1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][-a-zA-Z$._][-a-zA-Z$._0-9]*``'. Identifiers that require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves. The ``"\01"`` prefix
83 can be used on global variables to suppress mangling.
84 #. Unnamed values are represented as an unsigned numeric value with
85 their prefix. For example, ``%12``, ``@2``, ``%44``.
86 #. Constants, which are described in the section Constants_ below.
88 LLVM requires that values start with a prefix for two reasons: Compilers
89 don't need to worry about name clashes with reserved words, and the set
90 of reserved words may be expanded in the future without penalty.
91 Additionally, unnamed identifiers allow a compiler to quickly come up
92 with a temporary variable without having to avoid symbol table
95 Reserved words in LLVM are very similar to reserved words in other
96 languages. There are keywords for different opcodes ('``add``',
97 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
98 '``i32``', etc...), and others. These reserved words cannot conflict
99 with variable names, because none of them start with a prefix character
100 (``'%'`` or ``'@'``).
102 Here is an example of LLVM code to multiply the integer variable
109 %result = mul i32 %X, 8
111 After strength reduction:
115 %result = shl i32 %X, 3
121 %0 = add i32 %X, %X ; yields i32:%0
122 %1 = add i32 %0, %0 ; yields i32:%1
123 %result = add i32 %1, %1
125 This last way of multiplying ``%X`` by 8 illustrates several important
126 lexical features of LLVM:
128 #. Comments are delimited with a '``;``' and go until the end of line.
129 #. Unnamed temporaries are created when the result of a computation is
130 not assigned to a named value.
131 #. Unnamed temporaries are numbered sequentially (using a per-function
132 incrementing counter, starting with 0). Note that basic blocks and unnamed
133 function parameters are included in this numbering. For example, if the
134 entry basic block is not given a label name and all function parameters are
135 named, then it will get number 0.
137 It also shows a convention that we follow in this document. When
138 demonstrating instructions, we will follow an instruction with a comment
139 that defines the type and name of value produced.
147 LLVM programs are composed of ``Module``'s, each of which is a
148 translation unit of the input programs. Each module consists of
149 functions, global variables, and symbol table entries. Modules may be
150 combined together with the LLVM linker, which merges function (and
151 global variable) definitions, resolves forward declarations, and merges
152 symbol table entries. Here is an example of the "hello world" module:
156 ; Declare the string constant as a global constant.
157 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
159 ; External declaration of the puts function
160 declare i32 @puts(i8* nocapture) nounwind
162 ; Definition of main function
163 define i32 @main() { ; i32()*
164 ; Convert [13 x i8]* to i8 *...
165 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
167 ; Call puts function to write out the string to stdout.
168 call i32 @puts(i8* %cast210)
173 !0 = !{i32 42, null, !"string"}
176 This example is made up of a :ref:`global variable <globalvars>` named
177 "``.str``", an external declaration of the "``puts``" function, a
178 :ref:`function definition <functionstructure>` for "``main``" and
179 :ref:`named metadata <namedmetadatastructure>` "``foo``".
181 In general, a module is made up of a list of global values (where both
182 functions and global variables are global values). Global values are
183 represented by a pointer to a memory location (in this case, a pointer
184 to an array of char, and a pointer to a function), and have one of the
185 following :ref:`linkage types <linkage>`.
192 All Global Variables and Functions have one of the following types of
196 Global values with "``private``" linkage are only directly
197 accessible by objects in the current module. In particular, linking
198 code into a module with an private global value may cause the
199 private to be renamed as necessary to avoid collisions. Because the
200 symbol is private to the module, all references can be updated. This
201 doesn't show up in any symbol table in the object file.
203 Similar to private, but the value shows as a local symbol
204 (``STB_LOCAL`` in the case of ELF) in the object file. This
205 corresponds to the notion of the '``static``' keyword in C.
206 ``available_externally``
207 Globals with "``available_externally``" linkage are never emitted
208 into the object file corresponding to the LLVM module. They exist to
209 allow inlining and other optimizations to take place given knowledge
210 of the definition of the global, which is known to be somewhere
211 outside the module. Globals with ``available_externally`` linkage
212 are allowed to be discarded at will, and are otherwise the same as
213 ``linkonce_odr``. This linkage type is only allowed on definitions,
216 Globals with "``linkonce``" linkage are merged with other globals of
217 the same name when linkage occurs. This can be used to implement
218 some forms of inline functions, templates, or other code which must
219 be generated in each translation unit that uses it, but where the
220 body may be overridden with a more definitive definition later.
221 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
222 that ``linkonce`` linkage does not actually allow the optimizer to
223 inline the body of this function into callers because it doesn't
224 know if this definition of the function is the definitive definition
225 within the program or whether it will be overridden by a stronger
226 definition. To enable inlining and other optimizations, use
227 "``linkonce_odr``" linkage.
229 "``weak``" linkage has the same merging semantics as ``linkonce``
230 linkage, except that unreferenced globals with ``weak`` linkage may
231 not be discarded. This is used for globals that are declared "weak"
234 "``common``" linkage is most similar to "``weak``" linkage, but they
235 are used for tentative definitions in C, such as "``int X;``" at
236 global scope. Symbols with "``common``" linkage are merged in the
237 same way as ``weak symbols``, and they may not be deleted if
238 unreferenced. ``common`` symbols may not have an explicit section,
239 must have a zero initializer, and may not be marked
240 ':ref:`constant <globalvars>`'. Functions and aliases may not have
243 .. _linkage_appending:
246 "``appending``" linkage may only be applied to global variables of
247 pointer to array type. When two global variables with appending
248 linkage are linked together, the two global arrays are appended
249 together. This is the LLVM, typesafe, equivalent of having the
250 system linker append together "sections" with identical names when
253 The semantics of this linkage follow the ELF object file model: the
254 symbol is weak until linked, if not linked, the symbol becomes null
255 instead of being an undefined reference.
256 ``linkonce_odr``, ``weak_odr``
257 Some languages allow differing globals to be merged, such as two
258 functions with different semantics. Other languages, such as
259 ``C++``, ensure that only equivalent globals are ever merged (the
260 "one definition rule" --- "ODR"). Such languages can use the
261 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
262 global will only be merged with equivalent globals. These linkage
263 types are otherwise the same as their non-``odr`` versions.
265 If none of the above identifiers are used, the global is externally
266 visible, meaning that it participates in linkage and can be used to
267 resolve external symbol references.
269 It is illegal for a function *declaration* to have any linkage type
270 other than ``external`` or ``extern_weak``.
277 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
278 :ref:`invokes <i_invoke>` can all have an optional calling convention
279 specified for the call. The calling convention of any pair of dynamic
280 caller/callee must match, or the behavior of the program is undefined.
281 The following calling conventions are supported by LLVM, and more may be
284 "``ccc``" - The C calling convention
285 This calling convention (the default if no other calling convention
286 is specified) matches the target C calling conventions. This calling
287 convention supports varargs function calls and tolerates some
288 mismatch in the declared prototype and implemented declaration of
289 the function (as does normal C).
290 "``fastcc``" - The fast calling convention
291 This calling convention attempts to make calls as fast as possible
292 (e.g. by passing things in registers). This calling convention
293 allows the target to use whatever tricks it wants to produce fast
294 code for the target, without having to conform to an externally
295 specified ABI (Application Binary Interface). `Tail calls can only
296 be optimized when this, the GHC or the HiPE convention is
297 used. <CodeGenerator.html#id80>`_ This calling convention does not
298 support varargs and requires the prototype of all callees to exactly
299 match the prototype of the function definition.
300 "``coldcc``" - The cold calling convention
301 This calling convention attempts to make code in the caller as
302 efficient as possible under the assumption that the call is not
303 commonly executed. As such, these calls often preserve all registers
304 so that the call does not break any live ranges in the caller side.
305 This calling convention does not support varargs and requires the
306 prototype of all callees to exactly match the prototype of the
307 function definition. Furthermore the inliner doesn't consider such function
309 "``cc 10``" - GHC convention
310 This calling convention has been implemented specifically for use by
311 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
312 It passes everything in registers, going to extremes to achieve this
313 by disabling callee save registers. This calling convention should
314 not be used lightly but only for specific situations such as an
315 alternative to the *register pinning* performance technique often
316 used when implementing functional programming languages. At the
317 moment only X86 supports this convention and it has the following
320 - On *X86-32* only supports up to 4 bit type parameters. No
321 floating point types are supported.
322 - On *X86-64* only supports up to 10 bit type parameters and 6
323 floating point parameters.
325 This calling convention supports `tail call
326 optimization <CodeGenerator.html#id80>`_ but requires both the
327 caller and callee are using it.
328 "``cc 11``" - The HiPE calling convention
329 This calling convention has been implemented specifically for use by
330 the `High-Performance Erlang
331 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
332 native code compiler of the `Ericsson's Open Source Erlang/OTP
333 system <http://www.erlang.org/download.shtml>`_. It uses more
334 registers for argument passing than the ordinary C calling
335 convention and defines no callee-saved registers. The calling
336 convention properly supports `tail call
337 optimization <CodeGenerator.html#id80>`_ but requires that both the
338 caller and the callee use it. It uses a *register pinning*
339 mechanism, similar to GHC's convention, for keeping frequently
340 accessed runtime components pinned to specific hardware registers.
341 At the moment only X86 supports this convention (both 32 and 64
343 "``webkit_jscc``" - WebKit's JavaScript calling convention
344 This calling convention has been implemented for `WebKit FTL JIT
345 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
346 stack right to left (as cdecl does), and returns a value in the
347 platform's customary return register.
348 "``anyregcc``" - Dynamic calling convention for code patching
349 This is a special convention that supports patching an arbitrary code
350 sequence in place of a call site. This convention forces the call
351 arguments into registers but allows them to be dynamcially
352 allocated. This can currently only be used with calls to
353 llvm.experimental.patchpoint because only this intrinsic records
354 the location of its arguments in a side table. See :doc:`StackMaps`.
355 "``preserve_mostcc``" - The `PreserveMost` calling convention
356 This calling convention attempts to make the code in the caller as little
357 intrusive as possible. This calling convention behaves identical to the `C`
358 calling convention on how arguments and return values are passed, but it
359 uses a different set of caller/callee-saved registers. This alleviates the
360 burden of saving and recovering a large register set before and after the
361 call in the caller. If the arguments are passed in callee-saved registers,
362 then they will be preserved by the callee across the call. This doesn't
363 apply for values returned in callee-saved registers.
365 - On X86-64 the callee preserves all general purpose registers, except for
366 R11. R11 can be used as a scratch register. Floating-point registers
367 (XMMs/YMMs) are not preserved and need to be saved by the caller.
369 The idea behind this convention is to support calls to runtime functions
370 that have a hot path and a cold path. The hot path is usually a small piece
371 of code that doesn't many registers. The cold path might need to call out to
372 another function and therefore only needs to preserve the caller-saved
373 registers, which haven't already been saved by the caller. The
374 `PreserveMost` calling convention is very similar to the `cold` calling
375 convention in terms of caller/callee-saved registers, but they are used for
376 different types of function calls. `coldcc` is for function calls that are
377 rarely executed, whereas `preserve_mostcc` function calls are intended to be
378 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
379 doesn't prevent the inliner from inlining the function call.
381 This calling convention will be used by a future version of the ObjectiveC
382 runtime and should therefore still be considered experimental at this time.
383 Although this convention was created to optimize certain runtime calls to
384 the ObjectiveC runtime, it is not limited to this runtime and might be used
385 by other runtimes in the future too. The current implementation only
386 supports X86-64, but the intention is to support more architectures in the
388 "``preserve_allcc``" - The `PreserveAll` calling convention
389 This calling convention attempts to make the code in the caller even less
390 intrusive than the `PreserveMost` calling convention. This calling
391 convention also behaves identical to the `C` calling convention on how
392 arguments and return values are passed, but it uses a different set of
393 caller/callee-saved registers. This removes the burden of saving and
394 recovering a large register set before and after the call in the caller. If
395 the arguments are passed in callee-saved registers, then they will be
396 preserved by the callee across the call. This doesn't apply for values
397 returned in callee-saved registers.
399 - On X86-64 the callee preserves all general purpose registers, except for
400 R11. R11 can be used as a scratch register. Furthermore it also preserves
401 all floating-point registers (XMMs/YMMs).
403 The idea behind this convention is to support calls to runtime functions
404 that don't need to call out to any other functions.
406 This calling convention, like the `PreserveMost` calling convention, will be
407 used by a future version of the ObjectiveC runtime and should be considered
408 experimental at this time.
409 "``cc <n>``" - Numbered convention
410 Any calling convention may be specified by number, allowing
411 target-specific calling conventions to be used. Target specific
412 calling conventions start at 64.
414 More calling conventions can be added/defined on an as-needed basis, to
415 support Pascal conventions or any other well-known target-independent
418 .. _visibilitystyles:
423 All Global Variables and Functions have one of the following visibility
426 "``default``" - Default style
427 On targets that use the ELF object file format, default visibility
428 means that the declaration is visible to other modules and, in
429 shared libraries, means that the declared entity may be overridden.
430 On Darwin, default visibility means that the declaration is visible
431 to other modules. Default visibility corresponds to "external
432 linkage" in the language.
433 "``hidden``" - Hidden style
434 Two declarations of an object with hidden visibility refer to the
435 same object if they are in the same shared object. Usually, hidden
436 visibility indicates that the symbol will not be placed into the
437 dynamic symbol table, so no other module (executable or shared
438 library) can reference it directly.
439 "``protected``" - Protected style
440 On ELF, protected visibility indicates that the symbol will be
441 placed in the dynamic symbol table, but that references within the
442 defining module will bind to the local symbol. That is, the symbol
443 cannot be overridden by another module.
445 A symbol with ``internal`` or ``private`` linkage must have ``default``
453 All Global Variables, Functions and Aliases can have one of the following
457 "``dllimport``" causes the compiler to reference a function or variable via
458 a global pointer to a pointer that is set up by the DLL exporting the
459 symbol. On Microsoft Windows targets, the pointer name is formed by
460 combining ``__imp_`` and the function or variable name.
462 "``dllexport``" causes the compiler to provide a global pointer to a pointer
463 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
464 Microsoft Windows targets, the pointer name is formed by combining
465 ``__imp_`` and the function or variable name. Since this storage class
466 exists for defining a dll interface, the compiler, assembler and linker know
467 it is externally referenced and must refrain from deleting the symbol.
471 Thread Local Storage Models
472 ---------------------------
474 A variable may be defined as ``thread_local``, which means that it will
475 not be shared by threads (each thread will have a separated copy of the
476 variable). Not all targets support thread-local variables. Optionally, a
477 TLS model may be specified:
480 For variables that are only used within the current shared library.
482 For variables in modules that will not be loaded dynamically.
484 For variables defined in the executable and only used within it.
486 If no explicit model is given, the "general dynamic" model is used.
488 The models correspond to the ELF TLS models; see `ELF Handling For
489 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
490 more information on under which circumstances the different models may
491 be used. The target may choose a different TLS model if the specified
492 model is not supported, or if a better choice of model can be made.
494 A model can also be specified in a alias, but then it only governs how
495 the alias is accessed. It will not have any effect in the aliasee.
502 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
503 types <t_struct>`. Literal types are uniqued structurally, but identified types
504 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
505 to forward declare a type that is not yet available.
507 An example of a identified structure specification is:
511 %mytype = type { %mytype*, i32 }
513 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
514 literal types are uniqued in recent versions of LLVM.
521 Global variables define regions of memory allocated at compilation time
524 Global variables definitions must be initialized.
526 Global variables in other translation units can also be declared, in which
527 case they don't have an initializer.
529 Either global variable definitions or declarations may have an explicit section
530 to be placed in and may have an optional explicit alignment specified.
532 A variable may be defined as a global ``constant``, which indicates that
533 the contents of the variable will **never** be modified (enabling better
534 optimization, allowing the global data to be placed in the read-only
535 section of an executable, etc). Note that variables that need runtime
536 initialization cannot be marked ``constant`` as there is a store to the
539 LLVM explicitly allows *declarations* of global variables to be marked
540 constant, even if the final definition of the global is not. This
541 capability can be used to enable slightly better optimization of the
542 program, but requires the language definition to guarantee that
543 optimizations based on the 'constantness' are valid for the translation
544 units that do not include the definition.
546 As SSA values, global variables define pointer values that are in scope
547 (i.e. they dominate) all basic blocks in the program. Global variables
548 always define a pointer to their "content" type because they describe a
549 region of memory, and all memory objects in LLVM are accessed through
552 Global variables can be marked with ``unnamed_addr`` which indicates
553 that the address is not significant, only the content. Constants marked
554 like this can be merged with other constants if they have the same
555 initializer. Note that a constant with significant address *can* be
556 merged with a ``unnamed_addr`` constant, the result being a constant
557 whose address is significant.
559 A global variable may be declared to reside in a target-specific
560 numbered address space. For targets that support them, address spaces
561 may affect how optimizations are performed and/or what target
562 instructions are used to access the variable. The default address space
563 is zero. The address space qualifier must precede any other attributes.
565 LLVM allows an explicit section to be specified for globals. If the
566 target supports it, it will emit globals to the section specified.
567 Additionally, the global can placed in a comdat if the target has the necessary
570 By default, global initializers are optimized by assuming that global
571 variables defined within the module are not modified from their
572 initial values before the start of the global initializer. This is
573 true even for variables potentially accessible from outside the
574 module, including those with external linkage or appearing in
575 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
576 by marking the variable with ``externally_initialized``.
578 An explicit alignment may be specified for a global, which must be a
579 power of 2. If not present, or if the alignment is set to zero, the
580 alignment of the global is set by the target to whatever it feels
581 convenient. If an explicit alignment is specified, the global is forced
582 to have exactly that alignment. Targets and optimizers are not allowed
583 to over-align the global if the global has an assigned section. In this
584 case, the extra alignment could be observable: for example, code could
585 assume that the globals are densely packed in their section and try to
586 iterate over them as an array, alignment padding would break this
587 iteration. The maximum alignment is ``1 << 29``.
589 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
591 Variables and 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 placed in a COMDAT.
671 An explicit alignment may be specified for a function. If not present,
672 or if the alignment is set to zero, the alignment of the function is set
673 by the target to whatever it feels convenient. If an explicit alignment
674 is specified, the function is forced to have at least that much
675 alignment. All alignments must be a power of 2.
677 If the ``unnamed_addr`` attribute is given, the address is know to not
678 be significant and two identical functions can be merged.
682 define [linkage] [visibility] [DLLStorageClass]
684 <ResultType> @<FunctionName> ([argument list])
685 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
686 [align N] [gc] [prefix Constant] [prologue Constant] { ... }
688 The argument list is a comma seperated sequence of arguments where each
689 argument is of the following form
693 <type> [parameter Attrs] [name]
701 Aliases, unlike function or variables, don't create any new data. They
702 are just a new symbol and metadata for an existing position.
704 Aliases have a name and an aliasee that is either a global value or a
707 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
708 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
709 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
713 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy> @<Aliasee>
715 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
716 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
717 might not correctly handle dropping a weak symbol that is aliased.
719 Alias that are not ``unnamed_addr`` are guaranteed to have the same address as
720 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
723 Since aliases are only a second name, some restrictions apply, of which
724 some can only be checked when producing an object file:
726 * The expression defining the aliasee must be computable at assembly
727 time. Since it is just a name, no relocations can be used.
729 * No alias in the expression can be weak as the possibility of the
730 intermediate alias being overridden cannot be represented in an
733 * No global value in the expression can be a declaration, since that
734 would require a relocation, which is not possible.
741 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
743 Comdats have a name which represents the COMDAT key. All global objects that
744 specify this key will only end up in the final object file if the linker chooses
745 that key over some other key. Aliases are placed in the same COMDAT that their
746 aliasee computes to, if any.
748 Comdats have a selection kind to provide input on how the linker should
749 choose between keys in two different object files.
753 $<Name> = comdat SelectionKind
755 The selection kind must be one of the following:
758 The linker may choose any COMDAT key, the choice is arbitrary.
760 The linker may choose any COMDAT key but the sections must contain the
763 The linker will choose the section containing the largest COMDAT key.
765 The linker requires that only section with this COMDAT key exist.
767 The linker may choose any COMDAT key but the sections must contain the
770 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
771 ``any`` as a selection kind.
773 Here is an example of a COMDAT group where a function will only be selected if
774 the COMDAT key's section is the largest:
778 $foo = comdat largest
779 @foo = global i32 2, comdat($foo)
781 define void @bar() comdat($foo) {
785 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
791 @foo = global i32 2, comdat
794 In a COFF object file, this will create a COMDAT section with selection kind
795 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
796 and another COMDAT section with selection kind
797 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
798 section and contains the contents of the ``@bar`` symbol.
800 There are some restrictions on the properties of the global object.
801 It, or an alias to it, must have the same name as the COMDAT group when
803 The contents and size of this object may be used during link-time to determine
804 which COMDAT groups get selected depending on the selection kind.
805 Because the name of the object must match the name of the COMDAT group, the
806 linkage of the global object must not be local; local symbols can get renamed
807 if a collision occurs in the symbol table.
809 The combined use of COMDATS and section attributes may yield surprising results.
816 @g1 = global i32 42, section "sec", comdat($foo)
817 @g2 = global i32 42, section "sec", comdat($bar)
819 From the object file perspective, this requires the creation of two sections
820 with the same name. This is necessary because both globals belong to different
821 COMDAT groups and COMDATs, at the object file level, are represented by
824 Note that certain IR constructs like global variables and functions may create
825 COMDATs in the object file in addition to any which are specified using COMDAT
826 IR. This arises, for example, when a global variable has linkonce_odr linkage.
828 .. _namedmetadatastructure:
833 Named metadata is a collection of metadata. :ref:`Metadata
834 nodes <metadata>` (but not metadata strings) are the only valid
835 operands for a named metadata.
839 ; Some unnamed metadata nodes, which are referenced by the named metadata.
844 !name = !{!0, !1, !2}
851 The return type and each parameter of a function type may have a set of
852 *parameter attributes* associated with them. Parameter attributes are
853 used to communicate additional information about the result or
854 parameters of a function. Parameter attributes are considered to be part
855 of the function, not of the function type, so functions with different
856 parameter attributes can have the same function type.
858 Parameter attributes are simple keywords that follow the type specified.
859 If multiple parameter attributes are needed, they are space separated.
864 declare i32 @printf(i8* noalias nocapture, ...)
865 declare i32 @atoi(i8 zeroext)
866 declare signext i8 @returns_signed_char()
868 Note that any attributes for the function result (``nounwind``,
869 ``readonly``) come immediately after the argument list.
871 Currently, only the following parameter attributes are defined:
874 This indicates to the code generator that the parameter or return
875 value should be zero-extended to the extent required by the target's
876 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
877 the caller (for a parameter) or the callee (for a return value).
879 This indicates to the code generator that the parameter or return
880 value should be sign-extended to the extent required by the target's
881 ABI (which is usually 32-bits) by the caller (for a parameter) or
882 the callee (for a return value).
884 This indicates that this parameter or return value should be treated
885 in a special target-dependent fashion during while emitting code for
886 a function call or return (usually, by putting it in a register as
887 opposed to memory, though some targets use it to distinguish between
888 two different kinds of registers). Use of this attribute is
891 This indicates that the pointer parameter should really be passed by
892 value to the function. The attribute implies that a hidden copy of
893 the pointee is made between the caller and the callee, so the callee
894 is unable to modify the value in the caller. This attribute is only
895 valid on LLVM pointer arguments. It is generally used to pass
896 structs and arrays by value, but is also valid on pointers to
897 scalars. The copy is considered to belong to the caller not the
898 callee (for example, ``readonly`` functions should not write to
899 ``byval`` parameters). This is not a valid attribute for return
902 The byval attribute also supports specifying an alignment with the
903 align attribute. It indicates the alignment of the stack slot to
904 form and the known alignment of the pointer specified to the call
905 site. If the alignment is not specified, then the code generator
906 makes a target-specific assumption.
912 The ``inalloca`` argument attribute allows the caller to take the
913 address of outgoing stack arguments. An ``inalloca`` argument must
914 be a pointer to stack memory produced by an ``alloca`` instruction.
915 The alloca, or argument allocation, must also be tagged with the
916 inalloca keyword. Only the last argument may have the ``inalloca``
917 attribute, and that argument is guaranteed to be passed in memory.
919 An argument allocation may be used by a call at most once because
920 the call may deallocate it. The ``inalloca`` attribute cannot be
921 used in conjunction with other attributes that affect argument
922 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
923 ``inalloca`` attribute also disables LLVM's implicit lowering of
924 large aggregate return values, which means that frontend authors
925 must lower them with ``sret`` pointers.
927 When the call site is reached, the argument allocation must have
928 been the most recent stack allocation that is still live, or the
929 results are undefined. It is possible to allocate additional stack
930 space after an argument allocation and before its call site, but it
931 must be cleared off with :ref:`llvm.stackrestore
934 See :doc:`InAlloca` for more information on how to use this
938 This indicates that the pointer parameter specifies the address of a
939 structure that is the return value of the function in the source
940 program. This pointer must be guaranteed by the caller to be valid:
941 loads and stores to the structure may be assumed by the callee
942 not to trap and to be properly aligned. This may only be applied to
943 the first parameter. This is not a valid attribute for return
947 This indicates that the pointer value may be assumed by the optimizer to
948 have the specified alignment.
950 Note that this attribute has additional semantics when combined with the
956 This indicates that objects accessed via pointer values
957 :ref:`based <pointeraliasing>` on the argument or return value are not also
958 accessed, during the execution of the function, via pointer values not
959 *based* on the argument or return value. The attribute on a return value
960 also has additional semantics described below. The caller shares the
961 responsibility with the callee for ensuring that these requirements are met.
962 For further details, please see the discussion of the NoAlias response in
963 :ref:`alias analysis <Must, May, or No>`.
965 Note that this definition of ``noalias`` is intentionally similar
966 to the definition of ``restrict`` in C99 for function arguments.
968 For function return values, C99's ``restrict`` is not meaningful,
969 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
970 attribute on return values are stronger than the semantics of the attribute
971 when used on function arguments. On function return values, the ``noalias``
972 attribute indicates that the function acts like a system memory allocation
973 function, returning a pointer to allocated storage disjoint from the
974 storage for any other object accessible to the caller.
977 This indicates that the callee does not make any copies of the
978 pointer that outlive the callee itself. This is not a valid
979 attribute for return values.
984 This indicates that the pointer parameter can be excised using the
985 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
986 attribute for return values and can only be applied to one parameter.
989 This indicates that the function always returns the argument as its return
990 value. This is an optimization hint to the code generator when generating
991 the caller, allowing tail call optimization and omission of register saves
992 and restores in some cases; it is not checked or enforced when generating
993 the callee. The parameter and the function return type must be valid
994 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
995 valid attribute for return values and can only be applied to one parameter.
998 This indicates that the parameter or return pointer is not null. This
999 attribute may only be applied to pointer typed parameters. This is not
1000 checked or enforced by LLVM, the caller must ensure that the pointer
1001 passed in is non-null, or the callee must ensure that the returned pointer
1004 ``dereferenceable(<n>)``
1005 This indicates that the parameter or return pointer is dereferenceable. This
1006 attribute may only be applied to pointer typed parameters. A pointer that
1007 is dereferenceable can be loaded from speculatively without a risk of
1008 trapping. The number of bytes known to be dereferenceable must be provided
1009 in parentheses. It is legal for the number of bytes to be less than the
1010 size of the pointee type. The ``nonnull`` attribute does not imply
1011 dereferenceability (consider a pointer to one element past the end of an
1012 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1013 ``addrspace(0)`` (which is the default address space).
1017 Garbage Collector Names
1018 -----------------------
1020 Each function may specify a garbage collector name, which is simply a
1023 .. code-block:: llvm
1025 define void @f() gc "name" { ... }
1027 The compiler declares the supported values of *name*. Specifying a
1028 collector will cause the compiler to alter its output in order to
1029 support the named garbage collection algorithm.
1036 Prefix data is data associated with a function which the code
1037 generator will emit immediately before the function's entrypoint.
1038 The purpose of this feature is to allow frontends to associate
1039 language-specific runtime metadata with specific functions and make it
1040 available through the function pointer while still allowing the
1041 function pointer to be called.
