1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][a-zA-Z$._][a-zA-Z$._0-9]*``'. Identifiers that require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves. The ``"\01"`` prefix
83 can be used on global variables to suppress mangling.
84 #. Unnamed values are represented as an unsigned numeric value with
85 their prefix. For example, ``%12``, ``@2``, ``%44``.
86 #. Constants, which are described in the section Constants_ below.
88 LLVM requires that values start with a prefix for two reasons: Compilers
89 don't need to worry about name clashes with reserved words, and the set
90 of reserved words may be expanded in the future without penalty.
91 Additionally, unnamed identifiers allow a compiler to quickly come up
92 with a temporary variable without having to avoid symbol table
95 Reserved words in LLVM are very similar to reserved words in other
96 languages. There are keywords for different opcodes ('``add``',
97 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
98 '``i32``', etc...), and others. These reserved words cannot conflict
99 with variable names, because none of them start with a prefix character
100 (``'%'`` or ``'@'``).
102 Here is an example of LLVM code to multiply the integer variable
109 %result = mul i32 %X, 8
111 After strength reduction:
115 %result = shl i32 %X, 3
121 %0 = add i32 %X, %X ; yields i32:%0
122 %1 = add i32 %0, %0 ; yields i32:%1
123 %result = add i32 %1, %1
125 This last way of multiplying ``%X`` by 8 illustrates several important
126 lexical features of LLVM:
128 #. Comments are delimited with a '``;``' and go until the end of line.
129 #. Unnamed temporaries are created when the result of a computation is
130 not assigned to a named value.
131 #. Unnamed temporaries are numbered sequentially (using a per-function
132 incrementing counter, starting with 0). Note that basic blocks are
133 included in this numbering. For example, if the entry basic block is not
134 given a label name, then it will get number 0.
136 It also shows a convention that we follow in this document. When
137 demonstrating instructions, we will follow an instruction with a comment
138 that defines the type and name of value produced.
146 LLVM programs are composed of ``Module``'s, each of which is a
147 translation unit of the input programs. Each module consists of
148 functions, global variables, and symbol table entries. Modules may be
149 combined together with the LLVM linker, which merges function (and
150 global variable) definitions, resolves forward declarations, and merges
151 symbol table entries. Here is an example of the "hello world" module:
155 ; Declare the string constant as a global constant.
156 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
158 ; External declaration of the puts function
159 declare i32 @puts(i8* nocapture) nounwind
161 ; Definition of main function
162 define i32 @main() { ; i32()*
163 ; Convert [13 x i8]* to i8 *...
164 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
166 ; Call puts function to write out the string to stdout.
167 call i32 @puts(i8* %cast210)
172 !0 = metadata !{i32 42, null, metadata !"string"}
175 This example is made up of a :ref:`global variable <globalvars>` named
176 "``.str``", an external declaration of the "``puts``" function, a
177 :ref:`function definition <functionstructure>` for "``main``" and
178 :ref:`named metadata <namedmetadatastructure>` "``foo``".
180 In general, a module is made up of a list of global values (where both
181 functions and global variables are global values). Global values are
182 represented by a pointer to a memory location (in this case, a pointer
183 to an array of char, and a pointer to a function), and have one of the
184 following :ref:`linkage types <linkage>`.
191 All Global Variables and Functions have one of the following types of
195 Global values with "``private``" linkage are only directly
196 accessible by objects in the current module. In particular, linking
197 code into a module with an private global value may cause the
198 private to be renamed as necessary to avoid collisions. Because the
199 symbol is private to the module, all references can be updated. This
200 doesn't show up in any symbol table in the object file.
202 Similar to private, but the value shows as a local symbol
203 (``STB_LOCAL`` in the case of ELF) in the object file. This
204 corresponds to the notion of the '``static``' keyword in C.
205 ``available_externally``
206 Globals with "``available_externally``" linkage are never emitted
207 into the object file corresponding to the LLVM module. They exist to
208 allow inlining and other optimizations to take place given knowledge
209 of the definition of the global, which is known to be somewhere
210 outside the module. Globals with ``available_externally`` linkage
211 are allowed to be discarded at will, and are otherwise the same as
212 ``linkonce_odr``. This linkage type is only allowed on definitions,
215 Globals with "``linkonce``" linkage are merged with other globals of
216 the same name when linkage occurs. This can be used to implement
217 some forms of inline functions, templates, or other code which must
218 be generated in each translation unit that uses it, but where the
219 body may be overridden with a more definitive definition later.
220 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
221 that ``linkonce`` linkage does not actually allow the optimizer to
222 inline the body of this function into callers because it doesn't
223 know if this definition of the function is the definitive definition
224 within the program or whether it will be overridden by a stronger
225 definition. To enable inlining and other optimizations, use
226 "``linkonce_odr``" linkage.
228 "``weak``" linkage has the same merging semantics as ``linkonce``
229 linkage, except that unreferenced globals with ``weak`` linkage may
230 not be discarded. This is used for globals that are declared "weak"
233 "``common``" linkage is most similar to "``weak``" linkage, but they
234 are used for tentative definitions in C, such as "``int X;``" at
235 global scope. Symbols with "``common``" linkage are merged in the
236 same way as ``weak symbols``, and they may not be deleted if
237 unreferenced. ``common`` symbols may not have an explicit section,
238 must have a zero initializer, and may not be marked
239 ':ref:`constant <globalvars>`'. Functions and aliases may not have
242 .. _linkage_appending:
245 "``appending``" linkage may only be applied to global variables of
246 pointer to array type. When two global variables with appending
247 linkage are linked together, the two global arrays are appended
248 together. This is the LLVM, typesafe, equivalent of having the
249 system linker append together "sections" with identical names when
252 The semantics of this linkage follow the ELF object file model: the
253 symbol is weak until linked, if not linked, the symbol becomes null
254 instead of being an undefined reference.
255 ``linkonce_odr``, ``weak_odr``
256 Some languages allow differing globals to be merged, such as two
257 functions with different semantics. Other languages, such as
258 ``C++``, ensure that only equivalent globals are ever merged (the
259 "one definition rule" --- "ODR"). Such languages can use the
260 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
261 global will only be merged with equivalent globals. These linkage
262 types are otherwise the same as their non-``odr`` versions.
264 If none of the above identifiers are used, the global is externally
265 visible, meaning that it participates in linkage and can be used to
266 resolve external symbol references.
268 It is illegal for a function *declaration* to have any linkage type
269 other than ``external`` or ``extern_weak``.
276 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
277 :ref:`invokes <i_invoke>` can all have an optional calling convention
278 specified for the call. The calling convention of any pair of dynamic
279 caller/callee must match, or the behavior of the program is undefined.
280 The following calling conventions are supported by LLVM, and more may be
283 "``ccc``" - The C calling convention
284 This calling convention (the default if no other calling convention
285 is specified) matches the target C calling conventions. This calling
286 convention supports varargs function calls and tolerates some
287 mismatch in the declared prototype and implemented declaration of
288 the function (as does normal C).
289 "``fastcc``" - The fast calling convention
290 This calling convention attempts to make calls as fast as possible
291 (e.g. by passing things in registers). This calling convention
292 allows the target to use whatever tricks it wants to produce fast
293 code for the target, without having to conform to an externally
294 specified ABI (Application Binary Interface). `Tail calls can only
295 be optimized when this, the GHC or the HiPE convention is
296 used. <CodeGenerator.html#id80>`_ This calling convention does not
297 support varargs and requires the prototype of all callees to exactly
298 match the prototype of the function definition.
299 "``coldcc``" - The cold calling convention
300 This calling convention attempts to make code in the caller as
301 efficient as possible under the assumption that the call is not
302 commonly executed. As such, these calls often preserve all registers
303 so that the call does not break any live ranges in the caller side.
304 This calling convention does not support varargs and requires the
305 prototype of all callees to exactly match the prototype of the
306 function definition. Furthermore the inliner doesn't consider such function
308 "``cc 10``" - GHC convention
309 This calling convention has been implemented specifically for use by
310 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
311 It passes everything in registers, going to extremes to achieve this
312 by disabling callee save registers. This calling convention should
313 not be used lightly but only for specific situations such as an
314 alternative to the *register pinning* performance technique often
315 used when implementing functional programming languages. At the
316 moment only X86 supports this convention and it has the following
319 - On *X86-32* only supports up to 4 bit type parameters. No
320 floating point types are supported.
321 - On *X86-64* only supports up to 10 bit type parameters and 6
322 floating point parameters.
324 This calling convention supports `tail call
325 optimization <CodeGenerator.html#id80>`_ but requires both the
326 caller and callee are using it.
327 "``cc 11``" - The HiPE calling convention
328 This calling convention has been implemented specifically for use by
329 the `High-Performance Erlang
330 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
331 native code compiler of the `Ericsson's Open Source Erlang/OTP
332 system <http://www.erlang.org/download.shtml>`_. It uses more
333 registers for argument passing than the ordinary C calling
334 convention and defines no callee-saved registers. The calling
335 convention properly supports `tail call
336 optimization <CodeGenerator.html#id80>`_ but requires that both the
337 caller and the callee use it. It uses a *register pinning*
338 mechanism, similar to GHC's convention, for keeping frequently
339 accessed runtime components pinned to specific hardware registers.
340 At the moment only X86 supports this convention (both 32 and 64
342 "``webkit_jscc``" - WebKit's JavaScript calling convention
343 This calling convention has been implemented for `WebKit FTL JIT
344 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
345 stack right to left (as cdecl does), and returns a value in the
346 platform's customary return register.
347 "``anyregcc``" - Dynamic calling convention for code patching
348 This is a special convention that supports patching an arbitrary code
349 sequence in place of a call site. This convention forces the call
350 arguments into registers but allows them to be dynamcially
351 allocated. This can currently only be used with calls to
352 llvm.experimental.patchpoint because only this intrinsic records
353 the location of its arguments in a side table. See :doc:`StackMaps`.
354 "``preserve_mostcc``" - The `PreserveMost` calling convention
355 This calling convention attempts to make the code in the caller as little
356 intrusive as possible. This calling convention behaves identical to the `C`
357 calling convention on how arguments and return values are passed, but it
358 uses a different set of caller/callee-saved registers. This alleviates the
359 burden of saving and recovering a large register set before and after the
360 call in the caller. If the arguments are passed in callee-saved registers,
361 then they will be preserved by the callee across the call. This doesn't
362 apply for values returned in callee-saved registers.
364 - On X86-64 the callee preserves all general purpose registers, except for
365 R11. R11 can be used as a scratch register. Floating-point registers
366 (XMMs/YMMs) are not preserved and need to be saved by the caller.
368 The idea behind this convention is to support calls to runtime functions
369 that have a hot path and a cold path. The hot path is usually a small piece
370 of code that doesn't many registers. The cold path might need to call out to
371 another function and therefore only needs to preserve the caller-saved
372 registers, which haven't already been saved by the caller. The
373 `PreserveMost` calling convention is very similar to the `cold` calling
374 convention in terms of caller/callee-saved registers, but they are used for
375 different types of function calls. `coldcc` is for function calls that are
376 rarely executed, whereas `preserve_mostcc` function calls are intended to be
377 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
378 doesn't prevent the inliner from inlining the function call.
380 This calling convention will be used by a future version of the ObjectiveC
381 runtime and should therefore still be considered experimental at this time.
382 Although this convention was created to optimize certain runtime calls to
383 the ObjectiveC runtime, it is not limited to this runtime and might be used
384 by other runtimes in the future too. The current implementation only
385 supports X86-64, but the intention is to support more architectures in the
387 "``preserve_allcc``" - The `PreserveAll` calling convention
388 This calling convention attempts to make the code in the caller even less
389 intrusive than the `PreserveMost` calling convention. This calling
390 convention also behaves identical to the `C` calling convention on how
391 arguments and return values are passed, but it uses a different set of
392 caller/callee-saved registers. This removes the burden of saving and
393 recovering a large register set before and after the call in the caller. If
394 the arguments are passed in callee-saved registers, then they will be
395 preserved by the callee across the call. This doesn't apply for values
396 returned in callee-saved registers.
398 - On X86-64 the callee preserves all general purpose registers, except for
399 R11. R11 can be used as a scratch register. Furthermore it also preserves
400 all floating-point registers (XMMs/YMMs).
402 The idea behind this convention is to support calls to runtime functions
403 that don't need to call out to any other functions.
405 This calling convention, like the `PreserveMost` calling convention, will be
406 used by a future version of the ObjectiveC runtime and should be considered
407 experimental at this time.
408 "``cc <n>``" - Numbered convention
409 Any calling convention may be specified by number, allowing
410 target-specific calling conventions to be used. Target specific
411 calling conventions start at 64.
413 More calling conventions can be added/defined on an as-needed basis, to
414 support Pascal conventions or any other well-known target-independent
417 .. _visibilitystyles:
422 All Global Variables and Functions have one of the following visibility
425 "``default``" - Default style
426 On targets that use the ELF object file format, default visibility
427 means that the declaration is visible to other modules and, in
428 shared libraries, means that the declared entity may be overridden.
429 On Darwin, default visibility means that the declaration is visible
430 to other modules. Default visibility corresponds to "external
431 linkage" in the language.
432 "``hidden``" - Hidden style
433 Two declarations of an object with hidden visibility refer to the
434 same object if they are in the same shared object. Usually, hidden
435 visibility indicates that the symbol will not be placed into the
436 dynamic symbol table, so no other module (executable or shared
437 library) can reference it directly.
438 "``protected``" - Protected style
439 On ELF, protected visibility indicates that the symbol will be
440 placed in the dynamic symbol table, but that references within the
441 defining module will bind to the local symbol. That is, the symbol
442 cannot be overridden by another module.
444 A symbol with ``internal`` or ``private`` linkage must have ``default``
452 All Global Variables, Functions and Aliases can have one of the following
456 "``dllimport``" causes the compiler to reference a function or variable via
457 a global pointer to a pointer that is set up by the DLL exporting the
458 symbol. On Microsoft Windows targets, the pointer name is formed by
459 combining ``__imp_`` and the function or variable name.
461 "``dllexport``" causes the compiler to provide a global pointer to a pointer
462 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
463 Microsoft Windows targets, the pointer name is formed by combining
464 ``__imp_`` and the function or variable name. Since this storage class
465 exists for defining a dll interface, the compiler, assembler and linker know
466 it is externally referenced and must refrain from deleting the symbol.
470 Thread Local Storage Models
471 ---------------------------
473 A variable may be defined as ``thread_local``, which means that it will
474 not be shared by threads (each thread will have a separated copy of the
475 variable). Not all targets support thread-local variables. Optionally, a
476 TLS model may be specified:
479 For variables that are only used within the current shared library.
481 For variables in modules that will not be loaded dynamically.
483 For variables defined in the executable and only used within it.
485 If no explicit model is given, the "general dynamic" model is used.
487 The models correspond to the ELF TLS models; see `ELF Handling For
488 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
489 more information on under which circumstances the different models may
490 be used. The target may choose a different TLS model if the specified
491 model is not supported, or if a better choice of model can be made.
493 A model can also be specified in a alias, but then it only governs how
494 the alias is accessed. It will not have any effect in the aliasee.
501 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
502 types <t_struct>`. Literal types are uniqued structurally, but identified types
503 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
504 to forward declare a type that is not yet available.
506 An example of a identified structure specification is:
510 %mytype = type { %mytype*, i32 }
512 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
513 literal types are uniqued in recent versions of LLVM.
520 Global variables define regions of memory allocated at compilation time
523 Global variables definitions must be initialized.
525 Global variables in other translation units can also be declared, in which
526 case they don't have an initializer.
528 Either global variable definitions or declarations may have an explicit section
529 to be placed in and may have an optional explicit alignment specified.
531 A variable may be defined as a global ``constant``, which indicates that
532 the contents of the variable will **never** be modified (enabling better
533 optimization, allowing the global data to be placed in the read-only
534 section of an executable, etc). Note that variables that need runtime
535 initialization cannot be marked ``constant`` as there is a store to the
538 LLVM explicitly allows *declarations* of global variables to be marked
539 constant, even if the final definition of the global is not. This
540 capability can be used to enable slightly better optimization of the
541 program, but requires the language definition to guarantee that
542 optimizations based on the 'constantness' are valid for the translation
543 units that do not include the definition.
545 As SSA values, global variables define pointer values that are in scope
546 (i.e. they dominate) all basic blocks in the program. Global variables
547 always define a pointer to their "content" type because they describe a
548 region of memory, and all memory objects in LLVM are accessed through
551 Global variables can be marked with ``unnamed_addr`` which indicates
552 that the address is not significant, only the content. Constants marked
553 like this can be merged with other constants if they have the same
554 initializer. Note that a constant with significant address *can* be
555 merged with a ``unnamed_addr`` constant, the result being a constant
556 whose address is significant.
558 A global variable may be declared to reside in a target-specific
559 numbered address space. For targets that support them, address spaces
560 may affect how optimizations are performed and/or what target
561 instructions are used to access the variable. The default address space
562 is zero. The address space qualifier must precede any other attributes.
564 LLVM allows an explicit section to be specified for globals. If the
565 target supports it, it will emit globals to the section specified.
566 Additionally, the global can placed in a comdat if the target has the necessary
569 By default, global initializers are optimized by assuming that global
570 variables defined within the module are not modified from their
571 initial values before the start of the global initializer. This is
572 true even for variables potentially accessible from outside the
573 module, including those with external linkage or appearing in
574 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
575 by marking the variable with ``externally_initialized``.
577 An explicit alignment may be specified for a global, which must be a
578 power of 2. If not present, or if the alignment is set to zero, the
579 alignment of the global is set by the target to whatever it feels
580 convenient. If an explicit alignment is specified, the global is forced
581 to have exactly that alignment. Targets and optimizers are not allowed
582 to over-align the global if the global has an assigned section. In this
583 case, the extra alignment could be observable: for example, code could
584 assume that the globals are densely packed in their section and try to
585 iterate over them as an array, alignment padding would break this
586 iteration. The maximum alignment is ``1 << 29``.
588 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
590 Variables and aliasaes can have a
591 :ref:`Thread Local Storage Model <tls_model>`.
595 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
596 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
597 <global | constant> <Type> [<InitializerConstant>]
598 [, section "name"] [, align <Alignment>]
600 For example, the following defines a global in a numbered address space
601 with an initializer, section, and alignment:
605 @G = addrspace(5) constant float 1.0, section "foo", align 4
607 The following example just declares a global variable
611 @G = external global i32
613 The following example defines a thread-local global with the
614 ``initialexec`` TLS model:
618 @G = thread_local(initialexec) global i32 0, align 4
620 .. _functionstructure:
625 LLVM function definitions consist of the "``define``" keyword, an
626 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
627 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
628 an optional :ref:`calling convention <callingconv>`,
629 an optional ``unnamed_addr`` attribute, a return type, an optional
630 :ref:`parameter attribute <paramattrs>` for the return type, a function
631 name, a (possibly empty) argument list (each with optional :ref:`parameter
632 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
633 an optional section, an optional alignment,
634 an optional :ref:`comdat <langref_comdats>`,
635 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
636 curly brace, a list of basic blocks, and a closing curly brace.
638 LLVM function declarations consist of the "``declare``" keyword, an
639 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
640 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
641 an optional :ref:`calling convention <callingconv>`,
642 an optional ``unnamed_addr`` attribute, a return type, an optional
643 :ref:`parameter attribute <paramattrs>` for the return type, a function
644 name, a possibly empty list of arguments, an optional alignment, an optional
645 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
647 A function definition contains a list of basic blocks, forming the CFG (Control
648 Flow Graph) for the function. Each basic block may optionally start with a label
649 (giving the basic block a symbol table entry), contains a list of instructions,
650 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
651 function return). If an explicit label is not provided, a block is assigned an
652 implicit numbered label, using the next value from the same counter as used for
653 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
654 entry block does not have an explicit label, it will be assigned label "%0",
655 then the first unnamed temporary in that block will be "%1", etc.
657 The first basic block in a function is special in two ways: it is
658 immediately executed on entrance to the function, and it is not allowed
659 to have predecessor basic blocks (i.e. there can not be any branches to
660 the entry block of a function). Because the block can have no
661 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
663 LLVM allows an explicit section to be specified for functions. If the
664 target supports it, it will emit functions to the section specified.
665 Additionally, the function can placed in a COMDAT.
667 An explicit alignment may be specified for a function. If not present,
668 or if the alignment is set to zero, the alignment of the function is set
669 by the target to whatever it feels convenient. If an explicit alignment
670 is specified, the function is forced to have at least that much
671 alignment. All alignments must be a power of 2.
673 If the ``unnamed_addr`` attribute is given, the address is know to not
674 be significant and two identical functions can be merged.
678 define [linkage] [visibility] [DLLStorageClass]
680 <ResultType> @<FunctionName> ([argument list])
681 [unnamed_addr] [fn Attrs] [section "name"] [comdat $<ComdatName>]
682 [align N] [gc] [prefix Constant] { ... }
689 Aliases, unlike function or variables, don't create any new data. They
690 are just a new symbol and metadata for an existing position.
