, but it uses a different set of
393 caller/callee-saved registers. This removes the burden of saving and
394 recovering a large register set before and after the call in the caller. If
395 the arguments are passed in callee-saved registers, then they will be
396 preserved by the callee across the call. This doesn't apply for values
397 returned in callee-saved registers.
399 - On X86-64 the callee preserves all general purpose registers, except for
400 R11. R11 can be used as a scratch register. Furthermore it also preserves
401 all floating-point registers (XMMs/YMMs).
403 The idea behind this convention is to support calls to runtime functions
404 that don't need to call out to any other functions.
406 This calling convention, like the `PreserveMost` calling convention, will be
407 used by a future version of the ObjectiveC runtime and should be considered
408 experimental at this time.
409 "``cc <n>``" - Numbered convention
410 Any calling convention may be specified by number, allowing
411 target-specific calling conventions to be used. Target specific
412 calling conventions start at 64.
414 More calling conventions can be added/defined on an as-needed basis, to
415 support Pascal conventions or any other well-known target-independent
418 .. _visibilitystyles:
423 All Global Variables and Functions have one of the following visibility
426 "``default``" - Default style
427 On targets that use the ELF object file format, default visibility
428 means that the declaration is visible to other modules and, in
429 shared libraries, means that the declared entity may be overridden.
430 On Darwin, default visibility means that the declaration is visible
431 to other modules. Default visibility corresponds to "external
432 linkage" in the language.
433 "``hidden``" - Hidden style
434 Two declarations of an object with hidden visibility refer to the
435 same object if they are in the same shared object. Usually, hidden
436 visibility indicates that the symbol will not be placed into the
437 dynamic symbol table, so no other module (executable or shared
438 library) can reference it directly.
439 "``protected``" - Protected style
440 On ELF, protected visibility indicates that the symbol will be
441 placed in the dynamic symbol table, but that references within the
442 defining module will bind to the local symbol. That is, the symbol
443 cannot be overridden by another module.
445 A symbol with ``internal`` or ``private`` linkage must have ``default``
453 All Global Variables, Functions and Aliases can have one of the following
457 "``dllimport``" causes the compiler to reference a function or variable via
458 a global pointer to a pointer that is set up by the DLL exporting the
459 symbol. On Microsoft Windows targets, the pointer name is formed by
460 combining ``__imp_`` and the function or variable name.
462 "``dllexport``" causes the compiler to provide a global pointer to a pointer
463 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
464 Microsoft Windows targets, the pointer name is formed by combining
465 ``__imp_`` and the function or variable name. Since this storage class
466 exists for defining a dll interface, the compiler, assembler and linker know
467 it is externally referenced and must refrain from deleting the symbol.
471 Thread Local Storage Models
472 ---------------------------
474 A variable may be defined as ``thread_local``, which means that it will
475 not be shared by threads (each thread will have a separated copy of the
476 variable). Not all targets support thread-local variables. Optionally, a
477 TLS model may be specified:
480 For variables that are only used within the current shared library.
482 For variables in modules that will not be loaded dynamically.
484 For variables defined in the executable and only used within it.
486 If no explicit model is given, the "general dynamic" model is used.
488 The models correspond to the ELF TLS models; see `ELF Handling For
489 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
490 more information on under which circumstances the different models may
491 be used. The target may choose a different TLS model if the specified
492 model is not supported, or if a better choice of model can be made.
494 A model can also be specified in a alias, but then it only governs how
495 the alias is accessed. It will not have any effect in the aliasee.
502 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
503 types <t_struct>`. Literal types are uniqued structurally, but identified types
504 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
505 to forward declare a type that is not yet available.
507 An example of a identified structure specification is:
511 %mytype = type { %mytype*, i32 }
513 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
514 literal types are uniqued in recent versions of LLVM.
521 Global variables define regions of memory allocated at compilation time
524 Global variable definitions must be initialized.
526 Global variables in other translation units can also be declared, in which
527 case they don't have an initializer.
529 Either global variable definitions or declarations may have an explicit section
530 to be placed in and may have an optional explicit alignment specified.
532 A variable may be defined as a global ``constant``, which indicates that
533 the contents of the variable will **never** be modified (enabling better
534 optimization, allowing the global data to be placed in the read-only
535 section of an executable, etc). Note that variables that need runtime
536 initialization cannot be marked ``constant`` as there is a store to the
539 LLVM explicitly allows *declarations* of global variables to be marked
540 constant, even if the final definition of the global is not. This
541 capability can be used to enable slightly better optimization of the
542 program, but requires the language definition to guarantee that
543 optimizations based on the 'constantness' are valid for the translation
544 units that do not include the definition.
546 As SSA values, global variables define pointer values that are in scope
547 (i.e. they dominate) all basic blocks in the program. Global variables
548 always define a pointer to their "content" type because they describe a
549 region of memory, and all memory objects in LLVM are accessed through
552 Global variables can be marked with ``unnamed_addr`` which indicates
553 that the address is not significant, only the content. Constants marked
554 like this can be merged with other constants if they have the same
555 initializer. Note that a constant with significant address *can* be
556 merged with a ``unnamed_addr`` constant, the result being a constant
557 whose address is significant.
559 A global variable may be declared to reside in a target-specific
560 numbered address space. For targets that support them, address spaces
561 may affect how optimizations are performed and/or what target
562 instructions are used to access the variable. The default address space
563 is zero. The address space qualifier must precede any other attributes.
565 LLVM allows an explicit section to be specified for globals. If the
566 target supports it, it will emit globals to the section specified.
567 Additionally, the global can placed in a comdat if the target has the necessary
570 By default, global initializers are optimized by assuming that global
571 variables defined within the module are not modified from their
572 initial values before the start of the global initializer. This is
573 true even for variables potentially accessible from outside the
574 module, including those with external linkage or appearing in
575 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
576 by marking the variable with ``externally_initialized``.
578 An explicit alignment may be specified for a global, which must be a
579 power of 2. If not present, or if the alignment is set to zero, the
580 alignment of the global is set by the target to whatever it feels
581 convenient. If an explicit alignment is specified, the global is forced
582 to have exactly that alignment. Targets and optimizers are not allowed
583 to over-align the global if the global has an assigned section. In this
584 case, the extra alignment could be observable: for example, code could
585 assume that the globals are densely packed in their section and try to
586 iterate over them as an array, alignment padding would break this
587 iteration. The maximum alignment is ``1 << 29``.
589 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
591 Variables and aliases can have a
592 :ref:`Thread Local Storage Model <tls_model>`.
596 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
597 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
598 <global | constant> <Type> [<InitializerConstant>]
599 [, section "name"] [, comdat [($name)]]
600 [, align <Alignment>]
602 For example, the following defines a global in a numbered address space
603 with an initializer, section, and alignment:
607 @G = addrspace(5) constant float 1.0, section "foo", align 4
609 The following example just declares a global variable
613 @G = external global i32
615 The following example defines a thread-local global with the
616 ``initialexec`` TLS model:
620 @G = thread_local(initialexec) global i32 0, align 4
622 .. _functionstructure:
627 LLVM function definitions consist of the "``define``" keyword, an
628 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
629 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
630 an optional :ref:`calling convention <callingconv>`,
631 an optional ``unnamed_addr`` attribute, a return type, an optional
632 :ref:`parameter attribute <paramattrs>` for the return type, a function
633 name, a (possibly empty) argument list (each with optional :ref:`parameter
634 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
635 an optional section, an optional alignment,
636 an optional :ref:`comdat <langref_comdats>`,
637 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
638 an optional :ref:`prologue <prologuedata>`, an opening
639 curly brace, a list of basic blocks, and a closing curly brace.
641 LLVM function declarations consist of the "``declare``" keyword, an
642 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
643 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
644 an optional :ref:`calling convention <callingconv>`,
645 an optional ``unnamed_addr`` attribute, a return type, an optional
646 :ref:`parameter attribute <paramattrs>` for the return type, a function
647 name, a possibly empty list of arguments, an optional alignment, an optional
648 :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
649 and an optional :ref:`prologue <prologuedata>`.
651 A function definition contains a list of basic blocks, forming the CFG (Control
652 Flow Graph) for the function. Each basic block may optionally start with a label
653 (giving the basic block a symbol table entry), contains a list of instructions,
654 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
655 function return). If an explicit label is not provided, a block is assigned an
656 implicit numbered label, using the next value from the same counter as used for
657 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
658 entry block does not have an explicit label, it will be assigned label "%0",
659 then the first unnamed temporary in that block will be "%1", etc.
661 The first basic block in a function is special in two ways: it is
662 immediately executed on entrance to the function, and it is not allowed
663 to have predecessor basic blocks (i.e. there can not be any branches to
664 the entry block of a function). Because the block can have no
665 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
667 LLVM allows an explicit section to be specified for functions. If the
668 target supports it, it will emit functions to the section specified.
669 Additionally, the function can be placed in a COMDAT.
671 An explicit alignment may be specified for a function. If not present,
672 or if the alignment is set to zero, the alignment of the function is set
673 by the target to whatever it feels convenient. If an explicit alignment
674 is specified, the function is forced to have at least that much
675 alignment. All alignments must be a power of 2.
677 If the ``unnamed_addr`` attribute is given, the address is known to not
678 be significant and two identical functions can be merged.
682 define [linkage] [visibility] [DLLStorageClass]
684 <ResultType> @<FunctionName> ([argument list])
685 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
686 [align N] [gc] [prefix Constant] [prologue Constant] { ... }
688 The argument list is a comma seperated sequence of arguments where each
689 argument is of the following form
693 <type> [parameter Attrs] [name]
701 Aliases, unlike function or variables, don't create any new data. They
702 are just a new symbol and metadata for an existing position.
704 Aliases have a name and an aliasee that is either a global value or a
707 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
708 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
709 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
713 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy> @<Aliasee>
715 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
716 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
717 might not correctly handle dropping a weak symbol that is aliased.
719 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
720 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
723 Since aliases are only a second name, some restrictions apply, of which
724 some can only be checked when producing an object file:
726 * The expression defining the aliasee must be computable at assembly
727 time. Since it is just a name, no relocations can be used.
729 * No alias in the expression can be weak as the possibility of the
730 intermediate alias being overridden cannot be represented in an
733 * No global value in the expression can be a declaration, since that
734 would require a relocation, which is not possible.
741 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
743 Comdats have a name which represents the COMDAT key. All global objects that
744 specify this key will only end up in the final object file if the linker chooses
745 that key over some other key. Aliases are placed in the same COMDAT that their
746 aliasee computes to, if any.
748 Comdats have a selection kind to provide input on how the linker should
749 choose between keys in two different object files.
753 $<Name> = comdat SelectionKind
755 The selection kind must be one of the following:
758 The linker may choose any COMDAT key, the choice is arbitrary.
760 The linker may choose any COMDAT key but the sections must contain the
763 The linker will choose the section containing the largest COMDAT key.
765 The linker requires that only section with this COMDAT key exist.
767 The linker may choose any COMDAT key but the sections must contain the
770 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
771 ``any`` as a selection kind.
773 Here is an example of a COMDAT group where a function will only be selected if
774 the COMDAT key's section is the largest:
778 $foo = comdat largest
779 @foo = global i32 2, comdat($foo)
781 define void @bar() comdat($foo) {
785 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
791 @foo = global i32 2, comdat
794 In a COFF object file, this will create a COMDAT section with selection kind
795 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
796 and another COMDAT section with selection kind
797 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
798 section and contains the contents of the ``@bar`` symbol.
800 There are some restrictions on the properties of the global object.
801 It, or an alias to it, must have the same name as the COMDAT group when
803 The contents and size of this object may be used during link-time to determine
804 which COMDAT groups get selected depending on the selection kind.
805 Because the name of the object must match the name of the COMDAT group, the
806 linkage of the global object must not be local; local symbols can get renamed
807 if a collision occurs in the symbol table.
809 The combined use of COMDATS and section attributes may yield surprising results.
816 @g1 = global i32 42, section "sec", comdat($foo)
817 @g2 = global i32 42, section "sec", comdat($bar)
819 From the object file perspective, this requires the creation of two sections
820 with the same name. This is necessary because both globals belong to different
821 COMDAT groups and COMDATs, at the object file level, are represented by
824 Note that certain IR constructs like global variables and functions may create
825 COMDATs in the object file in addition to any which are specified using COMDAT
826 IR. This arises, for example, when a global variable has linkonce_odr linkage.
828 .. _namedmetadatastructure:
833 Named metadata is a collection of metadata. :ref:`Metadata
834 nodes <metadata>` (but not metadata strings) are the only valid
835 operands for a named metadata.
837 #. Named metadata are represented as a string of characters with the
838 metadata prefix. The rules for metadata names are the same as for
839 identifiers, but quoted names are not allowed. ``"\xx"`` type escapes
840 are still valid, which allows any character to be part of a name.
844 ; Some unnamed metadata nodes, which are referenced by the named metadata.
849 !name = !{!0, !1, !2}
856 The return type and each parameter of a function type may have a set of
857 *parameter attributes* associated with them. Parameter attributes are
858 used to communicate additional information about the result or
859 parameters of a function. Parameter attributes are considered to be part
860 of the function, not of the function type, so functions with different
861 parameter attributes can have the same function type.
863 Parameter attributes are simple keywords that follow the type specified.
864 If multiple parameter attributes are needed, they are space separated.
869 declare i32 @printf(i8* noalias nocapture, ...)
870 declare i32 @atoi(i8 zeroext)
871 declare signext i8 @returns_signed_char()
873 Note that any attributes for the function result (``nounwind``,
874 ``readonly``) come immediately after the argument list.
876 Currently, only the following parameter attributes are defined:
879 This indicates to the code generator that the parameter or return
880 value should be zero-extended to the extent required by the target's
881 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
882 the caller (for a parameter) or the callee (for a return value).
884 This indicates to the code generator that the parameter or return
885 value should be sign-extended to the extent required by the target's
886 ABI (which is usually 32-bits) by the caller (for a parameter) or
887 the callee (for a return value).
889 This indicates that this parameter or return value should be treated
890 in a special target-dependent fashion during while emitting code for
891 a function call or return (usually, by putting it in a register as
892 opposed to memory, though some targets use it to distinguish between
893 two different kinds of registers). Use of this attribute is
896 This indicates that the pointer parameter should really be passed by
897 value to the function. The attribute implies that a hidden copy of
898 the pointee is made between the caller and the callee, so the callee
899 is unable to modify the value in the caller. This attribute is only
900 valid on LLVM pointer arguments. It is generally used to pass
901 structs and arrays by value, but is also valid on pointers to
902 scalars. The copy is considered to belong to the caller not the
903 callee (for example, ``readonly`` functions should not write to
904 ``byval`` parameters). This is not a valid attribute for return
907 The byval attribute also supports specifying an alignment with the
908 align attribute. It indicates the alignment of the stack slot to
909 form and the known alignment of the pointer specified to the call
910 site. If the alignment is not specified, then the code generator
911 makes a target-specific assumption.
917 The ``inalloca`` argument attribute allows the caller to take the
918 address of outgoing stack arguments. An ``inalloca`` argument must
919 be a pointer to stack memory produced by an ``alloca`` instruction.
920 The alloca, or argument allocation, must also be tagged with the
921 inalloca keyword. Only the last argument may have the ``inalloca``
922 attribute, and that argument is guaranteed to be passed in memory.
924 An argument allocation may be used by a call at most once because
925 the call may deallocate it. The ``inalloca`` attribute cannot be
926 used in conjunction with other attributes that affect argument
927 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
928 ``inalloca`` attribute also disables LLVM's implicit lowering of
929 large aggregate return values, which means that frontend authors
930 must lower them with ``sret`` pointers.
932 When the call site is reached, the argument allocation must have
933 been the most recent stack allocation that is still live, or the
934 results are undefined. It is possible to allocate additional stack
935 space after an argument allocation and before its call site, but it
936 must be cleared off with :ref:`llvm.stackrestore
939 See :doc:`InAlloca` for more information on how to use this
943 This indicates that the pointer parameter specifies the address of a
944 structure that is the return value of the function in the source
945 program. This pointer must be guaranteed by the caller to be valid:
946 loads and stores to the structure may be assumed by the callee
947 not to trap and to be properly aligned. This may only be applied to
948 the first parameter. This is not a valid attribute for return
952 This indicates that the pointer value may be assumed by the optimizer to
953 have the specified alignment.
955 Note that this attribute has additional semantics when combined with the
961 This indicates that objects accessed via pointer values
962 :ref:`based <pointeraliasing>` on the argument or return value are not also
963 accessed, during the execution of the function, via pointer values not
964 *based* on the argument or return value. The attribute on a return value
965 also has additional semantics described below. The caller shares the
966 responsibility with the callee for ensuring that these requirements are met.
967 For further details, please see the discussion of the NoAlias response in
968 :ref:`alias analysis <Must, May, or No>`.
970 Note that this definition of ``noalias`` is intentionally similar
971 to the definition of ``restrict`` in C99 for function arguments.
973 For function return values, C99's ``restrict`` is not meaningful,
974 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
975 attribute on return values are stronger than the semantics of the attribute
976 when used on function arguments. On function return values, the ``noalias``
977 attribute indicates that the function acts like a system memory allocation
978 function, returning a pointer to allocated storage disjoint from the
979 storage for any other object accessible to the caller.
982 This indicates that the callee does not make any copies of the
983 pointer that outlive the callee itself. This is not a valid
984 attribute for return values.
989 This indicates that the pointer parameter can be excised using the
990 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
991 attribute for return values and can only be applied to one parameter.
994 This indicates that the function always returns the argument as its return
995 value. This is an optimization hint to the code generator when generating
996 the caller, allowing tail call optimization and omission of register saves
997 and restores in some cases; it is not checked or enforced when generating
998 the callee. The parameter and the function return type must be valid
999 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
1000 valid attribute for return values and can only be applied to one parameter.
1003 This indicates that the parameter or return pointer is not null. This
1004 attribute may only be applied to pointer typed parameters. This is not
1005 checked or enforced by LLVM, the caller must ensure that the pointer
1006 passed in is non-null, or the callee must ensure that the returned pointer
1009 ``dereferenceable(<n>)``
1010 This indicates that the parameter or return pointer is dereferenceable. This
1011 attribute may only be applied to pointer typed parameters. A pointer that
1012 is dereferenceable can be loaded from speculatively without a risk of
1013 trapping. The number of bytes known to be dereferenceable must be provided
1014 in parentheses. It is legal for the number of bytes to be less than the
1015 size of the pointee type. The ``nonnull`` attribute does not imply
1016 dereferenceability (consider a pointer to one element past the end of an
1017 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1018 ``addrspace(0)`` (which is the default address space).
1020 ``dereferenceable_or_null(<n>)``
1021 This indicates that the parameter or return value isn't both
1022 non-null and non-dereferenceable (up to ``<n>`` bytes) at the same
1023 time. All non-null pointers tagged with
1024 ``dereferenceable_or_null(<n>)`` are ``dereferenceable(<n>)``.
1025 For address space 0 ``dereferenceable_or_null(<n>)`` implies that
1026 a pointer is exactly one of ``dereferenceable(<n>)`` or ``null``,
1027 and in other address spaces ``dereferenceable_or_null(<n>)``
1028 implies that a pointer is at least one of ``dereferenceable(<n>)``
1029 or ``null`` (i.e. it may be both ``null`` and
1030 ``dereferenceable(<n>)``). This attribute may only be applied to
1031 pointer typed parameters.
1035 Garbage Collector Strategy Names
1036 --------------------------------
1038 Each function may specify a garbage collector strategy name, which is simply a
1041 .. code-block:: llvm
1043 define void @f() gc "name" { ... }
1045 The supported values of *name* includes those :ref:`built in to LLVM
1046 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1047 strategy will cause the compiler to alter its output in order to support the
1048 named garbage collection algorithm. Note that LLVM itself does not contain a
1049 garbage collector, this functionality is restricted to generating machine code
1050 which can interoperate with a collector provided externally.
1057 Prefix data is data associated with a function which the code
1058 generator will emit immediately before the function's entrypoint.
1059 The purpose of this feature is to allow frontends to associate
1060 language-specific runtime metadata with specific functions and make it
1061 available through the function pointer while still allowing the
1062 function pointer to be called.
1064 To access the data for a given function, a program may bitcast the
1065 function pointer to a pointer to the constant's type and dereference
1066 index -1. This implies that the IR symbol points just past the end of
1067 the prefix data. For instance, take the example of a function annotated
1068 with a single ``i32``,
1070 .. code-block:: llvm
1072 define void @f() prefix i32 123 { ... }
1074 The prefix data can be referenced as,
1076 .. code-block:: llvm
1078 %0 = bitcast void* () @f to i32*
1079 %a = getelementptr inbounds i32, i32* %0, i32 -1
1080 %b = load i32, i32* %a
1082 Prefix data is laid out as if it were an initializer for a global variable
1083 of the prefix data's type. The function will be placed such that the
1084 beginning of the prefix data is aligned. This means that if the size
1085 of the prefix data is not a multiple of the alignment size, the
1086 function's entrypoint will not be aligned. If alignment of the
1087 function's entrypoint is desired, padding must be added to the prefix
1090 A function may have prefix data but no body. This has similar semantics
1091 to the ``available_externally`` linkage in that the data may be used by the
1092 optimizers but will not be emitted in the object file.
1099 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1100 be inserted prior to the function body. This can be used for enabling
1101 function hot-patching and instrumentation.
1103 To maintain the semantics of ordinary function calls, the prologue data must
1104 have a particular format. Specifically, it must begin with a sequence of
1105 bytes which decode to a sequence of machine instructions, valid for the
1106 module's target, which transfer control to the point immediately succeeding
1107 the prologue data, without performing any other visible action. This allows
1108 the inliner and other passes to reason about the semantics of the function
1109 definition without needing to reason about the prologue data. Obviously this
1110 makes the format of the prologue data highly target dependent.
1112 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1113 which encodes the ``nop`` instruction:
1115 .. code-block:: llvm
1117 define void @f() prologue i8 144 { ... }
1119 Generally prologue data can be formed by encoding a relative branch instruction
1120 which skips the metadata, as in this example of valid prologue data for the
1121 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1123 .. code-block:: llvm
1125 %0 = type <{ i8, i8, i8* }>
1127 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1129 A function may have prologue data but no body. This has similar semantics
1130 to the ``available_externally`` linkage in that the data may be used by the
1131 optimizers but will not be emitted in the object file.
1138 Attribute groups are groups of attributes that are referenced by objects within
1139 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1140 functions will use the same set of attributes. In the degenerative case of a
1141 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1142 group will capture the important command line flags used to build that file.
1144 An attribute group is a module-level object. To use an attribute group, an
1145 object references the attribute group's ID (e.g. ``#37``). An object may refer
1146 to more than one attribute group. In that situation, the attributes from the
1147 different groups are merged.
1149 Here is an example of attribute groups for a function that should always be
1150 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1152 .. code-block:: llvm
1154 ; Target-independent attributes:
1155 attributes #0 = { alwaysinline alignstack=4 }
1157 ; Target-dependent attributes:
1158 attributes #1 = { "no-sse" }
1160 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1161 define void @f() #0 #1 { ... }
1168 Function attributes are set to communicate additional information about
1169 a function. Function attributes are considered to be part of the
1170 function, not of the function type, so functions with different function
1171 attributes can have the same function type.
1173 Function attributes are simple keywords that follow the type specified.
1174 If multiple attributes are needed, they are space separated. For
1177 .. code-block:: llvm
1179 define void @f() noinline { ... }
1180 define void @f() alwaysinline { ... }
1181 define void @f() alwaysinline optsize { ... }
1182 define void @f() optsize { ... }
1185 This attribute indicates that, when emitting the prologue and
1186 epilogue, the backend should forcibly align the stack pointer.
1187 Specify the desired alignment, which must be a power of two, in
1190 This attribute indicates that the inliner should attempt to inline
1191 this function into callers whenever possible, ignoring any active
1192 inlining size threshold for this caller.
1194 This indicates that the callee function at a call site should be
1195 recognized as a built-in function, even though the function's declaration
1196 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1197 direct calls to functions that are declared with the ``nobuiltin``
1200 This attribute indicates that this function is rarely called. When
1201 computing edge weights, basic blocks post-dominated by a cold
1202 function call are also considered to be cold; and, thus, given low
1205 This attribute indicates that the callee is dependent on a convergent
1206 thread execution pattern under certain parallel execution models.
1207 Transformations that are execution model agnostic may only move or
1208 tranform this call if the final location is control equivalent to its
1209 original position in the program, where control equivalence is defined as
1210 A dominates B and B post-dominates A, or vice versa.
1212 This attribute indicates that the source code contained a hint that
1213 inlining this function is desirable (such as the "inline" keyword in
1214 C/C++). It is just a hint; it imposes no requirements on the
1217 This attribute indicates that the function should be added to a
1218 jump-instruction table at code-generation time, and that all address-taken
1219 references to this function should be replaced with a reference to the
1220 appropriate jump-instruction-table function pointer. Note that this creates
1221 a new pointer for the original function, which means that code that depends
1222 on function-pointer identity can break. So, any function annotated with
1223 ``jumptable`` must also be ``unnamed_addr``.
1225 This attribute suggests that optimization passes and code generator
1226 passes make choices that keep the code size of this function as small
1227 as possible and perform optimizations that may sacrifice runtime
1228 performance in order to minimize the size of the generated code.
1230 This attribute disables prologue / epilogue emission for the
1231 function. This can have very system-specific consequences.
1233 This indicates that the callee function at a call site is not recognized as
1234 a built-in function. LLVM will retain the original call and not replace it
1235 with equivalent code based on the semantics of the built-in function, unless
1236 the call site uses the ``builtin`` attribute. This is valid at call sites
1237 and on function declarations and definitions.
1239 This attribute indicates that calls to the function cannot be
1240 duplicated. A call to a ``noduplicate`` function may be moved
1241 within its parent function, but may not be duplicated within
1242 its parent function.
1244 A function containing a ``noduplicate`` call may still
1245 be an inlining candidate, provided that the call is not
1246 duplicated by inlining. That implies that the function has
1247 internal linkage and only has one call site, so the original
1248 call is dead after inlining.
1250 This attributes disables implicit floating point instructions.
1252 This attribute indicates that the inliner should never inline this
1253 function in any situation. This attribute may not be used together
1254 with the ``alwaysinline`` attribute.
1256 This attribute suppresses lazy symbol binding for the function. This
1257 may make calls to the function faster, at the cost of extra program
1258 startup time if the function is not called during program startup.
1260 This attribute indicates that the code generator should not use a
1261 red zone, even if the target-specific ABI normally permits it.
1263 This function attribute indicates that the function never returns
1264 normally. This produces undefined behavior at runtime if the
1265 function ever does dynamically return.
1267 This function attribute indicates that the function never raises an
1268 exception. If the function does raise an exception, its runtime
1269 behavior is undefined. However, functions marked nounwind may still
1270 trap or generate asynchronous exceptions. Exception handling schemes
1271 that are recognized by LLVM to handle asynchronous exceptions, such
1272 as SEH, will still provide their implementation defined semantics.
1274 This function attribute indicates that the function is not optimized
1275 by any optimization or code generator passes with the
1276 exception of interprocedural optimization passes.
1277 This attribute cannot be used together with the ``alwaysinline``
1278 attribute; this attribute is also incompatible
1279 with the ``minsize`` attribute and the ``optsize`` attribute.
1281 This attribute requires the ``noinline`` attribute to be specified on
1282 the function as well, so the function is never inlined into any caller.
1283 Only functions with the ``alwaysinline`` attribute are valid
1284 candidates for inlining into the body of this function.
1286 This attribute suggests that optimization passes and code generator
1287 passes make choices that keep the code size of this function low,
1288 and otherwise do optimizations specifically to reduce code size as
1289 long as they do not significantly impact runtime performance.
1291 On a function, this attribute indicates that the function computes its
1292 result (or decides to unwind an exception) based strictly on its arguments,
1293 without dereferencing any pointer arguments or otherwise accessing
1294 any mutable state (e.g. memory, control registers, etc) visible to
1295 caller functions. It does not write through any pointer arguments
1296 (including ``byval`` arguments) and never changes any state visible
1297 to callers. This means that it cannot unwind exceptions by calling
1298 the ``C++`` exception throwing methods.
1300 On an argument, this attribute indicates that the function does not
1301 dereference that pointer argument, even though it may read or write the
1302 memory that the pointer points to if accessed through other pointers.
1304 On a function, this attribute indicates that the function does not write
1305 through any pointer arguments (including ``byval`` arguments) or otherwise
1306 modify any state (e.g. memory, control registers, etc) visible to
1307 caller functions. It may dereference pointer arguments and read
1308 state that may be set in the caller. A readonly function always
1309 returns the same value (or unwinds an exception identically) when
1310 called with the same set of arguments and global state. It cannot
1311 unwind an exception by calling the ``C++`` exception throwing
1314 On an argument, this attribute indicates that the function does not write
1315 through this pointer argument, even though it may write to the memory that
1316 the pointer points to.
1318 This attribute indicates that this function can return twice. The C
1319 ``setjmp`` is an example of such a function. The compiler disables
1320 some optimizations (like tail calls) in the caller of these
1322 ``sanitize_address``
1323 This attribute indicates that AddressSanitizer checks
1324 (dynamic address safety analysis) are enabled for this function.
1326 This attribute indicates that MemorySanitizer checks (dynamic detection
1327 of accesses to uninitialized memory) are enabled for this function.
1329 This attribute indicates that ThreadSanitizer checks
1330 (dynamic thread safety analysis) are enabled for this function.