1043 To access the data for a given function, a program may bitcast the
1044 function pointer to a pointer to the constant's type and dereference
1045 index -1. This implies that the IR symbol points just past the end of
1046 the prefix data. For instance, take the example of a function annotated
1047 with a single ``i32``,
1049 .. code-block:: llvm
1051 define void @f() prefix i32 123 { ... }
1053 The prefix data can be referenced as,
1055 .. code-block:: llvm
1057 %0 = bitcast *void () @f to *i32
1058 %a = getelementptr inbounds *i32 %0, i32 -1
1061 Prefix data is laid out as if it were an initializer for a global variable
1062 of the prefix data's type. The function will be placed such that the
1063 beginning of the prefix data is aligned. This means that if the size
1064 of the prefix data is not a multiple of the alignment size, the
1065 function's entrypoint will not be aligned. If alignment of the
1066 function's entrypoint is desired, padding must be added to the prefix
1069 A function may have prefix data but no body. This has similar semantics
1070 to the ``available_externally`` linkage in that the data may be used by the
1071 optimizers but will not be emitted in the object file.
1078 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1079 be inserted prior to the function body. This can be used for enabling
1080 function hot-patching and instrumentation.
1082 To maintain the semantics of ordinary function calls, the prologue data must
1083 have a particular format. Specifically, it must begin with a sequence of
1084 bytes which decode to a sequence of machine instructions, valid for the
1085 module's target, which transfer control to the point immediately succeeding
1086 the prologue data, without performing any other visible action. This allows
1087 the inliner and other passes to reason about the semantics of the function
1088 definition without needing to reason about the prologue data. Obviously this
1089 makes the format of the prologue data highly target dependent.
1091 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1092 which encodes the ``nop`` instruction:
1094 .. code-block:: llvm
1096 define void @f() prologue i8 144 { ... }
1098 Generally prologue data can be formed by encoding a relative branch instruction
1099 which skips the metadata, as in this example of valid prologue data for the
1100 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1102 .. code-block:: llvm
1104 %0 = type <{ i8, i8, i8* }>
1106 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1108 A function may have prologue data but no body. This has similar semantics
1109 to the ``available_externally`` linkage in that the data may be used by the
1110 optimizers but will not be emitted in the object file.
1117 Attribute groups are groups of attributes that are referenced by objects within
1118 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1119 functions will use the same set of attributes. In the degenerative case of a
1120 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1121 group will capture the important command line flags used to build that file.
1123 An attribute group is a module-level object. To use an attribute group, an
1124 object references the attribute group's ID (e.g. ``#37``). An object may refer
1125 to more than one attribute group. In that situation, the attributes from the
1126 different groups are merged.
1128 Here is an example of attribute groups for a function that should always be
1129 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1131 .. code-block:: llvm
1133 ; Target-independent attributes:
1134 attributes #0 = { alwaysinline alignstack=4 }
1136 ; Target-dependent attributes:
1137 attributes #1 = { "no-sse" }
1139 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1140 define void @f() #0 #1 { ... }
1147 Function attributes are set to communicate additional information about
1148 a function. Function attributes are considered to be part of the
1149 function, not of the function type, so functions with different function
1150 attributes can have the same function type.
1152 Function attributes are simple keywords that follow the type specified.
1153 If multiple attributes are needed, they are space separated. For
1156 .. code-block:: llvm
1158 define void @f() noinline { ... }
1159 define void @f() alwaysinline { ... }
1160 define void @f() alwaysinline optsize { ... }
1161 define void @f() optsize { ... }
1164 This attribute indicates that, when emitting the prologue and
1165 epilogue, the backend should forcibly align the stack pointer.
1166 Specify the desired alignment, which must be a power of two, in
1169 This attribute indicates that the inliner should attempt to inline
1170 this function into callers whenever possible, ignoring any active
1171 inlining size threshold for this caller.
1173 This indicates that the callee function at a call site should be
1174 recognized as a built-in function, even though the function's declaration
1175 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1176 direct calls to functions that are declared with the ``nobuiltin``
1179 This attribute indicates that this function is rarely called. When
1180 computing edge weights, basic blocks post-dominated by a cold
1181 function call are also considered to be cold; and, thus, given low
1184 This attribute indicates that the source code contained a hint that
1185 inlining this function is desirable (such as the "inline" keyword in
1186 C/C++). It is just a hint; it imposes no requirements on the
1189 This attribute indicates that the function should be added to a
1190 jump-instruction table at code-generation time, and that all address-taken
1191 references to this function should be replaced with a reference to the
1192 appropriate jump-instruction-table function pointer. Note that this creates
1193 a new pointer for the original function, which means that code that depends
1194 on function-pointer identity can break. So, any function annotated with
1195 ``jumptable`` must also be ``unnamed_addr``.
1197 This attribute suggests that optimization passes and code generator
1198 passes make choices that keep the code size of this function as small
1199 as possible and perform optimizations that may sacrifice runtime
1200 performance in order to minimize the size of the generated code.
1202 This attribute disables prologue / epilogue emission for the
1203 function. This can have very system-specific consequences.
1205 This indicates that the callee function at a call site is not recognized as
1206 a built-in function. LLVM will retain the original call and not replace it
1207 with equivalent code based on the semantics of the built-in function, unless
1208 the call site uses the ``builtin`` attribute. This is valid at call sites
1209 and on function declarations and definitions.
1211 This attribute indicates that calls to the function cannot be
1212 duplicated. A call to a ``noduplicate`` function may be moved
1213 within its parent function, but may not be duplicated within
1214 its parent function.
1216 A function containing a ``noduplicate`` call may still
1217 be an inlining candidate, provided that the call is not
1218 duplicated by inlining. That implies that the function has
1219 internal linkage and only has one call site, so the original
1220 call is dead after inlining.
1222 This attributes disables implicit floating point instructions.
1224 This attribute indicates that the inliner should never inline this
1225 function in any situation. This attribute may not be used together
1226 with the ``alwaysinline`` attribute.
1228 This attribute suppresses lazy symbol binding for the function. This
1229 may make calls to the function faster, at the cost of extra program
1230 startup time if the function is not called during program startup.
1232 This attribute indicates that the code generator should not use a
1233 red zone, even if the target-specific ABI normally permits it.
1235 This function attribute indicates that the function never returns
1236 normally. This produces undefined behavior at runtime if the
1237 function ever does dynamically return.
1239 This function attribute indicates that the function never raises an
1240 exception. If the function does raise an exception, its runtime
1241 behavior is undefined. However, functions marked nounwind may still
1242 trap or generate asynchronous exceptions. Exception handling schemes
1243 that are recognized by LLVM to handle asynchronous exceptions, such
1244 as SEH, will still provide their implementation defined semantics.
1246 This function attribute indicates that the function is not optimized
1247 by any optimization or code generator passes with the
1248 exception of interprocedural optimization passes.
1249 This attribute cannot be used together with the ``alwaysinline``
1250 attribute; this attribute is also incompatible
1251 with the ``minsize`` attribute and the ``optsize`` attribute.
1253 This attribute requires the ``noinline`` attribute to be specified on
1254 the function as well, so the function is never inlined into any caller.
1255 Only functions with the ``alwaysinline`` attribute are valid
1256 candidates for inlining into the body of this function.
1258 This attribute suggests that optimization passes and code generator
1259 passes make choices that keep the code size of this function low,
1260 and otherwise do optimizations specifically to reduce code size as
1261 long as they do not significantly impact runtime performance.
1263 On a function, this attribute indicates that the function computes its
1264 result (or decides to unwind an exception) based strictly on its arguments,
1265 without dereferencing any pointer arguments or otherwise accessing
1266 any mutable state (e.g. memory, control registers, etc) visible to
1267 caller functions. It does not write through any pointer arguments
1268 (including ``byval`` arguments) and never changes any state visible
1269 to callers. This means that it cannot unwind exceptions by calling
1270 the ``C++`` exception throwing methods.
1272 On an argument, this attribute indicates that the function does not
1273 dereference that pointer argument, even though it may read or write the
1274 memory that the pointer points to if accessed through other pointers.
1276 On a function, this attribute indicates that the function does not write
1277 through any pointer arguments (including ``byval`` arguments) or otherwise
1278 modify any state (e.g. memory, control registers, etc) visible to
1279 caller functions. It may dereference pointer arguments and read
1280 state that may be set in the caller. A readonly function always
1281 returns the same value (or unwinds an exception identically) when
1282 called with the same set of arguments and global state. It cannot
1283 unwind an exception by calling the ``C++`` exception throwing
1286 On an argument, this attribute indicates that the function does not write
1287 through this pointer argument, even though it may write to the memory that
1288 the pointer points to.
1290 This attribute indicates that this function can return twice. The C
1291 ``setjmp`` is an example of such a function. The compiler disables
1292 some optimizations (like tail calls) in the caller of these
1294 ``sanitize_address``
1295 This attribute indicates that AddressSanitizer checks
1296 (dynamic address safety analysis) are enabled for this function.
1298 This attribute indicates that MemorySanitizer checks (dynamic detection
1299 of accesses to uninitialized memory) are enabled for this function.
1301 This attribute indicates that ThreadSanitizer checks
1302 (dynamic thread safety analysis) are enabled for this function.
1304 This attribute indicates that the function should emit a stack
1305 smashing protector. It is in the form of a "canary" --- a random value
1306 placed on the stack before the local variables that's checked upon
1307 return from the function to see if it has been overwritten. A
1308 heuristic is used to determine if a function needs stack protectors
1309 or not. The heuristic used will enable protectors for functions with:
1311 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1312 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1313 - Calls to alloca() with variable sizes or constant sizes greater than
1314 ``ssp-buffer-size``.
1316 Variables that are identified as requiring a protector will be arranged
1317 on the stack such that they are adjacent to the stack protector guard.
1319 If a function that has an ``ssp`` attribute is inlined into a
1320 function that doesn't have an ``ssp`` attribute, then the resulting
1321 function will have an ``ssp`` attribute.
1323 This attribute indicates that the function should *always* emit a
1324 stack smashing protector. This overrides the ``ssp`` function
1327 Variables that are identified as requiring a protector will be arranged
1328 on the stack such that they are adjacent to the stack protector guard.
1329 The specific layout rules are:
1331 #. Large arrays and structures containing large arrays
1332 (``>= ssp-buffer-size``) are closest to the stack protector.
1333 #. Small arrays and structures containing small arrays
1334 (``< ssp-buffer-size``) are 2nd closest to the protector.
1335 #. Variables that have had their address taken are 3rd closest to the
1338 If a function that has an ``sspreq`` attribute is inlined into a
1339 function that doesn't have an ``sspreq`` attribute or which has an
1340 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1341 an ``sspreq`` attribute.
1343 This attribute indicates that the function should emit a stack smashing
1344 protector. This attribute causes a strong heuristic to be used when
1345 determining if a function needs stack protectors. The strong heuristic
1346 will enable protectors for functions with:
1348 - Arrays of any size and type
1349 - Aggregates containing an array of any size and type.
1350 - Calls to alloca().
1351 - Local variables that have had their address taken.
1353 Variables that are identified as requiring a protector will be arranged
1354 on the stack such that they are adjacent to the stack protector guard.
1355 The specific layout rules are:
1357 #. Large arrays and structures containing large arrays
1358 (``>= ssp-buffer-size``) are closest to the stack protector.
1359 #. Small arrays and structures containing small arrays
1360 (``< ssp-buffer-size``) are 2nd closest to the protector.
1361 #. Variables that have had their address taken are 3rd closest to the
1364 This overrides the ``ssp`` function attribute.
1366 If a function that has an ``sspstrong`` attribute is inlined into a
1367 function that doesn't have an ``sspstrong`` attribute, then the
1368 resulting function will have an ``sspstrong`` attribute.
1370 This attribute indicates that the ABI being targeted requires that
1371 an unwind table entry be produce for this function even if we can
1372 show that no exceptions passes by it. This is normally the case for
1373 the ELF x86-64 abi, but it can be disabled for some compilation
1378 Module-Level Inline Assembly
1379 ----------------------------
1381 Modules may contain "module-level inline asm" blocks, which corresponds
1382 to the GCC "file scope inline asm" blocks. These blocks are internally
1383 concatenated by LLVM and treated as a single unit, but may be separated
1384 in the ``.ll`` file if desired. The syntax is very simple:
1386 .. code-block:: llvm
1388 module asm "inline asm code goes here"
1389 module asm "more can go here"
1391 The strings can contain any character by escaping non-printable
1392 characters. The escape sequence used is simply "\\xx" where "xx" is the
1393 two digit hex code for the number.
1395 The inline asm code is simply printed to the machine code .s file when
1396 assembly code is generated.
1398 .. _langref_datalayout:
1403 A module may specify a target specific data layout string that specifies
1404 how data is to be laid out in memory. The syntax for the data layout is
1407 .. code-block:: llvm
1409 target datalayout = "layout specification"
1411 The *layout specification* consists of a list of specifications
1412 separated by the minus sign character ('-'). Each specification starts
1413 with a letter and may include other information after the letter to
1414 define some aspect of the data layout. The specifications accepted are
1418 Specifies that the target lays out data in big-endian form. That is,
1419 the bits with the most significance have the lowest address
1422 Specifies that the target lays out data in little-endian form. That
1423 is, the bits with the least significance have the lowest address
1426 Specifies the natural alignment of the stack in bits. Alignment
1427 promotion of stack variables is limited to the natural stack
1428 alignment to avoid dynamic stack realignment. The stack alignment
1429 must be a multiple of 8-bits. If omitted, the natural stack
1430 alignment defaults to "unspecified", which does not prevent any
1431 alignment promotions.
1432 ``p[n]:<size>:<abi>:<pref>``
1433 This specifies the *size* of a pointer and its ``<abi>`` and
1434 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1435 bits. The address space, ``n`` is optional, and if not specified,
1436 denotes the default address space 0. The value of ``n`` must be
1437 in the range [1,2^23).
1438 ``i<size>:<abi>:<pref>``
1439 This specifies the alignment for an integer type of a given bit
1440 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1441 ``v<size>:<abi>:<pref>``
1442 This specifies the alignment for a vector type of a given bit
1444 ``f<size>:<abi>:<pref>``
1445 This specifies the alignment for a floating point type of a given bit
1446 ``<size>``. Only values of ``<size>`` that are supported by the target
1447 will work. 32 (float) and 64 (double) are supported on all targets; 80
1448 or 128 (different flavors of long double) are also supported on some
1451 This specifies the alignment for an object of aggregate type.
1453 If present, specifies that llvm names are mangled in the output. The
1456 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1457 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1458 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1459 symbols get a ``_`` prefix.
1460 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1461 functions also get a suffix based on the frame size.
1462 ``n<size1>:<size2>:<size3>...``
1463 This specifies a set of native integer widths for the target CPU in
1464 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1465 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1466 this set are considered to support most general arithmetic operations
1469 On every specification that takes a ``<abi>:<pref>``, specifying the
1470 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1471 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1473 When constructing the data layout for a given target, LLVM starts with a
1474 default set of specifications which are then (possibly) overridden by
1475 the specifications in the ``datalayout`` keyword. The default
1476 specifications are given in this list:
1478 - ``E`` - big endian
1479 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1480 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1481 same as the default address space.
1482 - ``S0`` - natural stack alignment is unspecified
1483 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1484 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1485 - ``i16:16:16`` - i16 is 16-bit aligned
1486 - ``i32:32:32`` - i32 is 32-bit aligned
1487 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1488 alignment of 64-bits
1489 - ``f16:16:16`` - half is 16-bit aligned
1490 - ``f32:32:32`` - float is 32-bit aligned
1491 - ``f64:64:64`` - double is 64-bit aligned
1492 - ``f128:128:128`` - quad is 128-bit aligned
1493 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1494 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1495 - ``a:0:64`` - aggregates are 64-bit aligned
1497 When LLVM is determining the alignment for a given type, it uses the
1500 #. If the type sought is an exact match for one of the specifications,
1501 that specification is used.
1502 #. If no match is found, and the type sought is an integer type, then
1503 the smallest integer type that is larger than the bitwidth of the
1504 sought type is used. If none of the specifications are larger than
1505 the bitwidth then the largest integer type is used. For example,
1506 given the default specifications above, the i7 type will use the
1507 alignment of i8 (next largest) while both i65 and i256 will use the
1508 alignment of i64 (largest specified).
1509 #. If no match is found, and the type sought is a vector type, then the
1510 largest vector type that is smaller than the sought vector type will
1511 be used as a fall back. This happens because <128 x double> can be
1512 implemented in terms of 64 <2 x double>, for example.
1514 The function of the data layout string may not be what you expect.
1515 Notably, this is not a specification from the frontend of what alignment
1516 the code generator should use.
1518 Instead, if specified, the target data layout is required to match what
1519 the ultimate *code generator* expects. This string is used by the
1520 mid-level optimizers to improve code, and this only works if it matches
1521 what the ultimate code generator uses. If you would like to generate IR
1522 that does not embed this target-specific detail into the IR, then you
1523 don't have to specify the string. This will disable some optimizations
1524 that require precise layout information, but this also prevents those
1525 optimizations from introducing target specificity into the IR.
1532 A module may specify a target triple string that describes the target
1533 host. The syntax for the target triple is simply:
1535 .. code-block:: llvm
1537 target triple = "x86_64-apple-macosx10.7.0"
1539 The *target triple* string consists of a series of identifiers delimited
1540 by the minus sign character ('-'). The canonical forms are:
1544 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1545 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1547 This information is passed along to the backend so that it generates
1548 code for the proper architecture. It's possible to override this on the
1549 command line with the ``-mtriple`` command line option.
1551 .. _pointeraliasing:
1553 Pointer Aliasing Rules
1554 ----------------------
1556 Any memory access must be done through a pointer value associated with
1557 an address range of the memory access, otherwise the behavior is
1558 undefined. Pointer values are associated with address ranges according
1559 to the following rules:
1561 - A pointer value is associated with the addresses associated with any
1562 value it is *based* on.
1563 - An address of a global variable is associated with the address range
1564 of the variable's storage.
1565 - The result value of an allocation instruction is associated with the
1566 address range of the allocated storage.
1567 - A null pointer in the default address-space is associated with no
1569 - An integer constant other than zero or a pointer value returned from
1570 a function not defined within LLVM may be associated with address
1571 ranges allocated through mechanisms other than those provided by
1572 LLVM. Such ranges shall not overlap with any ranges of addresses
1573 allocated by mechanisms provided by LLVM.
1575 A pointer value is *based* on another pointer value according to the
1578 - A pointer value formed from a ``getelementptr`` operation is *based*
1579 on the first operand of the ``getelementptr``.
1580 - The result value of a ``bitcast`` is *based* on the operand of the
1582 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1583 values that contribute (directly or indirectly) to the computation of
1584 the pointer's value.
1585 - The "*based* on" relationship is transitive.
1587 Note that this definition of *"based"* is intentionally similar to the
1588 definition of *"based"* in C99, though it is slightly weaker.
1590 LLVM IR does not associate types with memory. The result type of a
1591 ``load`` merely indicates the size and alignment of the memory from
1592 which to load, as well as the interpretation of the value. The first
1593 operand type of a ``store`` similarly only indicates the size and
1594 alignment of the store.
1596 Consequently, type-based alias analysis, aka TBAA, aka
1597 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1598 :ref:`Metadata <metadata>` may be used to encode additional information
1599 which specialized optimization passes may use to implement type-based
1604 Volatile Memory Accesses
1605 ------------------------
1607 Certain memory accesses, such as :ref:`load <i_load>`'s,
1608 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1609 marked ``volatile``. The optimizers must not change the number of
1610 volatile operations or change their order of execution relative to other
1611 volatile operations. The optimizers *may* change the order of volatile
1612 operations relative to non-volatile operations. This is not Java's
1613 "volatile" and has no cross-thread synchronization behavior.
1615 IR-level volatile loads and stores cannot safely be optimized into
1616 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1617 flagged volatile. Likewise, the backend should never split or merge
1618 target-legal volatile load/store instructions.
1620 .. admonition:: Rationale
1622 Platforms may rely on volatile loads and stores of natively supported
1623 data width to be executed as single instruction. For example, in C
1624 this holds for an l-value of volatile primitive type with native
1625 hardware support, but not necessarily for aggregate types. The
1626 frontend upholds these expectations, which are intentionally
1627 unspecified in the IR. The rules above ensure that IR transformation
1628 do not violate the frontend's contract with the language.
1632 Memory Model for Concurrent Operations
1633 --------------------------------------
1635 The LLVM IR does not define any way to start parallel threads of
1636 execution or to register signal handlers. Nonetheless, there are
1637 platform-specific ways to create them, and we define LLVM IR's behavior
1638 in their presence. This model is inspired by the C++0x memory model.
1640 For a more informal introduction to this model, see the :doc:`Atomics`.
1642 We define a *happens-before* partial order as the least partial order
1645 - Is a superset of single-thread program order, and
1646 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1647 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1648 techniques, like pthread locks, thread creation, thread joining,
1649 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1650 Constraints <ordering>`).
1652 Note that program order does not introduce *happens-before* edges
1653 between a thread and signals executing inside that thread.
1655 Every (defined) read operation (load instructions, memcpy, atomic
1656 loads/read-modify-writes, etc.) R reads a series of bytes written by
1657 (defined) write operations (store instructions, atomic
1658 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1659 section, initialized globals are considered to have a write of the
1660 initializer which is atomic and happens before any other read or write
1661 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1662 may see any write to the same byte, except:
1664 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1665 write\ :sub:`2` happens before R\ :sub:`byte`, then
1666 R\ :sub:`byte` does not see write\ :sub:`1`.
1667 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1668 R\ :sub:`byte` does not see write\ :sub:`3`.
1670 Given that definition, R\ :sub:`byte` is defined as follows:
1672 - If R is volatile, the result is target-dependent. (Volatile is
1673 supposed to give guarantees which can support ``sig_atomic_t`` in
1674 C/C++, and may be used for accesses to addresses that do not behave
1675 like normal memory. It does not generally provide cross-thread
1677 - Otherwise, if there is no write to the same byte that happens before
1678 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1679 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1680 R\ :sub:`byte` returns the value written by that write.
1681 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1682 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1683 Memory Ordering Constraints <ordering>` section for additional
1684 constraints on how the choice is made.
1685 - Otherwise R\ :sub:`byte` returns ``undef``.
1687 R returns the value composed of the series of bytes it read. This
1688 implies that some bytes within the value may be ``undef`` **without**
1689 the entire value being ``undef``. Note that this only defines the
1690 semantics of the operation; it doesn't mean that targets will emit more
1691 than one instruction to read the series of bytes.
1693 Note that in cases where none of the atomic intrinsics are used, this
1694 model places only one restriction on IR transformations on top of what
1695 is required for single-threaded execution: introducing a store to a byte
1696 which might not otherwise be stored is not allowed in general.
1697 (Specifically, in the case where another thread might write to and read
1698 from an address, introducing a store can change a load that may see
1699 exactly one write into a load that may see multiple writes.)
1703 Atomic Memory Ordering Constraints
1704 ----------------------------------
1706 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1707 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1708 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1709 ordering parameters that determine which other atomic instructions on
1710 the same address they *synchronize with*. These semantics are borrowed
1711 from Java and C++0x, but are somewhat more colloquial. If these
1712 descriptions aren't precise enough, check those specs (see spec
1713 references in the :doc:`atomics guide <Atomics>`).
1714 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1715 differently since they don't take an address. See that instruction's
1716 documentation for details.
1718 For a simpler introduction to the ordering constraints, see the
1722 The set of values that can be read is governed by the happens-before
1723 partial order. A value cannot be read unless some operation wrote
1724 it. This is intended to provide a guarantee strong enough to model
1725 Java's non-volatile shared variables. This ordering cannot be
1726 specified for read-modify-write operations; it is not strong enough
1727 to make them atomic in any interesting way.
1729 In addition to the guarantees of ``unordered``, there is a single
1730 total order for modifications by ``monotonic`` operations on each
1731 address. All modification orders must be compatible with the
1732 happens-before order. There is no guarantee that the modification
1733 orders can be combined to a global total order for the whole program
1734 (and this often will not be possible). The read in an atomic
1735 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1736 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1737 order immediately before the value it writes. If one atomic read
1738 happens before another atomic read of the same address, the later
1739 read must see the same value or a later value in the address's
1740 modification order. This disallows reordering of ``monotonic`` (or
1741 stronger) operations on the same address. If an address is written
1742 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1743 read that address repeatedly, the other threads must eventually see
1744 the write. This corresponds to the C++0x/C1x
1745 ``memory_order_relaxed``.
1747 In addition to the guarantees of ``monotonic``, a
1748 *synchronizes-with* edge may be formed with a ``release`` operation.
1749 This is intended to model C++'s ``memory_order_acquire``.
1751 In addition to the guarantees of ``monotonic``, if this operation
1752 writes a value which is subsequently read by an ``acquire``
1753 operation, it *synchronizes-with* that operation. (This isn't a
1754 complete description; see the C++0x definition of a release
1755 sequence.) This corresponds to the C++0x/C1x
1756 ``memory_order_release``.
1757 ``acq_rel`` (acquire+release)
1758 Acts as both an ``acquire`` and ``release`` operation on its
1759 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1760 ``seq_cst`` (sequentially consistent)
1761 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1762 operation that only reads, ``release`` for an operation that only
1763 writes), there is a global total order on all
1764 sequentially-consistent operations on all addresses, which is
1765 consistent with the *happens-before* partial order and with the
1766 modification orders of all the affected addresses. Each
1767 sequentially-consistent read sees the last preceding write to the
1768 same address in this global order. This corresponds to the C++0x/C1x
1769 ``memory_order_seq_cst`` and Java volatile.
1773 If an atomic operation is marked ``singlethread``, it only *synchronizes
1774 with* or participates in modification and seq\_cst total orderings with
1775 other operations running in the same thread (for example, in signal
1783 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1784 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1785 :ref:`frem <i_frem>`) have the following flags that can set to enable
1786 otherwise unsafe floating point operations
1789 No NaNs - Allow optimizations to assume the arguments and result are not
1790 NaN. Such optimizations are required to retain defined behavior over
1791 NaNs, but the value of the result is undefined.
1794 No Infs - Allow optimizations to assume the arguments and result are not
1795 +/-Inf. Such optimizations are required to retain defined behavior over
1796 +/-Inf, but the value of the result is undefined.
1799 No Signed Zeros - Allow optimizations to treat the sign of a zero
1800 argument or result as insignificant.
1803 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1804 argument rather than perform division.
1807 Fast - Allow algebraically equivalent transformations that may
1808 dramatically change results in floating point (e.g. reassociate). This
1809 flag implies all the others.
1813 Use-list Order Directives
1814 -------------------------
1816 Use-list directives encode the in-memory order of each use-list, allowing the
1817 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1818 indexes that are assigned to the referenced value's uses. The referenced
1819 value's use-list is immediately sorted by these indexes.
1821 Use-list directives may appear at function scope or global scope. They are not
1822 instructions, and have no effect on the semantics of the IR. When they're at
1823 function scope, they must appear after the terminator of the final basic block.
1825 If basic blocks have their address taken via ``blockaddress()`` expressions,
1826 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1833 uselistorder <ty> <value>, { <order-indexes> }
1834 uselistorder_bb @function, %block { <order-indexes> }
1840 define void @foo(i32 %arg1, i32 %arg2) {
1842 ; ... instructions ...
1844 ; ... instructions ...
1846 ; At function scope.
1847 uselistorder i32 %arg1, { 1, 0, 2 }
1848 uselistorder label %bb, { 1, 0 }
1852 uselistorder i32* @global, { 1, 2, 0 }
1853 uselistorder i32 7, { 1, 0 }
1854 uselistorder i32 (i32) @bar, { 1, 0 }
1855 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1862 The LLVM type system is one of the most important features of the
1863 intermediate representation. Being typed enables a number of
1864 optimizations to be performed on the intermediate representation
1865 directly, without having to do extra analyses on the side before the
1866 transformation. A strong type system makes it easier to read the
1867 generated code and enables novel analyses and transformations that are
1868 not feasible to perform on normal three address code representations.
1878 The void type does not represent any value and has no size.
1896 The function type can be thought of as a function signature. It consists of a
1897 return type and a list of formal parameter types. The return type of a function
1898 type is a void type or first class type --- except for :ref:`label <t_label>`
1899 and :ref:`metadata <t_metadata>` types.