692 Aliases have a name and an aliasee that is either a global value or a
695 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
696 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
697 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
701 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy> @<Aliasee>
703 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
704 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
705 might not correctly handle dropping a weak symbol that is aliased.
707 Alias that are not ``unnamed_addr`` are guaranteed to have the same address as
708 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
711 Since aliases are only a second name, some restrictions apply, of which
712 some can only be checked when producing an object file:
714 * The expression defining the aliasee must be computable at assembly
715 time. Since it is just a name, no relocations can be used.
717 * No alias in the expression can be weak as the possibility of the
718 intermediate alias being overridden cannot be represented in an
721 * No global value in the expression can be a declaration, since that
722 would require a relocation, which is not possible.
729 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
731 Comdats have a name which represents the COMDAT key. All global objects that
732 specify this key will only end up in the final object file if the linker chooses
733 that key over some other key. Aliases are placed in the same COMDAT that their
734 aliasee computes to, if any.
736 Comdats have a selection kind to provide input on how the linker should
737 choose between keys in two different object files.
741 $<Name> = comdat SelectionKind
743 The selection kind must be one of the following:
746 The linker may choose any COMDAT key, the choice is arbitrary.
748 The linker may choose any COMDAT key but the sections must contain the
751 The linker will choose the section containing the largest COMDAT key.
753 The linker requires that only section with this COMDAT key exist.
755 The linker may choose any COMDAT key but the sections must contain the
758 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
759 ``any`` as a selection kind.
761 Here is an example of a COMDAT group where a function will only be selected if
762 the COMDAT key's section is the largest:
766 $foo = comdat largest
767 @foo = global i32 2, comdat $foo
769 define void @bar() comdat $foo {
773 In a COFF object file, this will create a COMDAT section with selection kind
774 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
775 and another COMDAT section with selection kind
776 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
777 section and contains the contents of the ``@baz`` symbol.
779 There are some restrictions on the properties of the global object.
780 It, or an alias to it, must have the same name as the COMDAT group when
782 The contents and size of this object may be used during link-time to determine
783 which COMDAT groups get selected depending on the selection kind.
784 Because the name of the object must match the name of the COMDAT group, the
785 linkage of the global object must not be local; local symbols can get renamed
786 if a collision occurs in the symbol table.
788 The combined use of COMDATS and section attributes may yield surprising results.
795 @g1 = global i32 42, section "sec", comdat $foo
796 @g2 = global i32 42, section "sec", comdat $bar
798 From the object file perspective, this requires the creation of two sections
799 with the same name. This is necessary because both globals belong to different
800 COMDAT groups and COMDATs, at the object file level, are represented by
803 Note that certain IR constructs like global variables and functions may create
804 COMDATs in the object file in addition to any which are specified using COMDAT
805 IR. This arises, for example, when a global variable has linkonce_odr linkage.
807 .. _namedmetadatastructure:
812 Named metadata is a collection of metadata. :ref:`Metadata
813 nodes <metadata>` (but not metadata strings) are the only valid
814 operands for a named metadata.
818 ; Some unnamed metadata nodes, which are referenced by the named metadata.
819 !0 = metadata !{metadata !"zero"}
820 !1 = metadata !{metadata !"one"}
821 !2 = metadata !{metadata !"two"}
823 !name = !{!0, !1, !2}
830 The return type and each parameter of a function type may have a set of
831 *parameter attributes* associated with them. Parameter attributes are
832 used to communicate additional information about the result or
833 parameters of a function. Parameter attributes are considered to be part
834 of the function, not of the function type, so functions with different
835 parameter attributes can have the same function type.
837 Parameter attributes are simple keywords that follow the type specified.
838 If multiple parameter attributes are needed, they are space separated.
843 declare i32 @printf(i8* noalias nocapture, ...)
844 declare i32 @atoi(i8 zeroext)
845 declare signext i8 @returns_signed_char()
847 Note that any attributes for the function result (``nounwind``,
848 ``readonly``) come immediately after the argument list.
850 Currently, only the following parameter attributes are defined:
853 This indicates to the code generator that the parameter or return
854 value should be zero-extended to the extent required by the target's
855 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
856 the caller (for a parameter) or the callee (for a return value).
858 This indicates to the code generator that the parameter or return
859 value should be sign-extended to the extent required by the target's
860 ABI (which is usually 32-bits) by the caller (for a parameter) or
861 the callee (for a return value).
863 This indicates that this parameter or return value should be treated
864 in a special target-dependent fashion during while emitting code for
865 a function call or return (usually, by putting it in a register as
866 opposed to memory, though some targets use it to distinguish between
867 two different kinds of registers). Use of this attribute is
870 This indicates that the pointer parameter should really be passed by
871 value to the function. The attribute implies that a hidden copy of
872 the pointee is made between the caller and the callee, so the callee
873 is unable to modify the value in the caller. This attribute is only
874 valid on LLVM pointer arguments. It is generally used to pass
875 structs and arrays by value, but is also valid on pointers to
876 scalars. The copy is considered to belong to the caller not the
877 callee (for example, ``readonly`` functions should not write to
878 ``byval`` parameters). This is not a valid attribute for return
881 The byval attribute also supports specifying an alignment with the
882 align attribute. It indicates the alignment of the stack slot to
883 form and the known alignment of the pointer specified to the call
884 site. If the alignment is not specified, then the code generator
885 makes a target-specific assumption.
891 The ``inalloca`` argument attribute allows the caller to take the
892 address of outgoing stack arguments. An ``inalloca`` argument must
893 be a pointer to stack memory produced by an ``alloca`` instruction.
894 The alloca, or argument allocation, must also be tagged with the
895 inalloca keyword. Only the last argument may have the ``inalloca``
896 attribute, and that argument is guaranteed to be passed in memory.
898 An argument allocation may be used by a call at most once because
899 the call may deallocate it. The ``inalloca`` attribute cannot be
900 used in conjunction with other attributes that affect argument
901 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
902 ``inalloca`` attribute also disables LLVM's implicit lowering of
903 large aggregate return values, which means that frontend authors
904 must lower them with ``sret`` pointers.
906 When the call site is reached, the argument allocation must have
907 been the most recent stack allocation that is still live, or the
908 results are undefined. It is possible to allocate additional stack
909 space after an argument allocation and before its call site, but it
910 must be cleared off with :ref:`llvm.stackrestore
913 See :doc:`InAlloca` for more information on how to use this
917 This indicates that the pointer parameter specifies the address of a
918 structure that is the return value of the function in the source
919 program. This pointer must be guaranteed by the caller to be valid:
920 loads and stores to the structure may be assumed by the callee
921 not to trap and to be properly aligned. This may only be applied to
922 the first parameter. This is not a valid attribute for return
926 This indicates that the pointer value may be assumed by the optimizer to
927 have the specified alignment.
929 Note that this attribute has additional semantics when combined with the
935 This indicates that pointer values :ref:`based <pointeraliasing>` on
936 the argument or return value do not alias pointer values that are
937 not *based* on it, ignoring certain "irrelevant" dependencies. For a
938 call to the parent function, dependencies between memory references
939 from before or after the call and from those during the call are
940 "irrelevant" to the ``noalias`` keyword for the arguments and return
941 value used in that call. The caller shares the responsibility with
942 the callee for ensuring that these requirements are met. For further
943 details, please see the discussion of the NoAlias response in :ref:`alias
944 analysis <Must, May, or No>`.
946 Note that this definition of ``noalias`` is intentionally similar
947 to the definition of ``restrict`` in C99 for function arguments,
948 though it is slightly weaker.
950 For function return values, C99's ``restrict`` is not meaningful,
951 while LLVM's ``noalias`` is.
953 This indicates that the callee does not make any copies of the
954 pointer that outlive the callee itself. This is not a valid
955 attribute for return values.
960 This indicates that the pointer parameter can be excised using the
961 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
962 attribute for return values and can only be applied to one parameter.
965 This indicates that the function always returns the argument as its return
966 value. This is an optimization hint to the code generator when generating
967 the caller, allowing tail call optimization and omission of register saves
968 and restores in some cases; it is not checked or enforced when generating
969 the callee. The parameter and the function return type must be valid
970 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
971 valid attribute for return values and can only be applied to one parameter.
974 This indicates that the parameter or return pointer is not null. This
975 attribute may only be applied to pointer typed parameters. This is not
976 checked or enforced by LLVM, the caller must ensure that the pointer
977 passed in is non-null, or the callee must ensure that the returned pointer
980 ``dereferenceable(<n>)``
981 This indicates that the parameter or return pointer is dereferenceable. This
982 attribute may only be applied to pointer typed parameters. A pointer that
983 is dereferenceable can be loaded from speculatively without a risk of
984 trapping. The number of bytes known to be dereferenceable must be provided
985 in parentheses. It is legal for the number of bytes to be less than the
986 size of the pointee type. The ``nonnull`` attribute does not imply
987 dereferenceability (consider a pointer to one element past the end of an
988 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
989 ``addrspace(0)`` (which is the default address space).
993 Garbage Collector Names
994 -----------------------
996 Each function may specify a garbage collector name, which is simply a
1001 define void @f() gc "name" { ... }
1003 The compiler declares the supported values of *name*. Specifying a
1004 collector will cause the compiler to alter its output in order to
1005 support the named garbage collection algorithm.
1012 Prefix data is data associated with a function which the code generator
1013 will emit immediately before the function body. The purpose of this feature
1014 is to allow frontends to associate language-specific runtime metadata with
1015 specific functions and make it available through the function pointer while
1016 still allowing the function pointer to be called. To access the data for a
1017 given function, a program may bitcast the function pointer to a pointer to
1018 the constant's type. This implies that the IR symbol points to the start
1021 To maintain the semantics of ordinary function calls, the prefix data must
1022 have a particular format. Specifically, it must begin with a sequence of
1023 bytes which decode to a sequence of machine instructions, valid for the
1024 module's target, which transfer control to the point immediately succeeding
1025 the prefix data, without performing any other visible action. This allows
1026 the inliner and other passes to reason about the semantics of the function
1027 definition without needing to reason about the prefix data. Obviously this
1028 makes the format of the prefix data highly target dependent.
1030 Prefix data is laid out as if it were an initializer for a global variable
1031 of the prefix data's type. No padding is automatically placed between the
1032 prefix data and the function body. If padding is required, it must be part
1035 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
1036 which encodes the ``nop`` instruction:
1038 .. code-block:: llvm
1040 define void @f() prefix i8 144 { ... }
1042 Generally prefix data can be formed by encoding a relative branch instruction
1043 which skips the metadata, as in this example of valid prefix data for the
1044 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1046 .. code-block:: llvm
1048 %0 = type <{ i8, i8, i8* }>
1050 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
1052 A function may have prefix data but no body. This has similar semantics
1053 to the ``available_externally`` linkage in that the data may be used by the
1054 optimizers but will not be emitted in the object file.
1061 Attribute groups are groups of attributes that are referenced by objects within
1062 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1063 functions will use the same set of attributes. In the degenerative case of a
1064 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1065 group will capture the important command line flags used to build that file.
1067 An attribute group is a module-level object. To use an attribute group, an
1068 object references the attribute group's ID (e.g. ``#37``). An object may refer
1069 to more than one attribute group. In that situation, the attributes from the
1070 different groups are merged.
1072 Here is an example of attribute groups for a function that should always be
1073 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1075 .. code-block:: llvm
1077 ; Target-independent attributes:
1078 attributes #0 = { alwaysinline alignstack=4 }
1080 ; Target-dependent attributes:
1081 attributes #1 = { "no-sse" }
1083 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1084 define void @f() #0 #1 { ... }
1091 Function attributes are set to communicate additional information about
1092 a function. Function attributes are considered to be part of the
1093 function, not of the function type, so functions with different function
1094 attributes can have the same function type.
1096 Function attributes are simple keywords that follow the type specified.
1097 If multiple attributes are needed, they are space separated. For
1100 .. code-block:: llvm
1102 define void @f() noinline { ... }
1103 define void @f() alwaysinline { ... }
1104 define void @f() alwaysinline optsize { ... }
1105 define void @f() optsize { ... }
1108 This attribute indicates that, when emitting the prologue and
1109 epilogue, the backend should forcibly align the stack pointer.
1110 Specify the desired alignment, which must be a power of two, in
1113 This attribute indicates that the inliner should attempt to inline
1114 this function into callers whenever possible, ignoring any active
1115 inlining size threshold for this caller.
1117 This indicates that the callee function at a call site should be
1118 recognized as a built-in function, even though the function's declaration
1119 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1120 direct calls to functions that are declared with the ``nobuiltin``
1123 This attribute indicates that this function is rarely called. When
1124 computing edge weights, basic blocks post-dominated by a cold
1125 function call are also considered to be cold; and, thus, given low
1128 This attribute indicates that the source code contained a hint that
1129 inlining this function is desirable (such as the "inline" keyword in
1130 C/C++). It is just a hint; it imposes no requirements on the
1133 This attribute indicates that the function should be added to a
1134 jump-instruction table at code-generation time, and that all address-taken
1135 references to this function should be replaced with a reference to the
1136 appropriate jump-instruction-table function pointer. Note that this creates
1137 a new pointer for the original function, which means that code that depends
1138 on function-pointer identity can break. So, any function annotated with
1139 ``jumptable`` must also be ``unnamed_addr``.
1141 This attribute suggests that optimization passes and code generator
1142 passes make choices that keep the code size of this function as small
1143 as possible and perform optimizations that may sacrifice runtime
1144 performance in order to minimize the size of the generated code.
1146 This attribute disables prologue / epilogue emission for the
1147 function. This can have very system-specific consequences.
1149 This indicates that the callee function at a call site is not recognized as
1150 a built-in function. LLVM will retain the original call and not replace it
1151 with equivalent code based on the semantics of the built-in function, unless
1152 the call site uses the ``builtin`` attribute. This is valid at call sites
1153 and on function declarations and definitions.
1155 This attribute indicates that calls to the function cannot be
1156 duplicated. A call to a ``noduplicate`` function may be moved
1157 within its parent function, but may not be duplicated within
1158 its parent function.
1160 A function containing a ``noduplicate`` call may still
1161 be an inlining candidate, provided that the call is not
1162 duplicated by inlining. That implies that the function has
1163 internal linkage and only has one call site, so the original
1164 call is dead after inlining.
1166 This attributes disables implicit floating point instructions.
1168 This attribute indicates that the inliner should never inline this
1169 function in any situation. This attribute may not be used together
1170 with the ``alwaysinline`` attribute.
1172 This attribute suppresses lazy symbol binding for the function. This
1173 may make calls to the function faster, at the cost of extra program
1174 startup time if the function is not called during program startup.
1176 This attribute indicates that the code generator should not use a
1177 red zone, even if the target-specific ABI normally permits it.
1179 This function attribute indicates that the function never returns
1180 normally. This produces undefined behavior at runtime if the
1181 function ever does dynamically return.
1183 This function attribute indicates that the function never returns
1184 with an unwind or exceptional control flow. If the function does
1185 unwind, its runtime behavior is undefined.
1187 This function attribute indicates that the function is not optimized
1188 by any optimization or code generator passes with the
1189 exception of interprocedural optimization passes.
1190 This attribute cannot be used together with the ``alwaysinline``
1191 attribute; this attribute is also incompatible
1192 with the ``minsize`` attribute and the ``optsize`` attribute.
1194 This attribute requires the ``noinline`` attribute to be specified on
1195 the function as well, so the function is never inlined into any caller.
1196 Only functions with the ``alwaysinline`` attribute are valid
1197 candidates for inlining into the body of this function.
1199 This attribute suggests that optimization passes and code generator
1200 passes make choices that keep the code size of this function low,
1201 and otherwise do optimizations specifically to reduce code size as
1202 long as they do not significantly impact runtime performance.
1204 On a function, this attribute indicates that the function computes its
1205 result (or decides to unwind an exception) based strictly on its arguments,
1206 without dereferencing any pointer arguments or otherwise accessing
1207 any mutable state (e.g. memory, control registers, etc) visible to
1208 caller functions. It does not write through any pointer arguments
1209 (including ``byval`` arguments) and never changes any state visible
1210 to callers. This means that it cannot unwind exceptions by calling
1211 the ``C++`` exception throwing methods.
1213 On an argument, this attribute indicates that the function does not
1214 dereference that pointer argument, even though it may read or write the
1215 memory that the pointer points to if accessed through other pointers.
1217 On a function, this attribute indicates that the function does not write
1218 through any pointer arguments (including ``byval`` arguments) or otherwise
1219 modify any state (e.g. memory, control registers, etc) visible to
1220 caller functions. It may dereference pointer arguments and read
1221 state that may be set in the caller. A readonly function always
1222 returns the same value (or unwinds an exception identically) when
1223 called with the same set of arguments and global state. It cannot
1224 unwind an exception by calling the ``C++`` exception throwing
1227 On an argument, this attribute indicates that the function does not write
1228 through this pointer argument, even though it may write to the memory that
1229 the pointer points to.
1231 This attribute indicates that this function can return twice. The C
1232 ``setjmp`` is an example of such a function. The compiler disables
1233 some optimizations (like tail calls) in the caller of these
1235 ``sanitize_address``
1236 This attribute indicates that AddressSanitizer checks
1237 (dynamic address safety analysis) are enabled for this function.
1239 This attribute indicates that MemorySanitizer checks (dynamic detection
1240 of accesses to uninitialized memory) are enabled for this function.
1242 This attribute indicates that ThreadSanitizer checks
1243 (dynamic thread safety analysis) are enabled for this function.
1245 This attribute indicates that the function should emit a stack
1246 smashing protector. It is in the form of a "canary" --- a random value
1247 placed on the stack before the local variables that's checked upon
1248 return from the function to see if it has been overwritten. A
1249 heuristic is used to determine if a function needs stack protectors
1250 or not. The heuristic used will enable protectors for functions with:
1252 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1253 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1254 - Calls to alloca() with variable sizes or constant sizes greater than
1255 ``ssp-buffer-size``.
1257 Variables that are identified as requiring a protector will be arranged
1258 on the stack such that they are adjacent to the stack protector guard.
1260 If a function that has an ``ssp`` attribute is inlined into a
1261 function that doesn't have an ``ssp`` attribute, then the resulting
1262 function will have an ``ssp`` attribute.
1264 This attribute indicates that the function should *always* emit a
1265 stack smashing protector. This overrides the ``ssp`` function
1268 Variables that are identified as requiring a protector will be arranged
1269 on the stack such that they are adjacent to the stack protector guard.
1270 The specific layout rules are:
1272 #. Large arrays and structures containing large arrays
1273 (``>= ssp-buffer-size``) are closest to the stack protector.
1274 #. Small arrays and structures containing small arrays
1275 (``< ssp-buffer-size``) are 2nd closest to the protector.
1276 #. Variables that have had their address taken are 3rd closest to the
1279 If a function that has an ``sspreq`` attribute is inlined into a
1280 function that doesn't have an ``sspreq`` attribute or which has an
1281 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1282 an ``sspreq`` attribute.
1284 This attribute indicates that the function should emit a stack smashing
1285 protector. This attribute causes a strong heuristic to be used when
1286 determining if a function needs stack protectors. The strong heuristic
1287 will enable protectors for functions with:
1289 - Arrays of any size and type
1290 - Aggregates containing an array of any size and type.
1291 - Calls to alloca().
1292 - Local variables that have had their address taken.
1294 Variables that are identified as requiring a protector will be arranged
1295 on the stack such that they are adjacent to the stack protector guard.
1296 The specific layout rules are:
1298 #. Large arrays and structures containing large arrays
1299 (``>= ssp-buffer-size``) are closest to the stack protector.
1300 #. Small arrays and structures containing small arrays
1301 (``< ssp-buffer-size``) are 2nd closest to the protector.
1302 #. Variables that have had their address taken are 3rd closest to the
1305 This overrides the ``ssp`` function attribute.
1307 If a function that has an ``sspstrong`` attribute is inlined into a
1308 function that doesn't have an ``sspstrong`` attribute, then the
1309 resulting function will have an ``sspstrong`` attribute.
1311 This attribute indicates that the ABI being targeted requires that
1312 an unwind table entry be produce for this function even if we can
1313 show that no exceptions passes by it. This is normally the case for
1314 the ELF x86-64 abi, but it can be disabled for some compilation
1319 Module-Level Inline Assembly
1320 ----------------------------
1322 Modules may contain "module-level inline asm" blocks, which corresponds
1323 to the GCC "file scope inline asm" blocks. These blocks are internally
1324 concatenated by LLVM and treated as a single unit, but may be separated
1325 in the ``.ll`` file if desired. The syntax is very simple:
1327 .. code-block:: llvm
1329 module asm "inline asm code goes here"
1330 module asm "more can go here"
1332 The strings can contain any character by escaping non-printable
1333 characters. The escape sequence used is simply "\\xx" where "xx" is the
1334 two digit hex code for the number.
1336 The inline asm code is simply printed to the machine code .s file when
1337 assembly code is generated.
