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 which 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.
83 #. Unnamed values are represented as an unsigned numeric value with
84 their prefix. For example, ``%12``, ``@2``, ``%44``.
85 #. Constants, which are described in the section Constants_ below.
87 LLVM requires that values start with a prefix for two reasons: Compilers
88 don't need to worry about name clashes with reserved words, and the set
89 of reserved words may be expanded in the future without penalty.
90 Additionally, unnamed identifiers allow a compiler to quickly come up
91 with a temporary variable without having to avoid symbol table
94 Reserved words in LLVM are very similar to reserved words in other
95 languages. There are keywords for different opcodes ('``add``',
96 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
97 '``i32``', etc...), and others. These reserved words cannot conflict
98 with variable names, because none of them start with a prefix character
101 Here is an example of LLVM code to multiply the integer variable
108 %result = mul i32 %X, 8
110 After strength reduction:
114 %result = shl i32 %X, 3
120 %0 = add i32 %X, %X ; yields i32:%0
121 %1 = add i32 %0, %0 ; yields i32:%1
122 %result = add i32 %1, %1
124 This last way of multiplying ``%X`` by 8 illustrates several important
125 lexical features of LLVM:
127 #. Comments are delimited with a '``;``' and go until the end of line.
128 #. Unnamed temporaries are created when the result of a computation is
129 not assigned to a named value.
130 #. Unnamed temporaries are numbered sequentially (using a per-function
131 incrementing counter, starting with 0). Note that basic blocks are
132 included in this numbering. For example, if the entry basic block is not
133 given a label name, then it will get number 0.
135 It also shows a convention that we follow in this document. When
136 demonstrating instructions, we will follow an instruction with a comment
137 that defines the type and name of value produced.
145 LLVM programs are composed of ``Module``'s, each of which is a
146 translation unit of the input programs. Each module consists of
147 functions, global variables, and symbol table entries. Modules may be
148 combined together with the LLVM linker, which merges function (and
149 global variable) definitions, resolves forward declarations, and merges
150 symbol table entries. Here is an example of the "hello world" module:
154 ; Declare the string constant as a global constant.
155 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
157 ; External declaration of the puts function
158 declare i32 @puts(i8* nocapture) nounwind
160 ; Definition of main function
161 define i32 @main() { ; i32()*
162 ; Convert [13 x i8]* to i8 *...
163 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
165 ; Call puts function to write out the string to stdout.
166 call i32 @puts(i8* %cast210)
171 !1 = metadata !{i32 42}
174 This example is made up of a :ref:`global variable <globalvars>` named
175 "``.str``", an external declaration of the "``puts``" function, a
176 :ref:`function definition <functionstructure>` for "``main``" and
177 :ref:`named metadata <namedmetadatastructure>` "``foo``".
179 In general, a module is made up of a list of global values (where both
180 functions and global variables are global values). Global values are
181 represented by a pointer to a memory location (in this case, a pointer
182 to an array of char, and a pointer to a function), and have one of the
183 following :ref:`linkage types <linkage>`.
190 All Global Variables and Functions have one of the following types of
194 Global values with "``private``" linkage are only directly
195 accessible by objects in the current module. In particular, linking
196 code into a module with an private global value may cause the
197 private to be renamed as necessary to avoid collisions. Because the
198 symbol is private to the module, all references can be updated. This
199 doesn't show up in any symbol table in the object file.
201 Similar to private, but the value shows as a local symbol
202 (``STB_LOCAL`` in the case of ELF) in the object file. This
203 corresponds to the notion of the '``static``' keyword in C.
204 ``available_externally``
205 Globals with "``available_externally``" linkage are never emitted
206 into the object file corresponding to the LLVM module. They exist to
207 allow inlining and other optimizations to take place given knowledge
208 of the definition of the global, which is known to be somewhere
209 outside the module. Globals with ``available_externally`` linkage
210 are allowed to be discarded at will, and are otherwise the same as
211 ``linkonce_odr``. This linkage type is only allowed on definitions,
214 Globals with "``linkonce``" linkage are merged with other globals of
215 the same name when linkage occurs. This can be used to implement
216 some forms of inline functions, templates, or other code which must
217 be generated in each translation unit that uses it, but where the
218 body may be overridden with a more definitive definition later.
219 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
220 that ``linkonce`` linkage does not actually allow the optimizer to
221 inline the body of this function into callers because it doesn't
222 know if this definition of the function is the definitive definition
223 within the program or whether it will be overridden by a stronger
224 definition. To enable inlining and other optimizations, use
225 "``linkonce_odr``" linkage.
227 "``weak``" linkage has the same merging semantics as ``linkonce``
228 linkage, except that unreferenced globals with ``weak`` linkage may
229 not be discarded. This is used for globals that are declared "weak"
232 "``common``" linkage is most similar to "``weak``" linkage, but they
233 are used for tentative definitions in C, such as "``int X;``" at
234 global scope. Symbols with "``common``" linkage are merged in the
235 same way as ``weak symbols``, and they may not be deleted if
236 unreferenced. ``common`` symbols may not have an explicit section,
237 must have a zero initializer, and may not be marked
238 ':ref:`constant <globalvars>`'. Functions and aliases may not have
241 .. _linkage_appending:
244 "``appending``" linkage may only be applied to global variables of
245 pointer to array type. When two global variables with appending
246 linkage are linked together, the two global arrays are appended
247 together. This is the LLVM, typesafe, equivalent of having the
248 system linker append together "sections" with identical names when
251 The semantics of this linkage follow the ELF object file model: the
252 symbol is weak until linked, if not linked, the symbol becomes null
253 instead of being an undefined reference.
254 ``linkonce_odr``, ``weak_odr``
255 Some languages allow differing globals to be merged, such as two
256 functions with different semantics. Other languages, such as
257 ``C++``, ensure that only equivalent globals are ever merged (the
258 "one definition rule" --- "ODR"). Such languages can use the
259 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
260 global will only be merged with equivalent globals. These linkage
261 types are otherwise the same as their non-``odr`` versions.
263 If none of the above identifiers are used, the global is externally
264 visible, meaning that it participates in linkage and can be used to
265 resolve external symbol references.
267 It is illegal for a function *declaration* to have any linkage type
268 other than ``external`` or ``extern_weak``.
275 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
276 :ref:`invokes <i_invoke>` can all have an optional calling convention
277 specified for the call. The calling convention of any pair of dynamic
278 caller/callee must match, or the behavior of the program is undefined.
279 The following calling conventions are supported by LLVM, and more may be
282 "``ccc``" - The C calling convention
283 This calling convention (the default if no other calling convention
284 is specified) matches the target C calling conventions. This calling
285 convention supports varargs function calls and tolerates some
286 mismatch in the declared prototype and implemented declaration of
287 the function (as does normal C).
288 "``fastcc``" - The fast calling convention
289 This calling convention attempts to make calls as fast as possible
290 (e.g. by passing things in registers). This calling convention
291 allows the target to use whatever tricks it wants to produce fast
292 code for the target, without having to conform to an externally
293 specified ABI (Application Binary Interface). `Tail calls can only
294 be optimized when this, the GHC or the HiPE convention is
295 used. <CodeGenerator.html#id80>`_ This calling convention does not
296 support varargs and requires the prototype of all callees to exactly
297 match the prototype of the function definition.
298 "``coldcc``" - The cold calling convention
299 This calling convention attempts to make code in the caller as
300 efficient as possible under the assumption that the call is not
301 commonly executed. As such, these calls often preserve all registers
302 so that the call does not break any live ranges in the caller side.
303 This calling convention does not support varargs and requires the
304 prototype of all callees to exactly match the prototype of the
305 function definition. Furthermore the inliner doesn't consider such function
307 "``cc 10``" - GHC convention
308 This calling convention has been implemented specifically for use by
309 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
310 It passes everything in registers, going to extremes to achieve this
311 by disabling callee save registers. This calling convention should
312 not be used lightly but only for specific situations such as an
313 alternative to the *register pinning* performance technique often
314 used when implementing functional programming languages. At the
315 moment only X86 supports this convention and it has the following
318 - On *X86-32* only supports up to 4 bit type parameters. No
319 floating point types are supported.
320 - On *X86-64* only supports up to 10 bit type parameters and 6
321 floating point parameters.
323 This calling convention supports `tail call
324 optimization <CodeGenerator.html#id80>`_ but requires both the
325 caller and callee are using it.
326 "``cc 11``" - The HiPE calling convention
327 This calling convention has been implemented specifically for use by
328 the `High-Performance Erlang
329 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
330 native code compiler of the `Ericsson's Open Source Erlang/OTP
331 system <http://www.erlang.org/download.shtml>`_. It uses more
332 registers for argument passing than the ordinary C calling
333 convention and defines no callee-saved registers. The calling
334 convention properly supports `tail call
335 optimization <CodeGenerator.html#id80>`_ but requires that both the
336 caller and the callee use it. It uses a *register pinning*
337 mechanism, similar to GHC's convention, for keeping frequently
338 accessed runtime components pinned to specific hardware registers.
339 At the moment only X86 supports this convention (both 32 and 64
341 "``webkit_jscc``" - WebKit's JavaScript calling convention
342 This calling convention has been implemented for `WebKit FTL JIT
343 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
344 stack right to left (as cdecl does), and returns a value in the
345 platform's customary return register.
346 "``anyregcc``" - Dynamic calling convention for code patching
347 This is a special convention that supports patching an arbitrary code
348 sequence in place of a call site. This convention forces the call
349 arguments into registers but allows them to be dynamcially
350 allocated. This can currently only be used with calls to
351 llvm.experimental.patchpoint because only this intrinsic records
352 the location of its arguments in a side table. See :doc:`StackMaps`.
353 "``preserve_mostcc``" - The `PreserveMost` calling convention
354 This calling convention attempts to make the code in the caller as little
355 intrusive as possible. This calling convention behaves identical to the `C`
356 calling convention on how arguments and return values are passed, but it
357 uses a different set of caller/callee-saved registers. This alleviates the
358 burden of saving and recovering a large register set before and after the
359 call in the caller. If the arguments are passed in callee-saved registers,
360 then they will be preserved by the callee across the call. This doesn't
361 apply for values returned in callee-saved registers.
363 - On X86-64 the callee preserves all general purpose registers, except for
364 R11. R11 can be used as a scratch register. Floating-point registers
365 (XMMs/YMMs) are not preserved and need to be saved by the caller.
367 The idea behind this convention is to support calls to runtime functions
368 that have a hot path and a cold path. The hot path is usually a small piece
369 of code that doesn't many registers. The cold path might need to call out to
370 another function and therefore only needs to preserve the caller-saved
371 registers, which haven't already been saved by the caller. The
372 `PreserveMost` calling convention is very similar to the `cold` calling
373 convention in terms of caller/callee-saved registers, but they are used for
374 different types of function calls. `coldcc` is for function calls that are
375 rarely executed, whereas `preserve_mostcc` function calls are intended to be
376 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
377 doesn't prevent the inliner from inlining the function call.
379 This calling convention will be used by a future version of the ObjectiveC
380 runtime and should therefore still be considered experimental at this time.
381 Although this convention was created to optimize certain runtime calls to
382 the ObjectiveC runtime, it is not limited to this runtime and might be used
383 by other runtimes in the future too. The current implementation only
384 supports X86-64, but the intention is to support more architectures in the
386 "``preserve_allcc``" - The `PreserveAll` calling convention
387 This calling convention attempts to make the code in the caller even less
388 intrusive than the `PreserveMost` calling convention. This calling
389 convention also behaves identical to the `C` calling convention on how
390 arguments and return values are passed, but it uses a different set of
391 caller/callee-saved registers. This removes the burden of saving and
392 recovering a large register set before and after the call in the caller. If
393 the arguments are passed in callee-saved registers, then they will be
394 preserved by the callee across the call. This doesn't apply for values
395 returned in callee-saved registers.
397 - On X86-64 the callee preserves all general purpose registers, except for
398 R11. R11 can be used as a scratch register. Furthermore it also preserves
399 all floating-point registers (XMMs/YMMs).
401 The idea behind this convention is to support calls to runtime functions
402 that don't need to call out to any other functions.
404 This calling convention, like the `PreserveMost` calling convention, will be
405 used by a future version of the ObjectiveC runtime and should be considered
406 experimental at this time.
407 "``cc <n>``" - Numbered convention
408 Any calling convention may be specified by number, allowing
409 target-specific calling conventions to be used. Target specific
410 calling conventions start at 64.
412 More calling conventions can be added/defined on an as-needed basis, to
413 support Pascal conventions or any other well-known target-independent
416 .. _visibilitystyles:
421 All Global Variables and Functions have one of the following visibility
424 "``default``" - Default style
425 On targets that use the ELF object file format, default visibility
426 means that the declaration is visible to other modules and, in
427 shared libraries, means that the declared entity may be overridden.
428 On Darwin, default visibility means that the declaration is visible
429 to other modules. Default visibility corresponds to "external
430 linkage" in the language.
431 "``hidden``" - Hidden style
432 Two declarations of an object with hidden visibility refer to the
433 same object if they are in the same shared object. Usually, hidden
434 visibility indicates that the symbol will not be placed into the
435 dynamic symbol table, so no other module (executable or shared
436 library) can reference it directly.
437 "``protected``" - Protected style
438 On ELF, protected visibility indicates that the symbol will be
439 placed in the dynamic symbol table, but that references within the
440 defining module will bind to the local symbol. That is, the symbol
441 cannot be overridden by another module.
443 A symbol with ``internal`` or ``private`` linkage must have ``default``
451 All Global Variables, Functions and Aliases can have one of the following
455 "``dllimport``" causes the compiler to reference a function or variable via
456 a global pointer to a pointer that is set up by the DLL exporting the
457 symbol. On Microsoft Windows targets, the pointer name is formed by
458 combining ``__imp_`` and the function or variable name.
460 "``dllexport``" causes the compiler to provide a global pointer to a pointer
461 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
462 Microsoft Windows targets, the pointer name is formed by combining
463 ``__imp_`` and the function or variable name. Since this storage class
464 exists for defining a dll interface, the compiler, assembler and linker know
465 it is externally referenced and must refrain from deleting the symbol.
469 Thread Local Storage Models
470 ---------------------------
472 A variable may be defined as ``thread_local``, which means that it will
473 not be shared by threads (each thread will have a separated copy of the
474 variable). Not all targets support thread-local variables. Optionally, a
475 TLS model may be specified:
478 For variables that are only used within the current shared library.
480 For variables in modules that will not be loaded dynamically.
482 For variables defined in the executable and only used within it.
484 If no explicit model is given, the "general dynamic" model is used.
486 The models correspond to the ELF TLS models; see `ELF Handling For
487 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
488 more information on under which circumstances the different models may
489 be used. The target may choose a different TLS model if the specified
490 model is not supported, or if a better choice of model can be made.
492 A model can also be specified in a alias, but then it only governs how
493 the alias is accessed. It will not have any effect in the aliasee.
500 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
501 types <t_struct>`. Literal types are uniqued structurally, but identified types
502 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
503 to forward declare a type which is not yet available.
505 An example of a identified structure specification is:
509 %mytype = type { %mytype*, i32 }
511 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
512 literal types are uniqued in recent versions of LLVM.
519 Global variables define regions of memory allocated at compilation time
522 Global variables definitions must be initialized.
524 Global variables in other translation units can also be declared, in which
525 case they don't have an initializer.
527 Either global variable definitions or declarations may have an explicit section
528 to be placed in and may have an optional explicit alignment specified.
530 A variable may be defined as a global ``constant``, which indicates that
531 the contents of the variable will **never** be modified (enabling better
532 optimization, allowing the global data to be placed in the read-only
533 section of an executable, etc). Note that variables that need runtime
534 initialization cannot be marked ``constant`` as there is a store to the
537 LLVM explicitly allows *declarations* of global variables to be marked
538 constant, even if the final definition of the global is not. This
539 capability can be used to enable slightly better optimization of the
540 program, but requires the language definition to guarantee that
541 optimizations based on the 'constantness' are valid for the translation
542 units that do not include the definition.
544 As SSA values, global variables define pointer values that are in scope
545 (i.e. they dominate) all basic blocks in the program. Global variables
546 always define a pointer to their "content" type because they describe a
547 region of memory, and all memory objects in LLVM are accessed through
550 Global variables can be marked with ``unnamed_addr`` which indicates
551 that the address is not significant, only the content. Constants marked
552 like this can be merged with other constants if they have the same
553 initializer. Note that a constant with significant address *can* be
554 merged with a ``unnamed_addr`` constant, the result being a constant
555 whose address is significant.
557 A global variable may be declared to reside in a target-specific
558 numbered address space. For targets that support them, address spaces
559 may affect how optimizations are performed and/or what target
560 instructions are used to access the variable. The default address space
561 is zero. The address space qualifier must precede any other attributes.
563 LLVM allows an explicit section to be specified for globals. If the
564 target supports it, it will emit globals to the section specified.
565 Additionally, the global can placed in a comdat if the target has the necessary
568 By default, global initializers are optimized by assuming that global
569 variables defined within the module are not modified from their
570 initial values before the start of the global initializer. This is
571 true even for variables potentially accessible from outside the
572 module, including those with external linkage or appearing in
573 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
574 by marking the variable with ``externally_initialized``.
576 An explicit alignment may be specified for a global, which must be a
577 power of 2. If not present, or if the alignment is set to zero, the
578 alignment of the global is set by the target to whatever it feels
579 convenient. If an explicit alignment is specified, the global is forced
580 to have exactly that alignment. Targets and optimizers are not allowed
581 to over-align the global if the global has an assigned section. In this
582 case, the extra alignment could be observable: for example, code could
583 assume that the globals are densely packed in their section and try to
584 iterate over them as an array, alignment padding would break this
585 iteration. The maximum alignment is ``1 << 29``.
587 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
589 Variables and aliasaes can have a
590 :ref:`Thread Local Storage Model <tls_model>`.
594 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
595 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
596 <global | constant> <Type> [<InitializerConstant>]
597 [, section "name"] [, align <Alignment>]
599 For example, the following defines a global in a numbered address space
600 with an initializer, section, and alignment:
604 @G = addrspace(5) constant float 1.0, section "foo", align 4
606 The following example just declares a global variable
610 @G = external global i32
612 The following example defines a thread-local global with the
613 ``initialexec`` TLS model:
617 @G = thread_local(initialexec) global i32 0, align 4
619 .. _functionstructure:
624 LLVM function definitions consist of the "``define``" keyword, an
625 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
626 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
627 an optional :ref:`calling convention <callingconv>`,
628 an optional ``unnamed_addr`` attribute, a return type, an optional
629 :ref:`parameter attribute <paramattrs>` for the return type, a function
630 name, a (possibly empty) argument list (each with optional :ref:`parameter
631 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
632 an optional section, an optional alignment,
633 an optional :ref:`comdat <langref_comdats>`,
634 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
635 curly brace, a list of basic blocks, and a closing curly brace.
637 LLVM function declarations consist of the "``declare``" keyword, an
638 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
639 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
640 an optional :ref:`calling convention <callingconv>`,
641 an optional ``unnamed_addr`` attribute, a return type, an optional
642 :ref:`parameter attribute <paramattrs>` for the return type, a function
643 name, a possibly empty list of arguments, an optional alignment, an optional
644 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
646 A function definition contains a list of basic blocks, forming the CFG (Control
647 Flow Graph) for the function. Each basic block may optionally start with a label
648 (giving the basic block a symbol table entry), contains a list of instructions,
649 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
650 function return). If an explicit label is not provided, a block is assigned an
651 implicit numbered label, using the next value from the same counter as used for
652 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
653 entry block does not have an explicit label, it will be assigned label "%0",
654 then the first unnamed temporary in that block will be "%1", etc.
656 The first basic block in a function is special in two ways: it is
657 immediately executed on entrance to the function, and it is not allowed
658 to have predecessor basic blocks (i.e. there can not be any branches to
659 the entry block of a function). Because the block can have no
660 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
662 LLVM allows an explicit section to be specified for functions. If the
663 target supports it, it will emit functions to the section specified.
664 Additionally, the function can placed in a COMDAT.
666 An explicit alignment may be specified for a function. If not present,
667 or if the alignment is set to zero, the alignment of the function is set
668 by the target to whatever it feels convenient. If an explicit alignment
669 is specified, the function is forced to have at least that much
670 alignment. All alignments must be a power of 2.
672 If the ``unnamed_addr`` attribute is given, the address is know to not
673 be significant and two identical functions can be merged.
677 define [linkage] [visibility] [DLLStorageClass]
679 <ResultType> @<FunctionName> ([argument list])
680 [unnamed_addr] [fn Attrs] [section "name"] [comdat $<ComdatName>]
681 [align N] [gc] [prefix Constant] { ... }
688 Aliases, unlike function or variables, don't create any new data. They
689 are just a new symbol and metadata for an existing position.
691 Aliases have a name and an aliasee that is either a global value or a
694 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
695 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
696 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
700 @<Name> = [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias [Linkage] <AliaseeTy> @<Aliasee>
702 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
703 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
704 might not correctly handle dropping a weak symbol that is aliased.
706 Alias that are not ``unnamed_addr`` are guaranteed to have the same address as
707 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
710 Since aliases are only a second name, some restrictions apply, of which
711 some can only be checked when producing an object file:
713 * The expression defining the aliasee must be computable at assembly
714 time. Since it is just a name, no relocations can be used.
716 * No alias in the expression can be weak as the possibility of the
717 intermediate alias being overridden cannot be represented in an
720 * No global value in the expression can be a declaration, since that
721 would require a relocation, which is not possible.
728 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
730 Comdats have a name which represents the COMDAT key. All global objects which
731 specify this key will only end up in the final object file if the linker chooses
732 that key over some other key. Aliases are placed in the same COMDAT that their
733 aliasee computes to, if any.
735 Comdats have a selection kind to provide input on how the linker should
736 choose between keys in two different object files.
740 $<Name> = comdat SelectionKind
742 The selection kind must be one of the following:
745 The linker may choose any COMDAT key, the choice is arbitrary.
747 The linker may choose any COMDAT key but the sections must contain the
750 The linker will choose the section containing the largest COMDAT key.
752 The linker requires that only section with this COMDAT key exist.
754 The linker may choose any COMDAT key but the sections must contain the
757 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
758 ``any`` as a selection kind.
760 Here is an example of a COMDAT group where a function will only be selected if
761 the COMDAT key's section is the largest:
765 $foo = comdat largest
766 @foo = global i32 2, comdat $foo
768 define void @bar() comdat $foo {
772 In a COFF object file, this will create a COMDAT section with selection kind
773 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
774 and another COMDAT section with selection kind
775 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
776 section and contains the contents of the ``@baz`` symbol.
778 There are some restrictions on the properties of the global object.
779 It, or an alias to it, must have the same name as the COMDAT group when
781 The contents and size of this object may be used during link-time to determine
782 which COMDAT groups get selected depending on the selection kind.
783 Because the name of the object must match the name of the COMDAT group, the
784 linkage of the global object must not be local; local symbols can get renamed
785 if a collision occurs in the symbol table.
787 The combined use of COMDATS and section attributes may yield surprising results.
794 @g1 = global i32 42, section "sec", comdat $foo
795 @g2 = global i32 42, section "sec", comdat $bar
797 From the object file perspective, this requires the creation of two sections
798 with the same name. This is necessary because both globals belong to different
799 COMDAT groups and COMDATs, at the object file level, are represented by
802 Note that certain IR constructs like global variables and functions may create
803 COMDATs in the object file in addition to any which are specified using COMDAT
804 IR. This arises, for example, when a global variable has linkonce_odr linkage.
806 .. _namedmetadatastructure:
811 Named metadata is a collection of metadata. :ref:`Metadata
812 nodes <metadata>` (but not metadata strings) are the only valid
813 operands for a named metadata.
817 ; Some unnamed metadata nodes, which are referenced by the named metadata.
818 !0 = metadata !{metadata !"zero"}
819 !1 = metadata !{metadata !"one"}
820 !2 = metadata !{metadata !"two"}
822 !name = !{!0, !1, !2}
829 The return type and each parameter of a function type may have a set of
830 *parameter attributes* associated with them. Parameter attributes are
831 used to communicate additional information about the result or
832 parameters of a function. Parameter attributes are considered to be part
833 of the function, not of the function type, so functions with different
834 parameter attributes can have the same function type.
836 Parameter attributes are simple keywords that follow the type specified.
837 If multiple parameter attributes are needed, they are space separated.
842 declare i32 @printf(i8* noalias nocapture, ...)
843 declare i32 @atoi(i8 zeroext)
844 declare signext i8 @returns_signed_char()
846 Note that any attributes for the function result (``nounwind``,
847 ``readonly``) come immediately after the argument list.
849 Currently, only the following parameter attributes are defined:
852 This indicates to the code generator that the parameter or return
853 value should be zero-extended to the extent required by the target's
854 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
855 the caller (for a parameter) or the callee (for a return value).
857 This indicates to the code generator that the parameter or return
858 value should be sign-extended to the extent required by the target's
859 ABI (which is usually 32-bits) by the caller (for a parameter) or
860 the callee (for a return value).
862 This indicates that this parameter or return value should be treated
863 in a special target-dependent fashion during while emitting code for
864 a function call or return (usually, by putting it in a register as
865 opposed to memory, though some targets use it to distinguish between
866 two different kinds of registers). Use of this attribute is
869 This indicates that the pointer parameter should really be passed by
870 value to the function. The attribute implies that a hidden copy of
871 the pointee is made between the caller and the callee, so the callee
872 is unable to modify the value in the caller. This attribute is only
873 valid on LLVM pointer arguments. It is generally used to pass
874 structs and arrays by value, but is also valid on pointers to
875 scalars. The copy is considered to belong to the caller not the
876 callee (for example, ``readonly`` functions should not write to
877 ``byval`` parameters). This is not a valid attribute for return
880 The byval attribute also supports specifying an alignment with the
881 align attribute. It indicates the alignment of the stack slot to
882 form and the known alignment of the pointer specified to the call
883 site. If the alignment is not specified, then the code generator
884 makes a target-specific assumption.
890 The ``inalloca`` argument attribute allows the caller to take the
891 address of outgoing stack arguments. An ``inalloca`` argument must
892 be a pointer to stack memory produced by an ``alloca`` instruction.
893 The alloca, or argument allocation, must also be tagged with the
894 inalloca keyword. Only the last argument may have the ``inalloca``
895 attribute, and that argument is guaranteed to be passed in memory.
897 An argument allocation may be used by a call at most once because
898 the call may deallocate it. The ``inalloca`` attribute cannot be
899 used in conjunction with other attributes that affect argument
900 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
901 ``inalloca`` attribute also disables LLVM's implicit lowering of
902 large aggregate return values, which means that frontend authors
903 must lower them with ``sret`` pointers.
905 When the call site is reached, the argument allocation must have
906 been the most recent stack allocation that is still live, or the
907 results are undefined. It is possible to allocate additional stack
908 space after an argument allocation and before its call site, but it
909 must be cleared off with :ref:`llvm.stackrestore
912 See :doc:`InAlloca` for more information on how to use this
916 This indicates that the pointer parameter specifies the address of a
917 structure that is the return value of the function in the source
918 program. This pointer must be guaranteed by the caller to be valid:
919 loads and stores to the structure may be assumed by the callee
920 not to trap and to be properly aligned. This may only be applied to
921 the first parameter. This is not a valid attribute for return
925 This indicates that the pointer value may be assumed by the optimizer to
926 have the specified alignment.