1905 <returntype> (<parameter list>)
1907 ...where '``<parameter list>``' is a comma-separated list of type
1908 specifiers. Optionally, the parameter list may include a type ``...``, which
1909 indicates that the function takes a variable number of arguments. Variable
1910 argument functions can access their arguments with the :ref:`variable argument
1911 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1912 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1916 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1917 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1918 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1919 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1920 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1921 | ``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. |
1922 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1923 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1924 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1931 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1932 Values of these types are the only ones which can be produced by
1940 These are the types that are valid in registers from CodeGen's perspective.
1949 The integer type is a very simple type that simply specifies an
1950 arbitrary bit width for the integer type desired. Any bit width from 1
1951 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1959 The number of bits the integer will occupy is specified by the ``N``
1965 +----------------+------------------------------------------------+
1966 | ``i1`` | a single-bit integer. |
1967 +----------------+------------------------------------------------+
1968 | ``i32`` | a 32-bit integer. |
1969 +----------------+------------------------------------------------+
1970 | ``i1942652`` | a really big integer of over 1 million bits. |
1971 +----------------+------------------------------------------------+
1975 Floating Point Types
1976 """"""""""""""""""""
1985 - 16-bit floating point value
1988 - 32-bit floating point value
1991 - 64-bit floating point value
1994 - 128-bit floating point value (112-bit mantissa)
1997 - 80-bit floating point value (X87)
2000 - 128-bit floating point value (two 64-bits)
2007 The x86_mmx type represents a value held in an MMX register on an x86
2008 machine. The operations allowed on it are quite limited: parameters and
2009 return values, load and store, and bitcast. User-specified MMX
2010 instructions are represented as intrinsic or asm calls with arguments
2011 and/or results of this type. There are no arrays, vectors or constants
2028 The pointer type is used to specify memory locations. Pointers are
2029 commonly used to reference objects in memory.
2031 Pointer types may have an optional address space attribute defining the
2032 numbered address space where the pointed-to object resides. The default
2033 address space is number zero. The semantics of non-zero address spaces
2034 are target-specific.
2036 Note that LLVM does not permit pointers to void (``void*``) nor does it
2037 permit pointers to labels (``label*``). Use ``i8*`` instead.
2047 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2048 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2049 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2050 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2051 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2052 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2053 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2062 A vector type is a simple derived type that represents a vector of
2063 elements. Vector types are used when multiple primitive data are
2064 operated in parallel using a single instruction (SIMD). A vector type
2065 requires a size (number of elements) and an underlying primitive data
2066 type. Vector types are considered :ref:`first class <t_firstclass>`.
2072 < <# elements> x <elementtype> >
2074 The number of elements is a constant integer value larger than 0;
2075 elementtype may be any integer, floating point or pointer type. Vectors
2076 of size zero are not allowed.
2080 +-------------------+--------------------------------------------------+
2081 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2082 +-------------------+--------------------------------------------------+
2083 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2084 +-------------------+--------------------------------------------------+
2085 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2086 +-------------------+--------------------------------------------------+
2087 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2088 +-------------------+--------------------------------------------------+
2097 The label type represents code labels.
2112 The metadata type represents embedded metadata. No derived types may be
2113 created from metadata except for :ref:`function <t_function>` arguments.
2126 Aggregate Types are a subset of derived types that can contain multiple
2127 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2128 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2138 The array type is a very simple derived type that arranges elements
2139 sequentially in memory. The array type requires a size (number of
2140 elements) and an underlying data type.
2146 [<# elements> x <elementtype>]
2148 The number of elements is a constant integer value; ``elementtype`` may
2149 be any type with a size.
2153 +------------------+--------------------------------------+
2154 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2155 +------------------+--------------------------------------+
2156 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2157 +------------------+--------------------------------------+
2158 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2159 +------------------+--------------------------------------+
2161 Here are some examples of multidimensional arrays:
2163 +-----------------------------+----------------------------------------------------------+
2164 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2165 +-----------------------------+----------------------------------------------------------+
2166 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2167 +-----------------------------+----------------------------------------------------------+
2168 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2169 +-----------------------------+----------------------------------------------------------+
2171 There is no restriction on indexing beyond the end of the array implied
2172 by a static type (though there are restrictions on indexing beyond the
2173 bounds of an allocated object in some cases). This means that
2174 single-dimension 'variable sized array' addressing can be implemented in
2175 LLVM with a zero length array type. An implementation of 'pascal style
2176 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2186 The structure type is used to represent a collection of data members
2187 together in memory. The elements of a structure may be any type that has
2190 Structures in memory are accessed using '``load``' and '``store``' by
2191 getting a pointer to a field with the '``getelementptr``' instruction.
2192 Structures in registers are accessed using the '``extractvalue``' and
2193 '``insertvalue``' instructions.
2195 Structures may optionally be "packed" structures, which indicate that
2196 the alignment of the struct is one byte, and that there is no padding
2197 between the elements. In non-packed structs, padding between field types
2198 is inserted as defined by the DataLayout string in the module, which is
2199 required to match what the underlying code generator expects.
2201 Structures can either be "literal" or "identified". A literal structure
2202 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2203 identified types are always defined at the top level with a name.
2204 Literal types are uniqued by their contents and can never be recursive
2205 or opaque since there is no way to write one. Identified types can be
2206 recursive, can be opaqued, and are never uniqued.
2212 %T1 = type { <type list> } ; Identified normal struct type
2213 %T2 = type <{ <type list> }> ; Identified packed struct type
2217 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2218 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2219 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2220 | ``{ 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``. |
2221 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2222 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2223 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2227 Opaque Structure Types
2228 """"""""""""""""""""""
2232 Opaque structure types are used to represent named structure types that
2233 do not have a body specified. This corresponds (for example) to the C
2234 notion of a forward declared structure.
2245 +--------------+-------------------+
2246 | ``opaque`` | An opaque type. |
2247 +--------------+-------------------+
2254 LLVM has several different basic types of constants. This section
2255 describes them all and their syntax.
2260 **Boolean constants**
2261 The two strings '``true``' and '``false``' are both valid constants
2263 **Integer constants**
2264 Standard integers (such as '4') are constants of the
2265 :ref:`integer <t_integer>` type. Negative numbers may be used with
2267 **Floating point constants**
2268 Floating point constants use standard decimal notation (e.g.
2269 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2270 hexadecimal notation (see below). The assembler requires the exact
2271 decimal value of a floating-point constant. For example, the
2272 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2273 decimal in binary. Floating point constants must have a :ref:`floating
2274 point <t_floating>` type.
2275 **Null pointer constants**
2276 The identifier '``null``' is recognized as a null pointer constant
2277 and must be of :ref:`pointer type <t_pointer>`.
2279 The one non-intuitive notation for constants is the hexadecimal form of
2280 floating point constants. For example, the form
2281 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2282 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2283 constants are required (and the only time that they are generated by the
2284 disassembler) is when a floating point constant must be emitted but it
2285 cannot be represented as a decimal floating point number in a reasonable
2286 number of digits. For example, NaN's, infinities, and other special
2287 values are represented in their IEEE hexadecimal format so that assembly
2288 and disassembly do not cause any bits to change in the constants.
2290 When using the hexadecimal form, constants of types half, float, and
2291 double are represented using the 16-digit form shown above (which
2292 matches the IEEE754 representation for double); half and float values
2293 must, however, be exactly representable as IEEE 754 half and single
2294 precision, respectively. Hexadecimal format is always used for long
2295 double, and there are three forms of long double. The 80-bit format used
2296 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2297 128-bit format used by PowerPC (two adjacent doubles) is represented by
2298 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2299 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2300 will only work if they match the long double format on your target.
2301 The IEEE 16-bit format (half precision) is represented by ``0xH``
2302 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2303 (sign bit at the left).
2305 There are no constants of type x86_mmx.
2307 .. _complexconstants:
2312 Complex constants are a (potentially recursive) combination of simple
2313 constants and smaller complex constants.
2315 **Structure constants**
2316 Structure constants are represented with notation similar to
2317 structure type definitions (a comma separated list of elements,
2318 surrounded by braces (``{}``)). For example:
2319 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2320 "``@G = external global i32``". Structure constants must have
2321 :ref:`structure type <t_struct>`, and the number and types of elements
2322 must match those specified by the type.
2324 Array constants are represented with notation similar to array type
2325 definitions (a comma separated list of elements, surrounded by
2326 square brackets (``[]``)). For example:
2327 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2328 :ref:`array type <t_array>`, and the number and types of elements must
2329 match those specified by the type. As a special case, character array
2330 constants may also be represented as a double-quoted string using the ``c``
2331 prefix. For example: "``c"Hello World\0A\00"``".
2332 **Vector constants**
2333 Vector constants are represented with notation similar to vector
2334 type definitions (a comma separated list of elements, surrounded by
2335 less-than/greater-than's (``<>``)). For example:
2336 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2337 must have :ref:`vector type <t_vector>`, and the number and types of
2338 elements must match those specified by the type.
2339 **Zero initialization**
2340 The string '``zeroinitializer``' can be used to zero initialize a
2341 value to zero of *any* type, including scalar and
2342 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2343 having to print large zero initializers (e.g. for large arrays) and
2344 is always exactly equivalent to using explicit zero initializers.
2346 A metadata node is a constant tuple without types. For example:
2347 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2348 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2349 Unlike other typed constants that are meant to be interpreted as part of
2350 the instruction stream, metadata is a place to attach additional
2351 information such as debug info.
2353 Global Variable and Function Addresses
2354 --------------------------------------
2356 The addresses of :ref:`global variables <globalvars>` and
2357 :ref:`functions <functionstructure>` are always implicitly valid
2358 (link-time) constants. These constants are explicitly referenced when
2359 the :ref:`identifier for the global <identifiers>` is used and always have
2360 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2363 .. code-block:: llvm
2367 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2374 The string '``undef``' can be used anywhere a constant is expected, and
2375 indicates that the user of the value may receive an unspecified
2376 bit-pattern. Undefined values may be of any type (other than '``label``'
2377 or '``void``') and be used anywhere a constant is permitted.
2379 Undefined values are useful because they indicate to the compiler that
2380 the program is well defined no matter what value is used. This gives the
2381 compiler more freedom to optimize. Here are some examples of
2382 (potentially surprising) transformations that are valid (in pseudo IR):
2384 .. code-block:: llvm
2394 This is safe because all of the output bits are affected by the undef
2395 bits. Any output bit can have a zero or one depending on the input bits.
2397 .. code-block:: llvm
2408 These logical operations have bits that are not always affected by the
2409 input. For example, if ``%X`` has a zero bit, then the output of the
2410 '``and``' operation will always be a zero for that bit, no matter what
2411 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2412 optimize or assume that the result of the '``and``' is '``undef``'.
2413 However, it is safe to assume that all bits of the '``undef``' could be
2414 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2415 all the bits of the '``undef``' operand to the '``or``' could be set,
2416 allowing the '``or``' to be folded to -1.
2418 .. code-block:: llvm
2420 %A = select undef, %X, %Y
2421 %B = select undef, 42, %Y
2422 %C = select %X, %Y, undef
2432 This set of examples shows that undefined '``select``' (and conditional
2433 branch) conditions can go *either way*, but they have to come from one
2434 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2435 both known to have a clear low bit, then ``%A`` would have to have a
2436 cleared low bit. However, in the ``%C`` example, the optimizer is
2437 allowed to assume that the '``undef``' operand could be the same as
2438 ``%Y``, allowing the whole '``select``' to be eliminated.
2440 .. code-block:: llvm
2442 %A = xor undef, undef
2459 This example points out that two '``undef``' operands are not
2460 necessarily the same. This can be surprising to people (and also matches
2461 C semantics) where they assume that "``X^X``" is always zero, even if
2462 ``X`` is undefined. This isn't true for a number of reasons, but the
2463 short answer is that an '``undef``' "variable" can arbitrarily change
2464 its value over its "live range". This is true because the variable
2465 doesn't actually *have a live range*. Instead, the value is logically
2466 read from arbitrary registers that happen to be around when needed, so
2467 the value is not necessarily consistent over time. In fact, ``%A`` and
2468 ``%C`` need to have the same semantics or the core LLVM "replace all
2469 uses with" concept would not hold.
2471 .. code-block:: llvm
2479 These examples show the crucial difference between an *undefined value*
2480 and *undefined behavior*. An undefined value (like '``undef``') is
2481 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2482 operation can be constant folded to '``undef``', because the '``undef``'
2483 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2484 However, in the second example, we can make a more aggressive
2485 assumption: because the ``undef`` is allowed to be an arbitrary value,
2486 we are allowed to assume that it could be zero. Since a divide by zero
2487 has *undefined behavior*, we are allowed to assume that the operation
2488 does not execute at all. This allows us to delete the divide and all
2489 code after it. Because the undefined operation "can't happen", the
2490 optimizer can assume that it occurs in dead code.
2492 .. code-block:: llvm
2494 a: store undef -> %X
2495 b: store %X -> undef
2500 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2501 value can be assumed to not have any effect; we can assume that the
2502 value is overwritten with bits that happen to match what was already
2503 there. However, a store *to* an undefined location could clobber
2504 arbitrary memory, therefore, it has undefined behavior.
2511 Poison values are similar to :ref:`undef values <undefvalues>`, however
2512 they also represent the fact that an instruction or constant expression
2513 that cannot evoke side effects has nevertheless detected a condition
2514 that results in undefined behavior.
2516 There is currently no way of representing a poison value in the IR; they
2517 only exist when produced by operations such as :ref:`add <i_add>` with
2520 Poison value behavior is defined in terms of value *dependence*:
2522 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2523 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2524 their dynamic predecessor basic block.
2525 - Function arguments depend on the corresponding actual argument values
2526 in the dynamic callers of their functions.
2527 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2528 instructions that dynamically transfer control back to them.
2529 - :ref:`Invoke <i_invoke>` instructions depend on the
2530 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2531 call instructions that dynamically transfer control back to them.
2532 - Non-volatile loads and stores depend on the most recent stores to all
2533 of the referenced memory addresses, following the order in the IR
2534 (including loads and stores implied by intrinsics such as
2535 :ref:`@llvm.memcpy <int_memcpy>`.)
2536 - An instruction with externally visible side effects depends on the
2537 most recent preceding instruction with externally visible side
2538 effects, following the order in the IR. (This includes :ref:`volatile
2539 operations <volatile>`.)
2540 - An instruction *control-depends* on a :ref:`terminator
2541 instruction <terminators>` if the terminator instruction has
2542 multiple successors and the instruction is always executed when
2543 control transfers to one of the successors, and may not be executed
2544 when control is transferred to another.
2545 - Additionally, an instruction also *control-depends* on a terminator
2546 instruction if the set of instructions it otherwise depends on would
2547 be different if the terminator had transferred control to a different
2549 - Dependence is transitive.
2551 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2552 with the additional effect that any instruction that has a *dependence*
2553 on a poison value has undefined behavior.
2555 Here are some examples:
2557 .. code-block:: llvm
2560 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2561 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2562 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2563 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2565 store i32 %poison, i32* @g ; Poison value stored to memory.
2566 %poison2 = load i32* @g ; Poison value loaded back from memory.
2568 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2570 %narrowaddr = bitcast i32* @g to i16*
2571 %wideaddr = bitcast i32* @g to i64*
2572 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2573 %poison4 = load i64* %wideaddr ; Returns a poison value.
2575 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2576 br i1 %cmp, label %true, label %end ; Branch to either destination.
2579 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2580 ; it has undefined behavior.
2584 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2585 ; Both edges into this PHI are
2586 ; control-dependent on %cmp, so this
2587 ; always results in a poison value.
2589 store volatile i32 0, i32* @g ; This would depend on the store in %true
2590 ; if %cmp is true, or the store in %entry
2591 ; otherwise, so this is undefined behavior.
2593 br i1 %cmp, label %second_true, label %second_end
2594 ; The same branch again, but this time the
2595 ; true block doesn't have side effects.
2602 store volatile i32 0, i32* @g ; This time, the instruction always depends
2603 ; on the store in %end. Also, it is
2604 ; control-equivalent to %end, so this is
2605 ; well-defined (ignoring earlier undefined
2606 ; behavior in this example).
2610 Addresses of Basic Blocks
2611 -------------------------
2613 ``blockaddress(@function, %block)``
2615 The '``blockaddress``' constant computes the address of the specified
2616 basic block in the specified function, and always has an ``i8*`` type.
2617 Taking the address of the entry block is illegal.
2619 This value only has defined behavior when used as an operand to the
2620 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2621 against null. Pointer equality tests between labels addresses results in
2622 undefined behavior --- though, again, comparison against null is ok, and
2623 no label is equal to the null pointer. This may be passed around as an
2624 opaque pointer sized value as long as the bits are not inspected. This
2625 allows ``ptrtoint`` and arithmetic to be performed on these values so
2626 long as the original value is reconstituted before the ``indirectbr``
2629 Finally, some targets may provide defined semantics when using the value
2630 as the operand to an inline assembly, but that is target specific.
2634 Constant Expressions
2635 --------------------
2637 Constant expressions are used to allow expressions involving other
2638 constants to be used as constants. Constant expressions may be of any
2639 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2640 that does not have side effects (e.g. load and call are not supported).
2641 The following is the syntax for constant expressions:
2643 ``trunc (CST to TYPE)``
2644 Truncate a constant to another type. The bit size of CST must be
2645 larger than the bit size of TYPE. Both types must be integers.
2646 ``zext (CST to TYPE)``
2647 Zero extend a constant to another type. The bit size of CST must be
2648 smaller than the bit size of TYPE. Both types must be integers.
2649 ``sext (CST to TYPE)``
2650 Sign extend a constant to another type. The bit size of CST must be
2651 smaller than the bit size of TYPE. Both types must be integers.
2652 ``fptrunc (CST to TYPE)``
2653 Truncate a floating point constant to another floating point type.
2654 The size of CST must be larger than the size of TYPE. Both types
2655 must be floating point.
2656 ``fpext (CST to TYPE)``
2657 Floating point extend a constant to another type. The size of CST
2658 must be smaller or equal to the size of TYPE. Both types must be
2660 ``fptoui (CST to TYPE)``
2661 Convert a floating point constant to the corresponding unsigned
2662 integer constant. TYPE must be a scalar or vector integer type. CST
2663 must be of scalar or vector floating point type. Both CST and TYPE
2664 must be scalars, or vectors of the same number of elements. If the
2665 value won't fit in the integer type, the results are undefined.
2666 ``fptosi (CST to TYPE)``
2667 Convert a floating point constant to the corresponding signed
2668 integer constant. TYPE must be a scalar or vector integer type. CST
2669 must be of scalar or vector floating point type. Both CST and TYPE
2670 must be scalars, or vectors of the same number of elements. If the
2671 value won't fit in the integer type, the results are undefined.
2672 ``uitofp (CST to TYPE)``
2673 Convert an unsigned integer constant to the corresponding floating
2674 point constant. TYPE must be a scalar or vector floating point type.
2675 CST must be of scalar or vector integer type. Both CST and TYPE must
2676 be scalars, or vectors of the same number of elements. If the value
2677 won't fit in the floating point type, the results are undefined.
2678 ``sitofp (CST to TYPE)``
2679 Convert a signed integer constant to the corresponding floating
2680 point constant. TYPE must be a scalar or vector floating point type.
2681 CST must be of scalar or vector integer type. Both CST and TYPE must
2682 be scalars, or vectors of the same number of elements. If the value
2683 won't fit in the floating point type, the results are undefined.
2684 ``ptrtoint (CST to TYPE)``
2685 Convert a pointer typed constant to the corresponding integer
2686 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2687 pointer type. The ``CST`` value is zero extended, truncated, or
2688 unchanged to make it fit in ``TYPE``.
2689 ``inttoptr (CST to TYPE)``
2690 Convert an integer constant to a pointer constant. TYPE must be a
2691 pointer type. CST must be of integer type. The CST value is zero
2692 extended, truncated, or unchanged to make it fit in a pointer size.
2693 This one is *really* dangerous!
2694 ``bitcast (CST to TYPE)``
2695 Convert a constant, CST, to another TYPE. The constraints of the
2696 operands are the same as those for the :ref:`bitcast
2697 instruction <i_bitcast>`.
2698 ``addrspacecast (CST to TYPE)``
2699 Convert a constant pointer or constant vector of pointer, CST, to another
2700 TYPE in a different address space. The constraints of the operands are the
2701 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2702 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2703 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2704 constants. As with the :ref:`getelementptr <i_getelementptr>`
2705 instruction, the index list may have zero or more indexes, which are
2706 required to make sense for the type of "CSTPTR".
2707 ``select (COND, VAL1, VAL2)``
2708 Perform the :ref:`select operation <i_select>` on constants.
2709 ``icmp COND (VAL1, VAL2)``
2710 Performs the :ref:`icmp operation <i_icmp>` on constants.
2711 ``fcmp COND (VAL1, VAL2)``
2712 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2713 ``extractelement (VAL, IDX)``
2714 Perform the :ref:`extractelement operation <i_extractelement>` on
2716 ``insertelement (VAL, ELT, IDX)``
2717 Perform the :ref:`insertelement operation <i_insertelement>` on
2719 ``shufflevector (VEC1, VEC2, IDXMASK)``
2720 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2722 ``extractvalue (VAL, IDX0, IDX1, ...)``
2723 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2724 constants. The index list is interpreted in a similar manner as
2725 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2726 least one index value must be specified.
2727 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2728 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2729 The index list is interpreted in a similar manner as indices in a
2730 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2731 value must be specified.
2732 ``OPCODE (LHS, RHS)``
2733 Perform the specified operation of the LHS and RHS constants. OPCODE
2734 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2735 binary <bitwiseops>` operations. The constraints on operands are
2736 the same as those for the corresponding instruction (e.g. no bitwise
2737 operations on floating point values are allowed).
2744 Inline Assembler Expressions
2745 ----------------------------
2747 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2748 Inline Assembly <moduleasm>`) through the use of a special value. This
2749 value represents the inline assembler as a string (containing the
2750 instructions to emit), a list of operand constraints (stored as a
2751 string), a flag that indicates whether or not the inline asm expression
2752 has side effects, and a flag indicating whether the function containing
2753 the asm needs to align its stack conservatively. An example inline
2754 assembler expression is:
2756 .. code-block:: llvm
2758 i32 (i32) asm "bswap $0", "=r,r"
2760 Inline assembler expressions may **only** be used as the callee operand
2761 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2762 Thus, typically we have:
2764 .. code-block:: llvm
2766 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2768 Inline asms with side effects not visible in the constraint list must be
2769 marked as having side effects. This is done through the use of the
2770 '``sideeffect``' keyword, like so:
2772 .. code-block:: llvm
2774 call void asm sideeffect "eieio", ""()
2776 In some cases inline asms will contain code that will not work unless
2777 the stack is aligned in some way, such as calls or SSE instructions on
2778 x86, yet will not contain code that does that alignment within the asm.
2779 The compiler should make conservative assumptions about what the asm
2780 might contain and should generate its usual stack alignment code in the
2781 prologue if the '``alignstack``' keyword is present:
2783 .. code-block:: llvm
2785 call void asm alignstack "eieio", ""()
2787 Inline asms also support using non-standard assembly dialects. The
2788 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2789 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2790 the only supported dialects. An example is:
2792 .. code-block:: llvm
2794 call void asm inteldialect "eieio", ""()
2796 If multiple keywords appear the '``sideeffect``' keyword must come
2797 first, the '``alignstack``' keyword second and the '``inteldialect``'
2803 The call instructions that wrap inline asm nodes may have a
2804 "``!srcloc``" MDNode attached to it that contains a list of constant
2805 integers. If present, the code generator will use the integer as the
2806 location cookie value when report errors through the ``LLVMContext``
2807 error reporting mechanisms. This allows a front-end to correlate backend
2808 errors that occur with inline asm back to the source code that produced
2811 .. code-block:: llvm
2813 call void asm sideeffect "something bad", ""(), !srcloc !42
2815 !42 = !{ i32 1234567 }
2817 It is up to the front-end to make sense of the magic numbers it places
2818 in the IR. If the MDNode contains multiple constants, the code generator
2819 will use the one that corresponds to the line of the asm that the error
2827 LLVM IR allows metadata to be attached to instructions in the program
2828 that can convey extra information about the code to the optimizers and
2829 code generator. One example application of metadata is source-level
2830 debug information. There are two metadata primitives: strings and nodes.
2832 Metadata does not have a type, and is not a value. If referenced from a
2833 ``call`` instruction, it uses the ``metadata`` type.
2835 All metadata are identified in syntax by a exclamation point ('``!``').
2837 Metadata Nodes and Metadata Strings
2838 -----------------------------------
2840 A metadata string is a string surrounded by double quotes. It can
2841 contain any character by escaping non-printable characters with
2842 "``\xx``" where "``xx``" is the two digit hex code. For example:
2845 Metadata nodes are represented with notation similar to structure
2846 constants (a comma separated list of elements, surrounded by braces and
2847 preceded by an exclamation point). Metadata nodes can have any values as
2848 their operand. For example:
2850 .. code-block:: llvm
2852 !{ !"test\00", i32 10}
2854 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
2856 .. code-block:: llvm
2858 !0 = distinct !{!"test\00", i32 10}
2860 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
2861 content. They can also occur when transformations cause uniquing collisions
2862 when metadata operands change.
2864 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2865 metadata nodes, which can be looked up in the module symbol table. For
2868 .. code-block:: llvm
2872 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2873 function is using two metadata arguments:
2875 .. code-block:: llvm
2877 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2879 Metadata can be attached with an instruction. Here metadata ``!21`` is
2880 attached to the ``add`` instruction using the ``!dbg`` identifier:
2882 .. code-block:: llvm
2884 %indvar.next = add i64 %indvar, 1, !dbg !21
2886 More information about specific metadata nodes recognized by the
2887 optimizers and code generator is found below.
2889 Specialized Metadata Nodes
2890 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2892 Specialized metadata nodes are custom data structures in metadata (as opposed
2893 to generic tuples). Their fields are labelled, and can be specified in any
2899 ``MDLocation`` nodes represent source debug locations. The ``scope:`` field is
2902 .. code-block:: llvm
2904 !0 = !MDLocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
2909 In LLVM IR, memory does not have types, so LLVM's own type system is not
2910 suitable for doing TBAA. Instead, metadata is added to the IR to
2911 describe a type system of a higher level language. This can be used to
2912 implement typical C/C++ TBAA, but it can also be used to implement
2913 custom alias analysis behavior for other languages.
2915 The current metadata format is very simple. TBAA metadata nodes have up
2916 to three fields, e.g.:
2918 .. code-block:: llvm
2920 !0 = !{ !"an example type tree" }
2921 !1 = !{ !"int", !0 }
2922 !2 = !{ !"float", !0 }
2923 !3 = !{ !"const float", !2, i64 1 }
2925 The first field is an identity field. It can be any value, usually a
2926 metadata string, which uniquely identifies the type. The most important
2927 name in the tree is the name of the root node. Two trees with different
2928 root node names are entirely disjoint, even if they have leaves with
2931 The second field identifies the type's parent node in the tree, or is
2932 null or omitted for a root node. A type is considered to alias all of
2933 its descendants and all of its ancestors in the tree. Also, a type is
2934 considered to alias all types in other trees, so that bitcode produced
2935 from multiple front-ends is handled conservatively.
2937 If the third field is present, it's an integer which if equal to 1
2938 indicates that the type is "constant" (meaning
2939 ``pointsToConstantMemory`` should return true; see `other useful
2940 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2942 '``tbaa.struct``' Metadata
2943 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2945 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2946 aggregate assignment operations in C and similar languages, however it
2947 is defined to copy a contiguous region of memory, which is more than
2948 strictly necessary for aggregate types which contain holes due to
2949 padding. Also, it doesn't contain any TBAA information about the fields
2952 ``!tbaa.struct`` metadata can describe which memory subregions in a
2953 memcpy are padding and what the TBAA tags of the struct are.
2955 The current metadata format is very simple. ``!tbaa.struct`` metadata
2956 nodes are a list of operands which are in conceptual groups of three.
2957 For each group of three, the first operand gives the byte offset of a
2958 field in bytes, the second gives its size in bytes, and the third gives
2961 .. code-block:: llvm
2963 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
2965 This describes a struct with two fields. The first is at offset 0 bytes
2966 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2967 and has size 4 bytes and has tbaa tag !2.
2969 Note that the fields need not be contiguous. In this example, there is a
2970 4 byte gap between the two fields. This gap represents padding which
2971 does not carry useful data and need not be preserved.
2973 '``noalias``' and '``alias.scope``' Metadata
2974 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2976 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
2977 noalias memory-access sets. This means that some collection of memory access
2978 instructions (loads, stores, memory-accessing calls, etc.) that carry
2979 ``noalias`` metadata can specifically be specified not to alias with some other
2980 collection of memory access instructions that carry ``alias.scope`` metadata.
2981 Each type of metadata specifies a list of scopes where each scope has an id and
2982 a domain. When evaluating an aliasing query, if for some some domain, the set
2983 of scopes with that domain in one instruction's ``alias.scope`` list is a
2984 subset of (or qual to) the set of scopes for that domain in another
2985 instruction's ``noalias`` list, then the two memory accesses are assumed not to
2988 The metadata identifying each domain is itself a list containing one or two
2989 entries. The first entry is the name of the domain. Note that if the name is a
2990 string then it can be combined accross functions and translation units. A
2991 self-reference can be used to create globally unique domain names. A
2992 descriptive string may optionally be provided as a second list entry.