1332 This attribute indicates that the function should emit a stack
1333 smashing protector. It is in the form of a "canary" --- a random value
1334 placed on the stack before the local variables that's checked upon
1335 return from the function to see if it has been overwritten. A
1336 heuristic is used to determine if a function needs stack protectors
1337 or not. The heuristic used will enable protectors for functions with:
1339 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1340 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1341 - Calls to alloca() with variable sizes or constant sizes greater than
1342 ``ssp-buffer-size``.
1344 Variables that are identified as requiring a protector will be arranged
1345 on the stack such that they are adjacent to the stack protector guard.
1347 If a function that has an ``ssp`` attribute is inlined into a
1348 function that doesn't have an ``ssp`` attribute, then the resulting
1349 function will have an ``ssp`` attribute.
1351 This attribute indicates that the function should *always* emit a
1352 stack smashing protector. This overrides the ``ssp`` function
1355 Variables that are identified as requiring a protector will be arranged
1356 on the stack such that they are adjacent to the stack protector guard.
1357 The specific layout rules are:
1359 #. Large arrays and structures containing large arrays
1360 (``>= ssp-buffer-size``) are closest to the stack protector.
1361 #. Small arrays and structures containing small arrays
1362 (``< ssp-buffer-size``) are 2nd closest to the protector.
1363 #. Variables that have had their address taken are 3rd closest to the
1366 If a function that has an ``sspreq`` attribute is inlined into a
1367 function that doesn't have an ``sspreq`` attribute or which has an
1368 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1369 an ``sspreq`` attribute.
1371 This attribute indicates that the function should emit a stack smashing
1372 protector. This attribute causes a strong heuristic to be used when
1373 determining if a function needs stack protectors. The strong heuristic
1374 will enable protectors for functions with:
1376 - Arrays of any size and type
1377 - Aggregates containing an array of any size and type.
1378 - Calls to alloca().
1379 - Local variables that have had their address taken.
1381 Variables that are identified as requiring a protector will be arranged
1382 on the stack such that they are adjacent to the stack protector guard.
1383 The specific layout rules are:
1385 #. Large arrays and structures containing large arrays
1386 (``>= ssp-buffer-size``) are closest to the stack protector.
1387 #. Small arrays and structures containing small arrays
1388 (``< ssp-buffer-size``) are 2nd closest to the protector.
1389 #. Variables that have had their address taken are 3rd closest to the
1392 This overrides the ``ssp`` function attribute.
1394 If a function that has an ``sspstrong`` attribute is inlined into a
1395 function that doesn't have an ``sspstrong`` attribute, then the
1396 resulting function will have an ``sspstrong`` attribute.
1398 This attribute indicates that the function will delegate to some other
1399 function with a tail call. The prototype of a thunk should not be used for
1400 optimization purposes. The caller is expected to cast the thunk prototype to
1401 match the thunk target prototype.
1403 This attribute indicates that the ABI being targeted requires that
1404 an unwind table entry be produce for this function even if we can
1405 show that no exceptions passes by it. This is normally the case for
1406 the ELF x86-64 abi, but it can be disabled for some compilation
1411 Module-Level Inline Assembly
1412 ----------------------------
1414 Modules may contain "module-level inline asm" blocks, which corresponds
1415 to the GCC "file scope inline asm" blocks. These blocks are internally
1416 concatenated by LLVM and treated as a single unit, but may be separated
1417 in the ``.ll`` file if desired. The syntax is very simple:
1419 .. code-block:: llvm
1421 module asm "inline asm code goes here"
1422 module asm "more can go here"
1424 The strings can contain any character by escaping non-printable
1425 characters. The escape sequence used is simply "\\xx" where "xx" is the
1426 two digit hex code for the number.
1428 The inline asm code is simply printed to the machine code .s file when
1429 assembly code is generated.
1431 .. _langref_datalayout:
1436 A module may specify a target specific data layout string that specifies
1437 how data is to be laid out in memory. The syntax for the data layout is
1440 .. code-block:: llvm
1442 target datalayout = "layout specification"
1444 The *layout specification* consists of a list of specifications
1445 separated by the minus sign character ('-'). Each specification starts
1446 with a letter and may include other information after the letter to
1447 define some aspect of the data layout. The specifications accepted are
1451 Specifies that the target lays out data in big-endian form. That is,
1452 the bits with the most significance have the lowest address
1455 Specifies that the target lays out data in little-endian form. That
1456 is, the bits with the least significance have the lowest address
1459 Specifies the natural alignment of the stack in bits. Alignment
1460 promotion of stack variables is limited to the natural stack
1461 alignment to avoid dynamic stack realignment. The stack alignment
1462 must be a multiple of 8-bits. If omitted, the natural stack
1463 alignment defaults to "unspecified", which does not prevent any
1464 alignment promotions.
1465 ``p[n]:<size>:<abi>:<pref>``
1466 This specifies the *size* of a pointer and its ``<abi>`` and
1467 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1468 bits. The address space, ``n`` is optional, and if not specified,
1469 denotes the default address space 0. The value of ``n`` must be
1470 in the range [1,2^23).
1471 ``i<size>:<abi>:<pref>``
1472 This specifies the alignment for an integer type of a given bit
1473 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1474 ``v<size>:<abi>:<pref>``
1475 This specifies the alignment for a vector type of a given bit
1477 ``f<size>:<abi>:<pref>``
1478 This specifies the alignment for a floating point type of a given bit
1479 ``<size>``. Only values of ``<size>`` that are supported by the target
1480 will work. 32 (float) and 64 (double) are supported on all targets; 80
1481 or 128 (different flavors of long double) are also supported on some
1484 This specifies the alignment for an object of aggregate type.
1486 If present, specifies that llvm names are mangled in the output. The
1489 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1490 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1491 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1492 symbols get a ``_`` prefix.
1493 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1494 functions also get a suffix based on the frame size.
1495 ``n<size1>:<size2>:<size3>...``
1496 This specifies a set of native integer widths for the target CPU in
1497 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1498 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1499 this set are considered to support most general arithmetic operations
1502 On every specification that takes a ``<abi>:<pref>``, specifying the
1503 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1504 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1506 When constructing the data layout for a given target, LLVM starts with a
1507 default set of specifications which are then (possibly) overridden by
1508 the specifications in the ``datalayout`` keyword. The default
1509 specifications are given in this list:
1511 - ``E`` - big endian
1512 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1513 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1514 same as the default address space.
1515 - ``S0`` - natural stack alignment is unspecified
1516 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1517 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1518 - ``i16:16:16`` - i16 is 16-bit aligned
1519 - ``i32:32:32`` - i32 is 32-bit aligned
1520 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1521 alignment of 64-bits
1522 - ``f16:16:16`` - half is 16-bit aligned
1523 - ``f32:32:32`` - float is 32-bit aligned
1524 - ``f64:64:64`` - double is 64-bit aligned
1525 - ``f128:128:128`` - quad is 128-bit aligned
1526 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1527 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1528 - ``a:0:64`` - aggregates are 64-bit aligned
1530 When LLVM is determining the alignment for a given type, it uses the
1533 #. If the type sought is an exact match for one of the specifications,
1534 that specification is used.
1535 #. If no match is found, and the type sought is an integer type, then
1536 the smallest integer type that is larger than the bitwidth of the
1537 sought type is used. If none of the specifications are larger than
1538 the bitwidth then the largest integer type is used. For example,
1539 given the default specifications above, the i7 type will use the
1540 alignment of i8 (next largest) while both i65 and i256 will use the
1541 alignment of i64 (largest specified).
1542 #. If no match is found, and the type sought is a vector type, then the
1543 largest vector type that is smaller than the sought vector type will
1544 be used as a fall back. This happens because <128 x double> can be
1545 implemented in terms of 64 <2 x double>, for example.
1547 The function of the data layout string may not be what you expect.
1548 Notably, this is not a specification from the frontend of what alignment
1549 the code generator should use.
1551 Instead, if specified, the target data layout is required to match what
1552 the ultimate *code generator* expects. This string is used by the
1553 mid-level optimizers to improve code, and this only works if it matches
1554 what the ultimate code generator uses. There is no way to generate IR
1555 that does not embed this target-specific detail into the IR. If you
1556 don't specify the string, the default specifications will be used to
1557 generate a Data Layout and the optimization phases will operate
1558 accordingly and introduce target specificity into the IR with respect to
1559 these default specifications.
1566 A module may specify a target triple string that describes the target
1567 host. The syntax for the target triple is simply:
1569 .. code-block:: llvm
1571 target triple = "x86_64-apple-macosx10.7.0"
1573 The *target triple* string consists of a series of identifiers delimited
1574 by the minus sign character ('-'). The canonical forms are:
1578 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1579 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1581 This information is passed along to the backend so that it generates
1582 code for the proper architecture. It's possible to override this on the
1583 command line with the ``-mtriple`` command line option.
1585 .. _pointeraliasing:
1587 Pointer Aliasing Rules
1588 ----------------------
1590 Any memory access must be done through a pointer value associated with
1591 an address range of the memory access, otherwise the behavior is
1592 undefined. Pointer values are associated with address ranges according
1593 to the following rules:
1595 - A pointer value is associated with the addresses associated with any
1596 value it is *based* on.
1597 - An address of a global variable is associated with the address range
1598 of the variable's storage.
1599 - The result value of an allocation instruction is associated with the
1600 address range of the allocated storage.
1601 - A null pointer in the default address-space is associated with no
1603 - An integer constant other than zero or a pointer value returned from
1604 a function not defined within LLVM may be associated with address
1605 ranges allocated through mechanisms other than those provided by
1606 LLVM. Such ranges shall not overlap with any ranges of addresses
1607 allocated by mechanisms provided by LLVM.
1609 A pointer value is *based* on another pointer value according to the
1612 - A pointer value formed from a ``getelementptr`` operation is *based*
1613 on the first value operand of the ``getelementptr``.
1614 - The result value of a ``bitcast`` is *based* on the operand of the
1616 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1617 values that contribute (directly or indirectly) to the computation of
1618 the pointer's value.
1619 - The "*based* on" relationship is transitive.
1621 Note that this definition of *"based"* is intentionally similar to the
1622 definition of *"based"* in C99, though it is slightly weaker.
1624 LLVM IR does not associate types with memory. The result type of a
1625 ``load`` merely indicates the size and alignment of the memory from
1626 which to load, as well as the interpretation of the value. The first
1627 operand type of a ``store`` similarly only indicates the size and
1628 alignment of the store.
1630 Consequently, type-based alias analysis, aka TBAA, aka
1631 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1632 :ref:`Metadata <metadata>` may be used to encode additional information
1633 which specialized optimization passes may use to implement type-based
1638 Volatile Memory Accesses
1639 ------------------------
1641 Certain memory accesses, such as :ref:`load <i_load>`'s,
1642 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1643 marked ``volatile``. The optimizers must not change the number of
1644 volatile operations or change their order of execution relative to other
1645 volatile operations. The optimizers *may* change the order of volatile
1646 operations relative to non-volatile operations. This is not Java's
1647 "volatile" and has no cross-thread synchronization behavior.
1649 IR-level volatile loads and stores cannot safely be optimized into
1650 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1651 flagged volatile. Likewise, the backend should never split or merge
1652 target-legal volatile load/store instructions.
1654 .. admonition:: Rationale
1656 Platforms may rely on volatile loads and stores of natively supported
1657 data width to be executed as single instruction. For example, in C
1658 this holds for an l-value of volatile primitive type with native
1659 hardware support, but not necessarily for aggregate types. The
1660 frontend upholds these expectations, which are intentionally
1661 unspecified in the IR. The rules above ensure that IR transformation
1662 do not violate the frontend's contract with the language.
1666 Memory Model for Concurrent Operations
1667 --------------------------------------
1669 The LLVM IR does not define any way to start parallel threads of
1670 execution or to register signal handlers. Nonetheless, there are
1671 platform-specific ways to create them, and we define LLVM IR's behavior
1672 in their presence. This model is inspired by the C++0x memory model.
1674 For a more informal introduction to this model, see the :doc:`Atomics`.
1676 We define a *happens-before* partial order as the least partial order
1679 - Is a superset of single-thread program order, and
1680 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1681 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1682 techniques, like pthread locks, thread creation, thread joining,
1683 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1684 Constraints <ordering>`).
1686 Note that program order does not introduce *happens-before* edges
1687 between a thread and signals executing inside that thread.
1689 Every (defined) read operation (load instructions, memcpy, atomic
1690 loads/read-modify-writes, etc.) R reads a series of bytes written by
1691 (defined) write operations (store instructions, atomic
1692 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1693 section, initialized globals are considered to have a write of the
1694 initializer which is atomic and happens before any other read or write
1695 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1696 may see any write to the same byte, except:
1698 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1699 write\ :sub:`2` happens before R\ :sub:`byte`, then
1700 R\ :sub:`byte` does not see write\ :sub:`1`.
1701 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1702 R\ :sub:`byte` does not see write\ :sub:`3`.
1704 Given that definition, R\ :sub:`byte` is defined as follows:
1706 - If R is volatile, the result is target-dependent. (Volatile is
1707 supposed to give guarantees which can support ``sig_atomic_t`` in
1708 C/C++, and may be used for accesses to addresses that do not behave
1709 like normal memory. It does not generally provide cross-thread
1711 - Otherwise, if there is no write to the same byte that happens before
1712 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1713 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1714 R\ :sub:`byte` returns the value written by that write.
1715 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1716 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1717 Memory Ordering Constraints <ordering>` section for additional
1718 constraints on how the choice is made.
1719 - Otherwise R\ :sub:`byte` returns ``undef``.
1721 R returns the value composed of the series of bytes it read. This
1722 implies that some bytes within the value may be ``undef`` **without**
1723 the entire value being ``undef``. Note that this only defines the
1724 semantics of the operation; it doesn't mean that targets will emit more
1725 than one instruction to read the series of bytes.
1727 Note that in cases where none of the atomic intrinsics are used, this
1728 model places only one restriction on IR transformations on top of what
1729 is required for single-threaded execution: introducing a store to a byte
1730 which might not otherwise be stored is not allowed in general.
1731 (Specifically, in the case where another thread might write to and read
1732 from an address, introducing a store can change a load that may see
1733 exactly one write into a load that may see multiple writes.)
1737 Atomic Memory Ordering Constraints
1738 ----------------------------------
1740 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1741 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1742 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1743 ordering parameters that determine which other atomic instructions on
1744 the same address they *synchronize with*. These semantics are borrowed
1745 from Java and C++0x, but are somewhat more colloquial. If these
1746 descriptions aren't precise enough, check those specs (see spec
1747 references in the :doc:`atomics guide <Atomics>`).
1748 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1749 differently since they don't take an address. See that instruction's
1750 documentation for details.
1752 For a simpler introduction to the ordering constraints, see the
1756 The set of values that can be read is governed by the happens-before
1757 partial order. A value cannot be read unless some operation wrote
1758 it. This is intended to provide a guarantee strong enough to model
1759 Java's non-volatile shared variables. This ordering cannot be
1760 specified for read-modify-write operations; it is not strong enough
1761 to make them atomic in any interesting way.
1763 In addition to the guarantees of ``unordered``, there is a single
1764 total order for modifications by ``monotonic`` operations on each
1765 address. All modification orders must be compatible with the
1766 happens-before order. There is no guarantee that the modification
1767 orders can be combined to a global total order for the whole program
1768 (and this often will not be possible). The read in an atomic
1769 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1770 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1771 order immediately before the value it writes. If one atomic read
1772 happens before another atomic read of the same address, the later
1773 read must see the same value or a later value in the address's
1774 modification order. This disallows reordering of ``monotonic`` (or
1775 stronger) operations on the same address. If an address is written
1776 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1777 read that address repeatedly, the other threads must eventually see
1778 the write. This corresponds to the C++0x/C1x
1779 ``memory_order_relaxed``.
1781 In addition to the guarantees of ``monotonic``, a
1782 *synchronizes-with* edge may be formed with a ``release`` operation.
1783 This is intended to model C++'s ``memory_order_acquire``.
1785 In addition to the guarantees of ``monotonic``, if this operation
1786 writes a value which is subsequently read by an ``acquire``
1787 operation, it *synchronizes-with* that operation. (This isn't a
1788 complete description; see the C++0x definition of a release
1789 sequence.) This corresponds to the C++0x/C1x
1790 ``memory_order_release``.
1791 ``acq_rel`` (acquire+release)
1792 Acts as both an ``acquire`` and ``release`` operation on its
1793 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1794 ``seq_cst`` (sequentially consistent)
1795 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1796 operation that only reads, ``release`` for an operation that only
1797 writes), there is a global total order on all
1798 sequentially-consistent operations on all addresses, which is
1799 consistent with the *happens-before* partial order and with the
1800 modification orders of all the affected addresses. Each
1801 sequentially-consistent read sees the last preceding write to the
1802 same address in this global order. This corresponds to the C++0x/C1x
1803 ``memory_order_seq_cst`` and Java volatile.
1807 If an atomic operation is marked ``singlethread``, it only *synchronizes
1808 with* or participates in modification and seq\_cst total orderings with
1809 other operations running in the same thread (for example, in signal
1817 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1818 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1819 :ref:`frem <i_frem>`) have the following flags that can be set to enable
1820 otherwise unsafe floating point operations
1823 No NaNs - Allow optimizations to assume the arguments and result are not
1824 NaN. Such optimizations are required to retain defined behavior over
1825 NaNs, but the value of the result is undefined.
1828 No Infs - Allow optimizations to assume the arguments and result are not
1829 +/-Inf. Such optimizations are required to retain defined behavior over
1830 +/-Inf, but the value of the result is undefined.
1833 No Signed Zeros - Allow optimizations to treat the sign of a zero
1834 argument or result as insignificant.
1837 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1838 argument rather than perform division.
1841 Fast - Allow algebraically equivalent transformations that may
1842 dramatically change results in floating point (e.g. reassociate). This
1843 flag implies all the others.
1847 Use-list Order Directives
1848 -------------------------
1850 Use-list directives encode the in-memory order of each use-list, allowing the
1851 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1852 indexes that are assigned to the referenced value's uses. The referenced
1853 value's use-list is immediately sorted by these indexes.
1855 Use-list directives may appear at function scope or global scope. They are not
1856 instructions, and have no effect on the semantics of the IR. When they're at
1857 function scope, they must appear after the terminator of the final basic block.
1859 If basic blocks have their address taken via ``blockaddress()`` expressions,
1860 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1867 uselistorder <ty> <value>, { <order-indexes> }
1868 uselistorder_bb @function, %block { <order-indexes> }
1874 define void @foo(i32 %arg1, i32 %arg2) {
1876 ; ... instructions ...
1878 ; ... instructions ...
1880 ; At function scope.
1881 uselistorder i32 %arg1, { 1, 0, 2 }
1882 uselistorder label %bb, { 1, 0 }
1886 uselistorder i32* @global, { 1, 2, 0 }
1887 uselistorder i32 7, { 1, 0 }
1888 uselistorder i32 (i32) @bar, { 1, 0 }
1889 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1896 The LLVM type system is one of the most important features of the
1897 intermediate representation. Being typed enables a number of
1898 optimizations to be performed on the intermediate representation
1899 directly, without having to do extra analyses on the side before the
1900 transformation. A strong type system makes it easier to read the
1901 generated code and enables novel analyses and transformations that are
1902 not feasible to perform on normal three address code representations.
1912 The void type does not represent any value and has no size.
1930 The function type can be thought of as a function signature. It consists of a
1931 return type and a list of formal parameter types. The return type of a function
1932 type is a void type or first class type --- except for :ref:`label <t_label>`
1933 and :ref:`metadata <t_metadata>` types.
1939 <returntype> (<parameter list>)
1941 ...where '``<parameter list>``' is a comma-separated list of type
1942 specifiers. Optionally, the parameter list may include a type ``...``, which
1943 indicates that the function takes a variable number of arguments. Variable
1944 argument functions can access their arguments with the :ref:`variable argument
1945 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1946 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1950 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1951 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1952 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1953 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1954 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1955 | ``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. |
1956 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1957 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1958 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1965 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1966 Values of these types are the only ones which can be produced by
1974 These are the types that are valid in registers from CodeGen's perspective.
1983 The integer type is a very simple type that simply specifies an
1984 arbitrary bit width for the integer type desired. Any bit width from 1
1985 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1993 The number of bits the integer will occupy is specified by the ``N``
1999 +----------------+------------------------------------------------+
2000 | ``i1`` | a single-bit integer. |
2001 +----------------+------------------------------------------------+
2002 | ``i32`` | a 32-bit integer. |
2003 +----------------+------------------------------------------------+
2004 | ``i1942652`` | a really big integer of over 1 million bits. |
2005 +----------------+------------------------------------------------+
2009 Floating Point Types
2010 """"""""""""""""""""
2019 - 16-bit floating point value
2022 - 32-bit floating point value
2025 - 64-bit floating point value
2028 - 128-bit floating point value (112-bit mantissa)
2031 - 80-bit floating point value (X87)
2034 - 128-bit floating point value (two 64-bits)
2041 The x86_mmx type represents a value held in an MMX register on an x86
2042 machine. The operations allowed on it are quite limited: parameters and
2043 return values, load and store, and bitcast. User-specified MMX
2044 instructions are represented as intrinsic or asm calls with arguments
2045 and/or results of this type. There are no arrays, vectors or constants
2062 The pointer type is used to specify memory locations. Pointers are
2063 commonly used to reference objects in memory.
2065 Pointer types may have an optional address space attribute defining the
2066 numbered address space where the pointed-to object resides. The default
2067 address space is number zero. The semantics of non-zero address spaces
2068 are target-specific.
2070 Note that LLVM does not permit pointers to void (``void*``) nor does it
2071 permit pointers to labels (``label*``). Use ``i8*`` instead.
2081 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2082 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2083 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2084 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2085 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2086 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2087 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2096 A vector type is a simple derived type that represents a vector of
2097 elements. Vector types are used when multiple primitive data are
2098 operated in parallel using a single instruction (SIMD). A vector type
2099 requires a size (number of elements) and an underlying primitive data
2100 type. Vector types are considered :ref:`first class <t_firstclass>`.
2106 < <# elements> x <elementtype> >
2108 The number of elements is a constant integer value larger than 0;
2109 elementtype may be any integer, floating point or pointer type. Vectors
2110 of size zero are not allowed.
2114 +-------------------+--------------------------------------------------+
2115 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2116 +-------------------+--------------------------------------------------+
2117 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2118 +-------------------+--------------------------------------------------+
2119 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2120 +-------------------+--------------------------------------------------+
2121 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2122 +-------------------+--------------------------------------------------+
2131 The label type represents code labels.
2146 The metadata type represents embedded metadata. No derived types may be
2147 created from metadata except for :ref:`function <t_function>` arguments.
2160 Aggregate Types are a subset of derived types that can contain multiple
2161 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2162 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2172 The array type is a very simple derived type that arranges elements
2173 sequentially in memory. The array type requires a size (number of
2174 elements) and an underlying data type.
2180 [<# elements> x <elementtype>]
2182 The number of elements is a constant integer value; ``elementtype`` may
2183 be any type with a size.
2187 +------------------+--------------------------------------+
2188 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2189 +------------------+--------------------------------------+
2190 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2191 +------------------+--------------------------------------+
2192 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2193 +------------------+--------------------------------------+
2195 Here are some examples of multidimensional arrays:
2197 +-----------------------------+----------------------------------------------------------+
2198 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2199 +-----------------------------+----------------------------------------------------------+
2200 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2201 +-----------------------------+----------------------------------------------------------+
2202 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2203 +-----------------------------+----------------------------------------------------------+
2205 There is no restriction on indexing beyond the end of the array implied
2206 by a static type (though there are restrictions on indexing beyond the
2207 bounds of an allocated object in some cases). This means that
2208 single-dimension 'variable sized array' addressing can be implemented in
2209 LLVM with a zero length array type. An implementation of 'pascal style
2210 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2220 The structure type is used to represent a collection of data members
2221 together in memory. The elements of a structure may be any type that has
2224 Structures in memory are accessed using '``load``' and '``store``' by
2225 getting a pointer to a field with the '``getelementptr``' instruction.
2226 Structures in registers are accessed using the '``extractvalue``' and
2227 '``insertvalue``' instructions.
2229 Structures may optionally be "packed" structures, which indicate that
2230 the alignment of the struct is one byte, and that there is no padding
2231 between the elements. In non-packed structs, padding between field types
2232 is inserted as defined by the DataLayout string in the module, which is
2233 required to match what the underlying code generator expects.
2235 Structures can either be "literal" or "identified". A literal structure
2236 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2237 identified types are always defined at the top level with a name.
2238 Literal types are uniqued by their contents and can never be recursive
2239 or opaque since there is no way to write one. Identified types can be
2240 recursive, can be opaqued, and are never uniqued.
2246 %T1 = type { <type list> } ; Identified normal struct type
2247 %T2 = type <{ <type list> }> ; Identified packed struct type
2251 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2252 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2253 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2254 | ``{ 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``. |
2255 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2256 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2257 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2261 Opaque Structure Types
2262 """"""""""""""""""""""
2266 Opaque structure types are used to represent named structure types that
2267 do not have a body specified. This corresponds (for example) to the C
2268 notion of a forward declared structure.
2279 +--------------+-------------------+
2280 | ``opaque`` | An opaque type. |
2281 +--------------+-------------------+
2288 LLVM has several different basic types of constants. This section
2289 describes them all and their syntax.
2294 **Boolean constants**
2295 The two strings '``true``' and '``false``' are both valid constants
2297 **Integer constants**
2298 Standard integers (such as '4') are constants of the
2299 :ref:`integer <t_integer>` type. Negative numbers may be used with
2301 **Floating point constants**
2302 Floating point constants use standard decimal notation (e.g.
2303 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2304 hexadecimal notation (see below). The assembler requires the exact
2305 decimal value of a floating-point constant. For example, the
2306 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2307 decimal in binary. Floating point constants must have a :ref:`floating
2308 point <t_floating>` type.
2309 **Null pointer constants**
2310 The identifier '``null``' is recognized as a null pointer constant
2311 and must be of :ref:`pointer type <t_pointer>`.
2313 The one non-intuitive notation for constants is the hexadecimal form of
2314 floating point constants. For example, the form
2315 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2316 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2317 constants are required (and the only time that they are generated by the
2318 disassembler) is when a floating point constant must be emitted but it
2319 cannot be represented as a decimal floating point number in a reasonable
2320 number of digits. For example, NaN's, infinities, and other special
2321 values are represented in their IEEE hexadecimal format so that assembly
2322 and disassembly do not cause any bits to change in the constants.
2324 When using the hexadecimal form, constants of types half, float, and
2325 double are represented using the 16-digit form shown above (which
2326 matches the IEEE754 representation for double); half and float values
2327 must, however, be exactly representable as IEEE 754 half and single
2328 precision, respectively. Hexadecimal format is always used for long
2329 double, and there are three forms of long double. The 80-bit format used
2330 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2331 128-bit format used by PowerPC (two adjacent doubles) is represented by
2332 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2333 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2334 will only work if they match the long double format on your target.
2335 The IEEE 16-bit format (half precision) is represented by ``0xH``
2336 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2337 (sign bit at the left).
2339 There are no constants of type x86_mmx.
2341 .. _complexconstants:
2346 Complex constants are a (potentially recursive) combination of simple
2347 constants and smaller complex constants.
2349 **Structure constants**
2350 Structure constants are represented with notation similar to
2351 structure type definitions (a comma separated list of elements,
2352 surrounded by braces (``{}``)). For example:
2353 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2354 "``@G = external global i32``". Structure constants must have
2355 :ref:`structure type <t_struct>`, and the number and types of elements
2356 must match those specified by the type.
2358 Array constants are represented with notation similar to array type
2359 definitions (a comma separated list of elements, surrounded by
2360 square brackets (``[]``)). For example:
2361 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2362 :ref:`array type <t_array>`, and the number and types of elements must
2363 match those specified by the type. As a special case, character array
2364 constants may also be represented as a double-quoted string using the ``c``
2365 prefix. For example: "``c"Hello World\0A\00"``".
2366 **Vector constants**
2367 Vector constants are represented with notation similar to vector
2368 type definitions (a comma separated list of elements, surrounded by
2369 less-than/greater-than's (``<>``)). For example:
2370 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2371 must have :ref:`vector type <t_vector>`, and the number and types of
2372 elements must match those specified by the type.
2373 **Zero initialization**
2374 The string '``zeroinitializer``' can be used to zero initialize a
2375 value to zero of *any* type, including scalar and
2376 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2377 having to print large zero initializers (e.g. for large arrays) and
2378 is always exactly equivalent to using explicit zero initializers.
2380 A metadata node is a constant tuple without types. For example:
2381 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2382 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2383 Unlike other typed constants that are meant to be interpreted as part of
2384 the instruction stream, metadata is a place to attach additional
2385 information such as debug info.
2387 Global Variable and Function Addresses
2388 --------------------------------------
2390 The addresses of :ref:`global variables <globalvars>` and
2391 :ref:`functions <functionstructure>` are always implicitly valid
2392 (link-time) constants. These constants are explicitly referenced when
2393 the :ref:`identifier for the global <identifiers>` is used and always have
2394 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2397 .. code-block:: llvm
2401 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2408 The string '``undef``' can be used anywhere a constant is expected, and
2409 indicates that the user of the value may receive an unspecified
2410 bit-pattern. Undefined values may be of any type (other than '``label``'
2411 or '``void``') and be used anywhere a constant is permitted.
2413 Undefined values are useful because they indicate to the compiler that
2414 the program is well defined no matter what value is used. This gives the
2415 compiler more freedom to optimize. Here are some examples of
2416 (potentially surprising) transformations that are valid (in pseudo IR):
2418 .. code-block:: llvm
2428 This is safe because all of the output bits are affected by the undef
2429 bits. Any output bit can have a zero or one depending on the input bits.