1339 .. _langref_datalayout:
1344 A module may specify a target specific data layout string that specifies
1345 how data is to be laid out in memory. The syntax for the data layout is
1348 .. code-block:: llvm
1350 target datalayout = "layout specification"
1352 The *layout specification* consists of a list of specifications
1353 separated by the minus sign character ('-'). Each specification starts
1354 with a letter and may include other information after the letter to
1355 define some aspect of the data layout. The specifications accepted are
1359 Specifies that the target lays out data in big-endian form. That is,
1360 the bits with the most significance have the lowest address
1363 Specifies that the target lays out data in little-endian form. That
1364 is, the bits with the least significance have the lowest address
1367 Specifies the natural alignment of the stack in bits. Alignment
1368 promotion of stack variables is limited to the natural stack
1369 alignment to avoid dynamic stack realignment. The stack alignment
1370 must be a multiple of 8-bits. If omitted, the natural stack
1371 alignment defaults to "unspecified", which does not prevent any
1372 alignment promotions.
1373 ``p[n]:<size>:<abi>:<pref>``
1374 This specifies the *size* of a pointer and its ``<abi>`` and
1375 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1376 bits. The address space, ``n`` is optional, and if not specified,
1377 denotes the default address space 0. The value of ``n`` must be
1378 in the range [1,2^23).
1379 ``i<size>:<abi>:<pref>``
1380 This specifies the alignment for an integer type of a given bit
1381 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1382 ``v<size>:<abi>:<pref>``
1383 This specifies the alignment for a vector type of a given bit
1385 ``f<size>:<abi>:<pref>``
1386 This specifies the alignment for a floating point type of a given bit
1387 ``<size>``. Only values of ``<size>`` that are supported by the target
1388 will work. 32 (float) and 64 (double) are supported on all targets; 80
1389 or 128 (different flavors of long double) are also supported on some
1392 This specifies the alignment for an object of aggregate type.
1394 If present, specifies that llvm names are mangled in the output. The
1397 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1398 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1399 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1400 symbols get a ``_`` prefix.
1401 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1402 functions also get a suffix based on the frame size.
1403 ``n<size1>:<size2>:<size3>...``
1404 This specifies a set of native integer widths for the target CPU in
1405 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1406 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1407 this set are considered to support most general arithmetic operations
1410 On every specification that takes a ``<abi>:<pref>``, specifying the
1411 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1412 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1414 When constructing the data layout for a given target, LLVM starts with a
1415 default set of specifications which are then (possibly) overridden by
1416 the specifications in the ``datalayout`` keyword. The default
1417 specifications are given in this list:
1419 - ``E`` - big endian
1420 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1421 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1422 same as the default address space.
1423 - ``S0`` - natural stack alignment is unspecified
1424 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1425 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1426 - ``i16:16:16`` - i16 is 16-bit aligned
1427 - ``i32:32:32`` - i32 is 32-bit aligned
1428 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1429 alignment of 64-bits
1430 - ``f16:16:16`` - half is 16-bit aligned
1431 - ``f32:32:32`` - float is 32-bit aligned
1432 - ``f64:64:64`` - double is 64-bit aligned
1433 - ``f128:128:128`` - quad is 128-bit aligned
1434 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1435 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1436 - ``a:0:64`` - aggregates are 64-bit aligned
1438 When LLVM is determining the alignment for a given type, it uses the
1441 #. If the type sought is an exact match for one of the specifications,
1442 that specification is used.
1443 #. If no match is found, and the type sought is an integer type, then
1444 the smallest integer type that is larger than the bitwidth of the
1445 sought type is used. If none of the specifications are larger than
1446 the bitwidth then the largest integer type is used. For example,
1447 given the default specifications above, the i7 type will use the
1448 alignment of i8 (next largest) while both i65 and i256 will use the
1449 alignment of i64 (largest specified).
1450 #. If no match is found, and the type sought is a vector type, then the
1451 largest vector type that is smaller than the sought vector type will
1452 be used as a fall back. This happens because <128 x double> can be
1453 implemented in terms of 64 <2 x double>, for example.
1455 The function of the data layout string may not be what you expect.
1456 Notably, this is not a specification from the frontend of what alignment
1457 the code generator should use.
1459 Instead, if specified, the target data layout is required to match what
1460 the ultimate *code generator* expects. This string is used by the
1461 mid-level optimizers to improve code, and this only works if it matches
1462 what the ultimate code generator uses. If you would like to generate IR
1463 that does not embed this target-specific detail into the IR, then you
1464 don't have to specify the string. This will disable some optimizations
1465 that require precise layout information, but this also prevents those
1466 optimizations from introducing target specificity into the IR.
1473 A module may specify a target triple string that describes the target
1474 host. The syntax for the target triple is simply:
1476 .. code-block:: llvm
1478 target triple = "x86_64-apple-macosx10.7.0"
1480 The *target triple* string consists of a series of identifiers delimited
1481 by the minus sign character ('-'). The canonical forms are:
1485 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1486 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1488 This information is passed along to the backend so that it generates
1489 code for the proper architecture. It's possible to override this on the
1490 command line with the ``-mtriple`` command line option.
1492 .. _pointeraliasing:
1494 Pointer Aliasing Rules
1495 ----------------------
1497 Any memory access must be done through a pointer value associated with
1498 an address range of the memory access, otherwise the behavior is
1499 undefined. Pointer values are associated with address ranges according
1500 to the following rules:
1502 - A pointer value is associated with the addresses associated with any
1503 value it is *based* on.
1504 - An address of a global variable is associated with the address range
1505 of the variable's storage.
1506 - The result value of an allocation instruction is associated with the
1507 address range of the allocated storage.
1508 - A null pointer in the default address-space is associated with no
1510 - An integer constant other than zero or a pointer value returned from
1511 a function not defined within LLVM may be associated with address
1512 ranges allocated through mechanisms other than those provided by
1513 LLVM. Such ranges shall not overlap with any ranges of addresses
1514 allocated by mechanisms provided by LLVM.
1516 A pointer value is *based* on another pointer value according to the
1519 - A pointer value formed from a ``getelementptr`` operation is *based*
1520 on the first operand of the ``getelementptr``.
1521 - The result value of a ``bitcast`` is *based* on the operand of the
1523 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1524 values that contribute (directly or indirectly) to the computation of
1525 the pointer's value.
1526 - The "*based* on" relationship is transitive.
1528 Note that this definition of *"based"* is intentionally similar to the
1529 definition of *"based"* in C99, though it is slightly weaker.
1531 LLVM IR does not associate types with memory. The result type of a
1532 ``load`` merely indicates the size and alignment of the memory from
1533 which to load, as well as the interpretation of the value. The first
1534 operand type of a ``store`` similarly only indicates the size and
1535 alignment of the store.
1537 Consequently, type-based alias analysis, aka TBAA, aka
1538 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1539 :ref:`Metadata <metadata>` may be used to encode additional information
1540 which specialized optimization passes may use to implement type-based
1545 Volatile Memory Accesses
1546 ------------------------
1548 Certain memory accesses, such as :ref:`load <i_load>`'s,
1549 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1550 marked ``volatile``. The optimizers must not change the number of
1551 volatile operations or change their order of execution relative to other
1552 volatile operations. The optimizers *may* change the order of volatile
1553 operations relative to non-volatile operations. This is not Java's
1554 "volatile" and has no cross-thread synchronization behavior.
1556 IR-level volatile loads and stores cannot safely be optimized into
1557 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1558 flagged volatile. Likewise, the backend should never split or merge
1559 target-legal volatile load/store instructions.
1561 .. admonition:: Rationale
1563 Platforms may rely on volatile loads and stores of natively supported
1564 data width to be executed as single instruction. For example, in C
1565 this holds for an l-value of volatile primitive type with native
1566 hardware support, but not necessarily for aggregate types. The
1567 frontend upholds these expectations, which are intentionally
1568 unspecified in the IR. The rules above ensure that IR transformation
1569 do not violate the frontend's contract with the language.
1573 Memory Model for Concurrent Operations
1574 --------------------------------------
1576 The LLVM IR does not define any way to start parallel threads of
1577 execution or to register signal handlers. Nonetheless, there are
1578 platform-specific ways to create them, and we define LLVM IR's behavior
1579 in their presence. This model is inspired by the C++0x memory model.
1581 For a more informal introduction to this model, see the :doc:`Atomics`.
1583 We define a *happens-before* partial order as the least partial order
1586 - Is a superset of single-thread program order, and
1587 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1588 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1589 techniques, like pthread locks, thread creation, thread joining,
1590 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1591 Constraints <ordering>`).
1593 Note that program order does not introduce *happens-before* edges
1594 between a thread and signals executing inside that thread.
1596 Every (defined) read operation (load instructions, memcpy, atomic
1597 loads/read-modify-writes, etc.) R reads a series of bytes written by
1598 (defined) write operations (store instructions, atomic
1599 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1600 section, initialized globals are considered to have a write of the
1601 initializer which is atomic and happens before any other read or write
1602 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1603 may see any write to the same byte, except:
1605 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1606 write\ :sub:`2` happens before R\ :sub:`byte`, then
1607 R\ :sub:`byte` does not see write\ :sub:`1`.
1608 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1609 R\ :sub:`byte` does not see write\ :sub:`3`.
1611 Given that definition, R\ :sub:`byte` is defined as follows:
1613 - If R is volatile, the result is target-dependent. (Volatile is
1614 supposed to give guarantees which can support ``sig_atomic_t`` in
1615 C/C++, and may be used for accesses to addresses that do not behave
1616 like normal memory. It does not generally provide cross-thread
1618 - Otherwise, if there is no write to the same byte that happens before
1619 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1620 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1621 R\ :sub:`byte` returns the value written by that write.
1622 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1623 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1624 Memory Ordering Constraints <ordering>` section for additional
1625 constraints on how the choice is made.
1626 - Otherwise R\ :sub:`byte` returns ``undef``.
1628 R returns the value composed of the series of bytes it read. This
1629 implies that some bytes within the value may be ``undef`` **without**
1630 the entire value being ``undef``. Note that this only defines the
1631 semantics of the operation; it doesn't mean that targets will emit more
1632 than one instruction to read the series of bytes.
1634 Note that in cases where none of the atomic intrinsics are used, this
1635 model places only one restriction on IR transformations on top of what
1636 is required for single-threaded execution: introducing a store to a byte
1637 which might not otherwise be stored is not allowed in general.
1638 (Specifically, in the case where another thread might write to and read
1639 from an address, introducing a store can change a load that may see
1640 exactly one write into a load that may see multiple writes.)
1644 Atomic Memory Ordering Constraints
1645 ----------------------------------
1647 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1648 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1649 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1650 ordering parameters that determine which other atomic instructions on
1651 the same address they *synchronize with*. These semantics are borrowed
1652 from Java and C++0x, but are somewhat more colloquial. If these
1653 descriptions aren't precise enough, check those specs (see spec
1654 references in the :doc:`atomics guide <Atomics>`).
1655 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1656 differently since they don't take an address. See that instruction's
1657 documentation for details.
1659 For a simpler introduction to the ordering constraints, see the
1663 The set of values that can be read is governed by the happens-before
1664 partial order. A value cannot be read unless some operation wrote
1665 it. This is intended to provide a guarantee strong enough to model
1666 Java's non-volatile shared variables. This ordering cannot be
1667 specified for read-modify-write operations; it is not strong enough
1668 to make them atomic in any interesting way.
1670 In addition to the guarantees of ``unordered``, there is a single
1671 total order for modifications by ``monotonic`` operations on each
1672 address. All modification orders must be compatible with the
1673 happens-before order. There is no guarantee that the modification
1674 orders can be combined to a global total order for the whole program
1675 (and this often will not be possible). The read in an atomic
1676 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1677 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1678 order immediately before the value it writes. If one atomic read
1679 happens before another atomic read of the same address, the later
1680 read must see the same value or a later value in the address's
1681 modification order. This disallows reordering of ``monotonic`` (or
1682 stronger) operations on the same address. If an address is written
1683 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1684 read that address repeatedly, the other threads must eventually see
1685 the write. This corresponds to the C++0x/C1x
1686 ``memory_order_relaxed``.
1688 In addition to the guarantees of ``monotonic``, a
1689 *synchronizes-with* edge may be formed with a ``release`` operation.
1690 This is intended to model C++'s ``memory_order_acquire``.
1692 In addition to the guarantees of ``monotonic``, if this operation
1693 writes a value which is subsequently read by an ``acquire``
1694 operation, it *synchronizes-with* that operation. (This isn't a
1695 complete description; see the C++0x definition of a release
1696 sequence.) This corresponds to the C++0x/C1x
1697 ``memory_order_release``.
1698 ``acq_rel`` (acquire+release)
1699 Acts as both an ``acquire`` and ``release`` operation on its
1700 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1701 ``seq_cst`` (sequentially consistent)
1702 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1703 operation that only reads, ``release`` for an operation that only
1704 writes), there is a global total order on all
1705 sequentially-consistent operations on all addresses, which is
1706 consistent with the *happens-before* partial order and with the
1707 modification orders of all the affected addresses. Each
1708 sequentially-consistent read sees the last preceding write to the
1709 same address in this global order. This corresponds to the C++0x/C1x
1710 ``memory_order_seq_cst`` and Java volatile.
1714 If an atomic operation is marked ``singlethread``, it only *synchronizes
1715 with* or participates in modification and seq\_cst total orderings with
1716 other operations running in the same thread (for example, in signal
1724 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1725 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1726 :ref:`frem <i_frem>`) have the following flags that can set to enable
1727 otherwise unsafe floating point operations
1730 No NaNs - Allow optimizations to assume the arguments and result are not
1731 NaN. Such optimizations are required to retain defined behavior over
1732 NaNs, but the value of the result is undefined.
1735 No Infs - Allow optimizations to assume the arguments and result are not
1736 +/-Inf. Such optimizations are required to retain defined behavior over
1737 +/-Inf, but the value of the result is undefined.
1740 No Signed Zeros - Allow optimizations to treat the sign of a zero
1741 argument or result as insignificant.
1744 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1745 argument rather than perform division.
1748 Fast - Allow algebraically equivalent transformations that may
1749 dramatically change results in floating point (e.g. reassociate). This
1750 flag implies all the others.
1757 The LLVM type system is one of the most important features of the
1758 intermediate representation. Being typed enables a number of
1759 optimizations to be performed on the intermediate representation
1760 directly, without having to do extra analyses on the side before the
1761 transformation. A strong type system makes it easier to read the
1762 generated code and enables novel analyses and transformations that are
1763 not feasible to perform on normal three address code representations.
1773 The void type does not represent any value and has no size.
1791 The function type can be thought of as a function signature. It consists of a
1792 return type and a list of formal parameter types. The return type of a function
1793 type is a void type or first class type --- except for :ref:`label <t_label>`
1794 and :ref:`metadata <t_metadata>` types.
1800 <returntype> (<parameter list>)
1802 ...where '``<parameter list>``' is a comma-separated list of type
1803 specifiers. Optionally, the parameter list may include a type ``...``, which
1804 indicates that the function takes a variable number of arguments. Variable
1805 argument functions can access their arguments with the :ref:`variable argument
1806 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1807 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1811 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1812 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1813 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1814 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1815 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1816 | ``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. |
1817 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1818 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1819 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1826 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1827 Values of these types are the only ones which can be produced by
1835 These are the types that are valid in registers from CodeGen's perspective.
1844 The integer type is a very simple type that simply specifies an
1845 arbitrary bit width for the integer type desired. Any bit width from 1
1846 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1854 The number of bits the integer will occupy is specified by the ``N``
1860 +----------------+------------------------------------------------+
1861 | ``i1`` | a single-bit integer. |
1862 +----------------+------------------------------------------------+
1863 | ``i32`` | a 32-bit integer. |
1864 +----------------+------------------------------------------------+
1865 | ``i1942652`` | a really big integer of over 1 million bits. |
1866 +----------------+------------------------------------------------+
1870 Floating Point Types
1871 """"""""""""""""""""
1880 - 16-bit floating point value
1883 - 32-bit floating point value
1886 - 64-bit floating point value
1889 - 128-bit floating point value (112-bit mantissa)
1892 - 80-bit floating point value (X87)
1895 - 128-bit floating point value (two 64-bits)
1902 The x86_mmx type represents a value held in an MMX register on an x86
1903 machine. The operations allowed on it are quite limited: parameters and
1904 return values, load and store, and bitcast. User-specified MMX
1905 instructions are represented as intrinsic or asm calls with arguments
1906 and/or results of this type. There are no arrays, vectors or constants
1923 The pointer type is used to specify memory locations. Pointers are
1924 commonly used to reference objects in memory.
1926 Pointer types may have an optional address space attribute defining the
1927 numbered address space where the pointed-to object resides. The default
1928 address space is number zero. The semantics of non-zero address spaces
1929 are target-specific.
1931 Note that LLVM does not permit pointers to void (``void*``) nor does it
1932 permit pointers to labels (``label*``). Use ``i8*`` instead.
1942 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1943 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1944 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1945 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1946 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1947 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1948 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1957 A vector type is a simple derived type that represents a vector of
1958 elements. Vector types are used when multiple primitive data are
1959 operated in parallel using a single instruction (SIMD). A vector type
1960 requires a size (number of elements) and an underlying primitive data
1961 type. Vector types are considered :ref:`first class <t_firstclass>`.
1967 < <# elements> x <elementtype> >
1969 The number of elements is a constant integer value larger than 0;
1970 elementtype may be any integer, floating point or pointer type. Vectors
1971 of size zero are not allowed.
1975 +-------------------+--------------------------------------------------+
1976 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1977 +-------------------+--------------------------------------------------+
1978 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1979 +-------------------+--------------------------------------------------+
1980 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1981 +-------------------+--------------------------------------------------+
1982 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1983 +-------------------+--------------------------------------------------+
1992 The label type represents code labels.
2007 The metadata type represents embedded metadata. No derived types may be
2008 created from metadata except for :ref:`function <t_function>` arguments.
2021 Aggregate Types are a subset of derived types that can contain multiple
2022 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2023 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2033 The array type is a very simple derived type that arranges elements
2034 sequentially in memory. The array type requires a size (number of
2035 elements) and an underlying data type.
2041 [<# elements> x <elementtype>]
2043 The number of elements is a constant integer value; ``elementtype`` may
2044 be any type with a size.
2048 +------------------+--------------------------------------+
2049 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2050 +------------------+--------------------------------------+
2051 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2052 +------------------+--------------------------------------+
2053 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2054 +------------------+--------------------------------------+
2056 Here are some examples of multidimensional arrays:
2058 +-----------------------------+----------------------------------------------------------+
2059 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2060 +-----------------------------+----------------------------------------------------------+
2061 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2062 +-----------------------------+----------------------------------------------------------+
2063 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2064 +-----------------------------+----------------------------------------------------------+
2066 There is no restriction on indexing beyond the end of the array implied
2067 by a static type (though there are restrictions on indexing beyond the
2068 bounds of an allocated object in some cases). This means that
2069 single-dimension 'variable sized array' addressing can be implemented in
2070 LLVM with a zero length array type. An implementation of 'pascal style
2071 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2081 The structure type is used to represent a collection of data members
2082 together in memory. The elements of a structure may be any type that has
2085 Structures in memory are accessed using '``load``' and '``store``' by
2086 getting a pointer to a field with the '``getelementptr``' instruction.
2087 Structures in registers are accessed using the '``extractvalue``' and
2088 '``insertvalue``' instructions.
2090 Structures may optionally be "packed" structures, which indicate that
2091 the alignment of the struct is one byte, and that there is no padding
2092 between the elements. In non-packed structs, padding between field types
2093 is inserted as defined by the DataLayout string in the module, which is
2094 required to match what the underlying code generator expects.
2096 Structures can either be "literal" or "identified". A literal structure
2097 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2098 identified types are always defined at the top level with a name.
2099 Literal types are uniqued by their contents and can never be recursive
2100 or opaque since there is no way to write one. Identified types can be
2101 recursive, can be opaqued, and are never uniqued.
2107 %T1 = type { <type list> } ; Identified normal struct type
2108 %T2 = type <{ <type list> }> ; Identified packed struct type
2112 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2113 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2114 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2115 | ``{ 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``. |
2116 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2117 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2118 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2122 Opaque Structure Types
2123 """"""""""""""""""""""
2127 Opaque structure types are used to represent named structure types that
2128 do not have a body specified. This corresponds (for example) to the C
2129 notion of a forward declared structure.
2140 +--------------+-------------------+
2141 | ``opaque`` | An opaque type. |
2142 +--------------+-------------------+
2149 LLVM has several different basic types of constants. This section
2150 describes them all and their syntax.
2155 **Boolean constants**
2156 The two strings '``true``' and '``false``' are both valid constants
2158 **Integer constants**
2159 Standard integers (such as '4') are constants of the
2160 :ref:`integer <t_integer>` type. Negative numbers may be used with
2162 **Floating point constants**
2163 Floating point constants use standard decimal notation (e.g.
2164 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2165 hexadecimal notation (see below). The assembler requires the exact
2166 decimal value of a floating-point constant. For example, the
2167 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2168 decimal in binary. Floating point constants must have a :ref:`floating
2169 point <t_floating>` type.
2170 **Null pointer constants**
2171 The identifier '``null``' is recognized as a null pointer constant
2172 and must be of :ref:`pointer type <t_pointer>`.