2994 The metadata identifying each scope is also itself a list containing two or
2995 three entries. The first entry is the name of the scope. Note that if the name
2996 is a string then it can be combined accross functions and translation units. A
2997 self-reference can be used to create globally unique scope names. A metadata
2998 reference to the scope's domain is the second entry. A descriptive string may
2999 optionally be provided as a third list entry.
3003 .. code-block:: llvm
3005 ; Two scope domains:
3009 ; Some scopes in these domains:
3015 !5 = !{!4} ; A list containing only scope !4
3019 ; These two instructions don't alias:
3020 %0 = load float* %c, align 4, !alias.scope !5
3021 store float %0, float* %arrayidx.i, align 4, !noalias !5
3023 ; These two instructions also don't alias (for domain !1, the set of scopes
3024 ; in the !alias.scope equals that in the !noalias list):
3025 %2 = load float* %c, align 4, !alias.scope !5
3026 store float %2, float* %arrayidx.i2, align 4, !noalias !6
3028 ; These two instructions don't alias (for domain !0, the set of scopes in
3029 ; the !noalias list is not a superset of, or equal to, the scopes in the
3030 ; !alias.scope list):
3031 %2 = load float* %c, align 4, !alias.scope !6
3032 store float %0, float* %arrayidx.i, align 4, !noalias !7
3034 '``fpmath``' Metadata
3035 ^^^^^^^^^^^^^^^^^^^^^
3037 ``fpmath`` metadata may be attached to any instruction of floating point
3038 type. It can be used to express the maximum acceptable error in the
3039 result of that instruction, in ULPs, thus potentially allowing the
3040 compiler to use a more efficient but less accurate method of computing
3041 it. ULP is defined as follows:
3043 If ``x`` is a real number that lies between two finite consecutive
3044 floating-point numbers ``a`` and ``b``, without being equal to one
3045 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
3046 distance between the two non-equal finite floating-point numbers
3047 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
3049 The metadata node shall consist of a single positive floating point
3050 number representing the maximum relative error, for example:
3052 .. code-block:: llvm
3054 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3056 '``range``' Metadata
3057 ^^^^^^^^^^^^^^^^^^^^
3059 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
3060 integer types. It expresses the possible ranges the loaded value or the value
3061 returned by the called function at this call site is in. The ranges are
3062 represented with a flattened list of integers. The loaded value or the value
3063 returned is known to be in the union of the ranges defined by each consecutive
3064 pair. Each pair has the following properties:
3066 - The type must match the type loaded by the instruction.
3067 - The pair ``a,b`` represents the range ``[a,b)``.
3068 - Both ``a`` and ``b`` are constants.
3069 - The range is allowed to wrap.
3070 - The range should not represent the full or empty set. That is,
3073 In addition, the pairs must be in signed order of the lower bound and
3074 they must be non-contiguous.
3078 .. code-block:: llvm
3080 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3081 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3082 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
3083 %d = invoke i8 @bar() to label %cont
3084 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
3086 !0 = !{ i8 0, i8 2 }
3087 !1 = !{ i8 255, i8 2 }
3088 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
3089 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
3094 It is sometimes useful to attach information to loop constructs. Currently,
3095 loop metadata is implemented as metadata attached to the branch instruction
3096 in the loop latch block. This type of metadata refer to a metadata node that is
3097 guaranteed to be separate for each loop. The loop identifier metadata is
3098 specified with the name ``llvm.loop``.
3100 The loop identifier metadata is implemented using a metadata that refers to
3101 itself to avoid merging it with any other identifier metadata, e.g.,
3102 during module linkage or function inlining. That is, each loop should refer
3103 to their own identification metadata even if they reside in separate functions.
3104 The following example contains loop identifier metadata for two separate loop
3107 .. code-block:: llvm
3112 The loop identifier metadata can be used to specify additional
3113 per-loop metadata. Any operands after the first operand can be treated
3114 as user-defined metadata. For example the ``llvm.loop.unroll.count``
3115 suggests an unroll factor to the loop unroller:
3117 .. code-block:: llvm
3119 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
3122 !1 = !{!"llvm.loop.unroll.count", i32 4}
3124 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
3125 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3127 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
3128 used to control per-loop vectorization and interleaving parameters such as
3129 vectorization width and interleave count. These metadata should be used in
3130 conjunction with ``llvm.loop`` loop identification metadata. The
3131 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
3132 optimization hints and the optimizer will only interleave and vectorize loops if
3133 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
3134 which contains information about loop-carried memory dependencies can be helpful
3135 in determining the safety of these transformations.
3137 '``llvm.loop.interleave.count``' Metadata
3138 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3140 This metadata suggests an interleave count to the loop interleaver.
3141 The first operand is the string ``llvm.loop.interleave.count`` and the
3142 second operand is an integer specifying the interleave count. For
3145 .. code-block:: llvm
3147 !0 = !{!"llvm.loop.interleave.count", i32 4}
3149 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
3150 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
3151 then the interleave count will be determined automatically.
3153 '``llvm.loop.vectorize.enable``' Metadata
3154 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3156 This metadata selectively enables or disables vectorization for the loop. The
3157 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
3158 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
3159 0 disables vectorization:
3161 .. code-block:: llvm
3163 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
3164 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
3166 '``llvm.loop.vectorize.width``' Metadata
3167 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3169 This metadata sets the target width of the vectorizer. The first
3170 operand is the string ``llvm.loop.vectorize.width`` and the second
3171 operand is an integer specifying the width. For example:
3173 .. code-block:: llvm
3175 !0 = !{!"llvm.loop.vectorize.width", i32 4}
3177 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
3178 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
3179 0 or if the loop does not have this metadata the width will be
3180 determined automatically.
3182 '``llvm.loop.unroll``'
3183 ^^^^^^^^^^^^^^^^^^^^^^
3185 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
3186 optimization hints such as the unroll factor. ``llvm.loop.unroll``
3187 metadata should be used in conjunction with ``llvm.loop`` loop
3188 identification metadata. The ``llvm.loop.unroll`` metadata are only
3189 optimization hints and the unrolling will only be performed if the
3190 optimizer believes it is safe to do so.
3192 '``llvm.loop.unroll.count``' Metadata
3193 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3195 This metadata suggests an unroll factor to the loop unroller. The
3196 first operand is the string ``llvm.loop.unroll.count`` and the second
3197 operand is a positive integer specifying the unroll factor. For
3200 .. code-block:: llvm
3202 !0 = !{!"llvm.loop.unroll.count", i32 4}
3204 If the trip count of the loop is less than the unroll count the loop
3205 will be partially unrolled.
3207 '``llvm.loop.unroll.disable``' Metadata
3208 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3210 This metadata either disables loop unrolling. The metadata has a single operand
3211 which is the string ``llvm.loop.unroll.disable``. For example:
3213 .. code-block:: llvm
3215 !0 = !{!"llvm.loop.unroll.disable"}
3217 '``llvm.loop.unroll.full``' Metadata
3218 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3220 This metadata either suggests that the loop should be unrolled fully. The
3221 metadata has a single operand which is the string ``llvm.loop.unroll.disable``.
3224 .. code-block:: llvm
3226 !0 = !{!"llvm.loop.unroll.full"}
3231 Metadata types used to annotate memory accesses with information helpful
3232 for optimizations are prefixed with ``llvm.mem``.
3234 '``llvm.mem.parallel_loop_access``' Metadata
3235 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3237 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
3238 or metadata containing a list of loop identifiers for nested loops.
3239 The metadata is attached to memory accessing instructions and denotes that
3240 no loop carried memory dependence exist between it and other instructions denoted
3241 with the same loop identifier.
3243 Precisely, given two instructions ``m1`` and ``m2`` that both have the
3244 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
3245 set of loops associated with that metadata, respectively, then there is no loop
3246 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
3249 As a special case, if all memory accessing instructions in a loop have
3250 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
3251 loop has no loop carried memory dependences and is considered to be a parallel
3254 Note that if not all memory access instructions have such metadata referring to
3255 the loop, then the loop is considered not being trivially parallel. Additional
3256 memory dependence analysis is required to make that determination. As a fail
3257 safe mechanism, this causes loops that were originally parallel to be considered
3258 sequential (if optimization passes that are unaware of the parallel semantics
3259 insert new memory instructions into the loop body).
3261 Example of a loop that is considered parallel due to its correct use of
3262 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
3263 metadata types that refer to the same loop identifier metadata.
3265 .. code-block:: llvm
3269 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
3271 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3273 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
3279 It is also possible to have nested parallel loops. In that case the
3280 memory accesses refer to a list of loop identifier metadata nodes instead of
3281 the loop identifier metadata node directly:
3283 .. code-block:: llvm
3287 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
3289 br label %inner.for.body
3293 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3295 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
3297 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
3301 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
3303 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
3305 outer.for.end: ; preds = %for.body
3307 !0 = !{!1, !2} ; a list of loop identifiers
3308 !1 = !{!1} ; an identifier for the inner loop
3309 !2 = !{!2} ; an identifier for the outer loop
3311 Module Flags Metadata
3312 =====================
3314 Information about the module as a whole is difficult to convey to LLVM's
3315 subsystems. The LLVM IR isn't sufficient to transmit this information.
3316 The ``llvm.module.flags`` named metadata exists in order to facilitate
3317 this. These flags are in the form of key / value pairs --- much like a
3318 dictionary --- making it easy for any subsystem who cares about a flag to
3321 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
3322 Each triplet has the following form:
3324 - The first element is a *behavior* flag, which specifies the behavior
3325 when two (or more) modules are merged together, and it encounters two
3326 (or more) metadata with the same ID. The supported behaviors are
3328 - The second element is a metadata string that is a unique ID for the
3329 metadata. Each module may only have one flag entry for each unique ID (not
3330 including entries with the **Require** behavior).
3331 - The third element is the value of the flag.
3333 When two (or more) modules are merged together, the resulting
3334 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
3335 each unique metadata ID string, there will be exactly one entry in the merged
3336 modules ``llvm.module.flags`` metadata table, and the value for that entry will
3337 be determined by the merge behavior flag, as described below. The only exception
3338 is that entries with the *Require* behavior are always preserved.
3340 The following behaviors are supported:
3351 Emits an error if two values disagree, otherwise the resulting value
3352 is that of the operands.
3356 Emits a warning if two values disagree. The result value will be the
3357 operand for the flag from the first module being linked.
3361 Adds a requirement that another module flag be present and have a
3362 specified value after linking is performed. The value must be a
3363 metadata pair, where the first element of the pair is the ID of the
3364 module flag to be restricted, and the second element of the pair is
3365 the value the module flag should be restricted to. This behavior can
3366 be used to restrict the allowable results (via triggering of an
3367 error) of linking IDs with the **Override** behavior.
3371 Uses the specified value, regardless of the behavior or value of the
3372 other module. If both modules specify **Override**, but the values
3373 differ, an error will be emitted.
3377 Appends the two values, which are required to be metadata nodes.
3381 Appends the two values, which are required to be metadata
3382 nodes. However, duplicate entries in the second list are dropped
3383 during the append operation.
3385 It is an error for a particular unique flag ID to have multiple behaviors,
3386 except in the case of **Require** (which adds restrictions on another metadata
3387 value) or **Override**.
3389 An example of module flags:
3391 .. code-block:: llvm
3393 !0 = !{ i32 1, !"foo", i32 1 }
3394 !1 = !{ i32 4, !"bar", i32 37 }
3395 !2 = !{ i32 2, !"qux", i32 42 }
3396 !3 = !{ i32 3, !"qux",
3401 !llvm.module.flags = !{ !0, !1, !2, !3 }
3403 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3404 if two or more ``!"foo"`` flags are seen is to emit an error if their
3405 values are not equal.
3407 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3408 behavior if two or more ``!"bar"`` flags are seen is to use the value
3411 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3412 behavior if two or more ``!"qux"`` flags are seen is to emit a
3413 warning if their values are not equal.
3415 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3421 The behavior is to emit an error if the ``llvm.module.flags`` does not
3422 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3425 Objective-C Garbage Collection Module Flags Metadata
3426 ----------------------------------------------------
3428 On the Mach-O platform, Objective-C stores metadata about garbage
3429 collection in a special section called "image info". The metadata
3430 consists of a version number and a bitmask specifying what types of
3431 garbage collection are supported (if any) by the file. If two or more
3432 modules are linked together their garbage collection metadata needs to
3433 be merged rather than appended together.
3435 The Objective-C garbage collection module flags metadata consists of the
3436 following key-value pairs:
3445 * - ``Objective-C Version``
3446 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3448 * - ``Objective-C Image Info Version``
3449 - **[Required]** --- The version of the image info section. Currently
3452 * - ``Objective-C Image Info Section``
3453 - **[Required]** --- The section to place the metadata. Valid values are
3454 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3455 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3456 Objective-C ABI version 2.
3458 * - ``Objective-C Garbage Collection``
3459 - **[Required]** --- Specifies whether garbage collection is supported or
3460 not. Valid values are 0, for no garbage collection, and 2, for garbage
3461 collection supported.
3463 * - ``Objective-C GC Only``
3464 - **[Optional]** --- Specifies that only garbage collection is supported.
3465 If present, its value must be 6. This flag requires that the
3466 ``Objective-C Garbage Collection`` flag have the value 2.
3468 Some important flag interactions:
3470 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3471 merged with a module with ``Objective-C Garbage Collection`` set to
3472 2, then the resulting module has the
3473 ``Objective-C Garbage Collection`` flag set to 0.
3474 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3475 merged with a module with ``Objective-C GC Only`` set to 6.
3477 Automatic Linker Flags Module Flags Metadata
3478 --------------------------------------------
3480 Some targets support embedding flags to the linker inside individual object
3481 files. Typically this is used in conjunction with language extensions which
3482 allow source files to explicitly declare the libraries they depend on, and have
3483 these automatically be transmitted to the linker via object files.
3485 These flags are encoded in the IR using metadata in the module flags section,
3486 using the ``Linker Options`` key. The merge behavior for this flag is required
3487 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3488 node which should be a list of other metadata nodes, each of which should be a
3489 list of metadata strings defining linker options.
3491 For example, the following metadata section specifies two separate sets of
3492 linker options, presumably to link against ``libz`` and the ``Cocoa``
3495 !0 = !{ i32 6, !"Linker Options",
3498 !{ !"-framework", !"Cocoa" } } }
3499 !llvm.module.flags = !{ !0 }
3501 The metadata encoding as lists of lists of options, as opposed to a collapsed
3502 list of options, is chosen so that the IR encoding can use multiple option
3503 strings to specify e.g., a single library, while still having that specifier be
3504 preserved as an atomic element that can be recognized by a target specific
3505 assembly writer or object file emitter.
3507 Each individual option is required to be either a valid option for the target's
3508 linker, or an option that is reserved by the target specific assembly writer or
3509 object file emitter. No other aspect of these options is defined by the IR.
3511 C type width Module Flags Metadata
3512 ----------------------------------
3514 The ARM backend emits a section into each generated object file describing the
3515 options that it was compiled with (in a compiler-independent way) to prevent
3516 linking incompatible objects, and to allow automatic library selection. Some
3517 of these options are not visible at the IR level, namely wchar_t width and enum
3520 To pass this information to the backend, these options are encoded in module
3521 flags metadata, using the following key-value pairs:
3531 - * 0 --- sizeof(wchar_t) == 4
3532 * 1 --- sizeof(wchar_t) == 2
3535 - * 0 --- Enums are at least as large as an ``int``.
3536 * 1 --- Enums are stored in the smallest integer type which can
3537 represent all of its values.
3539 For example, the following metadata section specifies that the module was
3540 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
3541 enum is the smallest type which can represent all of its values::
3543 !llvm.module.flags = !{!0, !1}
3544 !0 = !{i32 1, !"short_wchar", i32 1}
3545 !1 = !{i32 1, !"short_enum", i32 0}
3547 .. _intrinsicglobalvariables:
3549 Intrinsic Global Variables
3550 ==========================
3552 LLVM has a number of "magic" global variables that contain data that
3553 affect code generation or other IR semantics. These are documented here.
3554 All globals of this sort should have a section specified as
3555 "``llvm.metadata``". This section and all globals that start with
3556 "``llvm.``" are reserved for use by LLVM.
3560 The '``llvm.used``' Global Variable
3561 -----------------------------------
3563 The ``@llvm.used`` global is an array which has
3564 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3565 pointers to named global variables, functions and aliases which may optionally
3566 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3569 .. code-block:: llvm
3574 @llvm.used = appending global [2 x i8*] [
3576 i8* bitcast (i32* @Y to i8*)
3577 ], section "llvm.metadata"
3579 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3580 and linker are required to treat the symbol as if there is a reference to the
3581 symbol that it cannot see (which is why they have to be named). For example, if
3582 a variable has internal linkage and no references other than that from the
3583 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3584 references from inline asms and other things the compiler cannot "see", and
3585 corresponds to "``attribute((used))``" in GNU C.
3587 On some targets, the code generator must emit a directive to the
3588 assembler or object file to prevent the assembler and linker from
3589 molesting the symbol.
3591 .. _gv_llvmcompilerused:
3593 The '``llvm.compiler.used``' Global Variable
3594 --------------------------------------------
3596 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3597 directive, except that it only prevents the compiler from touching the
3598 symbol. On targets that support it, this allows an intelligent linker to
3599 optimize references to the symbol without being impeded as it would be
3602 This is a rare construct that should only be used in rare circumstances,
3603 and should not be exposed to source languages.
3605 .. _gv_llvmglobalctors:
3607 The '``llvm.global_ctors``' Global Variable
3608 -------------------------------------------
3610 .. code-block:: llvm
3612 %0 = type { i32, void ()*, i8* }
3613 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
3615 The ``@llvm.global_ctors`` array contains a list of constructor
3616 functions, priorities, and an optional associated global or function.
3617 The functions referenced by this array will be called in ascending order
3618 of priority (i.e. lowest first) when the module is loaded. The order of
3619 functions with the same priority is not defined.
3621 If the third field is present, non-null, and points to a global variable
3622 or function, the initializer function will only run if the associated
3623 data from the current module is not discarded.
3625 .. _llvmglobaldtors:
3627 The '``llvm.global_dtors``' Global Variable
3628 -------------------------------------------
3630 .. code-block:: llvm
3632 %0 = type { i32, void ()*, i8* }
3633 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
3635 The ``@llvm.global_dtors`` array contains a list of destructor
3636 functions, priorities, and an optional associated global or function.
3637 The functions referenced by this array will be called in descending
3638 order of priority (i.e. highest first) when the module is unloaded. The
3639 order of functions with the same priority is not defined.
3641 If the third field is present, non-null, and points to a global variable
3642 or function, the destructor function will only run if the associated
3643 data from the current module is not discarded.
3645 Instruction Reference
3646 =====================
3648 The LLVM instruction set consists of several different classifications
3649 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3650 instructions <binaryops>`, :ref:`bitwise binary
3651 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3652 :ref:`other instructions <otherops>`.
3656 Terminator Instructions
3657 -----------------------
3659 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3660 program ends with a "Terminator" instruction, which indicates which
3661 block should be executed after the current block is finished. These
3662 terminator instructions typically yield a '``void``' value: they produce
3663 control flow, not values (the one exception being the
3664 ':ref:`invoke <i_invoke>`' instruction).
3666 The terminator instructions are: ':ref:`ret <i_ret>`',
3667 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3668 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3669 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3673 '``ret``' Instruction
3674 ^^^^^^^^^^^^^^^^^^^^^
3681 ret <type> <value> ; Return a value from a non-void function
3682 ret void ; Return from void function
3687 The '``ret``' instruction is used to return control flow (and optionally
3688 a value) from a function back to the caller.
3690 There are two forms of the '``ret``' instruction: one that returns a
3691 value and then causes control flow, and one that just causes control
3697 The '``ret``' instruction optionally accepts a single argument, the
3698 return value. The type of the return value must be a ':ref:`first
3699 class <t_firstclass>`' type.
3701 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3702 return type and contains a '``ret``' instruction with no return value or
3703 a return value with a type that does not match its type, or if it has a
3704 void return type and contains a '``ret``' instruction with a return
3710 When the '``ret``' instruction is executed, control flow returns back to
3711 the calling function's context. If the caller is a
3712 ":ref:`call <i_call>`" instruction, execution continues at the
3713 instruction after the call. If the caller was an
3714 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3715 beginning of the "normal" destination block. If the instruction returns
3716 a value, that value shall set the call or invoke instruction's return
3722 .. code-block:: llvm
3724 ret i32 5 ; Return an integer value of 5
3725 ret void ; Return from a void function
3726 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3730 '``br``' Instruction
3731 ^^^^^^^^^^^^^^^^^^^^
3738 br i1 <cond>, label <iftrue>, label <iffalse>
3739 br label <dest> ; Unconditional branch
3744 The '``br``' instruction is used to cause control flow to transfer to a
3745 different basic block in the current function. There are two forms of
3746 this instruction, corresponding to a conditional branch and an
3747 unconditional branch.
3752 The conditional branch form of the '``br``' instruction takes a single
3753 '``i1``' value and two '``label``' values. The unconditional form of the
3754 '``br``' instruction takes a single '``label``' value as a target.
3759 Upon execution of a conditional '``br``' instruction, the '``i1``'
3760 argument is evaluated. If the value is ``true``, control flows to the
3761 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3762 to the '``iffalse``' ``label`` argument.
3767 .. code-block:: llvm
3770 %cond = icmp eq i32 %a, %b
3771 br i1 %cond, label %IfEqual, label %IfUnequal
3779 '``switch``' Instruction
3780 ^^^^^^^^^^^^^^^^^^^^^^^^
3787 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3792 The '``switch``' instruction is used to transfer control flow to one of
3793 several different places. It is a generalization of the '``br``'
3794 instruction, allowing a branch to occur to one of many possible
3800 The '``switch``' instruction uses three parameters: an integer
3801 comparison value '``value``', a default '``label``' destination, and an
3802 array of pairs of comparison value constants and '``label``'s. The table
3803 is not allowed to contain duplicate constant entries.
3808 The ``switch`` instruction specifies a table of values and destinations.
3809 When the '``switch``' instruction is executed, this table is searched
3810 for the given value. If the value is found, control flow is transferred
3811 to the corresponding destination; otherwise, control flow is transferred
3812 to the default destination.
3817 Depending on properties of the target machine and the particular
3818 ``switch`` instruction, this instruction may be code generated in
3819 different ways. For example, it could be generated as a series of
3820 chained conditional branches or with a lookup table.
3825 .. code-block:: llvm
3827 ; Emulate a conditional br instruction
3828 %Val = zext i1 %value to i32
3829 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3831 ; Emulate an unconditional br instruction
3832 switch i32 0, label %dest [ ]
3834 ; Implement a jump table:
3835 switch i32 %val, label %otherwise [ i32 0, label %onzero
3837 i32 2, label %ontwo ]
3841 '``indirectbr``' Instruction
3842 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3849 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3854 The '``indirectbr``' instruction implements an indirect branch to a
3855 label within the current function, whose address is specified by
3856 "``address``". Address must be derived from a
3857 :ref:`blockaddress <blockaddress>` constant.
3862 The '``address``' argument is the address of the label to jump to. The
3863 rest of the arguments indicate the full set of possible destinations
3864 that the address may point to. Blocks are allowed to occur multiple
3865 times in the destination list, though this isn't particularly useful.
3867 This destination list is required so that dataflow analysis has an
3868 accurate understanding of the CFG.
3873 Control transfers to the block specified in the address argument. All
3874 possible destination blocks must be listed in the label list, otherwise
3875 this instruction has undefined behavior. This implies that jumps to
3876 labels defined in other functions have undefined behavior as well.
3881 This is typically implemented with a jump through a register.
3886 .. code-block:: llvm
3888 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3892 '``invoke``' Instruction
3893 ^^^^^^^^^^^^^^^^^^^^^^^^
3900 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3901 to label <normal label> unwind label <exception label>
3906 The '``invoke``' instruction causes control to transfer to a specified
3907 function, with the possibility of control flow transfer to either the
3908 '``normal``' label or the '``exception``' label. If the callee function
3909 returns with the "``ret``" instruction, control flow will return to the
3910 "normal" label. If the callee (or any indirect callees) returns via the
3911 ":ref:`resume <i_resume>`" instruction or other exception handling
3912 mechanism, control is interrupted and continued at the dynamically
3913 nearest "exception" label.
3915 The '``exception``' label is a `landing
3916 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3917 '``exception``' label is required to have the
3918 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3919 information about the behavior of the program after unwinding happens,
3920 as its first non-PHI instruction. The restrictions on the
3921 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3922 instruction, so that the important information contained within the
3923 "``landingpad``" instruction can't be lost through normal code motion.
3928 This instruction requires several arguments:
3930 #. The optional "cconv" marker indicates which :ref:`calling
3931 convention <callingconv>` the call should use. If none is
3932 specified, the call defaults to using C calling conventions.
3933 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3934 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3936 #. '``ptr to function ty``': shall be the signature of the pointer to
3937 function value being invoked. In most cases, this is a direct
3938 function invocation, but indirect ``invoke``'s are just as possible,
3939 branching off an arbitrary pointer to function value.
3940 #. '``function ptr val``': An LLVM value containing a pointer to a
3941 function to be invoked.
3942 #. '``function args``': argument list whose types match the function
3943 signature argument types and parameter attributes. All arguments must
3944 be of :ref:`first class <t_firstclass>` type. If the function signature
3945 indicates the function accepts a variable number of arguments, the
3946 extra arguments can be specified.
3947 #. '``normal label``': the label reached when the called function
3948 executes a '``ret``' instruction.
3949 #. '``exception label``': the label reached when a callee returns via
3950 the :ref:`resume <i_resume>` instruction or other exception handling
3952 #. The optional :ref:`function attributes <fnattrs>` list. Only
3953 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3954 attributes are valid here.
3959 This instruction is designed to operate as a standard '``call``'
3960 instruction in most regards. The primary difference is that it
3961 establishes an association with a label, which is used by the runtime
3962 library to unwind the stack.
3964 This instruction is used in languages with destructors to ensure that
3965 proper cleanup is performed in the case of either a ``longjmp`` or a
3966 thrown exception. Additionally, this is important for implementation of
3967 '``catch``' clauses in high-level languages that support them.
3969 For the purposes of the SSA form, the definition of the value returned
3970 by the '``invoke``' instruction is deemed to occur on the edge from the
3971 current block to the "normal" label. If the callee unwinds then no
3972 return value is available.
3977 .. code-block:: llvm
3979 %retval = invoke i32 @Test(i32 15) to label %Continue
3980 unwind label %TestCleanup ; i32:retval set
3981 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3982 unwind label %TestCleanup ; i32:retval set
3986 '``resume``' Instruction
3987 ^^^^^^^^^^^^^^^^^^^^^^^^
3994 resume <type> <value>
3999 The '``resume``' instruction is a terminator instruction that has no
4005 The '``resume``' instruction requires one argument, which must have the
4006 same type as the result of any '``landingpad``' instruction in the same
4012 The '``resume``' instruction resumes propagation of an existing
4013 (in-flight) exception whose unwinding was interrupted with a
4014 :ref:`landingpad <i_landingpad>` instruction.
4019 .. code-block:: llvm
4021 resume { i8*, i32 } %exn
4025 '``unreachable``' Instruction
4026 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4038 The '``unreachable``' instruction has no defined semantics. This
4039 instruction is used to inform the optimizer that a particular portion of
4040 the code is not reachable. This can be used to indicate that the code
4041 after a no-return function cannot be reached, and other facts.
4046 The '``unreachable``' instruction has no defined semantics.
4053 Binary operators are used to do most of the computation in a program.
4054 They require two operands of the same type, execute an operation on
4055 them, and produce a single value. The operands might represent multiple
4056 data, as is the case with the :ref:`vector <t_vector>` data type. The
4057 result value has the same type as its operands.
4059 There are several different binary operators:
4063 '``add``' Instruction
4064 ^^^^^^^^^^^^^^^^^^^^^
4071 <result> = add <ty> <op1>, <op2> ; yields ty:result
4072 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
4073 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
4074 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
4079 The '``add``' instruction returns the sum of its two operands.
4084 The two arguments to the '``add``' instruction must be
4085 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4086 arguments must have identical types.
4091 The value produced is the integer sum of the two operands.
4093 If the sum has unsigned overflow, the result returned is the
4094 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4097 Because LLVM integers use a two's complement representation, this
4098 instruction is appropriate for both signed and unsigned integers.