2431 .. code-block:: llvm
2442 These logical operations have bits that are not always affected by the
2443 input. For example, if ``%X`` has a zero bit, then the output of the
2444 '``and``' operation will always be a zero for that bit, no matter what
2445 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2446 optimize or assume that the result of the '``and``' is '``undef``'.
2447 However, it is safe to assume that all bits of the '``undef``' could be
2448 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2449 all the bits of the '``undef``' operand to the '``or``' could be set,
2450 allowing the '``or``' to be folded to -1.
2452 .. code-block:: llvm
2454 %A = select undef, %X, %Y
2455 %B = select undef, 42, %Y
2456 %C = select %X, %Y, undef
2466 This set of examples shows that undefined '``select``' (and conditional
2467 branch) conditions can go *either way*, but they have to come from one
2468 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2469 both known to have a clear low bit, then ``%A`` would have to have a
2470 cleared low bit. However, in the ``%C`` example, the optimizer is
2471 allowed to assume that the '``undef``' operand could be the same as
2472 ``%Y``, allowing the whole '``select``' to be eliminated.
2474 .. code-block:: llvm
2476 %A = xor undef, undef
2493 This example points out that two '``undef``' operands are not
2494 necessarily the same. This can be surprising to people (and also matches
2495 C semantics) where they assume that "``X^X``" is always zero, even if
2496 ``X`` is undefined. This isn't true for a number of reasons, but the
2497 short answer is that an '``undef``' "variable" can arbitrarily change
2498 its value over its "live range". This is true because the variable
2499 doesn't actually *have a live range*. Instead, the value is logically
2500 read from arbitrary registers that happen to be around when needed, so
2501 the value is not necessarily consistent over time. In fact, ``%A`` and
2502 ``%C`` need to have the same semantics or the core LLVM "replace all
2503 uses with" concept would not hold.
2505 .. code-block:: llvm
2513 These examples show the crucial difference between an *undefined value*
2514 and *undefined behavior*. An undefined value (like '``undef``') is
2515 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2516 operation can be constant folded to '``undef``', because the '``undef``'
2517 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2518 However, in the second example, we can make a more aggressive
2519 assumption: because the ``undef`` is allowed to be an arbitrary value,
2520 we are allowed to assume that it could be zero. Since a divide by zero
2521 has *undefined behavior*, we are allowed to assume that the operation
2522 does not execute at all. This allows us to delete the divide and all
2523 code after it. Because the undefined operation "can't happen", the
2524 optimizer can assume that it occurs in dead code.
2526 .. code-block:: llvm
2528 a: store undef -> %X
2529 b: store %X -> undef
2534 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2535 value can be assumed to not have any effect; we can assume that the
2536 value is overwritten with bits that happen to match what was already
2537 there. However, a store *to* an undefined location could clobber
2538 arbitrary memory, therefore, it has undefined behavior.
2545 Poison values are similar to :ref:`undef values <undefvalues>`, however
2546 they also represent the fact that an instruction or constant expression
2547 that cannot evoke side effects has nevertheless detected a condition
2548 that results in undefined behavior.
2550 There is currently no way of representing a poison value in the IR; they
2551 only exist when produced by operations such as :ref:`add <i_add>` with
2554 Poison value behavior is defined in terms of value *dependence*:
2556 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2557 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2558 their dynamic predecessor basic block.
2559 - Function arguments depend on the corresponding actual argument values
2560 in the dynamic callers of their functions.
2561 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2562 instructions that dynamically transfer control back to them.
2563 - :ref:`Invoke <i_invoke>` instructions depend on the
2564 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2565 call instructions that dynamically transfer control back to them.
2566 - Non-volatile loads and stores depend on the most recent stores to all
2567 of the referenced memory addresses, following the order in the IR
2568 (including loads and stores implied by intrinsics such as
2569 :ref:`@llvm.memcpy <int_memcpy>`.)
2570 - An instruction with externally visible side effects depends on the
2571 most recent preceding instruction with externally visible side
2572 effects, following the order in the IR. (This includes :ref:`volatile
2573 operations <volatile>`.)
2574 - An instruction *control-depends* on a :ref:`terminator
2575 instruction <terminators>` if the terminator instruction has
2576 multiple successors and the instruction is always executed when
2577 control transfers to one of the successors, and may not be executed
2578 when control is transferred to another.
2579 - Additionally, an instruction also *control-depends* on a terminator
2580 instruction if the set of instructions it otherwise depends on would
2581 be different if the terminator had transferred control to a different
2583 - Dependence is transitive.
2585 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2586 with the additional effect that any instruction that has a *dependence*
2587 on a poison value has undefined behavior.
2589 Here are some examples:
2591 .. code-block:: llvm
2594 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2595 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2596 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2597 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2599 store i32 %poison, i32* @g ; Poison value stored to memory.
2600 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2602 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2604 %narrowaddr = bitcast i32* @g to i16*
2605 %wideaddr = bitcast i32* @g to i64*
2606 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2607 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2609 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2610 br i1 %cmp, label %true, label %end ; Branch to either destination.
2613 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2614 ; it has undefined behavior.
2618 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2619 ; Both edges into this PHI are
2620 ; control-dependent on %cmp, so this
2621 ; always results in a poison value.
2623 store volatile i32 0, i32* @g ; This would depend on the store in %true
2624 ; if %cmp is true, or the store in %entry
2625 ; otherwise, so this is undefined behavior.
2627 br i1 %cmp, label %second_true, label %second_end
2628 ; The same branch again, but this time the
2629 ; true block doesn't have side effects.
2636 store volatile i32 0, i32* @g ; This time, the instruction always depends
2637 ; on the store in %end. Also, it is
2638 ; control-equivalent to %end, so this is
2639 ; well-defined (ignoring earlier undefined
2640 ; behavior in this example).
2644 Addresses of Basic Blocks
2645 -------------------------
2647 ``blockaddress(@function, %block)``
2649 The '``blockaddress``' constant computes the address of the specified
2650 basic block in the specified function, and always has an ``i8*`` type.
2651 Taking the address of the entry block is illegal.
2653 This value only has defined behavior when used as an operand to the
2654 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2655 against null. Pointer equality tests between labels addresses results in
2656 undefined behavior --- though, again, comparison against null is ok, and
2657 no label is equal to the null pointer. This may be passed around as an
2658 opaque pointer sized value as long as the bits are not inspected. This
2659 allows ``ptrtoint`` and arithmetic to be performed on these values so
2660 long as the original value is reconstituted before the ``indirectbr``
2663 Finally, some targets may provide defined semantics when using the value
2664 as the operand to an inline assembly, but that is target specific.
2668 Constant Expressions
2669 --------------------
2671 Constant expressions are used to allow expressions involving other
2672 constants to be used as constants. Constant expressions may be of any
2673 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2674 that does not have side effects (e.g. load and call are not supported).
2675 The following is the syntax for constant expressions:
2677 ``trunc (CST to TYPE)``
2678 Truncate a constant to another type. The bit size of CST must be
2679 larger than the bit size of TYPE. Both types must be integers.
2680 ``zext (CST to TYPE)``
2681 Zero extend a constant to another type. The bit size of CST must be
2682 smaller than the bit size of TYPE. Both types must be integers.
2683 ``sext (CST to TYPE)``
2684 Sign extend a constant to another type. The bit size of CST must be
2685 smaller than the bit size of TYPE. Both types must be integers.
2686 ``fptrunc (CST to TYPE)``
2687 Truncate a floating point constant to another floating point type.
2688 The size of CST must be larger than the size of TYPE. Both types
2689 must be floating point.
2690 ``fpext (CST to TYPE)``
2691 Floating point extend a constant to another type. The size of CST
2692 must be smaller or equal to the size of TYPE. Both types must be
2694 ``fptoui (CST to TYPE)``
2695 Convert a floating point constant to the corresponding unsigned
2696 integer constant. TYPE must be a scalar or vector integer type. CST
2697 must be of scalar or vector floating point type. Both CST and TYPE
2698 must be scalars, or vectors of the same number of elements. If the
2699 value won't fit in the integer type, the results are undefined.
2700 ``fptosi (CST to TYPE)``
2701 Convert a floating point constant to the corresponding signed
2702 integer constant. TYPE must be a scalar or vector integer type. CST
2703 must be of scalar or vector floating point type. Both CST and TYPE
2704 must be scalars, or vectors of the same number of elements. If the
2705 value won't fit in the integer type, the results are undefined.
2706 ``uitofp (CST to TYPE)``
2707 Convert an unsigned integer constant to the corresponding floating
2708 point constant. TYPE must be a scalar or vector floating point type.
2709 CST must be of scalar or vector integer type. Both CST and TYPE must
2710 be scalars, or vectors of the same number of elements. If the value
2711 won't fit in the floating point type, the results are undefined.
2712 ``sitofp (CST to TYPE)``
2713 Convert a signed integer constant to the corresponding floating
2714 point constant. TYPE must be a scalar or vector floating point type.
2715 CST must be of scalar or vector integer type. Both CST and TYPE must
2716 be scalars, or vectors of the same number of elements. If the value
2717 won't fit in the floating point type, the results are undefined.
2718 ``ptrtoint (CST to TYPE)``
2719 Convert a pointer typed constant to the corresponding integer
2720 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2721 pointer type. The ``CST`` value is zero extended, truncated, or
2722 unchanged to make it fit in ``TYPE``.
2723 ``inttoptr (CST to TYPE)``
2724 Convert an integer constant to a pointer constant. TYPE must be a
2725 pointer type. CST must be of integer type. The CST value is zero
2726 extended, truncated, or unchanged to make it fit in a pointer size.
2727 This one is *really* dangerous!
2728 ``bitcast (CST to TYPE)``
2729 Convert a constant, CST, to another TYPE. The constraints of the
2730 operands are the same as those for the :ref:`bitcast
2731 instruction <i_bitcast>`.
2732 ``addrspacecast (CST to TYPE)``
2733 Convert a constant pointer or constant vector of pointer, CST, to another
2734 TYPE in a different address space. The constraints of the operands are the
2735 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2736 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2737 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2738 constants. As with the :ref:`getelementptr <i_getelementptr>`
2739 instruction, the index list may have zero or more indexes, which are
2740 required to make sense for the type of "pointer to TY".
2741 ``select (COND, VAL1, VAL2)``
2742 Perform the :ref:`select operation <i_select>` on constants.
2743 ``icmp COND (VAL1, VAL2)``
2744 Performs the :ref:`icmp operation <i_icmp>` on constants.
2745 ``fcmp COND (VAL1, VAL2)``
2746 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2747 ``extractelement (VAL, IDX)``
2748 Perform the :ref:`extractelement operation <i_extractelement>` on
2750 ``insertelement (VAL, ELT, IDX)``
2751 Perform the :ref:`insertelement operation <i_insertelement>` on
2753 ``shufflevector (VEC1, VEC2, IDXMASK)``
2754 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2756 ``extractvalue (VAL, IDX0, IDX1, ...)``
2757 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2758 constants. The index list is interpreted in a similar manner as
2759 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2760 least one index value must be specified.
2761 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2762 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2763 The index list is interpreted in a similar manner as indices in a
2764 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2765 value must be specified.
2766 ``OPCODE (LHS, RHS)``
2767 Perform the specified operation of the LHS and RHS constants. OPCODE
2768 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2769 binary <bitwiseops>` operations. The constraints on operands are
2770 the same as those for the corresponding instruction (e.g. no bitwise
2771 operations on floating point values are allowed).
2778 Inline Assembler Expressions
2779 ----------------------------
2781 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2782 Inline Assembly <moduleasm>`) through the use of a special value. This
2783 value represents the inline assembler as a string (containing the
2784 instructions to emit), a list of operand constraints (stored as a
2785 string), a flag that indicates whether or not the inline asm expression
2786 has side effects, and a flag indicating whether the function containing
2787 the asm needs to align its stack conservatively. An example inline
2788 assembler expression is:
2790 .. code-block:: llvm
2792 i32 (i32) asm "bswap $0", "=r,r"
2794 Inline assembler expressions may **only** be used as the callee operand
2795 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2796 Thus, typically we have:
2798 .. code-block:: llvm
2800 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2802 Inline asms with side effects not visible in the constraint list must be
2803 marked as having side effects. This is done through the use of the
2804 '``sideeffect``' keyword, like so:
2806 .. code-block:: llvm
2808 call void asm sideeffect "eieio", ""()
2810 In some cases inline asms will contain code that will not work unless
2811 the stack is aligned in some way, such as calls or SSE instructions on
2812 x86, yet will not contain code that does that alignment within the asm.
2813 The compiler should make conservative assumptions about what the asm
2814 might contain and should generate its usual stack alignment code in the
2815 prologue if the '``alignstack``' keyword is present:
2817 .. code-block:: llvm
2819 call void asm alignstack "eieio", ""()
2821 Inline asms also support using non-standard assembly dialects. The
2822 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2823 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2824 the only supported dialects. An example is:
2826 .. code-block:: llvm
2828 call void asm inteldialect "eieio", ""()
2830 If multiple keywords appear the '``sideeffect``' keyword must come
2831 first, the '``alignstack``' keyword second and the '``inteldialect``'
2837 The call instructions that wrap inline asm nodes may have a
2838 "``!srcloc``" MDNode attached to it that contains a list of constant
2839 integers. If present, the code generator will use the integer as the
2840 location cookie value when report errors through the ``LLVMContext``
2841 error reporting mechanisms. This allows a front-end to correlate backend
2842 errors that occur with inline asm back to the source code that produced
2845 .. code-block:: llvm
2847 call void asm sideeffect "something bad", ""(), !srcloc !42
2849 !42 = !{ i32 1234567 }
2851 It is up to the front-end to make sense of the magic numbers it places
2852 in the IR. If the MDNode contains multiple constants, the code generator
2853 will use the one that corresponds to the line of the asm that the error
2861 LLVM IR allows metadata to be attached to instructions in the program
2862 that can convey extra information about the code to the optimizers and
2863 code generator. One example application of metadata is source-level
2864 debug information. There are two metadata primitives: strings and nodes.
2866 Metadata does not have a type, and is not a value. If referenced from a
2867 ``call`` instruction, it uses the ``metadata`` type.
2869 All metadata are identified in syntax by a exclamation point ('``!``').
2871 .. _metadata-string:
2873 Metadata Nodes and Metadata Strings
2874 -----------------------------------
2876 A metadata string is a string surrounded by double quotes. It can
2877 contain any character by escaping non-printable characters with
2878 "``\xx``" where "``xx``" is the two digit hex code. For example:
2881 Metadata nodes are represented with notation similar to structure
2882 constants (a comma separated list of elements, surrounded by braces and
2883 preceded by an exclamation point). Metadata nodes can have any values as
2884 their operand. For example:
2886 .. code-block:: llvm
2888 !{ !"test\00", i32 10}
2890 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
2892 .. code-block:: llvm
2894 !0 = distinct !{!"test\00", i32 10}
2896 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
2897 content. They can also occur when transformations cause uniquing collisions
2898 when metadata operands change.
2900 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2901 metadata nodes, which can be looked up in the module symbol table. For
2904 .. code-block:: llvm
2908 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2909 function is using two metadata arguments:
2911 .. code-block:: llvm
2913 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2915 Metadata can be attached with an instruction. Here metadata ``!21`` is
2916 attached to the ``add`` instruction using the ``!dbg`` identifier:
2918 .. code-block:: llvm
2920 %indvar.next = add i64 %indvar, 1, !dbg !21
2922 More information about specific metadata nodes recognized by the
2923 optimizers and code generator is found below.
2925 .. _specialized-metadata:
2927 Specialized Metadata Nodes
2928 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2930 Specialized metadata nodes are custom data structures in metadata (as opposed
2931 to generic tuples). Their fields are labelled, and can be specified in any
2934 These aren't inherently debug info centric, but currently all the specialized
2935 metadata nodes are related to debug info.
2942 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
2943 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
2944 tuples containing the debug info to be emitted along with the compile unit,
2945 regardless of code optimizations (some nodes are only emitted if there are
2946 references to them from instructions).
2948 .. code-block:: llvm
2950 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
2951 isOptimized: true, flags: "-O2", runtimeVersion: 2,
2952 splitDebugFilename: "abc.debug", emissionKind: 1,
2953 enums: !2, retainedTypes: !3, subprograms: !4,
2954 globals: !5, imports: !6)
2956 Compile unit descriptors provide the root scope for objects declared in a
2957 specific compilation unit. File descriptors are defined using this scope.
2958 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
2959 keep track of subprograms, global variables, type information, and imported
2960 entities (declarations and namespaces).
2967 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
2969 .. code-block:: llvm
2971 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
2973 Files are sometimes used in ``scope:`` fields, and are the only valid target
2974 for ``file:`` fields.
2981 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
2982 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
2984 .. code-block:: llvm
2986 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
2987 encoding: DW_ATE_unsigned_char)
2988 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
2990 The ``encoding:`` describes the details of the type. Usually it's one of the
2993 .. code-block:: llvm
2999 DW_ATE_signed_char = 6
3001 DW_ATE_unsigned_char = 8
3003 .. _DISubroutineType:
3008 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3009 refers to a tuple; the first operand is the return type, while the rest are the
3010 types of the formal arguments in order. If the first operand is ``null``, that
3011 represents a function with no return value (such as ``void foo() {}`` in C++).
3013 .. code-block:: llvm
3015 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3016 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3017 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3024 ``DIDerivedType`` nodes represent types derived from other types, such as
3027 .. code-block:: llvm
3029 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3030 encoding: DW_ATE_unsigned_char)
3031 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3034 The following ``tag:`` values are valid:
3036 .. code-block:: llvm
3038 DW_TAG_formal_parameter = 5
3040 DW_TAG_pointer_type = 15
3041 DW_TAG_reference_type = 16
3043 DW_TAG_ptr_to_member_type = 31
3044 DW_TAG_const_type = 38
3045 DW_TAG_volatile_type = 53
3046 DW_TAG_restrict_type = 55
3048 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3049 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3050 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3051 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3052 argument of a subprogram.
3054 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3056 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3057 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3060 Note that the ``void *`` type is expressed as a type derived from NULL.
3062 .. _DICompositeType:
3067 ``DICompositeType`` nodes represent types composed of other types, like
3068 structures and unions. ``elements:`` points to a tuple of the composed types.
3070 If the source language supports ODR, the ``identifier:`` field gives the unique
3071 identifier used for type merging between modules. When specified, other types
3072 can refer to composite types indirectly via a :ref:`metadata string
3073 <metadata-string>` that matches their identifier.
3075 .. code-block:: llvm
3077 !0 = !DIEnumerator(name: "SixKind", value: 7)
3078 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3079 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3080 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3081 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3082 elements: !{!0, !1, !2})
3084 The following ``tag:`` values are valid:
3086 .. code-block:: llvm
3088 DW_TAG_array_type = 1
3089 DW_TAG_class_type = 2
3090 DW_TAG_enumeration_type = 4
3091 DW_TAG_structure_type = 19
3092 DW_TAG_union_type = 23
3093 DW_TAG_subroutine_type = 21
3094 DW_TAG_inheritance = 28
3097 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3098 descriptors <DISubrange>`, each representing the range of subscripts at that
3099 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3100 array type is a native packed vector.
3102 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3103 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3104 value for the set. All enumeration type descriptors are collected in the
3105 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3107 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3108 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3109 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3116 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3117 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3119 .. code-block:: llvm
3121 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3122 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3123 !2 = !DISubrange(count: -1) ; empty array.
3130 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3131 variants of :ref:`DICompositeType`.
3133 .. code-block:: llvm
3135 !0 = !DIEnumerator(name: "SixKind", value: 7)
3136 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3137 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3139 DITemplateTypeParameter
3140 """""""""""""""""""""""
3142 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3143 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3144 :ref:`DISubprogram` ``templateParams:`` fields.
3146 .. code-block:: llvm
3148 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3150 DITemplateValueParameter
3151 """"""""""""""""""""""""
3153 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3154 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3155 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3156 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3157 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3159 .. code-block:: llvm
3161 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3166 ``DINamespace`` nodes represent namespaces in the source language.
3168 .. code-block:: llvm
3170 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3175 ``DIGlobalVariable`` nodes represent global variables in the source language.
3177 .. code-block:: llvm
3179 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3180 file: !2, line: 7, type: !3, isLocal: true,
3181 isDefinition: false, variable: i32* @foo,
3184 All global variables should be referenced by the `globals:` field of a
3185 :ref:`compile unit <DICompileUnit>`.
3192 ``DISubprogram`` nodes represent functions from the source language. The
3193 ``variables:`` field points at :ref:`variables <DILocalVariable>` that must be
3194 retained, even if their IR counterparts are optimized out of the IR. The
3195 ``type:`` field must point at an :ref:`DISubroutineType`.
3197 .. code-block:: llvm
3199 !0 = !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3200 file: !2, line: 7, type: !3, isLocal: true,
3201 isDefinition: false, scopeLine: 8, containingType: !4,
3202 virtuality: DW_VIRTUALITY_pure_virtual, virtualIndex: 10,
3203 flags: DIFlagPrototyped, isOptimized: true,
3204 function: void ()* @_Z3foov,
3205 templateParams: !5, declaration: !6, variables: !7)
3212 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3213 <DISubprogram>`. The line number and column numbers are used to dinstinguish
3214 two lexical blocks at same depth. They are valid targets for ``scope:``
3217 .. code-block:: llvm
3219 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3221 Usually lexical blocks are ``distinct`` to prevent node merging based on
3224 .. _DILexicalBlockFile:
3229 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3230 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
3231 indicate textual inclusion, or the ``discriminator:`` field can be used to
3232 discriminate between control flow within a single block in the source language.
3234 .. code-block:: llvm
3236 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3237 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3238 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3245 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
3246 mandatory, and points at an :ref:`DILexicalBlockFile`, an
3247 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
3249 .. code-block:: llvm
3251 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3253 .. _DILocalVariable:
3258 ``DILocalVariable`` nodes represent local variables in the source language.
3259 Instead of ``DW_TAG_variable``, they use LLVM-specific fake tags to
3260 discriminate between local variables (``DW_TAG_auto_variable``) and subprogram
3261 arguments (``DW_TAG_arg_variable``). In the latter case, the ``arg:`` field
3262 specifies the argument position, and this variable will be included in the
3263 ``variables:`` field of its :ref:`DISubprogram`.
3265 .. code-block:: llvm
3267 !0 = !DILocalVariable(tag: DW_TAG_arg_variable, name: "this", arg: 0,
3268 scope: !3, file: !2, line: 7, type: !3,
3269 flags: DIFlagArtificial)
3270 !1 = !DILocalVariable(tag: DW_TAG_arg_variable, name: "x", arg: 1,
3271 scope: !4, file: !2, line: 7, type: !3)
3272 !1 = !DILocalVariable(tag: DW_TAG_auto_variable, name: "y",
3273 scope: !5, file: !2, line: 7, type: !3)
3278 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
3279 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
3280 describe how the referenced LLVM variable relates to the source language
3283 The current supported vocabulary is limited:
3285 - ``DW_OP_deref`` dereferences the working expression.
3286 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
3287 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
3288 here, respectively) of the variable piece from the working expression.
3290 .. code-block:: llvm
3292 !0 = !DIExpression(DW_OP_deref)
3293 !1 = !DIExpression(DW_OP_plus, 3)
3294 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
3295 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
3300 ``DIObjCProperty`` nodes represent Objective-C property nodes.
3302 .. code-block:: llvm
3304 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
3305 getter: "getFoo", attributes: 7, type: !2)
3310 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
3313 .. code-block:: llvm
3315 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
3316 entity: !1, line: 7)
3321 In LLVM IR, memory does not have types, so LLVM's own type system is not
3322 suitable for doing TBAA. Instead, metadata is added to the IR to
3323 describe a type system of a higher level language. This can be used to
3324 implement typical C/C++ TBAA, but it can also be used to implement
3325 custom alias analysis behavior for other languages.
3327 The current metadata format is very simple. TBAA metadata nodes have up
3328 to three fields, e.g.:
3330 .. code-block:: llvm
3332 !0 = !{ !"an example type tree" }
3333 !1 = !{ !"int", !0 }
3334 !2 = !{ !"float", !0 }
3335 !3 = !{ !"const float", !2, i64 1 }
3337 The first field is an identity field. It can be any value, usually a
3338 metadata string, which uniquely identifies the type. The most important
3339 name in the tree is the name of the root node. Two trees with different
3340 root node names are entirely disjoint, even if they have leaves with
3343 The second field identifies the type's parent node in the tree, or is
3344 null or omitted for a root node. A type is considered to alias all of
3345 its descendants and all of its ancestors in the tree. Also, a type is
3346 considered to alias all types in other trees, so that bitcode produced
3347 from multiple front-ends is handled conservatively.
3349 If the third field is present, it's an integer which if equal to 1
3350 indicates that the type is "constant" (meaning
3351 ``pointsToConstantMemory`` should return true; see `other useful
3352 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
3354 '``tbaa.struct``' Metadata
3355 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3357 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
3358 aggregate assignment operations in C and similar languages, however it
3359 is defined to copy a contiguous region of memory, which is more than
3360 strictly necessary for aggregate types which contain holes due to
3361 padding. Also, it doesn't contain any TBAA information about the fields
3364 ``!tbaa.struct`` metadata can describe which memory subregions in a
3365 memcpy are padding and what the TBAA tags of the struct are.
3367 The current metadata format is very simple. ``!tbaa.struct`` metadata
3368 nodes are a list of operands which are in conceptual groups of three.
3369 For each group of three, the first operand gives the byte offset of a
3370 field in bytes, the second gives its size in bytes, and the third gives
3373 .. code-block:: llvm
3375 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
3377 This describes a struct with two fields. The first is at offset 0 bytes
3378 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
3379 and has size 4 bytes and has tbaa tag !2.
3381 Note that the fields need not be contiguous. In this example, there is a
3382 4 byte gap between the two fields. This gap represents padding which
3383 does not carry useful data and need not be preserved.
3385 '``noalias``' and '``alias.scope``' Metadata
3386 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3388 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
3389 noalias memory-access sets. This means that some collection of memory access
3390 instructions (loads, stores, memory-accessing calls, etc.) that carry
3391 ``noalias`` metadata can specifically be specified not to alias with some other
3392 collection of memory access instructions that carry ``alias.scope`` metadata.
3393 Each type of metadata specifies a list of scopes where each scope has an id and
3394 a domain. When evaluating an aliasing query, if for some domain, the set
3395 of scopes with that domain in one instruction's ``alias.scope`` list is a
3396 subset of (or equal to) the set of scopes for that domain in another
3397 instruction's ``noalias`` list, then the two memory accesses are assumed not to
3400 The metadata identifying each domain is itself a list containing one or two
3401 entries. The first entry is the name of the domain. Note that if the name is a
3402 string then it can be combined accross functions and translation units. A
3403 self-reference can be used to create globally unique domain names. A
3404 descriptive string may optionally be provided as a second list entry.
3406 The metadata identifying each scope is also itself a list containing two or
3407 three entries. The first entry is the name of the scope. Note that if the name
3408 is a string then it can be combined accross functions and translation units. A
3409 self-reference can be used to create globally unique scope names. A metadata
3410 reference to the scope's domain is the second entry. A descriptive string may
3411 optionally be provided as a third list entry.
3415 .. code-block:: llvm
3417 ; Two scope domains:
3421 ; Some scopes in these domains:
3427 !5 = !{!4} ; A list containing only scope !4
3431 ; These two instructions don't alias:
3432 %0 = load float, float* %c, align 4, !alias.scope !5
3433 store float %0, float* %arrayidx.i, align 4, !noalias !5
3435 ; These two instructions also don't alias (for domain !1, the set of scopes
3436 ; in the !alias.scope equals that in the !noalias list):
3437 %2 = load float, float* %c, align 4, !alias.scope !5
3438 store float %2, float* %arrayidx.i2, align 4, !noalias !6
3440 ; These two instructions may alias (for domain !0, the set of scopes in
3441 ; the !noalias list is not a superset of, or equal to, the scopes in the
3442 ; !alias.scope list):
3443 %2 = load float, float* %c, align 4, !alias.scope !6
3444 store float %0, float* %arrayidx.i, align 4, !noalias !7
3446 '``fpmath``' Metadata
3447 ^^^^^^^^^^^^^^^^^^^^^
3449 ``fpmath`` metadata may be attached to any instruction of floating point
3450 type. It can be used to express the maximum acceptable error in the
3451 result of that instruction, in ULPs, thus potentially allowing the
3452 compiler to use a more efficient but less accurate method of computing
3453 it. ULP is defined as follows:
3455 If ``x`` is a real number that lies between two finite consecutive
3456 floating-point numbers ``a`` and ``b``, without being equal to one
3457 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
3458 distance between the two non-equal finite floating-point numbers
3459 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
3461 The metadata node shall consist of a single positive floating point
3462 number representing the maximum relative error, for example:
3464 .. code-block:: llvm
3466 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3470 '``range``' Metadata
3471 ^^^^^^^^^^^^^^^^^^^^
3473 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
3474 integer types. It expresses the possible ranges the loaded value or the value
3475 returned by the called function at this call site is in. The ranges are
3476 represented with a flattened list of integers. The loaded value or the value
3477 returned is known to be in the union of the ranges defined by each consecutive
3478 pair. Each pair has the following properties:
3480 - The type must match the type loaded by the instruction.
3481 - The pair ``a,b`` represents the range ``[a,b)``.
3482 - Both ``a`` and ``b`` are constants.
3483 - The range is allowed to wrap.
3484 - The range should not represent the full or empty set. That is,
3487 In addition, the pairs must be in signed order of the lower bound and
3488 they must be non-contiguous.