2174 The one non-intuitive notation for constants is the hexadecimal form of
2175 floating point constants. For example, the form
2176 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2177 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2178 constants are required (and the only time that they are generated by the
2179 disassembler) is when a floating point constant must be emitted but it
2180 cannot be represented as a decimal floating point number in a reasonable
2181 number of digits. For example, NaN's, infinities, and other special
2182 values are represented in their IEEE hexadecimal format so that assembly
2183 and disassembly do not cause any bits to change in the constants.
2185 When using the hexadecimal form, constants of types half, float, and
2186 double are represented using the 16-digit form shown above (which
2187 matches the IEEE754 representation for double); half and float values
2188 must, however, be exactly representable as IEEE 754 half and single
2189 precision, respectively. Hexadecimal format is always used for long
2190 double, and there are three forms of long double. The 80-bit format used
2191 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2192 128-bit format used by PowerPC (two adjacent doubles) is represented by
2193 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2194 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2195 will only work if they match the long double format on your target.
2196 The IEEE 16-bit format (half precision) is represented by ``0xH``
2197 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2198 (sign bit at the left).
2200 There are no constants of type x86_mmx.
2202 .. _complexconstants:
2207 Complex constants are a (potentially recursive) combination of simple
2208 constants and smaller complex constants.
2210 **Structure constants**
2211 Structure constants are represented with notation similar to
2212 structure type definitions (a comma separated list of elements,
2213 surrounded by braces (``{}``)). For example:
2214 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2215 "``@G = external global i32``". Structure constants must have
2216 :ref:`structure type <t_struct>`, and the number and types of elements
2217 must match those specified by the type.
2219 Array constants are represented with notation similar to array type
2220 definitions (a comma separated list of elements, surrounded by
2221 square brackets (``[]``)). For example:
2222 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2223 :ref:`array type <t_array>`, and the number and types of elements must
2224 match those specified by the type.
2225 **Vector constants**
2226 Vector constants are represented with notation similar to vector
2227 type definitions (a comma separated list of elements, surrounded by
2228 less-than/greater-than's (``<>``)). For example:
2229 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2230 must have :ref:`vector type <t_vector>`, and the number and types of
2231 elements must match those specified by the type.
2232 **Zero initialization**
2233 The string '``zeroinitializer``' can be used to zero initialize a
2234 value to zero of *any* type, including scalar and
2235 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2236 having to print large zero initializers (e.g. for large arrays) and
2237 is always exactly equivalent to using explicit zero initializers.
2239 A metadata node is a structure-like constant with :ref:`metadata
2240 type <t_metadata>`. For example:
2241 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2242 constants that are meant to be interpreted as part of the
2243 instruction stream, metadata is a place to attach additional
2244 information such as debug info.
2246 Global Variable and Function Addresses
2247 --------------------------------------
2249 The addresses of :ref:`global variables <globalvars>` and
2250 :ref:`functions <functionstructure>` are always implicitly valid
2251 (link-time) constants. These constants are explicitly referenced when
2252 the :ref:`identifier for the global <identifiers>` is used and always have
2253 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2256 .. code-block:: llvm
2260 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2267 The string '``undef``' can be used anywhere a constant is expected, and
2268 indicates that the user of the value may receive an unspecified
2269 bit-pattern. Undefined values may be of any type (other than '``label``'
2270 or '``void``') and be used anywhere a constant is permitted.
2272 Undefined values are useful because they indicate to the compiler that
2273 the program is well defined no matter what value is used. This gives the
2274 compiler more freedom to optimize. Here are some examples of
2275 (potentially surprising) transformations that are valid (in pseudo IR):
2277 .. code-block:: llvm
2287 This is safe because all of the output bits are affected by the undef
2288 bits. Any output bit can have a zero or one depending on the input bits.
2290 .. code-block:: llvm
2301 These logical operations have bits that are not always affected by the
2302 input. For example, if ``%X`` has a zero bit, then the output of the
2303 '``and``' operation will always be a zero for that bit, no matter what
2304 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2305 optimize or assume that the result of the '``and``' is '``undef``'.
2306 However, it is safe to assume that all bits of the '``undef``' could be
2307 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2308 all the bits of the '``undef``' operand to the '``or``' could be set,
2309 allowing the '``or``' to be folded to -1.
2311 .. code-block:: llvm
2313 %A = select undef, %X, %Y
2314 %B = select undef, 42, %Y
2315 %C = select %X, %Y, undef
2325 This set of examples shows that undefined '``select``' (and conditional
2326 branch) conditions can go *either way*, but they have to come from one
2327 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2328 both known to have a clear low bit, then ``%A`` would have to have a
2329 cleared low bit. However, in the ``%C`` example, the optimizer is
2330 allowed to assume that the '``undef``' operand could be the same as
2331 ``%Y``, allowing the whole '``select``' to be eliminated.
2333 .. code-block:: llvm
2335 %A = xor undef, undef
2352 This example points out that two '``undef``' operands are not
2353 necessarily the same. This can be surprising to people (and also matches
2354 C semantics) where they assume that "``X^X``" is always zero, even if
2355 ``X`` is undefined. This isn't true for a number of reasons, but the
2356 short answer is that an '``undef``' "variable" can arbitrarily change
2357 its value over its "live range". This is true because the variable
2358 doesn't actually *have a live range*. Instead, the value is logically
2359 read from arbitrary registers that happen to be around when needed, so
2360 the value is not necessarily consistent over time. In fact, ``%A`` and
2361 ``%C`` need to have the same semantics or the core LLVM "replace all
2362 uses with" concept would not hold.
2364 .. code-block:: llvm
2372 These examples show the crucial difference between an *undefined value*
2373 and *undefined behavior*. An undefined value (like '``undef``') is
2374 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2375 operation can be constant folded to '``undef``', because the '``undef``'
2376 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2377 However, in the second example, we can make a more aggressive
2378 assumption: because the ``undef`` is allowed to be an arbitrary value,
2379 we are allowed to assume that it could be zero. Since a divide by zero
2380 has *undefined behavior*, we are allowed to assume that the operation
2381 does not execute at all. This allows us to delete the divide and all
2382 code after it. Because the undefined operation "can't happen", the
2383 optimizer can assume that it occurs in dead code.
2385 .. code-block:: llvm
2387 a: store undef -> %X
2388 b: store %X -> undef
2393 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2394 value can be assumed to not have any effect; we can assume that the
2395 value is overwritten with bits that happen to match what was already
2396 there. However, a store *to* an undefined location could clobber
2397 arbitrary memory, therefore, it has undefined behavior.
2404 Poison values are similar to :ref:`undef values <undefvalues>`, however
2405 they also represent the fact that an instruction or constant expression
2406 that cannot evoke side effects has nevertheless detected a condition
2407 that results in undefined behavior.
2409 There is currently no way of representing a poison value in the IR; they
2410 only exist when produced by operations such as :ref:`add <i_add>` with
2413 Poison value behavior is defined in terms of value *dependence*:
2415 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2416 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2417 their dynamic predecessor basic block.
2418 - Function arguments depend on the corresponding actual argument values
2419 in the dynamic callers of their functions.
2420 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2421 instructions that dynamically transfer control back to them.
2422 - :ref:`Invoke <i_invoke>` instructions depend on the
2423 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2424 call instructions that dynamically transfer control back to them.
2425 - Non-volatile loads and stores depend on the most recent stores to all
2426 of the referenced memory addresses, following the order in the IR
2427 (including loads and stores implied by intrinsics such as
2428 :ref:`@llvm.memcpy <int_memcpy>`.)
2429 - An instruction with externally visible side effects depends on the
2430 most recent preceding instruction with externally visible side
2431 effects, following the order in the IR. (This includes :ref:`volatile
2432 operations <volatile>`.)
2433 - An instruction *control-depends* on a :ref:`terminator
2434 instruction <terminators>` if the terminator instruction has
2435 multiple successors and the instruction is always executed when
2436 control transfers to one of the successors, and may not be executed
2437 when control is transferred to another.
2438 - Additionally, an instruction also *control-depends* on a terminator
2439 instruction if the set of instructions it otherwise depends on would
2440 be different if the terminator had transferred control to a different
2442 - Dependence is transitive.
2444 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2445 with the additional effect that any instruction that has a *dependence*
2446 on a poison value has undefined behavior.
2448 Here are some examples:
2450 .. code-block:: llvm
2453 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2454 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2455 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2456 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2458 store i32 %poison, i32* @g ; Poison value stored to memory.
2459 %poison2 = load i32* @g ; Poison value loaded back from memory.
2461 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2463 %narrowaddr = bitcast i32* @g to i16*
2464 %wideaddr = bitcast i32* @g to i64*
2465 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2466 %poison4 = load i64* %wideaddr ; Returns a poison value.
2468 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2469 br i1 %cmp, label %true, label %end ; Branch to either destination.
2472 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2473 ; it has undefined behavior.
2477 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2478 ; Both edges into this PHI are
2479 ; control-dependent on %cmp, so this
2480 ; always results in a poison value.
2482 store volatile i32 0, i32* @g ; This would depend on the store in %true
2483 ; if %cmp is true, or the store in %entry
2484 ; otherwise, so this is undefined behavior.
2486 br i1 %cmp, label %second_true, label %second_end
2487 ; The same branch again, but this time the
2488 ; true block doesn't have side effects.
2495 store volatile i32 0, i32* @g ; This time, the instruction always depends
2496 ; on the store in %end. Also, it is
2497 ; control-equivalent to %end, so this is
2498 ; well-defined (ignoring earlier undefined
2499 ; behavior in this example).
2503 Addresses of Basic Blocks
2504 -------------------------
2506 ``blockaddress(@function, %block)``
2508 The '``blockaddress``' constant computes the address of the specified
2509 basic block in the specified function, and always has an ``i8*`` type.
2510 Taking the address of the entry block is illegal.
2512 This value only has defined behavior when used as an operand to the
2513 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2514 against null. Pointer equality tests between labels addresses results in
2515 undefined behavior --- though, again, comparison against null is ok, and
2516 no label is equal to the null pointer. This may be passed around as an
2517 opaque pointer sized value as long as the bits are not inspected. This
2518 allows ``ptrtoint`` and arithmetic to be performed on these values so
2519 long as the original value is reconstituted before the ``indirectbr``
2522 Finally, some targets may provide defined semantics when using the value
2523 as the operand to an inline assembly, but that is target specific.
2527 Constant Expressions
2528 --------------------
2530 Constant expressions are used to allow expressions involving other
2531 constants to be used as constants. Constant expressions may be of any
2532 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2533 that does not have side effects (e.g. load and call are not supported).
2534 The following is the syntax for constant expressions:
2536 ``trunc (CST to TYPE)``
2537 Truncate a constant to another type. The bit size of CST must be
2538 larger than the bit size of TYPE. Both types must be integers.
2539 ``zext (CST to TYPE)``
2540 Zero extend a constant to another type. The bit size of CST must be
2541 smaller than the bit size of TYPE. Both types must be integers.
2542 ``sext (CST to TYPE)``
2543 Sign extend a constant to another type. The bit size of CST must be
2544 smaller than the bit size of TYPE. Both types must be integers.
2545 ``fptrunc (CST to TYPE)``
2546 Truncate a floating point constant to another floating point type.
2547 The size of CST must be larger than the size of TYPE. Both types
2548 must be floating point.
2549 ``fpext (CST to TYPE)``
2550 Floating point extend a constant to another type. The size of CST
2551 must be smaller or equal to the size of TYPE. Both types must be
2553 ``fptoui (CST to TYPE)``
2554 Convert a floating point constant to the corresponding unsigned
2555 integer constant. TYPE must be a scalar or vector integer type. CST
2556 must be of scalar or vector floating point type. Both CST and TYPE
2557 must be scalars, or vectors of the same number of elements. If the
2558 value won't fit in the integer type, the results are undefined.
2559 ``fptosi (CST to TYPE)``
2560 Convert a floating point constant to the corresponding signed
2561 integer constant. TYPE must be a scalar or vector integer type. CST
2562 must be of scalar or vector floating point type. Both CST and TYPE
2563 must be scalars, or vectors of the same number of elements. If the
2564 value won't fit in the integer type, the results are undefined.
2565 ``uitofp (CST to TYPE)``
2566 Convert an unsigned integer constant to the corresponding floating
2567 point constant. TYPE must be a scalar or vector floating point type.
2568 CST must be of scalar or vector integer type. Both CST and TYPE must
2569 be scalars, or vectors of the same number of elements. If the value
2570 won't fit in the floating point type, the results are undefined.
2571 ``sitofp (CST to TYPE)``
2572 Convert a signed integer constant to the corresponding floating
2573 point constant. TYPE must be a scalar or vector floating point type.
2574 CST must be of scalar or vector integer type. Both CST and TYPE must
2575 be scalars, or vectors of the same number of elements. If the value
2576 won't fit in the floating point type, the results are undefined.
2577 ``ptrtoint (CST to TYPE)``
2578 Convert a pointer typed constant to the corresponding integer
2579 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2580 pointer type. The ``CST`` value is zero extended, truncated, or
2581 unchanged to make it fit in ``TYPE``.
2582 ``inttoptr (CST to TYPE)``
2583 Convert an integer constant to a pointer constant. TYPE must be a
2584 pointer type. CST must be of integer type. The CST value is zero
2585 extended, truncated, or unchanged to make it fit in a pointer size.
2586 This one is *really* dangerous!
2587 ``bitcast (CST to TYPE)``
2588 Convert a constant, CST, to another TYPE. The constraints of the
2589 operands are the same as those for the :ref:`bitcast
2590 instruction <i_bitcast>`.
2591 ``addrspacecast (CST to TYPE)``
2592 Convert a constant pointer or constant vector of pointer, CST, to another
2593 TYPE in a different address space. The constraints of the operands are the
2594 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2595 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2596 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2597 constants. As with the :ref:`getelementptr <i_getelementptr>`
2598 instruction, the index list may have zero or more indexes, which are
2599 required to make sense for the type of "CSTPTR".
2600 ``select (COND, VAL1, VAL2)``
2601 Perform the :ref:`select operation <i_select>` on constants.
2602 ``icmp COND (VAL1, VAL2)``
2603 Performs the :ref:`icmp operation <i_icmp>` on constants.
2604 ``fcmp COND (VAL1, VAL2)``
2605 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2606 ``extractelement (VAL, IDX)``
2607 Perform the :ref:`extractelement operation <i_extractelement>` on
2609 ``insertelement (VAL, ELT, IDX)``
2610 Perform the :ref:`insertelement operation <i_insertelement>` on
2612 ``shufflevector (VEC1, VEC2, IDXMASK)``
2613 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2615 ``extractvalue (VAL, IDX0, IDX1, ...)``
2616 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2617 constants. The index list is interpreted in a similar manner as
2618 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2619 least one index value must be specified.
2620 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2621 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2622 The index list is interpreted in a similar manner as indices in a
2623 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2624 value must be specified.
2625 ``OPCODE (LHS, RHS)``
2626 Perform the specified operation of the LHS and RHS constants. OPCODE
2627 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2628 binary <bitwiseops>` operations. The constraints on operands are
2629 the same as those for the corresponding instruction (e.g. no bitwise
2630 operations on floating point values are allowed).
2637 Inline Assembler Expressions
2638 ----------------------------
2640 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2641 Inline Assembly <moduleasm>`) through the use of a special value. This
2642 value represents the inline assembler as a string (containing the
2643 instructions to emit), a list of operand constraints (stored as a
2644 string), a flag that indicates whether or not the inline asm expression
2645 has side effects, and a flag indicating whether the function containing
2646 the asm needs to align its stack conservatively. An example inline
2647 assembler expression is:
2649 .. code-block:: llvm
2651 i32 (i32) asm "bswap $0", "=r,r"
2653 Inline assembler expressions may **only** be used as the callee operand
2654 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2655 Thus, typically we have:
2657 .. code-block:: llvm
2659 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2661 Inline asms with side effects not visible in the constraint list must be
2662 marked as having side effects. This is done through the use of the
2663 '``sideeffect``' keyword, like so:
2665 .. code-block:: llvm
2667 call void asm sideeffect "eieio", ""()
2669 In some cases inline asms will contain code that will not work unless
2670 the stack is aligned in some way, such as calls or SSE instructions on
2671 x86, yet will not contain code that does that alignment within the asm.
2672 The compiler should make conservative assumptions about what the asm
2673 might contain and should generate its usual stack alignment code in the
2674 prologue if the '``alignstack``' keyword is present:
2676 .. code-block:: llvm
2678 call void asm alignstack "eieio", ""()
2680 Inline asms also support using non-standard assembly dialects. The
2681 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2682 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2683 the only supported dialects. An example is:
2685 .. code-block:: llvm
2687 call void asm inteldialect "eieio", ""()
2689 If multiple keywords appear the '``sideeffect``' keyword must come
2690 first, the '``alignstack``' keyword second and the '``inteldialect``'
2696 The call instructions that wrap inline asm nodes may have a
2697 "``!srcloc``" MDNode attached to it that contains a list of constant
2698 integers. If present, the code generator will use the integer as the
2699 location cookie value when report errors through the ``LLVMContext``
2700 error reporting mechanisms. This allows a front-end to correlate backend
2701 errors that occur with inline asm back to the source code that produced
2704 .. code-block:: llvm
2706 call void asm sideeffect "something bad", ""(), !srcloc !42
2708 !42 = !{ i32 1234567 }
2710 It is up to the front-end to make sense of the magic numbers it places
2711 in the IR. If the MDNode contains multiple constants, the code generator
2712 will use the one that corresponds to the line of the asm that the error
2717 Metadata Nodes and Metadata Strings
2718 -----------------------------------
2720 LLVM IR allows metadata to be attached to instructions in the program
2721 that can convey extra information about the code to the optimizers and
2722 code generator. One example application of metadata is source-level
2723 debug information. There are two metadata primitives: strings and nodes.
2724 All metadata has the ``metadata`` type and is identified in syntax by a
2725 preceding exclamation point ('``!``').
2727 A metadata string is a string surrounded by double quotes. It can
2728 contain any character by escaping non-printable characters with
2729 "``\xx``" where "``xx``" is the two digit hex code. For example:
2732 Metadata nodes are represented with notation similar to structure
2733 constants (a comma separated list of elements, surrounded by braces and
2734 preceded by an exclamation point). Metadata nodes can have any values as
2735 their operand. For example:
2737 .. code-block:: llvm
2739 !{ metadata !"test\00", i32 10}
2741 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2742 metadata nodes, which can be looked up in the module symbol table. For
2745 .. code-block:: llvm
2747 !foo = metadata !{!4, !3}
2749 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2750 function is using two metadata arguments:
2752 .. code-block:: llvm
2754 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2756 Metadata can be attached with an instruction. Here metadata ``!21`` is
2757 attached to the ``add`` instruction using the ``!dbg`` identifier:
2759 .. code-block:: llvm
2761 %indvar.next = add i64 %indvar, 1, !dbg !21
2763 More information about specific metadata nodes recognized by the
2764 optimizers and code generator is found below.
2769 In LLVM IR, memory does not have types, so LLVM's own type system is not
2770 suitable for doing TBAA. Instead, metadata is added to the IR to
2771 describe a type system of a higher level language. This can be used to
2772 implement typical C/C++ TBAA, but it can also be used to implement
2773 custom alias analysis behavior for other languages.
2775 The current metadata format is very simple. TBAA metadata nodes have up
2776 to three fields, e.g.:
2778 .. code-block:: llvm
2780 !0 = metadata !{ metadata !"an example type tree" }
2781 !1 = metadata !{ metadata !"int", metadata !0 }
2782 !2 = metadata !{ metadata !"float", metadata !0 }
2783 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2785 The first field is an identity field. It can be any value, usually a
2786 metadata string, which uniquely identifies the type. The most important
2787 name in the tree is the name of the root node. Two trees with different
2788 root node names are entirely disjoint, even if they have leaves with
2791 The second field identifies the type's parent node in the tree, or is
2792 null or omitted for a root node. A type is considered to alias all of
2793 its descendants and all of its ancestors in the tree. Also, a type is
2794 considered to alias all types in other trees, so that bitcode produced
2795 from multiple front-ends is handled conservatively.
2797 If the third field is present, it's an integer which if equal to 1
2798 indicates that the type is "constant" (meaning
2799 ``pointsToConstantMemory`` should return true; see `other useful
2800 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2802 '``tbaa.struct``' Metadata
2803 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2805 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2806 aggregate assignment operations in C and similar languages, however it
2807 is defined to copy a contiguous region of memory, which is more than
2808 strictly necessary for aggregate types which contain holes due to
2809 padding. Also, it doesn't contain any TBAA information about the fields
2812 ``!tbaa.struct`` metadata can describe which memory subregions in a
2813 memcpy are padding and what the TBAA tags of the struct are.
2815 The current metadata format is very simple. ``!tbaa.struct`` metadata
2816 nodes are a list of operands which are in conceptual groups of three.
2817 For each group of three, the first operand gives the byte offset of a
2818 field in bytes, the second gives its size in bytes, and the third gives
2821 .. code-block:: llvm
2823 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2825 This describes a struct with two fields. The first is at offset 0 bytes
2826 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2827 and has size 4 bytes and has tbaa tag !2.