4100 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4101 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4102 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
4103 unsigned and/or signed overflow, respectively, occurs.
4108 .. code-block:: llvm
4110 <result> = add i32 4, %var ; yields i32:result = 4 + %var
4114 '``fadd``' Instruction
4115 ^^^^^^^^^^^^^^^^^^^^^^
4122 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4127 The '``fadd``' instruction returns the sum of its two operands.
4132 The two arguments to the '``fadd``' instruction must be :ref:`floating
4133 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4134 Both arguments must have identical types.
4139 The value produced is the floating point sum of the two operands. This
4140 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
4141 which are optimization hints to enable otherwise unsafe floating point
4147 .. code-block:: llvm
4149 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
4151 '``sub``' Instruction
4152 ^^^^^^^^^^^^^^^^^^^^^
4159 <result> = sub <ty> <op1>, <op2> ; yields ty:result
4160 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
4161 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
4162 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
4167 The '``sub``' instruction returns the difference of its two operands.
4169 Note that the '``sub``' instruction is used to represent the '``neg``'
4170 instruction present in most other intermediate representations.
4175 The two arguments to the '``sub``' instruction must be
4176 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4177 arguments must have identical types.
4182 The value produced is the integer difference of the two operands.
4184 If the difference has unsigned overflow, the result returned is the
4185 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4188 Because LLVM integers use a two's complement representation, this
4189 instruction is appropriate for both signed and unsigned integers.
4191 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4192 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4193 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
4194 unsigned and/or signed overflow, respectively, occurs.
4199 .. code-block:: llvm
4201 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
4202 <result> = sub i32 0, %val ; yields i32:result = -%var
4206 '``fsub``' Instruction
4207 ^^^^^^^^^^^^^^^^^^^^^^
4214 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4219 The '``fsub``' instruction returns the difference of its two operands.
4221 Note that the '``fsub``' instruction is used to represent the '``fneg``'
4222 instruction present in most other intermediate representations.
4227 The two arguments to the '``fsub``' instruction must be :ref:`floating
4228 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4229 Both arguments must have identical types.
4234 The value produced is the floating point difference of the two operands.
4235 This instruction can also take any number of :ref:`fast-math
4236 flags <fastmath>`, which are optimization hints to enable otherwise
4237 unsafe floating point optimizations:
4242 .. code-block:: llvm
4244 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
4245 <result> = fsub float -0.0, %val ; yields float:result = -%var
4247 '``mul``' Instruction
4248 ^^^^^^^^^^^^^^^^^^^^^
4255 <result> = mul <ty> <op1>, <op2> ; yields ty:result
4256 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
4257 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
4258 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
4263 The '``mul``' instruction returns the product of its two operands.
4268 The two arguments to the '``mul``' instruction must be
4269 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4270 arguments must have identical types.
4275 The value produced is the integer product of the two operands.
4277 If the result of the multiplication has unsigned overflow, the result
4278 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
4279 bit width of the result.
4281 Because LLVM integers use a two's complement representation, and the
4282 result is the same width as the operands, this instruction returns the
4283 correct result for both signed and unsigned integers. If a full product
4284 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
4285 sign-extended or zero-extended as appropriate to the width of the full
4288 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4289 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4290 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
4291 unsigned and/or signed overflow, respectively, occurs.
4296 .. code-block:: llvm
4298 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
4302 '``fmul``' Instruction
4303 ^^^^^^^^^^^^^^^^^^^^^^
4310 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4315 The '``fmul``' instruction returns the product of its two operands.
4320 The two arguments to the '``fmul``' instruction must be :ref:`floating
4321 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4322 Both arguments must have identical types.
4327 The value produced is the floating point product of the two operands.
4328 This instruction can also take any number of :ref:`fast-math
4329 flags <fastmath>`, which are optimization hints to enable otherwise
4330 unsafe floating point optimizations:
4335 .. code-block:: llvm
4337 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
4339 '``udiv``' Instruction
4340 ^^^^^^^^^^^^^^^^^^^^^^
4347 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
4348 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
4353 The '``udiv``' instruction returns the quotient of its two operands.
4358 The two arguments to the '``udiv``' instruction must be
4359 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4360 arguments must have identical types.
4365 The value produced is the unsigned integer quotient of the two operands.
4367 Note that unsigned integer division and signed integer division are
4368 distinct operations; for signed integer division, use '``sdiv``'.
4370 Division by zero leads to undefined behavior.
4372 If the ``exact`` keyword is present, the result value of the ``udiv`` is
4373 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
4374 such, "((a udiv exact b) mul b) == a").
4379 .. code-block:: llvm
4381 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
4383 '``sdiv``' Instruction
4384 ^^^^^^^^^^^^^^^^^^^^^^
4391 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
4392 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
4397 The '``sdiv``' instruction returns the quotient of its two operands.
4402 The two arguments to the '``sdiv``' instruction must be
4403 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4404 arguments must have identical types.
4409 The value produced is the signed integer quotient of the two operands
4410 rounded towards zero.
4412 Note that signed integer division and unsigned integer division are
4413 distinct operations; for unsigned integer division, use '``udiv``'.
4415 Division by zero leads to undefined behavior. Overflow also leads to
4416 undefined behavior; this is a rare case, but can occur, for example, by
4417 doing a 32-bit division of -2147483648 by -1.
4419 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4420 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4425 .. code-block:: llvm
4427 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
4431 '``fdiv``' Instruction
4432 ^^^^^^^^^^^^^^^^^^^^^^
4439 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4444 The '``fdiv``' instruction returns the quotient of its two operands.
4449 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4450 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4451 Both arguments must have identical types.
4456 The value produced is the floating point quotient of the two operands.
4457 This instruction can also take any number of :ref:`fast-math
4458 flags <fastmath>`, which are optimization hints to enable otherwise
4459 unsafe floating point optimizations:
4464 .. code-block:: llvm
4466 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
4468 '``urem``' Instruction
4469 ^^^^^^^^^^^^^^^^^^^^^^
4476 <result> = urem <ty> <op1>, <op2> ; yields ty:result
4481 The '``urem``' instruction returns the remainder from the unsigned
4482 division of its two arguments.
4487 The two arguments to the '``urem``' instruction must be
4488 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4489 arguments must have identical types.
4494 This instruction returns the unsigned integer *remainder* of a division.
4495 This instruction always performs an unsigned division to get the
4498 Note that unsigned integer remainder and signed integer remainder are
4499 distinct operations; for signed integer remainder, use '``srem``'.
4501 Taking the remainder of a division by zero leads to undefined behavior.
4506 .. code-block:: llvm
4508 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
4510 '``srem``' Instruction
4511 ^^^^^^^^^^^^^^^^^^^^^^
4518 <result> = srem <ty> <op1>, <op2> ; yields ty:result
4523 The '``srem``' instruction returns the remainder from the signed
4524 division of its two operands. This instruction can also take
4525 :ref:`vector <t_vector>` versions of the values in which case the elements
4531 The two arguments to the '``srem``' instruction must be
4532 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4533 arguments must have identical types.
4538 This instruction returns the *remainder* of a division (where the result
4539 is either zero or has the same sign as the dividend, ``op1``), not the
4540 *modulo* operator (where the result is either zero or has the same sign
4541 as the divisor, ``op2``) of a value. For more information about the
4542 difference, see `The Math
4543 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4544 table of how this is implemented in various languages, please see
4546 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4548 Note that signed integer remainder and unsigned integer remainder are
4549 distinct operations; for unsigned integer remainder, use '``urem``'.
4551 Taking the remainder of a division by zero leads to undefined behavior.
4552 Overflow also leads to undefined behavior; this is a rare case, but can
4553 occur, for example, by taking the remainder of a 32-bit division of
4554 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4555 rule lets srem be implemented using instructions that return both the
4556 result of the division and the remainder.)
4561 .. code-block:: llvm
4563 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
4567 '``frem``' Instruction
4568 ^^^^^^^^^^^^^^^^^^^^^^
4575 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4580 The '``frem``' instruction returns the remainder from the division of
4586 The two arguments to the '``frem``' instruction must be :ref:`floating
4587 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4588 Both arguments must have identical types.
4593 This instruction returns the *remainder* of a division. The remainder
4594 has the same sign as the dividend. This instruction can also take any
4595 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4596 to enable otherwise unsafe floating point optimizations:
4601 .. code-block:: llvm
4603 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
4607 Bitwise Binary Operations
4608 -------------------------
4610 Bitwise binary operators are used to do various forms of bit-twiddling
4611 in a program. They are generally very efficient instructions and can
4612 commonly be strength reduced from other instructions. They require two
4613 operands of the same type, execute an operation on them, and produce a
4614 single value. The resulting value is the same type as its operands.
4616 '``shl``' Instruction
4617 ^^^^^^^^^^^^^^^^^^^^^
4624 <result> = shl <ty> <op1>, <op2> ; yields ty:result
4625 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
4626 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
4627 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
4632 The '``shl``' instruction returns the first operand shifted to the left
4633 a specified number of bits.
4638 Both arguments to the '``shl``' instruction must be the same
4639 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4640 '``op2``' is treated as an unsigned value.
4645 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4646 where ``n`` is the width of the result. If ``op2`` is (statically or
4647 dynamically) negative or equal to or larger than the number of bits in
4648 ``op1``, the result is undefined. If the arguments are vectors, each
4649 vector element of ``op1`` is shifted by the corresponding shift amount
4652 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4653 value <poisonvalues>` if it shifts out any non-zero bits. If the
4654 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4655 value <poisonvalues>` if it shifts out any bits that disagree with the
4656 resultant sign bit. As such, NUW/NSW have the same semantics as they
4657 would if the shift were expressed as a mul instruction with the same
4658 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4663 .. code-block:: llvm
4665 <result> = shl i32 4, %var ; yields i32: 4 << %var
4666 <result> = shl i32 4, 2 ; yields i32: 16
4667 <result> = shl i32 1, 10 ; yields i32: 1024
4668 <result> = shl i32 1, 32 ; undefined
4669 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4671 '``lshr``' Instruction
4672 ^^^^^^^^^^^^^^^^^^^^^^
4679 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
4680 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
4685 The '``lshr``' instruction (logical shift right) returns the first
4686 operand shifted to the right a specified number of bits with zero fill.
4691 Both arguments to the '``lshr``' instruction must be the same
4692 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4693 '``op2``' is treated as an unsigned value.
4698 This instruction always performs a logical shift right operation. The
4699 most significant bits of the result will be filled with zero bits after
4700 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4701 than the number of bits in ``op1``, the result is undefined. If the
4702 arguments are vectors, each vector element of ``op1`` is shifted by the
4703 corresponding shift amount in ``op2``.
4705 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4706 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4712 .. code-block:: llvm
4714 <result> = lshr i32 4, 1 ; yields i32:result = 2
4715 <result> = lshr i32 4, 2 ; yields i32:result = 1
4716 <result> = lshr i8 4, 3 ; yields i8:result = 0
4717 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
4718 <result> = lshr i32 1, 32 ; undefined
4719 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4721 '``ashr``' Instruction
4722 ^^^^^^^^^^^^^^^^^^^^^^
4729 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
4730 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
4735 The '``ashr``' instruction (arithmetic shift right) returns the first
4736 operand shifted to the right a specified number of bits with sign
4742 Both arguments to the '``ashr``' instruction must be the same
4743 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4744 '``op2``' is treated as an unsigned value.
4749 This instruction always performs an arithmetic shift right operation,
4750 The most significant bits of the result will be filled with the sign bit
4751 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4752 than the number of bits in ``op1``, the result is undefined. If the
4753 arguments are vectors, each vector element of ``op1`` is shifted by the
4754 corresponding shift amount in ``op2``.
4756 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4757 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4763 .. code-block:: llvm
4765 <result> = ashr i32 4, 1 ; yields i32:result = 2
4766 <result> = ashr i32 4, 2 ; yields i32:result = 1
4767 <result> = ashr i8 4, 3 ; yields i8:result = 0
4768 <result> = ashr i8 -2, 1 ; yields i8:result = -1
4769 <result> = ashr i32 1, 32 ; undefined
4770 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4772 '``and``' Instruction
4773 ^^^^^^^^^^^^^^^^^^^^^
4780 <result> = and <ty> <op1>, <op2> ; yields ty:result
4785 The '``and``' instruction returns the bitwise logical and of its two
4791 The two arguments to the '``and``' instruction must be
4792 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4793 arguments must have identical types.
4798 The truth table used for the '``and``' instruction is:
4815 .. code-block:: llvm
4817 <result> = and i32 4, %var ; yields i32:result = 4 & %var
4818 <result> = and i32 15, 40 ; yields i32:result = 8
4819 <result> = and i32 4, 8 ; yields i32:result = 0
4821 '``or``' Instruction
4822 ^^^^^^^^^^^^^^^^^^^^
4829 <result> = or <ty> <op1>, <op2> ; yields ty:result
4834 The '``or``' instruction returns the bitwise logical inclusive or of its
4840 The two arguments to the '``or``' instruction must be
4841 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4842 arguments must have identical types.
4847 The truth table used for the '``or``' instruction is:
4866 <result> = or i32 4, %var ; yields i32:result = 4 | %var
4867 <result> = or i32 15, 40 ; yields i32:result = 47
4868 <result> = or i32 4, 8 ; yields i32:result = 12
4870 '``xor``' Instruction
4871 ^^^^^^^^^^^^^^^^^^^^^
4878 <result> = xor <ty> <op1>, <op2> ; yields ty:result
4883 The '``xor``' instruction returns the bitwise logical exclusive or of
4884 its two operands. The ``xor`` is used to implement the "one's
4885 complement" operation, which is the "~" operator in C.
4890 The two arguments to the '``xor``' instruction must be
4891 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4892 arguments must have identical types.
4897 The truth table used for the '``xor``' instruction is:
4914 .. code-block:: llvm
4916 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
4917 <result> = xor i32 15, 40 ; yields i32:result = 39
4918 <result> = xor i32 4, 8 ; yields i32:result = 12
4919 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
4924 LLVM supports several instructions to represent vector operations in a
4925 target-independent manner. These instructions cover the element-access
4926 and vector-specific operations needed to process vectors effectively.
4927 While LLVM does directly support these vector operations, many
4928 sophisticated algorithms will want to use target-specific intrinsics to
4929 take full advantage of a specific target.
4931 .. _i_extractelement:
4933 '``extractelement``' Instruction
4934 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4941 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
4946 The '``extractelement``' instruction extracts a single scalar element
4947 from a vector at a specified index.
4952 The first operand of an '``extractelement``' instruction is a value of
4953 :ref:`vector <t_vector>` type. The second operand is an index indicating
4954 the position from which to extract the element. The index may be a
4955 variable of any integer type.
4960 The result is a scalar of the same type as the element type of ``val``.
4961 Its value is the value at position ``idx`` of ``val``. If ``idx``
4962 exceeds the length of ``val``, the results are undefined.
4967 .. code-block:: llvm
4969 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4971 .. _i_insertelement:
4973 '``insertelement``' Instruction
4974 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4981 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
4986 The '``insertelement``' instruction inserts a scalar element into a
4987 vector at a specified index.
4992 The first operand of an '``insertelement``' instruction is a value of
4993 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4994 type must equal the element type of the first operand. The third operand
4995 is an index indicating the position at which to insert the value. The
4996 index may be a variable of any integer type.
5001 The result is a vector of the same type as ``val``. Its element values
5002 are those of ``val`` except at position ``idx``, where it gets the value
5003 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
5009 .. code-block:: llvm
5011 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
5013 .. _i_shufflevector:
5015 '``shufflevector``' Instruction
5016 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5023 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
5028 The '``shufflevector``' instruction constructs a permutation of elements
5029 from two input vectors, returning a vector with the same element type as
5030 the input and length that is the same as the shuffle mask.
5035 The first two operands of a '``shufflevector``' instruction are vectors
5036 with the same type. The third argument is a shuffle mask whose element
5037 type is always 'i32'. The result of the instruction is a vector whose
5038 length is the same as the shuffle mask and whose element type is the
5039 same as the element type of the first two operands.
5041 The shuffle mask operand is required to be a constant vector with either
5042 constant integer or undef values.
5047 The elements of the two input vectors are numbered from left to right
5048 across both of the vectors. The shuffle mask operand specifies, for each
5049 element of the result vector, which element of the two input vectors the
5050 result element gets. The element selector may be undef (meaning "don't
5051 care") and the second operand may be undef if performing a shuffle from
5057 .. code-block:: llvm
5059 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5060 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
5061 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
5062 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
5063 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
5064 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
5065 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5066 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
5068 Aggregate Operations
5069 --------------------
5071 LLVM supports several instructions for working with
5072 :ref:`aggregate <t_aggregate>` values.
5076 '``extractvalue``' Instruction
5077 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5084 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
5089 The '``extractvalue``' instruction extracts the value of a member field
5090 from an :ref:`aggregate <t_aggregate>` value.
5095 The first operand of an '``extractvalue``' instruction is a value of
5096 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
5097 constant indices to specify which value to extract in a similar manner
5098 as indices in a '``getelementptr``' instruction.
5100 The major differences to ``getelementptr`` indexing are:
5102 - Since the value being indexed is not a pointer, the first index is
5103 omitted and assumed to be zero.
5104 - At least one index must be specified.
5105 - Not only struct indices but also array indices must be in bounds.
5110 The result is the value at the position in the aggregate specified by
5116 .. code-block:: llvm
5118 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
5122 '``insertvalue``' Instruction
5123 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5130 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
5135 The '``insertvalue``' instruction inserts a value into a member field in
5136 an :ref:`aggregate <t_aggregate>` value.
5141 The first operand of an '``insertvalue``' instruction is a value of
5142 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
5143 a first-class value to insert. The following operands are constant
5144 indices indicating the position at which to insert the value in a
5145 similar manner as indices in a '``extractvalue``' instruction. The value
5146 to insert must have the same type as the value identified by the
5152 The result is an aggregate of the same type as ``val``. Its value is
5153 that of ``val`` except that the value at the position specified by the
5154 indices is that of ``elt``.
5159 .. code-block:: llvm
5161 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
5162 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
5163 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
5167 Memory Access and Addressing Operations
5168 ---------------------------------------
5170 A key design point of an SSA-based representation is how it represents
5171 memory. In LLVM, no memory locations are in SSA form, which makes things
5172 very simple. This section describes how to read, write, and allocate
5177 '``alloca``' Instruction
5178 ^^^^^^^^^^^^^^^^^^^^^^^^
5185 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
5190 The '``alloca``' instruction allocates memory on the stack frame of the
5191 currently executing function, to be automatically released when this
5192 function returns to its caller. The object is always allocated in the
5193 generic address space (address space zero).
5198 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
5199 bytes of memory on the runtime stack, returning a pointer of the
5200 appropriate type to the program. If "NumElements" is specified, it is
5201 the number of elements allocated, otherwise "NumElements" is defaulted
5202 to be one. If a constant alignment is specified, the value result of the
5203 allocation is guaranteed to be aligned to at least that boundary. The
5204 alignment may not be greater than ``1 << 29``. If not specified, or if
5205 zero, the target can choose to align the allocation on any convenient
5206 boundary compatible with the type.
5208 '``type``' may be any sized type.
5213 Memory is allocated; a pointer is returned. The operation is undefined
5214 if there is insufficient stack space for the allocation. '``alloca``'d
5215 memory is automatically released when the function returns. The
5216 '``alloca``' instruction is commonly used to represent automatic
5217 variables that must have an address available. When the function returns
5218 (either with the ``ret`` or ``resume`` instructions), the memory is
5219 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5220 The order in which memory is allocated (ie., which way the stack grows)
5226 .. code-block:: llvm
5228 %ptr = alloca i32 ; yields i32*:ptr
5229 %ptr = alloca i32, i32 4 ; yields i32*:ptr
5230 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
5231 %ptr = alloca i32, align 1024 ; yields i32*:ptr
5235 '``load``' Instruction
5236 ^^^^^^^^^^^^^^^^^^^^^^
5243 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !nonnull !<index>]
5244 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5245 !<index> = !{ i32 1 }
5250 The '``load``' instruction is used to read from memory.
5255 The argument to the ``load`` instruction specifies the memory address
5256 from which to load. The pointer must point to a :ref:`first
5257 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
5258 then the optimizer is not allowed to modify the number or order of
5259 execution of this ``load`` with other :ref:`volatile
5260 operations <volatile>`.
5262 If the ``load`` is marked as ``atomic``, it takes an extra
5263 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5264 ``release`` and ``acq_rel`` orderings are not valid on ``load``
5265 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5266 when they may see multiple atomic stores. The type of the pointee must
5267 be an integer type whose bit width is a power of two greater than or
5268 equal to eight and less than or equal to a target-specific size limit.
5269 ``align`` must be explicitly specified on atomic loads, and the load has
5270 undefined behavior if the alignment is not set to a value which is at
5271 least the size in bytes of the pointee. ``!nontemporal`` does not have
5272 any defined semantics for atomic loads.
5274 The optional constant ``align`` argument specifies the alignment of the
5275 operation (that is, the alignment of the memory address). A value of 0
5276 or an omitted ``align`` argument means that the operation has the ABI
5277 alignment for the target. It is the responsibility of the code emitter
5278 to ensure that the alignment information is correct. Overestimating the
5279 alignment results in undefined behavior. Underestimating the alignment
5280 may produce less efficient code. An alignment of 1 is always safe. The
5281 maximum possible alignment is ``1 << 29``.
5283 The optional ``!nontemporal`` metadata must reference a single
5284 metadata name ``<index>`` corresponding to a metadata node with one
5285 ``i32`` entry of value 1. The existence of the ``!nontemporal``
5286 metadata on the instruction tells the optimizer and code generator
5287 that this load is not expected to be reused in the cache. The code
5288 generator may select special instructions to save cache bandwidth, such
5289 as the ``MOVNT`` instruction on x86.
5291 The optional ``!invariant.load`` metadata must reference a single
5292 metadata name ``<index>`` corresponding to a metadata node with no
5293 entries. The existence of the ``!invariant.load`` metadata on the
5294 instruction tells the optimizer and code generator that the address
5295 operand to this load points to memory which can be assumed unchanged.
5296 Being invariant does not imply that a location is dereferenceable,
5297 but it does imply that once the location is known dereferenceable
5298 its value is henceforth unchanging.
5300 The optional ``!nonnull`` metadata must reference a single
5301 metadata name ``<index>`` corresponding to a metadata node with no
5302 entries. The existence of the ``!nonnull`` metadata on the
5303 instruction tells the optimizer that the value loaded is known to
5304 never be null. This is analogous to the ''nonnull'' attribute
5305 on parameters and return values. This metadata can only be applied
5306 to loads of a pointer type.
5311 The location of memory pointed to is loaded. If the value being loaded
5312 is of scalar type then the number of bytes read does not exceed the
5313 minimum number of bytes needed to hold all bits of the type. For
5314 example, loading an ``i24`` reads at most three bytes. When loading a
5315 value of a type like ``i20`` with a size that is not an integral number
5316 of bytes, the result is undefined if the value was not originally
5317 written using a store of the same type.
5322 .. code-block:: llvm
5324 %ptr = alloca i32 ; yields i32*:ptr
5325 store i32 3, i32* %ptr ; yields void
5326 %val = load i32* %ptr ; yields i32:val = i32 3
5330 '``store``' Instruction
5331 ^^^^^^^^^^^^^^^^^^^^^^^
5338 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
5339 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
5344 The '``store``' instruction is used to write to memory.
5349 There are two arguments to the ``store`` instruction: a value to store
5350 and an address at which to store it. The type of the ``<pointer>``
5351 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
5352 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
5353 then the optimizer is not allowed to modify the number or order of
5354 execution of this ``store`` with other :ref:`volatile
5355 operations <volatile>`.
5357 If the ``store`` is marked as ``atomic``, it takes an extra
5358 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5359 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
5360 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5361 when they may see multiple atomic stores. The type of the pointee must
5362 be an integer type whose bit width is a power of two greater than or
5363 equal to eight and less than or equal to a target-specific size limit.
5364 ``align`` must be explicitly specified on atomic stores, and the store
5365 has undefined behavior if the alignment is not set to a value which is
5366 at least the size in bytes of the pointee. ``!nontemporal`` does not
5367 have any defined semantics for atomic stores.
5369 The optional constant ``align`` argument specifies the alignment of the
5370 operation (that is, the alignment of the memory address). A value of 0
5371 or an omitted ``align`` argument means that the operation has the ABI
5372 alignment for the target. It is the responsibility of the code emitter
5373 to ensure that the alignment information is correct. Overestimating the
5374 alignment results in undefined behavior. Underestimating the
5375 alignment may produce less efficient code. An alignment of 1 is always
5376 safe. The maximum possible alignment is ``1 << 29``.
5378 The optional ``!nontemporal`` metadata must reference a single metadata
5379 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
5380 value 1. The existence of the ``!nontemporal`` metadata on the instruction
5381 tells the optimizer and code generator that this load is not expected to
5382 be reused in the cache. The code generator may select special
5383 instructions to save cache bandwidth, such as the MOVNT instruction on
5389 The contents of memory are updated to contain ``<value>`` at the
5390 location specified by the ``<pointer>`` operand. If ``<value>`` is
5391 of scalar type then the number of bytes written does not exceed the
5392 minimum number of bytes needed to hold all bits of the type. For
5393 example, storing an ``i24`` writes at most three bytes. When writing a
5394 value of a type like ``i20`` with a size that is not an integral number
5395 of bytes, it is unspecified what happens to the extra bits that do not
5396 belong to the type, but they will typically be overwritten.
5401 .. code-block:: llvm
5403 %ptr = alloca i32 ; yields i32*:ptr
5404 store i32 3, i32* %ptr ; yields void
5405 %val = load i32* %ptr ; yields i32:val = i32 3
5409 '``fence``' Instruction
5410 ^^^^^^^^^^^^^^^^^^^^^^^
5417 fence [singlethread] <ordering> ; yields void
5422 The '``fence``' instruction is used to introduce happens-before edges
5428 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5429 defines what *synchronizes-with* edges they add. They can only be given
5430 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5435 A fence A which has (at least) ``release`` ordering semantics
5436 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5437 semantics if and only if there exist atomic operations X and Y, both
5438 operating on some atomic object M, such that A is sequenced before X, X
5439 modifies M (either directly or through some side effect of a sequence
5440 headed by X), Y is sequenced before B, and Y observes M. This provides a
5441 *happens-before* dependency between A and B. Rather than an explicit
5442 ``fence``, one (but not both) of the atomic operations X or Y might
5443 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5444 still *synchronize-with* the explicit ``fence`` and establish the
5445 *happens-before* edge.
5447 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5448 ``acquire`` and ``release`` semantics specified above, participates in
5449 the global program order of other ``seq_cst`` operations and/or fences.
5451 The optional ":ref:`singlethread <singlethread>`" argument specifies
5452 that the fence only synchronizes with other fences in the same thread.
5453 (This is useful for interacting with signal handlers.)
5458 .. code-block:: llvm
5460 fence acquire ; yields void
5461 fence singlethread seq_cst ; yields void
5465 '``cmpxchg``' Instruction
5466 ^^^^^^^^^^^^^^^^^^^^^^^^^
5473 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
5478 The '``cmpxchg``' instruction is used to atomically modify memory. It
5479 loads a value in memory and compares it to a given value. If they are
5480 equal, it tries to store a new value into the memory.
5485 There are three arguments to the '``cmpxchg``' instruction: an address
5486 to operate on, a value to compare to the value currently be at that
5487 address, and a new value to place at that address if the compared values
5488 are equal. The type of '<cmp>' must be an integer type whose bit width
5489 is a power of two greater than or equal to eight and less than or equal
5490 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5491 type, and the type of '<pointer>' must be a pointer to that type. If the
5492 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5493 to modify the number or order of execution of this ``cmpxchg`` with
5494 other :ref:`volatile operations <volatile>`.
5496 The success and failure :ref:`ordering <ordering>` arguments specify how this
5497 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
5498 must be at least ``monotonic``, the ordering constraint on failure must be no
5499 stronger than that on success, and the failure ordering cannot be either
5500 ``release`` or ``acq_rel``.
5502 The optional "``singlethread``" argument declares that the ``cmpxchg``
5503 is only atomic with respect to code (usually signal handlers) running in
5504 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5505 respect to all other code in the system.
5507 The pointer passed into cmpxchg must have alignment greater than or
5508 equal to the size in memory of the operand.
5513 The contents of memory at the location specified by the '``<pointer>``' operand
5514 is read and compared to '``<cmp>``'; if the read value is the equal, the
5515 '``<new>``' is written. The original value at the location is returned, together
5516 with a flag indicating success (true) or failure (false).