928 Note that this attribute has additional semantics when combined with the
934 This indicates that pointer values :ref:`based <pointeraliasing>` on
935 the argument or return value do not alias pointer values which are
936 not *based* on it, ignoring certain "irrelevant" dependencies. For a
937 call to the parent function, dependencies between memory references
938 from before or after the call and from those during the call are
939 "irrelevant" to the ``noalias`` keyword for the arguments and return
940 value used in that call. The caller shares the responsibility with
941 the callee for ensuring that these requirements are met. For further
942 details, please see the discussion of the NoAlias response in :ref:`alias
943 analysis <Must, May, or No>`.
945 Note that this definition of ``noalias`` is intentionally similar
946 to the definition of ``restrict`` in C99 for function arguments,
947 though it is slightly weaker.
949 For function return values, C99's ``restrict`` is not meaningful,
950 while LLVM's ``noalias`` is.
952 This indicates that the callee does not make any copies of the
953 pointer that outlive the callee itself. This is not a valid
954 attribute for return values.
959 This indicates that the pointer parameter can be excised using the
960 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
961 attribute for return values and can only be applied to one parameter.
964 This indicates that the function always returns the argument as its return
965 value. This is an optimization hint to the code generator when generating
966 the caller, allowing tail call optimization and omission of register saves
967 and restores in some cases; it is not checked or enforced when generating
968 the callee. The parameter and the function return type must be valid
969 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
970 valid attribute for return values and can only be applied to one parameter.
973 This indicates that the parameter or return pointer is not null. This
974 attribute may only be applied to pointer typed parameters. This is not
975 checked or enforced by LLVM, the caller must ensure that the pointer
976 passed in is non-null, or the callee must ensure that the returned pointer
979 ``dereferenceable(<n>)``
980 This indicates that the parameter or return pointer is dereferenceable. This
981 attribute may only be applied to pointer typed parameters. A pointer that
982 is dereferenceable can be loaded from speculatively without a risk of
983 trapping. The number of bytes known to be dereferenceable must be provided
984 in parentheses. It is legal for the number of bytes to be less than the
985 size of the pointee type. The ``nonnull`` attribute does not imply
986 dereferenceability (consider a pointer to one element past the end of an
987 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
988 ``addrspace(0)`` (which is the default address space).
992 Garbage Collector Names
993 -----------------------
995 Each function may specify a garbage collector name, which is simply a
1000 define void @f() gc "name" { ... }
1002 The compiler declares the supported values of *name*. Specifying a
1003 collector which will cause the compiler to alter its output in order to
1004 support the named garbage collection algorithm.
1011 Prefix data is data associated with a function which the code generator
1012 will emit immediately before the function body. The purpose of this feature
1013 is to allow frontends to associate language-specific runtime metadata with
1014 specific functions and make it available through the function pointer while
1015 still allowing the function pointer to be called. To access the data for a
1016 given function, a program may bitcast the function pointer to a pointer to
1017 the constant's type. This implies that the IR symbol points to the start
1020 To maintain the semantics of ordinary function calls, the prefix data must
1021 have a particular format. Specifically, it must begin with a sequence of
1022 bytes which decode to a sequence of machine instructions, valid for the
1023 module's target, which transfer control to the point immediately succeeding
1024 the prefix data, without performing any other visible action. This allows
1025 the inliner and other passes to reason about the semantics of the function
1026 definition without needing to reason about the prefix data. Obviously this
1027 makes the format of the prefix data highly target dependent.
1029 Prefix data is laid out as if it were an initializer for a global variable
1030 of the prefix data's type. No padding is automatically placed between the
1031 prefix data and the function body. If padding is required, it must be part
1034 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
1035 which encodes the ``nop`` instruction:
1037 .. code-block:: llvm
1039 define void @f() prefix i8 144 { ... }
1041 Generally prefix data can be formed by encoding a relative branch instruction
1042 which skips the metadata, as in this example of valid prefix data for the
1043 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1045 .. code-block:: llvm
1047 %0 = type <{ i8, i8, i8* }>
1049 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
1051 A function may have prefix data but no body. This has similar semantics
1052 to the ``available_externally`` linkage in that the data may be used by the
1053 optimizers but will not be emitted in the object file.
1060 Attribute groups are groups of attributes that are referenced by objects within
1061 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1062 functions will use the same set of attributes. In the degenerative case of a
1063 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1064 group will capture the important command line flags used to build that file.
1066 An attribute group is a module-level object. To use an attribute group, an
1067 object references the attribute group's ID (e.g. ``#37``). An object may refer
1068 to more than one attribute group. In that situation, the attributes from the
1069 different groups are merged.
1071 Here is an example of attribute groups for a function that should always be
1072 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1074 .. code-block:: llvm
1076 ; Target-independent attributes:
1077 attributes #0 = { alwaysinline alignstack=4 }
1079 ; Target-dependent attributes:
1080 attributes #1 = { "no-sse" }
1082 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1083 define void @f() #0 #1 { ... }
1090 Function attributes are set to communicate additional information about
1091 a function. Function attributes are considered to be part of the
1092 function, not of the function type, so functions with different function
1093 attributes can have the same function type.
1095 Function attributes are simple keywords that follow the type specified.
1096 If multiple attributes are needed, they are space separated. For
1099 .. code-block:: llvm
1101 define void @f() noinline { ... }
1102 define void @f() alwaysinline { ... }
1103 define void @f() alwaysinline optsize { ... }
1104 define void @f() optsize { ... }
1107 This attribute indicates that, when emitting the prologue and
1108 epilogue, the backend should forcibly align the stack pointer.
1109 Specify the desired alignment, which must be a power of two, in
1112 This attribute indicates that the inliner should attempt to inline
1113 this function into callers whenever possible, ignoring any active
1114 inlining size threshold for this caller.
1116 This indicates that the callee function at a call site should be
1117 recognized as a built-in function, even though the function's declaration
1118 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1119 direct calls to functions which are declared with the ``nobuiltin``
1122 This attribute indicates that this function is rarely called. When
1123 computing edge weights, basic blocks post-dominated by a cold
1124 function call are also considered to be cold; and, thus, given low
1127 This attribute indicates that the source code contained a hint that
1128 inlining this function is desirable (such as the "inline" keyword in
1129 C/C++). It is just a hint; it imposes no requirements on the
1132 This attribute indicates that the function should be added to a
1133 jump-instruction table at code-generation time, and that all address-taken
1134 references to this function should be replaced with a reference to the
1135 appropriate jump-instruction-table function pointer. Note that this creates
1136 a new pointer for the original function, which means that code that depends
1137 on function-pointer identity can break. So, any function annotated with
1138 ``jumptable`` must also be ``unnamed_addr``.
1140 This attribute suggests that optimization passes and code generator
1141 passes make choices that keep the code size of this function as small
1142 as possible and perform optimizations that may sacrifice runtime
1143 performance in order to minimize the size of the generated code.
1145 This attribute disables prologue / epilogue emission for the
1146 function. This can have very system-specific consequences.
1148 This indicates that the callee function at a call site is not recognized as
1149 a built-in function. LLVM will retain the original call and not replace it
1150 with equivalent code based on the semantics of the built-in function, unless
1151 the call site uses the ``builtin`` attribute. This is valid at call sites
1152 and on function declarations and definitions.
1154 This attribute indicates that calls to the function cannot be
1155 duplicated. A call to a ``noduplicate`` function may be moved
1156 within its parent function, but may not be duplicated within
1157 its parent function.
1159 A function containing a ``noduplicate`` call may still
1160 be an inlining candidate, provided that the call is not
1161 duplicated by inlining. That implies that the function has
1162 internal linkage and only has one call site, so the original
1163 call is dead after inlining.
1165 This attributes disables implicit floating point instructions.
1167 This attribute indicates that the inliner should never inline this
1168 function in any situation. This attribute may not be used together
1169 with the ``alwaysinline`` attribute.
1171 This attribute suppresses lazy symbol binding for the function. This
1172 may make calls to the function faster, at the cost of extra program
1173 startup time if the function is not called during program startup.
1175 This attribute indicates that the code generator should not use a
1176 red zone, even if the target-specific ABI normally permits it.
1178 This function attribute indicates that the function never returns
1179 normally. This produces undefined behavior at runtime if the
1180 function ever does dynamically return.
1182 This function attribute indicates that the function never returns
1183 with an unwind or exceptional control flow. If the function does
1184 unwind, its runtime behavior is undefined.
1186 This function attribute indicates that the function is not optimized
1187 by any optimization or code generator passes with the
1188 exception of interprocedural optimization passes.
1189 This attribute cannot be used together with the ``alwaysinline``
1190 attribute; this attribute is also incompatible
1191 with the ``minsize`` attribute and the ``optsize`` attribute.
1193 This attribute requires the ``noinline`` attribute to be specified on
1194 the function as well, so the function is never inlined into any caller.
1195 Only functions with the ``alwaysinline`` attribute are valid
1196 candidates for inlining into the body of this function.
1198 This attribute suggests that optimization passes and code generator
1199 passes make choices that keep the code size of this function low,
1200 and otherwise do optimizations specifically to reduce code size as
1201 long as they do not significantly impact runtime performance.
1203 On a function, this attribute indicates that the function computes its
1204 result (or decides to unwind an exception) based strictly on its arguments,
1205 without dereferencing any pointer arguments or otherwise accessing
1206 any mutable state (e.g. memory, control registers, etc) visible to
1207 caller functions. It does not write through any pointer arguments
1208 (including ``byval`` arguments) and never changes any state visible
1209 to callers. This means that it cannot unwind exceptions by calling
1210 the ``C++`` exception throwing methods.
1212 On an argument, this attribute indicates that the function does not
1213 dereference that pointer argument, even though it may read or write the
1214 memory that the pointer points to if accessed through other pointers.
1216 On a function, this attribute indicates that the function does not write
1217 through any pointer arguments (including ``byval`` arguments) or otherwise
1218 modify any state (e.g. memory, control registers, etc) visible to
1219 caller functions. It may dereference pointer arguments and read
1220 state that may be set in the caller. A readonly function always
1221 returns the same value (or unwinds an exception identically) when
1222 called with the same set of arguments and global state. It cannot
1223 unwind an exception by calling the ``C++`` exception throwing
1226 On an argument, this attribute indicates that the function does not write
1227 through this pointer argument, even though it may write to the memory that
1228 the pointer points to.
1230 This attribute indicates that this function can return twice. The C
1231 ``setjmp`` is an example of such a function. The compiler disables
1232 some optimizations (like tail calls) in the caller of these
1234 ``sanitize_address``
1235 This attribute indicates that AddressSanitizer checks
1236 (dynamic address safety analysis) are enabled for this function.
1238 This attribute indicates that MemorySanitizer checks (dynamic detection
1239 of accesses to uninitialized memory) are enabled for this function.
1241 This attribute indicates that ThreadSanitizer checks
1242 (dynamic thread safety analysis) are enabled for this function.
1244 This attribute indicates that the function should emit a stack
1245 smashing protector. It is in the form of a "canary" --- a random value
1246 placed on the stack before the local variables that's checked upon
1247 return from the function to see if it has been overwritten. A
1248 heuristic is used to determine if a function needs stack protectors
1249 or not. The heuristic used will enable protectors for functions with:
1251 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1252 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1253 - Calls to alloca() with variable sizes or constant sizes greater than
1254 ``ssp-buffer-size``.
1256 Variables that are identified as requiring a protector will be arranged
1257 on the stack such that they are adjacent to the stack protector guard.
1259 If a function that has an ``ssp`` attribute is inlined into a
1260 function that doesn't have an ``ssp`` attribute, then the resulting
1261 function will have an ``ssp`` attribute.
1263 This attribute indicates that the function should *always* emit a
1264 stack smashing protector. This overrides the ``ssp`` function
1267 Variables that are identified as requiring a protector will be arranged
1268 on the stack such that they are adjacent to the stack protector guard.
1269 The specific layout rules are:
1271 #. Large arrays and structures containing large arrays
1272 (``>= ssp-buffer-size``) are closest to the stack protector.
1273 #. Small arrays and structures containing small arrays
1274 (``< ssp-buffer-size``) are 2nd closest to the protector.
1275 #. Variables that have had their address taken are 3rd closest to the
1278 If a function that has an ``sspreq`` attribute is inlined into a
1279 function that doesn't have an ``sspreq`` attribute or which has an
1280 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1281 an ``sspreq`` attribute.
1283 This attribute indicates that the function should emit a stack smashing
1284 protector. This attribute causes a strong heuristic to be used when
1285 determining if a function needs stack protectors. The strong heuristic
1286 will enable protectors for functions with:
1288 - Arrays of any size and type
1289 - Aggregates containing an array of any size and type.
1290 - Calls to alloca().
1291 - Local variables that have had their address taken.
1293 Variables that are identified as requiring a protector will be arranged
1294 on the stack such that they are adjacent to the stack protector guard.
1295 The specific layout rules are:
1297 #. Large arrays and structures containing large arrays
1298 (``>= ssp-buffer-size``) are closest to the stack protector.
1299 #. Small arrays and structures containing small arrays
1300 (``< ssp-buffer-size``) are 2nd closest to the protector.
1301 #. Variables that have had their address taken are 3rd closest to the
1304 This overrides the ``ssp`` function attribute.
1306 If a function that has an ``sspstrong`` attribute is inlined into a
1307 function that doesn't have an ``sspstrong`` attribute, then the
1308 resulting function will have an ``sspstrong`` attribute.
1310 This attribute indicates that the ABI being targeted requires that
1311 an unwind table entry be produce for this function even if we can
1312 show that no exceptions passes by it. This is normally the case for
1313 the ELF x86-64 abi, but it can be disabled for some compilation
1318 Module-Level Inline Assembly
1319 ----------------------------
1321 Modules may contain "module-level inline asm" blocks, which corresponds
1322 to the GCC "file scope inline asm" blocks. These blocks are internally
1323 concatenated by LLVM and treated as a single unit, but may be separated
1324 in the ``.ll`` file if desired. The syntax is very simple:
1326 .. code-block:: llvm
1328 module asm "inline asm code goes here"
1329 module asm "more can go here"
1331 The strings can contain any character by escaping non-printable
1332 characters. The escape sequence used is simply "\\xx" where "xx" is the
1333 two digit hex code for the number.
1335 The inline asm code is simply printed to the machine code .s file when
1336 assembly code is generated.
1338 .. _langref_datalayout:
1343 A module may specify a target specific data layout string that specifies
1344 how data is to be laid out in memory. The syntax for the data layout is
1347 .. code-block:: llvm
1349 target datalayout = "layout specification"
1351 The *layout specification* consists of a list of specifications
1352 separated by the minus sign character ('-'). Each specification starts
1353 with a letter and may include other information after the letter to
1354 define some aspect of the data layout. The specifications accepted are
1358 Specifies that the target lays out data in big-endian form. That is,
1359 the bits with the most significance have the lowest address
1362 Specifies that the target lays out data in little-endian form. That
1363 is, the bits with the least significance have the lowest address
1366 Specifies the natural alignment of the stack in bits. Alignment
1367 promotion of stack variables is limited to the natural stack
1368 alignment to avoid dynamic stack realignment. The stack alignment
1369 must be a multiple of 8-bits. If omitted, the natural stack
1370 alignment defaults to "unspecified", which does not prevent any
1371 alignment promotions.
1372 ``p[n]:<size>:<abi>:<pref>``
1373 This specifies the *size* of a pointer and its ``<abi>`` and
1374 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1375 bits. The address space, ``n`` is optional, and if not specified,
1376 denotes the default address space 0. The value of ``n`` must be
1377 in the range [1,2^23).
1378 ``i<size>:<abi>:<pref>``
1379 This specifies the alignment for an integer type of a given bit
1380 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1381 ``v<size>:<abi>:<pref>``
1382 This specifies the alignment for a vector type of a given bit
1384 ``f<size>:<abi>:<pref>``
1385 This specifies the alignment for a floating point type of a given bit
1386 ``<size>``. Only values of ``<size>`` that are supported by the target
1387 will work. 32 (float) and 64 (double) are supported on all targets; 80
1388 or 128 (different flavors of long double) are also supported on some
1391 This specifies the alignment for an object of aggregate type.
1393 If present, specifies that llvm names are mangled in the output. The
1396 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1397 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1398 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1399 symbols get a ``_`` prefix.
1400 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1401 functions also get a suffix based on the frame size.
1402 ``n<size1>:<size2>:<size3>...``
1403 This specifies a set of native integer widths for the target CPU in
1404 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1405 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1406 this set are considered to support most general arithmetic operations
1409 On every specification that takes a ``<abi>:<pref>``, specifying the
1410 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1411 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1413 When constructing the data layout for a given target, LLVM starts with a
1414 default set of specifications which are then (possibly) overridden by
1415 the specifications in the ``datalayout`` keyword. The default
1416 specifications are given in this list:
1418 - ``E`` - big endian
1419 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1420 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1421 same as the default address space.
1422 - ``S0`` - natural stack alignment is unspecified
1423 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1424 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1425 - ``i16:16:16`` - i16 is 16-bit aligned
1426 - ``i32:32:32`` - i32 is 32-bit aligned
1427 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1428 alignment of 64-bits
1429 - ``f16:16:16`` - half is 16-bit aligned
1430 - ``f32:32:32`` - float is 32-bit aligned
1431 - ``f64:64:64`` - double is 64-bit aligned
1432 - ``f128:128:128`` - quad is 128-bit aligned
1433 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1434 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1435 - ``a:0:64`` - aggregates are 64-bit aligned
1437 When LLVM is determining the alignment for a given type, it uses the
1440 #. If the type sought is an exact match for one of the specifications,
1441 that specification is used.
1442 #. If no match is found, and the type sought is an integer type, then
1443 the smallest integer type that is larger than the bitwidth of the
1444 sought type is used. If none of the specifications are larger than
1445 the bitwidth then the largest integer type is used. For example,
1446 given the default specifications above, the i7 type will use the
1447 alignment of i8 (next largest) while both i65 and i256 will use the
1448 alignment of i64 (largest specified).
1449 #. If no match is found, and the type sought is a vector type, then the
1450 largest vector type that is smaller than the sought vector type will
1451 be used as a fall back. This happens because <128 x double> can be
1452 implemented in terms of 64 <2 x double>, for example.
1454 The function of the data layout string may not be what you expect.
1455 Notably, this is not a specification from the frontend of what alignment
1456 the code generator should use.
1458 Instead, if specified, the target data layout is required to match what
1459 the ultimate *code generator* expects. This string is used by the
1460 mid-level optimizers to improve code, and this only works if it matches
1461 what the ultimate code generator uses. If you would like to generate IR
1462 that does not embed this target-specific detail into the IR, then you
1463 don't have to specify the string. This will disable some optimizations
1464 that require precise layout information, but this also prevents those
1465 optimizations from introducing target specificity into the IR.
1472 A module may specify a target triple string that describes the target
1473 host. The syntax for the target triple is simply:
1475 .. code-block:: llvm
1477 target triple = "x86_64-apple-macosx10.7.0"
1479 The *target triple* string consists of a series of identifiers delimited
1480 by the minus sign character ('-'). The canonical forms are:
1484 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1485 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1487 This information is passed along to the backend so that it generates
1488 code for the proper architecture. It's possible to override this on the
1489 command line with the ``-mtriple`` command line option.
1491 .. _pointeraliasing:
1493 Pointer Aliasing Rules
1494 ----------------------
1496 Any memory access must be done through a pointer value associated with
1497 an address range of the memory access, otherwise the behavior is
1498 undefined. Pointer values are associated with address ranges according
1499 to the following rules:
1501 - A pointer value is associated with the addresses associated with any
1502 value it is *based* on.
1503 - An address of a global variable is associated with the address range
1504 of the variable's storage.
1505 - The result value of an allocation instruction is associated with the
1506 address range of the allocated storage.
1507 - A null pointer in the default address-space is associated with no
1509 - An integer constant other than zero or a pointer value returned from
1510 a function not defined within LLVM may be associated with address
1511 ranges allocated through mechanisms other than those provided by
1512 LLVM. Such ranges shall not overlap with any ranges of addresses
1513 allocated by mechanisms provided by LLVM.
1515 A pointer value is *based* on another pointer value according to the
1518 - A pointer value formed from a ``getelementptr`` operation is *based*
1519 on the first operand of the ``getelementptr``.
1520 - The result value of a ``bitcast`` is *based* on the operand of the
1522 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1523 values that contribute (directly or indirectly) to the computation of
1524 the pointer's value.
1525 - The "*based* on" relationship is transitive.
1527 Note that this definition of *"based"* is intentionally similar to the
1528 definition of *"based"* in C99, though it is slightly weaker.
1530 LLVM IR does not associate types with memory. The result type of a
1531 ``load`` merely indicates the size and alignment of the memory from
1532 which to load, as well as the interpretation of the value. The first
1533 operand type of a ``store`` similarly only indicates the size and
1534 alignment of the store.
1536 Consequently, type-based alias analysis, aka TBAA, aka
1537 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1538 :ref:`Metadata <metadata>` may be used to encode additional information
1539 which specialized optimization passes may use to implement type-based
1544 Volatile Memory Accesses
1545 ------------------------
1547 Certain memory accesses, such as :ref:`load <i_load>`'s,
1548 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1549 marked ``volatile``. The optimizers must not change the number of
1550 volatile operations or change their order of execution relative to other
1551 volatile operations. The optimizers *may* change the order of volatile
1552 operations relative to non-volatile operations. This is not Java's
1553 "volatile" and has no cross-thread synchronization behavior.
1555 IR-level volatile loads and stores cannot safely be optimized into
1556 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1557 flagged volatile. Likewise, the backend should never split or merge
1558 target-legal volatile load/store instructions.
1560 .. admonition:: Rationale
1562 Platforms may rely on volatile loads and stores of natively supported
1563 data width to be executed as single instruction. For example, in C
1564 this holds for an l-value of volatile primitive type with native
1565 hardware support, but not necessarily for aggregate types. The
1566 frontend upholds these expectations, which are intentionally
1567 unspecified in the IR. The rules above ensure that IR transformation
1568 do not violate the frontend's contract with the language.
1572 Memory Model for Concurrent Operations
1573 --------------------------------------
1575 The LLVM IR does not define any way to start parallel threads of
1576 execution or to register signal handlers. Nonetheless, there are
1577 platform-specific ways to create them, and we define LLVM IR's behavior
1578 in their presence. This model is inspired by the C++0x memory model.
1580 For a more informal introduction to this model, see the :doc:`Atomics`.
1582 We define a *happens-before* partial order as the least partial order
1585 - Is a superset of single-thread program order, and
1586 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1587 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1588 techniques, like pthread locks, thread creation, thread joining,
1589 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1590 Constraints <ordering>`).
1592 Note that program order does not introduce *happens-before* edges
1593 between a thread and signals executing inside that thread.
1595 Every (defined) read operation (load instructions, memcpy, atomic
1596 loads/read-modify-writes, etc.) R reads a series of bytes written by
1597 (defined) write operations (store instructions, atomic
1598 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1599 section, initialized globals are considered to have a write of the
1600 initializer which is atomic and happens before any other read or write
1601 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1602 may see any write to the same byte, except:
1604 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1605 write\ :sub:`2` happens before R\ :sub:`byte`, then
1606 R\ :sub:`byte` does not see write\ :sub:`1`.
1607 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1608 R\ :sub:`byte` does not see write\ :sub:`3`.
1610 Given that definition, R\ :sub:`byte` is defined as follows:
1612 - If R is volatile, the result is target-dependent. (Volatile is
1613 supposed to give guarantees which can support ``sig_atomic_t`` in
1614 C/C++, and may be used for accesses to addresses which do not behave
1615 like normal memory. It does not generally provide cross-thread
1617 - Otherwise, if there is no write to the same byte that happens before
1618 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1619 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1620 R\ :sub:`byte` returns the value written by that write.
1621 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1622 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1623 Memory Ordering Constraints <ordering>` section for additional
1624 constraints on how the choice is made.
1625 - Otherwise R\ :sub:`byte` returns ``undef``.
1627 R returns the value composed of the series of bytes it read. This
1628 implies that some bytes within the value may be ``undef`` **without**
1629 the entire value being ``undef``. Note that this only defines the
1630 semantics of the operation; it doesn't mean that targets will emit more
1631 than one instruction to read the series of bytes.
1633 Note that in cases where none of the atomic intrinsics are used, this
1634 model places only one restriction on IR transformations on top of what
1635 is required for single-threaded execution: introducing a store to a byte
1636 which might not otherwise be stored is not allowed in general.
1637 (Specifically, in the case where another thread might write to and read
1638 from an address, introducing a store can change a load that may see
1639 exactly one write into a load that may see multiple writes.)
1643 Atomic Memory Ordering Constraints
1644 ----------------------------------
1646 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1647 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1648 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1649 ordering parameters that determine which other atomic instructions on
1650 the same address they *synchronize with*. These semantics are borrowed
1651 from Java and C++0x, but are somewhat more colloquial. If these
1652 descriptions aren't precise enough, check those specs (see spec
1653 references in the :doc:`atomics guide <Atomics>`).
1654 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1655 differently since they don't take an address. See that instruction's
1656 documentation for details.
1658 For a simpler introduction to the ordering constraints, see the
1662 The set of values that can be read is governed by the happens-before
1663 partial order. A value cannot be read unless some operation wrote
1664 it. This is intended to provide a guarantee strong enough to model
1665 Java's non-volatile shared variables. This ordering cannot be
1666 specified for read-modify-write operations; it is not strong enough
1667 to make them atomic in any interesting way.
1669 In addition to the guarantees of ``unordered``, there is a single
1670 total order for modifications by ``monotonic`` operations on each
1671 address. All modification orders must be compatible with the
1672 happens-before order. There is no guarantee that the modification
1673 orders can be combined to a global total order for the whole program
1674 (and this often will not be possible). The read in an atomic
1675 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1676 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1677 order immediately before the value it writes. If one atomic read
1678 happens before another atomic read of the same address, the later
1679 read must see the same value or a later value in the address's
1680 modification order. This disallows reordering of ``monotonic`` (or
1681 stronger) operations on the same address. If an address is written
1682 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1683 read that address repeatedly, the other threads must eventually see
1684 the write. This corresponds to the C++0x/C1x
1685 ``memory_order_relaxed``.
1687 In addition to the guarantees of ``monotonic``, a
1688 *synchronizes-with* edge may be formed with a ``release`` operation.
1689 This is intended to model C++'s ``memory_order_acquire``.
1691 In addition to the guarantees of ``monotonic``, if this operation
1692 writes a value which is subsequently read by an ``acquire``
1693 operation, it *synchronizes-with* that operation. (This isn't a
1694 complete description; see the C++0x definition of a release
1695 sequence.) This corresponds to the C++0x/C1x
1696 ``memory_order_release``.
1697 ``acq_rel`` (acquire+release)
1698 Acts as both an ``acquire`` and ``release`` operation on its
1699 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1700 ``seq_cst`` (sequentially consistent)
1701 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1702 operation which only reads, ``release`` for an operation which only
1703 writes), there is a global total order on all
1704 sequentially-consistent operations on all addresses, which is
1705 consistent with the *happens-before* partial order and with the
1706 modification orders of all the affected addresses. Each
1707 sequentially-consistent read sees the last preceding write to the
1708 same address in this global order. This corresponds to the C++0x/C1x
1709 ``memory_order_seq_cst`` and Java volatile.
1713 If an atomic operation is marked ``singlethread``, it only *synchronizes
1714 with* or participates in modification and seq\_cst total orderings with
1715 other operations running in the same thread (for example, in signal
1723 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1724 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1725 :ref:`frem <i_frem>`) have the following flags that can set to enable
1726 otherwise unsafe floating point operations
1729 No NaNs - Allow optimizations to assume the arguments and result are not
1730 NaN. Such optimizations are required to retain defined behavior over
1731 NaNs, but the value of the result is undefined.
1734 No Infs - Allow optimizations to assume the arguments and result are not
1735 +/-Inf. Such optimizations are required to retain defined behavior over
1736 +/-Inf, but the value of the result is undefined.