2829 Note that the fields need not be contiguous. In this example, there is a
2830 4 byte gap between the two fields. This gap represents padding which
2831 does not carry useful data and need not be preserved.
2833 '``noalias``' and '``alias.scope``' Metadata
2834 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2836 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
2837 noalias memory-access sets. This means that some collection of memory access
2838 instructions (loads, stores, memory-accessing calls, etc.) that carry
2839 ``noalias`` metadata can specifically be specified not to alias with some other
2840 collection of memory access instructions that carry ``alias.scope`` metadata.
2841 Each type of metadata specifies a list of scopes where each scope has an id and
2842 a domain. When evaluating an aliasing query, if for some some domain, the set
2843 of scopes with that domain in one instruction's ``alias.scope`` list is a
2844 subset of (or qual to) the set of scopes for that domain in another
2845 instruction's ``noalias`` list, then the two memory accesses are assumed not to
2848 The metadata identifying each domain is itself a list containing one or two
2849 entries. The first entry is the name of the domain. Note that if the name is a
2850 string then it can be combined accross functions and translation units. A
2851 self-reference can be used to create globally unique domain names. A
2852 descriptive string may optionally be provided as a second list entry.
2854 The metadata identifying each scope is also itself a list containing two or
2855 three entries. The first entry is the name of the scope. Note that if the name
2856 is a string then it can be combined accross functions and translation units. A
2857 self-reference can be used to create globally unique scope names. A metadata
2858 reference to the scope's domain is the second entry. A descriptive string may
2859 optionally be provided as a third list entry.
2863 .. code-block:: llvm
2865 ; Two scope domains:
2866 !0 = metadata !{metadata !0}
2867 !1 = metadata !{metadata !1}
2869 ; Some scopes in these domains:
2870 !2 = metadata !{metadata !2, metadata !0}
2871 !3 = metadata !{metadata !3, metadata !0}
2872 !4 = metadata !{metadata !4, metadata !1}
2875 !5 = metadata !{metadata !4} ; A list containing only scope !4
2876 !6 = metadata !{metadata !4, metadata !3, metadata !2}
2877 !7 = metadata !{metadata !3}
2879 ; These two instructions don't alias:
2880 %0 = load float* %c, align 4, !alias.scope !5
2881 store float %0, float* %arrayidx.i, align 4, !noalias !5
2883 ; These two instructions also don't alias (for domain !1, the set of scopes
2884 ; in the !alias.scope equals that in the !noalias list):
2885 %2 = load float* %c, align 4, !alias.scope !5
2886 store float %2, float* %arrayidx.i2, align 4, !noalias !6
2888 ; These two instructions don't alias (for domain !0, the set of scopes in
2889 ; the !noalias list is not a superset of, or equal to, the scopes in the
2890 ; !alias.scope list):
2891 %2 = load float* %c, align 4, !alias.scope !6
2892 store float %0, float* %arrayidx.i, align 4, !noalias !7
2894 '``fpmath``' Metadata
2895 ^^^^^^^^^^^^^^^^^^^^^
2897 ``fpmath`` metadata may be attached to any instruction of floating point
2898 type. It can be used to express the maximum acceptable error in the
2899 result of that instruction, in ULPs, thus potentially allowing the
2900 compiler to use a more efficient but less accurate method of computing
2901 it. ULP is defined as follows:
2903 If ``x`` is a real number that lies between two finite consecutive
2904 floating-point numbers ``a`` and ``b``, without being equal to one
2905 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2906 distance between the two non-equal finite floating-point numbers
2907 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2909 The metadata node shall consist of a single positive floating point
2910 number representing the maximum relative error, for example:
2912 .. code-block:: llvm
2914 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2916 '``range``' Metadata
2917 ^^^^^^^^^^^^^^^^^^^^
2919 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
2920 integer types. It expresses the possible ranges the loaded value or the value
2921 returned by the called function at this call site is in. The ranges are
2922 represented with a flattened list of integers. The loaded value or the value
2923 returned is known to be in the union of the ranges defined by each consecutive
2924 pair. Each pair has the following properties:
2926 - The type must match the type loaded by the instruction.
2927 - The pair ``a,b`` represents the range ``[a,b)``.
2928 - Both ``a`` and ``b`` are constants.
2929 - The range is allowed to wrap.
2930 - The range should not represent the full or empty set. That is,
2933 In addition, the pairs must be in signed order of the lower bound and
2934 they must be non-contiguous.
2938 .. code-block:: llvm
2940 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2941 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2942 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
2943 %d = invoke i8 @bar() to label %cont
2944 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
2946 !0 = metadata !{ i8 0, i8 2 }
2947 !1 = metadata !{ i8 255, i8 2 }
2948 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2949 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2954 It is sometimes useful to attach information to loop constructs. Currently,
2955 loop metadata is implemented as metadata attached to the branch instruction
2956 in the loop latch block. This type of metadata refer to a metadata node that is
2957 guaranteed to be separate for each loop. The loop identifier metadata is
2958 specified with the name ``llvm.loop``.
2960 The loop identifier metadata is implemented using a metadata that refers to
2961 itself to avoid merging it with any other identifier metadata, e.g.,
2962 during module linkage or function inlining. That is, each loop should refer
2963 to their own identification metadata even if they reside in separate functions.
2964 The following example contains loop identifier metadata for two separate loop
2967 .. code-block:: llvm
2969 !0 = metadata !{ metadata !0 }
2970 !1 = metadata !{ metadata !1 }
2972 The loop identifier metadata can be used to specify additional
2973 per-loop metadata. Any operands after the first operand can be treated
2974 as user-defined metadata. For example the ``llvm.loop.unroll.count``
2975 suggests an unroll factor to the loop unroller:
2977 .. code-block:: llvm
2979 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2981 !0 = metadata !{ metadata !0, metadata !1 }
2982 !1 = metadata !{ metadata !"llvm.loop.unroll.count", i32 4 }
2984 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
2985 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2987 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
2988 used to control per-loop vectorization and interleaving parameters such as
2989 vectorization width and interleave count. These metadata should be used in
2990 conjunction with ``llvm.loop`` loop identification metadata. The
2991 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
2992 optimization hints and the optimizer will only interleave and vectorize loops if
2993 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
2994 which contains information about loop-carried memory dependencies can be helpful
2995 in determining the safety of these transformations.
2997 '``llvm.loop.interleave.count``' Metadata
2998 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3000 This metadata suggests an interleave count to the loop interleaver.
3001 The first operand is the string ``llvm.loop.interleave.count`` and the
3002 second operand is an integer specifying the interleave count. For
3005 .. code-block:: llvm
3007 !0 = metadata !{ metadata !"llvm.loop.interleave.count", i32 4 }
3009 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
3010 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
3011 then the interleave count will be determined automatically.
3013 '``llvm.loop.vectorize.enable``' Metadata
3014 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3016 This metadata selectively enables or disables vectorization for the loop. The
3017 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
3018 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
3019 0 disables vectorization:
3021 .. code-block:: llvm
3023 !0 = metadata !{ metadata !"llvm.loop.vectorize.enable", i1 0 }
3024 !1 = metadata !{ metadata !"llvm.loop.vectorize.enable", i1 1 }
3026 '``llvm.loop.vectorize.width``' Metadata
3027 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3029 This metadata sets the target width of the vectorizer. The first
3030 operand is the string ``llvm.loop.vectorize.width`` and the second
3031 operand is an integer specifying the width. For example:
3033 .. code-block:: llvm
3035 !0 = metadata !{ metadata !"llvm.loop.vectorize.width", i32 4 }
3037 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
3038 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
3039 0 or if the loop does not have this metadata the width will be
3040 determined automatically.
3042 '``llvm.loop.unroll``'
3043 ^^^^^^^^^^^^^^^^^^^^^^
3045 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
3046 optimization hints such as the unroll factor. ``llvm.loop.unroll``
3047 metadata should be used in conjunction with ``llvm.loop`` loop
3048 identification metadata. The ``llvm.loop.unroll`` metadata are only
3049 optimization hints and the unrolling will only be performed if the
3050 optimizer believes it is safe to do so.
3052 '``llvm.loop.unroll.count``' Metadata
3053 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3055 This metadata suggests an unroll factor to the loop unroller. The
3056 first operand is the string ``llvm.loop.unroll.count`` and the second
3057 operand is a positive integer specifying the unroll factor. For
3060 .. code-block:: llvm
3062 !0 = metadata !{ metadata !"llvm.loop.unroll.count", i32 4 }
3064 If the trip count of the loop is less than the unroll count the loop
3065 will be partially unrolled.
3067 '``llvm.loop.unroll.disable``' Metadata
3068 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3070 This metadata either disables loop unrolling. The metadata has a single operand
3071 which is the string ``llvm.loop.unroll.disable``. For example:
3073 .. code-block:: llvm
3075 !0 = metadata !{ metadata !"llvm.loop.unroll.disable" }
3077 '``llvm.loop.unroll.full``' Metadata
3078 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3080 This metadata either suggests that the loop should be unrolled fully. The
3081 metadata has a single operand which is the string ``llvm.loop.unroll.disable``.
3084 .. code-block:: llvm
3086 !0 = metadata !{ metadata !"llvm.loop.unroll.full" }
3091 Metadata types used to annotate memory accesses with information helpful
3092 for optimizations are prefixed with ``llvm.mem``.
3094 '``llvm.mem.parallel_loop_access``' Metadata
3095 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3097 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
3098 or metadata containing a list of loop identifiers for nested loops.
3099 The metadata is attached to memory accessing instructions and denotes that
3100 no loop carried memory dependence exist between it and other instructions denoted
3101 with the same loop identifier.
3103 Precisely, given two instructions ``m1`` and ``m2`` that both have the
3104 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
3105 set of loops associated with that metadata, respectively, then there is no loop
3106 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
3109 As a special case, if all memory accessing instructions in a loop have
3110 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
3111 loop has no loop carried memory dependences and is considered to be a parallel
3114 Note that if not all memory access instructions have such metadata referring to
3115 the loop, then the loop is considered not being trivially parallel. Additional
3116 memory dependence analysis is required to make that determination. As a fail
3117 safe mechanism, this causes loops that were originally parallel to be considered
3118 sequential (if optimization passes that are unaware of the parallel semantics
3119 insert new memory instructions into the loop body).
3121 Example of a loop that is considered parallel due to its correct use of
3122 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
3123 metadata types that refer to the same loop identifier metadata.
3125 .. code-block:: llvm
3129 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
3131 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3133 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
3137 !0 = metadata !{ metadata !0 }
3139 It is also possible to have nested parallel loops. In that case the
3140 memory accesses refer to a list of loop identifier metadata nodes instead of
3141 the loop identifier metadata node directly:
3143 .. code-block:: llvm
3147 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
3149 br label %inner.for.body
3153 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3155 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
3157 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
3161 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
3163 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
3165 outer.for.end: ; preds = %for.body
3167 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
3168 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
3169 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
3171 Module Flags Metadata
3172 =====================
3174 Information about the module as a whole is difficult to convey to LLVM's
3175 subsystems. The LLVM IR isn't sufficient to transmit this information.
3176 The ``llvm.module.flags`` named metadata exists in order to facilitate
3177 this. These flags are in the form of key / value pairs --- much like a
3178 dictionary --- making it easy for any subsystem who cares about a flag to
3181 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
3182 Each triplet has the following form:
3184 - The first element is a *behavior* flag, which specifies the behavior
3185 when two (or more) modules are merged together, and it encounters two
3186 (or more) metadata with the same ID. The supported behaviors are
3188 - The second element is a metadata string that is a unique ID for the
3189 metadata. Each module may only have one flag entry for each unique ID (not
3190 including entries with the **Require** behavior).
3191 - The third element is the value of the flag.
3193 When two (or more) modules are merged together, the resulting
3194 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
3195 each unique metadata ID string, there will be exactly one entry in the merged
3196 modules ``llvm.module.flags`` metadata table, and the value for that entry will
3197 be determined by the merge behavior flag, as described below. The only exception
3198 is that entries with the *Require* behavior are always preserved.
3200 The following behaviors are supported:
3211 Emits an error if two values disagree, otherwise the resulting value
3212 is that of the operands.
3216 Emits a warning if two values disagree. The result value will be the
3217 operand for the flag from the first module being linked.
3221 Adds a requirement that another module flag be present and have a
3222 specified value after linking is performed. The value must be a
3223 metadata pair, where the first element of the pair is the ID of the
3224 module flag to be restricted, and the second element of the pair is
3225 the value the module flag should be restricted to. This behavior can
3226 be used to restrict the allowable results (via triggering of an
3227 error) of linking IDs with the **Override** behavior.
3231 Uses the specified value, regardless of the behavior or value of the
3232 other module. If both modules specify **Override**, but the values
3233 differ, an error will be emitted.
3237 Appends the two values, which are required to be metadata nodes.
3241 Appends the two values, which are required to be metadata
3242 nodes. However, duplicate entries in the second list are dropped
3243 during the append operation.
3245 It is an error for a particular unique flag ID to have multiple behaviors,
3246 except in the case of **Require** (which adds restrictions on another metadata
3247 value) or **Override**.
3249 An example of module flags:
3251 .. code-block:: llvm
3253 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3254 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3255 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3256 !3 = metadata !{ i32 3, metadata !"qux",
3258 metadata !"foo", i32 1
3261 !llvm.module.flags = !{ !0, !1, !2, !3 }
3263 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3264 if two or more ``!"foo"`` flags are seen is to emit an error if their
3265 values are not equal.
3267 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3268 behavior if two or more ``!"bar"`` flags are seen is to use the value
3271 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3272 behavior if two or more ``!"qux"`` flags are seen is to emit a
3273 warning if their values are not equal.
3275 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3279 metadata !{ metadata !"foo", i32 1 }
3281 The behavior is to emit an error if the ``llvm.module.flags`` does not
3282 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3285 Objective-C Garbage Collection Module Flags Metadata
3286 ----------------------------------------------------
3288 On the Mach-O platform, Objective-C stores metadata about garbage
3289 collection in a special section called "image info". The metadata
3290 consists of a version number and a bitmask specifying what types of
3291 garbage collection are supported (if any) by the file. If two or more
3292 modules are linked together their garbage collection metadata needs to
3293 be merged rather than appended together.
3295 The Objective-C garbage collection module flags metadata consists of the
3296 following key-value pairs:
3305 * - ``Objective-C Version``
3306 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3308 * - ``Objective-C Image Info Version``
3309 - **[Required]** --- The version of the image info section. Currently
3312 * - ``Objective-C Image Info Section``
3313 - **[Required]** --- The section to place the metadata. Valid values are
3314 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3315 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3316 Objective-C ABI version 2.
3318 * - ``Objective-C Garbage Collection``
3319 - **[Required]** --- Specifies whether garbage collection is supported or
3320 not. Valid values are 0, for no garbage collection, and 2, for garbage
3321 collection supported.
3323 * - ``Objective-C GC Only``
3324 - **[Optional]** --- Specifies that only garbage collection is supported.
3325 If present, its value must be 6. This flag requires that the
3326 ``Objective-C Garbage Collection`` flag have the value 2.
3328 Some important flag interactions:
3330 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3331 merged with a module with ``Objective-C Garbage Collection`` set to
3332 2, then the resulting module has the
3333 ``Objective-C Garbage Collection`` flag set to 0.
3334 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3335 merged with a module with ``Objective-C GC Only`` set to 6.
3337 Automatic Linker Flags Module Flags Metadata
3338 --------------------------------------------
3340 Some targets support embedding flags to the linker inside individual object
3341 files. Typically this is used in conjunction with language extensions which
3342 allow source files to explicitly declare the libraries they depend on, and have
3343 these automatically be transmitted to the linker via object files.
3345 These flags are encoded in the IR using metadata in the module flags section,
3346 using the ``Linker Options`` key. The merge behavior for this flag is required
3347 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3348 node which should be a list of other metadata nodes, each of which should be a
3349 list of metadata strings defining linker options.
3351 For example, the following metadata section specifies two separate sets of
3352 linker options, presumably to link against ``libz`` and the ``Cocoa``
3355 !0 = metadata !{ i32 6, metadata !"Linker Options",
3357 metadata !{ metadata !"-lz" },
3358 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3359 !llvm.module.flags = !{ !0 }
3361 The metadata encoding as lists of lists of options, as opposed to a collapsed
3362 list of options, is chosen so that the IR encoding can use multiple option
3363 strings to specify e.g., a single library, while still having that specifier be
3364 preserved as an atomic element that can be recognized by a target specific
3365 assembly writer or object file emitter.
3367 Each individual option is required to be either a valid option for the target's
3368 linker, or an option that is reserved by the target specific assembly writer or
3369 object file emitter. No other aspect of these options is defined by the IR.
3371 C type width Module Flags Metadata
3372 ----------------------------------
3374 The ARM backend emits a section into each generated object file describing the
3375 options that it was compiled with (in a compiler-independent way) to prevent
3376 linking incompatible objects, and to allow automatic library selection. Some
3377 of these options are not visible at the IR level, namely wchar_t width and enum
3380 To pass this information to the backend, these options are encoded in module
3381 flags metadata, using the following key-value pairs:
3391 - * 0 --- sizeof(wchar_t) == 4
3392 * 1 --- sizeof(wchar_t) == 2
3395 - * 0 --- Enums are at least as large as an ``int``.
3396 * 1 --- Enums are stored in the smallest integer type which can
3397 represent all of its values.
3399 For example, the following metadata section specifies that the module was
3400 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
3401 enum is the smallest type which can represent all of its values::
3403 !llvm.module.flags = !{!0, !1}
3404 !0 = metadata !{i32 1, metadata !"short_wchar", i32 1}
3405 !1 = metadata !{i32 1, metadata !"short_enum", i32 0}
3407 .. _intrinsicglobalvariables:
3409 Intrinsic Global Variables
3410 ==========================
3412 LLVM has a number of "magic" global variables that contain data that
3413 affect code generation or other IR semantics. These are documented here.
3414 All globals of this sort should have a section specified as
3415 "``llvm.metadata``". This section and all globals that start with
3416 "``llvm.``" are reserved for use by LLVM.
3420 The '``llvm.used``' Global Variable
3421 -----------------------------------
3423 The ``@llvm.used`` global is an array which has
3424 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3425 pointers to named global variables, functions and aliases which may optionally
3426 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3429 .. code-block:: llvm
3434 @llvm.used = appending global [2 x i8*] [
3436 i8* bitcast (i32* @Y to i8*)
3437 ], section "llvm.metadata"
3439 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3440 and linker are required to treat the symbol as if there is a reference to the
3441 symbol that it cannot see (which is why they have to be named). For example, if
3442 a variable has internal linkage and no references other than that from the
3443 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3444 references from inline asms and other things the compiler cannot "see", and
3445 corresponds to "``attribute((used))``" in GNU C.
3447 On some targets, the code generator must emit a directive to the
3448 assembler or object file to prevent the assembler and linker from
3449 molesting the symbol.
3451 .. _gv_llvmcompilerused:
3453 The '``llvm.compiler.used``' Global Variable
3454 --------------------------------------------
3456 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3457 directive, except that it only prevents the compiler from touching the
3458 symbol. On targets that support it, this allows an intelligent linker to
3459 optimize references to the symbol without being impeded as it would be
3462 This is a rare construct that should only be used in rare circumstances,
3463 and should not be exposed to source languages.
3465 .. _gv_llvmglobalctors:
3467 The '``llvm.global_ctors``' Global Variable
3468 -------------------------------------------
3470 .. code-block:: llvm
3472 %0 = type { i32, void ()*, i8* }
3473 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
3475 The ``@llvm.global_ctors`` array contains a list of constructor
3476 functions, priorities, and an optional associated global or function.
3477 The functions referenced by this array will be called in ascending order
3478 of priority (i.e. lowest first) when the module is loaded. The order of
3479 functions with the same priority is not defined.
3481 If the third field is present, non-null, and points to a global variable
3482 or function, the initializer function will only run if the associated
3483 data from the current module is not discarded.
3485 .. _llvmglobaldtors:
3487 The '``llvm.global_dtors``' Global Variable
3488 -------------------------------------------
3490 .. code-block:: llvm
3492 %0 = type { i32, void ()*, i8* }
3493 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
3495 The ``@llvm.global_dtors`` array contains a list of destructor
3496 functions, priorities, and an optional associated global or function.
3497 The functions referenced by this array will be called in descending
3498 order of priority (i.e. highest first) when the module is unloaded. The
3499 order of functions with the same priority is not defined.
3501 If the third field is present, non-null, and points to a global variable
3502 or function, the destructor function will only run if the associated
3503 data from the current module is not discarded.
3505 Instruction Reference
3506 =====================
3508 The LLVM instruction set consists of several different classifications
3509 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3510 instructions <binaryops>`, :ref:`bitwise binary
3511 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3512 :ref:`other instructions <otherops>`.