3492 .. code-block:: llvm
3494 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
3495 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3496 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
3497 %d = invoke i8 @bar() to label %cont
3498 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
3500 !0 = !{ i8 0, i8 2 }
3501 !1 = !{ i8 255, i8 2 }
3502 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
3503 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
3508 It is sometimes useful to attach information to loop constructs. Currently,
3509 loop metadata is implemented as metadata attached to the branch instruction
3510 in the loop latch block. This type of metadata refer to a metadata node that is
3511 guaranteed to be separate for each loop. The loop identifier metadata is
3512 specified with the name ``llvm.loop``.
3514 The loop identifier metadata is implemented using a metadata that refers to
3515 itself to avoid merging it with any other identifier metadata, e.g.,
3516 during module linkage or function inlining. That is, each loop should refer
3517 to their own identification metadata even if they reside in separate functions.
3518 The following example contains loop identifier metadata for two separate loop
3521 .. code-block:: llvm
3526 The loop identifier metadata can be used to specify additional
3527 per-loop metadata. Any operands after the first operand can be treated
3528 as user-defined metadata. For example the ``llvm.loop.unroll.count``
3529 suggests an unroll factor to the loop unroller:
3531 .. code-block:: llvm
3533 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
3536 !1 = !{!"llvm.loop.unroll.count", i32 4}
3538 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
3539 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3541 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
3542 used to control per-loop vectorization and interleaving parameters such as
3543 vectorization width and interleave count. These metadata should be used in
3544 conjunction with ``llvm.loop`` loop identification metadata. The
3545 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
3546 optimization hints and the optimizer will only interleave and vectorize loops if
3547 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
3548 which contains information about loop-carried memory dependencies can be helpful
3549 in determining the safety of these transformations.
3551 '``llvm.loop.interleave.count``' Metadata
3552 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3554 This metadata suggests an interleave count to the loop interleaver.
3555 The first operand is the string ``llvm.loop.interleave.count`` and the
3556 second operand is an integer specifying the interleave count. For
3559 .. code-block:: llvm
3561 !0 = !{!"llvm.loop.interleave.count", i32 4}
3563 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
3564 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
3565 then the interleave count will be determined automatically.
3567 '``llvm.loop.vectorize.enable``' Metadata
3568 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3570 This metadata selectively enables or disables vectorization for the loop. The
3571 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
3572 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
3573 0 disables vectorization:
3575 .. code-block:: llvm
3577 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
3578 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
3580 '``llvm.loop.vectorize.width``' Metadata
3581 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3583 This metadata sets the target width of the vectorizer. The first
3584 operand is the string ``llvm.loop.vectorize.width`` and the second
3585 operand is an integer specifying the width. For example:
3587 .. code-block:: llvm
3589 !0 = !{!"llvm.loop.vectorize.width", i32 4}
3591 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
3592 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
3593 0 or if the loop does not have this metadata the width will be
3594 determined automatically.
3596 '``llvm.loop.unroll``'
3597 ^^^^^^^^^^^^^^^^^^^^^^
3599 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
3600 optimization hints such as the unroll factor. ``llvm.loop.unroll``
3601 metadata should be used in conjunction with ``llvm.loop`` loop
3602 identification metadata. The ``llvm.loop.unroll`` metadata are only
3603 optimization hints and the unrolling will only be performed if the
3604 optimizer believes it is safe to do so.
3606 '``llvm.loop.unroll.count``' Metadata
3607 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3609 This metadata suggests an unroll factor to the loop unroller. The
3610 first operand is the string ``llvm.loop.unroll.count`` and the second
3611 operand is a positive integer specifying the unroll factor. For
3614 .. code-block:: llvm
3616 !0 = !{!"llvm.loop.unroll.count", i32 4}
3618 If the trip count of the loop is less than the unroll count the loop
3619 will be partially unrolled.
3621 '``llvm.loop.unroll.disable``' Metadata
3622 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3624 This metadata either disables loop unrolling. The metadata has a single operand
3625 which is the string ``llvm.loop.unroll.disable``. For example:
3627 .. code-block:: llvm
3629 !0 = !{!"llvm.loop.unroll.disable"}
3631 '``llvm.loop.unroll.runtime.disable``' Metadata
3632 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3634 This metadata either disables runtime loop unrolling. The metadata has a single
3635 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
3637 .. code-block:: llvm
3639 !0 = !{!"llvm.loop.unroll.runtime.disable"}
3641 '``llvm.loop.unroll.full``' Metadata
3642 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3644 This metadata either suggests that the loop should be unrolled fully. The
3645 metadata has a single operand which is the string ``llvm.loop.unroll.disable``.
3648 .. code-block:: llvm
3650 !0 = !{!"llvm.loop.unroll.full"}
3655 Metadata types used to annotate memory accesses with information helpful
3656 for optimizations are prefixed with ``llvm.mem``.
3658 '``llvm.mem.parallel_loop_access``' Metadata
3659 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3661 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
3662 or metadata containing a list of loop identifiers for nested loops.
3663 The metadata is attached to memory accessing instructions and denotes that
3664 no loop carried memory dependence exist between it and other instructions denoted
3665 with the same loop identifier.
3667 Precisely, given two instructions ``m1`` and ``m2`` that both have the
3668 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
3669 set of loops associated with that metadata, respectively, then there is no loop
3670 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
3673 As a special case, if all memory accessing instructions in a loop have
3674 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
3675 loop has no loop carried memory dependences and is considered to be a parallel
3678 Note that if not all memory access instructions have such metadata referring to
3679 the loop, then the loop is considered not being trivially parallel. Additional
3680 memory dependence analysis is required to make that determination. As a fail
3681 safe mechanism, this causes loops that were originally parallel to be considered
3682 sequential (if optimization passes that are unaware of the parallel semantics
3683 insert new memory instructions into the loop body).
3685 Example of a loop that is considered parallel due to its correct use of
3686 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
3687 metadata types that refer to the same loop identifier metadata.
3689 .. code-block:: llvm
3693 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
3695 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3697 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
3703 It is also possible to have nested parallel loops. In that case the
3704 memory accesses refer to a list of loop identifier metadata nodes instead of
3705 the loop identifier metadata node directly:
3707 .. code-block:: llvm
3711 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
3713 br label %inner.for.body
3717 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3719 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
3721 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
3725 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
3727 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
3729 outer.for.end: ; preds = %for.body
3731 !0 = !{!1, !2} ; a list of loop identifiers
3732 !1 = !{!1} ; an identifier for the inner loop
3733 !2 = !{!2} ; an identifier for the outer loop
3738 The ``llvm.bitsets`` global metadata is used to implement
3739 :doc:`bitsets <BitSets>`.
3741 Module Flags Metadata
3742 =====================
3744 Information about the module as a whole is difficult to convey to LLVM's
3745 subsystems. The LLVM IR isn't sufficient to transmit this information.
3746 The ``llvm.module.flags`` named metadata exists in order to facilitate
3747 this. These flags are in the form of key / value pairs --- much like a
3748 dictionary --- making it easy for any subsystem who cares about a flag to
3751 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
3752 Each triplet has the following form:
3754 - The first element is a *behavior* flag, which specifies the behavior
3755 when two (or more) modules are merged together, and it encounters two
3756 (or more) metadata with the same ID. The supported behaviors are
3758 - The second element is a metadata string that is a unique ID for the
3759 metadata. Each module may only have one flag entry for each unique ID (not
3760 including entries with the **Require** behavior).
3761 - The third element is the value of the flag.
3763 When two (or more) modules are merged together, the resulting
3764 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
3765 each unique metadata ID string, there will be exactly one entry in the merged
3766 modules ``llvm.module.flags`` metadata table, and the value for that entry will
3767 be determined by the merge behavior flag, as described below. The only exception
3768 is that entries with the *Require* behavior are always preserved.
3770 The following behaviors are supported:
3781 Emits an error if two values disagree, otherwise the resulting value
3782 is that of the operands.
3786 Emits a warning if two values disagree. The result value will be the
3787 operand for the flag from the first module being linked.
3791 Adds a requirement that another module flag be present and have a
3792 specified value after linking is performed. The value must be a
3793 metadata pair, where the first element of the pair is the ID of the
3794 module flag to be restricted, and the second element of the pair is
3795 the value the module flag should be restricted to. This behavior can
3796 be used to restrict the allowable results (via triggering of an
3797 error) of linking IDs with the **Override** behavior.
3801 Uses the specified value, regardless of the behavior or value of the
3802 other module. If both modules specify **Override**, but the values
3803 differ, an error will be emitted.
3807 Appends the two values, which are required to be metadata nodes.
3811 Appends the two values, which are required to be metadata
3812 nodes. However, duplicate entries in the second list are dropped
3813 during the append operation.
3815 It is an error for a particular unique flag ID to have multiple behaviors,
3816 except in the case of **Require** (which adds restrictions on another metadata
3817 value) or **Override**.
3819 An example of module flags:
3821 .. code-block:: llvm
3823 !0 = !{ i32 1, !"foo", i32 1 }
3824 !1 = !{ i32 4, !"bar", i32 37 }
3825 !2 = !{ i32 2, !"qux", i32 42 }
3826 !3 = !{ i32 3, !"qux",
3831 !llvm.module.flags = !{ !0, !1, !2, !3 }
3833 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3834 if two or more ``!"foo"`` flags are seen is to emit an error if their
3835 values are not equal.
3837 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3838 behavior if two or more ``!"bar"`` flags are seen is to use the value
3841 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3842 behavior if two or more ``!"qux"`` flags are seen is to emit a
3843 warning if their values are not equal.
3845 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3851 The behavior is to emit an error if the ``llvm.module.flags`` does not
3852 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3855 Objective-C Garbage Collection Module Flags Metadata
3856 ----------------------------------------------------
3858 On the Mach-O platform, Objective-C stores metadata about garbage
3859 collection in a special section called "image info". The metadata
3860 consists of a version number and a bitmask specifying what types of
3861 garbage collection are supported (if any) by the file. If two or more
3862 modules are linked together their garbage collection metadata needs to
3863 be merged rather than appended together.
3865 The Objective-C garbage collection module flags metadata consists of the
3866 following key-value pairs:
3875 * - ``Objective-C Version``
3876 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3878 * - ``Objective-C Image Info Version``
3879 - **[Required]** --- The version of the image info section. Currently
3882 * - ``Objective-C Image Info Section``
3883 - **[Required]** --- The section to place the metadata. Valid values are
3884 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3885 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3886 Objective-C ABI version 2.
3888 * - ``Objective-C Garbage Collection``
3889 - **[Required]** --- Specifies whether garbage collection is supported or
3890 not. Valid values are 0, for no garbage collection, and 2, for garbage
3891 collection supported.
3893 * - ``Objective-C GC Only``
3894 - **[Optional]** --- Specifies that only garbage collection is supported.
3895 If present, its value must be 6. This flag requires that the
3896 ``Objective-C Garbage Collection`` flag have the value 2.
3898 Some important flag interactions:
3900 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3901 merged with a module with ``Objective-C Garbage Collection`` set to
3902 2, then the resulting module has the
3903 ``Objective-C Garbage Collection`` flag set to 0.
3904 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3905 merged with a module with ``Objective-C GC Only`` set to 6.
3907 Automatic Linker Flags Module Flags Metadata
3908 --------------------------------------------
3910 Some targets support embedding flags to the linker inside individual object
3911 files. Typically this is used in conjunction with language extensions which
3912 allow source files to explicitly declare the libraries they depend on, and have
3913 these automatically be transmitted to the linker via object files.
3915 These flags are encoded in the IR using metadata in the module flags section,
3916 using the ``Linker Options`` key. The merge behavior for this flag is required
3917 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3918 node which should be a list of other metadata nodes, each of which should be a
3919 list of metadata strings defining linker options.
3921 For example, the following metadata section specifies two separate sets of
3922 linker options, presumably to link against ``libz`` and the ``Cocoa``
3925 !0 = !{ i32 6, !"Linker Options",
3928 !{ !"-framework", !"Cocoa" } } }
3929 !llvm.module.flags = !{ !0 }
3931 The metadata encoding as lists of lists of options, as opposed to a collapsed
3932 list of options, is chosen so that the IR encoding can use multiple option
3933 strings to specify e.g., a single library, while still having that specifier be
3934 preserved as an atomic element that can be recognized by a target specific
3935 assembly writer or object file emitter.
3937 Each individual option is required to be either a valid option for the target's
3938 linker, or an option that is reserved by the target specific assembly writer or
3939 object file emitter. No other aspect of these options is defined by the IR.
3941 C type width Module Flags Metadata
3942 ----------------------------------
3944 The ARM backend emits a section into each generated object file describing the
3945 options that it was compiled with (in a compiler-independent way) to prevent
3946 linking incompatible objects, and to allow automatic library selection. Some
3947 of these options are not visible at the IR level, namely wchar_t width and enum
3950 To pass this information to the backend, these options are encoded in module
3951 flags metadata, using the following key-value pairs:
3961 - * 0 --- sizeof(wchar_t) == 4
3962 * 1 --- sizeof(wchar_t) == 2
3965 - * 0 --- Enums are at least as large as an ``int``.
3966 * 1 --- Enums are stored in the smallest integer type which can
3967 represent all of its values.
3969 For example, the following metadata section specifies that the module was
3970 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
3971 enum is the smallest type which can represent all of its values::
3973 !llvm.module.flags = !{!0, !1}
3974 !0 = !{i32 1, !"short_wchar", i32 1}
3975 !1 = !{i32 1, !"short_enum", i32 0}
3977 .. _intrinsicglobalvariables:
3979 Intrinsic Global Variables
3980 ==========================
3982 LLVM has a number of "magic" global variables that contain data that
3983 affect code generation or other IR semantics. These are documented here.
3984 All globals of this sort should have a section specified as
3985 "``llvm.metadata``". This section and all globals that start with
3986 "``llvm.``" are reserved for use by LLVM.
3990 The '``llvm.used``' Global Variable
3991 -----------------------------------
3993 The ``@llvm.used`` global is an array which has
3994 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3995 pointers to named global variables, functions and aliases which may optionally
3996 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3999 .. code-block:: llvm
4004 @llvm.used = appending global [2 x i8*] [
4006 i8* bitcast (i32* @Y to i8*)
4007 ], section "llvm.metadata"
4009 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4010 and linker are required to treat the symbol as if there is a reference to the
4011 symbol that it cannot see (which is why they have to be named). For example, if
4012 a variable has internal linkage and no references other than that from the
4013 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4014 references from inline asms and other things the compiler cannot "see", and
4015 corresponds to "``attribute((used))``" in GNU C.
4017 On some targets, the code generator must emit a directive to the
4018 assembler or object file to prevent the assembler and linker from
4019 molesting the symbol.
4021 .. _gv_llvmcompilerused:
4023 The '``llvm.compiler.used``' Global Variable
4024 --------------------------------------------
4026 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4027 directive, except that it only prevents the compiler from touching the
4028 symbol. On targets that support it, this allows an intelligent linker to
4029 optimize references to the symbol without being impeded as it would be
4032 This is a rare construct that should only be used in rare circumstances,
4033 and should not be exposed to source languages.
4035 .. _gv_llvmglobalctors:
4037 The '``llvm.global_ctors``' Global Variable
4038 -------------------------------------------
4040 .. code-block:: llvm
4042 %0 = type { i32, void ()*, i8* }
4043 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4045 The ``@llvm.global_ctors`` array contains a list of constructor
4046 functions, priorities, and an optional associated global or function.
4047 The functions referenced by this array will be called in ascending order
4048 of priority (i.e. lowest first) when the module is loaded. The order of
4049 functions with the same priority is not defined.
4051 If the third field is present, non-null, and points to a global variable
4052 or function, the initializer function will only run if the associated
4053 data from the current module is not discarded.
4055 .. _llvmglobaldtors:
4057 The '``llvm.global_dtors``' Global Variable
4058 -------------------------------------------
4060 .. code-block:: llvm
4062 %0 = type { i32, void ()*, i8* }
4063 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4065 The ``@llvm.global_dtors`` array contains a list of destructor
4066 functions, priorities, and an optional associated global or function.
4067 The functions referenced by this array will be called in descending
4068 order of priority (i.e. highest first) when the module is unloaded. The
4069 order of functions with the same priority is not defined.
4071 If the third field is present, non-null, and points to a global variable
4072 or function, the destructor function will only run if the associated
4073 data from the current module is not discarded.
4075 Instruction Reference
4076 =====================
4078 The LLVM instruction set consists of several different classifications
4079 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4080 instructions <binaryops>`, :ref:`bitwise binary
4081 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4082 :ref:`other instructions <otherops>`.
4086 Terminator Instructions
4087 -----------------------
4089 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4090 program ends with a "Terminator" instruction, which indicates which
4091 block should be executed after the current block is finished. These
4092 terminator instructions typically yield a '``void``' value: they produce
4093 control flow, not values (the one exception being the
4094 ':ref:`invoke <i_invoke>`' instruction).
4096 The terminator instructions are: ':ref:`ret <i_ret>`',
4097 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4098 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4099 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
4103 '``ret``' Instruction
4104 ^^^^^^^^^^^^^^^^^^^^^
4111 ret <type> <value> ; Return a value from a non-void function
4112 ret void ; Return from void function
4117 The '``ret``' instruction is used to return control flow (and optionally
4118 a value) from a function back to the caller.
4120 There are two forms of the '``ret``' instruction: one that returns a
4121 value and then causes control flow, and one that just causes control
4127 The '``ret``' instruction optionally accepts a single argument, the
4128 return value. The type of the return value must be a ':ref:`first
4129 class <t_firstclass>`' type.
4131 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4132 return type and contains a '``ret``' instruction with no return value or
4133 a return value with a type that does not match its type, or if it has a
4134 void return type and contains a '``ret``' instruction with a return
4140 When the '``ret``' instruction is executed, control flow returns back to
4141 the calling function's context. If the caller is a
4142 ":ref:`call <i_call>`" instruction, execution continues at the
4143 instruction after the call. If the caller was an
4144 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4145 beginning of the "normal" destination block. If the instruction returns
4146 a value, that value shall set the call or invoke instruction's return
4152 .. code-block:: llvm
4154 ret i32 5 ; Return an integer value of 5
4155 ret void ; Return from a void function
4156 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4160 '``br``' Instruction
4161 ^^^^^^^^^^^^^^^^^^^^
4168 br i1 <cond>, label <iftrue>, label <iffalse>
4169 br label <dest> ; Unconditional branch
4174 The '``br``' instruction is used to cause control flow to transfer to a
4175 different basic block in the current function. There are two forms of
4176 this instruction, corresponding to a conditional branch and an
4177 unconditional branch.
4182 The conditional branch form of the '``br``' instruction takes a single
4183 '``i1``' value and two '``label``' values. The unconditional form of the
4184 '``br``' instruction takes a single '``label``' value as a target.
4189 Upon execution of a conditional '``br``' instruction, the '``i1``'
4190 argument is evaluated. If the value is ``true``, control flows to the
4191 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4192 to the '``iffalse``' ``label`` argument.
4197 .. code-block:: llvm
4200 %cond = icmp eq i32 %a, %b
4201 br i1 %cond, label %IfEqual, label %IfUnequal
4209 '``switch``' Instruction
4210 ^^^^^^^^^^^^^^^^^^^^^^^^
4217 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
4222 The '``switch``' instruction is used to transfer control flow to one of
4223 several different places. It is a generalization of the '``br``'
4224 instruction, allowing a branch to occur to one of many possible
4230 The '``switch``' instruction uses three parameters: an integer
4231 comparison value '``value``', a default '``label``' destination, and an
4232 array of pairs of comparison value constants and '``label``'s. The table
4233 is not allowed to contain duplicate constant entries.
4238 The ``switch`` instruction specifies a table of values and destinations.
4239 When the '``switch``' instruction is executed, this table is searched
4240 for the given value. If the value is found, control flow is transferred
4241 to the corresponding destination; otherwise, control flow is transferred
4242 to the default destination.
4247 Depending on properties of the target machine and the particular
4248 ``switch`` instruction, this instruction may be code generated in
4249 different ways. For example, it could be generated as a series of
4250 chained conditional branches or with a lookup table.
4255 .. code-block:: llvm
4257 ; Emulate a conditional br instruction
4258 %Val = zext i1 %value to i32
4259 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
4261 ; Emulate an unconditional br instruction
4262 switch i32 0, label %dest [ ]
4264 ; Implement a jump table:
4265 switch i32 %val, label %otherwise [ i32 0, label %onzero
4267 i32 2, label %ontwo ]
4271 '``indirectbr``' Instruction
4272 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4279 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
4284 The '``indirectbr``' instruction implements an indirect branch to a
4285 label within the current function, whose address is specified by
4286 "``address``". Address must be derived from a
4287 :ref:`blockaddress <blockaddress>` constant.
4292 The '``address``' argument is the address of the label to jump to. The
4293 rest of the arguments indicate the full set of possible destinations
4294 that the address may point to. Blocks are allowed to occur multiple
4295 times in the destination list, though this isn't particularly useful.
4297 This destination list is required so that dataflow analysis has an
4298 accurate understanding of the CFG.
4303 Control transfers to the block specified in the address argument. All
4304 possible destination blocks must be listed in the label list, otherwise
4305 this instruction has undefined behavior. This implies that jumps to
4306 labels defined in other functions have undefined behavior as well.
4311 This is typically implemented with a jump through a register.
4316 .. code-block:: llvm
4318 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
4322 '``invoke``' Instruction
4323 ^^^^^^^^^^^^^^^^^^^^^^^^
4330 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
4331 to label <normal label> unwind label <exception label>
4336 The '``invoke``' instruction causes control to transfer to a specified
4337 function, with the possibility of control flow transfer to either the
4338 '``normal``' label or the '``exception``' label. If the callee function
4339 returns with the "``ret``" instruction, control flow will return to the
4340 "normal" label. If the callee (or any indirect callees) returns via the
4341 ":ref:`resume <i_resume>`" instruction or other exception handling
4342 mechanism, control is interrupted and continued at the dynamically
4343 nearest "exception" label.
4345 The '``exception``' label is a `landing
4346 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
4347 '``exception``' label is required to have the
4348 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
4349 information about the behavior of the program after unwinding happens,
4350 as its first non-PHI instruction. The restrictions on the
4351 "``landingpad``" instruction's tightly couples it to the "``invoke``"
4352 instruction, so that the important information contained within the
4353 "``landingpad``" instruction can't be lost through normal code motion.
4358 This instruction requires several arguments:
4360 #. The optional "cconv" marker indicates which :ref:`calling
4361 convention <callingconv>` the call should use. If none is
4362 specified, the call defaults to using C calling conventions.
4363 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
4364 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
4366 #. '``ptr to function ty``': shall be the signature of the pointer to
4367 function value being invoked. In most cases, this is a direct
4368 function invocation, but indirect ``invoke``'s are just as possible,
4369 branching off an arbitrary pointer to function value.
4370 #. '``function ptr val``': An LLVM value containing a pointer to a
4371 function to be invoked.
4372 #. '``function args``': argument list whose types match the function
4373 signature argument types and parameter attributes. All arguments must
4374 be of :ref:`first class <t_firstclass>` type. If the function signature
4375 indicates the function accepts a variable number of arguments, the
4376 extra arguments can be specified.
4377 #. '``normal label``': the label reached when the called function
4378 executes a '``ret``' instruction.
4379 #. '``exception label``': the label reached when a callee returns via
4380 the :ref:`resume <i_resume>` instruction or other exception handling
4382 #. The optional :ref:`function attributes <fnattrs>` list. Only
4383 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
4384 attributes are valid here.
4389 This instruction is designed to operate as a standard '``call``'
4390 instruction in most regards. The primary difference is that it
4391 establishes an association with a label, which is used by the runtime
4392 library to unwind the stack.
4394 This instruction is used in languages with destructors to ensure that
4395 proper cleanup is performed in the case of either a ``longjmp`` or a
4396 thrown exception. Additionally, this is important for implementation of
4397 '``catch``' clauses in high-level languages that support them.
4399 For the purposes of the SSA form, the definition of the value returned
4400 by the '``invoke``' instruction is deemed to occur on the edge from the
4401 current block to the "normal" label. If the callee unwinds then no
4402 return value is available.
4407 .. code-block:: llvm
4409 %retval = invoke i32 @Test(i32 15) to label %Continue
4410 unwind label %TestCleanup ; i32:retval set
4411 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
4412 unwind label %TestCleanup ; i32:retval set
4416 '``resume``' Instruction
4417 ^^^^^^^^^^^^^^^^^^^^^^^^
4424 resume <type> <value>
4429 The '``resume``' instruction is a terminator instruction that has no
4435 The '``resume``' instruction requires one argument, which must have the
4436 same type as the result of any '``landingpad``' instruction in the same
4442 The '``resume``' instruction resumes propagation of an existing
4443 (in-flight) exception whose unwinding was interrupted with a
4444 :ref:`landingpad <i_landingpad>` instruction.
4449 .. code-block:: llvm
4451 resume { i8*, i32 } %exn
4455 '``unreachable``' Instruction
4456 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4468 The '``unreachable``' instruction has no defined semantics. This
4469 instruction is used to inform the optimizer that a particular portion of
4470 the code is not reachable. This can be used to indicate that the code
4471 after a no-return function cannot be reached, and other facts.
4476 The '``unreachable``' instruction has no defined semantics.
4483 Binary operators are used to do most of the computation in a program.
4484 They require two operands of the same type, execute an operation on
4485 them, and produce a single value. The operands might represent multiple
4486 data, as is the case with the :ref:`vector <t_vector>` data type. The
4487 result value has the same type as its operands.
4489 There are several different binary operators:
4493 '``add``' Instruction
4494 ^^^^^^^^^^^^^^^^^^^^^
4501 <result> = add <ty> <op1>, <op2> ; yields ty:result
4502 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
4503 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
4504 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
4509 The '``add``' instruction returns the sum of its two operands.
4514 The two arguments to the '``add``' instruction must be
4515 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4516 arguments must have identical types.
4521 The value produced is the integer sum of the two operands.
4523 If the sum has unsigned overflow, the result returned is the
4524 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4527 Because LLVM integers use a two's complement representation, this
4528 instruction is appropriate for both signed and unsigned integers.
4530 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4531 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4532 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
4533 unsigned and/or signed overflow, respectively, occurs.
4538 .. code-block:: llvm
4540 <result> = add i32 4, %var ; yields i32:result = 4 + %var
4544 '``fadd``' Instruction
4545 ^^^^^^^^^^^^^^^^^^^^^^
4552 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4557 The '``fadd``' instruction returns the sum of its two operands.
4562 The two arguments to the '``fadd``' instruction must be :ref:`floating
4563 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4564 Both arguments must have identical types.
4569 The value produced is the floating point sum of the two operands. This
4570 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
4571 which are optimization hints to enable otherwise unsafe floating point
4577 .. code-block:: llvm
4579 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
4581 '``sub``' Instruction
4582 ^^^^^^^^^^^^^^^^^^^^^
4589 <result> = sub <ty> <op1>, <op2> ; yields ty:result
4590 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
4591 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
4592 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
4597 The '``sub``' instruction returns the difference of its two operands.
4599 Note that the '``sub``' instruction is used to represent the '``neg``'
4600 instruction present in most other intermediate representations.
4605 The two arguments to the '``sub``' instruction must be
4606 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4607 arguments must have identical types.
4612 The value produced is the integer difference of the two operands.
4614 If the difference has unsigned overflow, the result returned is the
4615 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4618 Because LLVM integers use a two's complement representation, this
4619 instruction is appropriate for both signed and unsigned integers.
4621 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4622 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4623 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
4624 unsigned and/or signed overflow, respectively, occurs.
4629 .. code-block:: llvm
4631 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
4632 <result> = sub i32 0, %val ; yields i32:result = -%var
4636 '``fsub``' Instruction
4637 ^^^^^^^^^^^^^^^^^^^^^^
4644 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4649 The '``fsub``' instruction returns the difference of its two operands.
4651 Note that the '``fsub``' instruction is used to represent the '``fneg``'
4652 instruction present in most other intermediate representations.
4657 The two arguments to the '``fsub``' instruction must be :ref:`floating
4658 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4659 Both arguments must have identical types.
4664 The value produced is the floating point difference of the two operands.
4665 This instruction can also take any number of :ref:`fast-math
4666 flags <fastmath>`, which are optimization hints to enable otherwise
4667 unsafe floating point optimizations:
4672 .. code-block:: llvm
4674 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
4675 <result> = fsub float -0.0, %val ; yields float:result = -%var
4677 '``mul``' Instruction
4678 ^^^^^^^^^^^^^^^^^^^^^
4685 <result> = mul <ty> <op1>, <op2> ; yields ty:result
4686 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
4687 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
4688 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
4693 The '``mul``' instruction returns the product of its two operands.
4698 The two arguments to the '``mul``' instruction must be
4699 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4700 arguments must have identical types.
4705 The value produced is the integer product of the two operands.
4707 If the result of the multiplication has unsigned overflow, the result
4708 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
4709 bit width of the result.
4711 Because LLVM integers use a two's complement representation, and the
4712 result is the same width as the operands, this instruction returns the
4713 correct result for both signed and unsigned integers. If a full product
4714 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
4715 sign-extended or zero-extended as appropriate to the width of the full
4718 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4719 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4720 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
4721 unsigned and/or signed overflow, respectively, occurs.
4726 .. code-block:: llvm
4728 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
4732 '``fmul``' Instruction
4733 ^^^^^^^^^^^^^^^^^^^^^^
4740 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4745 The '``fmul``' instruction returns the product of its two operands.
4750 The two arguments to the '``fmul``' instruction must be :ref:`floating
4751 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4752 Both arguments must have identical types.