1739 No Signed Zeros - Allow optimizations to treat the sign of a zero
1740 argument or result as insignificant.
1743 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1744 argument rather than perform division.
1747 Fast - Allow algebraically equivalent transformations that may
1748 dramatically change results in floating point (e.g. reassociate). This
1749 flag implies all the others.
1756 The LLVM type system is one of the most important features of the
1757 intermediate representation. Being typed enables a number of
1758 optimizations to be performed on the intermediate representation
1759 directly, without having to do extra analyses on the side before the
1760 transformation. A strong type system makes it easier to read the
1761 generated code and enables novel analyses and transformations that are
1762 not feasible to perform on normal three address code representations.
1772 The void type does not represent any value and has no size.
1790 The function type can be thought of as a function signature. It consists of a
1791 return type and a list of formal parameter types. The return type of a function
1792 type is a void type or first class type --- except for :ref:`label <t_label>`
1793 and :ref:`metadata <t_metadata>` types.
1799 <returntype> (<parameter list>)
1801 ...where '``<parameter list>``' is a comma-separated list of type
1802 specifiers. Optionally, the parameter list may include a type ``...``, which
1803 indicates that the function takes a variable number of arguments. Variable
1804 argument functions can access their arguments with the :ref:`variable argument
1805 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1806 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1810 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1811 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1812 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1813 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1814 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1815 | ``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. |
1816 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1817 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1818 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1825 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1826 Values of these types are the only ones which can be produced by
1834 These are the types that are valid in registers from CodeGen's perspective.
1843 The integer type is a very simple type that simply specifies an
1844 arbitrary bit width for the integer type desired. Any bit width from 1
1845 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1853 The number of bits the integer will occupy is specified by the ``N``
1859 +----------------+------------------------------------------------+
1860 | ``i1`` | a single-bit integer. |
1861 +----------------+------------------------------------------------+
1862 | ``i32`` | a 32-bit integer. |
1863 +----------------+------------------------------------------------+
1864 | ``i1942652`` | a really big integer of over 1 million bits. |
1865 +----------------+------------------------------------------------+
1869 Floating Point Types
1870 """"""""""""""""""""
1879 - 16-bit floating point value
1882 - 32-bit floating point value
1885 - 64-bit floating point value
1888 - 128-bit floating point value (112-bit mantissa)
1891 - 80-bit floating point value (X87)
1894 - 128-bit floating point value (two 64-bits)
1901 The x86_mmx type represents a value held in an MMX register on an x86
1902 machine. The operations allowed on it are quite limited: parameters and
1903 return values, load and store, and bitcast. User-specified MMX
1904 instructions are represented as intrinsic or asm calls with arguments
1905 and/or results of this type. There are no arrays, vectors or constants
1922 The pointer type is used to specify memory locations. Pointers are
1923 commonly used to reference objects in memory.
1925 Pointer types may have an optional address space attribute defining the
1926 numbered address space where the pointed-to object resides. The default
1927 address space is number zero. The semantics of non-zero address spaces
1928 are target-specific.
1930 Note that LLVM does not permit pointers to void (``void*``) nor does it
1931 permit pointers to labels (``label*``). Use ``i8*`` instead.
1941 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1942 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1943 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1944 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1945 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1946 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1947 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1956 A vector type is a simple derived type that represents a vector of
1957 elements. Vector types are used when multiple primitive data are
1958 operated in parallel using a single instruction (SIMD). A vector type
1959 requires a size (number of elements) and an underlying primitive data
1960 type. Vector types are considered :ref:`first class <t_firstclass>`.
1966 < <# elements> x <elementtype> >
1968 The number of elements is a constant integer value larger than 0;
1969 elementtype may be any integer or floating point type, or a pointer to
1970 these types. Vectors of size zero are not allowed.
1974 +-------------------+--------------------------------------------------+
1975 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1976 +-------------------+--------------------------------------------------+
1977 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1978 +-------------------+--------------------------------------------------+
1979 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1980 +-------------------+--------------------------------------------------+
1981 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1982 +-------------------+--------------------------------------------------+
1991 The label type represents code labels.
2006 The metadata type represents embedded metadata. No derived types may be
2007 created from metadata except for :ref:`function <t_function>` arguments.
2020 Aggregate Types are a subset of derived types that can contain multiple
2021 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2022 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2032 The array type is a very simple derived type that arranges elements
2033 sequentially in memory. The array type requires a size (number of
2034 elements) and an underlying data type.
2040 [<# elements> x <elementtype>]
2042 The number of elements is a constant integer value; ``elementtype`` may
2043 be any type with a size.
2047 +------------------+--------------------------------------+
2048 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2049 +------------------+--------------------------------------+
2050 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2051 +------------------+--------------------------------------+
2052 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2053 +------------------+--------------------------------------+
2055 Here are some examples of multidimensional arrays:
2057 +-----------------------------+----------------------------------------------------------+
2058 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2059 +-----------------------------+----------------------------------------------------------+
2060 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2061 +-----------------------------+----------------------------------------------------------+
2062 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2063 +-----------------------------+----------------------------------------------------------+
2065 There is no restriction on indexing beyond the end of the array implied
2066 by a static type (though there are restrictions on indexing beyond the
2067 bounds of an allocated object in some cases). This means that
2068 single-dimension 'variable sized array' addressing can be implemented in
2069 LLVM with a zero length array type. An implementation of 'pascal style
2070 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2080 The structure type is used to represent a collection of data members
2081 together in memory. The elements of a structure may be any type that has
2084 Structures in memory are accessed using '``load``' and '``store``' by
2085 getting a pointer to a field with the '``getelementptr``' instruction.
2086 Structures in registers are accessed using the '``extractvalue``' and
2087 '``insertvalue``' instructions.
2089 Structures may optionally be "packed" structures, which indicate that
2090 the alignment of the struct is one byte, and that there is no padding
2091 between the elements. In non-packed structs, padding between field types
2092 is inserted as defined by the DataLayout string in the module, which is
2093 required to match what the underlying code generator expects.
2095 Structures can either be "literal" or "identified". A literal structure
2096 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2097 identified types are always defined at the top level with a name.
2098 Literal types are uniqued by their contents and can never be recursive
2099 or opaque since there is no way to write one. Identified types can be
2100 recursive, can be opaqued, and are never uniqued.
2106 %T1 = type { <type list> } ; Identified normal struct type
2107 %T2 = type <{ <type list> }> ; Identified packed struct type
2111 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2112 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2113 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2114 | ``{ 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``. |
2115 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2116 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2117 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2121 Opaque Structure Types
2122 """"""""""""""""""""""
2126 Opaque structure types are used to represent named structure types that
2127 do not have a body specified. This corresponds (for example) to the C
2128 notion of a forward declared structure.
2139 +--------------+-------------------+
2140 | ``opaque`` | An opaque type. |
2141 +--------------+-------------------+
2148 LLVM has several different basic types of constants. This section
2149 describes them all and their syntax.
2154 **Boolean constants**
2155 The two strings '``true``' and '``false``' are both valid constants
2157 **Integer constants**
2158 Standard integers (such as '4') are constants of the
2159 :ref:`integer <t_integer>` type. Negative numbers may be used with
2161 **Floating point constants**
2162 Floating point constants use standard decimal notation (e.g.
2163 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2164 hexadecimal notation (see below). The assembler requires the exact
2165 decimal value of a floating-point constant. For example, the
2166 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2167 decimal in binary. Floating point constants must have a :ref:`floating
2168 point <t_floating>` type.
2169 **Null pointer constants**
2170 The identifier '``null``' is recognized as a null pointer constant
2171 and must be of :ref:`pointer type <t_pointer>`.
2173 The one non-intuitive notation for constants is the hexadecimal form of
2174 floating point constants. For example, the form
2175 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2176 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2177 constants are required (and the only time that they are generated by the
2178 disassembler) is when a floating point constant must be emitted but it
2179 cannot be represented as a decimal floating point number in a reasonable
2180 number of digits. For example, NaN's, infinities, and other special
2181 values are represented in their IEEE hexadecimal format so that assembly
2182 and disassembly do not cause any bits to change in the constants.
2184 When using the hexadecimal form, constants of types half, float, and
2185 double are represented using the 16-digit form shown above (which
2186 matches the IEEE754 representation for double); half and float values
2187 must, however, be exactly representable as IEEE 754 half and single
2188 precision, respectively. Hexadecimal format is always used for long
2189 double, and there are three forms of long double. The 80-bit format used
2190 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2191 128-bit format used by PowerPC (two adjacent doubles) is represented by
2192 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2193 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2194 will only work if they match the long double format on your target.
2195 The IEEE 16-bit format (half precision) is represented by ``0xH``
2196 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2197 (sign bit at the left).
2199 There are no constants of type x86_mmx.
2201 .. _complexconstants:
2206 Complex constants are a (potentially recursive) combination of simple
2207 constants and smaller complex constants.
2209 **Structure constants**
2210 Structure constants are represented with notation similar to
2211 structure type definitions (a comma separated list of elements,
2212 surrounded by braces (``{}``)). For example:
2213 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2214 "``@G = external global i32``". Structure constants must have
2215 :ref:`structure type <t_struct>`, and the number and types of elements
2216 must match those specified by the type.
2218 Array constants are represented with notation similar to array type
2219 definitions (a comma separated list of elements, surrounded by
2220 square brackets (``[]``)). For example:
2221 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2222 :ref:`array type <t_array>`, and the number and types of elements must
2223 match those specified by the type.
2224 **Vector constants**
2225 Vector constants are represented with notation similar to vector
2226 type definitions (a comma separated list of elements, surrounded by
2227 less-than/greater-than's (``<>``)). For example:
2228 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2229 must have :ref:`vector type <t_vector>`, and the number and types of
2230 elements must match those specified by the type.
2231 **Zero initialization**
2232 The string '``zeroinitializer``' can be used to zero initialize a
2233 value to zero of *any* type, including scalar and
2234 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2235 having to print large zero initializers (e.g. for large arrays) and
2236 is always exactly equivalent to using explicit zero initializers.
2238 A metadata node is a structure-like constant with :ref:`metadata
2239 type <t_metadata>`. For example:
2240 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2241 constants that are meant to be interpreted as part of the
2242 instruction stream, metadata is a place to attach additional
2243 information such as debug info.
2245 Global Variable and Function Addresses
2246 --------------------------------------
2248 The addresses of :ref:`global variables <globalvars>` and
2249 :ref:`functions <functionstructure>` are always implicitly valid
2250 (link-time) constants. These constants are explicitly referenced when
2251 the :ref:`identifier for the global <identifiers>` is used and always have
2252 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2255 .. code-block:: llvm
2259 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2266 The string '``undef``' can be used anywhere a constant is expected, and
2267 indicates that the user of the value may receive an unspecified
2268 bit-pattern. Undefined values may be of any type (other than '``label``'
2269 or '``void``') and be used anywhere a constant is permitted.
2271 Undefined values are useful because they indicate to the compiler that
2272 the program is well defined no matter what value is used. This gives the
2273 compiler more freedom to optimize. Here are some examples of
2274 (potentially surprising) transformations that are valid (in pseudo IR):
2276 .. code-block:: llvm
2286 This is safe because all of the output bits are affected by the undef
2287 bits. Any output bit can have a zero or one depending on the input bits.
2289 .. code-block:: llvm
2300 These logical operations have bits that are not always affected by the
2301 input. For example, if ``%X`` has a zero bit, then the output of the
2302 '``and``' operation will always be a zero for that bit, no matter what
2303 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2304 optimize or assume that the result of the '``and``' is '``undef``'.
2305 However, it is safe to assume that all bits of the '``undef``' could be
2306 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2307 all the bits of the '``undef``' operand to the '``or``' could be set,
2308 allowing the '``or``' to be folded to -1.
2310 .. code-block:: llvm
2312 %A = select undef, %X, %Y
2313 %B = select undef, 42, %Y
2314 %C = select %X, %Y, undef
2324 This set of examples shows that undefined '``select``' (and conditional
2325 branch) conditions can go *either way*, but they have to come from one
2326 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2327 both known to have a clear low bit, then ``%A`` would have to have a
2328 cleared low bit. However, in the ``%C`` example, the optimizer is
2329 allowed to assume that the '``undef``' operand could be the same as
2330 ``%Y``, allowing the whole '``select``' to be eliminated.
2332 .. code-block:: llvm
2334 %A = xor undef, undef
2351 This example points out that two '``undef``' operands are not
2352 necessarily the same. This can be surprising to people (and also matches
2353 C semantics) where they assume that "``X^X``" is always zero, even if
2354 ``X`` is undefined. This isn't true for a number of reasons, but the
2355 short answer is that an '``undef``' "variable" can arbitrarily change
2356 its value over its "live range". This is true because the variable
2357 doesn't actually *have a live range*. Instead, the value is logically
2358 read from arbitrary registers that happen to be around when needed, so
2359 the value is not necessarily consistent over time. In fact, ``%A`` and
2360 ``%C`` need to have the same semantics or the core LLVM "replace all
2361 uses with" concept would not hold.
2363 .. code-block:: llvm
2371 These examples show the crucial difference between an *undefined value*
2372 and *undefined behavior*. An undefined value (like '``undef``') is
2373 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2374 operation can be constant folded to '``undef``', because the '``undef``'
2375 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2376 However, in the second example, we can make a more aggressive
2377 assumption: because the ``undef`` is allowed to be an arbitrary value,
2378 we are allowed to assume that it could be zero. Since a divide by zero
2379 has *undefined behavior*, we are allowed to assume that the operation
2380 does not execute at all. This allows us to delete the divide and all
2381 code after it. Because the undefined operation "can't happen", the
2382 optimizer can assume that it occurs in dead code.
2384 .. code-block:: llvm
2386 a: store undef -> %X
2387 b: store %X -> undef
2392 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2393 value can be assumed to not have any effect; we can assume that the
2394 value is overwritten with bits that happen to match what was already
2395 there. However, a store *to* an undefined location could clobber
2396 arbitrary memory, therefore, it has undefined behavior.
2403 Poison values are similar to :ref:`undef values <undefvalues>`, however
2404 they also represent the fact that an instruction or constant expression
2405 which cannot evoke side effects has nevertheless detected a condition
2406 which results in undefined behavior.
2408 There is currently no way of representing a poison value in the IR; they
2409 only exist when produced by operations such as :ref:`add <i_add>` with
2412 Poison value behavior is defined in terms of value *dependence*:
2414 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2415 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2416 their dynamic predecessor basic block.
2417 - Function arguments depend on the corresponding actual argument values
2418 in the dynamic callers of their functions.
2419 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2420 instructions that dynamically transfer control back to them.
2421 - :ref:`Invoke <i_invoke>` instructions depend on the
2422 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2423 call instructions that dynamically transfer control back to them.
2424 - Non-volatile loads and stores depend on the most recent stores to all
2425 of the referenced memory addresses, following the order in the IR
2426 (including loads and stores implied by intrinsics such as
2427 :ref:`@llvm.memcpy <int_memcpy>`.)
2428 - An instruction with externally visible side effects depends on the
2429 most recent preceding instruction with externally visible side
2430 effects, following the order in the IR. (This includes :ref:`volatile
2431 operations <volatile>`.)
2432 - An instruction *control-depends* on a :ref:`terminator
2433 instruction <terminators>` if the terminator instruction has
2434 multiple successors and the instruction is always executed when
2435 control transfers to one of the successors, and may not be executed
2436 when control is transferred to another.
2437 - Additionally, an instruction also *control-depends* on a terminator
2438 instruction if the set of instructions it otherwise depends on would
2439 be different if the terminator had transferred control to a different
2441 - Dependence is transitive.
2443 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2444 with the additional affect that any instruction which has a *dependence*
2445 on a poison value has undefined behavior.
2447 Here are some examples:
2449 .. code-block:: llvm
2452 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2453 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2454 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2455 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2457 store i32 %poison, i32* @g ; Poison value stored to memory.
2458 %poison2 = load i32* @g ; Poison value loaded back from memory.
2460 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2462 %narrowaddr = bitcast i32* @g to i16*
2463 %wideaddr = bitcast i32* @g to i64*
2464 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2465 %poison4 = load i64* %wideaddr ; Returns a poison value.
2467 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2468 br i1 %cmp, label %true, label %end ; Branch to either destination.
2471 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2472 ; it has undefined behavior.
2476 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2477 ; Both edges into this PHI are
2478 ; control-dependent on %cmp, so this
2479 ; always results in a poison value.
2481 store volatile i32 0, i32* @g ; This would depend on the store in %true
2482 ; if %cmp is true, or the store in %entry
2483 ; otherwise, so this is undefined behavior.
2485 br i1 %cmp, label %second_true, label %second_end
2486 ; The same branch again, but this time the
2487 ; true block doesn't have side effects.
2494 store volatile i32 0, i32* @g ; This time, the instruction always depends
2495 ; on the store in %end. Also, it is
2496 ; control-equivalent to %end, so this is
2497 ; well-defined (ignoring earlier undefined
2498 ; behavior in this example).
2502 Addresses of Basic Blocks
2503 -------------------------
2505 ``blockaddress(@function, %block)``
2507 The '``blockaddress``' constant computes the address of the specified
2508 basic block in the specified function, and always has an ``i8*`` type.
2509 Taking the address of the entry block is illegal.
2511 This value only has defined behavior when used as an operand to the
2512 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2513 against null. Pointer equality tests between labels addresses results in
2514 undefined behavior --- though, again, comparison against null is ok, and
2515 no label is equal to the null pointer. This may be passed around as an
2516 opaque pointer sized value as long as the bits are not inspected. This
2517 allows ``ptrtoint`` and arithmetic to be performed on these values so
2518 long as the original value is reconstituted before the ``indirectbr``
2521 Finally, some targets may provide defined semantics when using the value
2522 as the operand to an inline assembly, but that is target specific.
2526 Constant Expressions
2527 --------------------
2529 Constant expressions are used to allow expressions involving other
2530 constants to be used as constants. Constant expressions may be of any
2531 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2532 that does not have side effects (e.g. load and call are not supported).
2533 The following is the syntax for constant expressions:
2535 ``trunc (CST to TYPE)``
2536 Truncate a constant to another type. The bit size of CST must be
2537 larger than the bit size of TYPE. Both types must be integers.
2538 ``zext (CST to TYPE)``
2539 Zero extend a constant to another type. The bit size of CST must be
2540 smaller than the bit size of TYPE. Both types must be integers.
2541 ``sext (CST to TYPE)``
2542 Sign extend a constant to another type. The bit size of CST must be
2543 smaller than the bit size of TYPE. Both types must be integers.
2544 ``fptrunc (CST to TYPE)``
2545 Truncate a floating point constant to another floating point type.
2546 The size of CST must be larger than the size of TYPE. Both types
2547 must be floating point.
2548 ``fpext (CST to TYPE)``
2549 Floating point extend a constant to another type. The size of CST
2550 must be smaller or equal to the size of TYPE. Both types must be
2552 ``fptoui (CST to TYPE)``
2553 Convert a floating point constant to the corresponding unsigned
2554 integer constant. TYPE must be a scalar or vector integer type. CST
2555 must be of scalar or vector floating point type. Both CST and TYPE
2556 must be scalars, or vectors of the same number of elements. If the
2557 value won't fit in the integer type, the results are undefined.
2558 ``fptosi (CST to TYPE)``
2559 Convert a floating point constant to the corresponding signed
2560 integer constant. TYPE must be a scalar or vector integer type. CST
2561 must be of scalar or vector floating point type. Both CST and TYPE
2562 must be scalars, or vectors of the same number of elements. If the
2563 value won't fit in the integer type, the results are undefined.
2564 ``uitofp (CST to TYPE)``
2565 Convert an unsigned integer constant to the corresponding floating
2566 point constant. TYPE must be a scalar or vector floating point type.
2567 CST must be of scalar or vector integer type. Both CST and TYPE must
2568 be scalars, or vectors of the same number of elements. If the value
2569 won't fit in the floating point type, the results are undefined.
2570 ``sitofp (CST to TYPE)``
2571 Convert a signed integer constant to the corresponding floating
2572 point constant. TYPE must be a scalar or vector floating point type.
2573 CST must be of scalar or vector integer type. Both CST and TYPE must
2574 be scalars, or vectors of the same number of elements. If the value
2575 won't fit in the floating point type, the results are undefined.
2576 ``ptrtoint (CST to TYPE)``
2577 Convert a pointer typed constant to the corresponding integer
2578 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2579 pointer type. The ``CST`` value is zero extended, truncated, or
2580 unchanged to make it fit in ``TYPE``.
2581 ``inttoptr (CST to TYPE)``
2582 Convert an integer constant to a pointer constant. TYPE must be a
2583 pointer type. CST must be of integer type. The CST value is zero
2584 extended, truncated, or unchanged to make it fit in a pointer size.
2585 This one is *really* dangerous!
2586 ``bitcast (CST to TYPE)``
2587 Convert a constant, CST, to another TYPE. The constraints of the
2588 operands are the same as those for the :ref:`bitcast
2589 instruction <i_bitcast>`.
2590 ``addrspacecast (CST to TYPE)``
2591 Convert a constant pointer or constant vector of pointer, CST, to another
2592 TYPE in a different address space. The constraints of the operands are the
2593 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2594 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2595 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2596 constants. As with the :ref:`getelementptr <i_getelementptr>`
2597 instruction, the index list may have zero or more indexes, which are
2598 required to make sense for the type of "CSTPTR".
2599 ``select (COND, VAL1, VAL2)``
2600 Perform the :ref:`select operation <i_select>` on constants.
2601 ``icmp COND (VAL1, VAL2)``
2602 Performs the :ref:`icmp operation <i_icmp>` on constants.
2603 ``fcmp COND (VAL1, VAL2)``
2604 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2605 ``extractelement (VAL, IDX)``
2606 Perform the :ref:`extractelement operation <i_extractelement>` on
2608 ``insertelement (VAL, ELT, IDX)``
2609 Perform the :ref:`insertelement operation <i_insertelement>` on
2611 ``shufflevector (VEC1, VEC2, IDXMASK)``
2612 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2614 ``extractvalue (VAL, IDX0, IDX1, ...)``
2615 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2616 constants. The index list is interpreted in a similar manner as
2617 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2618 least one index value must be specified.
2619 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2620 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2621 The index list is interpreted in a similar manner as indices in a
2622 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2623 value must be specified.
2624 ``OPCODE (LHS, RHS)``
2625 Perform the specified operation of the LHS and RHS constants. OPCODE
2626 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2627 binary <bitwiseops>` operations. The constraints on operands are
2628 the same as those for the corresponding instruction (e.g. no bitwise
2629 operations on floating point values are allowed).
2636 Inline Assembler Expressions
2637 ----------------------------
2639 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2640 Inline Assembly <moduleasm>`) through the use of a special value. This
2641 value represents the inline assembler as a string (containing the
2642 instructions to emit), a list of operand constraints (stored as a
2643 string), a flag that indicates whether or not the inline asm expression
2644 has side effects, and a flag indicating whether the function containing
2645 the asm needs to align its stack conservatively. An example inline
2646 assembler expression is:
2648 .. code-block:: llvm
2650 i32 (i32) asm "bswap $0", "=r,r"
2652 Inline assembler expressions may **only** be used as the callee operand
2653 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2654 Thus, typically we have:
2656 .. code-block:: llvm
2658 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2660 Inline asms with side effects not visible in the constraint list must be
2661 marked as having side effects. This is done through the use of the
2662 '``sideeffect``' keyword, like so:
2664 .. code-block:: llvm
2666 call void asm sideeffect "eieio", ""()
2668 In some cases inline asms will contain code that will not work unless
2669 the stack is aligned in some way, such as calls or SSE instructions on
2670 x86, yet will not contain code that does that alignment within the asm.
2671 The compiler should make conservative assumptions about what the asm
2672 might contain and should generate its usual stack alignment code in the
2673 prologue if the '``alignstack``' keyword is present:
2675 .. code-block:: llvm
2677 call void asm alignstack "eieio", ""()
2679 Inline asms also support using non-standard assembly dialects. The
2680 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2681 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2682 the only supported dialects. An example is:
2684 .. code-block:: llvm
2686 call void asm inteldialect "eieio", ""()
2688 If multiple keywords appear the '``sideeffect``' keyword must come
2689 first, the '``alignstack``' keyword second and the '``inteldialect``'
2695 The call instructions that wrap inline asm nodes may have a
2696 "``!srcloc``" MDNode attached to it that contains a list of constant
2697 integers. If present, the code generator will use the integer as the
2698 location cookie value when report errors through the ``LLVMContext``
2699 error reporting mechanisms. This allows a front-end to correlate backend
2700 errors that occur with inline asm back to the source code that produced
2703 .. code-block:: llvm
2705 call void asm sideeffect "something bad", ""(), !srcloc !42
2707 !42 = !{ i32 1234567 }
2709 It is up to the front-end to make sense of the magic numbers it places
2710 in the IR. If the MDNode contains multiple constants, the code generator
2711 will use the one that corresponds to the line of the asm that the error
2716 Metadata Nodes and Metadata Strings
2717 -----------------------------------
2719 LLVM IR allows metadata to be attached to instructions in the program
2720 that can convey extra information about the code to the optimizers and
2721 code generator. One example application of metadata is source-level
2722 debug information. There are two metadata primitives: strings and nodes.
2723 All metadata has the ``metadata`` type and is identified in syntax by a
2724 preceding exclamation point ('``!``').
2726 A metadata string is a string surrounded by double quotes. It can
2727 contain any character by escaping non-printable characters with
2728 "``\xx``" where "``xx``" is the two digit hex code. For example:
2731 Metadata nodes are represented with notation similar to structure
2732 constants (a comma separated list of elements, surrounded by braces and
2733 preceded by an exclamation point). Metadata nodes can have any values as
2734 their operand. For example:
2736 .. code-block:: llvm
2738 !{ metadata !"test\00", i32 10}
2740 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2741 metadata nodes, which can be looked up in the module symbol table. For
2744 .. code-block:: llvm
2746 !foo = metadata !{!4, !3}
2748 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2749 function is using two metadata arguments:
2751 .. code-block:: llvm
2753 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2755 Metadata can be attached with an instruction. Here metadata ``!21`` is
2756 attached to the ``add`` instruction using the ``!dbg`` identifier:
2758 .. code-block:: llvm
2760 %indvar.next = add i64 %indvar, 1, !dbg !21
2762 More information about specific metadata nodes recognized by the
2763 optimizers and code generator is found below.
2768 In LLVM IR, memory does not have types, so LLVM's own type system is not
2769 suitable for doing TBAA. Instead, metadata is added to the IR to
2770 describe a type system of a higher level language. This can be used to
2771 implement typical C/C++ TBAA, but it can also be used to implement
2772 custom alias analysis behavior for other languages.
2774 The current metadata format is very simple. TBAA metadata nodes have up
2775 to three fields, e.g.:
2777 .. code-block:: llvm
2779 !0 = metadata !{ metadata !"an example type tree" }
2780 !1 = metadata !{ metadata !"int", metadata !0 }
2781 !2 = metadata !{ metadata !"float", metadata !0 }
2782 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2784 The first field is an identity field. It can be any value, usually a
2785 metadata string, which uniquely identifies the type. The most important
2786 name in the tree is the name of the root node. Two trees with different
2787 root node names are entirely disjoint, even if they have leaves with
2790 The second field identifies the type's parent node in the tree, or is
2791 null or omitted for a root node. A type is considered to alias all of
2792 its descendants and all of its ancestors in the tree. Also, a type is
2793 considered to alias all types in other trees, so that bitcode produced
2794 from multiple front-ends is handled conservatively.