3516 Terminator Instructions
3517 -----------------------
3519 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3520 program ends with a "Terminator" instruction, which indicates which
3521 block should be executed after the current block is finished. These
3522 terminator instructions typically yield a '``void``' value: they produce
3523 control flow, not values (the one exception being the
3524 ':ref:`invoke <i_invoke>`' instruction).
3526 The terminator instructions are: ':ref:`ret <i_ret>`',
3527 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3528 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3529 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3533 '``ret``' Instruction
3534 ^^^^^^^^^^^^^^^^^^^^^
3541 ret <type> <value> ; Return a value from a non-void function
3542 ret void ; Return from void function
3547 The '``ret``' instruction is used to return control flow (and optionally
3548 a value) from a function back to the caller.
3550 There are two forms of the '``ret``' instruction: one that returns a
3551 value and then causes control flow, and one that just causes control
3557 The '``ret``' instruction optionally accepts a single argument, the
3558 return value. The type of the return value must be a ':ref:`first
3559 class <t_firstclass>`' type.
3561 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3562 return type and contains a '``ret``' instruction with no return value or
3563 a return value with a type that does not match its type, or if it has a
3564 void return type and contains a '``ret``' instruction with a return
3570 When the '``ret``' instruction is executed, control flow returns back to
3571 the calling function's context. If the caller is a
3572 ":ref:`call <i_call>`" instruction, execution continues at the
3573 instruction after the call. If the caller was an
3574 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3575 beginning of the "normal" destination block. If the instruction returns
3576 a value, that value shall set the call or invoke instruction's return
3582 .. code-block:: llvm
3584 ret i32 5 ; Return an integer value of 5
3585 ret void ; Return from a void function
3586 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3590 '``br``' Instruction
3591 ^^^^^^^^^^^^^^^^^^^^
3598 br i1 <cond>, label <iftrue>, label <iffalse>
3599 br label <dest> ; Unconditional branch
3604 The '``br``' instruction is used to cause control flow to transfer to a
3605 different basic block in the current function. There are two forms of
3606 this instruction, corresponding to a conditional branch and an
3607 unconditional branch.
3612 The conditional branch form of the '``br``' instruction takes a single
3613 '``i1``' value and two '``label``' values. The unconditional form of the
3614 '``br``' instruction takes a single '``label``' value as a target.
3619 Upon execution of a conditional '``br``' instruction, the '``i1``'
3620 argument is evaluated. If the value is ``true``, control flows to the
3621 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3622 to the '``iffalse``' ``label`` argument.
3627 .. code-block:: llvm
3630 %cond = icmp eq i32 %a, %b
3631 br i1 %cond, label %IfEqual, label %IfUnequal
3639 '``switch``' Instruction
3640 ^^^^^^^^^^^^^^^^^^^^^^^^
3647 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3652 The '``switch``' instruction is used to transfer control flow to one of
3653 several different places. It is a generalization of the '``br``'
3654 instruction, allowing a branch to occur to one of many possible
3660 The '``switch``' instruction uses three parameters: an integer
3661 comparison value '``value``', a default '``label``' destination, and an
3662 array of pairs of comparison value constants and '``label``'s. The table
3663 is not allowed to contain duplicate constant entries.
3668 The ``switch`` instruction specifies a table of values and destinations.
3669 When the '``switch``' instruction is executed, this table is searched
3670 for the given value. If the value is found, control flow is transferred
3671 to the corresponding destination; otherwise, control flow is transferred
3672 to the default destination.
3677 Depending on properties of the target machine and the particular
3678 ``switch`` instruction, this instruction may be code generated in
3679 different ways. For example, it could be generated as a series of
3680 chained conditional branches or with a lookup table.
3685 .. code-block:: llvm
3687 ; Emulate a conditional br instruction
3688 %Val = zext i1 %value to i32
3689 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3691 ; Emulate an unconditional br instruction
3692 switch i32 0, label %dest [ ]
3694 ; Implement a jump table:
3695 switch i32 %val, label %otherwise [ i32 0, label %onzero
3697 i32 2, label %ontwo ]
3701 '``indirectbr``' Instruction
3702 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3709 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3714 The '``indirectbr``' instruction implements an indirect branch to a
3715 label within the current function, whose address is specified by
3716 "``address``". Address must be derived from a
3717 :ref:`blockaddress <blockaddress>` constant.
3722 The '``address``' argument is the address of the label to jump to. The
3723 rest of the arguments indicate the full set of possible destinations
3724 that the address may point to. Blocks are allowed to occur multiple
3725 times in the destination list, though this isn't particularly useful.
3727 This destination list is required so that dataflow analysis has an
3728 accurate understanding of the CFG.
3733 Control transfers to the block specified in the address argument. All
3734 possible destination blocks must be listed in the label list, otherwise
3735 this instruction has undefined behavior. This implies that jumps to
3736 labels defined in other functions have undefined behavior as well.
3741 This is typically implemented with a jump through a register.
3746 .. code-block:: llvm
3748 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3752 '``invoke``' Instruction
3753 ^^^^^^^^^^^^^^^^^^^^^^^^
3760 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3761 to label <normal label> unwind label <exception label>
3766 The '``invoke``' instruction causes control to transfer to a specified
3767 function, with the possibility of control flow transfer to either the
3768 '``normal``' label or the '``exception``' label. If the callee function
3769 returns with the "``ret``" instruction, control flow will return to the
3770 "normal" label. If the callee (or any indirect callees) returns via the
3771 ":ref:`resume <i_resume>`" instruction or other exception handling
3772 mechanism, control is interrupted and continued at the dynamically
3773 nearest "exception" label.
3775 The '``exception``' label is a `landing
3776 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3777 '``exception``' label is required to have the
3778 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3779 information about the behavior of the program after unwinding happens,
3780 as its first non-PHI instruction. The restrictions on the
3781 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3782 instruction, so that the important information contained within the
3783 "``landingpad``" instruction can't be lost through normal code motion.
3788 This instruction requires several arguments:
3790 #. The optional "cconv" marker indicates which :ref:`calling
3791 convention <callingconv>` the call should use. If none is
3792 specified, the call defaults to using C calling conventions.
3793 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3794 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3796 #. '``ptr to function ty``': shall be the signature of the pointer to
3797 function value being invoked. In most cases, this is a direct
3798 function invocation, but indirect ``invoke``'s are just as possible,
3799 branching off an arbitrary pointer to function value.
3800 #. '``function ptr val``': An LLVM value containing a pointer to a
3801 function to be invoked.
3802 #. '``function args``': argument list whose types match the function
3803 signature argument types and parameter attributes. All arguments must
3804 be of :ref:`first class <t_firstclass>` type. If the function signature
3805 indicates the function accepts a variable number of arguments, the
3806 extra arguments can be specified.
3807 #. '``normal label``': the label reached when the called function
3808 executes a '``ret``' instruction.
3809 #. '``exception label``': the label reached when a callee returns via
3810 the :ref:`resume <i_resume>` instruction or other exception handling
3812 #. The optional :ref:`function attributes <fnattrs>` list. Only
3813 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3814 attributes are valid here.
3819 This instruction is designed to operate as a standard '``call``'
3820 instruction in most regards. The primary difference is that it
3821 establishes an association with a label, which is used by the runtime
3822 library to unwind the stack.
3824 This instruction is used in languages with destructors to ensure that
3825 proper cleanup is performed in the case of either a ``longjmp`` or a
3826 thrown exception. Additionally, this is important for implementation of
3827 '``catch``' clauses in high-level languages that support them.
3829 For the purposes of the SSA form, the definition of the value returned
3830 by the '``invoke``' instruction is deemed to occur on the edge from the
3831 current block to the "normal" label. If the callee unwinds then no
3832 return value is available.
3837 .. code-block:: llvm
3839 %retval = invoke i32 @Test(i32 15) to label %Continue
3840 unwind label %TestCleanup ; i32:retval set
3841 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3842 unwind label %TestCleanup ; i32:retval set
3846 '``resume``' Instruction
3847 ^^^^^^^^^^^^^^^^^^^^^^^^
3854 resume <type> <value>
3859 The '``resume``' instruction is a terminator instruction that has no
3865 The '``resume``' instruction requires one argument, which must have the
3866 same type as the result of any '``landingpad``' instruction in the same
3872 The '``resume``' instruction resumes propagation of an existing
3873 (in-flight) exception whose unwinding was interrupted with a
3874 :ref:`landingpad <i_landingpad>` instruction.
3879 .. code-block:: llvm
3881 resume { i8*, i32 } %exn
3885 '``unreachable``' Instruction
3886 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3898 The '``unreachable``' instruction has no defined semantics. This
3899 instruction is used to inform the optimizer that a particular portion of
3900 the code is not reachable. This can be used to indicate that the code
3901 after a no-return function cannot be reached, and other facts.
3906 The '``unreachable``' instruction has no defined semantics.
3913 Binary operators are used to do most of the computation in a program.
3914 They require two operands of the same type, execute an operation on
3915 them, and produce a single value. The operands might represent multiple
3916 data, as is the case with the :ref:`vector <t_vector>` data type. The
3917 result value has the same type as its operands.
3919 There are several different binary operators:
3923 '``add``' Instruction
3924 ^^^^^^^^^^^^^^^^^^^^^
3931 <result> = add <ty> <op1>, <op2> ; yields ty:result
3932 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
3933 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
3934 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
3939 The '``add``' instruction returns the sum of its two operands.
3944 The two arguments to the '``add``' instruction must be
3945 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3946 arguments must have identical types.
3951 The value produced is the integer sum of the two operands.
3953 If the sum has unsigned overflow, the result returned is the
3954 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3957 Because LLVM integers use a two's complement representation, this
3958 instruction is appropriate for both signed and unsigned integers.
3960 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3961 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3962 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3963 unsigned and/or signed overflow, respectively, occurs.
3968 .. code-block:: llvm
3970 <result> = add i32 4, %var ; yields i32:result = 4 + %var
3974 '``fadd``' Instruction
3975 ^^^^^^^^^^^^^^^^^^^^^^
3982 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
3987 The '``fadd``' instruction returns the sum of its two operands.
3992 The two arguments to the '``fadd``' instruction must be :ref:`floating
3993 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3994 Both arguments must have identical types.
3999 The value produced is the floating point sum of the two operands. This
4000 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
4001 which are optimization hints to enable otherwise unsafe floating point
4007 .. code-block:: llvm
4009 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
4011 '``sub``' Instruction
4012 ^^^^^^^^^^^^^^^^^^^^^
4019 <result> = sub <ty> <op1>, <op2> ; yields ty:result
4020 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
4021 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
4022 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
4027 The '``sub``' instruction returns the difference of its two operands.
4029 Note that the '``sub``' instruction is used to represent the '``neg``'
4030 instruction present in most other intermediate representations.
4035 The two arguments to the '``sub``' instruction must be
4036 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4037 arguments must have identical types.
4042 The value produced is the integer difference of the two operands.
4044 If the difference has unsigned overflow, the result returned is the
4045 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4048 Because LLVM integers use a two's complement representation, this
4049 instruction is appropriate for both signed and unsigned integers.
4051 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4052 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4053 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
4054 unsigned and/or signed overflow, respectively, occurs.
4059 .. code-block:: llvm
4061 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
4062 <result> = sub i32 0, %val ; yields i32:result = -%var
4066 '``fsub``' Instruction
4067 ^^^^^^^^^^^^^^^^^^^^^^
4074 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4079 The '``fsub``' instruction returns the difference of its two operands.
4081 Note that the '``fsub``' instruction is used to represent the '``fneg``'
4082 instruction present in most other intermediate representations.
4087 The two arguments to the '``fsub``' instruction must be :ref:`floating
4088 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4089 Both arguments must have identical types.
4094 The value produced is the floating point difference of the two operands.
4095 This instruction can also take any number of :ref:`fast-math
4096 flags <fastmath>`, which are optimization hints to enable otherwise
4097 unsafe floating point optimizations:
4102 .. code-block:: llvm
4104 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
4105 <result> = fsub float -0.0, %val ; yields float:result = -%var
4107 '``mul``' Instruction
4108 ^^^^^^^^^^^^^^^^^^^^^
4115 <result> = mul <ty> <op1>, <op2> ; yields ty:result
4116 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
4117 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
4118 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
4123 The '``mul``' instruction returns the product of its two operands.
4128 The two arguments to the '``mul``' instruction must be
4129 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4130 arguments must have identical types.
4135 The value produced is the integer product of the two operands.
4137 If the result of the multiplication has unsigned overflow, the result
4138 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
4139 bit width of the result.
4141 Because LLVM integers use a two's complement representation, and the
4142 result is the same width as the operands, this instruction returns the
4143 correct result for both signed and unsigned integers. If a full product
4144 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
4145 sign-extended or zero-extended as appropriate to the width of the full
4148 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4149 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4150 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
4151 unsigned and/or signed overflow, respectively, occurs.
4156 .. code-block:: llvm
4158 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
4162 '``fmul``' Instruction
4163 ^^^^^^^^^^^^^^^^^^^^^^
4170 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4175 The '``fmul``' instruction returns the product of its two operands.
4180 The two arguments to the '``fmul``' instruction must be :ref:`floating
4181 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4182 Both arguments must have identical types.
4187 The value produced is the floating point product of the two operands.
4188 This instruction can also take any number of :ref:`fast-math
4189 flags <fastmath>`, which are optimization hints to enable otherwise
4190 unsafe floating point optimizations:
4195 .. code-block:: llvm
4197 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
4199 '``udiv``' Instruction
4200 ^^^^^^^^^^^^^^^^^^^^^^
4207 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
4208 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
4213 The '``udiv``' instruction returns the quotient of its two operands.
4218 The two arguments to the '``udiv``' instruction must be
4219 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4220 arguments must have identical types.
4225 The value produced is the unsigned integer quotient of the two operands.
4227 Note that unsigned integer division and signed integer division are
4228 distinct operations; for signed integer division, use '``sdiv``'.
4230 Division by zero leads to undefined behavior.
4232 If the ``exact`` keyword is present, the result value of the ``udiv`` is
4233 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
4234 such, "((a udiv exact b) mul b) == a").
4239 .. code-block:: llvm
4241 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
4243 '``sdiv``' Instruction
4244 ^^^^^^^^^^^^^^^^^^^^^^
4251 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
4252 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
4257 The '``sdiv``' instruction returns the quotient of its two operands.
4262 The two arguments to the '``sdiv``' instruction must be
4263 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4264 arguments must have identical types.
4269 The value produced is the signed integer quotient of the two operands
4270 rounded towards zero.
4272 Note that signed integer division and unsigned integer division are
4273 distinct operations; for unsigned integer division, use '``udiv``'.
4275 Division by zero leads to undefined behavior. Overflow also leads to
4276 undefined behavior; this is a rare case, but can occur, for example, by
4277 doing a 32-bit division of -2147483648 by -1.
4279 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4280 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4285 .. code-block:: llvm
4287 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
4291 '``fdiv``' Instruction
4292 ^^^^^^^^^^^^^^^^^^^^^^
4299 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4304 The '``fdiv``' instruction returns the quotient of its two operands.
4309 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4310 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4311 Both arguments must have identical types.
4316 The value produced is the floating point quotient of the two operands.
4317 This instruction can also take any number of :ref:`fast-math
4318 flags <fastmath>`, which are optimization hints to enable otherwise
4319 unsafe floating point optimizations:
4324 .. code-block:: llvm
4326 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
4328 '``urem``' Instruction
4329 ^^^^^^^^^^^^^^^^^^^^^^
4336 <result> = urem <ty> <op1>, <op2> ; yields ty:result
4341 The '``urem``' instruction returns the remainder from the unsigned
4342 division of its two arguments.
4347 The two arguments to the '``urem``' instruction must be
4348 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4349 arguments must have identical types.
4354 This instruction returns the unsigned integer *remainder* of a division.
4355 This instruction always performs an unsigned division to get the
4358 Note that unsigned integer remainder and signed integer remainder are
4359 distinct operations; for signed integer remainder, use '``srem``'.
4361 Taking the remainder of a division by zero leads to undefined behavior.
4366 .. code-block:: llvm
4368 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
4370 '``srem``' Instruction
4371 ^^^^^^^^^^^^^^^^^^^^^^
4378 <result> = srem <ty> <op1>, <op2> ; yields ty:result
4383 The '``srem``' instruction returns the remainder from the signed
4384 division of its two operands. This instruction can also take
4385 :ref:`vector <t_vector>` versions of the values in which case the elements
4391 The two arguments to the '``srem``' instruction must be
4392 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4393 arguments must have identical types.
4398 This instruction returns the *remainder* of a division (where the result
4399 is either zero or has the same sign as the dividend, ``op1``), not the
4400 *modulo* operator (where the result is either zero or has the same sign
4401 as the divisor, ``op2``) of a value. For more information about the
4402 difference, see `The Math
4403 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4404 table of how this is implemented in various languages, please see
4406 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4408 Note that signed integer remainder and unsigned integer remainder are
4409 distinct operations; for unsigned integer remainder, use '``urem``'.
4411 Taking the remainder of a division by zero leads to undefined behavior.
4412 Overflow also leads to undefined behavior; this is a rare case, but can
4413 occur, for example, by taking the remainder of a 32-bit division of
4414 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4415 rule lets srem be implemented using instructions that return both the
4416 result of the division and the remainder.)
4421 .. code-block:: llvm
4423 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
4427 '``frem``' Instruction
4428 ^^^^^^^^^^^^^^^^^^^^^^
4435 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4440 The '``frem``' instruction returns the remainder from the division of
4446 The two arguments to the '``frem``' instruction must be :ref:`floating
4447 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4448 Both arguments must have identical types.
4453 This instruction returns the *remainder* of a division. The remainder
4454 has the same sign as the dividend. This instruction can also take any
4455 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4456 to enable otherwise unsafe floating point optimizations:
4461 .. code-block:: llvm
4463 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
4467 Bitwise Binary Operations
4468 -------------------------
4470 Bitwise binary operators are used to do various forms of bit-twiddling
4471 in a program. They are generally very efficient instructions and can
4472 commonly be strength reduced from other instructions. They require two
4473 operands of the same type, execute an operation on them, and produce a
4474 single value. The resulting value is the same type as its operands.
4476 '``shl``' Instruction
4477 ^^^^^^^^^^^^^^^^^^^^^
4484 <result> = shl <ty> <op1>, <op2> ; yields ty:result
4485 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
4486 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
4487 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
4492 The '``shl``' instruction returns the first operand shifted to the left
4493 a specified number of bits.
4498 Both arguments to the '``shl``' instruction must be the same
4499 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4500 '``op2``' is treated as an unsigned value.
4505 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4506 where ``n`` is the width of the result. If ``op2`` is (statically or
4507 dynamically) negative or equal to or larger than the number of bits in
4508 ``op1``, the result is undefined. If the arguments are vectors, each
4509 vector element of ``op1`` is shifted by the corresponding shift amount
4512 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4513 value <poisonvalues>` if it shifts out any non-zero bits. If the
4514 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4515 value <poisonvalues>` if it shifts out any bits that disagree with the
4516 resultant sign bit. As such, NUW/NSW have the same semantics as they
4517 would if the shift were expressed as a mul instruction with the same
4518 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4523 .. code-block:: llvm
4525 <result> = shl i32 4, %var ; yields i32: 4 << %var
4526 <result> = shl i32 4, 2 ; yields i32: 16
4527 <result> = shl i32 1, 10 ; yields i32: 1024
4528 <result> = shl i32 1, 32 ; undefined
4529 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4531 '``lshr``' Instruction
4532 ^^^^^^^^^^^^^^^^^^^^^^
4539 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
4540 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
4545 The '``lshr``' instruction (logical shift right) returns the first
4546 operand shifted to the right a specified number of bits with zero fill.
4551 Both arguments to the '``lshr``' instruction must be the same
4552 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4553 '``op2``' is treated as an unsigned value.
4558 This instruction always performs a logical shift right operation. The
4559 most significant bits of the result will be filled with zero bits after
4560 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4561 than the number of bits in ``op1``, the result is undefined. If the
4562 arguments are vectors, each vector element of ``op1`` is shifted by the
4563 corresponding shift amount in ``op2``.
4565 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4566 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4572 .. code-block:: llvm
4574 <result> = lshr i32 4, 1 ; yields i32:result = 2
4575 <result> = lshr i32 4, 2 ; yields i32:result = 1
4576 <result> = lshr i8 4, 3 ; yields i8:result = 0
4577 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
4578 <result> = lshr i32 1, 32 ; undefined
4579 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4581 '``ashr``' Instruction
4582 ^^^^^^^^^^^^^^^^^^^^^^
4589 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
4590 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
4595 The '``ashr``' instruction (arithmetic shift right) returns the first
4596 operand shifted to the right a specified number of bits with sign
4602 Both arguments to the '``ashr``' instruction must be the same
4603 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4604 '``op2``' is treated as an unsigned value.
4609 This instruction always performs an arithmetic shift right operation,
4610 The most significant bits of the result will be filled with the sign bit
4611 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4612 than the number of bits in ``op1``, the result is undefined. If the
4613 arguments are vectors, each vector element of ``op1`` is shifted by the
4614 corresponding shift amount in ``op2``.