4757 The value produced is the floating point product of the two operands.
4758 This instruction can also take any number of :ref:`fast-math
4759 flags <fastmath>`, which are optimization hints to enable otherwise
4760 unsafe floating point optimizations:
4765 .. code-block:: llvm
4767 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
4769 '``udiv``' Instruction
4770 ^^^^^^^^^^^^^^^^^^^^^^
4777 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
4778 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
4783 The '``udiv``' instruction returns the quotient of its two operands.
4788 The two arguments to the '``udiv``' instruction must be
4789 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4790 arguments must have identical types.
4795 The value produced is the unsigned integer quotient of the two operands.
4797 Note that unsigned integer division and signed integer division are
4798 distinct operations; for signed integer division, use '``sdiv``'.
4800 Division by zero leads to undefined behavior.
4802 If the ``exact`` keyword is present, the result value of the ``udiv`` is
4803 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
4804 such, "((a udiv exact b) mul b) == a").
4809 .. code-block:: llvm
4811 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
4813 '``sdiv``' Instruction
4814 ^^^^^^^^^^^^^^^^^^^^^^
4821 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
4822 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
4827 The '``sdiv``' instruction returns the quotient of its two operands.
4832 The two arguments to the '``sdiv``' instruction must be
4833 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4834 arguments must have identical types.
4839 The value produced is the signed integer quotient of the two operands
4840 rounded towards zero.
4842 Note that signed integer division and unsigned integer division are
4843 distinct operations; for unsigned integer division, use '``udiv``'.
4845 Division by zero leads to undefined behavior. Overflow also leads to
4846 undefined behavior; this is a rare case, but can occur, for example, by
4847 doing a 32-bit division of -2147483648 by -1.
4849 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4850 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4855 .. code-block:: llvm
4857 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
4861 '``fdiv``' Instruction
4862 ^^^^^^^^^^^^^^^^^^^^^^
4869 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4874 The '``fdiv``' instruction returns the quotient of its two operands.
4879 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4880 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4881 Both arguments must have identical types.
4886 The value produced is the floating point quotient of the two operands.
4887 This instruction can also take any number of :ref:`fast-math
4888 flags <fastmath>`, which are optimization hints to enable otherwise
4889 unsafe floating point optimizations:
4894 .. code-block:: llvm
4896 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
4898 '``urem``' Instruction
4899 ^^^^^^^^^^^^^^^^^^^^^^
4906 <result> = urem <ty> <op1>, <op2> ; yields ty:result
4911 The '``urem``' instruction returns the remainder from the unsigned
4912 division of its two arguments.
4917 The two arguments to the '``urem``' instruction must be
4918 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4919 arguments must have identical types.
4924 This instruction returns the unsigned integer *remainder* of a division.
4925 This instruction always performs an unsigned division to get the
4928 Note that unsigned integer remainder and signed integer remainder are
4929 distinct operations; for signed integer remainder, use '``srem``'.
4931 Taking the remainder of a division by zero leads to undefined behavior.
4936 .. code-block:: llvm
4938 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
4940 '``srem``' Instruction
4941 ^^^^^^^^^^^^^^^^^^^^^^
4948 <result> = srem <ty> <op1>, <op2> ; yields ty:result
4953 The '``srem``' instruction returns the remainder from the signed
4954 division of its two operands. This instruction can also take
4955 :ref:`vector <t_vector>` versions of the values in which case the elements
4961 The two arguments to the '``srem``' instruction must be
4962 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4963 arguments must have identical types.
4968 This instruction returns the *remainder* of a division (where the result
4969 is either zero or has the same sign as the dividend, ``op1``), not the
4970 *modulo* operator (where the result is either zero or has the same sign
4971 as the divisor, ``op2``) of a value. For more information about the
4972 difference, see `The Math
4973 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4974 table of how this is implemented in various languages, please see
4976 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4978 Note that signed integer remainder and unsigned integer remainder are
4979 distinct operations; for unsigned integer remainder, use '``urem``'.
4981 Taking the remainder of a division by zero leads to undefined behavior.
4982 Overflow also leads to undefined behavior; this is a rare case, but can
4983 occur, for example, by taking the remainder of a 32-bit division of
4984 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4985 rule lets srem be implemented using instructions that return both the
4986 result of the division and the remainder.)
4991 .. code-block:: llvm
4993 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
4997 '``frem``' Instruction
4998 ^^^^^^^^^^^^^^^^^^^^^^
5005 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5010 The '``frem``' instruction returns the remainder from the division of
5016 The two arguments to the '``frem``' instruction must be :ref:`floating
5017 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5018 Both arguments must have identical types.
5023 This instruction returns the *remainder* of a division. The remainder
5024 has the same sign as the dividend. This instruction can also take any
5025 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
5026 to enable otherwise unsafe floating point optimizations:
5031 .. code-block:: llvm
5033 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
5037 Bitwise Binary Operations
5038 -------------------------
5040 Bitwise binary operators are used to do various forms of bit-twiddling
5041 in a program. They are generally very efficient instructions and can
5042 commonly be strength reduced from other instructions. They require two
5043 operands of the same type, execute an operation on them, and produce a
5044 single value. The resulting value is the same type as its operands.
5046 '``shl``' Instruction
5047 ^^^^^^^^^^^^^^^^^^^^^
5054 <result> = shl <ty> <op1>, <op2> ; yields ty:result
5055 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
5056 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
5057 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
5062 The '``shl``' instruction returns the first operand shifted to the left
5063 a specified number of bits.
5068 Both arguments to the '``shl``' instruction must be the same
5069 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5070 '``op2``' is treated as an unsigned value.
5075 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
5076 where ``n`` is the width of the result. If ``op2`` is (statically or
5077 dynamically) equal to or larger than the number of bits in
5078 ``op1``, the result is undefined. If the arguments are vectors, each
5079 vector element of ``op1`` is shifted by the corresponding shift amount
5082 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
5083 value <poisonvalues>` if it shifts out any non-zero bits. If the
5084 ``nsw`` keyword is present, then the shift produces a :ref:`poison
5085 value <poisonvalues>` if it shifts out any bits that disagree with the
5086 resultant sign bit. As such, NUW/NSW have the same semantics as they
5087 would if the shift were expressed as a mul instruction with the same
5088 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
5093 .. code-block:: llvm
5095 <result> = shl i32 4, %var ; yields i32: 4 << %var
5096 <result> = shl i32 4, 2 ; yields i32: 16
5097 <result> = shl i32 1, 10 ; yields i32: 1024
5098 <result> = shl i32 1, 32 ; undefined
5099 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
5101 '``lshr``' Instruction
5102 ^^^^^^^^^^^^^^^^^^^^^^
5109 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
5110 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
5115 The '``lshr``' instruction (logical shift right) returns the first
5116 operand shifted to the right a specified number of bits with zero fill.
5121 Both arguments to the '``lshr``' instruction must be the same
5122 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5123 '``op2``' is treated as an unsigned value.
5128 This instruction always performs a logical shift right operation. The
5129 most significant bits of the result will be filled with zero bits after
5130 the shift. If ``op2`` is (statically or dynamically) equal to or larger
5131 than the number of bits in ``op1``, the result is undefined. If the
5132 arguments are vectors, each vector element of ``op1`` is shifted by the
5133 corresponding shift amount in ``op2``.
5135 If the ``exact`` keyword is present, the result value of the ``lshr`` is
5136 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5142 .. code-block:: llvm
5144 <result> = lshr i32 4, 1 ; yields i32:result = 2
5145 <result> = lshr i32 4, 2 ; yields i32:result = 1
5146 <result> = lshr i8 4, 3 ; yields i8:result = 0
5147 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
5148 <result> = lshr i32 1, 32 ; undefined
5149 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
5151 '``ashr``' Instruction
5152 ^^^^^^^^^^^^^^^^^^^^^^
5159 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
5160 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
5165 The '``ashr``' instruction (arithmetic shift right) returns the first
5166 operand shifted to the right a specified number of bits with sign
5172 Both arguments to the '``ashr``' instruction must be the same
5173 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5174 '``op2``' is treated as an unsigned value.
5179 This instruction always performs an arithmetic shift right operation,
5180 The most significant bits of the result will be filled with the sign bit
5181 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
5182 than the number of bits in ``op1``, the result is undefined. If the
5183 arguments are vectors, each vector element of ``op1`` is shifted by the
5184 corresponding shift amount in ``op2``.
5186 If the ``exact`` keyword is present, the result value of the ``ashr`` is
5187 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5193 .. code-block:: llvm
5195 <result> = ashr i32 4, 1 ; yields i32:result = 2
5196 <result> = ashr i32 4, 2 ; yields i32:result = 1
5197 <result> = ashr i8 4, 3 ; yields i8:result = 0
5198 <result> = ashr i8 -2, 1 ; yields i8:result = -1
5199 <result> = ashr i32 1, 32 ; undefined
5200 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
5202 '``and``' Instruction
5203 ^^^^^^^^^^^^^^^^^^^^^
5210 <result> = and <ty> <op1>, <op2> ; yields ty:result
5215 The '``and``' instruction returns the bitwise logical and of its two
5221 The two arguments to the '``and``' instruction must be
5222 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5223 arguments must have identical types.
5228 The truth table used for the '``and``' instruction is:
5245 .. code-block:: llvm
5247 <result> = and i32 4, %var ; yields i32:result = 4 & %var
5248 <result> = and i32 15, 40 ; yields i32:result = 8
5249 <result> = and i32 4, 8 ; yields i32:result = 0
5251 '``or``' Instruction
5252 ^^^^^^^^^^^^^^^^^^^^
5259 <result> = or <ty> <op1>, <op2> ; yields ty:result
5264 The '``or``' instruction returns the bitwise logical inclusive or of its
5270 The two arguments to the '``or``' instruction must be
5271 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5272 arguments must have identical types.
5277 The truth table used for the '``or``' instruction is:
5296 <result> = or i32 4, %var ; yields i32:result = 4 | %var
5297 <result> = or i32 15, 40 ; yields i32:result = 47
5298 <result> = or i32 4, 8 ; yields i32:result = 12
5300 '``xor``' Instruction
5301 ^^^^^^^^^^^^^^^^^^^^^
5308 <result> = xor <ty> <op1>, <op2> ; yields ty:result
5313 The '``xor``' instruction returns the bitwise logical exclusive or of
5314 its two operands. The ``xor`` is used to implement the "one's
5315 complement" operation, which is the "~" operator in C.
5320 The two arguments to the '``xor``' instruction must be
5321 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5322 arguments must have identical types.
5327 The truth table used for the '``xor``' instruction is:
5344 .. code-block:: llvm
5346 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
5347 <result> = xor i32 15, 40 ; yields i32:result = 39
5348 <result> = xor i32 4, 8 ; yields i32:result = 12
5349 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
5354 LLVM supports several instructions to represent vector operations in a
5355 target-independent manner. These instructions cover the element-access
5356 and vector-specific operations needed to process vectors effectively.
5357 While LLVM does directly support these vector operations, many
5358 sophisticated algorithms will want to use target-specific intrinsics to
5359 take full advantage of a specific target.
5361 .. _i_extractelement:
5363 '``extractelement``' Instruction
5364 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5371 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
5376 The '``extractelement``' instruction extracts a single scalar element
5377 from a vector at a specified index.
5382 The first operand of an '``extractelement``' instruction is a value of
5383 :ref:`vector <t_vector>` type. The second operand is an index indicating
5384 the position from which to extract the element. The index may be a
5385 variable of any integer type.
5390 The result is a scalar of the same type as the element type of ``val``.
5391 Its value is the value at position ``idx`` of ``val``. If ``idx``
5392 exceeds the length of ``val``, the results are undefined.
5397 .. code-block:: llvm
5399 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
5401 .. _i_insertelement:
5403 '``insertelement``' Instruction
5404 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5411 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
5416 The '``insertelement``' instruction inserts a scalar element into a
5417 vector at a specified index.
5422 The first operand of an '``insertelement``' instruction is a value of
5423 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
5424 type must equal the element type of the first operand. The third operand
5425 is an index indicating the position at which to insert the value. The
5426 index may be a variable of any integer type.
5431 The result is a vector of the same type as ``val``. Its element values
5432 are those of ``val`` except at position ``idx``, where it gets the value
5433 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
5439 .. code-block:: llvm
5441 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
5443 .. _i_shufflevector:
5445 '``shufflevector``' Instruction
5446 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5453 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
5458 The '``shufflevector``' instruction constructs a permutation of elements
5459 from two input vectors, returning a vector with the same element type as
5460 the input and length that is the same as the shuffle mask.
5465 The first two operands of a '``shufflevector``' instruction are vectors
5466 with the same type. The third argument is a shuffle mask whose element
5467 type is always 'i32'. The result of the instruction is a vector whose
5468 length is the same as the shuffle mask and whose element type is the
5469 same as the element type of the first two operands.
5471 The shuffle mask operand is required to be a constant vector with either
5472 constant integer or undef values.
5477 The elements of the two input vectors are numbered from left to right
5478 across both of the vectors. The shuffle mask operand specifies, for each
5479 element of the result vector, which element of the two input vectors the
5480 result element gets. The element selector may be undef (meaning "don't
5481 care") and the second operand may be undef if performing a shuffle from
5487 .. code-block:: llvm
5489 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5490 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
5491 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
5492 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
5493 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
5494 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
5495 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5496 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
5498 Aggregate Operations
5499 --------------------
5501 LLVM supports several instructions for working with
5502 :ref:`aggregate <t_aggregate>` values.
5506 '``extractvalue``' Instruction
5507 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5514 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
5519 The '``extractvalue``' instruction extracts the value of a member field
5520 from an :ref:`aggregate <t_aggregate>` value.
5525 The first operand of an '``extractvalue``' instruction is a value of
5526 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
5527 constant indices to specify which value to extract in a similar manner
5528 as indices in a '``getelementptr``' instruction.
5530 The major differences to ``getelementptr`` indexing are:
5532 - Since the value being indexed is not a pointer, the first index is
5533 omitted and assumed to be zero.
5534 - At least one index must be specified.
5535 - Not only struct indices but also array indices must be in bounds.
5540 The result is the value at the position in the aggregate specified by
5546 .. code-block:: llvm
5548 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
5552 '``insertvalue``' Instruction
5553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5560 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
5565 The '``insertvalue``' instruction inserts a value into a member field in
5566 an :ref:`aggregate <t_aggregate>` value.
5571 The first operand of an '``insertvalue``' instruction is a value of
5572 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
5573 a first-class value to insert. The following operands are constant
5574 indices indicating the position at which to insert the value in a
5575 similar manner as indices in a '``extractvalue``' instruction. The value
5576 to insert must have the same type as the value identified by the
5582 The result is an aggregate of the same type as ``val``. Its value is
5583 that of ``val`` except that the value at the position specified by the
5584 indices is that of ``elt``.
5589 .. code-block:: llvm
5591 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
5592 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
5593 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
5597 Memory Access and Addressing Operations
5598 ---------------------------------------
5600 A key design point of an SSA-based representation is how it represents
5601 memory. In LLVM, no memory locations are in SSA form, which makes things
5602 very simple. This section describes how to read, write, and allocate
5607 '``alloca``' Instruction
5608 ^^^^^^^^^^^^^^^^^^^^^^^^
5615 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
5620 The '``alloca``' instruction allocates memory on the stack frame of the
5621 currently executing function, to be automatically released when this
5622 function returns to its caller. The object is always allocated in the
5623 generic address space (address space zero).
5628 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
5629 bytes of memory on the runtime stack, returning a pointer of the
5630 appropriate type to the program. If "NumElements" is specified, it is
5631 the number of elements allocated, otherwise "NumElements" is defaulted
5632 to be one. If a constant alignment is specified, the value result of the
5633 allocation is guaranteed to be aligned to at least that boundary. The
5634 alignment may not be greater than ``1 << 29``. If not specified, or if
5635 zero, the target can choose to align the allocation on any convenient
5636 boundary compatible with the type.
5638 '``type``' may be any sized type.
5643 Memory is allocated; a pointer is returned. The operation is undefined
5644 if there is insufficient stack space for the allocation. '``alloca``'d
5645 memory is automatically released when the function returns. The
5646 '``alloca``' instruction is commonly used to represent automatic
5647 variables that must have an address available. When the function returns
5648 (either with the ``ret`` or ``resume`` instructions), the memory is
5649 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5650 The order in which memory is allocated (ie., which way the stack grows)
5656 .. code-block:: llvm
5658 %ptr = alloca i32 ; yields i32*:ptr
5659 %ptr = alloca i32, i32 4 ; yields i32*:ptr
5660 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
5661 %ptr = alloca i32, align 1024 ; yields i32*:ptr
5665 '``load``' Instruction
5666 ^^^^^^^^^^^^^^^^^^^^^^
5673 <result> = load [volatile] <ty>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !nonnull !<index>][, !dereferenceable !<index>][, !dereferenceable_or_null !<index>]
5674 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5675 !<index> = !{ i32 1 }
5680 The '``load``' instruction is used to read from memory.
5685 The argument to the ``load`` instruction specifies the memory address
5686 from which to load. The type specified must be a :ref:`first
5687 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
5688 then the optimizer is not allowed to modify the number or order of
5689 execution of this ``load`` with other :ref:`volatile
5690 operations <volatile>`.
5692 If the ``load`` is marked as ``atomic``, it takes an extra
5693 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5694 ``release`` and ``acq_rel`` orderings are not valid on ``load``
5695 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5696 when they may see multiple atomic stores. The type of the pointee must
5697 be an integer type whose bit width is a power of two greater than or
5698 equal to eight and less than or equal to a target-specific size limit.
5699 ``align`` must be explicitly specified on atomic loads, and the load has
5700 undefined behavior if the alignment is not set to a value which is at
5701 least the size in bytes of the pointee. ``!nontemporal`` does not have
5702 any defined semantics for atomic loads.
5704 The optional constant ``align`` argument specifies the alignment of the
5705 operation (that is, the alignment of the memory address). A value of 0
5706 or an omitted ``align`` argument means that the operation has the ABI
5707 alignment for the target. It is the responsibility of the code emitter
5708 to ensure that the alignment information is correct. Overestimating the
5709 alignment results in undefined behavior. Underestimating the alignment
5710 may produce less efficient code. An alignment of 1 is always safe. The
5711 maximum possible alignment is ``1 << 29``.
5713 The optional ``!nontemporal`` metadata must reference a single
5714 metadata name ``<index>`` corresponding to a metadata node with one
5715 ``i32`` entry of value 1. The existence of the ``!nontemporal``
5716 metadata on the instruction tells the optimizer and code generator
5717 that this load is not expected to be reused in the cache. The code
5718 generator may select special instructions to save cache bandwidth, such
5719 as the ``MOVNT`` instruction on x86.
5721 The optional ``!invariant.load`` metadata must reference a single
5722 metadata name ``<index>`` corresponding to a metadata node with no
5723 entries. The existence of the ``!invariant.load`` metadata on the
5724 instruction tells the optimizer and code generator that the address
5725 operand to this load points to memory which can be assumed unchanged.
5726 Being invariant does not imply that a location is dereferenceable,
5727 but it does imply that once the location is known dereferenceable
5728 its value is henceforth unchanging.
5730 The optional ``!nonnull`` metadata must reference a single
5731 metadata name ``<index>`` corresponding to a metadata node with no
5732 entries. The existence of the ``!nonnull`` metadata on the
5733 instruction tells the optimizer that the value loaded is known to
5734 never be null. This is analogous to the ''nonnull'' attribute
5735 on parameters and return values. This metadata can only be applied
5736 to loads of a pointer type.
5738 The optional ``!dereferenceable`` metadata must reference a single
5739 metadata name ``<index>`` corresponding to a metadata node with one ``i64``
5740 entry. The existence of the ``!dereferenceable`` metadata on the instruction
5741 tells the optimizer that the value loaded is known to be dereferenceable.
5742 The number of bytes known to be dereferenceable is specified by the integer
5743 value in the metadata node. This is analogous to the ''dereferenceable''
5744 attribute on parameters and return values. This metadata can only be applied
5745 to loads of a pointer type.
5747 The optional ``!dereferenceable_or_null`` metadata must reference a single
5748 metadata name ``<index>`` corresponding to a metadata node with one ``i64``
5749 entry. The existence of the ``!dereferenceable_or_null`` metadata on the
5750 instruction tells the optimizer that the value loaded is known to be either
5751 dereferenceable or null.
5752 The number of bytes known to be dereferenceable is specified by the integer
5753 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
5754 attribute on parameters and return values. This metadata can only be applied
5755 to loads of a pointer type.
5760 The location of memory pointed to is loaded. If the value being loaded
5761 is of scalar type then the number of bytes read does not exceed the
5762 minimum number of bytes needed to hold all bits of the type. For
5763 example, loading an ``i24`` reads at most three bytes. When loading a
5764 value of a type like ``i20`` with a size that is not an integral number
5765 of bytes, the result is undefined if the value was not originally
5766 written using a store of the same type.
5771 .. code-block:: llvm
5773 %ptr = alloca i32 ; yields i32*:ptr
5774 store i32 3, i32* %ptr ; yields void
5775 %val = load i32, i32* %ptr ; yields i32:val = i32 3
5779 '``store``' Instruction
5780 ^^^^^^^^^^^^^^^^^^^^^^^
5787 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
5788 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
5793 The '``store``' instruction is used to write to memory.
5798 There are two arguments to the ``store`` instruction: a value to store
5799 and an address at which to store it. The type of the ``<pointer>``
5800 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
5801 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
5802 then the optimizer is not allowed to modify the number or order of
5803 execution of this ``store`` with other :ref:`volatile
5804 operations <volatile>`.
5806 If the ``store`` is marked as ``atomic``, it takes an extra
5807 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5808 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
5809 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5810 when they may see multiple atomic stores. The type of the pointee must
5811 be an integer type whose bit width is a power of two greater than or
5812 equal to eight and less than or equal to a target-specific size limit.
5813 ``align`` must be explicitly specified on atomic stores, and the store
5814 has undefined behavior if the alignment is not set to a value which is
5815 at least the size in bytes of the pointee. ``!nontemporal`` does not
5816 have any defined semantics for atomic stores.
5818 The optional constant ``align`` argument specifies the alignment of the
5819 operation (that is, the alignment of the memory address). A value of 0
5820 or an omitted ``align`` argument means that the operation has the ABI
5821 alignment for the target. It is the responsibility of the code emitter
5822 to ensure that the alignment information is correct. Overestimating the
5823 alignment results in undefined behavior. Underestimating the
5824 alignment may produce less efficient code. An alignment of 1 is always
5825 safe. The maximum possible alignment is ``1 << 29``.
5827 The optional ``!nontemporal`` metadata must reference a single metadata
5828 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
5829 value 1. The existence of the ``!nontemporal`` metadata on the instruction
5830 tells the optimizer and code generator that this load is not expected to
5831 be reused in the cache. The code generator may select special
5832 instructions to save cache bandwidth, such as the MOVNT instruction on
5838 The contents of memory are updated to contain ``<value>`` at the
5839 location specified by the ``<pointer>`` operand. If ``<value>`` is
5840 of scalar type then the number of bytes written does not exceed the
5841 minimum number of bytes needed to hold all bits of the type. For
5842 example, storing an ``i24`` writes at most three bytes. When writing a
5843 value of a type like ``i20`` with a size that is not an integral number
5844 of bytes, it is unspecified what happens to the extra bits that do not
5845 belong to the type, but they will typically be overwritten.
5850 .. code-block:: llvm
5852 %ptr = alloca i32 ; yields i32*:ptr
5853 store i32 3, i32* %ptr ; yields void
5854 %val = load i32* %ptr ; yields i32:val = i32 3
5858 '``fence``' Instruction
5859 ^^^^^^^^^^^^^^^^^^^^^^^
5866 fence [singlethread] <ordering> ; yields void
5871 The '``fence``' instruction is used to introduce happens-before edges
5877 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5878 defines what *synchronizes-with* edges they add. They can only be given
5879 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5884 A fence A which has (at least) ``release`` ordering semantics
5885 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5886 semantics if and only if there exist atomic operations X and Y, both
5887 operating on some atomic object M, such that A is sequenced before X, X
5888 modifies M (either directly or through some side effect of a sequence
5889 headed by X), Y is sequenced before B, and Y observes M. This provides a
5890 *happens-before* dependency between A and B. Rather than an explicit
5891 ``fence``, one (but not both) of the atomic operations X or Y might
5892 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5893 still *synchronize-with* the explicit ``fence`` and establish the
5894 *happens-before* edge.
5896 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5897 ``acquire`` and ``release`` semantics specified above, participates in
5898 the global program order of other ``seq_cst`` operations and/or fences.
5900 The optional ":ref:`singlethread <singlethread>`" argument specifies
5901 that the fence only synchronizes with other fences in the same thread.
5902 (This is useful for interacting with signal handlers.)
5907 .. code-block:: llvm
5909 fence acquire ; yields void
5910 fence singlethread seq_cst ; yields void
5914 '``cmpxchg``' Instruction
5915 ^^^^^^^^^^^^^^^^^^^^^^^^^
5922 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
5927 The '``cmpxchg``' instruction is used to atomically modify memory. It
5928 loads a value in memory and compares it to a given value. If they are
5929 equal, it tries to store a new value into the memory.
5934 There are three arguments to the '``cmpxchg``' instruction: an address
5935 to operate on, a value to compare to the value currently be at that
5936 address, and a new value to place at that address if the compared values
5937 are equal. The type of '<cmp>' must be an integer type whose bit width
5938 is a power of two greater than or equal to eight and less than or equal
5939 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5940 type, and the type of '<pointer>' must be a pointer to that type. If the
5941 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5942 to modify the number or order of execution of this ``cmpxchg`` with
5943 other :ref:`volatile operations <volatile>`.
5945 The success and failure :ref:`ordering <ordering>` arguments specify how this
5946 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
5947 must be at least ``monotonic``, the ordering constraint on failure must be no
5948 stronger than that on success, and the failure ordering cannot be either
5949 ``release`` or ``acq_rel``.
5951 The optional "``singlethread``" argument declares that the ``cmpxchg``
5952 is only atomic with respect to code (usually signal handlers) running in
5953 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5954 respect to all other code in the system.
5956 The pointer passed into cmpxchg must have alignment greater than or
5957 equal to the size in memory of the operand.
5962 The contents of memory at the location specified by the '``<pointer>``' operand
5963 is read and compared to '``<cmp>``'; if the read value is the equal, the
5964 '``<new>``' is written. The original value at the location is returned, together
5965 with a flag indicating success (true) or failure (false).
5967 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
5968 permitted: the operation may not write ``<new>`` even if the comparison
5971 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
5972 if the value loaded equals ``cmp``.
5974 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5975 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5976 load with an ordering parameter determined the second ordering parameter.
5981 .. code-block:: llvm
5984 %orig = atomic load i32, i32* %ptr unordered ; yields i32
5988 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5989 %squared = mul i32 %cmp, %cmp
5990 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
5991 %value_loaded = extractvalue { i32, i1 } %val_success, 0
5992 %success = extractvalue { i32, i1 } %val_success, 1
5993 br i1 %success, label %done, label %loop
6000 '``atomicrmw``' Instruction
6001 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6008 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
6013 The '``atomicrmw``' instruction is used to atomically modify memory.
6018 There are three arguments to the '``atomicrmw``' instruction: an
6019 operation to apply, an address whose value to modify, an argument to the
6020 operation. The operation must be one of the following keywords:
6034 The type of '<value>' must be an integer type whose bit width is a power
6035 of two greater than or equal to eight and less than or equal to a
6036 target-specific size limit. The type of the '``<pointer>``' operand must
6037 be a pointer to that type. If the ``atomicrmw`` is marked as
6038 ``volatile``, then the optimizer is not allowed to modify the number or
6039 order of execution of this ``atomicrmw`` with other :ref:`volatile
6040 operations <volatile>`.
6045 The contents of memory at the location specified by the '``<pointer>``'
6046 operand are atomically read, modified, and written back. The original
6047 value at the location is returned. The modification is specified by the
6050 - xchg: ``*ptr = val``
6051 - add: ``*ptr = *ptr + val``
6052 - sub: ``*ptr = *ptr - val``
6053 - and: ``*ptr = *ptr & val``
6054 - nand: ``*ptr = ~(*ptr & val)``
6055 - or: ``*ptr = *ptr | val``
6056 - xor: ``*ptr = *ptr ^ val``
6057 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
6058 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
6059 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
6061 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
6067 .. code-block:: llvm
6069 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
6071 .. _i_getelementptr:
6073 '``getelementptr``' Instruction
6074 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6081 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
6082 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
6083 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
6088 The '``getelementptr``' instruction is used to get the address of a
6089 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
6090 address calculation only and does not access memory.
6095 The first argument is always a type used as the basis for the calculations.
6096 The second argument is always a pointer or a vector of pointers, and is the
6097 base address to start from. The remaining arguments are indices
6098 that indicate which of the elements of the aggregate object are indexed.
6099 The interpretation of each index is dependent on the type being indexed
6100 into. The first index always indexes the pointer value given as the
6101 first argument, the second index indexes a value of the type pointed to
6102 (not necessarily the value directly pointed to, since the first index
6103 can be non-zero), etc. The first type indexed into must be a pointer
6104 value, subsequent types can be arrays, vectors, and structs. Note that
6105 subsequent types being indexed into can never be pointers, since that
6106 would require loading the pointer before continuing calculation.
6108 The type of each index argument depends on the type it is indexing into.
6109 When indexing into a (optionally packed) structure, only ``i32`` integer
6110 **constants** are allowed (when using a vector of indices they must all
6111 be the **same** ``i32`` integer constant). When indexing into an array,
6112 pointer or vector, integers of any width are allowed, and they are not
6113 required to be constant. These integers are treated as signed values
6116 For example, let's consider a C code fragment and how it gets compiled
6132 int *foo(struct ST *s) {
6133 return &s[1].Z.B[5][13];
6136 The LLVM code generated by Clang is:
6138 .. code-block:: llvm
6140 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
6141 %struct.ST = type { i32, double, %struct.RT }
6143 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
6145 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
6152 In the example above, the first index is indexing into the
6153 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
6154 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
6155 indexes into the third element of the structure, yielding a
6156 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
6157 structure. The third index indexes into the second element of the
6158 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
6159 dimensions of the array are subscripted into, yielding an '``i32``'
6160 type. The '``getelementptr``' instruction returns a pointer to this
6161 element, thus computing a value of '``i32*``' type.