2796 If the third field is present, it's an integer which if equal to 1
2797 indicates that the type is "constant" (meaning
2798 ``pointsToConstantMemory`` should return true; see `other useful
2799 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2801 '``tbaa.struct``' Metadata
2802 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2804 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2805 aggregate assignment operations in C and similar languages, however it
2806 is defined to copy a contiguous region of memory, which is more than
2807 strictly necessary for aggregate types which contain holes due to
2808 padding. Also, it doesn't contain any TBAA information about the fields
2811 ``!tbaa.struct`` metadata can describe which memory subregions in a
2812 memcpy are padding and what the TBAA tags of the struct are.
2814 The current metadata format is very simple. ``!tbaa.struct`` metadata
2815 nodes are a list of operands which are in conceptual groups of three.
2816 For each group of three, the first operand gives the byte offset of a
2817 field in bytes, the second gives its size in bytes, and the third gives
2820 .. code-block:: llvm
2822 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2824 This describes a struct with two fields. The first is at offset 0 bytes
2825 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2826 and has size 4 bytes and has tbaa tag !2.
2828 Note that the fields need not be contiguous. In this example, there is a
2829 4 byte gap between the two fields. This gap represents padding which
2830 does not carry useful data and need not be preserved.
2832 '``noalias``' and '``alias.scope``' Metadata
2833 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2835 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
2836 noalias memory-access sets. This means that some collection of memory access
2837 instructions (loads, stores, memory-accessing calls, etc.) that carry
2838 ``noalias`` metadata can specifically be specified not to alias with some other
2839 collection of memory access instructions that carry ``alias.scope`` metadata.
2840 Each type of metadata specifies a list of scopes, and when evaluating an
2841 aliasing query, if one of the instructions has a scope in its ``alias.scope``
2842 list that is identical to a scope in the other instruction's ``noalias`` list,
2843 or is a descendant (in the scope hierarchy) of a scope in the other
2844 instruction's ``noalias`` list , then the two memory accesses are assumed not
2847 The metadata identifying each scope is itself a list containing one or two
2848 entries. The first entry is the name of the scope. Note that if the name is a
2849 string then it can be combined accross functions and translation units. A
2850 self-reference can be used to create globally unique scope names.
2851 Optionally, a metadata reference to a parent scope can be provided as a second
2856 .. code-block:: llvm
2858 ; A root scope (which doubles as a list of itself):
2859 !0 = metadata !{metadata !0}
2861 ; Two child scopes (which must be self-referential to avoid being "uniqued"):
2862 !1 = metadata !{metadata !2} ; A list containing only scope !2
2863 !2 = metadata !{metadata !2, metadata !0} ; Scope !2 is a descendant of scope !0
2865 !3 = metadata !{metadata !4} ; A list containing only scope !4
2866 !4 = metadata !{metadata !4, metadata !0} ; Scope !4 is a descendant of scope !0
2868 ; These two instructions don't alias:
2869 %0 = load float* %c, align 4, !alias.scope !0
2870 store float %0, float* %arrayidx.i, align 4, !noalias !0
2872 ; These two instructions may alias (scope !2 and scope !4 are peers):
2873 %2 = load float* %c, align 4, !alias.scope !1
2874 store float %2, float* %arrayidx.i2, align 4, !noalias !3
2876 ; These two instructions don't alias (scope !2 is a descendant of scope !0
2877 ; and the store does not alias with anything in scope !0 or any of its descendants):
2878 %2 = load float* %c, align 4, !alias.scope !1
2879 store float %0, float* %arrayidx.i, align 4, !noalias !0
2881 ; These two instructions may alias:
2882 %2 = load float* %c, align 4, !alias.scope !0
2883 store float %0, float* %arrayidx.i, align 4, !noalias !1
2885 '``fpmath``' Metadata
2886 ^^^^^^^^^^^^^^^^^^^^^
2888 ``fpmath`` metadata may be attached to any instruction of floating point
2889 type. It can be used to express the maximum acceptable error in the
2890 result of that instruction, in ULPs, thus potentially allowing the
2891 compiler to use a more efficient but less accurate method of computing
2892 it. ULP is defined as follows:
2894 If ``x`` is a real number that lies between two finite consecutive
2895 floating-point numbers ``a`` and ``b``, without being equal to one
2896 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2897 distance between the two non-equal finite floating-point numbers
2898 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2900 The metadata node shall consist of a single positive floating point
2901 number representing the maximum relative error, for example:
2903 .. code-block:: llvm
2905 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2907 '``range``' Metadata
2908 ^^^^^^^^^^^^^^^^^^^^
2910 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
2911 integer types. It expresses the possible ranges the loaded value or the value
2912 returned by the called function at this call site is in. The ranges are
2913 represented with a flattened list of integers. The loaded value or the value
2914 returned is known to be in the union of the ranges defined by each consecutive
2915 pair. Each pair has the following properties:
2917 - The type must match the type loaded by the instruction.
2918 - The pair ``a,b`` represents the range ``[a,b)``.
2919 - Both ``a`` and ``b`` are constants.
2920 - The range is allowed to wrap.
2921 - The range should not represent the full or empty set. That is,
2924 In addition, the pairs must be in signed order of the lower bound and
2925 they must be non-contiguous.
2929 .. code-block:: llvm
2931 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2932 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2933 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
2934 %d = invoke i8 @bar() to label %cont
2935 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
2937 !0 = metadata !{ i8 0, i8 2 }
2938 !1 = metadata !{ i8 255, i8 2 }
2939 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2940 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2945 It is sometimes useful to attach information to loop constructs. Currently,
2946 loop metadata is implemented as metadata attached to the branch instruction
2947 in the loop latch block. This type of metadata refer to a metadata node that is
2948 guaranteed to be separate for each loop. The loop identifier metadata is
2949 specified with the name ``llvm.loop``.
2951 The loop identifier metadata is implemented using a metadata that refers to
2952 itself to avoid merging it with any other identifier metadata, e.g.,
2953 during module linkage or function inlining. That is, each loop should refer
2954 to their own identification metadata even if they reside in separate functions.
2955 The following example contains loop identifier metadata for two separate loop
2958 .. code-block:: llvm
2960 !0 = metadata !{ metadata !0 }
2961 !1 = metadata !{ metadata !1 }
2963 The loop identifier metadata can be used to specify additional
2964 per-loop metadata. Any operands after the first operand can be treated
2965 as user-defined metadata. For example the ``llvm.loop.unroll.count``
2966 suggests an unroll factor to the loop unroller:
2968 .. code-block:: llvm
2970 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2972 !0 = metadata !{ metadata !0, metadata !1 }
2973 !1 = metadata !{ metadata !"llvm.loop.unroll.count", i32 4 }
2975 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
2976 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2978 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
2979 used to control per-loop vectorization and interleaving parameters such as
2980 vectorization width and interleave count. These metadata should be used in
2981 conjunction with ``llvm.loop`` loop identification metadata. The
2982 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
2983 optimization hints and the optimizer will only interleave and vectorize loops if
2984 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
2985 which contains information about loop-carried memory dependencies can be helpful
2986 in determining the safety of these transformations.
2988 '``llvm.loop.interleave.count``' Metadata
2989 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2991 This metadata suggests an interleave count to the loop interleaver.
2992 The first operand is the string ``llvm.loop.interleave.count`` and the
2993 second operand is an integer specifying the interleave count. For
2996 .. code-block:: llvm
2998 !0 = metadata !{ metadata !"llvm.loop.interleave.count", i32 4 }
3000 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
3001 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
3002 then the interleave count will be determined automatically.
3004 '``llvm.loop.vectorize.enable``' Metadata
3005 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3007 This metadata selectively enables or disables vectorization for the loop. The
3008 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
3009 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
3010 0 disables vectorization:
3012 .. code-block:: llvm
3014 !0 = metadata !{ metadata !"llvm.loop.vectorize.enable", i1 0 }
3015 !1 = metadata !{ metadata !"llvm.loop.vectorize.enable", i1 1 }
3017 '``llvm.loop.vectorize.width``' Metadata
3018 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3020 This metadata sets the target width of the vectorizer. The first
3021 operand is the string ``llvm.loop.vectorize.width`` and the second
3022 operand is an integer specifying the width. For example:
3024 .. code-block:: llvm
3026 !0 = metadata !{ metadata !"llvm.loop.vectorize.width", i32 4 }
3028 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
3029 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
3030 0 or if the loop does not have this metadata the width will be
3031 determined automatically.
3033 '``llvm.loop.unroll``'
3034 ^^^^^^^^^^^^^^^^^^^^^^
3036 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
3037 optimization hints such as the unroll factor. ``llvm.loop.unroll``
3038 metadata should be used in conjunction with ``llvm.loop`` loop
3039 identification metadata. The ``llvm.loop.unroll`` metadata are only
3040 optimization hints and the unrolling will only be performed if the
3041 optimizer believes it is safe to do so.
3043 '``llvm.loop.unroll.count``' Metadata
3044 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3046 This metadata suggests an unroll factor to the loop unroller. The
3047 first operand is the string ``llvm.loop.unroll.count`` and the second
3048 operand is a positive integer specifying the unroll factor. For
3051 .. code-block:: llvm
3053 !0 = metadata !{ metadata !"llvm.loop.unroll.count", i32 4 }
3055 If the trip count of the loop is less than the unroll count the loop
3056 will be partially unrolled.
3058 '``llvm.loop.unroll.disable``' Metadata
3059 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3061 This metadata either disables loop unrolling. The metadata has a single operand
3062 which is the string ``llvm.loop.unroll.disable``. For example:
3064 .. code-block:: llvm
3066 !0 = metadata !{ metadata !"llvm.loop.unroll.disable" }
3068 '``llvm.loop.unroll.full``' Metadata
3069 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3071 This metadata either suggests that the loop should be unrolled fully. The
3072 metadata has a single operand which is the string ``llvm.loop.unroll.disable``.
3075 .. code-block:: llvm
3077 !0 = metadata !{ metadata !"llvm.loop.unroll.full" }
3082 Metadata types used to annotate memory accesses with information helpful
3083 for optimizations are prefixed with ``llvm.mem``.
3085 '``llvm.mem.parallel_loop_access``' Metadata
3086 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3088 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
3089 or metadata containing a list of loop identifiers for nested loops.
3090 The metadata is attached to memory accessing instructions and denotes that
3091 no loop carried memory dependence exist between it and other instructions denoted
3092 with the same loop identifier.
3094 Precisely, given two instructions ``m1`` and ``m2`` that both have the
3095 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
3096 set of loops associated with that metadata, respectively, then there is no loop
3097 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
3100 As a special case, if all memory accessing instructions in a loop have
3101 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
3102 loop has no loop carried memory dependences and is considered to be a parallel
3105 Note that if not all memory access instructions have such metadata referring to
3106 the loop, then the loop is considered not being trivially parallel. Additional
3107 memory dependence analysis is required to make that determination. As a fail
3108 safe mechanism, this causes loops that were originally parallel to be considered
3109 sequential (if optimization passes that are unaware of the parallel semantics
3110 insert new memory instructions into the loop body).
3112 Example of a loop that is considered parallel due to its correct use of
3113 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
3114 metadata types that refer to the same loop identifier metadata.
3116 .. code-block:: llvm
3120 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
3122 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3124 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
3128 !0 = metadata !{ metadata !0 }
3130 It is also possible to have nested parallel loops. In that case the
3131 memory accesses refer to a list of loop identifier metadata nodes instead of
3132 the loop identifier metadata node directly:
3134 .. code-block:: llvm
3138 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
3140 br label %inner.for.body
3144 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3146 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
3148 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
3152 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
3154 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
3156 outer.for.end: ; preds = %for.body
3158 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
3159 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
3160 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
3162 Module Flags Metadata
3163 =====================
3165 Information about the module as a whole is difficult to convey to LLVM's
3166 subsystems. The LLVM IR isn't sufficient to transmit this information.
3167 The ``llvm.module.flags`` named metadata exists in order to facilitate
3168 this. These flags are in the form of key / value pairs --- much like a
3169 dictionary --- making it easy for any subsystem who cares about a flag to
3172 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
3173 Each triplet has the following form:
3175 - The first element is a *behavior* flag, which specifies the behavior
3176 when two (or more) modules are merged together, and it encounters two
3177 (or more) metadata with the same ID. The supported behaviors are
3179 - The second element is a metadata string that is a unique ID for the
3180 metadata. Each module may only have one flag entry for each unique ID (not
3181 including entries with the **Require** behavior).
3182 - The third element is the value of the flag.
3184 When two (or more) modules are merged together, the resulting
3185 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
3186 each unique metadata ID string, there will be exactly one entry in the merged
3187 modules ``llvm.module.flags`` metadata table, and the value for that entry will
3188 be determined by the merge behavior flag, as described below. The only exception
3189 is that entries with the *Require* behavior are always preserved.
3191 The following behaviors are supported:
3202 Emits an error if two values disagree, otherwise the resulting value
3203 is that of the operands.
3207 Emits a warning if two values disagree. The result value will be the
3208 operand for the flag from the first module being linked.
3212 Adds a requirement that another module flag be present and have a
3213 specified value after linking is performed. The value must be a
3214 metadata pair, where the first element of the pair is the ID of the
3215 module flag to be restricted, and the second element of the pair is
3216 the value the module flag should be restricted to. This behavior can
3217 be used to restrict the allowable results (via triggering of an
3218 error) of linking IDs with the **Override** behavior.
3222 Uses the specified value, regardless of the behavior or value of the
3223 other module. If both modules specify **Override**, but the values
3224 differ, an error will be emitted.
3228 Appends the two values, which are required to be metadata nodes.
3232 Appends the two values, which are required to be metadata
3233 nodes. However, duplicate entries in the second list are dropped
3234 during the append operation.
3236 It is an error for a particular unique flag ID to have multiple behaviors,
3237 except in the case of **Require** (which adds restrictions on another metadata
3238 value) or **Override**.
3240 An example of module flags:
3242 .. code-block:: llvm
3244 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3245 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3246 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3247 !3 = metadata !{ i32 3, metadata !"qux",
3249 metadata !"foo", i32 1
3252 !llvm.module.flags = !{ !0, !1, !2, !3 }
3254 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3255 if two or more ``!"foo"`` flags are seen is to emit an error if their
3256 values are not equal.
3258 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3259 behavior if two or more ``!"bar"`` flags are seen is to use the value
3262 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3263 behavior if two or more ``!"qux"`` flags are seen is to emit a
3264 warning if their values are not equal.
3266 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3270 metadata !{ metadata !"foo", i32 1 }
3272 The behavior is to emit an error if the ``llvm.module.flags`` does not
3273 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3276 Objective-C Garbage Collection Module Flags Metadata
3277 ----------------------------------------------------
3279 On the Mach-O platform, Objective-C stores metadata about garbage
3280 collection in a special section called "image info". The metadata
3281 consists of a version number and a bitmask specifying what types of
3282 garbage collection are supported (if any) by the file. If two or more
3283 modules are linked together their garbage collection metadata needs to
3284 be merged rather than appended together.
3286 The Objective-C garbage collection module flags metadata consists of the
3287 following key-value pairs:
3296 * - ``Objective-C Version``
3297 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3299 * - ``Objective-C Image Info Version``
3300 - **[Required]** --- The version of the image info section. Currently
3303 * - ``Objective-C Image Info Section``
3304 - **[Required]** --- The section to place the metadata. Valid values are
3305 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3306 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3307 Objective-C ABI version 2.
3309 * - ``Objective-C Garbage Collection``
3310 - **[Required]** --- Specifies whether garbage collection is supported or
3311 not. Valid values are 0, for no garbage collection, and 2, for garbage
3312 collection supported.
3314 * - ``Objective-C GC Only``
3315 - **[Optional]** --- Specifies that only garbage collection is supported.
3316 If present, its value must be 6. This flag requires that the
3317 ``Objective-C Garbage Collection`` flag have the value 2.
3319 Some important flag interactions:
3321 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3322 merged with a module with ``Objective-C Garbage Collection`` set to
3323 2, then the resulting module has the
3324 ``Objective-C Garbage Collection`` flag set to 0.
3325 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3326 merged with a module with ``Objective-C GC Only`` set to 6.
3328 Automatic Linker Flags Module Flags Metadata
3329 --------------------------------------------
3331 Some targets support embedding flags to the linker inside individual object
3332 files. Typically this is used in conjunction with language extensions which
3333 allow source files to explicitly declare the libraries they depend on, and have
3334 these automatically be transmitted to the linker via object files.
3336 These flags are encoded in the IR using metadata in the module flags section,
3337 using the ``Linker Options`` key. The merge behavior for this flag is required
3338 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3339 node which should be a list of other metadata nodes, each of which should be a
3340 list of metadata strings defining linker options.
3342 For example, the following metadata section specifies two separate sets of
3343 linker options, presumably to link against ``libz`` and the ``Cocoa``
3346 !0 = metadata !{ i32 6, metadata !"Linker Options",
3348 metadata !{ metadata !"-lz" },
3349 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3350 !llvm.module.flags = !{ !0 }
3352 The metadata encoding as lists of lists of options, as opposed to a collapsed
3353 list of options, is chosen so that the IR encoding can use multiple option
3354 strings to specify e.g., a single library, while still having that specifier be
3355 preserved as an atomic element that can be recognized by a target specific
3356 assembly writer or object file emitter.
3358 Each individual option is required to be either a valid option for the target's
3359 linker, or an option that is reserved by the target specific assembly writer or
3360 object file emitter. No other aspect of these options is defined by the IR.
3362 C type width Module Flags Metadata
3363 ----------------------------------
3365 The ARM backend emits a section into each generated object file describing the
3366 options that it was compiled with (in a compiler-independent way) to prevent
3367 linking incompatible objects, and to allow automatic library selection. Some
3368 of these options are not visible at the IR level, namely wchar_t width and enum
3371 To pass this information to the backend, these options are encoded in module
3372 flags metadata, using the following key-value pairs:
3382 - * 0 --- sizeof(wchar_t) == 4
3383 * 1 --- sizeof(wchar_t) == 2
3386 - * 0 --- Enums are at least as large as an ``int``.
3387 * 1 --- Enums are stored in the smallest integer type which can
3388 represent all of its values.
3390 For example, the following metadata section specifies that the module was
3391 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
3392 enum is the smallest type which can represent all of its values::
3394 !llvm.module.flags = !{!0, !1}
3395 !0 = metadata !{i32 1, metadata !"short_wchar", i32 1}
3396 !1 = metadata !{i32 1, metadata !"short_enum", i32 0}
3398 .. _intrinsicglobalvariables:
3400 Intrinsic Global Variables
3401 ==========================
3403 LLVM has a number of "magic" global variables that contain data that
3404 affect code generation or other IR semantics. These are documented here.
3405 All globals of this sort should have a section specified as
3406 "``llvm.metadata``". This section and all globals that start with
3407 "``llvm.``" are reserved for use by LLVM.
3411 The '``llvm.used``' Global Variable
3412 -----------------------------------
3414 The ``@llvm.used`` global is an array which has
3415 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3416 pointers to named global variables, functions and aliases which may optionally
3417 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3420 .. code-block:: llvm
3425 @llvm.used = appending global [2 x i8*] [
3427 i8* bitcast (i32* @Y to i8*)
3428 ], section "llvm.metadata"
3430 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3431 and linker are required to treat the symbol as if there is a reference to the
3432 symbol that it cannot see (which is why they have to be named). For example, if
3433 a variable has internal linkage and no references other than that from the
3434 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3435 references from inline asms and other things the compiler cannot "see", and
3436 corresponds to "``attribute((used))``" in GNU C.
3438 On some targets, the code generator must emit a directive to the
3439 assembler or object file to prevent the assembler and linker from
3440 molesting the symbol.
3442 .. _gv_llvmcompilerused:
3444 The '``llvm.compiler.used``' Global Variable
3445 --------------------------------------------
3447 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3448 directive, except that it only prevents the compiler from touching the
3449 symbol. On targets that support it, this allows an intelligent linker to
3450 optimize references to the symbol without being impeded as it would be
3453 This is a rare construct that should only be used in rare circumstances,
3454 and should not be exposed to source languages.
3456 .. _gv_llvmglobalctors:
3458 The '``llvm.global_ctors``' Global Variable
3459 -------------------------------------------
3461 .. code-block:: llvm
3463 %0 = type { i32, void ()*, i8* }
3464 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
3466 The ``@llvm.global_ctors`` array contains a list of constructor
3467 functions, priorities, and an optional associated global or function.
3468 The functions referenced by this array will be called in ascending order
3469 of priority (i.e. lowest first) when the module is loaded. The order of
3470 functions with the same priority is not defined.
3472 If the third field is present, non-null, and points to a global variable
3473 or function, the initializer function will only run if the associated
3474 data from the current module is not discarded.
3476 .. _llvmglobaldtors:
3478 The '``llvm.global_dtors``' Global Variable
3479 -------------------------------------------
3481 .. code-block:: llvm
3483 %0 = type { i32, void ()*, i8* }
3484 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
3486 The ``@llvm.global_dtors`` array contains a list of destructor
3487 functions, priorities, and an optional associated global or function.
3488 The functions referenced by this array will be called in descending
3489 order of priority (i.e. highest first) when the module is unloaded. The
3490 order of functions with the same priority is not defined.
3492 If the third field is present, non-null, and points to a global variable
3493 or function, the destructor function will only run if the associated
3494 data from the current module is not discarded.
3496 Instruction Reference
3497 =====================
3499 The LLVM instruction set consists of several different classifications
3500 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3501 instructions <binaryops>`, :ref:`bitwise binary
3502 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3503 :ref:`other instructions <otherops>`.
3507 Terminator Instructions
3508 -----------------------
3510 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3511 program ends with a "Terminator" instruction, which indicates which
3512 block should be executed after the current block is finished. These
3513 terminator instructions typically yield a '``void``' value: they produce
3514 control flow, not values (the one exception being the
3515 ':ref:`invoke <i_invoke>`' instruction).
3517 The terminator instructions are: ':ref:`ret <i_ret>`',
3518 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3519 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3520 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3524 '``ret``' Instruction
3525 ^^^^^^^^^^^^^^^^^^^^^
3532 ret <type> <value> ; Return a value from a non-void function
3533 ret void ; Return from void function
3538 The '``ret``' instruction is used to return control flow (and optionally
3539 a value) from a function back to the caller.
3541 There are two forms of the '``ret``' instruction: one that returns a
3542 value and then causes control flow, and one that just causes control
3548 The '``ret``' instruction optionally accepts a single argument, the
3549 return value. The type of the return value must be a ':ref:`first
3550 class <t_firstclass>`' type.
3552 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3553 return type and contains a '``ret``' instruction with no return value or
3554 a return value with a type that does not match its type, or if it has a
3555 void return type and contains a '``ret``' instruction with a return
3561 When the '``ret``' instruction is executed, control flow returns back to
3562 the calling function's context. If the caller is a
3563 ":ref:`call <i_call>`" instruction, execution continues at the
3564 instruction after the call. If the caller was an
3565 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3566 beginning of the "normal" destination block. If the instruction returns
3567 a value, that value shall set the call or invoke instruction's return
3573 .. code-block:: llvm
3575 ret i32 5 ; Return an integer value of 5
3576 ret void ; Return from a void function
3577 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3581 '``br``' Instruction
3582 ^^^^^^^^^^^^^^^^^^^^
3589 br i1 <cond>, label <iftrue>, label <iffalse>
3590 br label <dest> ; Unconditional branch
3595 The '``br``' instruction is used to cause control flow to transfer to a
3596 different basic block in the current function. There are two forms of
3597 this instruction, corresponding to a conditional branch and an
3598 unconditional branch.
3603 The conditional branch form of the '``br``' instruction takes a single
3604 '``i1``' value and two '``label``' values. The unconditional form of the
3605 '``br``' instruction takes a single '``label``' value as a target.
3610 Upon execution of a conditional '``br``' instruction, the '``i1``'
3611 argument is evaluated. If the value is ``true``, control flows to the
3612 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3613 to the '``iffalse``' ``label`` argument.
3618 .. code-block:: llvm
3621 %cond = icmp eq i32 %a, %b
3622 br i1 %cond, label %IfEqual, label %IfUnequal
3630 '``switch``' Instruction
3631 ^^^^^^^^^^^^^^^^^^^^^^^^
3638 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3643 The '``switch``' instruction is used to transfer control flow to one of
3644 several different places. It is a generalization of the '``br``'
3645 instruction, allowing a branch to occur to one of many possible
3651 The '``switch``' instruction uses three parameters: an integer
3652 comparison value '``value``', a default '``label``' destination, and an
3653 array of pairs of comparison value constants and '``label``'s. The table
3654 is not allowed to contain duplicate constant entries.
3659 The ``switch`` instruction specifies a table of values and destinations.
3660 When the '``switch``' instruction is executed, this table is searched
3661 for the given value. If the value is found, control flow is transferred
3662 to the corresponding destination; otherwise, control flow is transferred
3663 to the default destination.
3668 Depending on properties of the target machine and the particular
3669 ``switch`` instruction, this instruction may be code generated in
3670 different ways. For example, it could be generated as a series of
3671 chained conditional branches or with a lookup table.
3676 .. code-block:: llvm
3678 ; Emulate a conditional br instruction
3679 %Val = zext i1 %value to i32
3680 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3682 ; Emulate an unconditional br instruction
3683 switch i32 0, label %dest [ ]
3685 ; Implement a jump table:
3686 switch i32 %val, label %otherwise [ i32 0, label %onzero
3688 i32 2, label %ontwo ]
3692 '``indirectbr``' Instruction
3693 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3700 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3705 The '``indirectbr``' instruction implements an indirect branch to a
3706 label within the current function, whose address is specified by
3707 "``address``". Address must be derived from a
3708 :ref:`blockaddress <blockaddress>` constant.
3713 The '``address``' argument is the address of the label to jump to. The
3714 rest of the arguments indicate the full set of possible destinations
3715 that the address may point to. Blocks are allowed to occur multiple
3716 times in the destination list, though this isn't particularly useful.
3718 This destination list is required so that dataflow analysis has an
3719 accurate understanding of the CFG.
3724 Control transfers to the block specified in the address argument. All
3725 possible destination blocks must be listed in the label list, otherwise
3726 this instruction has undefined behavior. This implies that jumps to
3727 labels defined in other functions have undefined behavior as well.
3732 This is typically implemented with a jump through a register.
3737 .. code-block:: llvm
3739 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3743 '``invoke``' Instruction
3744 ^^^^^^^^^^^^^^^^^^^^^^^^
3751 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3752 to label <normal label> unwind label <exception label>
3757 The '``invoke``' instruction causes control to transfer to a specified
3758 function, with the possibility of control flow transfer to either the
3759 '``normal``' label or the '``exception``' label. If the callee function
3760 returns with the "``ret``" instruction, control flow will return to the
3761 "normal" label. If the callee (or any indirect callees) returns via the
3762 ":ref:`resume <i_resume>`" instruction or other exception handling
3763 mechanism, control is interrupted and continued at the dynamically
3764 nearest "exception" label.
3766 The '``exception``' label is a `landing
3767 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3768 '``exception``' label is required to have the
3769 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3770 information about the behavior of the program after unwinding happens,
3771 as its first non-PHI instruction. The restrictions on the
3772 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3773 instruction, so that the important information contained within the
3774 "``landingpad``" instruction can't be lost through normal code motion.