4616 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4617 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4623 .. code-block:: llvm
4625 <result> = ashr i32 4, 1 ; yields i32:result = 2
4626 <result> = ashr i32 4, 2 ; yields i32:result = 1
4627 <result> = ashr i8 4, 3 ; yields i8:result = 0
4628 <result> = ashr i8 -2, 1 ; yields i8:result = -1
4629 <result> = ashr i32 1, 32 ; undefined
4630 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4632 '``and``' Instruction
4633 ^^^^^^^^^^^^^^^^^^^^^
4640 <result> = and <ty> <op1>, <op2> ; yields ty:result
4645 The '``and``' instruction returns the bitwise logical and of its two
4651 The two arguments to the '``and``' instruction must be
4652 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4653 arguments must have identical types.
4658 The truth table used for the '``and``' instruction is:
4675 .. code-block:: llvm
4677 <result> = and i32 4, %var ; yields i32:result = 4 & %var
4678 <result> = and i32 15, 40 ; yields i32:result = 8
4679 <result> = and i32 4, 8 ; yields i32:result = 0
4681 '``or``' Instruction
4682 ^^^^^^^^^^^^^^^^^^^^
4689 <result> = or <ty> <op1>, <op2> ; yields ty:result
4694 The '``or``' instruction returns the bitwise logical inclusive or of its
4700 The two arguments to the '``or``' instruction must be
4701 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4702 arguments must have identical types.
4707 The truth table used for the '``or``' instruction is:
4726 <result> = or i32 4, %var ; yields i32:result = 4 | %var
4727 <result> = or i32 15, 40 ; yields i32:result = 47
4728 <result> = or i32 4, 8 ; yields i32:result = 12
4730 '``xor``' Instruction
4731 ^^^^^^^^^^^^^^^^^^^^^
4738 <result> = xor <ty> <op1>, <op2> ; yields ty:result
4743 The '``xor``' instruction returns the bitwise logical exclusive or of
4744 its two operands. The ``xor`` is used to implement the "one's
4745 complement" operation, which is the "~" operator in C.
4750 The two arguments to the '``xor``' instruction must be
4751 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4752 arguments must have identical types.
4757 The truth table used for the '``xor``' instruction is:
4774 .. code-block:: llvm
4776 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
4777 <result> = xor i32 15, 40 ; yields i32:result = 39
4778 <result> = xor i32 4, 8 ; yields i32:result = 12
4779 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
4784 LLVM supports several instructions to represent vector operations in a
4785 target-independent manner. These instructions cover the element-access
4786 and vector-specific operations needed to process vectors effectively.
4787 While LLVM does directly support these vector operations, many
4788 sophisticated algorithms will want to use target-specific intrinsics to
4789 take full advantage of a specific target.
4791 .. _i_extractelement:
4793 '``extractelement``' Instruction
4794 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4801 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
4806 The '``extractelement``' instruction extracts a single scalar element
4807 from a vector at a specified index.
4812 The first operand of an '``extractelement``' instruction is a value of
4813 :ref:`vector <t_vector>` type. The second operand is an index indicating
4814 the position from which to extract the element. The index may be a
4815 variable of any integer type.
4820 The result is a scalar of the same type as the element type of ``val``.
4821 Its value is the value at position ``idx`` of ``val``. If ``idx``
4822 exceeds the length of ``val``, the results are undefined.
4827 .. code-block:: llvm
4829 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4831 .. _i_insertelement:
4833 '``insertelement``' Instruction
4834 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4841 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
4846 The '``insertelement``' instruction inserts a scalar element into a
4847 vector at a specified index.
4852 The first operand of an '``insertelement``' instruction is a value of
4853 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4854 type must equal the element type of the first operand. The third operand
4855 is an index indicating the position at which to insert the value. The
4856 index may be a variable of any integer type.
4861 The result is a vector of the same type as ``val``. Its element values
4862 are those of ``val`` except at position ``idx``, where it gets the value
4863 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4869 .. code-block:: llvm
4871 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4873 .. _i_shufflevector:
4875 '``shufflevector``' Instruction
4876 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4883 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4888 The '``shufflevector``' instruction constructs a permutation of elements
4889 from two input vectors, returning a vector with the same element type as
4890 the input and length that is the same as the shuffle mask.
4895 The first two operands of a '``shufflevector``' instruction are vectors
4896 with the same type. The third argument is a shuffle mask whose element
4897 type is always 'i32'. The result of the instruction is a vector whose
4898 length is the same as the shuffle mask and whose element type is the
4899 same as the element type of the first two operands.
4901 The shuffle mask operand is required to be a constant vector with either
4902 constant integer or undef values.
4907 The elements of the two input vectors are numbered from left to right
4908 across both of the vectors. The shuffle mask operand specifies, for each
4909 element of the result vector, which element of the two input vectors the
4910 result element gets. The element selector may be undef (meaning "don't
4911 care") and the second operand may be undef if performing a shuffle from
4917 .. code-block:: llvm
4919 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4920 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4921 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4922 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4923 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4924 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4925 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4926 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4928 Aggregate Operations
4929 --------------------
4931 LLVM supports several instructions for working with
4932 :ref:`aggregate <t_aggregate>` values.
4936 '``extractvalue``' Instruction
4937 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4944 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4949 The '``extractvalue``' instruction extracts the value of a member field
4950 from an :ref:`aggregate <t_aggregate>` value.
4955 The first operand of an '``extractvalue``' instruction is a value of
4956 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4957 constant indices to specify which value to extract in a similar manner
4958 as indices in a '``getelementptr``' instruction.
4960 The major differences to ``getelementptr`` indexing are:
4962 - Since the value being indexed is not a pointer, the first index is
4963 omitted and assumed to be zero.
4964 - At least one index must be specified.
4965 - Not only struct indices but also array indices must be in bounds.
4970 The result is the value at the position in the aggregate specified by
4976 .. code-block:: llvm
4978 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4982 '``insertvalue``' Instruction
4983 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4990 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4995 The '``insertvalue``' instruction inserts a value into a member field in
4996 an :ref:`aggregate <t_aggregate>` value.
5001 The first operand of an '``insertvalue``' instruction is a value of
5002 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
5003 a first-class value to insert. The following operands are constant
5004 indices indicating the position at which to insert the value in a
5005 similar manner as indices in a '``extractvalue``' instruction. The value
5006 to insert must have the same type as the value identified by the
5012 The result is an aggregate of the same type as ``val``. Its value is
5013 that of ``val`` except that the value at the position specified by the
5014 indices is that of ``elt``.
5019 .. code-block:: llvm
5021 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
5022 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
5023 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
5027 Memory Access and Addressing Operations
5028 ---------------------------------------
5030 A key design point of an SSA-based representation is how it represents
5031 memory. In LLVM, no memory locations are in SSA form, which makes things
5032 very simple. This section describes how to read, write, and allocate
5037 '``alloca``' Instruction
5038 ^^^^^^^^^^^^^^^^^^^^^^^^
5045 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
5050 The '``alloca``' instruction allocates memory on the stack frame of the
5051 currently executing function, to be automatically released when this
5052 function returns to its caller. The object is always allocated in the
5053 generic address space (address space zero).
5058 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
5059 bytes of memory on the runtime stack, returning a pointer of the
5060 appropriate type to the program. If "NumElements" is specified, it is
5061 the number of elements allocated, otherwise "NumElements" is defaulted
5062 to be one. If a constant alignment is specified, the value result of the
5063 allocation is guaranteed to be aligned to at least that boundary. The
5064 alignment may not be greater than ``1 << 29``. If not specified, or if
5065 zero, the target can choose to align the allocation on any convenient
5066 boundary compatible with the type.
5068 '``type``' may be any sized type.
5073 Memory is allocated; a pointer is returned. The operation is undefined
5074 if there is insufficient stack space for the allocation. '``alloca``'d
5075 memory is automatically released when the function returns. The
5076 '``alloca``' instruction is commonly used to represent automatic
5077 variables that must have an address available. When the function returns
5078 (either with the ``ret`` or ``resume`` instructions), the memory is
5079 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5080 The order in which memory is allocated (ie., which way the stack grows)
5086 .. code-block:: llvm
5088 %ptr = alloca i32 ; yields i32*:ptr
5089 %ptr = alloca i32, i32 4 ; yields i32*:ptr
5090 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
5091 %ptr = alloca i32, align 1024 ; yields i32*:ptr
5095 '``load``' Instruction
5096 ^^^^^^^^^^^^^^^^^^^^^^
5103 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
5104 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5105 !<index> = !{ i32 1 }
5110 The '``load``' instruction is used to read from memory.
5115 The argument to the ``load`` instruction specifies the memory address
5116 from which to load. The pointer must point to a :ref:`first
5117 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
5118 then the optimizer is not allowed to modify the number or order of
5119 execution of this ``load`` with other :ref:`volatile
5120 operations <volatile>`.
5122 If the ``load`` is marked as ``atomic``, it takes an extra
5123 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5124 ``release`` and ``acq_rel`` orderings are not valid on ``load``
5125 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5126 when they may see multiple atomic stores. The type of the pointee must
5127 be an integer type whose bit width is a power of two greater than or
5128 equal to eight and less than or equal to a target-specific size limit.
5129 ``align`` must be explicitly specified on atomic loads, and the load has
5130 undefined behavior if the alignment is not set to a value which is at
5131 least the size in bytes of the pointee. ``!nontemporal`` does not have
5132 any defined semantics for atomic loads.
5134 The optional constant ``align`` argument specifies the alignment of the
5135 operation (that is, the alignment of the memory address). A value of 0
5136 or an omitted ``align`` argument means that the operation has the ABI
5137 alignment for the target. It is the responsibility of the code emitter
5138 to ensure that the alignment information is correct. Overestimating the
5139 alignment results in undefined behavior. Underestimating the alignment
5140 may produce less efficient code. An alignment of 1 is always safe. The
5141 maximum possible alignment is ``1 << 29``.
5143 The optional ``!nontemporal`` metadata must reference a single
5144 metadata name ``<index>`` corresponding to a metadata node with one
5145 ``i32`` entry of value 1. The existence of the ``!nontemporal``
5146 metadata on the instruction tells the optimizer and code generator
5147 that this load is not expected to be reused in the cache. The code
5148 generator may select special instructions to save cache bandwidth, such
5149 as the ``MOVNT`` instruction on x86.
5151 The optional ``!invariant.load`` metadata must reference a single
5152 metadata name ``<index>`` corresponding to a metadata node with no
5153 entries. The existence of the ``!invariant.load`` metadata on the
5154 instruction tells the optimizer and code generator that this load
5155 address points to memory which does not change value during program
5156 execution. The optimizer may then move this load around, for example, by
5157 hoisting it out of loops using loop invariant code motion.
5162 The location of memory pointed to is loaded. If the value being loaded
5163 is of scalar type then the number of bytes read does not exceed the
5164 minimum number of bytes needed to hold all bits of the type. For
5165 example, loading an ``i24`` reads at most three bytes. When loading a
5166 value of a type like ``i20`` with a size that is not an integral number
5167 of bytes, the result is undefined if the value was not originally
5168 written using a store of the same type.
5173 .. code-block:: llvm
5175 %ptr = alloca i32 ; yields i32*:ptr
5176 store i32 3, i32* %ptr ; yields void
5177 %val = load i32* %ptr ; yields i32:val = i32 3
5181 '``store``' Instruction
5182 ^^^^^^^^^^^^^^^^^^^^^^^
5189 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
5190 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
5195 The '``store``' instruction is used to write to memory.
5200 There are two arguments to the ``store`` instruction: a value to store
5201 and an address at which to store it. The type of the ``<pointer>``
5202 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
5203 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
5204 then the optimizer is not allowed to modify the number or order of
5205 execution of this ``store`` with other :ref:`volatile
5206 operations <volatile>`.
5208 If the ``store`` is marked as ``atomic``, it takes an extra
5209 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5210 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
5211 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5212 when they may see multiple atomic stores. The type of the pointee must
5213 be an integer type whose bit width is a power of two greater than or
5214 equal to eight and less than or equal to a target-specific size limit.
5215 ``align`` must be explicitly specified on atomic stores, and the store
5216 has undefined behavior if the alignment is not set to a value which is
5217 at least the size in bytes of the pointee. ``!nontemporal`` does not
5218 have any defined semantics for atomic stores.
5220 The optional constant ``align`` argument specifies the alignment of the
5221 operation (that is, the alignment of the memory address). A value of 0
5222 or an omitted ``align`` argument means that the operation has the ABI
5223 alignment for the target. It is the responsibility of the code emitter
5224 to ensure that the alignment information is correct. Overestimating the
5225 alignment results in undefined behavior. Underestimating the
5226 alignment may produce less efficient code. An alignment of 1 is always
5227 safe. The maximum possible alignment is ``1 << 29``.
5229 The optional ``!nontemporal`` metadata must reference a single metadata
5230 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
5231 value 1. The existence of the ``!nontemporal`` metadata on the instruction
5232 tells the optimizer and code generator that this load is not expected to
5233 be reused in the cache. The code generator may select special
5234 instructions to save cache bandwidth, such as the MOVNT instruction on
5240 The contents of memory are updated to contain ``<value>`` at the
5241 location specified by the ``<pointer>`` operand. If ``<value>`` is
5242 of scalar type then the number of bytes written does not exceed the
5243 minimum number of bytes needed to hold all bits of the type. For
5244 example, storing an ``i24`` writes at most three bytes. When writing a
5245 value of a type like ``i20`` with a size that is not an integral number
5246 of bytes, it is unspecified what happens to the extra bits that do not
5247 belong to the type, but they will typically be overwritten.
5252 .. code-block:: llvm
5254 %ptr = alloca i32 ; yields i32*:ptr
5255 store i32 3, i32* %ptr ; yields void
5256 %val = load i32* %ptr ; yields i32:val = i32 3
5260 '``fence``' Instruction
5261 ^^^^^^^^^^^^^^^^^^^^^^^
5268 fence [singlethread] <ordering> ; yields void
5273 The '``fence``' instruction is used to introduce happens-before edges
5279 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5280 defines what *synchronizes-with* edges they add. They can only be given
5281 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5286 A fence A which has (at least) ``release`` ordering semantics
5287 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5288 semantics if and only if there exist atomic operations X and Y, both
5289 operating on some atomic object M, such that A is sequenced before X, X
5290 modifies M (either directly or through some side effect of a sequence
5291 headed by X), Y is sequenced before B, and Y observes M. This provides a
5292 *happens-before* dependency between A and B. Rather than an explicit
5293 ``fence``, one (but not both) of the atomic operations X or Y might
5294 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5295 still *synchronize-with* the explicit ``fence`` and establish the
5296 *happens-before* edge.
5298 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5299 ``acquire`` and ``release`` semantics specified above, participates in
5300 the global program order of other ``seq_cst`` operations and/or fences.
5302 The optional ":ref:`singlethread <singlethread>`" argument specifies
5303 that the fence only synchronizes with other fences in the same thread.
5304 (This is useful for interacting with signal handlers.)
5309 .. code-block:: llvm
5311 fence acquire ; yields void
5312 fence singlethread seq_cst ; yields void
5316 '``cmpxchg``' Instruction
5317 ^^^^^^^^^^^^^^^^^^^^^^^^^
5324 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
5329 The '``cmpxchg``' instruction is used to atomically modify memory. It
5330 loads a value in memory and compares it to a given value. If they are
5331 equal, it tries to store a new value into the memory.
5336 There are three arguments to the '``cmpxchg``' instruction: an address
5337 to operate on, a value to compare to the value currently be at that
5338 address, and a new value to place at that address if the compared values
5339 are equal. The type of '<cmp>' must be an integer type whose bit width
5340 is a power of two greater than or equal to eight and less than or equal
5341 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5342 type, and the type of '<pointer>' must be a pointer to that type. If the
5343 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5344 to modify the number or order of execution of this ``cmpxchg`` with
5345 other :ref:`volatile operations <volatile>`.
5347 The success and failure :ref:`ordering <ordering>` arguments specify how this
5348 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
5349 must be at least ``monotonic``, the ordering constraint on failure must be no
5350 stronger than that on success, and the failure ordering cannot be either
5351 ``release`` or ``acq_rel``.
5353 The optional "``singlethread``" argument declares that the ``cmpxchg``
5354 is only atomic with respect to code (usually signal handlers) running in
5355 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5356 respect to all other code in the system.
5358 The pointer passed into cmpxchg must have alignment greater than or
5359 equal to the size in memory of the operand.
5364 The contents of memory at the location specified by the '``<pointer>``' operand
5365 is read and compared to '``<cmp>``'; if the read value is the equal, the
5366 '``<new>``' is written. The original value at the location is returned, together
5367 with a flag indicating success (true) or failure (false).
5369 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
5370 permitted: the operation may not write ``<new>`` even if the comparison
5373 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
5374 if the value loaded equals ``cmp``.
5376 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5377 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5378 load with an ordering parameter determined the second ordering parameter.
5383 .. code-block:: llvm
5386 %orig = atomic load i32* %ptr unordered ; yields i32
5390 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5391 %squared = mul i32 %cmp, %cmp
5392 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
5393 %value_loaded = extractvalue { i32, i1 } %val_success, 0
5394 %success = extractvalue { i32, i1 } %val_success, 1
5395 br i1 %success, label %done, label %loop
5402 '``atomicrmw``' Instruction
5403 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5410 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
5415 The '``atomicrmw``' instruction is used to atomically modify memory.
5420 There are three arguments to the '``atomicrmw``' instruction: an
5421 operation to apply, an address whose value to modify, an argument to the
5422 operation. The operation must be one of the following keywords:
5436 The type of '<value>' must be an integer type whose bit width is a power
5437 of two greater than or equal to eight and less than or equal to a
5438 target-specific size limit. The type of the '``<pointer>``' operand must
5439 be a pointer to that type. If the ``atomicrmw`` is marked as
5440 ``volatile``, then the optimizer is not allowed to modify the number or
5441 order of execution of this ``atomicrmw`` with other :ref:`volatile
5442 operations <volatile>`.
5447 The contents of memory at the location specified by the '``<pointer>``'
5448 operand are atomically read, modified, and written back. The original
5449 value at the location is returned. The modification is specified by the
5452 - xchg: ``*ptr = val``
5453 - add: ``*ptr = *ptr + val``
5454 - sub: ``*ptr = *ptr - val``
5455 - and: ``*ptr = *ptr & val``
5456 - nand: ``*ptr = ~(*ptr & val)``
5457 - or: ``*ptr = *ptr | val``
5458 - xor: ``*ptr = *ptr ^ val``
5459 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5460 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5461 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5463 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5469 .. code-block:: llvm
5471 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
5473 .. _i_getelementptr:
5475 '``getelementptr``' Instruction
5476 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5483 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5484 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5485 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5490 The '``getelementptr``' instruction is used to get the address of a
5491 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5492 address calculation only and does not access memory.
5497 The first argument is always a pointer or a vector of pointers, and
5498 forms the basis of the calculation. The remaining arguments are indices
5499 that indicate which of the elements of the aggregate object are indexed.
5500 The interpretation of each index is dependent on the type being indexed
5501 into. The first index always indexes the pointer value given as the
5502 first argument, the second index indexes a value of the type pointed to
5503 (not necessarily the value directly pointed to, since the first index
5504 can be non-zero), etc. The first type indexed into must be a pointer
5505 value, subsequent types can be arrays, vectors, and structs. Note that
5506 subsequent types being indexed into can never be pointers, since that
5507 would require loading the pointer before continuing calculation.
5509 The type of each index argument depends on the type it is indexing into.
5510 When indexing into a (optionally packed) structure, only ``i32`` integer
5511 **constants** are allowed (when using a vector of indices they must all
5512 be the **same** ``i32`` integer constant). When indexing into an array,
5513 pointer or vector, integers of any width are allowed, and they are not
5514 required to be constant. These integers are treated as signed values
5517 For example, let's consider a C code fragment and how it gets compiled
5533 int *foo(struct ST *s) {
5534 return &s[1].Z.B[5][13];
5537 The LLVM code generated by Clang is:
5539 .. code-block:: llvm
5541 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5542 %struct.ST = type { i32, double, %struct.RT }
5544 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5546 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5553 In the example above, the first index is indexing into the
5554 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5555 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5556 indexes into the third element of the structure, yielding a
5557 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5558 structure. The third index indexes into the second element of the
5559 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5560 dimensions of the array are subscripted into, yielding an '``i32``'
5561 type. The '``getelementptr``' instruction returns a pointer to this
5562 element, thus computing a value of '``i32*``' type.
5564 Note that it is perfectly legal to index partially through a structure,
5565 returning a pointer to an inner element. Because of this, the LLVM code
5566 for the given testcase is equivalent to:
5568 .. code-block:: llvm
5570 define i32* @foo(%struct.ST* %s) {
5571 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5572 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5573 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5574 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5575 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5579 If the ``inbounds`` keyword is present, the result value of the
5580 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5581 pointer is not an *in bounds* address of an allocated object, or if any
5582 of the addresses that would be formed by successive addition of the
5583 offsets implied by the indices to the base address with infinitely
5584 precise signed arithmetic are not an *in bounds* address of that
5585 allocated object. The *in bounds* addresses for an allocated object are
5586 all the addresses that point into the object, plus the address one byte
5587 past the end. In cases where the base is a vector of pointers the
5588 ``inbounds`` keyword applies to each of the computations element-wise.