6163 Note that it is perfectly legal to index partially through a structure,
6164 returning a pointer to an inner element. Because of this, the LLVM code
6165 for the given testcase is equivalent to:
6167 .. code-block:: llvm
6169 define i32* @foo(%struct.ST* %s) {
6170 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
6171 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
6172 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
6173 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
6174 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
6178 If the ``inbounds`` keyword is present, the result value of the
6179 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
6180 pointer is not an *in bounds* address of an allocated object, or if any
6181 of the addresses that would be formed by successive addition of the
6182 offsets implied by the indices to the base address with infinitely
6183 precise signed arithmetic are not an *in bounds* address of that
6184 allocated object. The *in bounds* addresses for an allocated object are
6185 all the addresses that point into the object, plus the address one byte
6186 past the end. In cases where the base is a vector of pointers the
6187 ``inbounds`` keyword applies to each of the computations element-wise.
6189 If the ``inbounds`` keyword is not present, the offsets are added to the
6190 base address with silently-wrapping two's complement arithmetic. If the
6191 offsets have a different width from the pointer, they are sign-extended
6192 or truncated to the width of the pointer. The result value of the
6193 ``getelementptr`` may be outside the object pointed to by the base
6194 pointer. The result value may not necessarily be used to access memory
6195 though, even if it happens to point into allocated storage. See the
6196 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
6199 The getelementptr instruction is often confusing. For some more insight
6200 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
6205 .. code-block:: llvm
6207 ; yields [12 x i8]*:aptr
6208 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
6210 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
6212 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
6214 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
6216 In cases where the pointer argument is a vector of pointers, each index
6217 must be a vector with the same number of elements. For example:
6219 .. code-block:: llvm
6221 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets,
6223 Conversion Operations
6224 ---------------------
6226 The instructions in this category are the conversion instructions
6227 (casting) which all take a single operand and a type. They perform
6228 various bit conversions on the operand.
6230 '``trunc .. to``' Instruction
6231 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6238 <result> = trunc <ty> <value> to <ty2> ; yields ty2
6243 The '``trunc``' instruction truncates its operand to the type ``ty2``.
6248 The '``trunc``' instruction takes a value to trunc, and a type to trunc
6249 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
6250 of the same number of integers. The bit size of the ``value`` must be
6251 larger than the bit size of the destination type, ``ty2``. Equal sized
6252 types are not allowed.
6257 The '``trunc``' instruction truncates the high order bits in ``value``
6258 and converts the remaining bits to ``ty2``. Since the source size must
6259 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
6260 It will always truncate bits.
6265 .. code-block:: llvm
6267 %X = trunc i32 257 to i8 ; yields i8:1
6268 %Y = trunc i32 123 to i1 ; yields i1:true
6269 %Z = trunc i32 122 to i1 ; yields i1:false
6270 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
6272 '``zext .. to``' Instruction
6273 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6280 <result> = zext <ty> <value> to <ty2> ; yields ty2
6285 The '``zext``' instruction zero extends its operand to type ``ty2``.
6290 The '``zext``' instruction takes a value to cast, and a type to cast it
6291 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
6292 the same number of integers. The bit size of the ``value`` must be
6293 smaller than the bit size of the destination type, ``ty2``.
6298 The ``zext`` fills the high order bits of the ``value`` with zero bits
6299 until it reaches the size of the destination type, ``ty2``.
6301 When zero extending from i1, the result will always be either 0 or 1.
6306 .. code-block:: llvm
6308 %X = zext i32 257 to i64 ; yields i64:257
6309 %Y = zext i1 true to i32 ; yields i32:1
6310 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
6312 '``sext .. to``' Instruction
6313 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6320 <result> = sext <ty> <value> to <ty2> ; yields ty2
6325 The '``sext``' sign extends ``value`` to the type ``ty2``.
6330 The '``sext``' instruction takes a value to cast, and a type to cast it
6331 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
6332 the same number of integers. The bit size of the ``value`` must be
6333 smaller than the bit size of the destination type, ``ty2``.
6338 The '``sext``' instruction performs a sign extension by copying the sign
6339 bit (highest order bit) of the ``value`` until it reaches the bit size
6340 of the type ``ty2``.
6342 When sign extending from i1, the extension always results in -1 or 0.
6347 .. code-block:: llvm
6349 %X = sext i8 -1 to i16 ; yields i16 :65535
6350 %Y = sext i1 true to i32 ; yields i32:-1
6351 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
6353 '``fptrunc .. to``' Instruction
6354 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6361 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
6366 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
6371 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
6372 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
6373 The size of ``value`` must be larger than the size of ``ty2``. This
6374 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
6379 The '``fptrunc``' instruction truncates a ``value`` from a larger
6380 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
6381 point <t_floating>` type. If the value cannot fit within the
6382 destination type, ``ty2``, then the results are undefined.
6387 .. code-block:: llvm
6389 %X = fptrunc double 123.0 to float ; yields float:123.0
6390 %Y = fptrunc double 1.0E+300 to float ; yields undefined
6392 '``fpext .. to``' Instruction
6393 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6400 <result> = fpext <ty> <value> to <ty2> ; yields ty2
6405 The '``fpext``' extends a floating point ``value`` to a larger floating
6411 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
6412 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
6413 to. The source type must be smaller than the destination type.
6418 The '``fpext``' instruction extends the ``value`` from a smaller
6419 :ref:`floating point <t_floating>` type to a larger :ref:`floating
6420 point <t_floating>` type. The ``fpext`` cannot be used to make a
6421 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
6422 *no-op cast* for a floating point cast.
6427 .. code-block:: llvm
6429 %X = fpext float 3.125 to double ; yields double:3.125000e+00
6430 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
6432 '``fptoui .. to``' Instruction
6433 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6440 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
6445 The '``fptoui``' converts a floating point ``value`` to its unsigned
6446 integer equivalent of type ``ty2``.
6451 The '``fptoui``' instruction takes a value to cast, which must be a
6452 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6453 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6454 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6455 type with the same number of elements as ``ty``
6460 The '``fptoui``' instruction converts its :ref:`floating
6461 point <t_floating>` operand into the nearest (rounding towards zero)
6462 unsigned integer value. If the value cannot fit in ``ty2``, the results
6468 .. code-block:: llvm
6470 %X = fptoui double 123.0 to i32 ; yields i32:123
6471 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
6472 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
6474 '``fptosi .. to``' Instruction
6475 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6482 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
6487 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
6488 ``value`` to type ``ty2``.
6493 The '``fptosi``' instruction takes a value to cast, which must be a
6494 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6495 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6496 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6497 type with the same number of elements as ``ty``
6502 The '``fptosi``' instruction converts its :ref:`floating
6503 point <t_floating>` operand into the nearest (rounding towards zero)
6504 signed integer value. If the value cannot fit in ``ty2``, the results
6510 .. code-block:: llvm
6512 %X = fptosi double -123.0 to i32 ; yields i32:-123
6513 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
6514 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
6516 '``uitofp .. to``' Instruction
6517 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6524 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
6529 The '``uitofp``' instruction regards ``value`` as an unsigned integer
6530 and converts that value to the ``ty2`` type.
6535 The '``uitofp``' instruction takes a value to cast, which must be a
6536 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6537 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6538 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6539 type with the same number of elements as ``ty``
6544 The '``uitofp``' instruction interprets its operand as an unsigned
6545 integer quantity and converts it to the corresponding floating point
6546 value. If the value cannot fit in the floating point value, the results
6552 .. code-block:: llvm
6554 %X = uitofp i32 257 to float ; yields float:257.0
6555 %Y = uitofp i8 -1 to double ; yields double:255.0
6557 '``sitofp .. to``' Instruction
6558 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6565 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
6570 The '``sitofp``' instruction regards ``value`` as a signed integer and
6571 converts that value to the ``ty2`` type.
6576 The '``sitofp``' instruction takes a value to cast, which must be a
6577 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6578 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6579 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6580 type with the same number of elements as ``ty``
6585 The '``sitofp``' instruction interprets its operand as a signed integer
6586 quantity and converts it to the corresponding floating point value. If
6587 the value cannot fit in the floating point value, the results are
6593 .. code-block:: llvm
6595 %X = sitofp i32 257 to float ; yields float:257.0
6596 %Y = sitofp i8 -1 to double ; yields double:-1.0
6600 '``ptrtoint .. to``' Instruction
6601 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6608 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
6613 The '``ptrtoint``' instruction converts the pointer or a vector of
6614 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
6619 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
6620 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
6621 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
6622 a vector of integers type.
6627 The '``ptrtoint``' instruction converts ``value`` to integer type
6628 ``ty2`` by interpreting the pointer value as an integer and either
6629 truncating or zero extending that value to the size of the integer type.
6630 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
6631 ``value`` is larger than ``ty2`` then a truncation is done. If they are
6632 the same size, then nothing is done (*no-op cast*) other than a type
6638 .. code-block:: llvm
6640 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
6641 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
6642 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
6646 '``inttoptr .. to``' Instruction
6647 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6654 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
6659 The '``inttoptr``' instruction converts an integer ``value`` to a
6660 pointer type, ``ty2``.
6665 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
6666 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
6672 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
6673 applying either a zero extension or a truncation depending on the size
6674 of the integer ``value``. If ``value`` is larger than the size of a
6675 pointer then a truncation is done. If ``value`` is smaller than the size
6676 of a pointer then a zero extension is done. If they are the same size,
6677 nothing is done (*no-op cast*).
6682 .. code-block:: llvm
6684 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
6685 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
6686 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
6687 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
6691 '``bitcast .. to``' Instruction
6692 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6699 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
6704 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
6710 The '``bitcast``' instruction takes a value to cast, which must be a
6711 non-aggregate first class value, and a type to cast it to, which must
6712 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
6713 bit sizes of ``value`` and the destination type, ``ty2``, must be
6714 identical. If the source type is a pointer, the destination type must
6715 also be a pointer of the same size. This instruction supports bitwise
6716 conversion of vectors to integers and to vectors of other types (as
6717 long as they have the same size).
6722 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
6723 is always a *no-op cast* because no bits change with this
6724 conversion. The conversion is done as if the ``value`` had been stored
6725 to memory and read back as type ``ty2``. Pointer (or vector of
6726 pointers) types may only be converted to other pointer (or vector of
6727 pointers) types with the same address space through this instruction.
6728 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
6729 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
6734 .. code-block:: llvm
6736 %X = bitcast i8 255 to i8 ; yields i8 :-1
6737 %Y = bitcast i32* %x to sint* ; yields sint*:%x
6738 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
6739 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
6741 .. _i_addrspacecast:
6743 '``addrspacecast .. to``' Instruction
6744 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6751 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
6756 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
6757 address space ``n`` to type ``pty2`` in address space ``m``.
6762 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
6763 to cast and a pointer type to cast it to, which must have a different
6769 The '``addrspacecast``' instruction converts the pointer value
6770 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
6771 value modification, depending on the target and the address space
6772 pair. Pointer conversions within the same address space must be
6773 performed with the ``bitcast`` instruction. Note that if the address space
6774 conversion is legal then both result and operand refer to the same memory
6780 .. code-block:: llvm
6782 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
6783 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
6784 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
6791 The instructions in this category are the "miscellaneous" instructions,
6792 which defy better classification.
6796 '``icmp``' Instruction
6797 ^^^^^^^^^^^^^^^^^^^^^^
6804 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6809 The '``icmp``' instruction returns a boolean value or a vector of
6810 boolean values based on comparison of its two integer, integer vector,
6811 pointer, or pointer vector operands.
6816 The '``icmp``' instruction takes three operands. The first operand is
6817 the condition code indicating the kind of comparison to perform. It is
6818 not a value, just a keyword. The possible condition code are:
6821 #. ``ne``: not equal
6822 #. ``ugt``: unsigned greater than
6823 #. ``uge``: unsigned greater or equal
6824 #. ``ult``: unsigned less than
6825 #. ``ule``: unsigned less or equal
6826 #. ``sgt``: signed greater than
6827 #. ``sge``: signed greater or equal
6828 #. ``slt``: signed less than
6829 #. ``sle``: signed less or equal
6831 The remaining two arguments must be :ref:`integer <t_integer>` or
6832 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
6833 must also be identical types.
6838 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
6839 code given as ``cond``. The comparison performed always yields either an
6840 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
6842 #. ``eq``: yields ``true`` if the operands are equal, ``false``
6843 otherwise. No sign interpretation is necessary or performed.
6844 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
6845 otherwise. No sign interpretation is necessary or performed.
6846 #. ``ugt``: interprets the operands as unsigned values and yields
6847 ``true`` if ``op1`` is greater than ``op2``.
6848 #. ``uge``: interprets the operands as unsigned values and yields
6849 ``true`` if ``op1`` is greater than or equal to ``op2``.
6850 #. ``ult``: interprets the operands as unsigned values and yields
6851 ``true`` if ``op1`` is less than ``op2``.
6852 #. ``ule``: interprets the operands as unsigned values and yields
6853 ``true`` if ``op1`` is less than or equal to ``op2``.
6854 #. ``sgt``: interprets the operands as signed values and yields ``true``
6855 if ``op1`` is greater than ``op2``.
6856 #. ``sge``: interprets the operands as signed values and yields ``true``
6857 if ``op1`` is greater than or equal to ``op2``.
6858 #. ``slt``: interprets the operands as signed values and yields ``true``
6859 if ``op1`` is less than ``op2``.
6860 #. ``sle``: interprets the operands as signed values and yields ``true``
6861 if ``op1`` is less than or equal to ``op2``.
6863 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
6864 are compared as if they were integers.
6866 If the operands are integer vectors, then they are compared element by
6867 element. The result is an ``i1`` vector with the same number of elements
6868 as the values being compared. Otherwise, the result is an ``i1``.
6873 .. code-block:: llvm
6875 <result> = icmp eq i32 4, 5 ; yields: result=false
6876 <result> = icmp ne float* %X, %X ; yields: result=false
6877 <result> = icmp ult i16 4, 5 ; yields: result=true
6878 <result> = icmp sgt i16 4, 5 ; yields: result=false
6879 <result> = icmp ule i16 -4, 5 ; yields: result=false
6880 <result> = icmp sge i16 4, 5 ; yields: result=false
6882 Note that the code generator does not yet support vector types with the
6883 ``icmp`` instruction.
6887 '``fcmp``' Instruction
6888 ^^^^^^^^^^^^^^^^^^^^^^
6895 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6900 The '``fcmp``' instruction returns a boolean value or vector of boolean
6901 values based on comparison of its operands.
6903 If the operands are floating point scalars, then the result type is a
6904 boolean (:ref:`i1 <t_integer>`).
6906 If the operands are floating point vectors, then the result type is a
6907 vector of boolean with the same number of elements as the operands being
6913 The '``fcmp``' instruction takes three operands. The first operand is
6914 the condition code indicating the kind of comparison to perform. It is
6915 not a value, just a keyword. The possible condition code are:
6917 #. ``false``: no comparison, always returns false
6918 #. ``oeq``: ordered and equal
6919 #. ``ogt``: ordered and greater than
6920 #. ``oge``: ordered and greater than or equal
6921 #. ``olt``: ordered and less than
6922 #. ``ole``: ordered and less than or equal
6923 #. ``one``: ordered and not equal
6924 #. ``ord``: ordered (no nans)
6925 #. ``ueq``: unordered or equal
6926 #. ``ugt``: unordered or greater than
6927 #. ``uge``: unordered or greater than or equal
6928 #. ``ult``: unordered or less than
6929 #. ``ule``: unordered or less than or equal
6930 #. ``une``: unordered or not equal
6931 #. ``uno``: unordered (either nans)
6932 #. ``true``: no comparison, always returns true
6934 *Ordered* means that neither operand is a QNAN while *unordered* means
6935 that either operand may be a QNAN.
6937 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6938 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6939 type. They must have identical types.
6944 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6945 condition code given as ``cond``. If the operands are vectors, then the
6946 vectors are compared element by element. Each comparison performed
6947 always yields an :ref:`i1 <t_integer>` result, as follows:
6949 #. ``false``: always yields ``false``, regardless of operands.
6950 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6951 is equal to ``op2``.
6952 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6953 is greater than ``op2``.
6954 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6955 is greater than or equal to ``op2``.
6956 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6957 is less than ``op2``.
6958 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6959 is less than or equal to ``op2``.
6960 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6961 is not equal to ``op2``.
6962 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6963 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6965 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6966 greater than ``op2``.
6967 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6968 greater than or equal to ``op2``.
6969 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6971 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6972 less than or equal to ``op2``.
6973 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6974 not equal to ``op2``.
6975 #. ``uno``: yields ``true`` if either operand is a QNAN.
6976 #. ``true``: always yields ``true``, regardless of operands.
6981 .. code-block:: llvm
6983 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6984 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6985 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6986 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6988 Note that the code generator does not yet support vector types with the
6989 ``fcmp`` instruction.
6993 '``phi``' Instruction
6994 ^^^^^^^^^^^^^^^^^^^^^
7001 <result> = phi <ty> [ <val0>, <label0>], ...
7006 The '``phi``' instruction is used to implement the φ node in the SSA
7007 graph representing the function.
7012 The type of the incoming values is specified with the first type field.
7013 After this, the '``phi``' instruction takes a list of pairs as
7014 arguments, with one pair for each predecessor basic block of the current
7015 block. Only values of :ref:`first class <t_firstclass>` type may be used as
7016 the value arguments to the PHI node. Only labels may be used as the
7019 There must be no non-phi instructions between the start of a basic block
7020 and the PHI instructions: i.e. PHI instructions must be first in a basic
7023 For the purposes of the SSA form, the use of each incoming value is
7024 deemed to occur on the edge from the corresponding predecessor block to
7025 the current block (but after any definition of an '``invoke``'
7026 instruction's return value on the same edge).
7031 At runtime, the '``phi``' instruction logically takes on the value
7032 specified by the pair corresponding to the predecessor basic block that
7033 executed just prior to the current block.
7038 .. code-block:: llvm
7040 Loop: ; Infinite loop that counts from 0 on up...
7041 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
7042 %nextindvar = add i32 %indvar, 1
7047 '``select``' Instruction
7048 ^^^^^^^^^^^^^^^^^^^^^^^^
7055 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
7057 selty is either i1 or {<N x i1>}
7062 The '``select``' instruction is used to choose one value based on a
7063 condition, without IR-level branching.
7068 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
7069 values indicating the condition, and two values of the same :ref:`first
7070 class <t_firstclass>` type.
7075 If the condition is an i1 and it evaluates to 1, the instruction returns
7076 the first value argument; otherwise, it returns the second value
7079 If the condition is a vector of i1, then the value arguments must be
7080 vectors of the same size, and the selection is done element by element.
7082 If the condition is an i1 and the value arguments are vectors of the
7083 same size, then an entire vector is selected.
7088 .. code-block:: llvm
7090 %X = select i1 true, i8 17, i8 42 ; yields i8:17
7094 '``call``' Instruction
7095 ^^^^^^^^^^^^^^^^^^^^^^
7102 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
7107 The '``call``' instruction represents a simple function call.
7112 This instruction requires several arguments:
7114 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
7115 should perform tail call optimization. The ``tail`` marker is a hint that
7116 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
7117 means that the call must be tail call optimized in order for the program to
7118 be correct. The ``musttail`` marker provides these guarantees:
7120 #. The call will not cause unbounded stack growth if it is part of a
7121 recursive cycle in the call graph.
7122 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
7125 Both markers imply that the callee does not access allocas or varargs from
7126 the caller. Calls marked ``musttail`` must obey the following additional
7129 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
7130 or a pointer bitcast followed by a ret instruction.
7131 - The ret instruction must return the (possibly bitcasted) value
7132 produced by the call or void.
7133 - The caller and callee prototypes must match. Pointer types of
7134 parameters or return types may differ in pointee type, but not
7136 - The calling conventions of the caller and callee must match.
7137 - All ABI-impacting function attributes, such as sret, byval, inreg,
7138 returned, and inalloca, must match.
7139 - The callee must be varargs iff the caller is varargs. Bitcasting a
7140 non-varargs function to the appropriate varargs type is legal so
7141 long as the non-varargs prefixes obey the other rules.
7143 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
7144 the following conditions are met:
7146 - Caller and callee both have the calling convention ``fastcc``.
7147 - The call is in tail position (ret immediately follows call and ret
7148 uses value of call or is void).
7149 - Option ``-tailcallopt`` is enabled, or
7150 ``llvm::GuaranteedTailCallOpt`` is ``true``.
7151 - `Platform-specific constraints are
7152 met. <CodeGenerator.html#tailcallopt>`_
7154 #. The optional "cconv" marker indicates which :ref:`calling
7155 convention <callingconv>` the call should use. If none is
7156 specified, the call defaults to using C calling conventions. The
7157 calling convention of the call must match the calling convention of
7158 the target function, or else the behavior is undefined.
7159 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
7160 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
7162 #. '``ty``': the type of the call instruction itself which is also the
7163 type of the return value. Functions that return no value are marked
7165 #. '``fnty``': shall be the signature of the pointer to function value
7166 being invoked. The argument types must match the types implied by
7167 this signature. This type can be omitted if the function is not
7168 varargs and if the function type does not return a pointer to a
7170 #. '``fnptrval``': An LLVM value containing a pointer to a function to
7171 be invoked. In most cases, this is a direct function invocation, but
7172 indirect ``call``'s are just as possible, calling an arbitrary pointer
7174 #. '``function args``': argument list whose types match the function
7175 signature argument types and parameter attributes. All arguments must
7176 be of :ref:`first class <t_firstclass>` type. If the function signature
7177 indicates the function accepts a variable number of arguments, the
7178 extra arguments can be specified.
7179 #. The optional :ref:`function attributes <fnattrs>` list. Only
7180 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
7181 attributes are valid here.
7186 The '``call``' instruction is used to cause control flow to transfer to
7187 a specified function, with its incoming arguments bound to the specified
7188 values. Upon a '``ret``' instruction in the called function, control
7189 flow continues with the instruction after the function call, and the
7190 return value of the function is bound to the result argument.
7195 .. code-block:: llvm
7197 %retval = call i32 @test(i32 %argc)
7198 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
7199 %X = tail call i32 @foo() ; yields i32
7200 %Y = tail call fastcc i32 @foo() ; yields i32
7201 call void %foo(i8 97 signext)
7203 %struct.A = type { i32, i8 }
7204 %r = call %struct.A @foo() ; yields { i32, i8 }
7205 %gr = extractvalue %struct.A %r, 0 ; yields i32
7206 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
7207 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
7208 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
7210 llvm treats calls to some functions with names and arguments that match
7211 the standard C99 library as being the C99 library functions, and may
7212 perform optimizations or generate code for them under that assumption.
7213 This is something we'd like to change in the future to provide better
7214 support for freestanding environments and non-C-based languages.
7218 '``va_arg``' Instruction
7219 ^^^^^^^^^^^^^^^^^^^^^^^^
7226 <resultval> = va_arg <va_list*> <arglist>, <argty>
7231 The '``va_arg``' instruction is used to access arguments passed through
7232 the "variable argument" area of a function call. It is used to implement
7233 the ``va_arg`` macro in C.
7238 This instruction takes a ``va_list*`` value and the type of the
7239 argument. It returns a value of the specified argument type and
7240 increments the ``va_list`` to point to the next argument. The actual
7241 type of ``va_list`` is target specific.
7246 The '``va_arg``' instruction loads an argument of the specified type
7247 from the specified ``va_list`` and causes the ``va_list`` to point to
7248 the next argument. For more information, see the variable argument
7249 handling :ref:`Intrinsic Functions <int_varargs>`.
7251 It is legal for this instruction to be called in a function which does
7252 not take a variable number of arguments, for example, the ``vfprintf``
7255 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
7256 function <intrinsics>` because it takes a type as an argument.
7261 See the :ref:`variable argument processing <int_varargs>` section.
7263 Note that the code generator does not yet fully support va\_arg on many
7264 targets. Also, it does not currently support va\_arg with aggregate
7265 types on any target.
7269 '``landingpad``' Instruction
7270 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7277 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
7278 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
7280 <clause> := catch <type> <value>
7281 <clause> := filter <array constant type> <array constant>
7286 The '``landingpad``' instruction is used by `LLVM's exception handling
7287 system <ExceptionHandling.html#overview>`_ to specify that a basic block
7288 is a landing pad --- one where the exception lands, and corresponds to the
7289 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
7290 defines values supplied by the personality function (``pers_fn``) upon
7291 re-entry to the function. The ``resultval`` has the type ``resultty``.
7296 This instruction takes a ``pers_fn`` value. This is the personality
7297 function associated with the unwinding mechanism. The optional
7298 ``cleanup`` flag indicates that the landing pad block is a cleanup.
7300 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
7301 contains the global variable representing the "type" that may be caught
7302 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
7303 clause takes an array constant as its argument. Use
7304 "``[0 x i8**] undef``" for a filter which cannot throw. The
7305 '``landingpad``' instruction must contain *at least* one ``clause`` or
7306 the ``cleanup`` flag.
7311 The '``landingpad``' instruction defines the values which are set by the
7312 personality function (``pers_fn``) upon re-entry to the function, and
7313 therefore the "result type" of the ``landingpad`` instruction. As with
7314 calling conventions, how the personality function results are
7315 represented in LLVM IR is target specific.
7317 The clauses are applied in order from top to bottom. If two
7318 ``landingpad`` instructions are merged together through inlining, the
7319 clauses from the calling function are appended to the list of clauses.
7320 When the call stack is being unwound due to an exception being thrown,
7321 the exception is compared against each ``clause`` in turn. If it doesn't
7322 match any of the clauses, and the ``cleanup`` flag is not set, then
7323 unwinding continues further up the call stack.
7325 The ``landingpad`` instruction has several restrictions:
7327 - A landing pad block is a basic block which is the unwind destination
7328 of an '``invoke``' instruction.
7329 - A landing pad block must have a '``landingpad``' instruction as its
7330 first non-PHI instruction.
7331 - There can be only one '``landingpad``' instruction within the landing
7333 - A basic block that is not a landing pad block may not include a
7334 '``landingpad``' instruction.
7335 - All '``landingpad``' instructions in a function must have the same
7336 personality function.
7341 .. code-block:: llvm
7343 ;; A landing pad which can catch an integer.
7344 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7346 ;; A landing pad that is a cleanup.
7347 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7349 ;; A landing pad which can catch an integer and can only throw a double.
7350 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7352 filter [1 x i8**] [@_ZTId]
7359 LLVM supports the notion of an "intrinsic function". These functions
7360 have well known names and semantics and are required to follow certain
7361 restrictions. Overall, these intrinsics represent an extension mechanism
7362 for the LLVM language that does not require changing all of the
7363 transformations in LLVM when adding to the language (or the bitcode
7364 reader/writer, the parser, etc...).
7366 Intrinsic function names must all start with an "``llvm.``" prefix. This
7367 prefix is reserved in LLVM for intrinsic names; thus, function names may
7368 not begin with this prefix. Intrinsic functions must always be external
7369 functions: you cannot define the body of intrinsic functions. Intrinsic
7370 functions may only be used in call or invoke instructions: it is illegal
7371 to take the address of an intrinsic function. Additionally, because
7372 intrinsic functions are part of the LLVM language, it is required if any
7373 are added that they be documented here.
7375 Some intrinsic functions can be overloaded, i.e., the intrinsic
7376 represents a family of functions that perform the same operation but on
7377 different data types. Because LLVM can represent over 8 million
7378 different integer types, overloading is used commonly to allow an
7379 intrinsic function to operate on any integer type. One or more of the
7380 argument types or the result type can be overloaded to accept any
7381 integer type. Argument types may also be defined as exactly matching a
7382 previous argument's type or the result type. This allows an intrinsic
7383 function which accepts multiple arguments, but needs all of them to be
7384 of the same type, to only be overloaded with respect to a single
7385 argument or the result.
7387 Overloaded intrinsics will have the names of its overloaded argument
7388 types encoded into its function name, each preceded by a period. Only
7389 those types which are overloaded result in a name suffix. Arguments
7390 whose type is matched against another type do not. For example, the
7391 ``llvm.ctpop`` function can take an integer of any width and returns an
7392 integer of exactly the same integer width. This leads to a family of
7393 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
7394 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
7395 overloaded, and only one type suffix is required. Because the argument's
7396 type is matched against the return type, it does not require its own
7399 To learn how to add an intrinsic function, please see the `Extending
7400 LLVM Guide <ExtendingLLVM.html>`_.
7404 Variable Argument Handling Intrinsics
7405 -------------------------------------
7407 Variable argument support is defined in LLVM with the
7408 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
7409 functions. These functions are related to the similarly named macros
7410 defined in the ``<stdarg.h>`` header file.
7412 All of these functions operate on arguments that use a target-specific
7413 value type "``va_list``". The LLVM assembly language reference manual
7414 does not define what this type is, so all transformations should be
7415 prepared to handle these functions regardless of the type used.
7417 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
7418 variable argument handling intrinsic functions are used.
7420 .. code-block:: llvm
7422 ; This struct is different for every platform. For most platforms,
7423 ; it is merely an i8*.