5518 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
5519 permitted: the operation may not write ``<new>`` even if the comparison
5522 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
5523 if the value loaded equals ``cmp``.
5525 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5526 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5527 load with an ordering parameter determined the second ordering parameter.
5532 .. code-block:: llvm
5535 %orig = atomic load i32* %ptr unordered ; yields i32
5539 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5540 %squared = mul i32 %cmp, %cmp
5541 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
5542 %value_loaded = extractvalue { i32, i1 } %val_success, 0
5543 %success = extractvalue { i32, i1 } %val_success, 1
5544 br i1 %success, label %done, label %loop
5551 '``atomicrmw``' Instruction
5552 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5559 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
5564 The '``atomicrmw``' instruction is used to atomically modify memory.
5569 There are three arguments to the '``atomicrmw``' instruction: an
5570 operation to apply, an address whose value to modify, an argument to the
5571 operation. The operation must be one of the following keywords:
5585 The type of '<value>' must be an integer type whose bit width is a power
5586 of two greater than or equal to eight and less than or equal to a
5587 target-specific size limit. The type of the '``<pointer>``' operand must
5588 be a pointer to that type. If the ``atomicrmw`` is marked as
5589 ``volatile``, then the optimizer is not allowed to modify the number or
5590 order of execution of this ``atomicrmw`` with other :ref:`volatile
5591 operations <volatile>`.
5596 The contents of memory at the location specified by the '``<pointer>``'
5597 operand are atomically read, modified, and written back. The original
5598 value at the location is returned. The modification is specified by the
5601 - xchg: ``*ptr = val``
5602 - add: ``*ptr = *ptr + val``
5603 - sub: ``*ptr = *ptr - val``
5604 - and: ``*ptr = *ptr & val``
5605 - nand: ``*ptr = ~(*ptr & val)``
5606 - or: ``*ptr = *ptr | val``
5607 - xor: ``*ptr = *ptr ^ val``
5608 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5609 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5610 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5612 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5618 .. code-block:: llvm
5620 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
5622 .. _i_getelementptr:
5624 '``getelementptr``' Instruction
5625 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5632 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5633 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5634 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5639 The '``getelementptr``' instruction is used to get the address of a
5640 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5641 address calculation only and does not access memory.
5646 The first argument is always a pointer or a vector of pointers, and
5647 forms the basis of the calculation. The remaining arguments are indices
5648 that indicate which of the elements of the aggregate object are indexed.
5649 The interpretation of each index is dependent on the type being indexed
5650 into. The first index always indexes the pointer value given as the
5651 first argument, the second index indexes a value of the type pointed to
5652 (not necessarily the value directly pointed to, since the first index
5653 can be non-zero), etc. The first type indexed into must be a pointer
5654 value, subsequent types can be arrays, vectors, and structs. Note that
5655 subsequent types being indexed into can never be pointers, since that
5656 would require loading the pointer before continuing calculation.
5658 The type of each index argument depends on the type it is indexing into.
5659 When indexing into a (optionally packed) structure, only ``i32`` integer
5660 **constants** are allowed (when using a vector of indices they must all
5661 be the **same** ``i32`` integer constant). When indexing into an array,
5662 pointer or vector, integers of any width are allowed, and they are not
5663 required to be constant. These integers are treated as signed values
5666 For example, let's consider a C code fragment and how it gets compiled
5682 int *foo(struct ST *s) {
5683 return &s[1].Z.B[5][13];
5686 The LLVM code generated by Clang is:
5688 .. code-block:: llvm
5690 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5691 %struct.ST = type { i32, double, %struct.RT }
5693 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5695 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5702 In the example above, the first index is indexing into the
5703 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5704 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5705 indexes into the third element of the structure, yielding a
5706 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5707 structure. The third index indexes into the second element of the
5708 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5709 dimensions of the array are subscripted into, yielding an '``i32``'
5710 type. The '``getelementptr``' instruction returns a pointer to this
5711 element, thus computing a value of '``i32*``' type.
5713 Note that it is perfectly legal to index partially through a structure,
5714 returning a pointer to an inner element. Because of this, the LLVM code
5715 for the given testcase is equivalent to:
5717 .. code-block:: llvm
5719 define i32* @foo(%struct.ST* %s) {
5720 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5721 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5722 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5723 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5724 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5728 If the ``inbounds`` keyword is present, the result value of the
5729 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5730 pointer is not an *in bounds* address of an allocated object, or if any
5731 of the addresses that would be formed by successive addition of the
5732 offsets implied by the indices to the base address with infinitely
5733 precise signed arithmetic are not an *in bounds* address of that
5734 allocated object. The *in bounds* addresses for an allocated object are
5735 all the addresses that point into the object, plus the address one byte
5736 past the end. In cases where the base is a vector of pointers the
5737 ``inbounds`` keyword applies to each of the computations element-wise.
5739 If the ``inbounds`` keyword is not present, the offsets are added to the
5740 base address with silently-wrapping two's complement arithmetic. If the
5741 offsets have a different width from the pointer, they are sign-extended
5742 or truncated to the width of the pointer. The result value of the
5743 ``getelementptr`` may be outside the object pointed to by the base
5744 pointer. The result value may not necessarily be used to access memory
5745 though, even if it happens to point into allocated storage. See the
5746 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5749 The getelementptr instruction is often confusing. For some more insight
5750 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5755 .. code-block:: llvm
5757 ; yields [12 x i8]*:aptr
5758 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5760 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5762 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5764 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5766 In cases where the pointer argument is a vector of pointers, each index
5767 must be a vector with the same number of elements. For example:
5769 .. code-block:: llvm
5771 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5773 Conversion Operations
5774 ---------------------
5776 The instructions in this category are the conversion instructions
5777 (casting) which all take a single operand and a type. They perform
5778 various bit conversions on the operand.
5780 '``trunc .. to``' Instruction
5781 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5788 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5793 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5798 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5799 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5800 of the same number of integers. The bit size of the ``value`` must be
5801 larger than the bit size of the destination type, ``ty2``. Equal sized
5802 types are not allowed.
5807 The '``trunc``' instruction truncates the high order bits in ``value``
5808 and converts the remaining bits to ``ty2``. Since the source size must
5809 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5810 It will always truncate bits.
5815 .. code-block:: llvm
5817 %X = trunc i32 257 to i8 ; yields i8:1
5818 %Y = trunc i32 123 to i1 ; yields i1:true
5819 %Z = trunc i32 122 to i1 ; yields i1:false
5820 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5822 '``zext .. to``' Instruction
5823 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5830 <result> = zext <ty> <value> to <ty2> ; yields ty2
5835 The '``zext``' instruction zero extends its operand to type ``ty2``.
5840 The '``zext``' instruction takes a value to cast, and a type to cast it
5841 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5842 the same number of integers. The bit size of the ``value`` must be
5843 smaller than the bit size of the destination type, ``ty2``.
5848 The ``zext`` fills the high order bits of the ``value`` with zero bits
5849 until it reaches the size of the destination type, ``ty2``.
5851 When zero extending from i1, the result will always be either 0 or 1.
5856 .. code-block:: llvm
5858 %X = zext i32 257 to i64 ; yields i64:257
5859 %Y = zext i1 true to i32 ; yields i32:1
5860 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5862 '``sext .. to``' Instruction
5863 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5870 <result> = sext <ty> <value> to <ty2> ; yields ty2
5875 The '``sext``' sign extends ``value`` to the type ``ty2``.
5880 The '``sext``' instruction takes a value to cast, and a type to cast it
5881 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5882 the same number of integers. The bit size of the ``value`` must be
5883 smaller than the bit size of the destination type, ``ty2``.
5888 The '``sext``' instruction performs a sign extension by copying the sign
5889 bit (highest order bit) of the ``value`` until it reaches the bit size
5890 of the type ``ty2``.
5892 When sign extending from i1, the extension always results in -1 or 0.
5897 .. code-block:: llvm
5899 %X = sext i8 -1 to i16 ; yields i16 :65535
5900 %Y = sext i1 true to i32 ; yields i32:-1
5901 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5903 '``fptrunc .. to``' Instruction
5904 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5911 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5916 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5921 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5922 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5923 The size of ``value`` must be larger than the size of ``ty2``. This
5924 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5929 The '``fptrunc``' instruction truncates a ``value`` from a larger
5930 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5931 point <t_floating>` type. If the value cannot fit within the
5932 destination type, ``ty2``, then the results are undefined.
5937 .. code-block:: llvm
5939 %X = fptrunc double 123.0 to float ; yields float:123.0
5940 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5942 '``fpext .. to``' Instruction
5943 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5950 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5955 The '``fpext``' extends a floating point ``value`` to a larger floating
5961 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5962 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5963 to. The source type must be smaller than the destination type.
5968 The '``fpext``' instruction extends the ``value`` from a smaller
5969 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5970 point <t_floating>` type. The ``fpext`` cannot be used to make a
5971 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5972 *no-op cast* for a floating point cast.
5977 .. code-block:: llvm
5979 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5980 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5982 '``fptoui .. to``' Instruction
5983 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5990 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5995 The '``fptoui``' converts a floating point ``value`` to its unsigned
5996 integer equivalent of type ``ty2``.
6001 The '``fptoui``' instruction takes a value to cast, which must be a
6002 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6003 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6004 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6005 type with the same number of elements as ``ty``
6010 The '``fptoui``' instruction converts its :ref:`floating
6011 point <t_floating>` operand into the nearest (rounding towards zero)
6012 unsigned integer value. If the value cannot fit in ``ty2``, the results
6018 .. code-block:: llvm
6020 %X = fptoui double 123.0 to i32 ; yields i32:123
6021 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
6022 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
6024 '``fptosi .. to``' Instruction
6025 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6032 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
6037 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
6038 ``value`` to type ``ty2``.
6043 The '``fptosi``' instruction takes a value to cast, which must be a
6044 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6045 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6046 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6047 type with the same number of elements as ``ty``
6052 The '``fptosi``' instruction converts its :ref:`floating
6053 point <t_floating>` operand into the nearest (rounding towards zero)
6054 signed integer value. If the value cannot fit in ``ty2``, the results
6060 .. code-block:: llvm
6062 %X = fptosi double -123.0 to i32 ; yields i32:-123
6063 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
6064 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
6066 '``uitofp .. to``' Instruction
6067 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6074 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
6079 The '``uitofp``' instruction regards ``value`` as an unsigned integer
6080 and converts that value to the ``ty2`` type.
6085 The '``uitofp``' instruction takes a value to cast, which must be a
6086 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6087 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6088 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6089 type with the same number of elements as ``ty``
6094 The '``uitofp``' instruction interprets its operand as an unsigned
6095 integer quantity and converts it to the corresponding floating point
6096 value. If the value cannot fit in the floating point value, the results
6102 .. code-block:: llvm
6104 %X = uitofp i32 257 to float ; yields float:257.0
6105 %Y = uitofp i8 -1 to double ; yields double:255.0
6107 '``sitofp .. to``' Instruction
6108 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6115 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
6120 The '``sitofp``' instruction regards ``value`` as a signed integer and
6121 converts that value to the ``ty2`` type.
6126 The '``sitofp``' instruction takes a value to cast, which must be a
6127 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6128 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6129 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6130 type with the same number of elements as ``ty``
6135 The '``sitofp``' instruction interprets its operand as a signed integer
6136 quantity and converts it to the corresponding floating point value. If
6137 the value cannot fit in the floating point value, the results are
6143 .. code-block:: llvm
6145 %X = sitofp i32 257 to float ; yields float:257.0
6146 %Y = sitofp i8 -1 to double ; yields double:-1.0
6150 '``ptrtoint .. to``' Instruction
6151 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6158 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
6163 The '``ptrtoint``' instruction converts the pointer or a vector of
6164 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
6169 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
6170 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
6171 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
6172 a vector of integers type.
6177 The '``ptrtoint``' instruction converts ``value`` to integer type
6178 ``ty2`` by interpreting the pointer value as an integer and either
6179 truncating or zero extending that value to the size of the integer type.
6180 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
6181 ``value`` is larger than ``ty2`` then a truncation is done. If they are
6182 the same size, then nothing is done (*no-op cast*) other than a type
6188 .. code-block:: llvm
6190 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
6191 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
6192 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
6196 '``inttoptr .. to``' Instruction
6197 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6204 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
6209 The '``inttoptr``' instruction converts an integer ``value`` to a
6210 pointer type, ``ty2``.
6215 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
6216 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
6222 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
6223 applying either a zero extension or a truncation depending on the size
6224 of the integer ``value``. If ``value`` is larger than the size of a
6225 pointer then a truncation is done. If ``value`` is smaller than the size
6226 of a pointer then a zero extension is done. If they are the same size,
6227 nothing is done (*no-op cast*).
6232 .. code-block:: llvm
6234 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
6235 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
6236 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
6237 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
6241 '``bitcast .. to``' Instruction
6242 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6249 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
6254 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
6260 The '``bitcast``' instruction takes a value to cast, which must be a
6261 non-aggregate first class value, and a type to cast it to, which must
6262 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
6263 bit sizes of ``value`` and the destination type, ``ty2``, must be
6264 identical. If the source type is a pointer, the destination type must
6265 also be a pointer of the same size. This instruction supports bitwise
6266 conversion of vectors to integers and to vectors of other types (as
6267 long as they have the same size).
6272 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
6273 is always a *no-op cast* because no bits change with this
6274 conversion. The conversion is done as if the ``value`` had been stored
6275 to memory and read back as type ``ty2``. Pointer (or vector of
6276 pointers) types may only be converted to other pointer (or vector of
6277 pointers) types with the same address space through this instruction.
6278 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
6279 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
6284 .. code-block:: llvm
6286 %X = bitcast i8 255 to i8 ; yields i8 :-1
6287 %Y = bitcast i32* %x to sint* ; yields sint*:%x
6288 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
6289 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
6291 .. _i_addrspacecast:
6293 '``addrspacecast .. to``' Instruction
6294 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6301 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
6306 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
6307 address space ``n`` to type ``pty2`` in address space ``m``.
6312 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
6313 to cast and a pointer type to cast it to, which must have a different
6319 The '``addrspacecast``' instruction converts the pointer value
6320 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
6321 value modification, depending on the target and the address space
6322 pair. Pointer conversions within the same address space must be
6323 performed with the ``bitcast`` instruction. Note that if the address space
6324 conversion is legal then both result and operand refer to the same memory
6330 .. code-block:: llvm
6332 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
6333 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
6334 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
6341 The instructions in this category are the "miscellaneous" instructions,
6342 which defy better classification.
6346 '``icmp``' Instruction
6347 ^^^^^^^^^^^^^^^^^^^^^^
6354 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6359 The '``icmp``' instruction returns a boolean value or a vector of
6360 boolean values based on comparison of its two integer, integer vector,
6361 pointer, or pointer vector operands.
6366 The '``icmp``' instruction takes three operands. The first operand is
6367 the condition code indicating the kind of comparison to perform. It is
6368 not a value, just a keyword. The possible condition code are:
6371 #. ``ne``: not equal
6372 #. ``ugt``: unsigned greater than
6373 #. ``uge``: unsigned greater or equal
6374 #. ``ult``: unsigned less than
6375 #. ``ule``: unsigned less or equal
6376 #. ``sgt``: signed greater than
6377 #. ``sge``: signed greater or equal
6378 #. ``slt``: signed less than
6379 #. ``sle``: signed less or equal
6381 The remaining two arguments must be :ref:`integer <t_integer>` or
6382 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
6383 must also be identical types.
6388 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
6389 code given as ``cond``. The comparison performed always yields either an
6390 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
6392 #. ``eq``: yields ``true`` if the operands are equal, ``false``
6393 otherwise. No sign interpretation is necessary or performed.
6394 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
6395 otherwise. No sign interpretation is necessary or performed.
6396 #. ``ugt``: interprets the operands as unsigned values and yields
6397 ``true`` if ``op1`` is greater than ``op2``.
6398 #. ``uge``: interprets the operands as unsigned values and yields
6399 ``true`` if ``op1`` is greater than or equal to ``op2``.
6400 #. ``ult``: interprets the operands as unsigned values and yields
6401 ``true`` if ``op1`` is less than ``op2``.
6402 #. ``ule``: interprets the operands as unsigned values and yields
6403 ``true`` if ``op1`` is less than or equal to ``op2``.
6404 #. ``sgt``: interprets the operands as signed values and yields ``true``
6405 if ``op1`` is greater than ``op2``.
6406 #. ``sge``: interprets the operands as signed values and yields ``true``
6407 if ``op1`` is greater than or equal to ``op2``.
6408 #. ``slt``: interprets the operands as signed values and yields ``true``
6409 if ``op1`` is less than ``op2``.
6410 #. ``sle``: interprets the operands as signed values and yields ``true``
6411 if ``op1`` is less than or equal to ``op2``.
6413 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
6414 are compared as if they were integers.
6416 If the operands are integer vectors, then they are compared element by
6417 element. The result is an ``i1`` vector with the same number of elements
6418 as the values being compared. Otherwise, the result is an ``i1``.
6423 .. code-block:: llvm
6425 <result> = icmp eq i32 4, 5 ; yields: result=false
6426 <result> = icmp ne float* %X, %X ; yields: result=false
6427 <result> = icmp ult i16 4, 5 ; yields: result=true
6428 <result> = icmp sgt i16 4, 5 ; yields: result=false
6429 <result> = icmp ule i16 -4, 5 ; yields: result=false
6430 <result> = icmp sge i16 4, 5 ; yields: result=false
6432 Note that the code generator does not yet support vector types with the
6433 ``icmp`` instruction.
6437 '``fcmp``' Instruction
6438 ^^^^^^^^^^^^^^^^^^^^^^
6445 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6450 The '``fcmp``' instruction returns a boolean value or vector of boolean
6451 values based on comparison of its operands.
6453 If the operands are floating point scalars, then the result type is a
6454 boolean (:ref:`i1 <t_integer>`).
6456 If the operands are floating point vectors, then the result type is a
6457 vector of boolean with the same number of elements as the operands being
6463 The '``fcmp``' instruction takes three operands. The first operand is
6464 the condition code indicating the kind of comparison to perform. It is
6465 not a value, just a keyword. The possible condition code are:
6467 #. ``false``: no comparison, always returns false
6468 #. ``oeq``: ordered and equal
6469 #. ``ogt``: ordered and greater than
6470 #. ``oge``: ordered and greater than or equal
6471 #. ``olt``: ordered and less than
6472 #. ``ole``: ordered and less than or equal
6473 #. ``one``: ordered and not equal
6474 #. ``ord``: ordered (no nans)
6475 #. ``ueq``: unordered or equal
6476 #. ``ugt``: unordered or greater than
6477 #. ``uge``: unordered or greater than or equal
6478 #. ``ult``: unordered or less than
6479 #. ``ule``: unordered or less than or equal
6480 #. ``une``: unordered or not equal
6481 #. ``uno``: unordered (either nans)
6482 #. ``true``: no comparison, always returns true
6484 *Ordered* means that neither operand is a QNAN while *unordered* means
6485 that either operand may be a QNAN.
6487 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6488 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6489 type. They must have identical types.
6494 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6495 condition code given as ``cond``. If the operands are vectors, then the
6496 vectors are compared element by element. Each comparison performed
6497 always yields an :ref:`i1 <t_integer>` result, as follows:
6499 #. ``false``: always yields ``false``, regardless of operands.
6500 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6501 is equal to ``op2``.
6502 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6503 is greater than ``op2``.
6504 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6505 is greater than or equal to ``op2``.
6506 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6507 is less than ``op2``.
6508 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6509 is less than or equal to ``op2``.
6510 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6511 is not equal to ``op2``.
6512 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6513 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6515 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6516 greater than ``op2``.
6517 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6518 greater than or equal to ``op2``.
6519 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6521 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6522 less than or equal to ``op2``.
6523 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6524 not equal to ``op2``.
6525 #. ``uno``: yields ``true`` if either operand is a QNAN.
6526 #. ``true``: always yields ``true``, regardless of operands.
6531 .. code-block:: llvm
6533 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6534 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6535 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6536 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6538 Note that the code generator does not yet support vector types with the
6539 ``fcmp`` instruction.
6543 '``phi``' Instruction
6544 ^^^^^^^^^^^^^^^^^^^^^
6551 <result> = phi <ty> [ <val0>, <label0>], ...
6556 The '``phi``' instruction is used to implement the φ node in the SSA
6557 graph representing the function.
6562 The type of the incoming values is specified with the first type field.
6563 After this, the '``phi``' instruction takes a list of pairs as
6564 arguments, with one pair for each predecessor basic block of the current
6565 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6566 the value arguments to the PHI node. Only labels may be used as the
6569 There must be no non-phi instructions between the start of a basic block
6570 and the PHI instructions: i.e. PHI instructions must be first in a basic
6573 For the purposes of the SSA form, the use of each incoming value is
6574 deemed to occur on the edge from the corresponding predecessor block to
6575 the current block (but after any definition of an '``invoke``'
6576 instruction's return value on the same edge).
6581 At runtime, the '``phi``' instruction logically takes on the value
6582 specified by the pair corresponding to the predecessor basic block that
6583 executed just prior to the current block.
6588 .. code-block:: llvm
6590 Loop: ; Infinite loop that counts from 0 on up...
6591 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6592 %nextindvar = add i32 %indvar, 1
6597 '``select``' Instruction
6598 ^^^^^^^^^^^^^^^^^^^^^^^^
6605 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6607 selty is either i1 or {<N x i1>}
6612 The '``select``' instruction is used to choose one value based on a
6613 condition, without IR-level branching.
6618 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6619 values indicating the condition, and two values of the same :ref:`first
6620 class <t_firstclass>` type. If the val1/val2 are vectors and the
6621 condition is a scalar, then entire vectors are selected, not individual
6627 If the condition is an i1 and it evaluates to 1, the instruction returns
6628 the first value argument; otherwise, it returns the second value
6631 If the condition is a vector of i1, then the value arguments must be
6632 vectors of the same size, and the selection is done element by element.
6637 .. code-block:: llvm
6639 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6643 '``call``' Instruction
6644 ^^^^^^^^^^^^^^^^^^^^^^
6651 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6656 The '``call``' instruction represents a simple function call.
6661 This instruction requires several arguments:
6663 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
6664 should perform tail call optimization. The ``tail`` marker is a hint that
6665 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
6666 means that the call must be tail call optimized in order for the program to
6667 be correct. The ``musttail`` marker provides these guarantees:
6669 #. The call will not cause unbounded stack growth if it is part of a
6670 recursive cycle in the call graph.
6671 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
6674 Both markers imply that the callee does not access allocas or varargs from
6675 the caller. Calls marked ``musttail`` must obey the following additional
6678 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
6679 or a pointer bitcast followed by a ret instruction.
6680 - The ret instruction must return the (possibly bitcasted) value
6681 produced by the call or void.
6682 - The caller and callee prototypes must match. Pointer types of
6683 parameters or return types may differ in pointee type, but not
6685 - The calling conventions of the caller and callee must match.
6686 - All ABI-impacting function attributes, such as sret, byval, inreg,
6687 returned, and inalloca, must match.
6688 - The callee must be varargs iff the caller is varargs. Bitcasting a
6689 non-varargs function to the appropriate varargs type is legal so
6690 long as the non-varargs prefixes obey the other rules.
6692 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
6693 the following conditions are met:
6695 - Caller and callee both have the calling convention ``fastcc``.
6696 - The call is in tail position (ret immediately follows call and ret
6697 uses value of call or is void).
6698 - Option ``-tailcallopt`` is enabled, or
6699 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6700 - `Platform-specific constraints are
6701 met. <CodeGenerator.html#tailcallopt>`_
6703 #. The optional "cconv" marker indicates which :ref:`calling
6704 convention <callingconv>` the call should use. If none is
6705 specified, the call defaults to using C calling conventions. The
6706 calling convention of the call must match the calling convention of
6707 the target function, or else the behavior is undefined.
6708 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6709 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6711 #. '``ty``': the type of the call instruction itself which is also the
6712 type of the return value. Functions that return no value are marked
6714 #. '``fnty``': shall be the signature of the pointer to function value
6715 being invoked. The argument types must match the types implied by
6716 this signature. This type can be omitted if the function is not
6717 varargs and if the function type does not return a pointer to a
6719 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6720 be invoked. In most cases, this is a direct function invocation, but
6721 indirect ``call``'s are just as possible, calling an arbitrary pointer
6723 #. '``function args``': argument list whose types match the function
6724 signature argument types and parameter attributes. All arguments must
6725 be of :ref:`first class <t_firstclass>` type. If the function signature
6726 indicates the function accepts a variable number of arguments, the
6727 extra arguments can be specified.
6728 #. The optional :ref:`function attributes <fnattrs>` list. Only
6729 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6730 attributes are valid here.
6735 The '``call``' instruction is used to cause control flow to transfer to
6736 a specified function, with its incoming arguments bound to the specified
6737 values. Upon a '``ret``' instruction in the called function, control
6738 flow continues with the instruction after the function call, and the
6739 return value of the function is bound to the result argument.
6744 .. code-block:: llvm
6746 %retval = call i32 @test(i32 %argc)
6747 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6748 %X = tail call i32 @foo() ; yields i32
6749 %Y = tail call fastcc i32 @foo() ; yields i32
6750 call void %foo(i8 97 signext)
6752 %struct.A = type { i32, i8 }
6753 %r = call %struct.A @foo() ; yields { i32, i8 }
6754 %gr = extractvalue %struct.A %r, 0 ; yields i32
6755 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6756 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6757 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6759 llvm treats calls to some functions with names and arguments that match
6760 the standard C99 library as being the C99 library functions, and may
6761 perform optimizations or generate code for them under that assumption.
6762 This is something we'd like to change in the future to provide better
6763 support for freestanding environments and non-C-based languages.
6767 '``va_arg``' Instruction
6768 ^^^^^^^^^^^^^^^^^^^^^^^^
6775 <resultval> = va_arg <va_list*> <arglist>, <argty>
6780 The '``va_arg``' instruction is used to access arguments passed through
6781 the "variable argument" area of a function call. It is used to implement
6782 the ``va_arg`` macro in C.
6787 This instruction takes a ``va_list*`` value and the type of the
6788 argument. It returns a value of the specified argument type and
6789 increments the ``va_list`` to point to the next argument. The actual
6790 type of ``va_list`` is target specific.
6795 The '``va_arg``' instruction loads an argument of the specified type
6796 from the specified ``va_list`` and causes the ``va_list`` to point to
6797 the next argument. For more information, see the variable argument
6798 handling :ref:`Intrinsic Functions <int_varargs>`.
6800 It is legal for this instruction to be called in a function which does
6801 not take a variable number of arguments, for example, the ``vfprintf``
6804 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6805 function <intrinsics>` because it takes a type as an argument.
6810 See the :ref:`variable argument processing <int_varargs>` section.
6812 Note that the code generator does not yet fully support va\_arg on many
6813 targets. Also, it does not currently support va\_arg with aggregate
6814 types on any target.
6818 '``landingpad``' Instruction
6819 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6826 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6827 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6829 <clause> := catch <type> <value>
6830 <clause> := filter <array constant type> <array constant>
6835 The '``landingpad``' instruction is used by `LLVM's exception handling
6836 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6837 is a landing pad --- one where the exception lands, and corresponds to the
6838 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6839 defines values supplied by the personality function (``pers_fn``) upon
6840 re-entry to the function. The ``resultval`` has the type ``resultty``.
6845 This instruction takes a ``pers_fn`` value. This is the personality
6846 function associated with the unwinding mechanism. The optional
6847 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6849 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6850 contains the global variable representing the "type" that may be caught
6851 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6852 clause takes an array constant as its argument. Use
6853 "``[0 x i8**] undef``" for a filter which cannot throw. The
6854 '``landingpad``' instruction must contain *at least* one ``clause`` or
6855 the ``cleanup`` flag.
6860 The '``landingpad``' instruction defines the values which are set by the
6861 personality function (``pers_fn``) upon re-entry to the function, and
6862 therefore the "result type" of the ``landingpad`` instruction. As with
6863 calling conventions, how the personality function results are
6864 represented in LLVM IR is target specific.
6866 The clauses are applied in order from top to bottom. If two
6867 ``landingpad`` instructions are merged together through inlining, the
6868 clauses from the calling function are appended to the list of clauses.
6869 When the call stack is being unwound due to an exception being thrown,
6870 the exception is compared against each ``clause`` in turn. If it doesn't
6871 match any of the clauses, and the ``cleanup`` flag is not set, then
6872 unwinding continues further up the call stack.
6874 The ``landingpad`` instruction has several restrictions:
6876 - A landing pad block is a basic block which is the unwind destination
6877 of an '``invoke``' instruction.