3779 This instruction requires several arguments:
3781 #. The optional "cconv" marker indicates which :ref:`calling
3782 convention <callingconv>` the call should use. If none is
3783 specified, the call defaults to using C calling conventions.
3784 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3785 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3787 #. '``ptr to function ty``': shall be the signature of the pointer to
3788 function value being invoked. In most cases, this is a direct
3789 function invocation, but indirect ``invoke``'s are just as possible,
3790 branching off an arbitrary pointer to function value.
3791 #. '``function ptr val``': An LLVM value containing a pointer to a
3792 function to be invoked.
3793 #. '``function args``': argument list whose types match the function
3794 signature argument types and parameter attributes. All arguments must
3795 be of :ref:`first class <t_firstclass>` type. If the function signature
3796 indicates the function accepts a variable number of arguments, the
3797 extra arguments can be specified.
3798 #. '``normal label``': the label reached when the called function
3799 executes a '``ret``' instruction.
3800 #. '``exception label``': the label reached when a callee returns via
3801 the :ref:`resume <i_resume>` instruction or other exception handling
3803 #. The optional :ref:`function attributes <fnattrs>` list. Only
3804 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3805 attributes are valid here.
3810 This instruction is designed to operate as a standard '``call``'
3811 instruction in most regards. The primary difference is that it
3812 establishes an association with a label, which is used by the runtime
3813 library to unwind the stack.
3815 This instruction is used in languages with destructors to ensure that
3816 proper cleanup is performed in the case of either a ``longjmp`` or a
3817 thrown exception. Additionally, this is important for implementation of
3818 '``catch``' clauses in high-level languages that support them.
3820 For the purposes of the SSA form, the definition of the value returned
3821 by the '``invoke``' instruction is deemed to occur on the edge from the
3822 current block to the "normal" label. If the callee unwinds then no
3823 return value is available.
3828 .. code-block:: llvm
3830 %retval = invoke i32 @Test(i32 15) to label %Continue
3831 unwind label %TestCleanup ; i32:retval set
3832 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3833 unwind label %TestCleanup ; i32:retval set
3837 '``resume``' Instruction
3838 ^^^^^^^^^^^^^^^^^^^^^^^^
3845 resume <type> <value>
3850 The '``resume``' instruction is a terminator instruction that has no
3856 The '``resume``' instruction requires one argument, which must have the
3857 same type as the result of any '``landingpad``' instruction in the same
3863 The '``resume``' instruction resumes propagation of an existing
3864 (in-flight) exception whose unwinding was interrupted with a
3865 :ref:`landingpad <i_landingpad>` instruction.
3870 .. code-block:: llvm
3872 resume { i8*, i32 } %exn
3876 '``unreachable``' Instruction
3877 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3889 The '``unreachable``' instruction has no defined semantics. This
3890 instruction is used to inform the optimizer that a particular portion of
3891 the code is not reachable. This can be used to indicate that the code
3892 after a no-return function cannot be reached, and other facts.
3897 The '``unreachable``' instruction has no defined semantics.
3904 Binary operators are used to do most of the computation in a program.
3905 They require two operands of the same type, execute an operation on
3906 them, and produce a single value. The operands might represent multiple
3907 data, as is the case with the :ref:`vector <t_vector>` data type. The
3908 result value has the same type as its operands.
3910 There are several different binary operators:
3914 '``add``' Instruction
3915 ^^^^^^^^^^^^^^^^^^^^^
3922 <result> = add <ty> <op1>, <op2> ; yields ty:result
3923 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
3924 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
3925 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
3930 The '``add``' instruction returns the sum of its two operands.
3935 The two arguments to the '``add``' instruction must be
3936 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3937 arguments must have identical types.
3942 The value produced is the integer sum of the two operands.
3944 If the sum has unsigned overflow, the result returned is the
3945 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3948 Because LLVM integers use a two's complement representation, this
3949 instruction is appropriate for both signed and unsigned integers.
3951 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3952 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3953 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3954 unsigned and/or signed overflow, respectively, occurs.
3959 .. code-block:: llvm
3961 <result> = add i32 4, %var ; yields i32:result = 4 + %var
3965 '``fadd``' Instruction
3966 ^^^^^^^^^^^^^^^^^^^^^^
3973 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
3978 The '``fadd``' instruction returns the sum of its two operands.
3983 The two arguments to the '``fadd``' instruction must be :ref:`floating
3984 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3985 Both arguments must have identical types.
3990 The value produced is the floating point sum of the two operands. This
3991 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3992 which are optimization hints to enable otherwise unsafe floating point
3998 .. code-block:: llvm
4000 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
4002 '``sub``' Instruction
4003 ^^^^^^^^^^^^^^^^^^^^^
4010 <result> = sub <ty> <op1>, <op2> ; yields ty:result
4011 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
4012 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
4013 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
4018 The '``sub``' instruction returns the difference of its two operands.
4020 Note that the '``sub``' instruction is used to represent the '``neg``'
4021 instruction present in most other intermediate representations.
4026 The two arguments to the '``sub``' instruction must be
4027 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4028 arguments must have identical types.
4033 The value produced is the integer difference of the two operands.
4035 If the difference has unsigned overflow, the result returned is the
4036 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4039 Because LLVM integers use a two's complement representation, this
4040 instruction is appropriate for both signed and unsigned integers.
4042 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4043 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4044 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
4045 unsigned and/or signed overflow, respectively, occurs.
4050 .. code-block:: llvm
4052 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
4053 <result> = sub i32 0, %val ; yields i32:result = -%var
4057 '``fsub``' Instruction
4058 ^^^^^^^^^^^^^^^^^^^^^^
4065 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4070 The '``fsub``' instruction returns the difference of its two operands.
4072 Note that the '``fsub``' instruction is used to represent the '``fneg``'
4073 instruction present in most other intermediate representations.
4078 The two arguments to the '``fsub``' instruction must be :ref:`floating
4079 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4080 Both arguments must have identical types.
4085 The value produced is the floating point difference of the two operands.
4086 This instruction can also take any number of :ref:`fast-math
4087 flags <fastmath>`, which are optimization hints to enable otherwise
4088 unsafe floating point optimizations:
4093 .. code-block:: llvm
4095 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
4096 <result> = fsub float -0.0, %val ; yields float:result = -%var
4098 '``mul``' Instruction
4099 ^^^^^^^^^^^^^^^^^^^^^
4106 <result> = mul <ty> <op1>, <op2> ; yields ty:result
4107 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
4108 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
4109 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
4114 The '``mul``' instruction returns the product of its two operands.
4119 The two arguments to the '``mul``' instruction must be
4120 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4121 arguments must have identical types.
4126 The value produced is the integer product of the two operands.
4128 If the result of the multiplication has unsigned overflow, the result
4129 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
4130 bit width of the result.
4132 Because LLVM integers use a two's complement representation, and the
4133 result is the same width as the operands, this instruction returns the
4134 correct result for both signed and unsigned integers. If a full product
4135 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
4136 sign-extended or zero-extended as appropriate to the width of the full
4139 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4140 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4141 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
4142 unsigned and/or signed overflow, respectively, occurs.
4147 .. code-block:: llvm
4149 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
4153 '``fmul``' Instruction
4154 ^^^^^^^^^^^^^^^^^^^^^^
4161 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4166 The '``fmul``' instruction returns the product of its two operands.
4171 The two arguments to the '``fmul``' instruction must be :ref:`floating
4172 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4173 Both arguments must have identical types.
4178 The value produced is the floating point product of the two operands.
4179 This instruction can also take any number of :ref:`fast-math
4180 flags <fastmath>`, which are optimization hints to enable otherwise
4181 unsafe floating point optimizations:
4186 .. code-block:: llvm
4188 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
4190 '``udiv``' Instruction
4191 ^^^^^^^^^^^^^^^^^^^^^^
4198 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
4199 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
4204 The '``udiv``' instruction returns the quotient of its two operands.
4209 The two arguments to the '``udiv``' instruction must be
4210 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4211 arguments must have identical types.
4216 The value produced is the unsigned integer quotient of the two operands.
4218 Note that unsigned integer division and signed integer division are
4219 distinct operations; for signed integer division, use '``sdiv``'.
4221 Division by zero leads to undefined behavior.
4223 If the ``exact`` keyword is present, the result value of the ``udiv`` is
4224 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
4225 such, "((a udiv exact b) mul b) == a").
4230 .. code-block:: llvm
4232 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
4234 '``sdiv``' Instruction
4235 ^^^^^^^^^^^^^^^^^^^^^^
4242 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
4243 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
4248 The '``sdiv``' instruction returns the quotient of its two operands.
4253 The two arguments to the '``sdiv``' instruction must be
4254 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4255 arguments must have identical types.
4260 The value produced is the signed integer quotient of the two operands
4261 rounded towards zero.
4263 Note that signed integer division and unsigned integer division are
4264 distinct operations; for unsigned integer division, use '``udiv``'.
4266 Division by zero leads to undefined behavior. Overflow also leads to
4267 undefined behavior; this is a rare case, but can occur, for example, by
4268 doing a 32-bit division of -2147483648 by -1.
4270 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4271 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4276 .. code-block:: llvm
4278 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
4282 '``fdiv``' Instruction
4283 ^^^^^^^^^^^^^^^^^^^^^^
4290 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4295 The '``fdiv``' instruction returns the quotient of its two operands.
4300 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4301 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4302 Both arguments must have identical types.
4307 The value produced is the floating point quotient of the two operands.
4308 This instruction can also take any number of :ref:`fast-math
4309 flags <fastmath>`, which are optimization hints to enable otherwise
4310 unsafe floating point optimizations:
4315 .. code-block:: llvm
4317 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
4319 '``urem``' Instruction
4320 ^^^^^^^^^^^^^^^^^^^^^^
4327 <result> = urem <ty> <op1>, <op2> ; yields ty:result
4332 The '``urem``' instruction returns the remainder from the unsigned
4333 division of its two arguments.
4338 The two arguments to the '``urem``' instruction must be
4339 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4340 arguments must have identical types.
4345 This instruction returns the unsigned integer *remainder* of a division.
4346 This instruction always performs an unsigned division to get the
4349 Note that unsigned integer remainder and signed integer remainder are
4350 distinct operations; for signed integer remainder, use '``srem``'.
4352 Taking the remainder of a division by zero leads to undefined behavior.
4357 .. code-block:: llvm
4359 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
4361 '``srem``' Instruction
4362 ^^^^^^^^^^^^^^^^^^^^^^
4369 <result> = srem <ty> <op1>, <op2> ; yields ty:result
4374 The '``srem``' instruction returns the remainder from the signed
4375 division of its two operands. This instruction can also take
4376 :ref:`vector <t_vector>` versions of the values in which case the elements
4382 The two arguments to the '``srem``' instruction must be
4383 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4384 arguments must have identical types.
4389 This instruction returns the *remainder* of a division (where the result
4390 is either zero or has the same sign as the dividend, ``op1``), not the
4391 *modulo* operator (where the result is either zero or has the same sign
4392 as the divisor, ``op2``) of a value. For more information about the
4393 difference, see `The Math
4394 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4395 table of how this is implemented in various languages, please see
4397 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4399 Note that signed integer remainder and unsigned integer remainder are
4400 distinct operations; for unsigned integer remainder, use '``urem``'.
4402 Taking the remainder of a division by zero leads to undefined behavior.
4403 Overflow also leads to undefined behavior; this is a rare case, but can
4404 occur, for example, by taking the remainder of a 32-bit division of
4405 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4406 rule lets srem be implemented using instructions that return both the
4407 result of the division and the remainder.)
4412 .. code-block:: llvm
4414 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
4418 '``frem``' Instruction
4419 ^^^^^^^^^^^^^^^^^^^^^^
4426 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4431 The '``frem``' instruction returns the remainder from the division of
4437 The two arguments to the '``frem``' instruction must be :ref:`floating
4438 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4439 Both arguments must have identical types.
4444 This instruction returns the *remainder* of a division. The remainder
4445 has the same sign as the dividend. This instruction can also take any
4446 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4447 to enable otherwise unsafe floating point optimizations:
4452 .. code-block:: llvm
4454 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
4458 Bitwise Binary Operations
4459 -------------------------
4461 Bitwise binary operators are used to do various forms of bit-twiddling
4462 in a program. They are generally very efficient instructions and can
4463 commonly be strength reduced from other instructions. They require two
4464 operands of the same type, execute an operation on them, and produce a
4465 single value. The resulting value is the same type as its operands.
4467 '``shl``' Instruction
4468 ^^^^^^^^^^^^^^^^^^^^^
4475 <result> = shl <ty> <op1>, <op2> ; yields ty:result
4476 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
4477 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
4478 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
4483 The '``shl``' instruction returns the first operand shifted to the left
4484 a specified number of bits.
4489 Both arguments to the '``shl``' instruction must be the same
4490 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4491 '``op2``' is treated as an unsigned value.
4496 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4497 where ``n`` is the width of the result. If ``op2`` is (statically or
4498 dynamically) negative or equal to or larger than the number of bits in
4499 ``op1``, the result is undefined. If the arguments are vectors, each
4500 vector element of ``op1`` is shifted by the corresponding shift amount
4503 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4504 value <poisonvalues>` if it shifts out any non-zero bits. If the
4505 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4506 value <poisonvalues>` if it shifts out any bits that disagree with the
4507 resultant sign bit. As such, NUW/NSW have the same semantics as they
4508 would if the shift were expressed as a mul instruction with the same
4509 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4514 .. code-block:: llvm
4516 <result> = shl i32 4, %var ; yields i32: 4 << %var
4517 <result> = shl i32 4, 2 ; yields i32: 16
4518 <result> = shl i32 1, 10 ; yields i32: 1024
4519 <result> = shl i32 1, 32 ; undefined
4520 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4522 '``lshr``' Instruction
4523 ^^^^^^^^^^^^^^^^^^^^^^
4530 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
4531 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
4536 The '``lshr``' instruction (logical shift right) returns the first
4537 operand shifted to the right a specified number of bits with zero fill.
4542 Both arguments to the '``lshr``' instruction must be the same
4543 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4544 '``op2``' is treated as an unsigned value.
4549 This instruction always performs a logical shift right operation. The
4550 most significant bits of the result will be filled with zero bits after
4551 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4552 than the number of bits in ``op1``, the result is undefined. If the
4553 arguments are vectors, each vector element of ``op1`` is shifted by the
4554 corresponding shift amount in ``op2``.
4556 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4557 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4563 .. code-block:: llvm
4565 <result> = lshr i32 4, 1 ; yields i32:result = 2
4566 <result> = lshr i32 4, 2 ; yields i32:result = 1
4567 <result> = lshr i8 4, 3 ; yields i8:result = 0
4568 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
4569 <result> = lshr i32 1, 32 ; undefined
4570 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4572 '``ashr``' Instruction
4573 ^^^^^^^^^^^^^^^^^^^^^^
4580 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
4581 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
4586 The '``ashr``' instruction (arithmetic shift right) returns the first
4587 operand shifted to the right a specified number of bits with sign
4593 Both arguments to the '``ashr``' instruction must be the same
4594 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4595 '``op2``' is treated as an unsigned value.
4600 This instruction always performs an arithmetic shift right operation,
4601 The most significant bits of the result will be filled with the sign bit
4602 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4603 than the number of bits in ``op1``, the result is undefined. If the
4604 arguments are vectors, each vector element of ``op1`` is shifted by the
4605 corresponding shift amount in ``op2``.
4607 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4608 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4614 .. code-block:: llvm
4616 <result> = ashr i32 4, 1 ; yields i32:result = 2
4617 <result> = ashr i32 4, 2 ; yields i32:result = 1
4618 <result> = ashr i8 4, 3 ; yields i8:result = 0
4619 <result> = ashr i8 -2, 1 ; yields i8:result = -1
4620 <result> = ashr i32 1, 32 ; undefined
4621 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4623 '``and``' Instruction
4624 ^^^^^^^^^^^^^^^^^^^^^
4631 <result> = and <ty> <op1>, <op2> ; yields ty:result
4636 The '``and``' instruction returns the bitwise logical and of its two
4642 The two arguments to the '``and``' instruction must be
4643 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4644 arguments must have identical types.
4649 The truth table used for the '``and``' instruction is:
4666 .. code-block:: llvm
4668 <result> = and i32 4, %var ; yields i32:result = 4 & %var
4669 <result> = and i32 15, 40 ; yields i32:result = 8
4670 <result> = and i32 4, 8 ; yields i32:result = 0
4672 '``or``' Instruction
4673 ^^^^^^^^^^^^^^^^^^^^
4680 <result> = or <ty> <op1>, <op2> ; yields ty:result
4685 The '``or``' instruction returns the bitwise logical inclusive or of its
4691 The two arguments to the '``or``' instruction must be
4692 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4693 arguments must have identical types.
4698 The truth table used for the '``or``' instruction is:
4717 <result> = or i32 4, %var ; yields i32:result = 4 | %var
4718 <result> = or i32 15, 40 ; yields i32:result = 47
4719 <result> = or i32 4, 8 ; yields i32:result = 12
4721 '``xor``' Instruction
4722 ^^^^^^^^^^^^^^^^^^^^^
4729 <result> = xor <ty> <op1>, <op2> ; yields ty:result
4734 The '``xor``' instruction returns the bitwise logical exclusive or of
4735 its two operands. The ``xor`` is used to implement the "one's
4736 complement" operation, which is the "~" operator in C.
4741 The two arguments to the '``xor``' instruction must be
4742 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4743 arguments must have identical types.
4748 The truth table used for the '``xor``' instruction is:
4765 .. code-block:: llvm
4767 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
4768 <result> = xor i32 15, 40 ; yields i32:result = 39
4769 <result> = xor i32 4, 8 ; yields i32:result = 12
4770 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
4775 LLVM supports several instructions to represent vector operations in a
4776 target-independent manner. These instructions cover the element-access
4777 and vector-specific operations needed to process vectors effectively.
4778 While LLVM does directly support these vector operations, many
4779 sophisticated algorithms will want to use target-specific intrinsics to
4780 take full advantage of a specific target.
4782 .. _i_extractelement:
4784 '``extractelement``' Instruction
4785 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4792 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
4797 The '``extractelement``' instruction extracts a single scalar element
4798 from a vector at a specified index.
4803 The first operand of an '``extractelement``' instruction is a value of
4804 :ref:`vector <t_vector>` type. The second operand is an index indicating
4805 the position from which to extract the element. The index may be a
4806 variable of any integer type.
4811 The result is a scalar of the same type as the element type of ``val``.
4812 Its value is the value at position ``idx`` of ``val``. If ``idx``
4813 exceeds the length of ``val``, the results are undefined.
4818 .. code-block:: llvm
4820 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4822 .. _i_insertelement:
4824 '``insertelement``' Instruction
4825 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4832 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
4837 The '``insertelement``' instruction inserts a scalar element into a
4838 vector at a specified index.
4843 The first operand of an '``insertelement``' instruction is a value of
4844 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4845 type must equal the element type of the first operand. The third operand
4846 is an index indicating the position at which to insert the value. The
4847 index may be a variable of any integer type.
4852 The result is a vector of the same type as ``val``. Its element values
4853 are those of ``val`` except at position ``idx``, where it gets the value
4854 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4860 .. code-block:: llvm
4862 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4864 .. _i_shufflevector:
4866 '``shufflevector``' Instruction
4867 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4874 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4879 The '``shufflevector``' instruction constructs a permutation of elements
4880 from two input vectors, returning a vector with the same element type as
4881 the input and length that is the same as the shuffle mask.
4886 The first two operands of a '``shufflevector``' instruction are vectors
4887 with the same type. The third argument is a shuffle mask whose element
4888 type is always 'i32'. The result of the instruction is a vector whose
4889 length is the same as the shuffle mask and whose element type is the
4890 same as the element type of the first two operands.
4892 The shuffle mask operand is required to be a constant vector with either
4893 constant integer or undef values.
4898 The elements of the two input vectors are numbered from left to right
4899 across both of the vectors. The shuffle mask operand specifies, for each
4900 element of the result vector, which element of the two input vectors the
4901 result element gets. The element selector may be undef (meaning "don't
4902 care") and the second operand may be undef if performing a shuffle from
4908 .. code-block:: llvm
4910 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4911 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4912 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4913 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4914 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4915 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4916 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4917 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4919 Aggregate Operations
4920 --------------------
4922 LLVM supports several instructions for working with
4923 :ref:`aggregate <t_aggregate>` values.
4927 '``extractvalue``' Instruction
4928 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4935 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4940 The '``extractvalue``' instruction extracts the value of a member field
4941 from an :ref:`aggregate <t_aggregate>` value.
4946 The first operand of an '``extractvalue``' instruction is a value of
4947 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4948 constant indices to specify which value to extract in a similar manner
4949 as indices in a '``getelementptr``' instruction.
4951 The major differences to ``getelementptr`` indexing are:
4953 - Since the value being indexed is not a pointer, the first index is
4954 omitted and assumed to be zero.
4955 - At least one index must be specified.
4956 - Not only struct indices but also array indices must be in bounds.
4961 The result is the value at the position in the aggregate specified by
4967 .. code-block:: llvm
4969 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4973 '``insertvalue``' Instruction
4974 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4981 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4986 The '``insertvalue``' instruction inserts a value into a member field in
4987 an :ref:`aggregate <t_aggregate>` value.
4992 The first operand of an '``insertvalue``' instruction is a value of
4993 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4994 a first-class value to insert. The following operands are constant
4995 indices indicating the position at which to insert the value in a
4996 similar manner as indices in a '``extractvalue``' instruction. The value
4997 to insert must have the same type as the value identified by the
5003 The result is an aggregate of the same type as ``val``. Its value is
5004 that of ``val`` except that the value at the position specified by the
5005 indices is that of ``elt``.
5010 .. code-block:: llvm
5012 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
5013 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
5014 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
5018 Memory Access and Addressing Operations
5019 ---------------------------------------
5021 A key design point of an SSA-based representation is how it represents
5022 memory. In LLVM, no memory locations are in SSA form, which makes things
5023 very simple. This section describes how to read, write, and allocate
5028 '``alloca``' Instruction
5029 ^^^^^^^^^^^^^^^^^^^^^^^^
5036 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
5041 The '``alloca``' instruction allocates memory on the stack frame of the
5042 currently executing function, to be automatically released when this
5043 function returns to its caller. The object is always allocated in the
5044 generic address space (address space zero).
5049 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
5050 bytes of memory on the runtime stack, returning a pointer of the
5051 appropriate type to the program. If "NumElements" is specified, it is
5052 the number of elements allocated, otherwise "NumElements" is defaulted
5053 to be one. If a constant alignment is specified, the value result of the
5054 allocation is guaranteed to be aligned to at least that boundary. The
5055 alignment may not be greater than ``1 << 29``. If not specified, or if
5056 zero, the target can choose to align the allocation on any convenient
5057 boundary compatible with the type.
5059 '``type``' may be any sized type.
5064 Memory is allocated; a pointer is returned. The operation is undefined
5065 if there is insufficient stack space for the allocation. '``alloca``'d
5066 memory is automatically released when the function returns. The
5067 '``alloca``' instruction is commonly used to represent automatic
5068 variables that must have an address available. When the function returns
5069 (either with the ``ret`` or ``resume`` instructions), the memory is
5070 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5071 The order in which memory is allocated (ie., which way the stack grows)
5077 .. code-block:: llvm
5079 %ptr = alloca i32 ; yields i32*:ptr
5080 %ptr = alloca i32, i32 4 ; yields i32*:ptr
5081 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
5082 %ptr = alloca i32, align 1024 ; yields i32*:ptr
5086 '``load``' Instruction
5087 ^^^^^^^^^^^^^^^^^^^^^^
5094 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
5095 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5096 !<index> = !{ i32 1 }
5101 The '``load``' instruction is used to read from memory.
5106 The argument to the ``load`` instruction specifies the memory address
5107 from which to load. The pointer must point to a :ref:`first
5108 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
5109 then the optimizer is not allowed to modify the number or order of
5110 execution of this ``load`` with other :ref:`volatile
5111 operations <volatile>`.
5113 If the ``load`` is marked as ``atomic``, it takes an extra
5114 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5115 ``release`` and ``acq_rel`` orderings are not valid on ``load``
5116 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5117 when they may see multiple atomic stores. The type of the pointee must
5118 be an integer type whose bit width is a power of two greater than or
5119 equal to eight and less than or equal to a target-specific size limit.
5120 ``align`` must be explicitly specified on atomic loads, and the load has
5121 undefined behavior if the alignment is not set to a value which is at
5122 least the size in bytes of the pointee. ``!nontemporal`` does not have
5123 any defined semantics for atomic loads.
5125 The optional constant ``align`` argument specifies the alignment of the
5126 operation (that is, the alignment of the memory address). A value of 0
5127 or an omitted ``align`` argument means that the operation has the ABI
5128 alignment for the target. It is the responsibility of the code emitter
5129 to ensure that the alignment information is correct. Overestimating the
5130 alignment results in undefined behavior. Underestimating the alignment
5131 may produce less efficient code. An alignment of 1 is always safe. The
5132 maximum possible alignment is ``1 << 29``.
5134 The optional ``!nontemporal`` metadata must reference a single
5135 metadata name ``<index>`` corresponding to a metadata node with one
5136 ``i32`` entry of value 1. The existence of the ``!nontemporal``
5137 metadata on the instruction tells the optimizer and code generator
5138 that this load is not expected to be reused in the cache. The code
5139 generator may select special instructions to save cache bandwidth, such
5140 as the ``MOVNT`` instruction on x86.
5142 The optional ``!invariant.load`` metadata must reference a single
5143 metadata name ``<index>`` corresponding to a metadata node with no
5144 entries. The existence of the ``!invariant.load`` metadata on the
5145 instruction tells the optimizer and code generator that this load
5146 address points to memory which does not change value during program
5147 execution. The optimizer may then move this load around, for example, by
5148 hoisting it out of loops using loop invariant code motion.
5153 The location of memory pointed to is loaded. If the value being loaded
5154 is of scalar type then the number of bytes read does not exceed the
5155 minimum number of bytes needed to hold all bits of the type. For
5156 example, loading an ``i24`` reads at most three bytes. When loading a
5157 value of a type like ``i20`` with a size that is not an integral number
5158 of bytes, the result is undefined if the value was not originally
5159 written using a store of the same type.
5164 .. code-block:: llvm
5166 %ptr = alloca i32 ; yields i32*:ptr
5167 store i32 3, i32* %ptr ; yields void
5168 %val = load i32* %ptr ; yields i32:val = i32 3
5172 '``store``' Instruction
5173 ^^^^^^^^^^^^^^^^^^^^^^^
5180 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
5181 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
5186 The '``store``' instruction is used to write to memory.
5191 There are two arguments to the ``store`` instruction: a value to store
5192 and an address at which to store it. The type of the ``<pointer>``
5193 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
5194 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
5195 then the optimizer is not allowed to modify the number or order of
5196 execution of this ``store`` with other :ref:`volatile
5197 operations <volatile>`.
5199 If the ``store`` is marked as ``atomic``, it takes an extra
5200 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5201 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
5202 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5203 when they may see multiple atomic stores. The type of the pointee must
5204 be an integer type whose bit width is a power of two greater than or
5205 equal to eight and less than or equal to a target-specific size limit.
5206 ``align`` must be explicitly specified on atomic stores, and the store
5207 has undefined behavior if the alignment is not set to a value which is
5208 at least the size in bytes of the pointee. ``!nontemporal`` does not
5209 have any defined semantics for atomic stores.