7424 %struct.va_list = type { i8* }
7426 ; For Unix x86_64 platforms, va_list is the following struct:
7427 ; %struct.va_list = type { i32, i32, i8*, i8* }
7429 define i32 @test(i32 %X, ...) {
7430 ; Initialize variable argument processing
7431 %ap = alloca %struct.va_list
7432 %ap2 = bitcast %struct.va_list* %ap to i8*
7433 call void @llvm.va_start(i8* %ap2)
7435 ; Read a single integer argument
7436 %tmp = va_arg i8* %ap2, i32
7438 ; Demonstrate usage of llvm.va_copy and llvm.va_end
7440 %aq2 = bitcast i8** %aq to i8*
7441 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
7442 call void @llvm.va_end(i8* %aq2)
7444 ; Stop processing of arguments.
7445 call void @llvm.va_end(i8* %ap2)
7449 declare void @llvm.va_start(i8*)
7450 declare void @llvm.va_copy(i8*, i8*)
7451 declare void @llvm.va_end(i8*)
7455 '``llvm.va_start``' Intrinsic
7456 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7463 declare void @llvm.va_start(i8* <arglist>)
7468 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
7469 subsequent use by ``va_arg``.
7474 The argument is a pointer to a ``va_list`` element to initialize.
7479 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
7480 available in C. In a target-dependent way, it initializes the
7481 ``va_list`` element to which the argument points, so that the next call
7482 to ``va_arg`` will produce the first variable argument passed to the
7483 function. Unlike the C ``va_start`` macro, this intrinsic does not need
7484 to know the last argument of the function as the compiler can figure
7487 '``llvm.va_end``' Intrinsic
7488 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7495 declare void @llvm.va_end(i8* <arglist>)
7500 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
7501 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
7506 The argument is a pointer to a ``va_list`` to destroy.
7511 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
7512 available in C. In a target-dependent way, it destroys the ``va_list``
7513 element to which the argument points. Calls to
7514 :ref:`llvm.va_start <int_va_start>` and
7515 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
7520 '``llvm.va_copy``' Intrinsic
7521 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7528 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
7533 The '``llvm.va_copy``' intrinsic copies the current argument position
7534 from the source argument list to the destination argument list.
7539 The first argument is a pointer to a ``va_list`` element to initialize.
7540 The second argument is a pointer to a ``va_list`` element to copy from.
7545 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
7546 available in C. In a target-dependent way, it copies the source
7547 ``va_list`` element into the destination ``va_list`` element. This
7548 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
7549 arbitrarily complex and require, for example, memory allocation.
7551 Accurate Garbage Collection Intrinsics
7552 --------------------------------------
7554 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
7555 (GC) requires the frontend to generate code containing appropriate intrinsic
7556 calls and select an appropriate GC strategy which knows how to lower these
7557 intrinsics in a manner which is appropriate for the target collector.
7559 These intrinsics allow identification of :ref:`GC roots on the
7560 stack <int_gcroot>`, as well as garbage collector implementations that
7561 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
7562 Frontends for type-safe garbage collected languages should generate
7563 these intrinsics to make use of the LLVM garbage collectors. For more
7564 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
7566 Experimental Statepoint Intrinsics
7567 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7569 LLVM provides an second experimental set of intrinsics for describing garbage
7570 collection safepoints in compiled code. These intrinsics are an alternative
7571 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
7572 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
7573 differences in approach are covered in the `Garbage Collection with LLVM
7574 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
7575 described in :doc:`Statepoints`.
7579 '``llvm.gcroot``' Intrinsic
7580 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7587 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
7592 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
7593 the code generator, and allows some metadata to be associated with it.
7598 The first argument specifies the address of a stack object that contains
7599 the root pointer. The second pointer (which must be either a constant or
7600 a global value address) contains the meta-data to be associated with the
7606 At runtime, a call to this intrinsic stores a null pointer into the
7607 "ptrloc" location. At compile-time, the code generator generates
7608 information to allow the runtime to find the pointer at GC safe points.
7609 The '``llvm.gcroot``' intrinsic may only be used in a function which
7610 :ref:`specifies a GC algorithm <gc>`.
7614 '``llvm.gcread``' Intrinsic
7615 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7622 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
7627 The '``llvm.gcread``' intrinsic identifies reads of references from heap
7628 locations, allowing garbage collector implementations that require read
7634 The second argument is the address to read from, which should be an
7635 address allocated from the garbage collector. The first object is a
7636 pointer to the start of the referenced object, if needed by the language
7637 runtime (otherwise null).
7642 The '``llvm.gcread``' intrinsic has the same semantics as a load
7643 instruction, but may be replaced with substantially more complex code by
7644 the garbage collector runtime, as needed. The '``llvm.gcread``'
7645 intrinsic may only be used in a function which :ref:`specifies a GC
7650 '``llvm.gcwrite``' Intrinsic
7651 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7658 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
7663 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
7664 locations, allowing garbage collector implementations that require write
7665 barriers (such as generational or reference counting collectors).
7670 The first argument is the reference to store, the second is the start of
7671 the object to store it to, and the third is the address of the field of
7672 Obj to store to. If the runtime does not require a pointer to the
7673 object, Obj may be null.
7678 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
7679 instruction, but may be replaced with substantially more complex code by
7680 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
7681 intrinsic may only be used in a function which :ref:`specifies a GC
7684 Code Generator Intrinsics
7685 -------------------------
7687 These intrinsics are provided by LLVM to expose special features that
7688 may only be implemented with code generator support.
7690 '``llvm.returnaddress``' Intrinsic
7691 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7698 declare i8 *@llvm.returnaddress(i32 <level>)
7703 The '``llvm.returnaddress``' intrinsic attempts to compute a
7704 target-specific value indicating the return address of the current
7705 function or one of its callers.
7710 The argument to this intrinsic indicates which function to return the
7711 address for. Zero indicates the calling function, one indicates its
7712 caller, etc. The argument is **required** to be a constant integer
7718 The '``llvm.returnaddress``' intrinsic either returns a pointer
7719 indicating the return address of the specified call frame, or zero if it
7720 cannot be identified. The value returned by this intrinsic is likely to
7721 be incorrect or 0 for arguments other than zero, so it should only be
7722 used for debugging purposes.
7724 Note that calling this intrinsic does not prevent function inlining or
7725 other aggressive transformations, so the value returned may not be that
7726 of the obvious source-language caller.
7728 '``llvm.frameaddress``' Intrinsic
7729 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7736 declare i8* @llvm.frameaddress(i32 <level>)
7741 The '``llvm.frameaddress``' intrinsic attempts to return the
7742 target-specific frame pointer value for the specified stack frame.
7747 The argument to this intrinsic indicates which function to return the
7748 frame pointer for. Zero indicates the calling function, one indicates
7749 its caller, etc. The argument is **required** to be a constant integer
7755 The '``llvm.frameaddress``' intrinsic either returns a pointer
7756 indicating the frame address of the specified call frame, or zero if it
7757 cannot be identified. The value returned by this intrinsic is likely to
7758 be incorrect or 0 for arguments other than zero, so it should only be
7759 used for debugging purposes.
7761 Note that calling this intrinsic does not prevent function inlining or
7762 other aggressive transformations, so the value returned may not be that
7763 of the obvious source-language caller.
7765 '``llvm.frameescape``' and '``llvm.framerecover``' Intrinsics
7766 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7773 declare void @llvm.frameescape(...)
7774 declare i8* @llvm.framerecover(i8* %func, i8* %fp, i32 %idx)
7779 The '``llvm.frameescape``' intrinsic escapes offsets of a collection of static
7780 allocas, and the '``llvm.framerecover``' intrinsic applies those offsets to a
7781 live frame pointer to recover the address of the allocation. The offset is
7782 computed during frame layout of the caller of ``llvm.frameescape``.
7787 All arguments to '``llvm.frameescape``' must be pointers to static allocas or
7788 casts of static allocas. Each function can only call '``llvm.frameescape``'
7789 once, and it can only do so from the entry block.
7791 The ``func`` argument to '``llvm.framerecover``' must be a constant
7792 bitcasted pointer to a function defined in the current module. The code
7793 generator cannot determine the frame allocation offset of functions defined in
7796 The ``fp`` argument to '``llvm.framerecover``' must be a frame
7797 pointer of a call frame that is currently live. The return value of
7798 '``llvm.frameaddress``' is one way to produce such a value, but most platforms
7799 also expose the frame pointer through stack unwinding mechanisms.
7801 The ``idx`` argument to '``llvm.framerecover``' indicates which alloca passed to
7802 '``llvm.frameescape``' to recover. It is zero-indexed.
7807 These intrinsics allow a group of functions to access one stack memory
7808 allocation in an ancestor stack frame. The memory returned from
7809 '``llvm.frameallocate``' may be allocated prior to stack realignment, so the
7810 memory is only aligned to the ABI-required stack alignment. Each function may
7811 only call '``llvm.frameallocate``' one or zero times from the function entry
7812 block. The frame allocation intrinsic inhibits inlining, as any frame
7813 allocations in the inlined function frame are likely to be at a different
7814 offset from the one used by '``llvm.framerecover``' called with the
7817 .. _int_read_register:
7818 .. _int_write_register:
7820 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
7821 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7828 declare i32 @llvm.read_register.i32(metadata)
7829 declare i64 @llvm.read_register.i64(metadata)
7830 declare void @llvm.write_register.i32(metadata, i32 @value)
7831 declare void @llvm.write_register.i64(metadata, i64 @value)
7837 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
7838 provides access to the named register. The register must be valid on
7839 the architecture being compiled to. The type needs to be compatible
7840 with the register being read.
7845 The '``llvm.read_register``' intrinsic returns the current value of the
7846 register, where possible. The '``llvm.write_register``' intrinsic sets
7847 the current value of the register, where possible.
7849 This is useful to implement named register global variables that need
7850 to always be mapped to a specific register, as is common practice on
7851 bare-metal programs including OS kernels.
7853 The compiler doesn't check for register availability or use of the used
7854 register in surrounding code, including inline assembly. Because of that,
7855 allocatable registers are not supported.
7857 Warning: So far it only works with the stack pointer on selected
7858 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
7859 work is needed to support other registers and even more so, allocatable
7864 '``llvm.stacksave``' Intrinsic
7865 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7872 declare i8* @llvm.stacksave()
7877 The '``llvm.stacksave``' intrinsic is used to remember the current state
7878 of the function stack, for use with
7879 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
7880 implementing language features like scoped automatic variable sized
7886 This intrinsic returns a opaque pointer value that can be passed to
7887 :ref:`llvm.stackrestore <int_stackrestore>`. When an
7888 ``llvm.stackrestore`` intrinsic is executed with a value saved from
7889 ``llvm.stacksave``, it effectively restores the state of the stack to
7890 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
7891 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
7892 were allocated after the ``llvm.stacksave`` was executed.
7894 .. _int_stackrestore:
7896 '``llvm.stackrestore``' Intrinsic
7897 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7904 declare void @llvm.stackrestore(i8* %ptr)
7909 The '``llvm.stackrestore``' intrinsic is used to restore the state of
7910 the function stack to the state it was in when the corresponding
7911 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
7912 useful for implementing language features like scoped automatic variable
7913 sized arrays in C99.
7918 See the description for :ref:`llvm.stacksave <int_stacksave>`.
7920 '``llvm.prefetch``' Intrinsic
7921 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7928 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7933 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
7934 insert a prefetch instruction if supported; otherwise, it is a noop.
7935 Prefetches have no effect on the behavior of the program but can change
7936 its performance characteristics.
7941 ``address`` is the address to be prefetched, ``rw`` is the specifier
7942 determining if the fetch should be for a read (0) or write (1), and
7943 ``locality`` is a temporal locality specifier ranging from (0) - no
7944 locality, to (3) - extremely local keep in cache. The ``cache type``
7945 specifies whether the prefetch is performed on the data (1) or
7946 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
7947 arguments must be constant integers.
7952 This intrinsic does not modify the behavior of the program. In
7953 particular, prefetches cannot trap and do not produce a value. On
7954 targets that support this intrinsic, the prefetch can provide hints to
7955 the processor cache for better performance.
7957 '``llvm.pcmarker``' Intrinsic
7958 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7965 declare void @llvm.pcmarker(i32 <id>)
7970 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7971 Counter (PC) in a region of code to simulators and other tools. The
7972 method is target specific, but it is expected that the marker will use
7973 exported symbols to transmit the PC of the marker. The marker makes no
7974 guarantees that it will remain with any specific instruction after
7975 optimizations. It is possible that the presence of a marker will inhibit
7976 optimizations. The intended use is to be inserted after optimizations to
7977 allow correlations of simulation runs.
7982 ``id`` is a numerical id identifying the marker.
7987 This intrinsic does not modify the behavior of the program. Backends
7988 that do not support this intrinsic may ignore it.
7990 '``llvm.readcyclecounter``' Intrinsic
7991 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7998 declare i64 @llvm.readcyclecounter()
8003 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
8004 counter register (or similar low latency, high accuracy clocks) on those
8005 targets that support it. On X86, it should map to RDTSC. On Alpha, it
8006 should map to RPCC. As the backing counters overflow quickly (on the
8007 order of 9 seconds on alpha), this should only be used for small
8013 When directly supported, reading the cycle counter should not modify any
8014 memory. Implementations are allowed to either return a application
8015 specific value or a system wide value. On backends without support, this
8016 is lowered to a constant 0.
8018 Note that runtime support may be conditional on the privilege-level code is
8019 running at and the host platform.
8021 '``llvm.clear_cache``' Intrinsic
8022 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8029 declare void @llvm.clear_cache(i8*, i8*)
8034 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
8035 in the specified range to the execution unit of the processor. On
8036 targets with non-unified instruction and data cache, the implementation
8037 flushes the instruction cache.
8042 On platforms with coherent instruction and data caches (e.g. x86), this
8043 intrinsic is a nop. On platforms with non-coherent instruction and data
8044 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
8045 instructions or a system call, if cache flushing requires special
8048 The default behavior is to emit a call to ``__clear_cache`` from the run
8051 This instrinsic does *not* empty the instruction pipeline. Modifications
8052 of the current function are outside the scope of the intrinsic.
8054 '``llvm.instrprof_increment``' Intrinsic
8055 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8062 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
8063 i32 <num-counters>, i32 <index>)
8068 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
8069 frontend for use with instrumentation based profiling. These will be
8070 lowered by the ``-instrprof`` pass to generate execution counts of a
8076 The first argument is a pointer to a global variable containing the
8077 name of the entity being instrumented. This should generally be the
8078 (mangled) function name for a set of counters.
8080 The second argument is a hash value that can be used by the consumer
8081 of the profile data to detect changes to the instrumented source, and
8082 the third is the number of counters associated with ``name``. It is an
8083 error if ``hash`` or ``num-counters`` differ between two instances of
8084 ``instrprof_increment`` that refer to the same name.
8086 The last argument refers to which of the counters for ``name`` should
8087 be incremented. It should be a value between 0 and ``num-counters``.
8092 This intrinsic represents an increment of a profiling counter. It will
8093 cause the ``-instrprof`` pass to generate the appropriate data
8094 structures and the code to increment the appropriate value, in a
8095 format that can be written out by a compiler runtime and consumed via
8096 the ``llvm-profdata`` tool.
8098 Standard C Library Intrinsics
8099 -----------------------------
8101 LLVM provides intrinsics for a few important standard C library
8102 functions. These intrinsics allow source-language front-ends to pass
8103 information about the alignment of the pointer arguments to the code
8104 generator, providing opportunity for more efficient code generation.
8108 '``llvm.memcpy``' Intrinsic
8109 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8114 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
8115 integer bit width and for different address spaces. Not all targets
8116 support all bit widths however.
8120 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
8121 i32 <len>, i32 <align>, i1 <isvolatile>)
8122 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
8123 i64 <len>, i32 <align>, i1 <isvolatile>)
8128 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
8129 source location to the destination location.
8131 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
8132 intrinsics do not return a value, takes extra alignment/isvolatile
8133 arguments and the pointers can be in specified address spaces.
8138 The first argument is a pointer to the destination, the second is a
8139 pointer to the source. The third argument is an integer argument
8140 specifying the number of bytes to copy, the fourth argument is the
8141 alignment of the source and destination locations, and the fifth is a
8142 boolean indicating a volatile access.
8144 If the call to this intrinsic has an alignment value that is not 0 or 1,
8145 then the caller guarantees that both the source and destination pointers
8146 are aligned to that boundary.
8148 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
8149 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
8150 very cleanly specified and it is unwise to depend on it.
8155 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
8156 source location to the destination location, which are not allowed to
8157 overlap. It copies "len" bytes of memory over. If the argument is known
8158 to be aligned to some boundary, this can be specified as the fourth
8159 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
8161 '``llvm.memmove``' Intrinsic
8162 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8167 This is an overloaded intrinsic. You can use llvm.memmove on any integer
8168 bit width and for different address space. Not all targets support all
8173 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
8174 i32 <len>, i32 <align>, i1 <isvolatile>)
8175 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
8176 i64 <len>, i32 <align>, i1 <isvolatile>)
8181 The '``llvm.memmove.*``' intrinsics move a block of memory from the
8182 source location to the destination location. It is similar to the
8183 '``llvm.memcpy``' intrinsic but allows the two memory locations to
8186 Note that, unlike the standard libc function, the ``llvm.memmove.*``
8187 intrinsics do not return a value, takes extra alignment/isvolatile
8188 arguments and the pointers can be in specified address spaces.
8193 The first argument is a pointer to the destination, the second is a
8194 pointer to the source. The third argument is an integer argument
8195 specifying the number of bytes to copy, the fourth argument is the
8196 alignment of the source and destination locations, and the fifth is a
8197 boolean indicating a volatile access.
8199 If the call to this intrinsic has an alignment value that is not 0 or 1,
8200 then the caller guarantees that the source and destination pointers are
8201 aligned to that boundary.
8203 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
8204 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
8205 not very cleanly specified and it is unwise to depend on it.
8210 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
8211 source location to the destination location, which may overlap. It
8212 copies "len" bytes of memory over. If the argument is known to be
8213 aligned to some boundary, this can be specified as the fourth argument,
8214 otherwise it should be set to 0 or 1 (both meaning no alignment).
8216 '``llvm.memset.*``' Intrinsics
8217 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8222 This is an overloaded intrinsic. You can use llvm.memset on any integer
8223 bit width and for different address spaces. However, not all targets
8224 support all bit widths.
8228 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
8229 i32 <len>, i32 <align>, i1 <isvolatile>)
8230 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
8231 i64 <len>, i32 <align>, i1 <isvolatile>)
8236 The '``llvm.memset.*``' intrinsics fill a block of memory with a
8237 particular byte value.
8239 Note that, unlike the standard libc function, the ``llvm.memset``
8240 intrinsic does not return a value and takes extra alignment/volatile
8241 arguments. Also, the destination can be in an arbitrary address space.
8246 The first argument is a pointer to the destination to fill, the second
8247 is the byte value with which to fill it, the third argument is an
8248 integer argument specifying the number of bytes to fill, and the fourth
8249 argument is the known alignment of the destination location.
8251 If the call to this intrinsic has an alignment value that is not 0 or 1,
8252 then the caller guarantees that the destination pointer is aligned to
8255 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
8256 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
8257 very cleanly specified and it is unwise to depend on it.
8262 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
8263 at the destination location. If the argument is known to be aligned to
8264 some boundary, this can be specified as the fourth argument, otherwise
8265 it should be set to 0 or 1 (both meaning no alignment).
8267 '``llvm.sqrt.*``' Intrinsic
8268 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8273 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
8274 floating point or vector of floating point type. Not all targets support
8279 declare float @llvm.sqrt.f32(float %Val)
8280 declare double @llvm.sqrt.f64(double %Val)
8281 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
8282 declare fp128 @llvm.sqrt.f128(fp128 %Val)
8283 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
8288 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
8289 returning the same value as the libm '``sqrt``' functions would. Unlike
8290 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
8291 negative numbers other than -0.0 (which allows for better optimization,
8292 because there is no need to worry about errno being set).
8293 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
8298 The argument and return value are floating point numbers of the same
8304 This function returns the sqrt of the specified operand if it is a
8305 nonnegative floating point number.
8307 '``llvm.powi.*``' Intrinsic
8308 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8313 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
8314 floating point or vector of floating point type. Not all targets support
8319 declare float @llvm.powi.f32(float %Val, i32 %power)
8320 declare double @llvm.powi.f64(double %Val, i32 %power)
8321 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
8322 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
8323 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
8328 The '``llvm.powi.*``' intrinsics return the first operand raised to the
8329 specified (positive or negative) power. The order of evaluation of
8330 multiplications is not defined. When a vector of floating point type is
8331 used, the second argument remains a scalar integer value.
8336 The second argument is an integer power, and the first is a value to
8337 raise to that power.
8342 This function returns the first value raised to the second power with an
8343 unspecified sequence of rounding operations.
8345 '``llvm.sin.*``' Intrinsic
8346 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8351 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
8352 floating point or vector of floating point type. Not all targets support
8357 declare float @llvm.sin.f32(float %Val)
8358 declare double @llvm.sin.f64(double %Val)
8359 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
8360 declare fp128 @llvm.sin.f128(fp128 %Val)
8361 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
8366 The '``llvm.sin.*``' intrinsics return the sine of the operand.
8371 The argument and return value are floating point numbers of the same
8377 This function returns the sine of the specified operand, returning the
8378 same values as the libm ``sin`` functions would, and handles error
8379 conditions in the same way.
8381 '``llvm.cos.*``' Intrinsic
8382 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8387 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
8388 floating point or vector of floating point type. Not all targets support
8393 declare float @llvm.cos.f32(float %Val)
8394 declare double @llvm.cos.f64(double %Val)
8395 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
8396 declare fp128 @llvm.cos.f128(fp128 %Val)
8397 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
8402 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
8407 The argument and return value are floating point numbers of the same
8413 This function returns the cosine of the specified operand, returning the
8414 same values as the libm ``cos`` functions would, and handles error
8415 conditions in the same way.
8417 '``llvm.pow.*``' Intrinsic
8418 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8423 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
8424 floating point or vector of floating point type. Not all targets support
8429 declare float @llvm.pow.f32(float %Val, float %Power)
8430 declare double @llvm.pow.f64(double %Val, double %Power)
8431 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
8432 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
8433 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
8438 The '``llvm.pow.*``' intrinsics return the first operand raised to the
8439 specified (positive or negative) power.
8444 The second argument is a floating point power, and the first is a value
8445 to raise to that power.
8450 This function returns the first value raised to the second power,
8451 returning the same values as the libm ``pow`` functions would, and
8452 handles error conditions in the same way.
8454 '``llvm.exp.*``' Intrinsic
8455 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8460 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
8461 floating point or vector of floating point type. Not all targets support
8466 declare float @llvm.exp.f32(float %Val)
8467 declare double @llvm.exp.f64(double %Val)
8468 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
8469 declare fp128 @llvm.exp.f128(fp128 %Val)
8470 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
8475 The '``llvm.exp.*``' intrinsics perform the exp function.
8480 The argument and return value are floating point numbers of the same
8486 This function returns the same values as the libm ``exp`` functions
8487 would, and handles error conditions in the same way.
8489 '``llvm.exp2.*``' Intrinsic
8490 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8495 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
8496 floating point or vector of floating point type. Not all targets support
8501 declare float @llvm.exp2.f32(float %Val)
8502 declare double @llvm.exp2.f64(double %Val)
8503 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
8504 declare fp128 @llvm.exp2.f128(fp128 %Val)
8505 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
8510 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
8515 The argument and return value are floating point numbers of the same
8521 This function returns the same values as the libm ``exp2`` functions
8522 would, and handles error conditions in the same way.
8524 '``llvm.log.*``' Intrinsic
8525 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8530 This is an overloaded intrinsic. You can use ``llvm.log`` on any
8531 floating point or vector of floating point type. Not all targets support
8536 declare float @llvm.log.f32(float %Val)
8537 declare double @llvm.log.f64(double %Val)
8538 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
8539 declare fp128 @llvm.log.f128(fp128 %Val)
8540 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
8545 The '``llvm.log.*``' intrinsics perform the log function.
8550 The argument and return value are floating point numbers of the same
8556 This function returns the same values as the libm ``log`` functions
8557 would, and handles error conditions in the same way.
8559 '``llvm.log10.*``' Intrinsic
8560 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8565 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
8566 floating point or vector of floating point type. Not all targets support
8571 declare float @llvm.log10.f32(float %Val)
8572 declare double @llvm.log10.f64(double %Val)
8573 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
8574 declare fp128 @llvm.log10.f128(fp128 %Val)
8575 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
8580 The '``llvm.log10.*``' intrinsics perform the log10 function.
8585 The argument and return value are floating point numbers of the same
8591 This function returns the same values as the libm ``log10`` functions
8592 would, and handles error conditions in the same way.
8594 '``llvm.log2.*``' Intrinsic
8595 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8600 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
8601 floating point or vector of floating point type. Not all targets support
8606 declare float @llvm.log2.f32(float %Val)
8607 declare double @llvm.log2.f64(double %Val)
8608 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
8609 declare fp128 @llvm.log2.f128(fp128 %Val)
8610 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
8615 The '``llvm.log2.*``' intrinsics perform the log2 function.
8620 The argument and return value are floating point numbers of the same
8626 This function returns the same values as the libm ``log2`` functions
8627 would, and handles error conditions in the same way.
8629 '``llvm.fma.*``' Intrinsic
8630 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8635 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
8636 floating point or vector of floating point type. Not all targets support
8641 declare float @llvm.fma.f32(float %a, float %b, float %c)
8642 declare double @llvm.fma.f64(double %a, double %b, double %c)
8643 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
8644 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
8645 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
8650 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
8656 The argument and return value are floating point numbers of the same
8662 This function returns the same values as the libm ``fma`` functions
8663 would, and does not set errno.
8665 '``llvm.fabs.*``' Intrinsic
8666 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8671 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
8672 floating point or vector of floating point type. Not all targets support
8677 declare float @llvm.fabs.f32(float %Val)
8678 declare double @llvm.fabs.f64(double %Val)
8679 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
8680 declare fp128 @llvm.fabs.f128(fp128 %Val)
8681 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
8686 The '``llvm.fabs.*``' intrinsics return the absolute value of the
8692 The argument and return value are floating point numbers of the same
8698 This function returns the same values as the libm ``fabs`` functions
8699 would, and handles error conditions in the same way.
8701 '``llvm.minnum.*``' Intrinsic
8702 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8707 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
8708 floating point or vector of floating point type. Not all targets support
8713 declare float @llvm.minnum.f32(float %Val0, float %Val1)
8714 declare double @llvm.minnum.f64(double %Val0, double %Val1)
8715 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8716 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
8717 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8722 The '``llvm.minnum.*``' intrinsics return the minimum of the two
8729 The arguments and return value are floating point numbers of the same
8735 Follows the IEEE-754 semantics for minNum, which also match for libm's
8738 If either operand is a NaN, returns the other non-NaN operand. Returns
8739 NaN only if both operands are NaN. If the operands compare equal,
8740 returns a value that compares equal to both operands. This means that
8741 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8743 '``llvm.maxnum.*``' Intrinsic
8744 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8749 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
8750 floating point or vector of floating point type. Not all targets support
8755 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
8756 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
8757 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8758 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
8759 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8764 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
8771 The arguments and return value are floating point numbers of the same
8776 Follows the IEEE-754 semantics for maxNum, which also match for libm's
8779 If either operand is a NaN, returns the other non-NaN operand. Returns
8780 NaN only if both operands are NaN. If the operands compare equal,
8781 returns a value that compares equal to both operands. This means that
8782 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8784 '``llvm.copysign.*``' Intrinsic
8785 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8790 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
8791 floating point or vector of floating point type. Not all targets support
8796 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
8797 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
8798 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
8799 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
8800 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
8805 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
8806 first operand and the sign of the second operand.
8811 The arguments and return value are floating point numbers of the same
8817 This function returns the same values as the libm ``copysign``
8818 functions would, and handles error conditions in the same way.
8820 '``llvm.floor.*``' Intrinsic
8821 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8826 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
8827 floating point or vector of floating point type. Not all targets support
8832 declare float @llvm.floor.f32(float %Val)
8833 declare double @llvm.floor.f64(double %Val)
8834 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
8835 declare fp128 @llvm.floor.f128(fp128 %Val)
8836 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
8841 The '``llvm.floor.*``' intrinsics return the floor of the operand.
8846 The argument and return value are floating point numbers of the same
8852 This function returns the same values as the libm ``floor`` functions
8853 would, and handles error conditions in the same way.
8855 '``llvm.ceil.*``' Intrinsic
8856 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8861 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
8862 floating point or vector of floating point type. Not all targets support
8867 declare float @llvm.ceil.f32(float %Val)
8868 declare double @llvm.ceil.f64(double %Val)
8869 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
8870 declare fp128 @llvm.ceil.f128(fp128 %Val)
8871 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
8876 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
8881 The argument and return value are floating point numbers of the same
8887 This function returns the same values as the libm ``ceil`` functions
8888 would, and handles error conditions in the same way.
8890 '``llvm.trunc.*``' Intrinsic
8891 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8896 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
8897 floating point or vector of floating point type. Not all targets support
8902 declare float @llvm.trunc.f32(float %Val)
8903 declare double @llvm.trunc.f64(double %Val)
8904 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
8905 declare fp128 @llvm.trunc.f128(fp128 %Val)
8906 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
8911 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
8912 nearest integer not larger in magnitude than the operand.
8917 The argument and return value are floating point numbers of the same
8923 This function returns the same values as the libm ``trunc`` functions
8924 would, and handles error conditions in the same way.
8926 '``llvm.rint.*``' Intrinsic
8927 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8932 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
8933 floating point or vector of floating point type. Not all targets support
8938 declare float @llvm.rint.f32(float %Val)
8939 declare double @llvm.rint.f64(double %Val)
8940 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
8941 declare fp128 @llvm.rint.f128(fp128 %Val)
8942 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
8947 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
8948 nearest integer. It may raise an inexact floating-point exception if the
8949 operand isn't an integer.