6878 - A landing pad block must have a '``landingpad``' instruction as its
6879 first non-PHI instruction.
6880 - There can be only one '``landingpad``' instruction within the landing
6882 - A basic block that is not a landing pad block may not include a
6883 '``landingpad``' instruction.
6884 - All '``landingpad``' instructions in a function must have the same
6885 personality function.
6890 .. code-block:: llvm
6892 ;; A landing pad which can catch an integer.
6893 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6895 ;; A landing pad that is a cleanup.
6896 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6898 ;; A landing pad which can catch an integer and can only throw a double.
6899 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6901 filter [1 x i8**] [@_ZTId]
6908 LLVM supports the notion of an "intrinsic function". These functions
6909 have well known names and semantics and are required to follow certain
6910 restrictions. Overall, these intrinsics represent an extension mechanism
6911 for the LLVM language that does not require changing all of the
6912 transformations in LLVM when adding to the language (or the bitcode
6913 reader/writer, the parser, etc...).
6915 Intrinsic function names must all start with an "``llvm.``" prefix. This
6916 prefix is reserved in LLVM for intrinsic names; thus, function names may
6917 not begin with this prefix. Intrinsic functions must always be external
6918 functions: you cannot define the body of intrinsic functions. Intrinsic
6919 functions may only be used in call or invoke instructions: it is illegal
6920 to take the address of an intrinsic function. Additionally, because
6921 intrinsic functions are part of the LLVM language, it is required if any
6922 are added that they be documented here.
6924 Some intrinsic functions can be overloaded, i.e., the intrinsic
6925 represents a family of functions that perform the same operation but on
6926 different data types. Because LLVM can represent over 8 million
6927 different integer types, overloading is used commonly to allow an
6928 intrinsic function to operate on any integer type. One or more of the
6929 argument types or the result type can be overloaded to accept any
6930 integer type. Argument types may also be defined as exactly matching a
6931 previous argument's type or the result type. This allows an intrinsic
6932 function which accepts multiple arguments, but needs all of them to be
6933 of the same type, to only be overloaded with respect to a single
6934 argument or the result.
6936 Overloaded intrinsics will have the names of its overloaded argument
6937 types encoded into its function name, each preceded by a period. Only
6938 those types which are overloaded result in a name suffix. Arguments
6939 whose type is matched against another type do not. For example, the
6940 ``llvm.ctpop`` function can take an integer of any width and returns an
6941 integer of exactly the same integer width. This leads to a family of
6942 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6943 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6944 overloaded, and only one type suffix is required. Because the argument's
6945 type is matched against the return type, it does not require its own
6948 To learn how to add an intrinsic function, please see the `Extending
6949 LLVM Guide <ExtendingLLVM.html>`_.
6953 Variable Argument Handling Intrinsics
6954 -------------------------------------
6956 Variable argument support is defined in LLVM with the
6957 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6958 functions. These functions are related to the similarly named macros
6959 defined in the ``<stdarg.h>`` header file.
6961 All of these functions operate on arguments that use a target-specific
6962 value type "``va_list``". The LLVM assembly language reference manual
6963 does not define what this type is, so all transformations should be
6964 prepared to handle these functions regardless of the type used.
6966 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6967 variable argument handling intrinsic functions are used.
6969 .. code-block:: llvm
6971 ; This struct is different for every platform. For most platforms,
6972 ; it is merely an i8*.
6973 %struct.va_list = type { i8* }
6975 ; For Unix x86_64 platforms, va_list is the following struct:
6976 ; %struct.va_list = type { i32, i32, i8*, i8* }
6978 define i32 @test(i32 %X, ...) {
6979 ; Initialize variable argument processing
6980 %ap = alloca %struct.va_list
6981 %ap2 = bitcast %struct.va_list* %ap to i8*
6982 call void @llvm.va_start(i8* %ap2)
6984 ; Read a single integer argument
6985 %tmp = va_arg i8* %ap2, i32
6987 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6989 %aq2 = bitcast i8** %aq to i8*
6990 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6991 call void @llvm.va_end(i8* %aq2)
6993 ; Stop processing of arguments.
6994 call void @llvm.va_end(i8* %ap2)
6998 declare void @llvm.va_start(i8*)
6999 declare void @llvm.va_copy(i8*, i8*)
7000 declare void @llvm.va_end(i8*)
7004 '``llvm.va_start``' Intrinsic
7005 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7012 declare void @llvm.va_start(i8* <arglist>)
7017 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
7018 subsequent use by ``va_arg``.
7023 The argument is a pointer to a ``va_list`` element to initialize.
7028 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
7029 available in C. In a target-dependent way, it initializes the
7030 ``va_list`` element to which the argument points, so that the next call
7031 to ``va_arg`` will produce the first variable argument passed to the
7032 function. Unlike the C ``va_start`` macro, this intrinsic does not need
7033 to know the last argument of the function as the compiler can figure
7036 '``llvm.va_end``' Intrinsic
7037 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7044 declare void @llvm.va_end(i8* <arglist>)
7049 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
7050 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
7055 The argument is a pointer to a ``va_list`` to destroy.
7060 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
7061 available in C. In a target-dependent way, it destroys the ``va_list``
7062 element to which the argument points. Calls to
7063 :ref:`llvm.va_start <int_va_start>` and
7064 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
7069 '``llvm.va_copy``' Intrinsic
7070 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7077 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
7082 The '``llvm.va_copy``' intrinsic copies the current argument position
7083 from the source argument list to the destination argument list.
7088 The first argument is a pointer to a ``va_list`` element to initialize.
7089 The second argument is a pointer to a ``va_list`` element to copy from.
7094 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
7095 available in C. In a target-dependent way, it copies the source
7096 ``va_list`` element into the destination ``va_list`` element. This
7097 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
7098 arbitrarily complex and require, for example, memory allocation.
7100 Accurate Garbage Collection Intrinsics
7101 --------------------------------------
7103 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
7104 (GC) requires the implementation and generation of these intrinsics.
7105 These intrinsics allow identification of :ref:`GC roots on the
7106 stack <int_gcroot>`, as well as garbage collector implementations that
7107 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
7108 Front-ends for type-safe garbage collected languages should generate
7109 these intrinsics to make use of the LLVM garbage collectors. For more
7110 details, see `Accurate Garbage Collection with
7111 LLVM <GarbageCollection.html>`_.
7113 The garbage collection intrinsics only operate on objects in the generic
7114 address space (address space zero).
7118 '``llvm.gcroot``' Intrinsic
7119 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7126 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
7131 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
7132 the code generator, and allows some metadata to be associated with it.
7137 The first argument specifies the address of a stack object that contains
7138 the root pointer. The second pointer (which must be either a constant or
7139 a global value address) contains the meta-data to be associated with the
7145 At runtime, a call to this intrinsic stores a null pointer into the
7146 "ptrloc" location. At compile-time, the code generator generates
7147 information to allow the runtime to find the pointer at GC safe points.
7148 The '``llvm.gcroot``' intrinsic may only be used in a function which
7149 :ref:`specifies a GC algorithm <gc>`.
7153 '``llvm.gcread``' Intrinsic
7154 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7161 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
7166 The '``llvm.gcread``' intrinsic identifies reads of references from heap
7167 locations, allowing garbage collector implementations that require read
7173 The second argument is the address to read from, which should be an
7174 address allocated from the garbage collector. The first object is a
7175 pointer to the start of the referenced object, if needed by the language
7176 runtime (otherwise null).
7181 The '``llvm.gcread``' intrinsic has the same semantics as a load
7182 instruction, but may be replaced with substantially more complex code by
7183 the garbage collector runtime, as needed. The '``llvm.gcread``'
7184 intrinsic may only be used in a function which :ref:`specifies a GC
7189 '``llvm.gcwrite``' Intrinsic
7190 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7197 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
7202 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
7203 locations, allowing garbage collector implementations that require write
7204 barriers (such as generational or reference counting collectors).
7209 The first argument is the reference to store, the second is the start of
7210 the object to store it to, and the third is the address of the field of
7211 Obj to store to. If the runtime does not require a pointer to the
7212 object, Obj may be null.
7217 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
7218 instruction, but may be replaced with substantially more complex code by
7219 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
7220 intrinsic may only be used in a function which :ref:`specifies a GC
7223 Code Generator Intrinsics
7224 -------------------------
7226 These intrinsics are provided by LLVM to expose special features that
7227 may only be implemented with code generator support.
7229 '``llvm.returnaddress``' Intrinsic
7230 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7237 declare i8 *@llvm.returnaddress(i32 <level>)
7242 The '``llvm.returnaddress``' intrinsic attempts to compute a
7243 target-specific value indicating the return address of the current
7244 function or one of its callers.
7249 The argument to this intrinsic indicates which function to return the
7250 address for. Zero indicates the calling function, one indicates its
7251 caller, etc. The argument is **required** to be a constant integer
7257 The '``llvm.returnaddress``' intrinsic either returns a pointer
7258 indicating the return address of the specified call frame, or zero if it
7259 cannot be identified. The value returned by this intrinsic is likely to
7260 be incorrect or 0 for arguments other than zero, so it should only be
7261 used for debugging purposes.
7263 Note that calling this intrinsic does not prevent function inlining or
7264 other aggressive transformations, so the value returned may not be that
7265 of the obvious source-language caller.
7267 '``llvm.frameaddress``' Intrinsic
7268 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7275 declare i8* @llvm.frameaddress(i32 <level>)
7280 The '``llvm.frameaddress``' intrinsic attempts to return the
7281 target-specific frame pointer value for the specified stack frame.
7286 The argument to this intrinsic indicates which function to return the
7287 frame pointer for. Zero indicates the calling function, one indicates
7288 its caller, etc. The argument is **required** to be a constant integer
7294 The '``llvm.frameaddress``' intrinsic either returns a pointer
7295 indicating the frame address of the specified call frame, or zero if it
7296 cannot be identified. The value returned by this intrinsic is likely to
7297 be incorrect or 0 for arguments other than zero, so it should only be
7298 used for debugging purposes.
7300 Note that calling this intrinsic does not prevent function inlining or
7301 other aggressive transformations, so the value returned may not be that
7302 of the obvious source-language caller.
7304 '``llvm.frameallocate``' and '``llvm.framerecover``' Intrinsics
7305 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7312 declare i8* @llvm.frameallocate(i32 %size)
7313 declare i8* @llvm.framerecover(i8* %func, i8* %fp)
7318 The '``llvm.frameallocate``' intrinsic allocates stack memory at some fixed
7319 offset from the frame pointer, and the '``llvm.framerecover``'
7320 intrinsic applies that offset to a live frame pointer to recover the address of
7321 the allocation. The offset is computed during frame layout of the caller of
7322 ``llvm.frameallocate``.
7327 The ``size`` argument to '``llvm.frameallocate``' must be a constant integer
7328 indicating the amount of stack memory to allocate. As with allocas, allocating
7329 zero bytes is legal, but the result is undefined.
7331 The ``func`` argument to '``llvm.framerecover``' must be a constant
7332 bitcasted pointer to a function defined in the current module. The code
7333 generator cannot determine the frame allocation offset of functions defined in
7336 The ``fp`` argument to '``llvm.framerecover``' must be a frame
7337 pointer of a call frame that is currently live. The return value of
7338 '``llvm.frameaddress``' is one way to produce such a value, but most platforms
7339 also expose the frame pointer through stack unwinding mechanisms.
7344 These intrinsics allow a group of functions to access one stack memory
7345 allocation in an ancestor stack frame. The memory returned from
7346 '``llvm.frameallocate``' may be allocated prior to stack realignment, so the
7347 memory is only aligned to the ABI-required stack alignment. Each function may
7348 only call '``llvm.frameallocate``' one or zero times from the function entry
7349 block. The frame allocation intrinsic inhibits inlining, as any frame
7350 allocations in the inlined function frame are likely to be at a different
7351 offset from the one used by '``llvm.framerecover``' called with the
7354 .. _int_read_register:
7355 .. _int_write_register:
7357 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
7358 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7365 declare i32 @llvm.read_register.i32(metadata)
7366 declare i64 @llvm.read_register.i64(metadata)
7367 declare void @llvm.write_register.i32(metadata, i32 @value)
7368 declare void @llvm.write_register.i64(metadata, i64 @value)
7374 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
7375 provides access to the named register. The register must be valid on
7376 the architecture being compiled to. The type needs to be compatible
7377 with the register being read.
7382 The '``llvm.read_register``' intrinsic returns the current value of the
7383 register, where possible. The '``llvm.write_register``' intrinsic sets
7384 the current value of the register, where possible.
7386 This is useful to implement named register global variables that need
7387 to always be mapped to a specific register, as is common practice on
7388 bare-metal programs including OS kernels.
7390 The compiler doesn't check for register availability or use of the used
7391 register in surrounding code, including inline assembly. Because of that,
7392 allocatable registers are not supported.
7394 Warning: So far it only works with the stack pointer on selected
7395 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
7396 work is needed to support other registers and even more so, allocatable
7401 '``llvm.stacksave``' Intrinsic
7402 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7409 declare i8* @llvm.stacksave()
7414 The '``llvm.stacksave``' intrinsic is used to remember the current state
7415 of the function stack, for use with
7416 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
7417 implementing language features like scoped automatic variable sized
7423 This intrinsic returns a opaque pointer value that can be passed to
7424 :ref:`llvm.stackrestore <int_stackrestore>`. When an
7425 ``llvm.stackrestore`` intrinsic is executed with a value saved from
7426 ``llvm.stacksave``, it effectively restores the state of the stack to
7427 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
7428 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
7429 were allocated after the ``llvm.stacksave`` was executed.
7431 .. _int_stackrestore:
7433 '``llvm.stackrestore``' Intrinsic
7434 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7441 declare void @llvm.stackrestore(i8* %ptr)
7446 The '``llvm.stackrestore``' intrinsic is used to restore the state of
7447 the function stack to the state it was in when the corresponding
7448 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
7449 useful for implementing language features like scoped automatic variable
7450 sized arrays in C99.
7455 See the description for :ref:`llvm.stacksave <int_stacksave>`.
7457 '``llvm.prefetch``' Intrinsic
7458 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7465 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7470 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
7471 insert a prefetch instruction if supported; otherwise, it is a noop.
7472 Prefetches have no effect on the behavior of the program but can change
7473 its performance characteristics.
7478 ``address`` is the address to be prefetched, ``rw`` is the specifier
7479 determining if the fetch should be for a read (0) or write (1), and
7480 ``locality`` is a temporal locality specifier ranging from (0) - no
7481 locality, to (3) - extremely local keep in cache. The ``cache type``
7482 specifies whether the prefetch is performed on the data (1) or
7483 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
7484 arguments must be constant integers.
7489 This intrinsic does not modify the behavior of the program. In
7490 particular, prefetches cannot trap and do not produce a value. On
7491 targets that support this intrinsic, the prefetch can provide hints to
7492 the processor cache for better performance.
7494 '``llvm.pcmarker``' Intrinsic
7495 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7502 declare void @llvm.pcmarker(i32 <id>)
7507 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7508 Counter (PC) in a region of code to simulators and other tools. The
7509 method is target specific, but it is expected that the marker will use
7510 exported symbols to transmit the PC of the marker. The marker makes no
7511 guarantees that it will remain with any specific instruction after
7512 optimizations. It is possible that the presence of a marker will inhibit
7513 optimizations. The intended use is to be inserted after optimizations to
7514 allow correlations of simulation runs.
7519 ``id`` is a numerical id identifying the marker.
7524 This intrinsic does not modify the behavior of the program. Backends
7525 that do not support this intrinsic may ignore it.
7527 '``llvm.readcyclecounter``' Intrinsic
7528 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7535 declare i64 @llvm.readcyclecounter()
7540 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
7541 counter register (or similar low latency, high accuracy clocks) on those
7542 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7543 should map to RPCC. As the backing counters overflow quickly (on the
7544 order of 9 seconds on alpha), this should only be used for small
7550 When directly supported, reading the cycle counter should not modify any
7551 memory. Implementations are allowed to either return a application
7552 specific value or a system wide value. On backends without support, this
7553 is lowered to a constant 0.
7555 Note that runtime support may be conditional on the privilege-level code is
7556 running at and the host platform.
7558 '``llvm.clear_cache``' Intrinsic
7559 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7566 declare void @llvm.clear_cache(i8*, i8*)
7571 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
7572 in the specified range to the execution unit of the processor. On
7573 targets with non-unified instruction and data cache, the implementation
7574 flushes the instruction cache.
7579 On platforms with coherent instruction and data caches (e.g. x86), this
7580 intrinsic is a nop. On platforms with non-coherent instruction and data
7581 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
7582 instructions or a system call, if cache flushing requires special
7585 The default behavior is to emit a call to ``__clear_cache`` from the run
7588 This instrinsic does *not* empty the instruction pipeline. Modifications
7589 of the current function are outside the scope of the intrinsic.
7591 '``llvm.instrprof_increment``' Intrinsic
7592 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7599 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
7600 i32 <num-counters>, i32 <index>)
7605 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
7606 frontend for use with instrumentation based profiling. These will be
7607 lowered by the ``-instrprof`` pass to generate execution counts of a
7613 The first argument is a pointer to a global variable containing the
7614 name of the entity being instrumented. This should generally be the
7615 (mangled) function name for a set of counters.
7617 The second argument is a hash value that can be used by the consumer
7618 of the profile data to detect changes to the instrumented source, and
7619 the third is the number of counters associated with ``name``. It is an
7620 error if ``hash`` or ``num-counters`` differ between two instances of
7621 ``instrprof_increment`` that refer to the same name.
7623 The last argument refers to which of the counters for ``name`` should
7624 be incremented. It should be a value between 0 and ``num-counters``.
7629 This intrinsic represents an increment of a profiling counter. It will
7630 cause the ``-instrprof`` pass to generate the appropriate data
7631 structures and the code to increment the appropriate value, in a
7632 format that can be written out by a compiler runtime and consumed via
7633 the ``llvm-profdata`` tool.
7635 Standard C Library Intrinsics
7636 -----------------------------
7638 LLVM provides intrinsics for a few important standard C library
7639 functions. These intrinsics allow source-language front-ends to pass
7640 information about the alignment of the pointer arguments to the code
7641 generator, providing opportunity for more efficient code generation.
7645 '``llvm.memcpy``' Intrinsic
7646 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7651 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
7652 integer bit width and for different address spaces. Not all targets
7653 support all bit widths however.
7657 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7658 i32 <len>, i32 <align>, i1 <isvolatile>)
7659 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7660 i64 <len>, i32 <align>, i1 <isvolatile>)
7665 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7666 source location to the destination location.
7668 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
7669 intrinsics do not return a value, takes extra alignment/isvolatile
7670 arguments and the pointers can be in specified address spaces.
7675 The first argument is a pointer to the destination, the second is a
7676 pointer to the source. The third argument is an integer argument
7677 specifying the number of bytes to copy, the fourth argument is the
7678 alignment of the source and destination locations, and the fifth is a
7679 boolean indicating a volatile access.
7681 If the call to this intrinsic has an alignment value that is not 0 or 1,
7682 then the caller guarantees that both the source and destination pointers
7683 are aligned to that boundary.
7685 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
7686 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7687 very cleanly specified and it is unwise to depend on it.
7692 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7693 source location to the destination location, which are not allowed to
7694 overlap. It copies "len" bytes of memory over. If the argument is known
7695 to be aligned to some boundary, this can be specified as the fourth
7696 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
7698 '``llvm.memmove``' Intrinsic
7699 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7704 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7705 bit width and for different address space. Not all targets support all
7710 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7711 i32 <len>, i32 <align>, i1 <isvolatile>)
7712 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7713 i64 <len>, i32 <align>, i1 <isvolatile>)
7718 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7719 source location to the destination location. It is similar to the
7720 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7723 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7724 intrinsics do not return a value, takes extra alignment/isvolatile
7725 arguments and the pointers can be in specified address spaces.
7730 The first argument is a pointer to the destination, the second is a
7731 pointer to the source. The third argument is an integer argument
7732 specifying the number of bytes to copy, the fourth argument is the
7733 alignment of the source and destination locations, and the fifth is a
7734 boolean indicating a volatile access.
7736 If the call to this intrinsic has an alignment value that is not 0 or 1,
7737 then the caller guarantees that the source and destination pointers are
7738 aligned to that boundary.
7740 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7741 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7742 not very cleanly specified and it is unwise to depend on it.
7747 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7748 source location to the destination location, which may overlap. It
7749 copies "len" bytes of memory over. If the argument is known to be
7750 aligned to some boundary, this can be specified as the fourth argument,
7751 otherwise it should be set to 0 or 1 (both meaning no alignment).
7753 '``llvm.memset.*``' Intrinsics
7754 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7759 This is an overloaded intrinsic. You can use llvm.memset on any integer
7760 bit width and for different address spaces. However, not all targets
7761 support all bit widths.
7765 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7766 i32 <len>, i32 <align>, i1 <isvolatile>)
7767 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7768 i64 <len>, i32 <align>, i1 <isvolatile>)
7773 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7774 particular byte value.
7776 Note that, unlike the standard libc function, the ``llvm.memset``
7777 intrinsic does not return a value and takes extra alignment/volatile
7778 arguments. Also, the destination can be in an arbitrary address space.
7783 The first argument is a pointer to the destination to fill, the second
7784 is the byte value with which to fill it, the third argument is an
7785 integer argument specifying the number of bytes to fill, and the fourth
7786 argument is the known alignment of the destination location.
7788 If the call to this intrinsic has an alignment value that is not 0 or 1,
7789 then the caller guarantees that the destination pointer is aligned to
7792 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7793 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7794 very cleanly specified and it is unwise to depend on it.
7799 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7800 at the destination location. If the argument is known to be aligned to
7801 some boundary, this can be specified as the fourth argument, otherwise
7802 it should be set to 0 or 1 (both meaning no alignment).
7804 '``llvm.sqrt.*``' Intrinsic
7805 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7810 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7811 floating point or vector of floating point type. Not all targets support
7816 declare float @llvm.sqrt.f32(float %Val)
7817 declare double @llvm.sqrt.f64(double %Val)
7818 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7819 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7820 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7825 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7826 returning the same value as the libm '``sqrt``' functions would. Unlike
7827 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7828 negative numbers other than -0.0 (which allows for better optimization,
7829 because there is no need to worry about errno being set).
7830 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7835 The argument and return value are floating point numbers of the same
7841 This function returns the sqrt of the specified operand if it is a
7842 nonnegative floating point number.
7844 '``llvm.powi.*``' Intrinsic
7845 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7850 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7851 floating point or vector of floating point type. Not all targets support
7856 declare float @llvm.powi.f32(float %Val, i32 %power)
7857 declare double @llvm.powi.f64(double %Val, i32 %power)
7858 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7859 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7860 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7865 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7866 specified (positive or negative) power. The order of evaluation of
7867 multiplications is not defined. When a vector of floating point type is
7868 used, the second argument remains a scalar integer value.
7873 The second argument is an integer power, and the first is a value to
7874 raise to that power.
7879 This function returns the first value raised to the second power with an
7880 unspecified sequence of rounding operations.
7882 '``llvm.sin.*``' Intrinsic
7883 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7888 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7889 floating point or vector of floating point type. Not all targets support
7894 declare float @llvm.sin.f32(float %Val)
7895 declare double @llvm.sin.f64(double %Val)
7896 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7897 declare fp128 @llvm.sin.f128(fp128 %Val)
7898 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7903 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7908 The argument and return value are floating point numbers of the same
7914 This function returns the sine of the specified operand, returning the
7915 same values as the libm ``sin`` functions would, and handles error
7916 conditions in the same way.
7918 '``llvm.cos.*``' Intrinsic
7919 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7924 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7925 floating point or vector of floating point type. Not all targets support
7930 declare float @llvm.cos.f32(float %Val)
7931 declare double @llvm.cos.f64(double %Val)
7932 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7933 declare fp128 @llvm.cos.f128(fp128 %Val)
7934 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7939 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7944 The argument and return value are floating point numbers of the same
7950 This function returns the cosine of the specified operand, returning the
7951 same values as the libm ``cos`` functions would, and handles error
7952 conditions in the same way.
7954 '``llvm.pow.*``' Intrinsic
7955 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7960 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7961 floating point or vector of floating point type. Not all targets support
7966 declare float @llvm.pow.f32(float %Val, float %Power)
7967 declare double @llvm.pow.f64(double %Val, double %Power)
7968 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7969 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7970 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7975 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7976 specified (positive or negative) power.
7981 The second argument is a floating point power, and the first is a value
7982 to raise to that power.
7987 This function returns the first value raised to the second power,
7988 returning the same values as the libm ``pow`` functions would, and
7989 handles error conditions in the same way.
7991 '``llvm.exp.*``' Intrinsic
7992 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7997 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7998 floating point or vector of floating point type. Not all targets support
8003 declare float @llvm.exp.f32(float %Val)
8004 declare double @llvm.exp.f64(double %Val)
8005 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
8006 declare fp128 @llvm.exp.f128(fp128 %Val)
8007 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
8012 The '``llvm.exp.*``' intrinsics perform the exp function.
8017 The argument and return value are floating point numbers of the same
8023 This function returns the same values as the libm ``exp`` functions
8024 would, and handles error conditions in the same way.
8026 '``llvm.exp2.*``' Intrinsic
8027 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8032 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
8033 floating point or vector of floating point type. Not all targets support
8038 declare float @llvm.exp2.f32(float %Val)
8039 declare double @llvm.exp2.f64(double %Val)
8040 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
8041 declare fp128 @llvm.exp2.f128(fp128 %Val)
8042 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
8047 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
8052 The argument and return value are floating point numbers of the same
8058 This function returns the same values as the libm ``exp2`` functions
8059 would, and handles error conditions in the same way.
8061 '``llvm.log.*``' Intrinsic
8062 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8067 This is an overloaded intrinsic. You can use ``llvm.log`` on any
8068 floating point or vector of floating point type. Not all targets support
8073 declare float @llvm.log.f32(float %Val)
8074 declare double @llvm.log.f64(double %Val)
8075 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
8076 declare fp128 @llvm.log.f128(fp128 %Val)
8077 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
8082 The '``llvm.log.*``' intrinsics perform the log function.
8087 The argument and return value are floating point numbers of the same
8093 This function returns the same values as the libm ``log`` functions
8094 would, and handles error conditions in the same way.
8096 '``llvm.log10.*``' Intrinsic
8097 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8102 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
8103 floating point or vector of floating point type. Not all targets support
8108 declare float @llvm.log10.f32(float %Val)
8109 declare double @llvm.log10.f64(double %Val)
8110 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
8111 declare fp128 @llvm.log10.f128(fp128 %Val)
8112 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
8117 The '``llvm.log10.*``' intrinsics perform the log10 function.
8122 The argument and return value are floating point numbers of the same
8128 This function returns the same values as the libm ``log10`` functions
8129 would, and handles error conditions in the same way.
8131 '``llvm.log2.*``' Intrinsic
8132 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8137 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
8138 floating point or vector of floating point type. Not all targets support
8143 declare float @llvm.log2.f32(float %Val)
8144 declare double @llvm.log2.f64(double %Val)
8145 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
8146 declare fp128 @llvm.log2.f128(fp128 %Val)
8147 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
8152 The '``llvm.log2.*``' intrinsics perform the log2 function.
8157 The argument and return value are floating point numbers of the same
8163 This function returns the same values as the libm ``log2`` functions
8164 would, and handles error conditions in the same way.
8166 '``llvm.fma.*``' Intrinsic
8167 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8172 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
8173 floating point or vector of floating point type. Not all targets support
8178 declare float @llvm.fma.f32(float %a, float %b, float %c)
8179 declare double @llvm.fma.f64(double %a, double %b, double %c)
8180 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
8181 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
8182 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
8187 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
8193 The argument and return value are floating point numbers of the same
8199 This function returns the same values as the libm ``fma`` functions
8200 would, and does not set errno.
8202 '``llvm.fabs.*``' Intrinsic
8203 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8208 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
8209 floating point or vector of floating point type. Not all targets support
8214 declare float @llvm.fabs.f32(float %Val)
8215 declare double @llvm.fabs.f64(double %Val)
8216 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
8217 declare fp128 @llvm.fabs.f128(fp128 %Val)
8218 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
8223 The '``llvm.fabs.*``' intrinsics return the absolute value of the
8229 The argument and return value are floating point numbers of the same
8235 This function returns the same values as the libm ``fabs`` functions
8236 would, and handles error conditions in the same way.