5211 The optional constant ``align`` argument specifies the alignment of the
5212 operation (that is, the alignment of the memory address). A value of 0
5213 or an omitted ``align`` argument means that the operation has the ABI
5214 alignment for the target. It is the responsibility of the code emitter
5215 to ensure that the alignment information is correct. Overestimating the
5216 alignment results in undefined behavior. Underestimating the
5217 alignment may produce less efficient code. An alignment of 1 is always
5218 safe. The maximum possible alignment is ``1 << 29``.
5220 The optional ``!nontemporal`` metadata must reference a single metadata
5221 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
5222 value 1. The existence of the ``!nontemporal`` metadata on the instruction
5223 tells the optimizer and code generator that this load is not expected to
5224 be reused in the cache. The code generator may select special
5225 instructions to save cache bandwidth, such as the MOVNT instruction on
5231 The contents of memory are updated to contain ``<value>`` at the
5232 location specified by the ``<pointer>`` operand. If ``<value>`` is
5233 of scalar type then the number of bytes written does not exceed the
5234 minimum number of bytes needed to hold all bits of the type. For
5235 example, storing an ``i24`` writes at most three bytes. When writing a
5236 value of a type like ``i20`` with a size that is not an integral number
5237 of bytes, it is unspecified what happens to the extra bits that do not
5238 belong to the type, but they will typically be overwritten.
5243 .. code-block:: llvm
5245 %ptr = alloca i32 ; yields i32*:ptr
5246 store i32 3, i32* %ptr ; yields void
5247 %val = load i32* %ptr ; yields i32:val = i32 3
5251 '``fence``' Instruction
5252 ^^^^^^^^^^^^^^^^^^^^^^^
5259 fence [singlethread] <ordering> ; yields void
5264 The '``fence``' instruction is used to introduce happens-before edges
5270 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5271 defines what *synchronizes-with* edges they add. They can only be given
5272 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5277 A fence A which has (at least) ``release`` ordering semantics
5278 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5279 semantics if and only if there exist atomic operations X and Y, both
5280 operating on some atomic object M, such that A is sequenced before X, X
5281 modifies M (either directly or through some side effect of a sequence
5282 headed by X), Y is sequenced before B, and Y observes M. This provides a
5283 *happens-before* dependency between A and B. Rather than an explicit
5284 ``fence``, one (but not both) of the atomic operations X or Y might
5285 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5286 still *synchronize-with* the explicit ``fence`` and establish the
5287 *happens-before* edge.
5289 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5290 ``acquire`` and ``release`` semantics specified above, participates in
5291 the global program order of other ``seq_cst`` operations and/or fences.
5293 The optional ":ref:`singlethread <singlethread>`" argument specifies
5294 that the fence only synchronizes with other fences in the same thread.
5295 (This is useful for interacting with signal handlers.)
5300 .. code-block:: llvm
5302 fence acquire ; yields void
5303 fence singlethread seq_cst ; yields void
5307 '``cmpxchg``' Instruction
5308 ^^^^^^^^^^^^^^^^^^^^^^^^^
5315 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
5320 The '``cmpxchg``' instruction is used to atomically modify memory. It
5321 loads a value in memory and compares it to a given value. If they are
5322 equal, it tries to store a new value into the memory.
5327 There are three arguments to the '``cmpxchg``' instruction: an address
5328 to operate on, a value to compare to the value currently be at that
5329 address, and a new value to place at that address if the compared values
5330 are equal. The type of '<cmp>' must be an integer type whose bit width
5331 is a power of two greater than or equal to eight and less than or equal
5332 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5333 type, and the type of '<pointer>' must be a pointer to that type. If the
5334 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5335 to modify the number or order of execution of this ``cmpxchg`` with
5336 other :ref:`volatile operations <volatile>`.
5338 The success and failure :ref:`ordering <ordering>` arguments specify how this
5339 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
5340 must be at least ``monotonic``, the ordering constraint on failure must be no
5341 stronger than that on success, and the failure ordering cannot be either
5342 ``release`` or ``acq_rel``.
5344 The optional "``singlethread``" argument declares that the ``cmpxchg``
5345 is only atomic with respect to code (usually signal handlers) running in
5346 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5347 respect to all other code in the system.
5349 The pointer passed into cmpxchg must have alignment greater than or
5350 equal to the size in memory of the operand.
5355 The contents of memory at the location specified by the '``<pointer>``' operand
5356 is read and compared to '``<cmp>``'; if the read value is the equal, the
5357 '``<new>``' is written. The original value at the location is returned, together
5358 with a flag indicating success (true) or failure (false).
5360 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
5361 permitted: the operation may not write ``<new>`` even if the comparison
5364 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
5365 if the value loaded equals ``cmp``.
5367 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5368 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5369 load with an ordering parameter determined the second ordering parameter.
5374 .. code-block:: llvm
5377 %orig = atomic load i32* %ptr unordered ; yields i32
5381 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5382 %squared = mul i32 %cmp, %cmp
5383 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
5384 %value_loaded = extractvalue { i32, i1 } %val_success, 0
5385 %success = extractvalue { i32, i1 } %val_success, 1
5386 br i1 %success, label %done, label %loop
5393 '``atomicrmw``' Instruction
5394 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5401 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
5406 The '``atomicrmw``' instruction is used to atomically modify memory.
5411 There are three arguments to the '``atomicrmw``' instruction: an
5412 operation to apply, an address whose value to modify, an argument to the
5413 operation. The operation must be one of the following keywords:
5427 The type of '<value>' must be an integer type whose bit width is a power
5428 of two greater than or equal to eight and less than or equal to a
5429 target-specific size limit. The type of the '``<pointer>``' operand must
5430 be a pointer to that type. If the ``atomicrmw`` is marked as
5431 ``volatile``, then the optimizer is not allowed to modify the number or
5432 order of execution of this ``atomicrmw`` with other :ref:`volatile
5433 operations <volatile>`.
5438 The contents of memory at the location specified by the '``<pointer>``'
5439 operand are atomically read, modified, and written back. The original
5440 value at the location is returned. The modification is specified by the
5443 - xchg: ``*ptr = val``
5444 - add: ``*ptr = *ptr + val``
5445 - sub: ``*ptr = *ptr - val``
5446 - and: ``*ptr = *ptr & val``
5447 - nand: ``*ptr = ~(*ptr & val)``
5448 - or: ``*ptr = *ptr | val``
5449 - xor: ``*ptr = *ptr ^ val``
5450 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5451 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5452 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5454 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5460 .. code-block:: llvm
5462 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
5464 .. _i_getelementptr:
5466 '``getelementptr``' Instruction
5467 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5474 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5475 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5476 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5481 The '``getelementptr``' instruction is used to get the address of a
5482 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5483 address calculation only and does not access memory.
5488 The first argument is always a pointer or a vector of pointers, and
5489 forms the basis of the calculation. The remaining arguments are indices
5490 that indicate which of the elements of the aggregate object are indexed.
5491 The interpretation of each index is dependent on the type being indexed
5492 into. The first index always indexes the pointer value given as the
5493 first argument, the second index indexes a value of the type pointed to
5494 (not necessarily the value directly pointed to, since the first index
5495 can be non-zero), etc. The first type indexed into must be a pointer
5496 value, subsequent types can be arrays, vectors, and structs. Note that
5497 subsequent types being indexed into can never be pointers, since that
5498 would require loading the pointer before continuing calculation.
5500 The type of each index argument depends on the type it is indexing into.
5501 When indexing into a (optionally packed) structure, only ``i32`` integer
5502 **constants** are allowed (when using a vector of indices they must all
5503 be the **same** ``i32`` integer constant). When indexing into an array,
5504 pointer or vector, integers of any width are allowed, and they are not
5505 required to be constant. These integers are treated as signed values
5508 For example, let's consider a C code fragment and how it gets compiled
5524 int *foo(struct ST *s) {
5525 return &s[1].Z.B[5][13];
5528 The LLVM code generated by Clang is:
5530 .. code-block:: llvm
5532 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5533 %struct.ST = type { i32, double, %struct.RT }
5535 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5537 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5544 In the example above, the first index is indexing into the
5545 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5546 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5547 indexes into the third element of the structure, yielding a
5548 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5549 structure. The third index indexes into the second element of the
5550 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5551 dimensions of the array are subscripted into, yielding an '``i32``'
5552 type. The '``getelementptr``' instruction returns a pointer to this
5553 element, thus computing a value of '``i32*``' type.
5555 Note that it is perfectly legal to index partially through a structure,
5556 returning a pointer to an inner element. Because of this, the LLVM code
5557 for the given testcase is equivalent to:
5559 .. code-block:: llvm
5561 define i32* @foo(%struct.ST* %s) {
5562 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5563 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5564 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5565 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5566 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5570 If the ``inbounds`` keyword is present, the result value of the
5571 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5572 pointer is not an *in bounds* address of an allocated object, or if any
5573 of the addresses that would be formed by successive addition of the
5574 offsets implied by the indices to the base address with infinitely
5575 precise signed arithmetic are not an *in bounds* address of that
5576 allocated object. The *in bounds* addresses for an allocated object are
5577 all the addresses that point into the object, plus the address one byte
5578 past the end. In cases where the base is a vector of pointers the
5579 ``inbounds`` keyword applies to each of the computations element-wise.
5581 If the ``inbounds`` keyword is not present, the offsets are added to the
5582 base address with silently-wrapping two's complement arithmetic. If the
5583 offsets have a different width from the pointer, they are sign-extended
5584 or truncated to the width of the pointer. The result value of the
5585 ``getelementptr`` may be outside the object pointed to by the base
5586 pointer. The result value may not necessarily be used to access memory
5587 though, even if it happens to point into allocated storage. See the
5588 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5591 The getelementptr instruction is often confusing. For some more insight
5592 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5597 .. code-block:: llvm
5599 ; yields [12 x i8]*:aptr
5600 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5602 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5604 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5606 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5608 In cases where the pointer argument is a vector of pointers, each index
5609 must be a vector with the same number of elements. For example:
5611 .. code-block:: llvm
5613 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5615 Conversion Operations
5616 ---------------------
5618 The instructions in this category are the conversion instructions
5619 (casting) which all take a single operand and a type. They perform
5620 various bit conversions on the operand.
5622 '``trunc .. to``' Instruction
5623 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5630 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5635 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5640 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5641 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5642 of the same number of integers. The bit size of the ``value`` must be
5643 larger than the bit size of the destination type, ``ty2``. Equal sized
5644 types are not allowed.
5649 The '``trunc``' instruction truncates the high order bits in ``value``
5650 and converts the remaining bits to ``ty2``. Since the source size must
5651 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5652 It will always truncate bits.
5657 .. code-block:: llvm
5659 %X = trunc i32 257 to i8 ; yields i8:1
5660 %Y = trunc i32 123 to i1 ; yields i1:true
5661 %Z = trunc i32 122 to i1 ; yields i1:false
5662 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5664 '``zext .. to``' Instruction
5665 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5672 <result> = zext <ty> <value> to <ty2> ; yields ty2
5677 The '``zext``' instruction zero extends its operand to type ``ty2``.
5682 The '``zext``' instruction takes a value to cast, and a type to cast it
5683 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5684 the same number of integers. The bit size of the ``value`` must be
5685 smaller than the bit size of the destination type, ``ty2``.
5690 The ``zext`` fills the high order bits of the ``value`` with zero bits
5691 until it reaches the size of the destination type, ``ty2``.
5693 When zero extending from i1, the result will always be either 0 or 1.
5698 .. code-block:: llvm
5700 %X = zext i32 257 to i64 ; yields i64:257
5701 %Y = zext i1 true to i32 ; yields i32:1
5702 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5704 '``sext .. to``' Instruction
5705 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5712 <result> = sext <ty> <value> to <ty2> ; yields ty2
5717 The '``sext``' sign extends ``value`` to the type ``ty2``.
5722 The '``sext``' instruction takes a value to cast, and a type to cast it
5723 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5724 the same number of integers. The bit size of the ``value`` must be
5725 smaller than the bit size of the destination type, ``ty2``.
5730 The '``sext``' instruction performs a sign extension by copying the sign
5731 bit (highest order bit) of the ``value`` until it reaches the bit size
5732 of the type ``ty2``.
5734 When sign extending from i1, the extension always results in -1 or 0.
5739 .. code-block:: llvm
5741 %X = sext i8 -1 to i16 ; yields i16 :65535
5742 %Y = sext i1 true to i32 ; yields i32:-1
5743 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5745 '``fptrunc .. to``' Instruction
5746 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5753 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5758 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5763 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5764 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5765 The size of ``value`` must be larger than the size of ``ty2``. This
5766 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5771 The '``fptrunc``' instruction truncates a ``value`` from a larger
5772 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5773 point <t_floating>` type. If the value cannot fit within the
5774 destination type, ``ty2``, then the results are undefined.
5779 .. code-block:: llvm
5781 %X = fptrunc double 123.0 to float ; yields float:123.0
5782 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5784 '``fpext .. to``' Instruction
5785 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5792 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5797 The '``fpext``' extends a floating point ``value`` to a larger floating
5803 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5804 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5805 to. The source type must be smaller than the destination type.
5810 The '``fpext``' instruction extends the ``value`` from a smaller
5811 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5812 point <t_floating>` type. The ``fpext`` cannot be used to make a
5813 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5814 *no-op cast* for a floating point cast.
5819 .. code-block:: llvm
5821 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5822 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5824 '``fptoui .. to``' Instruction
5825 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5832 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5837 The '``fptoui``' converts a floating point ``value`` to its unsigned
5838 integer equivalent of type ``ty2``.
5843 The '``fptoui``' instruction takes a value to cast, which must be a
5844 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5845 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5846 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5847 type with the same number of elements as ``ty``
5852 The '``fptoui``' instruction converts its :ref:`floating
5853 point <t_floating>` operand into the nearest (rounding towards zero)
5854 unsigned integer value. If the value cannot fit in ``ty2``, the results
5860 .. code-block:: llvm
5862 %X = fptoui double 123.0 to i32 ; yields i32:123
5863 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5864 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5866 '``fptosi .. to``' Instruction
5867 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5874 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5879 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5880 ``value`` to type ``ty2``.
5885 The '``fptosi``' instruction takes a value to cast, which must be a
5886 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5887 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5888 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5889 type with the same number of elements as ``ty``
5894 The '``fptosi``' instruction converts its :ref:`floating
5895 point <t_floating>` operand into the nearest (rounding towards zero)
5896 signed integer value. If the value cannot fit in ``ty2``, the results
5902 .. code-block:: llvm
5904 %X = fptosi double -123.0 to i32 ; yields i32:-123
5905 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5906 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5908 '``uitofp .. to``' Instruction
5909 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5916 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5921 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5922 and converts that value to the ``ty2`` type.
5927 The '``uitofp``' instruction takes a value to cast, which must be a
5928 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5929 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5930 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5931 type with the same number of elements as ``ty``
5936 The '``uitofp``' instruction interprets its operand as an unsigned
5937 integer quantity and converts it to the corresponding floating point
5938 value. If the value cannot fit in the floating point value, the results
5944 .. code-block:: llvm
5946 %X = uitofp i32 257 to float ; yields float:257.0
5947 %Y = uitofp i8 -1 to double ; yields double:255.0
5949 '``sitofp .. to``' Instruction
5950 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5957 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5962 The '``sitofp``' instruction regards ``value`` as a signed integer and
5963 converts that value to the ``ty2`` type.
5968 The '``sitofp``' instruction takes a value to cast, which must be a
5969 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5970 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5971 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5972 type with the same number of elements as ``ty``
5977 The '``sitofp``' instruction interprets its operand as a signed integer
5978 quantity and converts it to the corresponding floating point value. If
5979 the value cannot fit in the floating point value, the results are
5985 .. code-block:: llvm
5987 %X = sitofp i32 257 to float ; yields float:257.0
5988 %Y = sitofp i8 -1 to double ; yields double:-1.0
5992 '``ptrtoint .. to``' Instruction
5993 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6000 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
6005 The '``ptrtoint``' instruction converts the pointer or a vector of
6006 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
6011 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
6012 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
6013 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
6014 a vector of integers type.
6019 The '``ptrtoint``' instruction converts ``value`` to integer type
6020 ``ty2`` by interpreting the pointer value as an integer and either
6021 truncating or zero extending that value to the size of the integer type.
6022 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
6023 ``value`` is larger than ``ty2`` then a truncation is done. If they are
6024 the same size, then nothing is done (*no-op cast*) other than a type
6030 .. code-block:: llvm
6032 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
6033 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
6034 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
6038 '``inttoptr .. to``' Instruction
6039 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6046 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
6051 The '``inttoptr``' instruction converts an integer ``value`` to a
6052 pointer type, ``ty2``.
6057 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
6058 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
6064 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
6065 applying either a zero extension or a truncation depending on the size
6066 of the integer ``value``. If ``value`` is larger than the size of a
6067 pointer then a truncation is done. If ``value`` is smaller than the size
6068 of a pointer then a zero extension is done. If they are the same size,
6069 nothing is done (*no-op cast*).
6074 .. code-block:: llvm
6076 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
6077 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
6078 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
6079 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
6083 '``bitcast .. to``' Instruction
6084 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6091 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
6096 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
6102 The '``bitcast``' instruction takes a value to cast, which must be a
6103 non-aggregate first class value, and a type to cast it to, which must
6104 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
6105 bit sizes of ``value`` and the destination type, ``ty2``, must be
6106 identical. If the source type is a pointer, the destination type must
6107 also be a pointer of the same size. This instruction supports bitwise
6108 conversion of vectors to integers and to vectors of other types (as
6109 long as they have the same size).
6114 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
6115 is always a *no-op cast* because no bits change with this
6116 conversion. The conversion is done as if the ``value`` had been stored
6117 to memory and read back as type ``ty2``. Pointer (or vector of
6118 pointers) types may only be converted to other pointer (or vector of
6119 pointers) types with the same address space through this instruction.
6120 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
6121 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
6126 .. code-block:: llvm
6128 %X = bitcast i8 255 to i8 ; yields i8 :-1
6129 %Y = bitcast i32* %x to sint* ; yields sint*:%x
6130 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
6131 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
6133 .. _i_addrspacecast:
6135 '``addrspacecast .. to``' Instruction
6136 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6143 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
6148 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
6149 address space ``n`` to type ``pty2`` in address space ``m``.
6154 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
6155 to cast and a pointer type to cast it to, which must have a different
6161 The '``addrspacecast``' instruction converts the pointer value
6162 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
6163 value modification, depending on the target and the address space
6164 pair. Pointer conversions within the same address space must be
6165 performed with the ``bitcast`` instruction. Note that if the address space
6166 conversion is legal then both result and operand refer to the same memory
6172 .. code-block:: llvm
6174 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
6175 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
6176 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
6183 The instructions in this category are the "miscellaneous" instructions,
6184 which defy better classification.
6188 '``icmp``' Instruction
6189 ^^^^^^^^^^^^^^^^^^^^^^
6196 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6201 The '``icmp``' instruction returns a boolean value or a vector of
6202 boolean values based on comparison of its two integer, integer vector,
6203 pointer, or pointer vector operands.
6208 The '``icmp``' instruction takes three operands. The first operand is
6209 the condition code indicating the kind of comparison to perform. It is
6210 not a value, just a keyword. The possible condition code are:
6213 #. ``ne``: not equal
6214 #. ``ugt``: unsigned greater than
6215 #. ``uge``: unsigned greater or equal
6216 #. ``ult``: unsigned less than
6217 #. ``ule``: unsigned less or equal
6218 #. ``sgt``: signed greater than
6219 #. ``sge``: signed greater or equal
6220 #. ``slt``: signed less than
6221 #. ``sle``: signed less or equal
6223 The remaining two arguments must be :ref:`integer <t_integer>` or
6224 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
6225 must also be identical types.
6230 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
6231 code given as ``cond``. The comparison performed always yields either an
6232 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
6234 #. ``eq``: yields ``true`` if the operands are equal, ``false``
6235 otherwise. No sign interpretation is necessary or performed.
6236 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
6237 otherwise. No sign interpretation is necessary or performed.
6238 #. ``ugt``: interprets the operands as unsigned values and yields
6239 ``true`` if ``op1`` is greater than ``op2``.
6240 #. ``uge``: interprets the operands as unsigned values and yields
6241 ``true`` if ``op1`` is greater than or equal to ``op2``.
6242 #. ``ult``: interprets the operands as unsigned values and yields
6243 ``true`` if ``op1`` is less than ``op2``.
6244 #. ``ule``: interprets the operands as unsigned values and yields
6245 ``true`` if ``op1`` is less than or equal to ``op2``.
6246 #. ``sgt``: interprets the operands as signed values and yields ``true``
6247 if ``op1`` is greater than ``op2``.
6248 #. ``sge``: interprets the operands as signed values and yields ``true``
6249 if ``op1`` is greater than or equal to ``op2``.
6250 #. ``slt``: interprets the operands as signed values and yields ``true``
6251 if ``op1`` is less than ``op2``.
6252 #. ``sle``: interprets the operands as signed values and yields ``true``
6253 if ``op1`` is less than or equal to ``op2``.
6255 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
6256 are compared as if they were integers.
6258 If the operands are integer vectors, then they are compared element by
6259 element. The result is an ``i1`` vector with the same number of elements
6260 as the values being compared. Otherwise, the result is an ``i1``.
6265 .. code-block:: llvm
6267 <result> = icmp eq i32 4, 5 ; yields: result=false
6268 <result> = icmp ne float* %X, %X ; yields: result=false
6269 <result> = icmp ult i16 4, 5 ; yields: result=true
6270 <result> = icmp sgt i16 4, 5 ; yields: result=false
6271 <result> = icmp ule i16 -4, 5 ; yields: result=false
6272 <result> = icmp sge i16 4, 5 ; yields: result=false
6274 Note that the code generator does not yet support vector types with the
6275 ``icmp`` instruction.
6279 '``fcmp``' Instruction
6280 ^^^^^^^^^^^^^^^^^^^^^^
6287 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6292 The '``fcmp``' instruction returns a boolean value or vector of boolean
6293 values based on comparison of its operands.
6295 If the operands are floating point scalars, then the result type is a
6296 boolean (:ref:`i1 <t_integer>`).
6298 If the operands are floating point vectors, then the result type is a
6299 vector of boolean with the same number of elements as the operands being
6305 The '``fcmp``' instruction takes three operands. The first operand is
6306 the condition code indicating the kind of comparison to perform. It is
6307 not a value, just a keyword. The possible condition code are:
6309 #. ``false``: no comparison, always returns false
6310 #. ``oeq``: ordered and equal
6311 #. ``ogt``: ordered and greater than
6312 #. ``oge``: ordered and greater than or equal
6313 #. ``olt``: ordered and less than
6314 #. ``ole``: ordered and less than or equal
6315 #. ``one``: ordered and not equal
6316 #. ``ord``: ordered (no nans)
6317 #. ``ueq``: unordered or equal
6318 #. ``ugt``: unordered or greater than
6319 #. ``uge``: unordered or greater than or equal
6320 #. ``ult``: unordered or less than
6321 #. ``ule``: unordered or less than or equal
6322 #. ``une``: unordered or not equal
6323 #. ``uno``: unordered (either nans)
6324 #. ``true``: no comparison, always returns true
6326 *Ordered* means that neither operand is a QNAN while *unordered* means
6327 that either operand may be a QNAN.
6329 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6330 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6331 type. They must have identical types.
6336 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6337 condition code given as ``cond``. If the operands are vectors, then the
6338 vectors are compared element by element. Each comparison performed
6339 always yields an :ref:`i1 <t_integer>` result, as follows:
6341 #. ``false``: always yields ``false``, regardless of operands.
6342 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6343 is equal to ``op2``.
6344 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6345 is greater than ``op2``.
6346 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6347 is greater than or equal to ``op2``.
6348 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6349 is less than ``op2``.
6350 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6351 is less than or equal to ``op2``.
6352 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6353 is not equal to ``op2``.
6354 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6355 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6357 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6358 greater than ``op2``.
6359 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6360 greater than or equal to ``op2``.
6361 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6363 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6364 less than or equal to ``op2``.
6365 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6366 not equal to ``op2``.
6367 #. ``uno``: yields ``true`` if either operand is a QNAN.
6368 #. ``true``: always yields ``true``, regardless of operands.
6373 .. code-block:: llvm
6375 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6376 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6377 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6378 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6380 Note that the code generator does not yet support vector types with the
6381 ``fcmp`` instruction.
6385 '``phi``' Instruction
6386 ^^^^^^^^^^^^^^^^^^^^^
6393 <result> = phi <ty> [ <val0>, <label0>], ...
6398 The '``phi``' instruction is used to implement the φ node in the SSA
6399 graph representing the function.
6404 The type of the incoming values is specified with the first type field.
6405 After this, the '``phi``' instruction takes a list of pairs as
6406 arguments, with one pair for each predecessor basic block of the current
6407 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6408 the value arguments to the PHI node. Only labels may be used as the
6411 There must be no non-phi instructions between the start of a basic block
6412 and the PHI instructions: i.e. PHI instructions must be first in a basic
6415 For the purposes of the SSA form, the use of each incoming value is
6416 deemed to occur on the edge from the corresponding predecessor block to
6417 the current block (but after any definition of an '``invoke``'
6418 instruction's return value on the same edge).
6423 At runtime, the '``phi``' instruction logically takes on the value
6424 specified by the pair corresponding to the predecessor basic block that
6425 executed just prior to the current block.
6430 .. code-block:: llvm
6432 Loop: ; Infinite loop that counts from 0 on up...
6433 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6434 %nextindvar = add i32 %indvar, 1
6439 '``select``' Instruction
6440 ^^^^^^^^^^^^^^^^^^^^^^^^
6447 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6449 selty is either i1 or {<N x i1>}
6454 The '``select``' instruction is used to choose one value based on a
6455 condition, without IR-level branching.
6460 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6461 values indicating the condition, and two values of the same :ref:`first
6462 class <t_firstclass>` type. If the val1/val2 are vectors and the
6463 condition is a scalar, then entire vectors are selected, not individual
6469 If the condition is an i1 and it evaluates to 1, the instruction returns
6470 the first value argument; otherwise, it returns the second value
6473 If the condition is a vector of i1, then the value arguments must be
6474 vectors of the same size, and the selection is done element by element.
6479 .. code-block:: llvm
6481 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6485 '``call``' Instruction
6486 ^^^^^^^^^^^^^^^^^^^^^^
6493 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6498 The '``call``' instruction represents a simple function call.
6503 This instruction requires several arguments:
6505 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
6506 should perform tail call optimization. The ``tail`` marker is a hint that
6507 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
6508 means that the call must be tail call optimized in order for the program to
6509 be correct. The ``musttail`` marker provides these guarantees:
6511 #. The call will not cause unbounded stack growth if it is part of a
6512 recursive cycle in the call graph.
6513 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
6516 Both markers imply that the callee does not access allocas or varargs from
6517 the caller. Calls marked ``musttail`` must obey the following additional
6520 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
6521 or a pointer bitcast followed by a ret instruction.
6522 - The ret instruction must return the (possibly bitcasted) value
6523 produced by the call or void.
6524 - The caller and callee prototypes must match. Pointer types of
6525 parameters or return types may differ in pointee type, but not
6527 - The calling conventions of the caller and callee must match.
6528 - All ABI-impacting function attributes, such as sret, byval, inreg,
6529 returned, and inalloca, must match.
6531 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
6532 the following conditions are met:
6534 - Caller and callee both have the calling convention ``fastcc``.
6535 - The call is in tail position (ret immediately follows call and ret
6536 uses value of call or is void).
6537 - Option ``-tailcallopt`` is enabled, or
6538 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6539 - `Platform-specific constraints are
6540 met. <CodeGenerator.html#tailcallopt>`_
6542 #. The optional "cconv" marker indicates which :ref:`calling
6543 convention <callingconv>` the call should use. If none is
6544 specified, the call defaults to using C calling conventions. The
6545 calling convention of the call must match the calling convention of
6546 the target function, or else the behavior is undefined.