8954 The argument and return value are floating point numbers of the same
8960 This function returns the same values as the libm ``rint`` functions
8961 would, and handles error conditions in the same way.
8963 '``llvm.nearbyint.*``' Intrinsic
8964 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8969 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
8970 floating point or vector of floating point type. Not all targets support
8975 declare float @llvm.nearbyint.f32(float %Val)
8976 declare double @llvm.nearbyint.f64(double %Val)
8977 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
8978 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
8979 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
8984 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
8990 The argument and return value are floating point numbers of the same
8996 This function returns the same values as the libm ``nearbyint``
8997 functions would, and handles error conditions in the same way.
8999 '``llvm.round.*``' Intrinsic
9000 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9005 This is an overloaded intrinsic. You can use ``llvm.round`` on any
9006 floating point or vector of floating point type. Not all targets support
9011 declare float @llvm.round.f32(float %Val)
9012 declare double @llvm.round.f64(double %Val)
9013 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
9014 declare fp128 @llvm.round.f128(fp128 %Val)
9015 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
9020 The '``llvm.round.*``' intrinsics returns the operand rounded to the
9026 The argument and return value are floating point numbers of the same
9032 This function returns the same values as the libm ``round``
9033 functions would, and handles error conditions in the same way.
9035 Bit Manipulation Intrinsics
9036 ---------------------------
9038 LLVM provides intrinsics for a few important bit manipulation
9039 operations. These allow efficient code generation for some algorithms.
9041 '``llvm.bswap.*``' Intrinsics
9042 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9047 This is an overloaded intrinsic function. You can use bswap on any
9048 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
9052 declare i16 @llvm.bswap.i16(i16 <id>)
9053 declare i32 @llvm.bswap.i32(i32 <id>)
9054 declare i64 @llvm.bswap.i64(i64 <id>)
9059 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
9060 values with an even number of bytes (positive multiple of 16 bits).
9061 These are useful for performing operations on data that is not in the
9062 target's native byte order.
9067 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
9068 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
9069 intrinsic returns an i32 value that has the four bytes of the input i32
9070 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
9071 returned i32 will have its bytes in 3, 2, 1, 0 order. The
9072 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
9073 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
9076 '``llvm.ctpop.*``' Intrinsic
9077 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9082 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
9083 bit width, or on any vector with integer elements. Not all targets
9084 support all bit widths or vector types, however.
9088 declare i8 @llvm.ctpop.i8(i8 <src>)
9089 declare i16 @llvm.ctpop.i16(i16 <src>)
9090 declare i32 @llvm.ctpop.i32(i32 <src>)
9091 declare i64 @llvm.ctpop.i64(i64 <src>)
9092 declare i256 @llvm.ctpop.i256(i256 <src>)
9093 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
9098 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
9104 The only argument is the value to be counted. The argument may be of any
9105 integer type, or a vector with integer elements. The return type must
9106 match the argument type.
9111 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
9112 each element of a vector.
9114 '``llvm.ctlz.*``' Intrinsic
9115 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9120 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
9121 integer bit width, or any vector whose elements are integers. Not all
9122 targets support all bit widths or vector types, however.
9126 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
9127 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
9128 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
9129 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
9130 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
9131 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
9136 The '``llvm.ctlz``' family of intrinsic functions counts the number of
9137 leading zeros in a variable.
9142 The first argument is the value to be counted. This argument may be of
9143 any integer type, or a vector with integer element type. The return
9144 type must match the first argument type.
9146 The second argument must be a constant and is a flag to indicate whether
9147 the intrinsic should ensure that a zero as the first argument produces a
9148 defined result. Historically some architectures did not provide a
9149 defined result for zero values as efficiently, and many algorithms are
9150 now predicated on avoiding zero-value inputs.
9155 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
9156 zeros in a variable, or within each element of the vector. If
9157 ``src == 0`` then the result is the size in bits of the type of ``src``
9158 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9159 ``llvm.ctlz(i32 2) = 30``.
9161 '``llvm.cttz.*``' Intrinsic
9162 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9167 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
9168 integer bit width, or any vector of integer elements. Not all targets
9169 support all bit widths or vector types, however.
9173 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
9174 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
9175 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
9176 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
9177 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
9178 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
9183 The '``llvm.cttz``' family of intrinsic functions counts the number of
9189 The first argument is the value to be counted. This argument may be of
9190 any integer type, or a vector with integer element type. The return
9191 type must match the first argument type.
9193 The second argument must be a constant and is a flag to indicate whether
9194 the intrinsic should ensure that a zero as the first argument produces a
9195 defined result. Historically some architectures did not provide a
9196 defined result for zero values as efficiently, and many algorithms are
9197 now predicated on avoiding zero-value inputs.
9202 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
9203 zeros in a variable, or within each element of a vector. If ``src == 0``
9204 then the result is the size in bits of the type of ``src`` if
9205 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9206 ``llvm.cttz(2) = 1``.
9210 Arithmetic with Overflow Intrinsics
9211 -----------------------------------
9213 LLVM provides intrinsics for some arithmetic with overflow operations.
9215 '``llvm.sadd.with.overflow.*``' Intrinsics
9216 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9221 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
9222 on any integer bit width.
9226 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
9227 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9228 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
9233 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9234 a signed addition of the two arguments, and indicate whether an overflow
9235 occurred during the signed summation.
9240 The arguments (%a and %b) and the first element of the result structure
9241 may be of integer types of any bit width, but they must have the same
9242 bit width. The second element of the result structure must be of type
9243 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9249 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9250 a signed addition of the two variables. They return a structure --- the
9251 first element of which is the signed summation, and the second element
9252 of which is a bit specifying if the signed summation resulted in an
9258 .. code-block:: llvm
9260 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9261 %sum = extractvalue {i32, i1} %res, 0
9262 %obit = extractvalue {i32, i1} %res, 1
9263 br i1 %obit, label %overflow, label %normal
9265 '``llvm.uadd.with.overflow.*``' Intrinsics
9266 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9271 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
9272 on any integer bit width.
9276 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
9277 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
9278 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
9283 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9284 an unsigned addition of the two arguments, and indicate whether a carry
9285 occurred during the unsigned summation.
9290 The arguments (%a and %b) and the first element of the result structure
9291 may be of integer types of any bit width, but they must have the same
9292 bit width. The second element of the result structure must be of type
9293 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9299 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9300 an unsigned addition of the two arguments. They return a structure --- the
9301 first element of which is the sum, and the second element of which is a
9302 bit specifying if the unsigned summation resulted in a carry.
9307 .. code-block:: llvm
9309 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
9310 %sum = extractvalue {i32, i1} %res, 0
9311 %obit = extractvalue {i32, i1} %res, 1
9312 br i1 %obit, label %carry, label %normal
9314 '``llvm.ssub.with.overflow.*``' Intrinsics
9315 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9320 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
9321 on any integer bit width.
9325 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
9326 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
9327 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
9332 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
9333 a signed subtraction of the two arguments, and indicate whether an
9334 overflow occurred during the signed subtraction.
9339 The arguments (%a and %b) and the first element of the result structure
9340 may be of integer types of any bit width, but they must have the same
9341 bit width. The second element of the result structure must be of type
9342 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9348 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
9349 a signed subtraction of the two arguments. They return a structure --- the
9350 first element of which is the subtraction, and the second element of
9351 which is a bit specifying if the signed subtraction resulted in an
9357 .. code-block:: llvm
9359 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
9360 %sum = extractvalue {i32, i1} %res, 0
9361 %obit = extractvalue {i32, i1} %res, 1
9362 br i1 %obit, label %overflow, label %normal
9364 '``llvm.usub.with.overflow.*``' Intrinsics
9365 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9370 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
9371 on any integer bit width.
9375 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
9376 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
9377 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
9382 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
9383 an unsigned subtraction of the two arguments, and indicate whether an
9384 overflow occurred during the unsigned subtraction.
9389 The arguments (%a and %b) and the first element of the result structure
9390 may be of integer types of any bit width, but they must have the same
9391 bit width. The second element of the result structure must be of type
9392 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9398 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
9399 an unsigned subtraction of the two arguments. They return a structure ---
9400 the first element of which is the subtraction, and the second element of
9401 which is a bit specifying if the unsigned subtraction resulted in an
9407 .. code-block:: llvm
9409 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
9410 %sum = extractvalue {i32, i1} %res, 0
9411 %obit = extractvalue {i32, i1} %res, 1
9412 br i1 %obit, label %overflow, label %normal
9414 '``llvm.smul.with.overflow.*``' Intrinsics
9415 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9420 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
9421 on any integer bit width.
9425 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
9426 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
9427 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
9432 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
9433 a signed multiplication of the two arguments, and indicate whether an
9434 overflow occurred during the signed multiplication.
9439 The arguments (%a and %b) and the first element of the result structure
9440 may be of integer types of any bit width, but they must have the same
9441 bit width. The second element of the result structure must be of type
9442 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9448 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
9449 a signed multiplication of the two arguments. They return a structure ---
9450 the first element of which is the multiplication, and the second element
9451 of which is a bit specifying if the signed multiplication resulted in an
9457 .. code-block:: llvm
9459 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
9460 %sum = extractvalue {i32, i1} %res, 0
9461 %obit = extractvalue {i32, i1} %res, 1
9462 br i1 %obit, label %overflow, label %normal
9464 '``llvm.umul.with.overflow.*``' Intrinsics
9465 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9470 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
9471 on any integer bit width.
9475 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
9476 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9477 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
9482 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9483 a unsigned multiplication of the two arguments, and indicate whether an
9484 overflow occurred during the unsigned multiplication.
9489 The arguments (%a and %b) and the first element of the result structure
9490 may be of integer types of any bit width, but they must have the same
9491 bit width. The second element of the result structure must be of type
9492 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9498 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9499 an unsigned multiplication of the two arguments. They return a structure ---
9500 the first element of which is the multiplication, and the second
9501 element of which is a bit specifying if the unsigned multiplication
9502 resulted in an overflow.
9507 .. code-block:: llvm
9509 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9510 %sum = extractvalue {i32, i1} %res, 0
9511 %obit = extractvalue {i32, i1} %res, 1
9512 br i1 %obit, label %overflow, label %normal
9514 Specialised Arithmetic Intrinsics
9515 ---------------------------------
9517 '``llvm.fmuladd.*``' Intrinsic
9518 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9525 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
9526 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
9531 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
9532 expressions that can be fused if the code generator determines that (a) the
9533 target instruction set has support for a fused operation, and (b) that the
9534 fused operation is more efficient than the equivalent, separate pair of mul
9535 and add instructions.
9540 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
9541 multiplicands, a and b, and an addend c.
9550 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
9552 is equivalent to the expression a \* b + c, except that rounding will
9553 not be performed between the multiplication and addition steps if the
9554 code generator fuses the operations. Fusion is not guaranteed, even if
9555 the target platform supports it. If a fused multiply-add is required the
9556 corresponding llvm.fma.\* intrinsic function should be used
9557 instead. This never sets errno, just as '``llvm.fma.*``'.
9562 .. code-block:: llvm
9564 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
9566 Half Precision Floating Point Intrinsics
9567 ----------------------------------------
9569 For most target platforms, half precision floating point is a
9570 storage-only format. This means that it is a dense encoding (in memory)
9571 but does not support computation in the format.
9573 This means that code must first load the half-precision floating point
9574 value as an i16, then convert it to float with
9575 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
9576 then be performed on the float value (including extending to double
9577 etc). To store the value back to memory, it is first converted to float
9578 if needed, then converted to i16 with
9579 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
9582 .. _int_convert_to_fp16:
9584 '``llvm.convert.to.fp16``' Intrinsic
9585 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9592 declare i16 @llvm.convert.to.fp16.f32(float %a)
9593 declare i16 @llvm.convert.to.fp16.f64(double %a)
9598 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9599 conventional floating point type to half precision floating point format.
9604 The intrinsic function contains single argument - the value to be
9610 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9611 conventional floating point format to half precision floating point format. The
9612 return value is an ``i16`` which contains the converted number.
9617 .. code-block:: llvm
9619 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
9620 store i16 %res, i16* @x, align 2
9622 .. _int_convert_from_fp16:
9624 '``llvm.convert.from.fp16``' Intrinsic
9625 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9632 declare float @llvm.convert.from.fp16.f32(i16 %a)
9633 declare double @llvm.convert.from.fp16.f64(i16 %a)
9638 The '``llvm.convert.from.fp16``' intrinsic function performs a
9639 conversion from half precision floating point format to single precision
9640 floating point format.
9645 The intrinsic function contains single argument - the value to be
9651 The '``llvm.convert.from.fp16``' intrinsic function performs a
9652 conversion from half single precision floating point format to single
9653 precision floating point format. The input half-float value is
9654 represented by an ``i16`` value.
9659 .. code-block:: llvm
9661 %a = load i16, i16* @x, align 2
9662 %res = call float @llvm.convert.from.fp16(i16 %a)
9669 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
9670 prefix), are described in the `LLVM Source Level
9671 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
9674 Exception Handling Intrinsics
9675 -----------------------------
9677 The LLVM exception handling intrinsics (which all start with
9678 ``llvm.eh.`` prefix), are described in the `LLVM Exception
9679 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
9683 Trampoline Intrinsics
9684 ---------------------
9686 These intrinsics make it possible to excise one parameter, marked with
9687 the :ref:`nest <nest>` attribute, from a function. The result is a
9688 callable function pointer lacking the nest parameter - the caller does
9689 not need to provide a value for it. Instead, the value to use is stored
9690 in advance in a "trampoline", a block of memory usually allocated on the
9691 stack, which also contains code to splice the nest value into the
9692 argument list. This is used to implement the GCC nested function address
9695 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
9696 then the resulting function pointer has signature ``i32 (i32, i32)*``.
9697 It can be created as follows:
9699 .. code-block:: llvm
9701 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
9702 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
9703 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
9704 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
9705 %fp = bitcast i8* %p to i32 (i32, i32)*
9707 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
9708 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
9712 '``llvm.init.trampoline``' Intrinsic
9713 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9720 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
9725 This fills the memory pointed to by ``tramp`` with executable code,
9726 turning it into a trampoline.
9731 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
9732 pointers. The ``tramp`` argument must point to a sufficiently large and
9733 sufficiently aligned block of memory; this memory is written to by the
9734 intrinsic. Note that the size and the alignment are target-specific -
9735 LLVM currently provides no portable way of determining them, so a
9736 front-end that generates this intrinsic needs to have some
9737 target-specific knowledge. The ``func`` argument must hold a function
9738 bitcast to an ``i8*``.
9743 The block of memory pointed to by ``tramp`` is filled with target
9744 dependent code, turning it into a function. Then ``tramp`` needs to be
9745 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
9746 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
9747 function's signature is the same as that of ``func`` with any arguments
9748 marked with the ``nest`` attribute removed. At most one such ``nest``
9749 argument is allowed, and it must be of pointer type. Calling the new
9750 function is equivalent to calling ``func`` with the same argument list,
9751 but with ``nval`` used for the missing ``nest`` argument. If, after
9752 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
9753 modified, then the effect of any later call to the returned function
9754 pointer is undefined.
9758 '``llvm.adjust.trampoline``' Intrinsic
9759 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9766 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
9771 This performs any required machine-specific adjustment to the address of
9772 a trampoline (passed as ``tramp``).
9777 ``tramp`` must point to a block of memory which already has trampoline
9778 code filled in by a previous call to
9779 :ref:`llvm.init.trampoline <int_it>`.
9784 On some architectures the address of the code to be executed needs to be
9785 different than the address where the trampoline is actually stored. This
9786 intrinsic returns the executable address corresponding to ``tramp``
9787 after performing the required machine specific adjustments. The pointer
9788 returned can then be :ref:`bitcast and executed <int_trampoline>`.
9790 .. _int_mload_mstore:
9792 Masked Vector Load and Store Intrinsics
9793 ---------------------------------------
9795 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.
9799 '``llvm.masked.load.*``' Intrinsics
9800 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9804 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
9808 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
9809 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
9814 Reads a vector from memory according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes. The masked-off lanes in the result vector are taken from the corresponding lanes of the '``passthru``' operand.
9820 The first operand is the base pointer for the load. The second operand is the alignment of the source location. It must be a constant integer value. The third operand, mask, is a vector of boolean values with the same number of elements as the return type. The fourth is a pass-through value that is used to fill the masked-off lanes of the result. The return type, underlying type of the base pointer and the type of the '``passthru``' operand are the same vector types.
9826 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.
9827 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.
9832 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
9834 ;; The result of the two following instructions is identical aside from potential memory access exception
9835 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
9836 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
9840 '``llvm.masked.store.*``' Intrinsics
9841 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9845 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
9849 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
9850 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
9855 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.
9860 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.
9866 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.
9867 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.
9871 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
9873 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
9874 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
9875 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
9876 store <16 x float> %res, <16 x float>* %ptr, align 4
9879 Masked Vector Gather and Scatter Intrinsics
9880 -------------------------------------------
9882 LLVM provides intrinsics for vector gather and scatter operations. They are similar to :ref:`Masked Vector Load and Store <int_mload_mstore>`, except they are designed for arbitrary memory accesses, rather than sequential memory accesses. Gather and scatter also employ a mask operand, which holds one bit per vector element, switching the associated vector lane on or off. The memory addresses corresponding to the "off" lanes are not accessed. When all bits are off, no memory is accessed.
9886 '``llvm.masked.gather.*``' Intrinsics
9887 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9891 This is an overloaded intrinsic. The loaded data are multiple scalar values of any integer or floating point data type gathered together into one vector.
9895 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
9896 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
9901 Reads scalar values from arbitrary memory locations and gathers them into one vector. The memory locations are provided in the vector of pointers '``ptrs``'. The memory is accessed according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes. The masked-off lanes in the result vector are taken from the corresponding lanes of the '``passthru``' operand.
9907 The first operand is a vector of pointers which holds all memory addresses to read. The second operand is an alignment of the source addresses. It must be a constant integer value. The third operand, mask, is a vector of boolean values with the same number of elements as the return type. The fourth is a pass-through value that is used to fill the masked-off lanes of the result. The return type, underlying type of the vector of pointers and the type of the '``passthru``' operand are the same vector types.
9913 The '``llvm.masked.gather``' intrinsic is designed for conditional reading of multiple scalar values from arbitrary memory locations in a single IR operation. It is useful for targets that support vector masked gathers and allows vectorizing basic blocks with data and control divergence. Other targets may support this intrinsic differently, for example by lowering it into a sequence of scalar load operations.
9914 The semantics of this operation are equivalent to a sequence of conditional scalar loads with subsequent gathering all loaded values into a single vector. The mask restricts memory access to certain lanes and facilitates vectorization of predicated basic blocks.
9919 %res = call <4 x double> @llvm.masked.gather.v4f64 (<4 x double*> %ptrs, i32 8, <4 x i1>%mask, <4 x double> <true, true, true, true>)
9921 ;; The gather with all-true mask is equivalent to the following instruction sequence
9922 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
9923 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
9924 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
9925 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
9927 %val0 = load double, double* %ptr0, align 8
9928 %val1 = load double, double* %ptr1, align 8
9929 %val2 = load double, double* %ptr2, align 8
9930 %val3 = load double, double* %ptr3, align 8
9932 %vec0 = insertelement <4 x double>undef, %val0, 0
9933 %vec01 = insertelement <4 x double>%vec0, %val1, 1
9934 %vec012 = insertelement <4 x double>%vec01, %val2, 2
9935 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
9939 '``llvm.masked.scatter.*``' Intrinsics
9940 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9944 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type. Each vector element is stored in an arbitrary memory addresses. Scatter with overlapping addresses is guaranteed to be ordered from least-significant to most-significant element.
9948 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
9949 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
9954 Writes each element from the value vector to the corresponding memory address. The memory addresses are represented as a vector of pointers. Writing is done according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes.
9959 The first operand is a vector value to be written to memory. The second operand is a vector of pointers, pointing to where the value elements should be stored. It has the same underlying type as the value operand. The third operand is an alignment of the destination addresses. The fourth operand, mask, is a vector of boolean values. The types of the mask and the value operand must have the same number of vector elements.
9965 The '``llvm.masked.scatter``' intrinsics is designed for writing selected vector elements to arbitrary memory addresses in a single IR operation. The operation may be conditional, when not all bits in the mask are switched on. It is useful for targets that support vector masked scatter and allows vectorizing basic blocks with data and control divergency. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar store operations.
9969 ;; This instruction unconditionaly stores data vector in multiple addresses
9970 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
9972 ;; It is equivalent to a list of scalar stores
9973 %val0 = extractelement <8 x i32> %value, i32 0
9974 %val1 = extractelement <8 x i32> %value, i32 1
9976 %val7 = extractelement <8 x i32> %value, i32 7
9977 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
9978 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
9980 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
9981 ;; Note: the order of the following stores is important when they overlap:
9982 store i32 %val0, i32* %ptr0, align 4
9983 store i32 %val1, i32* %ptr1, align 4
9985 store i32 %val7, i32* %ptr7, align 4
9991 This class of intrinsics provides information about the lifetime of
9992 memory objects and ranges where variables are immutable.
9996 '``llvm.lifetime.start``' Intrinsic
9997 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10004 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
10009 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
10015 The first argument is a constant integer representing the size of the
10016 object, or -1 if it is variable sized. The second argument is a pointer
10022 This intrinsic indicates that before this point in the code, the value
10023 of the memory pointed to by ``ptr`` is dead. This means that it is known
10024 to never be used and has an undefined value. A load from the pointer
10025 that precedes this intrinsic can be replaced with ``'undef'``.
10029 '``llvm.lifetime.end``' Intrinsic
10030 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10037 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
10042 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
10048 The first argument is a constant integer representing the size of the
10049 object, or -1 if it is variable sized. The second argument is a pointer
10055 This intrinsic indicates that after this point in the code, the value of
10056 the memory pointed to by ``ptr`` is dead. This means that it is known to
10057 never be used and has an undefined value. Any stores into the memory
10058 object following this intrinsic may be removed as dead.
10060 '``llvm.invariant.start``' Intrinsic
10061 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10068 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
10073 The '``llvm.invariant.start``' intrinsic specifies that the contents of
10074 a memory object will not change.
10079 The first argument is a constant integer representing the size of the
10080 object, or -1 if it is variable sized. The second argument is a pointer
10086 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
10087 the return value, the referenced memory location is constant and
10090 '``llvm.invariant.end``' Intrinsic
10091 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10098 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
10103 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
10104 memory object are mutable.
10109 The first argument is the matching ``llvm.invariant.start`` intrinsic.
10110 The second argument is a constant integer representing the size of the
10111 object, or -1 if it is variable sized and the third argument is a
10112 pointer to the object.
10117 This intrinsic indicates that the memory is mutable again.
10122 This class of intrinsics is designed to be generic and has no specific
10125 '``llvm.var.annotation``' Intrinsic
10126 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10133 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
10138 The '``llvm.var.annotation``' intrinsic.
10143 The first argument is a pointer to a value, the second is a pointer to a
10144 global string, the third is a pointer to a global string which is the
10145 source file name, and the last argument is the line number.
10150 This intrinsic allows annotation of local variables with arbitrary
10151 strings. This can be useful for special purpose optimizations that want
10152 to look for these annotations. These have no other defined use; they are
10153 ignored by code generation and optimization.
10155 '``llvm.ptr.annotation.*``' Intrinsic
10156 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10161 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
10162 pointer to an integer of any width. *NOTE* you must specify an address space for
10163 the pointer. The identifier for the default address space is the integer
10168 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
10169 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
10170 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
10171 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
10172 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
10177 The '``llvm.ptr.annotation``' intrinsic.
10182 The first argument is a pointer to an integer value of arbitrary bitwidth
10183 (result of some expression), the second is a pointer to a global string, the
10184 third is a pointer to a global string which is the source file name, and the
10185 last argument is the line number. It returns the value of the first argument.
10190 This intrinsic allows annotation of a pointer to an integer with arbitrary
10191 strings. This can be useful for special purpose optimizations that want to look
10192 for these annotations. These have no other defined use; they are ignored by code
10193 generation and optimization.
10195 '``llvm.annotation.*``' Intrinsic
10196 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10201 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
10202 any integer bit width.
10206 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
10207 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
10208 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
10209 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
10210 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
10215 The '``llvm.annotation``' intrinsic.
10220 The first argument is an integer value (result of some expression), the
10221 second is a pointer to a global string, the third is a pointer to a
10222 global string which is the source file name, and the last argument is
10223 the line number. It returns the value of the first argument.
10228 This intrinsic allows annotations to be put on arbitrary expressions
10229 with arbitrary strings. This can be useful for special purpose
10230 optimizations that want to look for these annotations. These have no
10231 other defined use; they are ignored by code generation and optimization.
10233 '``llvm.trap``' Intrinsic
10234 ^^^^^^^^^^^^^^^^^^^^^^^^^
10241 declare void @llvm.trap() noreturn nounwind
10246 The '``llvm.trap``' intrinsic.
10256 This intrinsic is lowered to the target dependent trap instruction. If
10257 the target does not have a trap instruction, this intrinsic will be
10258 lowered to a call of the ``abort()`` function.
10260 '``llvm.debugtrap``' Intrinsic
10261 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10268 declare void @llvm.debugtrap() nounwind
10273 The '``llvm.debugtrap``' intrinsic.
10283 This intrinsic is lowered to code which is intended to cause an
10284 execution trap with the intention of requesting the attention of a
10287 '``llvm.stackprotector``' Intrinsic
10288 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10295 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
10300 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
10301 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
10302 is placed on the stack before local variables.
10307 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
10308 The first argument is the value loaded from the stack guard
10309 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
10310 enough space to hold the value of the guard.
10315 This intrinsic causes the prologue/epilogue inserter to force the position of
10316 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
10317 to ensure that if a local variable on the stack is overwritten, it will destroy
10318 the value of the guard. When the function exits, the guard on the stack is
10319 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
10320 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
10321 calling the ``__stack_chk_fail()`` function.
10323 '``llvm.stackprotectorcheck``' Intrinsic
10324 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10331 declare void @llvm.stackprotectorcheck(i8** <guard>)
10336 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
10337 created stack protector and if they are not equal calls the
10338 ``__stack_chk_fail()`` function.
10343 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
10344 the variable ``@__stack_chk_guard``.
10349 This intrinsic is provided to perform the stack protector check by comparing
10350 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
10351 values do not match call the ``__stack_chk_fail()`` function.
10353 The reason to provide this as an IR level intrinsic instead of implementing it
10354 via other IR operations is that in order to perform this operation at the IR
10355 level without an intrinsic, one would need to create additional basic blocks to
10356 handle the success/failure cases. This makes it difficult to stop the stack
10357 protector check from disrupting sibling tail calls in Codegen. With this
10358 intrinsic, we are able to generate the stack protector basic blocks late in
10359 codegen after the tail call decision has occurred.
10361 '``llvm.objectsize``' Intrinsic
10362 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10369 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
10370 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
10375 The ``llvm.objectsize`` intrinsic is designed to provide information to
10376 the optimizers to determine at compile time whether a) an operation
10377 (like memcpy) will overflow a buffer that corresponds to an object, or
10378 b) that a runtime check for overflow isn't necessary. An object in this
10379 context means an allocation of a specific class, structure, array, or
10385 The ``llvm.objectsize`` intrinsic takes two arguments. The first
10386 argument is a pointer to or into the ``object``. The second argument is
10387 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
10388 or -1 (if false) when the object size is unknown. The second argument
10389 only accepts constants.
10394 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
10395 the size of the object concerned. If the size cannot be determined at
10396 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
10397 on the ``min`` argument).
10399 '``llvm.expect``' Intrinsic
10400 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10405 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
10410 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
10411 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
10412 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
10417 The ``llvm.expect`` intrinsic provides information about expected (the
10418 most probable) value of ``val``, which can be used by optimizers.
10423 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
10424 a value. The second argument is an expected value, this needs to be a
10425 constant value, variables are not allowed.
10430 This intrinsic is lowered to the ``val``.
10434 '``llvm.assume``' Intrinsic
10435 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10442 declare void @llvm.assume(i1 %cond)
10447 The ``llvm.assume`` allows the optimizer to assume that the provided
10448 condition is true. This information can then be used in simplifying other parts
10454 The condition which the optimizer may assume is always true.
10459 The intrinsic allows the optimizer to assume that the provided condition is
10460 always true whenever the control flow reaches the intrinsic call. No code is
10461 generated for this intrinsic, and instructions that contribute only to the
10462 provided condition are not used for code generation. If the condition is
10463 violated during execution, the behavior is undefined.
10465 Note that the optimizer might limit the transformations performed on values
10466 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
10467 only used to form the intrinsic's input argument. This might prove undesirable
10468 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
10469 sufficient overall improvement in code quality. For this reason,
10470 ``llvm.assume`` should not be used to document basic mathematical invariants
10471 that the optimizer can otherwise deduce or facts that are of little use to the
10476 '``llvm.bitset.test``' Intrinsic
10477 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10484 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
10490 The first argument is a pointer to be tested. The second argument is a
10491 metadata string containing the name of a :doc:`bitset <BitSets>`.
10496 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
10497 member of the given bitset.
10499 '``llvm.donothing``' Intrinsic
10500 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10507 declare void @llvm.donothing() nounwind readnone
10512 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
10513 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
10514 with an invoke instruction.
10524 This intrinsic does nothing, and it's removed by optimizers and ignored
10527 Stack Map Intrinsics
10528 --------------------
10530 LLVM provides experimental intrinsics to support runtime patching
10531 mechanisms commonly desired in dynamic language JITs. These intrinsics
10532 are described in :doc:`StackMaps`.