8238 '``llvm.minnum.*``' Intrinsic
8239 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8244 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
8245 floating point or vector of floating point type. Not all targets support
8250 declare float @llvm.minnum.f32(float %Val0, float %Val1)
8251 declare double @llvm.minnum.f64(double %Val0, double %Val1)
8252 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8253 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
8254 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8259 The '``llvm.minnum.*``' intrinsics return the minimum of the two
8266 The arguments and return value are floating point numbers of the same
8272 Follows the IEEE-754 semantics for minNum, which also match for libm's
8275 If either operand is a NaN, returns the other non-NaN operand. Returns
8276 NaN only if both operands are NaN. If the operands compare equal,
8277 returns a value that compares equal to both operands. This means that
8278 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8280 '``llvm.maxnum.*``' Intrinsic
8281 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8286 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
8287 floating point or vector of floating point type. Not all targets support
8292 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
8293 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
8294 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8295 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
8296 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8301 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
8308 The arguments and return value are floating point numbers of the same
8313 Follows the IEEE-754 semantics for maxNum, which also match for libm's
8316 If either operand is a NaN, returns the other non-NaN operand. Returns
8317 NaN only if both operands are NaN. If the operands compare equal,
8318 returns a value that compares equal to both operands. This means that
8319 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8321 '``llvm.copysign.*``' Intrinsic
8322 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8327 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
8328 floating point or vector of floating point type. Not all targets support
8333 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
8334 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
8335 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
8336 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
8337 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
8342 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
8343 first operand and the sign of the second operand.
8348 The arguments and return value are floating point numbers of the same
8354 This function returns the same values as the libm ``copysign``
8355 functions would, and handles error conditions in the same way.
8357 '``llvm.floor.*``' Intrinsic
8358 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8363 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
8364 floating point or vector of floating point type. Not all targets support
8369 declare float @llvm.floor.f32(float %Val)
8370 declare double @llvm.floor.f64(double %Val)
8371 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
8372 declare fp128 @llvm.floor.f128(fp128 %Val)
8373 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
8378 The '``llvm.floor.*``' intrinsics return the floor of the operand.
8383 The argument and return value are floating point numbers of the same
8389 This function returns the same values as the libm ``floor`` functions
8390 would, and handles error conditions in the same way.
8392 '``llvm.ceil.*``' Intrinsic
8393 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8398 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
8399 floating point or vector of floating point type. Not all targets support
8404 declare float @llvm.ceil.f32(float %Val)
8405 declare double @llvm.ceil.f64(double %Val)
8406 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
8407 declare fp128 @llvm.ceil.f128(fp128 %Val)
8408 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
8413 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
8418 The argument and return value are floating point numbers of the same
8424 This function returns the same values as the libm ``ceil`` functions
8425 would, and handles error conditions in the same way.
8427 '``llvm.trunc.*``' Intrinsic
8428 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8433 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
8434 floating point or vector of floating point type. Not all targets support
8439 declare float @llvm.trunc.f32(float %Val)
8440 declare double @llvm.trunc.f64(double %Val)
8441 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
8442 declare fp128 @llvm.trunc.f128(fp128 %Val)
8443 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
8448 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
8449 nearest integer not larger in magnitude than the operand.
8454 The argument and return value are floating point numbers of the same
8460 This function returns the same values as the libm ``trunc`` functions
8461 would, and handles error conditions in the same way.
8463 '``llvm.rint.*``' Intrinsic
8464 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8469 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
8470 floating point or vector of floating point type. Not all targets support
8475 declare float @llvm.rint.f32(float %Val)
8476 declare double @llvm.rint.f64(double %Val)
8477 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
8478 declare fp128 @llvm.rint.f128(fp128 %Val)
8479 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
8484 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
8485 nearest integer. It may raise an inexact floating-point exception if the
8486 operand isn't an integer.
8491 The argument and return value are floating point numbers of the same
8497 This function returns the same values as the libm ``rint`` functions
8498 would, and handles error conditions in the same way.
8500 '``llvm.nearbyint.*``' Intrinsic
8501 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8506 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
8507 floating point or vector of floating point type. Not all targets support
8512 declare float @llvm.nearbyint.f32(float %Val)
8513 declare double @llvm.nearbyint.f64(double %Val)
8514 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
8515 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
8516 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
8521 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
8527 The argument and return value are floating point numbers of the same
8533 This function returns the same values as the libm ``nearbyint``
8534 functions would, and handles error conditions in the same way.
8536 '``llvm.round.*``' Intrinsic
8537 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8542 This is an overloaded intrinsic. You can use ``llvm.round`` on any
8543 floating point or vector of floating point type. Not all targets support
8548 declare float @llvm.round.f32(float %Val)
8549 declare double @llvm.round.f64(double %Val)
8550 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
8551 declare fp128 @llvm.round.f128(fp128 %Val)
8552 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
8557 The '``llvm.round.*``' intrinsics returns the operand rounded to the
8563 The argument and return value are floating point numbers of the same
8569 This function returns the same values as the libm ``round``
8570 functions would, and handles error conditions in the same way.
8572 Bit Manipulation Intrinsics
8573 ---------------------------
8575 LLVM provides intrinsics for a few important bit manipulation
8576 operations. These allow efficient code generation for some algorithms.
8578 '``llvm.bswap.*``' Intrinsics
8579 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8584 This is an overloaded intrinsic function. You can use bswap on any
8585 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
8589 declare i16 @llvm.bswap.i16(i16 <id>)
8590 declare i32 @llvm.bswap.i32(i32 <id>)
8591 declare i64 @llvm.bswap.i64(i64 <id>)
8596 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
8597 values with an even number of bytes (positive multiple of 16 bits).
8598 These are useful for performing operations on data that is not in the
8599 target's native byte order.
8604 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
8605 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
8606 intrinsic returns an i32 value that has the four bytes of the input i32
8607 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
8608 returned i32 will have its bytes in 3, 2, 1, 0 order. The
8609 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
8610 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
8613 '``llvm.ctpop.*``' Intrinsic
8614 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8619 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
8620 bit width, or on any vector with integer elements. Not all targets
8621 support all bit widths or vector types, however.
8625 declare i8 @llvm.ctpop.i8(i8 <src>)
8626 declare i16 @llvm.ctpop.i16(i16 <src>)
8627 declare i32 @llvm.ctpop.i32(i32 <src>)
8628 declare i64 @llvm.ctpop.i64(i64 <src>)
8629 declare i256 @llvm.ctpop.i256(i256 <src>)
8630 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
8635 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
8641 The only argument is the value to be counted. The argument may be of any
8642 integer type, or a vector with integer elements. The return type must
8643 match the argument type.
8648 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
8649 each element of a vector.
8651 '``llvm.ctlz.*``' Intrinsic
8652 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8657 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
8658 integer bit width, or any vector whose elements are integers. Not all
8659 targets support all bit widths or vector types, however.
8663 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
8664 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
8665 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
8666 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
8667 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
8668 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8673 The '``llvm.ctlz``' family of intrinsic functions counts the number of
8674 leading zeros in a variable.
8679 The first argument is the value to be counted. This argument may be of
8680 any integer type, or a vector with integer element type. The return
8681 type must match the first argument type.
8683 The second argument must be a constant and is a flag to indicate whether
8684 the intrinsic should ensure that a zero as the first argument produces a
8685 defined result. Historically some architectures did not provide a
8686 defined result for zero values as efficiently, and many algorithms are
8687 now predicated on avoiding zero-value inputs.
8692 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
8693 zeros in a variable, or within each element of the vector. If
8694 ``src == 0`` then the result is the size in bits of the type of ``src``
8695 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8696 ``llvm.ctlz(i32 2) = 30``.
8698 '``llvm.cttz.*``' Intrinsic
8699 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8704 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
8705 integer bit width, or any vector of integer elements. Not all targets
8706 support all bit widths or vector types, however.
8710 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
8711 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
8712 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
8713 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
8714 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
8715 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8720 The '``llvm.cttz``' family of intrinsic functions counts the number of
8726 The first argument is the value to be counted. This argument may be of
8727 any integer type, or a vector with integer element type. The return
8728 type must match the first argument type.
8730 The second argument must be a constant and is a flag to indicate whether
8731 the intrinsic should ensure that a zero as the first argument produces a
8732 defined result. Historically some architectures did not provide a
8733 defined result for zero values as efficiently, and many algorithms are
8734 now predicated on avoiding zero-value inputs.
8739 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
8740 zeros in a variable, or within each element of a vector. If ``src == 0``
8741 then the result is the size in bits of the type of ``src`` if
8742 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8743 ``llvm.cttz(2) = 1``.
8745 Arithmetic with Overflow Intrinsics
8746 -----------------------------------
8748 LLVM provides intrinsics for some arithmetic with overflow operations.
8750 '``llvm.sadd.with.overflow.*``' Intrinsics
8751 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8756 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
8757 on any integer bit width.
8761 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8762 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8763 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8768 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8769 a signed addition of the two arguments, and indicate whether an overflow
8770 occurred during the signed summation.
8775 The arguments (%a and %b) and the first element of the result structure
8776 may be of integer types of any bit width, but they must have the same
8777 bit width. The second element of the result structure must be of type
8778 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8784 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8785 a signed addition of the two variables. They return a structure --- the
8786 first element of which is the signed summation, and the second element
8787 of which is a bit specifying if the signed summation resulted in an
8793 .. code-block:: llvm
8795 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8796 %sum = extractvalue {i32, i1} %res, 0
8797 %obit = extractvalue {i32, i1} %res, 1
8798 br i1 %obit, label %overflow, label %normal
8800 '``llvm.uadd.with.overflow.*``' Intrinsics
8801 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8806 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8807 on any integer bit width.
8811 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8812 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8813 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8818 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8819 an unsigned addition of the two arguments, and indicate whether a carry
8820 occurred during the unsigned summation.
8825 The arguments (%a and %b) and the first element of the result structure
8826 may be of integer types of any bit width, but they must have the same
8827 bit width. The second element of the result structure must be of type
8828 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8834 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8835 an unsigned addition of the two arguments. They return a structure --- the
8836 first element of which is the sum, and the second element of which is a
8837 bit specifying if the unsigned summation resulted in a carry.
8842 .. code-block:: llvm
8844 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8845 %sum = extractvalue {i32, i1} %res, 0
8846 %obit = extractvalue {i32, i1} %res, 1
8847 br i1 %obit, label %carry, label %normal
8849 '``llvm.ssub.with.overflow.*``' Intrinsics
8850 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8855 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8856 on any integer bit width.
8860 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8861 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8862 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8867 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8868 a signed subtraction of the two arguments, and indicate whether an
8869 overflow occurred during the signed subtraction.
8874 The arguments (%a and %b) and the first element of the result structure
8875 may be of integer types of any bit width, but they must have the same
8876 bit width. The second element of the result structure must be of type
8877 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8883 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8884 a signed subtraction of the two arguments. They return a structure --- the
8885 first element of which is the subtraction, and the second element of
8886 which is a bit specifying if the signed subtraction resulted in an
8892 .. code-block:: llvm
8894 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8895 %sum = extractvalue {i32, i1} %res, 0
8896 %obit = extractvalue {i32, i1} %res, 1
8897 br i1 %obit, label %overflow, label %normal
8899 '``llvm.usub.with.overflow.*``' Intrinsics
8900 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8905 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8906 on any integer bit width.
8910 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8911 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8912 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8917 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8918 an unsigned subtraction of the two arguments, and indicate whether an
8919 overflow occurred during the unsigned subtraction.
8924 The arguments (%a and %b) and the first element of the result structure
8925 may be of integer types of any bit width, but they must have the same
8926 bit width. The second element of the result structure must be of type
8927 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8933 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8934 an unsigned subtraction of the two arguments. They return a structure ---
8935 the first element of which is the subtraction, and the second element of
8936 which is a bit specifying if the unsigned subtraction resulted in an
8942 .. code-block:: llvm
8944 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8945 %sum = extractvalue {i32, i1} %res, 0
8946 %obit = extractvalue {i32, i1} %res, 1
8947 br i1 %obit, label %overflow, label %normal
8949 '``llvm.smul.with.overflow.*``' Intrinsics
8950 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8955 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8956 on any integer bit width.
8960 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8961 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8962 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8967 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8968 a signed multiplication of the two arguments, and indicate whether an
8969 overflow occurred during the signed multiplication.
8974 The arguments (%a and %b) and the first element of the result structure
8975 may be of integer types of any bit width, but they must have the same
8976 bit width. The second element of the result structure must be of type
8977 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8983 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8984 a signed multiplication of the two arguments. They return a structure ---
8985 the first element of which is the multiplication, and the second element
8986 of which is a bit specifying if the signed multiplication resulted in an
8992 .. code-block:: llvm
8994 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8995 %sum = extractvalue {i32, i1} %res, 0
8996 %obit = extractvalue {i32, i1} %res, 1
8997 br i1 %obit, label %overflow, label %normal
8999 '``llvm.umul.with.overflow.*``' Intrinsics
9000 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9005 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
9006 on any integer bit width.
9010 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
9011 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9012 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
9017 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9018 a unsigned multiplication of the two arguments, and indicate whether an
9019 overflow occurred during the unsigned multiplication.
9024 The arguments (%a and %b) and the first element of the result structure
9025 may be of integer types of any bit width, but they must have the same
9026 bit width. The second element of the result structure must be of type
9027 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9033 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9034 an unsigned multiplication of the two arguments. They return a structure ---
9035 the first element of which is the multiplication, and the second
9036 element of which is a bit specifying if the unsigned multiplication
9037 resulted in an overflow.
9042 .. code-block:: llvm
9044 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9045 %sum = extractvalue {i32, i1} %res, 0
9046 %obit = extractvalue {i32, i1} %res, 1
9047 br i1 %obit, label %overflow, label %normal
9049 Specialised Arithmetic Intrinsics
9050 ---------------------------------
9052 '``llvm.fmuladd.*``' Intrinsic
9053 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9060 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
9061 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
9066 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
9067 expressions that can be fused if the code generator determines that (a) the
9068 target instruction set has support for a fused operation, and (b) that the
9069 fused operation is more efficient than the equivalent, separate pair of mul
9070 and add instructions.
9075 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
9076 multiplicands, a and b, and an addend c.
9085 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
9087 is equivalent to the expression a \* b + c, except that rounding will
9088 not be performed between the multiplication and addition steps if the
9089 code generator fuses the operations. Fusion is not guaranteed, even if
9090 the target platform supports it. If a fused multiply-add is required the
9091 corresponding llvm.fma.\* intrinsic function should be used
9092 instead. This never sets errno, just as '``llvm.fma.*``'.
9097 .. code-block:: llvm
9099 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
9101 Half Precision Floating Point Intrinsics
9102 ----------------------------------------
9104 For most target platforms, half precision floating point is a
9105 storage-only format. This means that it is a dense encoding (in memory)
9106 but does not support computation in the format.
9108 This means that code must first load the half-precision floating point
9109 value as an i16, then convert it to float with
9110 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
9111 then be performed on the float value (including extending to double
9112 etc). To store the value back to memory, it is first converted to float
9113 if needed, then converted to i16 with
9114 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
9117 .. _int_convert_to_fp16:
9119 '``llvm.convert.to.fp16``' Intrinsic
9120 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9127 declare i16 @llvm.convert.to.fp16.f32(float %a)
9128 declare i16 @llvm.convert.to.fp16.f64(double %a)
9133 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9134 conventional floating point type to half precision floating point format.
9139 The intrinsic function contains single argument - the value to be
9145 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9146 conventional floating point format to half precision floating point format. The
9147 return value is an ``i16`` which contains the converted number.
9152 .. code-block:: llvm
9154 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
9155 store i16 %res, i16* @x, align 2
9157 .. _int_convert_from_fp16:
9159 '``llvm.convert.from.fp16``' Intrinsic
9160 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9167 declare float @llvm.convert.from.fp16.f32(i16 %a)
9168 declare double @llvm.convert.from.fp16.f64(i16 %a)
9173 The '``llvm.convert.from.fp16``' intrinsic function performs a
9174 conversion from half precision floating point format to single precision
9175 floating point format.
9180 The intrinsic function contains single argument - the value to be
9186 The '``llvm.convert.from.fp16``' intrinsic function performs a
9187 conversion from half single precision floating point format to single
9188 precision floating point format. The input half-float value is
9189 represented by an ``i16`` value.
9194 .. code-block:: llvm
9196 %a = load i16* @x, align 2
9197 %res = call float @llvm.convert.from.fp16(i16 %a)
9202 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
9203 prefix), are described in the `LLVM Source Level
9204 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
9207 Exception Handling Intrinsics
9208 -----------------------------
9210 The LLVM exception handling intrinsics (which all start with
9211 ``llvm.eh.`` prefix), are described in the `LLVM Exception
9212 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
9216 Trampoline Intrinsics
9217 ---------------------
9219 These intrinsics make it possible to excise one parameter, marked with
9220 the :ref:`nest <nest>` attribute, from a function. The result is a
9221 callable function pointer lacking the nest parameter - the caller does
9222 not need to provide a value for it. Instead, the value to use is stored
9223 in advance in a "trampoline", a block of memory usually allocated on the
9224 stack, which also contains code to splice the nest value into the
9225 argument list. This is used to implement the GCC nested function address
9228 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
9229 then the resulting function pointer has signature ``i32 (i32, i32)*``.
9230 It can be created as follows:
9232 .. code-block:: llvm
9234 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
9235 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
9236 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
9237 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
9238 %fp = bitcast i8* %p to i32 (i32, i32)*
9240 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
9241 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
9245 '``llvm.init.trampoline``' Intrinsic
9246 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9253 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
9258 This fills the memory pointed to by ``tramp`` with executable code,
9259 turning it into a trampoline.
9264 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
9265 pointers. The ``tramp`` argument must point to a sufficiently large and
9266 sufficiently aligned block of memory; this memory is written to by the
9267 intrinsic. Note that the size and the alignment are target-specific -
9268 LLVM currently provides no portable way of determining them, so a
9269 front-end that generates this intrinsic needs to have some
9270 target-specific knowledge. The ``func`` argument must hold a function
9271 bitcast to an ``i8*``.
9276 The block of memory pointed to by ``tramp`` is filled with target
9277 dependent code, turning it into a function. Then ``tramp`` needs to be
9278 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
9279 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
9280 function's signature is the same as that of ``func`` with any arguments
9281 marked with the ``nest`` attribute removed. At most one such ``nest``
9282 argument is allowed, and it must be of pointer type. Calling the new
9283 function is equivalent to calling ``func`` with the same argument list,
9284 but with ``nval`` used for the missing ``nest`` argument. If, after
9285 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
9286 modified, then the effect of any later call to the returned function
9287 pointer is undefined.
9291 '``llvm.adjust.trampoline``' Intrinsic
9292 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9299 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
9304 This performs any required machine-specific adjustment to the address of
9305 a trampoline (passed as ``tramp``).
9310 ``tramp`` must point to a block of memory which already has trampoline
9311 code filled in by a previous call to
9312 :ref:`llvm.init.trampoline <int_it>`.
9317 On some architectures the address of the code to be executed needs to be
9318 different than the address where the trampoline is actually stored. This
9319 intrinsic returns the executable address corresponding to ``tramp``
9320 after performing the required machine specific adjustments. The pointer
9321 returned can then be :ref:`bitcast and executed <int_trampoline>`.
9323 Masked Vector Load and Store Intrinsics
9324 ---------------------------------------
9326 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.
9330 '``llvm.masked.load.*``' Intrinsics
9331 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9335 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
9339 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
9340 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
9345 Reads a vector from memory according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes. The masked-off lanes in the result vector are taken from the corresponding lanes in the passthru operand.
9351 The first operand is the base pointer for the load. The second operand is the alignment of the source location. It must be a constant integer value. The third operand, mask, is a vector of boolean 'i1' values with the same number of elements as the return type. The fourth is a pass-through value that is used to fill the masked-off lanes of the result. The return type, underlying type of the base pointer and the type of passthru operand are the same vector types.
9357 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.
9358 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.
9363 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
9365 ;; The result of the two following instructions is identical aside from potential memory access exception
9366 %loadlal = load <16 x float>* %ptr, align 4
9367 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
9371 '``llvm.masked.store.*``' Intrinsics
9372 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9376 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
9380 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
9381 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
9386 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.
9391 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.
9397 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.
9398 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.
9402 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
9404 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
9405 %oldval = load <16 x float>* %ptr, align 4
9406 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
9407 store <16 x float> %res, <16 x float>* %ptr, align 4
9413 This class of intrinsics provides information about the lifetime of
9414 memory objects and ranges where variables are immutable.
9418 '``llvm.lifetime.start``' Intrinsic
9419 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9426 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
9431 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
9437 The first argument is a constant integer representing the size of the
9438 object, or -1 if it is variable sized. The second argument is a pointer
9444 This intrinsic indicates that before this point in the code, the value
9445 of the memory pointed to by ``ptr`` is dead. This means that it is known
9446 to never be used and has an undefined value. A load from the pointer
9447 that precedes this intrinsic can be replaced with ``'undef'``.
9451 '``llvm.lifetime.end``' Intrinsic
9452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9459 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
9464 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
9470 The first argument is a constant integer representing the size of the
9471 object, or -1 if it is variable sized. The second argument is a pointer
9477 This intrinsic indicates that after this point in the code, the value of
9478 the memory pointed to by ``ptr`` is dead. This means that it is known to
9479 never be used and has an undefined value. Any stores into the memory
9480 object following this intrinsic may be removed as dead.
9482 '``llvm.invariant.start``' Intrinsic
9483 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9490 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
9495 The '``llvm.invariant.start``' intrinsic specifies that the contents of
9496 a memory object will not change.
9501 The first argument is a constant integer representing the size of the
9502 object, or -1 if it is variable sized. The second argument is a pointer
9508 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
9509 the return value, the referenced memory location is constant and
9512 '``llvm.invariant.end``' Intrinsic
9513 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9520 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
9525 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
9526 memory object are mutable.
9531 The first argument is the matching ``llvm.invariant.start`` intrinsic.
9532 The second argument is a constant integer representing the size of the
9533 object, or -1 if it is variable sized and the third argument is a
9534 pointer to the object.
9539 This intrinsic indicates that the memory is mutable again.
9544 This class of intrinsics is designed to be generic and has no specific
9547 '``llvm.var.annotation``' Intrinsic
9548 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9555 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9560 The '``llvm.var.annotation``' intrinsic.
9565 The first argument is a pointer to a value, the second is a pointer to a
9566 global string, the third is a pointer to a global string which is the
9567 source file name, and the last argument is the line number.
9572 This intrinsic allows annotation of local variables with arbitrary
9573 strings. This can be useful for special purpose optimizations that want
9574 to look for these annotations. These have no other defined use; they are
9575 ignored by code generation and optimization.
9577 '``llvm.ptr.annotation.*``' Intrinsic
9578 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9583 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
9584 pointer to an integer of any width. *NOTE* you must specify an address space for
9585 the pointer. The identifier for the default address space is the integer
9590 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9591 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
9592 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
9593 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
9594 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
9599 The '``llvm.ptr.annotation``' intrinsic.
9604 The first argument is a pointer to an integer value of arbitrary bitwidth
9605 (result of some expression), the second is a pointer to a global string, the
9606 third is a pointer to a global string which is the source file name, and the
9607 last argument is the line number. It returns the value of the first argument.
9612 This intrinsic allows annotation of a pointer to an integer with arbitrary
9613 strings. This can be useful for special purpose optimizations that want to look
9614 for these annotations. These have no other defined use; they are ignored by code
9615 generation and optimization.
9617 '``llvm.annotation.*``' Intrinsic
9618 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9623 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
9624 any integer bit width.
9628 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
9629 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
9630 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
9631 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
9632 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
9637 The '``llvm.annotation``' intrinsic.
9642 The first argument is an integer value (result of some expression), the
9643 second is a pointer to a global string, the third is a pointer to a
9644 global string which is the source file name, and the last argument is
9645 the line number. It returns the value of the first argument.
9650 This intrinsic allows annotations to be put on arbitrary expressions
9651 with arbitrary strings. This can be useful for special purpose
9652 optimizations that want to look for these annotations. These have no
9653 other defined use; they are ignored by code generation and optimization.
9655 '``llvm.trap``' Intrinsic
9656 ^^^^^^^^^^^^^^^^^^^^^^^^^
9663 declare void @llvm.trap() noreturn nounwind
9668 The '``llvm.trap``' intrinsic.
9678 This intrinsic is lowered to the target dependent trap instruction. If
9679 the target does not have a trap instruction, this intrinsic will be
9680 lowered to a call of the ``abort()`` function.
9682 '``llvm.debugtrap``' Intrinsic
9683 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9690 declare void @llvm.debugtrap() nounwind
9695 The '``llvm.debugtrap``' intrinsic.
9705 This intrinsic is lowered to code which is intended to cause an
9706 execution trap with the intention of requesting the attention of a
9709 '``llvm.stackprotector``' Intrinsic
9710 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9717 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
9722 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
9723 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
9724 is placed on the stack before local variables.
9729 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
9730 The first argument is the value loaded from the stack guard
9731 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
9732 enough space to hold the value of the guard.
9737 This intrinsic causes the prologue/epilogue inserter to force the position of
9738 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
9739 to ensure that if a local variable on the stack is overwritten, it will destroy
9740 the value of the guard. When the function exits, the guard on the stack is
9741 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
9742 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
9743 calling the ``__stack_chk_fail()`` function.
9745 '``llvm.stackprotectorcheck``' Intrinsic
9746 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9753 declare void @llvm.stackprotectorcheck(i8** <guard>)
9758 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
9759 created stack protector and if they are not equal calls the
9760 ``__stack_chk_fail()`` function.
9765 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
9766 the variable ``@__stack_chk_guard``.
9771 This intrinsic is provided to perform the stack protector check by comparing
9772 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
9773 values do not match call the ``__stack_chk_fail()`` function.
9775 The reason to provide this as an IR level intrinsic instead of implementing it
9776 via other IR operations is that in order to perform this operation at the IR
9777 level without an intrinsic, one would need to create additional basic blocks to
9778 handle the success/failure cases. This makes it difficult to stop the stack
9779 protector check from disrupting sibling tail calls in Codegen. With this
9780 intrinsic, we are able to generate the stack protector basic blocks late in
9781 codegen after the tail call decision has occurred.
9783 '``llvm.objectsize``' Intrinsic
9784 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9791 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
9792 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
9797 The ``llvm.objectsize`` intrinsic is designed to provide information to
9798 the optimizers to determine at compile time whether a) an operation
9799 (like memcpy) will overflow a buffer that corresponds to an object, or
9800 b) that a runtime check for overflow isn't necessary. An object in this
9801 context means an allocation of a specific class, structure, array, or
9807 The ``llvm.objectsize`` intrinsic takes two arguments. The first
9808 argument is a pointer to or into the ``object``. The second argument is
9809 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
9810 or -1 (if false) when the object size is unknown. The second argument
9811 only accepts constants.
9816 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
9817 the size of the object concerned. If the size cannot be determined at
9818 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
9819 on the ``min`` argument).
9821 '``llvm.expect``' Intrinsic
9822 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9827 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
9832 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
9833 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9834 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9839 The ``llvm.expect`` intrinsic provides information about expected (the
9840 most probable) value of ``val``, which can be used by optimizers.
9845 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
9846 a value. The second argument is an expected value, this needs to be a
9847 constant value, variables are not allowed.
9852 This intrinsic is lowered to the ``val``.
9854 '``llvm.assume``' Intrinsic
9855 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9862 declare void @llvm.assume(i1 %cond)
9867 The ``llvm.assume`` allows the optimizer to assume that the provided
9868 condition is true. This information can then be used in simplifying other parts
9874 The condition which the optimizer may assume is always true.
9879 The intrinsic allows the optimizer to assume that the provided condition is
9880 always true whenever the control flow reaches the intrinsic call. No code is
9881 generated for this intrinsic, and instructions that contribute only to the
9882 provided condition are not used for code generation. If the condition is
9883 violated during execution, the behavior is undefined.
9885 Note that the optimizer might limit the transformations performed on values
9886 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
9887 only used to form the intrinsic's input argument. This might prove undesirable
9888 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
9889 sufficient overall improvement in code quality. For this reason,
9890 ``llvm.assume`` should not be used to document basic mathematical invariants
9891 that the optimizer can otherwise deduce or facts that are of little use to the
9894 '``llvm.donothing``' Intrinsic
9895 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9902 declare void @llvm.donothing() nounwind readnone
9907 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
9908 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
9909 with an invoke instruction.
9919 This intrinsic does nothing, and it's removed by optimizers and ignored
9922 Stack Map Intrinsics
9923 --------------------
9925 LLVM provides experimental intrinsics to support runtime patching
9926 mechanisms commonly desired in dynamic language JITs. These intrinsics
9927 are described in :doc:`StackMaps`.