6547 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6548 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6550 #. '``ty``': the type of the call instruction itself which is also the
6551 type of the return value. Functions that return no value are marked
6553 #. '``fnty``': shall be the signature of the pointer to function value
6554 being invoked. The argument types must match the types implied by
6555 this signature. This type can be omitted if the function is not
6556 varargs and if the function type does not return a pointer to a
6558 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6559 be invoked. In most cases, this is a direct function invocation, but
6560 indirect ``call``'s are just as possible, calling an arbitrary pointer
6562 #. '``function args``': argument list whose types match the function
6563 signature argument types and parameter attributes. All arguments must
6564 be of :ref:`first class <t_firstclass>` type. If the function signature
6565 indicates the function accepts a variable number of arguments, the
6566 extra arguments can be specified.
6567 #. The optional :ref:`function attributes <fnattrs>` list. Only
6568 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6569 attributes are valid here.
6574 The '``call``' instruction is used to cause control flow to transfer to
6575 a specified function, with its incoming arguments bound to the specified
6576 values. Upon a '``ret``' instruction in the called function, control
6577 flow continues with the instruction after the function call, and the
6578 return value of the function is bound to the result argument.
6583 .. code-block:: llvm
6585 %retval = call i32 @test(i32 %argc)
6586 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6587 %X = tail call i32 @foo() ; yields i32
6588 %Y = tail call fastcc i32 @foo() ; yields i32
6589 call void %foo(i8 97 signext)
6591 %struct.A = type { i32, i8 }
6592 %r = call %struct.A @foo() ; yields { i32, i8 }
6593 %gr = extractvalue %struct.A %r, 0 ; yields i32
6594 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6595 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6596 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6598 llvm treats calls to some functions with names and arguments that match
6599 the standard C99 library as being the C99 library functions, and may
6600 perform optimizations or generate code for them under that assumption.
6601 This is something we'd like to change in the future to provide better
6602 support for freestanding environments and non-C-based languages.
6606 '``va_arg``' Instruction
6607 ^^^^^^^^^^^^^^^^^^^^^^^^
6614 <resultval> = va_arg <va_list*> <arglist>, <argty>
6619 The '``va_arg``' instruction is used to access arguments passed through
6620 the "variable argument" area of a function call. It is used to implement
6621 the ``va_arg`` macro in C.
6626 This instruction takes a ``va_list*`` value and the type of the
6627 argument. It returns a value of the specified argument type and
6628 increments the ``va_list`` to point to the next argument. The actual
6629 type of ``va_list`` is target specific.
6634 The '``va_arg``' instruction loads an argument of the specified type
6635 from the specified ``va_list`` and causes the ``va_list`` to point to
6636 the next argument. For more information, see the variable argument
6637 handling :ref:`Intrinsic Functions <int_varargs>`.
6639 It is legal for this instruction to be called in a function which does
6640 not take a variable number of arguments, for example, the ``vfprintf``
6643 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6644 function <intrinsics>` because it takes a type as an argument.
6649 See the :ref:`variable argument processing <int_varargs>` section.
6651 Note that the code generator does not yet fully support va\_arg on many
6652 targets. Also, it does not currently support va\_arg with aggregate
6653 types on any target.
6657 '``landingpad``' Instruction
6658 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6665 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6666 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6668 <clause> := catch <type> <value>
6669 <clause> := filter <array constant type> <array constant>
6674 The '``landingpad``' instruction is used by `LLVM's exception handling
6675 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6676 is a landing pad --- one where the exception lands, and corresponds to the
6677 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6678 defines values supplied by the personality function (``pers_fn``) upon
6679 re-entry to the function. The ``resultval`` has the type ``resultty``.
6684 This instruction takes a ``pers_fn`` value. This is the personality
6685 function associated with the unwinding mechanism. The optional
6686 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6688 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6689 contains the global variable representing the "type" that may be caught
6690 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6691 clause takes an array constant as its argument. Use
6692 "``[0 x i8**] undef``" for a filter which cannot throw. The
6693 '``landingpad``' instruction must contain *at least* one ``clause`` or
6694 the ``cleanup`` flag.
6699 The '``landingpad``' instruction defines the values which are set by the
6700 personality function (``pers_fn``) upon re-entry to the function, and
6701 therefore the "result type" of the ``landingpad`` instruction. As with
6702 calling conventions, how the personality function results are
6703 represented in LLVM IR is target specific.
6705 The clauses are applied in order from top to bottom. If two
6706 ``landingpad`` instructions are merged together through inlining, the
6707 clauses from the calling function are appended to the list of clauses.
6708 When the call stack is being unwound due to an exception being thrown,
6709 the exception is compared against each ``clause`` in turn. If it doesn't
6710 match any of the clauses, and the ``cleanup`` flag is not set, then
6711 unwinding continues further up the call stack.
6713 The ``landingpad`` instruction has several restrictions:
6715 - A landing pad block is a basic block which is the unwind destination
6716 of an '``invoke``' instruction.
6717 - A landing pad block must have a '``landingpad``' instruction as its
6718 first non-PHI instruction.
6719 - There can be only one '``landingpad``' instruction within the landing
6721 - A basic block that is not a landing pad block may not include a
6722 '``landingpad``' instruction.
6723 - All '``landingpad``' instructions in a function must have the same
6724 personality function.
6729 .. code-block:: llvm
6731 ;; A landing pad which can catch an integer.
6732 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6734 ;; A landing pad that is a cleanup.
6735 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6737 ;; A landing pad which can catch an integer and can only throw a double.
6738 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6740 filter [1 x i8**] [@_ZTId]
6747 LLVM supports the notion of an "intrinsic function". These functions
6748 have well known names and semantics and are required to follow certain
6749 restrictions. Overall, these intrinsics represent an extension mechanism
6750 for the LLVM language that does not require changing all of the
6751 transformations in LLVM when adding to the language (or the bitcode
6752 reader/writer, the parser, etc...).
6754 Intrinsic function names must all start with an "``llvm.``" prefix. This
6755 prefix is reserved in LLVM for intrinsic names; thus, function names may
6756 not begin with this prefix. Intrinsic functions must always be external
6757 functions: you cannot define the body of intrinsic functions. Intrinsic
6758 functions may only be used in call or invoke instructions: it is illegal
6759 to take the address of an intrinsic function. Additionally, because
6760 intrinsic functions are part of the LLVM language, it is required if any
6761 are added that they be documented here.
6763 Some intrinsic functions can be overloaded, i.e., the intrinsic
6764 represents a family of functions that perform the same operation but on
6765 different data types. Because LLVM can represent over 8 million
6766 different integer types, overloading is used commonly to allow an
6767 intrinsic function to operate on any integer type. One or more of the
6768 argument types or the result type can be overloaded to accept any
6769 integer type. Argument types may also be defined as exactly matching a
6770 previous argument's type or the result type. This allows an intrinsic
6771 function which accepts multiple arguments, but needs all of them to be
6772 of the same type, to only be overloaded with respect to a single
6773 argument or the result.
6775 Overloaded intrinsics will have the names of its overloaded argument
6776 types encoded into its function name, each preceded by a period. Only
6777 those types which are overloaded result in a name suffix. Arguments
6778 whose type is matched against another type do not. For example, the
6779 ``llvm.ctpop`` function can take an integer of any width and returns an
6780 integer of exactly the same integer width. This leads to a family of
6781 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6782 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6783 overloaded, and only one type suffix is required. Because the argument's
6784 type is matched against the return type, it does not require its own
6787 To learn how to add an intrinsic function, please see the `Extending
6788 LLVM Guide <ExtendingLLVM.html>`_.
6792 Variable Argument Handling Intrinsics
6793 -------------------------------------
6795 Variable argument support is defined in LLVM with the
6796 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6797 functions. These functions are related to the similarly named macros
6798 defined in the ``<stdarg.h>`` header file.
6800 All of these functions operate on arguments that use a target-specific
6801 value type "``va_list``". The LLVM assembly language reference manual
6802 does not define what this type is, so all transformations should be
6803 prepared to handle these functions regardless of the type used.
6805 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6806 variable argument handling intrinsic functions are used.
6808 .. code-block:: llvm
6810 define i32 @test(i32 %X, ...) {
6811 ; Initialize variable argument processing
6813 %ap2 = bitcast i8** %ap to i8*
6814 call void @llvm.va_start(i8* %ap2)
6816 ; Read a single integer argument
6817 %tmp = va_arg i8** %ap, i32
6819 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6821 %aq2 = bitcast i8** %aq to i8*
6822 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6823 call void @llvm.va_end(i8* %aq2)
6825 ; Stop processing of arguments.
6826 call void @llvm.va_end(i8* %ap2)
6830 declare void @llvm.va_start(i8*)
6831 declare void @llvm.va_copy(i8*, i8*)
6832 declare void @llvm.va_end(i8*)
6836 '``llvm.va_start``' Intrinsic
6837 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6844 declare void @llvm.va_start(i8* <arglist>)
6849 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6850 subsequent use by ``va_arg``.
6855 The argument is a pointer to a ``va_list`` element to initialize.
6860 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6861 available in C. In a target-dependent way, it initializes the
6862 ``va_list`` element to which the argument points, so that the next call
6863 to ``va_arg`` will produce the first variable argument passed to the
6864 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6865 to know the last argument of the function as the compiler can figure
6868 '``llvm.va_end``' Intrinsic
6869 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6876 declare void @llvm.va_end(i8* <arglist>)
6881 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6882 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6887 The argument is a pointer to a ``va_list`` to destroy.
6892 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6893 available in C. In a target-dependent way, it destroys the ``va_list``
6894 element to which the argument points. Calls to
6895 :ref:`llvm.va_start <int_va_start>` and
6896 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6901 '``llvm.va_copy``' Intrinsic
6902 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6909 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6914 The '``llvm.va_copy``' intrinsic copies the current argument position
6915 from the source argument list to the destination argument list.
6920 The first argument is a pointer to a ``va_list`` element to initialize.
6921 The second argument is a pointer to a ``va_list`` element to copy from.
6926 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6927 available in C. In a target-dependent way, it copies the source
6928 ``va_list`` element into the destination ``va_list`` element. This
6929 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6930 arbitrarily complex and require, for example, memory allocation.
6932 Accurate Garbage Collection Intrinsics
6933 --------------------------------------
6935 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6936 (GC) requires the implementation and generation of these intrinsics.
6937 These intrinsics allow identification of :ref:`GC roots on the
6938 stack <int_gcroot>`, as well as garbage collector implementations that
6939 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6940 Front-ends for type-safe garbage collected languages should generate
6941 these intrinsics to make use of the LLVM garbage collectors. For more
6942 details, see `Accurate Garbage Collection with
6943 LLVM <GarbageCollection.html>`_.
6945 The garbage collection intrinsics only operate on objects in the generic
6946 address space (address space zero).
6950 '``llvm.gcroot``' Intrinsic
6951 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6958 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6963 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6964 the code generator, and allows some metadata to be associated with it.
6969 The first argument specifies the address of a stack object that contains
6970 the root pointer. The second pointer (which must be either a constant or
6971 a global value address) contains the meta-data to be associated with the
6977 At runtime, a call to this intrinsic stores a null pointer into the
6978 "ptrloc" location. At compile-time, the code generator generates
6979 information to allow the runtime to find the pointer at GC safe points.
6980 The '``llvm.gcroot``' intrinsic may only be used in a function which
6981 :ref:`specifies a GC algorithm <gc>`.
6985 '``llvm.gcread``' Intrinsic
6986 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6993 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6998 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6999 locations, allowing garbage collector implementations that require read
7005 The second argument is the address to read from, which should be an
7006 address allocated from the garbage collector. The first object is a
7007 pointer to the start of the referenced object, if needed by the language
7008 runtime (otherwise null).
7013 The '``llvm.gcread``' intrinsic has the same semantics as a load
7014 instruction, but may be replaced with substantially more complex code by
7015 the garbage collector runtime, as needed. The '``llvm.gcread``'
7016 intrinsic may only be used in a function which :ref:`specifies a GC
7021 '``llvm.gcwrite``' Intrinsic
7022 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7029 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
7034 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
7035 locations, allowing garbage collector implementations that require write
7036 barriers (such as generational or reference counting collectors).
7041 The first argument is the reference to store, the second is the start of
7042 the object to store it to, and the third is the address of the field of
7043 Obj to store to. If the runtime does not require a pointer to the
7044 object, Obj may be null.
7049 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
7050 instruction, but may be replaced with substantially more complex code by
7051 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
7052 intrinsic may only be used in a function which :ref:`specifies a GC
7055 Code Generator Intrinsics
7056 -------------------------
7058 These intrinsics are provided by LLVM to expose special features that
7059 may only be implemented with code generator support.
7061 '``llvm.returnaddress``' Intrinsic
7062 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7069 declare i8 *@llvm.returnaddress(i32 <level>)
7074 The '``llvm.returnaddress``' intrinsic attempts to compute a
7075 target-specific value indicating the return address of the current
7076 function or one of its callers.
7081 The argument to this intrinsic indicates which function to return the
7082 address for. Zero indicates the calling function, one indicates its
7083 caller, etc. The argument is **required** to be a constant integer
7089 The '``llvm.returnaddress``' intrinsic either returns a pointer
7090 indicating the return address of the specified call frame, or zero if it
7091 cannot be identified. The value returned by this intrinsic is likely to
7092 be incorrect or 0 for arguments other than zero, so it should only be
7093 used for debugging purposes.
7095 Note that calling this intrinsic does not prevent function inlining or
7096 other aggressive transformations, so the value returned may not be that
7097 of the obvious source-language caller.
7099 '``llvm.frameaddress``' Intrinsic
7100 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7107 declare i8* @llvm.frameaddress(i32 <level>)
7112 The '``llvm.frameaddress``' intrinsic attempts to return the
7113 target-specific frame pointer value for the specified stack frame.
7118 The argument to this intrinsic indicates which function to return the
7119 frame pointer for. Zero indicates the calling function, one indicates
7120 its caller, etc. The argument is **required** to be a constant integer
7126 The '``llvm.frameaddress``' intrinsic either returns a pointer
7127 indicating the frame address of the specified call frame, or zero if it
7128 cannot be identified. The value returned by this intrinsic is likely to
7129 be incorrect or 0 for arguments other than zero, so it should only be
7130 used for debugging purposes.
7132 Note that calling this intrinsic does not prevent function inlining or
7133 other aggressive transformations, so the value returned may not be that
7134 of the obvious source-language caller.
7136 .. _int_read_register:
7137 .. _int_write_register:
7139 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
7140 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7147 declare i32 @llvm.read_register.i32(metadata)
7148 declare i64 @llvm.read_register.i64(metadata)
7149 declare void @llvm.write_register.i32(metadata, i32 @value)
7150 declare void @llvm.write_register.i64(metadata, i64 @value)
7151 !0 = metadata !{metadata !"sp\00"}
7156 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
7157 provides access to the named register. The register must be valid on
7158 the architecture being compiled to. The type needs to be compatible
7159 with the register being read.
7164 The '``llvm.read_register``' intrinsic returns the current value of the
7165 register, where possible. The '``llvm.write_register``' intrinsic sets
7166 the current value of the register, where possible.
7168 This is useful to implement named register global variables that need
7169 to always be mapped to a specific register, as is common practice on
7170 bare-metal programs including OS kernels.
7172 The compiler doesn't check for register availability or use of the used
7173 register in surrounding code, including inline assembly. Because of that,
7174 allocatable registers are not supported.
7176 Warning: So far it only works with the stack pointer on selected
7177 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
7178 work is needed to support other registers and even more so, allocatable
7183 '``llvm.stacksave``' Intrinsic
7184 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7191 declare i8* @llvm.stacksave()
7196 The '``llvm.stacksave``' intrinsic is used to remember the current state
7197 of the function stack, for use with
7198 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
7199 implementing language features like scoped automatic variable sized
7205 This intrinsic returns a opaque pointer value that can be passed to
7206 :ref:`llvm.stackrestore <int_stackrestore>`. When an
7207 ``llvm.stackrestore`` intrinsic is executed with a value saved from
7208 ``llvm.stacksave``, it effectively restores the state of the stack to
7209 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
7210 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
7211 were allocated after the ``llvm.stacksave`` was executed.
7213 .. _int_stackrestore:
7215 '``llvm.stackrestore``' Intrinsic
7216 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7223 declare void @llvm.stackrestore(i8* %ptr)
7228 The '``llvm.stackrestore``' intrinsic is used to restore the state of
7229 the function stack to the state it was in when the corresponding
7230 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
7231 useful for implementing language features like scoped automatic variable
7232 sized arrays in C99.
7237 See the description for :ref:`llvm.stacksave <int_stacksave>`.
7239 '``llvm.prefetch``' Intrinsic
7240 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7247 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7252 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
7253 insert a prefetch instruction if supported; otherwise, it is a noop.
7254 Prefetches have no effect on the behavior of the program but can change
7255 its performance characteristics.
7260 ``address`` is the address to be prefetched, ``rw`` is the specifier
7261 determining if the fetch should be for a read (0) or write (1), and
7262 ``locality`` is a temporal locality specifier ranging from (0) - no
7263 locality, to (3) - extremely local keep in cache. The ``cache type``
7264 specifies whether the prefetch is performed on the data (1) or
7265 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
7266 arguments must be constant integers.
7271 This intrinsic does not modify the behavior of the program. In
7272 particular, prefetches cannot trap and do not produce a value. On
7273 targets that support this intrinsic, the prefetch can provide hints to
7274 the processor cache for better performance.
7276 '``llvm.pcmarker``' Intrinsic
7277 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7284 declare void @llvm.pcmarker(i32 <id>)
7289 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7290 Counter (PC) in a region of code to simulators and other tools. The
7291 method is target specific, but it is expected that the marker will use
7292 exported symbols to transmit the PC of the marker. The marker makes no
7293 guarantees that it will remain with any specific instruction after
7294 optimizations. It is possible that the presence of a marker will inhibit
7295 optimizations. The intended use is to be inserted after optimizations to
7296 allow correlations of simulation runs.
7301 ``id`` is a numerical id identifying the marker.
7306 This intrinsic does not modify the behavior of the program. Backends
7307 that do not support this intrinsic may ignore it.
7309 '``llvm.readcyclecounter``' Intrinsic
7310 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7317 declare i64 @llvm.readcyclecounter()
7322 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
7323 counter register (or similar low latency, high accuracy clocks) on those
7324 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7325 should map to RPCC. As the backing counters overflow quickly (on the
7326 order of 9 seconds on alpha), this should only be used for small
7332 When directly supported, reading the cycle counter should not modify any
7333 memory. Implementations are allowed to either return a application
7334 specific value or a system wide value. On backends without support, this
7335 is lowered to a constant 0.
7337 Note that runtime support may be conditional on the privilege-level code is
7338 running at and the host platform.
7340 '``llvm.clear_cache``' Intrinsic
7341 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7348 declare void @llvm.clear_cache(i8*, i8*)
7353 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
7354 in the specified range to the execution unit of the processor. On
7355 targets with non-unified instruction and data cache, the implementation
7356 flushes the instruction cache.
7361 On platforms with coherent instruction and data caches (e.g. x86), this
7362 intrinsic is a nop. On platforms with non-coherent instruction and data
7363 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
7364 instructions or a system call, if cache flushing requires special
7367 The default behavior is to emit a call to ``__clear_cache`` from the run
7370 This instrinsic does *not* empty the instruction pipeline. Modifications
7371 of the current function are outside the scope of the intrinsic.
7373 Standard C Library Intrinsics
7374 -----------------------------
7376 LLVM provides intrinsics for a few important standard C library
7377 functions. These intrinsics allow source-language front-ends to pass
7378 information about the alignment of the pointer arguments to the code
7379 generator, providing opportunity for more efficient code generation.
7383 '``llvm.memcpy``' Intrinsic
7384 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7389 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
7390 integer bit width and for different address spaces. Not all targets
7391 support all bit widths however.
7395 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7396 i32 <len>, i32 <align>, i1 <isvolatile>)
7397 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7398 i64 <len>, i32 <align>, i1 <isvolatile>)
7403 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7404 source location to the destination location.
7406 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
7407 intrinsics do not return a value, takes extra alignment/isvolatile
7408 arguments and the pointers can be in specified address spaces.
7413 The first argument is a pointer to the destination, the second is a
7414 pointer to the source. The third argument is an integer argument
7415 specifying the number of bytes to copy, the fourth argument is the
7416 alignment of the source and destination locations, and the fifth is a
7417 boolean indicating a volatile access.
7419 If the call to this intrinsic has an alignment value that is not 0 or 1,
7420 then the caller guarantees that both the source and destination pointers
7421 are aligned to that boundary.
7423 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
7424 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7425 very cleanly specified and it is unwise to depend on it.
7430 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7431 source location to the destination location, which are not allowed to
7432 overlap. It copies "len" bytes of memory over. If the argument is known
7433 to be aligned to some boundary, this can be specified as the fourth
7434 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
7436 '``llvm.memmove``' Intrinsic
7437 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7442 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7443 bit width and for different address space. Not all targets support all
7448 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7449 i32 <len>, i32 <align>, i1 <isvolatile>)
7450 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7451 i64 <len>, i32 <align>, i1 <isvolatile>)
7456 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7457 source location to the destination location. It is similar to the
7458 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7461 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7462 intrinsics do not return a value, takes extra alignment/isvolatile
7463 arguments and the pointers can be in specified address spaces.
7468 The first argument is a pointer to the destination, the second is a
7469 pointer to the source. The third argument is an integer argument
7470 specifying the number of bytes to copy, the fourth argument is the
7471 alignment of the source and destination locations, and the fifth is a
7472 boolean indicating a volatile access.
7474 If the call to this intrinsic has an alignment value that is not 0 or 1,
7475 then the caller guarantees that the source and destination pointers are
7476 aligned to that boundary.
7478 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7479 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7480 not very cleanly specified and it is unwise to depend on it.
7485 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7486 source location to the destination location, which may overlap. It
7487 copies "len" bytes of memory over. If the argument is known to be
7488 aligned to some boundary, this can be specified as the fourth argument,
7489 otherwise it should be set to 0 or 1 (both meaning no alignment).
7491 '``llvm.memset.*``' Intrinsics
7492 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7497 This is an overloaded intrinsic. You can use llvm.memset on any integer
7498 bit width and for different address spaces. However, not all targets
7499 support all bit widths.
7503 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7504 i32 <len>, i32 <align>, i1 <isvolatile>)
7505 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7506 i64 <len>, i32 <align>, i1 <isvolatile>)
7511 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7512 particular byte value.
7514 Note that, unlike the standard libc function, the ``llvm.memset``
7515 intrinsic does not return a value and takes extra alignment/volatile
7516 arguments. Also, the destination can be in an arbitrary address space.
7521 The first argument is a pointer to the destination to fill, the second
7522 is the byte value with which to fill it, the third argument is an
7523 integer argument specifying the number of bytes to fill, and the fourth
7524 argument is the known alignment of the destination location.
7526 If the call to this intrinsic has an alignment value that is not 0 or 1,
7527 then the caller guarantees that the destination pointer is aligned to
7530 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7531 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7532 very cleanly specified and it is unwise to depend on it.
7537 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7538 at the destination location. If the argument is known to be aligned to
7539 some boundary, this can be specified as the fourth argument, otherwise
7540 it should be set to 0 or 1 (both meaning no alignment).
7542 '``llvm.sqrt.*``' Intrinsic
7543 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7548 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7549 floating point or vector of floating point type. Not all targets support
7554 declare float @llvm.sqrt.f32(float %Val)
7555 declare double @llvm.sqrt.f64(double %Val)
7556 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7557 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7558 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7563 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7564 returning the same value as the libm '``sqrt``' functions would. Unlike
7565 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7566 negative numbers other than -0.0 (which allows for better optimization,
7567 because there is no need to worry about errno being set).
7568 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7573 The argument and return value are floating point numbers of the same
7579 This function returns the sqrt of the specified operand if it is a
7580 nonnegative floating point number.
7582 '``llvm.powi.*``' Intrinsic
7583 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7588 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7589 floating point or vector of floating point type. Not all targets support
7594 declare float @llvm.powi.f32(float %Val, i32 %power)
7595 declare double @llvm.powi.f64(double %Val, i32 %power)
7596 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7597 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7598 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7603 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7604 specified (positive or negative) power. The order of evaluation of
7605 multiplications is not defined. When a vector of floating point type is
7606 used, the second argument remains a scalar integer value.
7611 The second argument is an integer power, and the first is a value to
7612 raise to that power.
7617 This function returns the first value raised to the second power with an
7618 unspecified sequence of rounding operations.
7620 '``llvm.sin.*``' Intrinsic
7621 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7626 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7627 floating point or vector of floating point type. Not all targets support
7632 declare float @llvm.sin.f32(float %Val)
7633 declare double @llvm.sin.f64(double %Val)
7634 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7635 declare fp128 @llvm.sin.f128(fp128 %Val)
7636 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7641 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7646 The argument and return value are floating point numbers of the same
7652 This function returns the sine of the specified operand, returning the
7653 same values as the libm ``sin`` functions would, and handles error
7654 conditions in the same way.
7656 '``llvm.cos.*``' Intrinsic
7657 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7662 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7663 floating point or vector of floating point type. Not all targets support
7668 declare float @llvm.cos.f32(float %Val)
7669 declare double @llvm.cos.f64(double %Val)
7670 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7671 declare fp128 @llvm.cos.f128(fp128 %Val)
7672 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7677 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7682 The argument and return value are floating point numbers of the same
7688 This function returns the cosine of the specified operand, returning the
7689 same values as the libm ``cos`` functions would, and handles error
7690 conditions in the same way.
7692 '``llvm.pow.*``' Intrinsic
7693 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7698 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7699 floating point or vector of floating point type. Not all targets support
7704 declare float @llvm.pow.f32(float %Val, float %Power)
7705 declare double @llvm.pow.f64(double %Val, double %Power)
7706 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7707 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7708 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7713 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7714 specified (positive or negative) power.
7719 The second argument is a floating point power, and the first is a value
7720 to raise to that power.
7725 This function returns the first value raised to the second power,
7726 returning the same values as the libm ``pow`` functions would, and
7727 handles error conditions in the same way.
7729 '``llvm.exp.*``' Intrinsic
7730 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7735 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7736 floating point or vector of floating point type. Not all targets support
7741 declare float @llvm.exp.f32(float %Val)
7742 declare double @llvm.exp.f64(double %Val)
7743 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7744 declare fp128 @llvm.exp.f128(fp128 %Val)
7745 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7750 The '``llvm.exp.*``' intrinsics perform the exp function.
7755 The argument and return value are floating point numbers of the same
7761 This function returns the same values as the libm ``exp`` functions
7762 would, and handles error conditions in the same way.
7764 '``llvm.exp2.*``' Intrinsic
7765 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7770 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7771 floating point or vector of floating point type. Not all targets support
7776 declare float @llvm.exp2.f32(float %Val)
7777 declare double @llvm.exp2.f64(double %Val)
7778 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7779 declare fp128 @llvm.exp2.f128(fp128 %Val)
7780 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7785 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7790 The argument and return value are floating point numbers of the same
7796 This function returns the same values as the libm ``exp2`` functions
7797 would, and handles error conditions in the same way.
7799 '``llvm.log.*``' Intrinsic
7800 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7805 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7806 floating point or vector of floating point type. Not all targets support
7811 declare float @llvm.log.f32(float %Val)
7812 declare double @llvm.log.f64(double %Val)
7813 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7814 declare fp128 @llvm.log.f128(fp128 %Val)
7815 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7820 The '``llvm.log.*``' intrinsics perform the log function.
7825 The argument and return value are floating point numbers of the same
7831 This function returns the same values as the libm ``log`` functions
7832 would, and handles error conditions in the same way.
7834 '``llvm.log10.*``' Intrinsic
7835 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7840 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7841 floating point or vector of floating point type. Not all targets support
7846 declare float @llvm.log10.f32(float %Val)
7847 declare double @llvm.log10.f64(double %Val)
7848 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7849 declare fp128 @llvm.log10.f128(fp128 %Val)
7850 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7855 The '``llvm.log10.*``' intrinsics perform the log10 function.
7860 The argument and return value are floating point numbers of the same
7866 This function returns the same values as the libm ``log10`` functions
7867 would, and handles error conditions in the same way.
7869 '``llvm.log2.*``' Intrinsic
7870 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7875 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7876 floating point or vector of floating point type. Not all targets support
7881 declare float @llvm.log2.f32(float %Val)
7882 declare double @llvm.log2.f64(double %Val)
7883 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7884 declare fp128 @llvm.log2.f128(fp128 %Val)
7885 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7890 The '``llvm.log2.*``' intrinsics perform the log2 function.
7895 The argument and return value are floating point numbers of the same
7901 This function returns the same values as the libm ``log2`` functions
7902 would, and handles error conditions in the same way.
7904 '``llvm.fma.*``' Intrinsic
7905 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7910 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7911 floating point or vector of floating point type. Not all targets support
7916 declare float @llvm.fma.f32(float %a, float %b, float %c)
7917 declare double @llvm.fma.f64(double %a, double %b, double %c)
7918 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7919 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7920 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7925 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7931 The argument and return value are floating point numbers of the same
7937 This function returns the same values as the libm ``fma`` functions
7938 would, and does not set errno.
7940 '``llvm.fabs.*``' Intrinsic
7941 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7946 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7947 floating point or vector of floating point type. Not all targets support
7952 declare float @llvm.fabs.f32(float %Val)
7953 declare double @llvm.fabs.f64(double %Val)
7954 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7955 declare fp128 @llvm.fabs.f128(fp128 %Val)
7956 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7961 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7967 The argument and return value are floating point numbers of the same
7973 This function returns the same values as the libm ``fabs`` functions
7974 would, and handles error conditions in the same way.
7976 '``llvm.copysign.*``' Intrinsic
7977 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7982 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7983 floating point or vector of floating point type. Not all targets support
7988 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7989 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7990 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7991 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7992 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7997 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7998 first operand and the sign of the second operand.
8003 The arguments and return value are floating point numbers of the same
8009 This function returns the same values as the libm ``copysign``
8010 functions would, and handles error conditions in the same way.
8012 '``llvm.floor.*``' Intrinsic
8013 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8018 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
8019 floating point or vector of floating point type. Not all targets support
8024 declare float @llvm.floor.f32(float %Val)
8025 declare double @llvm.floor.f64(double %Val)
8026 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
8027 declare fp128 @llvm.floor.f128(fp128 %Val)
8028 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
8033 The '``llvm.floor.*``' intrinsics return the floor of the operand.
8038 The argument and return value are floating point numbers of the same
8044 This function returns the same values as the libm ``floor`` functions
8045 would, and handles error conditions in the same way.
8047 '``llvm.ceil.*``' Intrinsic
8048 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8053 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
8054 floating point or vector of floating point type. Not all targets support
8059 declare float @llvm.ceil.f32(float %Val)
8060 declare double @llvm.ceil.f64(double %Val)
8061 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
8062 declare fp128 @llvm.ceil.f128(fp128 %Val)
8063 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
8068 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
8073 The argument and return value are floating point numbers of the same
8079 This function returns the same values as the libm ``ceil`` functions
8080 would, and handles error conditions in the same way.
8082 '``llvm.trunc.*``' Intrinsic
8083 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8088 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
8089 floating point or vector of floating point type. Not all targets support
8094 declare float @llvm.trunc.f32(float %Val)
8095 declare double @llvm.trunc.f64(double %Val)
8096 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
8097 declare fp128 @llvm.trunc.f128(fp128 %Val)
8098 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
8103 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
8104 nearest integer not larger in magnitude than the operand.
8109 The argument and return value are floating point numbers of the same
8115 This function returns the same values as the libm ``trunc`` functions
8116 would, and handles error conditions in the same way.
8118 '``llvm.rint.*``' Intrinsic
8119 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8124 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
8125 floating point or vector of floating point type. Not all targets support
8130 declare float @llvm.rint.f32(float %Val)
8131 declare double @llvm.rint.f64(double %Val)
8132 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
8133 declare fp128 @llvm.rint.f128(fp128 %Val)
8134 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
8139 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
8140 nearest integer. It may raise an inexact floating-point exception if the
8141 operand isn't an integer.
8146 The argument and return value are floating point numbers of the same
8152 This function returns the same values as the libm ``rint`` functions
8153 would, and handles error conditions in the same way.
8155 '``llvm.nearbyint.*``' Intrinsic
8156 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8161 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
8162 floating point or vector of floating point type. Not all targets support
8167 declare float @llvm.nearbyint.f32(float %Val)
8168 declare double @llvm.nearbyint.f64(double %Val)
8169 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
8170 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
8171 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
8176 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
8182 The argument and return value are floating point numbers of the same
8188 This function returns the same values as the libm ``nearbyint``
8189 functions would, and handles error conditions in the same way.
8191 '``llvm.round.*``' Intrinsic
8192 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8197 This is an overloaded intrinsic. You can use ``llvm.round`` on any
8198 floating point or vector of floating point type. Not all targets support
8203 declare float @llvm.round.f32(float %Val)
8204 declare double @llvm.round.f64(double %Val)
8205 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
8206 declare fp128 @llvm.round.f128(fp128 %Val)
8207 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
8212 The '``llvm.round.*``' intrinsics returns the operand rounded to the
8218 The argument and return value are floating point numbers of the same
8224 This function returns the same values as the libm ``round``
8225 functions would, and handles error conditions in the same way.
8227 Bit Manipulation Intrinsics
8228 ---------------------------
8230 LLVM provides intrinsics for a few important bit manipulation
8231 operations. These allow efficient code generation for some algorithms.
8233 '``llvm.bswap.*``' Intrinsics
8234 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8239 This is an overloaded intrinsic function. You can use bswap on any
8240 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
8244 declare i16 @llvm.bswap.i16(i16 <id>)
8245 declare i32 @llvm.bswap.i32(i32 <id>)
8246 declare i64 @llvm.bswap.i64(i64 <id>)
8251 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
8252 values with an even number of bytes (positive multiple of 16 bits).
8253 These are useful for performing operations on data that is not in the
8254 target's native byte order.
8259 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
8260 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
8261 intrinsic returns an i32 value that has the four bytes of the input i32
8262 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
8263 returned i32 will have its bytes in 3, 2, 1, 0 order. The
8264 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
8265 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
8268 '``llvm.ctpop.*``' Intrinsic
8269 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8274 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
8275 bit width, or on any vector with integer elements. Not all targets
8276 support all bit widths or vector types, however.
8280 declare i8 @llvm.ctpop.i8(i8 <src>)
8281 declare i16 @llvm.ctpop.i16(i16 <src>)
8282 declare i32 @llvm.ctpop.i32(i32 <src>)
8283 declare i64 @llvm.ctpop.i64(i64 <src>)
8284 declare i256 @llvm.ctpop.i256(i256 <src>)
8285 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
8290 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
8296 The only argument is the value to be counted. The argument may be of any
8297 integer type, or a vector with integer elements. The return type must
8298 match the argument type.
8303 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
8304 each element of a vector.
8306 '``llvm.ctlz.*``' Intrinsic
8307 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8312 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
8313 integer bit width, or any vector whose elements are integers. Not all
8314 targets support all bit widths or vector types, however.
8318 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
8319 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
8320 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
8321 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
8322 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
8323 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8328 The '``llvm.ctlz``' family of intrinsic functions counts the number of
8329 leading zeros in a variable.
8334 The first argument is the value to be counted. This argument may be of
8335 any integer type, or a vectory with integer element type. The return
8336 type must match the first argument type.
8338 The second argument must be a constant and is a flag to indicate whether
8339 the intrinsic should ensure that a zero as the first argument produces a
8340 defined result. Historically some architectures did not provide a
8341 defined result for zero values as efficiently, and many algorithms are
8342 now predicated on avoiding zero-value inputs.
8347 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
8348 zeros in a variable, or within each element of the vector. If
8349 ``src == 0`` then the result is the size in bits of the type of ``src``
8350 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8351 ``llvm.ctlz(i32 2) = 30``.
8353 '``llvm.cttz.*``' Intrinsic
8354 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8359 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
8360 integer bit width, or any vector of integer elements. Not all targets
8361 support all bit widths or vector types, however.
8365 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
8366 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
8367 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
8368 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
8369 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
8370 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8375 The '``llvm.cttz``' family of intrinsic functions counts the number of
8381 The first argument is the value to be counted. This argument may be of
8382 any integer type, or a vectory with integer element type. The return
8383 type must match the first argument type.
8385 The second argument must be a constant and is a flag to indicate whether
8386 the intrinsic should ensure that a zero as the first argument produces a
8387 defined result. Historically some architectures did not provide a
8388 defined result for zero values as efficiently, and many algorithms are
8389 now predicated on avoiding zero-value inputs.
8394 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
8395 zeros in a variable, or within each element of a vector. If ``src == 0``
8396 then the result is the size in bits of the type of ``src`` if
8397 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8398 ``llvm.cttz(2) = 1``.
8400 Arithmetic with Overflow Intrinsics
8401 -----------------------------------
8403 LLVM provides intrinsics for some arithmetic with overflow operations.
8405 '``llvm.sadd.with.overflow.*``' Intrinsics
8406 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8411 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
8412 on any integer bit width.
8416 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8417 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8418 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8423 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8424 a signed addition of the two arguments, and indicate whether an overflow
8425 occurred during the signed summation.
8430 The arguments (%a and %b) and the first element of the result structure
8431 may be of integer types of any bit width, but they must have the same
8432 bit width. The second element of the result structure must be of type
8433 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8439 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8440 a signed addition of the two variables. They return a structure --- the
8441 first element of which is the signed summation, and the second element
8442 of which is a bit specifying if the signed summation resulted in an
8448 .. code-block:: llvm
8450 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8451 %sum = extractvalue {i32, i1} %res, 0
8452 %obit = extractvalue {i32, i1} %res, 1
8453 br i1 %obit, label %overflow, label %normal
8455 '``llvm.uadd.with.overflow.*``' Intrinsics
8456 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8461 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8462 on any integer bit width.
8466 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8467 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8468 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8473 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8474 an unsigned addition of the two arguments, and indicate whether a carry
8475 occurred during the unsigned summation.
8480 The arguments (%a and %b) and the first element of the result structure
8481 may be of integer types of any bit width, but they must have the same
8482 bit width. The second element of the result structure must be of type
8483 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8489 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8490 an unsigned addition of the two arguments. They return a structure --- the
8491 first element of which is the sum, and the second element of which is a
8492 bit specifying if the unsigned summation resulted in a carry.
8497 .. code-block:: llvm
8499 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8500 %sum = extractvalue {i32, i1} %res, 0
8501 %obit = extractvalue {i32, i1} %res, 1
8502 br i1 %obit, label %carry, label %normal
8504 '``llvm.ssub.with.overflow.*``' Intrinsics
8505 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8510 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8511 on any integer bit width.
8515 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8516 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8517 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8522 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8523 a signed subtraction of the two arguments, and indicate whether an
8524 overflow occurred during the signed subtraction.
8529 The arguments (%a and %b) and the first element of the result structure
8530 may be of integer types of any bit width, but they must have the same
8531 bit width. The second element of the result structure must be of type
8532 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8538 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8539 a signed subtraction of the two arguments. They return a structure --- the
8540 first element of which is the subtraction, and the second element of
8541 which is a bit specifying if the signed subtraction resulted in an
8547 .. code-block:: llvm
8549 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8550 %sum = extractvalue {i32, i1} %res, 0
8551 %obit = extractvalue {i32, i1} %res, 1
8552 br i1 %obit, label %overflow, label %normal
8554 '``llvm.usub.with.overflow.*``' Intrinsics
8555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8560 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8561 on any integer bit width.
8565 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8566 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8567 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8572 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8573 an unsigned subtraction of the two arguments, and indicate whether an
8574 overflow occurred during the unsigned subtraction.
8579 The arguments (%a and %b) and the first element of the result structure
8580 may be of integer types of any bit width, but they must have the same
8581 bit width. The second element of the result structure must be of type
8582 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8588 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8589 an unsigned subtraction of the two arguments. They return a structure ---
8590 the first element of which is the subtraction, and the second element of
8591 which is a bit specifying if the unsigned subtraction resulted in an
8597 .. code-block:: llvm
8599 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8600 %sum = extractvalue {i32, i1} %res, 0
8601 %obit = extractvalue {i32, i1} %res, 1
8602 br i1 %obit, label %overflow, label %normal
8604 '``llvm.smul.with.overflow.*``' Intrinsics
8605 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8610 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8611 on any integer bit width.
8615 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8616 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8617 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8622 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8623 a signed multiplication of the two arguments, and indicate whether an
8624 overflow occurred during the signed multiplication.
8629 The arguments (%a and %b) and the first element of the result structure
8630 may be of integer types of any bit width, but they must have the same
8631 bit width. The second element of the result structure must be of type
8632 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8638 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8639 a signed multiplication of the two arguments. They return a structure ---
8640 the first element of which is the multiplication, and the second element
8641 of which is a bit specifying if the signed multiplication resulted in an
8647 .. code-block:: llvm
8649 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8650 %sum = extractvalue {i32, i1} %res, 0
8651 %obit = extractvalue {i32, i1} %res, 1
8652 br i1 %obit, label %overflow, label %normal
8654 '``llvm.umul.with.overflow.*``' Intrinsics
8655 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8660 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8661 on any integer bit width.
8665 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8666 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8667 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8672 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8673 a unsigned multiplication of the two arguments, and indicate whether an
8674 overflow occurred during the unsigned multiplication.
8679 The arguments (%a and %b) and the first element of the result structure
8680 may be of integer types of any bit width, but they must have the same
8681 bit width. The second element of the result structure must be of type
8682 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8688 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8689 an unsigned multiplication of the two arguments. They return a structure ---
8690 the first element of which is the multiplication, and the second
8691 element of which is a bit specifying if the unsigned multiplication
8692 resulted in an overflow.
8697 .. code-block:: llvm
8699 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8700 %sum = extractvalue {i32, i1} %res, 0
8701 %obit = extractvalue {i32, i1} %res, 1
8702 br i1 %obit, label %overflow, label %normal
8704 Specialised Arithmetic Intrinsics
8705 ---------------------------------
8707 '``llvm.fmuladd.*``' Intrinsic
8708 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8715 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8716 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8721 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8722 expressions that can be fused if the code generator determines that (a) the
8723 target instruction set has support for a fused operation, and (b) that the
8724 fused operation is more efficient than the equivalent, separate pair of mul
8725 and add instructions.
8730 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8731 multiplicands, a and b, and an addend c.
8740 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8742 is equivalent to the expression a \* b + c, except that rounding will
8743 not be performed between the multiplication and addition steps if the
8744 code generator fuses the operations. Fusion is not guaranteed, even if
8745 the target platform supports it. If a fused multiply-add is required the
8746 corresponding llvm.fma.\* intrinsic function should be used
8747 instead. This never sets errno, just as '``llvm.fma.*``'.
8752 .. code-block:: llvm
8754 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
8756 Half Precision Floating Point Intrinsics
8757 ----------------------------------------
8759 For most target platforms, half precision floating point is a
8760 storage-only format. This means that it is a dense encoding (in memory)
8761 but does not support computation in the format.
8763 This means that code must first load the half-precision floating point
8764 value as an i16, then convert it to float with
8765 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8766 then be performed on the float value (including extending to double
8767 etc). To store the value back to memory, it is first converted to float
8768 if needed, then converted to i16 with
8769 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8772 .. _int_convert_to_fp16:
8774 '``llvm.convert.to.fp16``' Intrinsic
8775 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8782 declare i16 @llvm.convert.to.fp16.f32(float %a)
8783 declare i16 @llvm.convert.to.fp16.f64(double %a)
8788 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
8789 conventional floating point type to half precision floating point format.
8794 The intrinsic function contains single argument - the value to be
8800 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
8801 conventional floating point format to half precision floating point format. The
8802 return value is an ``i16`` which contains the converted number.
8807 .. code-block:: llvm
8809 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
8810 store i16 %res, i16* @x, align 2
8812 .. _int_convert_from_fp16:
8814 '``llvm.convert.from.fp16``' Intrinsic
8815 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8822 declare float @llvm.convert.from.fp16.f32(i16 %a)
8823 declare double @llvm.convert.from.fp16.f64(i16 %a)
8828 The '``llvm.convert.from.fp16``' intrinsic function performs a
8829 conversion from half precision floating point format to single precision
8830 floating point format.
8835 The intrinsic function contains single argument - the value to be
8841 The '``llvm.convert.from.fp16``' intrinsic function performs a
8842 conversion from half single precision floating point format to single
8843 precision floating point format. The input half-float value is
8844 represented by an ``i16`` value.
8849 .. code-block:: llvm
8851 %a = load i16* @x, align 2
8852 %res = call float @llvm.convert.from.fp16(i16 %a)
8857 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8858 prefix), are described in the `LLVM Source Level
8859 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8862 Exception Handling Intrinsics
8863 -----------------------------
8865 The LLVM exception handling intrinsics (which all start with
8866 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8867 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8871 Trampoline Intrinsics
8872 ---------------------
8874 These intrinsics make it possible to excise one parameter, marked with
8875 the :ref:`nest <nest>` attribute, from a function. The result is a
8876 callable function pointer lacking the nest parameter - the caller does
8877 not need to provide a value for it. Instead, the value to use is stored
8878 in advance in a "trampoline", a block of memory usually allocated on the
8879 stack, which also contains code to splice the nest value into the
8880 argument list. This is used to implement the GCC nested function address
8883 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8884 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8885 It can be created as follows:
8887 .. code-block:: llvm
8889 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8890 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8891 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8892 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8893 %fp = bitcast i8* %p to i32 (i32, i32)*
8895 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8896 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8900 '``llvm.init.trampoline``' Intrinsic
8901 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8908 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8913 This fills the memory pointed to by ``tramp`` with executable code,
8914 turning it into a trampoline.
8919 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8920 pointers. The ``tramp`` argument must point to a sufficiently large and
8921 sufficiently aligned block of memory; this memory is written to by the
8922 intrinsic. Note that the size and the alignment are target-specific -
8923 LLVM currently provides no portable way of determining them, so a
8924 front-end that generates this intrinsic needs to have some
8925 target-specific knowledge. The ``func`` argument must hold a function
8926 bitcast to an ``i8*``.
8931 The block of memory pointed to by ``tramp`` is filled with target
8932 dependent code, turning it into a function. Then ``tramp`` needs to be
8933 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8934 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8935 function's signature is the same as that of ``func`` with any arguments
8936 marked with the ``nest`` attribute removed. At most one such ``nest``
8937 argument is allowed, and it must be of pointer type. Calling the new
8938 function is equivalent to calling ``func`` with the same argument list,
8939 but with ``nval`` used for the missing ``nest`` argument. If, after
8940 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8941 modified, then the effect of any later call to the returned function
8942 pointer is undefined.
8946 '``llvm.adjust.trampoline``' Intrinsic
8947 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8954 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8959 This performs any required machine-specific adjustment to the address of
8960 a trampoline (passed as ``tramp``).
8965 ``tramp`` must point to a block of memory which already has trampoline
8966 code filled in by a previous call to
8967 :ref:`llvm.init.trampoline <int_it>`.
8972 On some architectures the address of the code to be executed needs to be
8973 different than the address where the trampoline is actually stored. This
8974 intrinsic returns the executable address corresponding to ``tramp``
8975 after performing the required machine specific adjustments. The pointer
8976 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8981 This class of intrinsics provides information about the lifetime of
8982 memory objects and ranges where variables are immutable.
8986 '``llvm.lifetime.start``' Intrinsic
8987 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8994 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8999 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
9005 The first argument is a constant integer representing the size of the
9006 object, or -1 if it is variable sized. The second argument is a pointer
9012 This intrinsic indicates that before this point in the code, the value
9013 of the memory pointed to by ``ptr`` is dead. This means that it is known
9014 to never be used and has an undefined value. A load from the pointer
9015 that precedes this intrinsic can be replaced with ``'undef'``.
9019 '``llvm.lifetime.end``' Intrinsic
9020 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9027 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
9032 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
9038 The first argument is a constant integer representing the size of the
9039 object, or -1 if it is variable sized. The second argument is a pointer
9045 This intrinsic indicates that after this point in the code, the value of
9046 the memory pointed to by ``ptr`` is dead. This means that it is known to
9047 never be used and has an undefined value. Any stores into the memory
9048 object following this intrinsic may be removed as dead.
9050 '``llvm.invariant.start``' Intrinsic
9051 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9058 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
9063 The '``llvm.invariant.start``' intrinsic specifies that the contents of
9064 a memory object will not change.
9069 The first argument is a constant integer representing the size of the
9070 object, or -1 if it is variable sized. The second argument is a pointer
9076 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
9077 the return value, the referenced memory location is constant and
9080 '``llvm.invariant.end``' Intrinsic
9081 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9088 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
9093 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
9094 memory object are mutable.
9099 The first argument is the matching ``llvm.invariant.start`` intrinsic.
9100 The second argument is a constant integer representing the size of the
9101 object, or -1 if it is variable sized and the third argument is a
9102 pointer to the object.
9107 This intrinsic indicates that the memory is mutable again.
9112 This class of intrinsics is designed to be generic and has no specific
9115 '``llvm.var.annotation``' Intrinsic
9116 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9123 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9128 The '``llvm.var.annotation``' intrinsic.
9133 The first argument is a pointer to a value, the second is a pointer to a
9134 global string, the third is a pointer to a global string which is the
9135 source file name, and the last argument is the line number.
9140 This intrinsic allows annotation of local variables with arbitrary
9141 strings. This can be useful for special purpose optimizations that want
9142 to look for these annotations. These have no other defined use; they are
9143 ignored by code generation and optimization.
9145 '``llvm.ptr.annotation.*``' Intrinsic
9146 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9151 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
9152 pointer to an integer of any width. *NOTE* you must specify an address space for
9153 the pointer. The identifier for the default address space is the integer
9158 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9159 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
9160 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
9161 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
9162 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
9167 The '``llvm.ptr.annotation``' intrinsic.
9172 The first argument is a pointer to an integer value of arbitrary bitwidth
9173 (result of some expression), the second is a pointer to a global string, the
9174 third is a pointer to a global string which is the source file name, and the
9175 last argument is the line number. It returns the value of the first argument.
9180 This intrinsic allows annotation of a pointer to an integer with arbitrary
9181 strings. This can be useful for special purpose optimizations that want to look
9182 for these annotations. These have no other defined use; they are ignored by code
9183 generation and optimization.
9185 '``llvm.annotation.*``' Intrinsic
9186 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9191 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
9192 any integer bit width.
9196 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
9197 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
9198 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
9199 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
9200 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
9205 The '``llvm.annotation``' intrinsic.
9210 The first argument is an integer value (result of some expression), the
9211 second is a pointer to a global string, the third is a pointer to a
9212 global string which is the source file name, and the last argument is
9213 the line number. It returns the value of the first argument.
9218 This intrinsic allows annotations to be put on arbitrary expressions
9219 with arbitrary strings. This can be useful for special purpose
9220 optimizations that want to look for these annotations. These have no
9221 other defined use; they are ignored by code generation and optimization.
9223 '``llvm.trap``' Intrinsic
9224 ^^^^^^^^^^^^^^^^^^^^^^^^^
9231 declare void @llvm.trap() noreturn nounwind
9236 The '``llvm.trap``' intrinsic.
9246 This intrinsic is lowered to the target dependent trap instruction. If
9247 the target does not have a trap instruction, this intrinsic will be
9248 lowered to a call of the ``abort()`` function.
9250 '``llvm.debugtrap``' Intrinsic
9251 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9258 declare void @llvm.debugtrap() nounwind
9263 The '``llvm.debugtrap``' intrinsic.
9273 This intrinsic is lowered to code which is intended to cause an
9274 execution trap with the intention of requesting the attention of a
9277 '``llvm.stackprotector``' Intrinsic
9278 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9285 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
9290 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
9291 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
9292 is placed on the stack before local variables.
9297 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
9298 The first argument is the value loaded from the stack guard
9299 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
9300 enough space to hold the value of the guard.
9305 This intrinsic causes the prologue/epilogue inserter to force the position of
9306 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
9307 to ensure that if a local variable on the stack is overwritten, it will destroy
9308 the value of the guard. When the function exits, the guard on the stack is
9309 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
9310 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
9311 calling the ``__stack_chk_fail()`` function.
9313 '``llvm.stackprotectorcheck``' Intrinsic
9314 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9321 declare void @llvm.stackprotectorcheck(i8** <guard>)
9326 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
9327 created stack protector and if they are not equal calls the
9328 ``__stack_chk_fail()`` function.
9333 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
9334 the variable ``@__stack_chk_guard``.
9339 This intrinsic is provided to perform the stack protector check by comparing
9340 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
9341 values do not match call the ``__stack_chk_fail()`` function.
9343 The reason to provide this as an IR level intrinsic instead of implementing it
9344 via other IR operations is that in order to perform this operation at the IR
9345 level without an intrinsic, one would need to create additional basic blocks to
9346 handle the success/failure cases. This makes it difficult to stop the stack
9347 protector check from disrupting sibling tail calls in Codegen. With this
9348 intrinsic, we are able to generate the stack protector basic blocks late in
9349 codegen after the tail call decision has occurred.
9351 '``llvm.objectsize``' Intrinsic
9352 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9359 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
9360 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
9365 The ``llvm.objectsize`` intrinsic is designed to provide information to
9366 the optimizers to determine at compile time whether a) an operation
9367 (like memcpy) will overflow a buffer that corresponds to an object, or
9368 b) that a runtime check for overflow isn't necessary. An object in this
9369 context means an allocation of a specific class, structure, array, or
9375 The ``llvm.objectsize`` intrinsic takes two arguments. The first
9376 argument is a pointer to or into the ``object``. The second argument is
9377 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
9378 or -1 (if false) when the object size is unknown. The second argument
9379 only accepts constants.
9384 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
9385 the size of the object concerned. If the size cannot be determined at
9386 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
9387 on the ``min`` argument).
9389 '``llvm.expect``' Intrinsic
9390 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9395 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
9400 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
9401 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9402 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9407 The ``llvm.expect`` intrinsic provides information about expected (the
9408 most probable) value of ``val``, which can be used by optimizers.
9413 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
9414 a value. The second argument is an expected value, this needs to be a
9415 constant value, variables are not allowed.
9420 This intrinsic is lowered to the ``val``.
9422 '``llvm.donothing``' Intrinsic
9423 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9430 declare void @llvm.donothing() nounwind readnone
9435 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
9436 only intrinsic that can be called with an invoke instruction.
9446 This intrinsic does nothing, and it's removed by optimizers and ignored
9449 Stack Map Intrinsics
9450 --------------------
9452 LLVM provides experimental intrinsics to support runtime patching
9453 mechanisms commonly desired in dynamic language JITs. These intrinsics
9454 are described in :doc:`StackMaps`.