5590 If the ``inbounds`` keyword is not present, the offsets are added to the
5591 base address with silently-wrapping two's complement arithmetic. If the
5592 offsets have a different width from the pointer, they are sign-extended
5593 or truncated to the width of the pointer. The result value of the
5594 ``getelementptr`` may be outside the object pointed to by the base
5595 pointer. The result value may not necessarily be used to access memory
5596 though, even if it happens to point into allocated storage. See the
5597 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5600 The getelementptr instruction is often confusing. For some more insight
5601 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5606 .. code-block:: llvm
5608 ; yields [12 x i8]*:aptr
5609 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5611 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5613 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5615 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5617 In cases where the pointer argument is a vector of pointers, each index
5618 must be a vector with the same number of elements. For example:
5620 .. code-block:: llvm
5622 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5624 Conversion Operations
5625 ---------------------
5627 The instructions in this category are the conversion instructions
5628 (casting) which all take a single operand and a type. They perform
5629 various bit conversions on the operand.
5631 '``trunc .. to``' Instruction
5632 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5639 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5644 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5649 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5650 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5651 of the same number of integers. The bit size of the ``value`` must be
5652 larger than the bit size of the destination type, ``ty2``. Equal sized
5653 types are not allowed.
5658 The '``trunc``' instruction truncates the high order bits in ``value``
5659 and converts the remaining bits to ``ty2``. Since the source size must
5660 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5661 It will always truncate bits.
5666 .. code-block:: llvm
5668 %X = trunc i32 257 to i8 ; yields i8:1
5669 %Y = trunc i32 123 to i1 ; yields i1:true
5670 %Z = trunc i32 122 to i1 ; yields i1:false
5671 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5673 '``zext .. to``' Instruction
5674 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5681 <result> = zext <ty> <value> to <ty2> ; yields ty2
5686 The '``zext``' instruction zero extends its operand to type ``ty2``.
5691 The '``zext``' instruction takes a value to cast, and a type to cast it
5692 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5693 the same number of integers. The bit size of the ``value`` must be
5694 smaller than the bit size of the destination type, ``ty2``.
5699 The ``zext`` fills the high order bits of the ``value`` with zero bits
5700 until it reaches the size of the destination type, ``ty2``.
5702 When zero extending from i1, the result will always be either 0 or 1.
5707 .. code-block:: llvm
5709 %X = zext i32 257 to i64 ; yields i64:257
5710 %Y = zext i1 true to i32 ; yields i32:1
5711 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5713 '``sext .. to``' Instruction
5714 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5721 <result> = sext <ty> <value> to <ty2> ; yields ty2
5726 The '``sext``' sign extends ``value`` to the type ``ty2``.
5731 The '``sext``' instruction takes a value to cast, and a type to cast it
5732 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5733 the same number of integers. The bit size of the ``value`` must be
5734 smaller than the bit size of the destination type, ``ty2``.
5739 The '``sext``' instruction performs a sign extension by copying the sign
5740 bit (highest order bit) of the ``value`` until it reaches the bit size
5741 of the type ``ty2``.
5743 When sign extending from i1, the extension always results in -1 or 0.
5748 .. code-block:: llvm
5750 %X = sext i8 -1 to i16 ; yields i16 :65535
5751 %Y = sext i1 true to i32 ; yields i32:-1
5752 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5754 '``fptrunc .. to``' Instruction
5755 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5762 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5767 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5772 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5773 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5774 The size of ``value`` must be larger than the size of ``ty2``. This
5775 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5780 The '``fptrunc``' instruction truncates a ``value`` from a larger
5781 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5782 point <t_floating>` type. If the value cannot fit within the
5783 destination type, ``ty2``, then the results are undefined.
5788 .. code-block:: llvm
5790 %X = fptrunc double 123.0 to float ; yields float:123.0
5791 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5793 '``fpext .. to``' Instruction
5794 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5801 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5806 The '``fpext``' extends a floating point ``value`` to a larger floating
5812 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5813 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5814 to. The source type must be smaller than the destination type.
5819 The '``fpext``' instruction extends the ``value`` from a smaller
5820 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5821 point <t_floating>` type. The ``fpext`` cannot be used to make a
5822 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5823 *no-op cast* for a floating point cast.
5828 .. code-block:: llvm
5830 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5831 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5833 '``fptoui .. to``' Instruction
5834 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5841 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5846 The '``fptoui``' converts a floating point ``value`` to its unsigned
5847 integer equivalent of type ``ty2``.
5852 The '``fptoui``' instruction takes a value to cast, which must be a
5853 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5854 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5855 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5856 type with the same number of elements as ``ty``
5861 The '``fptoui``' instruction converts its :ref:`floating
5862 point <t_floating>` operand into the nearest (rounding towards zero)
5863 unsigned integer value. If the value cannot fit in ``ty2``, the results
5869 .. code-block:: llvm
5871 %X = fptoui double 123.0 to i32 ; yields i32:123
5872 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5873 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5875 '``fptosi .. to``' Instruction
5876 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5883 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5888 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5889 ``value`` to type ``ty2``.
5894 The '``fptosi``' instruction takes a value to cast, which must be a
5895 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5896 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5897 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5898 type with the same number of elements as ``ty``
5903 The '``fptosi``' instruction converts its :ref:`floating
5904 point <t_floating>` operand into the nearest (rounding towards zero)
5905 signed integer value. If the value cannot fit in ``ty2``, the results
5911 .. code-block:: llvm
5913 %X = fptosi double -123.0 to i32 ; yields i32:-123
5914 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5915 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5917 '``uitofp .. to``' Instruction
5918 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5925 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5930 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5931 and converts that value to the ``ty2`` type.
5936 The '``uitofp``' instruction takes a value to cast, which must be a
5937 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5938 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5939 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5940 type with the same number of elements as ``ty``
5945 The '``uitofp``' instruction interprets its operand as an unsigned
5946 integer quantity and converts it to the corresponding floating point
5947 value. If the value cannot fit in the floating point value, the results
5953 .. code-block:: llvm
5955 %X = uitofp i32 257 to float ; yields float:257.0
5956 %Y = uitofp i8 -1 to double ; yields double:255.0
5958 '``sitofp .. to``' Instruction
5959 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5966 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5971 The '``sitofp``' instruction regards ``value`` as a signed integer and
5972 converts that value to the ``ty2`` type.
5977 The '``sitofp``' instruction takes a value to cast, which must be a
5978 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5979 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5980 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5981 type with the same number of elements as ``ty``
5986 The '``sitofp``' instruction interprets its operand as a signed integer
5987 quantity and converts it to the corresponding floating point value. If
5988 the value cannot fit in the floating point value, the results are
5994 .. code-block:: llvm
5996 %X = sitofp i32 257 to float ; yields float:257.0
5997 %Y = sitofp i8 -1 to double ; yields double:-1.0
6001 '``ptrtoint .. to``' Instruction
6002 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6009 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
6014 The '``ptrtoint``' instruction converts the pointer or a vector of
6015 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
6020 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
6021 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
6022 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
6023 a vector of integers type.
6028 The '``ptrtoint``' instruction converts ``value`` to integer type
6029 ``ty2`` by interpreting the pointer value as an integer and either
6030 truncating or zero extending that value to the size of the integer type.
6031 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
6032 ``value`` is larger than ``ty2`` then a truncation is done. If they are
6033 the same size, then nothing is done (*no-op cast*) other than a type
6039 .. code-block:: llvm
6041 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
6042 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
6043 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
6047 '``inttoptr .. to``' Instruction
6048 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6055 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
6060 The '``inttoptr``' instruction converts an integer ``value`` to a
6061 pointer type, ``ty2``.
6066 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
6067 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
6073 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
6074 applying either a zero extension or a truncation depending on the size
6075 of the integer ``value``. If ``value`` is larger than the size of a
6076 pointer then a truncation is done. If ``value`` is smaller than the size
6077 of a pointer then a zero extension is done. If they are the same size,
6078 nothing is done (*no-op cast*).
6083 .. code-block:: llvm
6085 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
6086 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
6087 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
6088 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
6092 '``bitcast .. to``' Instruction
6093 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6100 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
6105 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
6111 The '``bitcast``' instruction takes a value to cast, which must be a
6112 non-aggregate first class value, and a type to cast it to, which must
6113 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
6114 bit sizes of ``value`` and the destination type, ``ty2``, must be
6115 identical. If the source type is a pointer, the destination type must
6116 also be a pointer of the same size. This instruction supports bitwise
6117 conversion of vectors to integers and to vectors of other types (as
6118 long as they have the same size).
6123 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
6124 is always a *no-op cast* because no bits change with this
6125 conversion. The conversion is done as if the ``value`` had been stored
6126 to memory and read back as type ``ty2``. Pointer (or vector of
6127 pointers) types may only be converted to other pointer (or vector of
6128 pointers) types with the same address space through this instruction.
6129 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
6130 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
6135 .. code-block:: llvm
6137 %X = bitcast i8 255 to i8 ; yields i8 :-1
6138 %Y = bitcast i32* %x to sint* ; yields sint*:%x
6139 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
6140 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
6142 .. _i_addrspacecast:
6144 '``addrspacecast .. to``' Instruction
6145 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6152 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
6157 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
6158 address space ``n`` to type ``pty2`` in address space ``m``.
6163 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
6164 to cast and a pointer type to cast it to, which must have a different
6170 The '``addrspacecast``' instruction converts the pointer value
6171 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
6172 value modification, depending on the target and the address space
6173 pair. Pointer conversions within the same address space must be
6174 performed with the ``bitcast`` instruction. Note that if the address space
6175 conversion is legal then both result and operand refer to the same memory
6181 .. code-block:: llvm
6183 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
6184 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
6185 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
6192 The instructions in this category are the "miscellaneous" instructions,
6193 which defy better classification.
6197 '``icmp``' Instruction
6198 ^^^^^^^^^^^^^^^^^^^^^^
6205 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6210 The '``icmp``' instruction returns a boolean value or a vector of
6211 boolean values based on comparison of its two integer, integer vector,
6212 pointer, or pointer vector operands.
6217 The '``icmp``' instruction takes three operands. The first operand is
6218 the condition code indicating the kind of comparison to perform. It is
6219 not a value, just a keyword. The possible condition code are:
6222 #. ``ne``: not equal
6223 #. ``ugt``: unsigned greater than
6224 #. ``uge``: unsigned greater or equal
6225 #. ``ult``: unsigned less than
6226 #. ``ule``: unsigned less or equal
6227 #. ``sgt``: signed greater than
6228 #. ``sge``: signed greater or equal
6229 #. ``slt``: signed less than
6230 #. ``sle``: signed less or equal
6232 The remaining two arguments must be :ref:`integer <t_integer>` or
6233 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
6234 must also be identical types.
6239 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
6240 code given as ``cond``. The comparison performed always yields either an
6241 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
6243 #. ``eq``: yields ``true`` if the operands are equal, ``false``
6244 otherwise. No sign interpretation is necessary or performed.
6245 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
6246 otherwise. No sign interpretation is necessary or performed.
6247 #. ``ugt``: interprets the operands as unsigned values and yields
6248 ``true`` if ``op1`` is greater than ``op2``.
6249 #. ``uge``: interprets the operands as unsigned values and yields
6250 ``true`` if ``op1`` is greater than or equal to ``op2``.
6251 #. ``ult``: interprets the operands as unsigned values and yields
6252 ``true`` if ``op1`` is less than ``op2``.
6253 #. ``ule``: interprets the operands as unsigned values and yields
6254 ``true`` if ``op1`` is less than or equal to ``op2``.
6255 #. ``sgt``: interprets the operands as signed values and yields ``true``
6256 if ``op1`` is greater than ``op2``.
6257 #. ``sge``: interprets the operands as signed values and yields ``true``
6258 if ``op1`` is greater than or equal to ``op2``.
6259 #. ``slt``: interprets the operands as signed values and yields ``true``
6260 if ``op1`` is less than ``op2``.
6261 #. ``sle``: interprets the operands as signed values and yields ``true``
6262 if ``op1`` is less than or equal to ``op2``.
6264 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
6265 are compared as if they were integers.
6267 If the operands are integer vectors, then they are compared element by
6268 element. The result is an ``i1`` vector with the same number of elements
6269 as the values being compared. Otherwise, the result is an ``i1``.
6274 .. code-block:: llvm
6276 <result> = icmp eq i32 4, 5 ; yields: result=false
6277 <result> = icmp ne float* %X, %X ; yields: result=false
6278 <result> = icmp ult i16 4, 5 ; yields: result=true
6279 <result> = icmp sgt i16 4, 5 ; yields: result=false
6280 <result> = icmp ule i16 -4, 5 ; yields: result=false
6281 <result> = icmp sge i16 4, 5 ; yields: result=false
6283 Note that the code generator does not yet support vector types with the
6284 ``icmp`` instruction.
6288 '``fcmp``' Instruction
6289 ^^^^^^^^^^^^^^^^^^^^^^
6296 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6301 The '``fcmp``' instruction returns a boolean value or vector of boolean
6302 values based on comparison of its operands.
6304 If the operands are floating point scalars, then the result type is a
6305 boolean (:ref:`i1 <t_integer>`).
6307 If the operands are floating point vectors, then the result type is a
6308 vector of boolean with the same number of elements as the operands being
6314 The '``fcmp``' instruction takes three operands. The first operand is
6315 the condition code indicating the kind of comparison to perform. It is
6316 not a value, just a keyword. The possible condition code are:
6318 #. ``false``: no comparison, always returns false
6319 #. ``oeq``: ordered and equal
6320 #. ``ogt``: ordered and greater than
6321 #. ``oge``: ordered and greater than or equal
6322 #. ``olt``: ordered and less than
6323 #. ``ole``: ordered and less than or equal
6324 #. ``one``: ordered and not equal
6325 #. ``ord``: ordered (no nans)
6326 #. ``ueq``: unordered or equal
6327 #. ``ugt``: unordered or greater than
6328 #. ``uge``: unordered or greater than or equal
6329 #. ``ult``: unordered or less than
6330 #. ``ule``: unordered or less than or equal
6331 #. ``une``: unordered or not equal
6332 #. ``uno``: unordered (either nans)
6333 #. ``true``: no comparison, always returns true
6335 *Ordered* means that neither operand is a QNAN while *unordered* means
6336 that either operand may be a QNAN.
6338 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6339 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6340 type. They must have identical types.
6345 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6346 condition code given as ``cond``. If the operands are vectors, then the
6347 vectors are compared element by element. Each comparison performed
6348 always yields an :ref:`i1 <t_integer>` result, as follows:
6350 #. ``false``: always yields ``false``, regardless of operands.
6351 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6352 is equal to ``op2``.
6353 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6354 is greater than ``op2``.
6355 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6356 is greater than or equal to ``op2``.
6357 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6358 is less than ``op2``.
6359 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6360 is less than or equal to ``op2``.
6361 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6362 is not equal to ``op2``.
6363 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6364 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6366 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6367 greater than ``op2``.
6368 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6369 greater than or equal to ``op2``.
6370 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6372 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6373 less than or equal to ``op2``.
6374 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6375 not equal to ``op2``.
6376 #. ``uno``: yields ``true`` if either operand is a QNAN.
6377 #. ``true``: always yields ``true``, regardless of operands.
6382 .. code-block:: llvm
6384 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6385 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6386 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6387 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6389 Note that the code generator does not yet support vector types with the
6390 ``fcmp`` instruction.
6394 '``phi``' Instruction
6395 ^^^^^^^^^^^^^^^^^^^^^
6402 <result> = phi <ty> [ <val0>, <label0>], ...
6407 The '``phi``' instruction is used to implement the φ node in the SSA
6408 graph representing the function.
6413 The type of the incoming values is specified with the first type field.
6414 After this, the '``phi``' instruction takes a list of pairs as
6415 arguments, with one pair for each predecessor basic block of the current
6416 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6417 the value arguments to the PHI node. Only labels may be used as the
6420 There must be no non-phi instructions between the start of a basic block
6421 and the PHI instructions: i.e. PHI instructions must be first in a basic
6424 For the purposes of the SSA form, the use of each incoming value is
6425 deemed to occur on the edge from the corresponding predecessor block to
6426 the current block (but after any definition of an '``invoke``'
6427 instruction's return value on the same edge).
6432 At runtime, the '``phi``' instruction logically takes on the value
6433 specified by the pair corresponding to the predecessor basic block that
6434 executed just prior to the current block.
6439 .. code-block:: llvm
6441 Loop: ; Infinite loop that counts from 0 on up...
6442 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6443 %nextindvar = add i32 %indvar, 1
6448 '``select``' Instruction
6449 ^^^^^^^^^^^^^^^^^^^^^^^^
6456 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6458 selty is either i1 or {<N x i1>}
6463 The '``select``' instruction is used to choose one value based on a
6464 condition, without IR-level branching.
6469 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6470 values indicating the condition, and two values of the same :ref:`first
6471 class <t_firstclass>` type. If the val1/val2 are vectors and the
6472 condition is a scalar, then entire vectors are selected, not individual
6478 If the condition is an i1 and it evaluates to 1, the instruction returns
6479 the first value argument; otherwise, it returns the second value
6482 If the condition is a vector of i1, then the value arguments must be
6483 vectors of the same size, and the selection is done element by element.
6488 .. code-block:: llvm
6490 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6494 '``call``' Instruction
6495 ^^^^^^^^^^^^^^^^^^^^^^
6502 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6507 The '``call``' instruction represents a simple function call.
6512 This instruction requires several arguments:
6514 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
6515 should perform tail call optimization. The ``tail`` marker is a hint that
6516 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
6517 means that the call must be tail call optimized in order for the program to
6518 be correct. The ``musttail`` marker provides these guarantees:
6520 #. The call will not cause unbounded stack growth if it is part of a
6521 recursive cycle in the call graph.
6522 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
6525 Both markers imply that the callee does not access allocas or varargs from
6526 the caller. Calls marked ``musttail`` must obey the following additional
6529 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
6530 or a pointer bitcast followed by a ret instruction.
6531 - The ret instruction must return the (possibly bitcasted) value
6532 produced by the call or void.
6533 - The caller and callee prototypes must match. Pointer types of
6534 parameters or return types may differ in pointee type, but not
6536 - The calling conventions of the caller and callee must match.
6537 - All ABI-impacting function attributes, such as sret, byval, inreg,
6538 returned, and inalloca, must match.
6540 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
6541 the following conditions are met:
6543 - Caller and callee both have the calling convention ``fastcc``.
6544 - The call is in tail position (ret immediately follows call and ret
6545 uses value of call or is void).
6546 - Option ``-tailcallopt`` is enabled, or
6547 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6548 - `Platform-specific constraints are
6549 met. <CodeGenerator.html#tailcallopt>`_
6551 #. The optional "cconv" marker indicates which :ref:`calling
6552 convention <callingconv>` the call should use. If none is
6553 specified, the call defaults to using C calling conventions. The
6554 calling convention of the call must match the calling convention of
6555 the target function, or else the behavior is undefined.
6556 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6557 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6559 #. '``ty``': the type of the call instruction itself which is also the
6560 type of the return value. Functions that return no value are marked
6562 #. '``fnty``': shall be the signature of the pointer to function value
6563 being invoked. The argument types must match the types implied by
6564 this signature. This type can be omitted if the function is not
6565 varargs and if the function type does not return a pointer to a
6567 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6568 be invoked. In most cases, this is a direct function invocation, but
6569 indirect ``call``'s are just as possible, calling an arbitrary pointer
6571 #. '``function args``': argument list whose types match the function
6572 signature argument types and parameter attributes. All arguments must
6573 be of :ref:`first class <t_firstclass>` type. If the function signature
6574 indicates the function accepts a variable number of arguments, the
6575 extra arguments can be specified.
6576 #. The optional :ref:`function attributes <fnattrs>` list. Only
6577 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6578 attributes are valid here.
6583 The '``call``' instruction is used to cause control flow to transfer to
6584 a specified function, with its incoming arguments bound to the specified
6585 values. Upon a '``ret``' instruction in the called function, control
6586 flow continues with the instruction after the function call, and the
6587 return value of the function is bound to the result argument.
6592 .. code-block:: llvm
6594 %retval = call i32 @test(i32 %argc)
6595 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6596 %X = tail call i32 @foo() ; yields i32
6597 %Y = tail call fastcc i32 @foo() ; yields i32
6598 call void %foo(i8 97 signext)
6600 %struct.A = type { i32, i8 }
6601 %r = call %struct.A @foo() ; yields { i32, i8 }
6602 %gr = extractvalue %struct.A %r, 0 ; yields i32
6603 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6604 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6605 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6607 llvm treats calls to some functions with names and arguments that match
6608 the standard C99 library as being the C99 library functions, and may
6609 perform optimizations or generate code for them under that assumption.
6610 This is something we'd like to change in the future to provide better
6611 support for freestanding environments and non-C-based languages.
6615 '``va_arg``' Instruction
6616 ^^^^^^^^^^^^^^^^^^^^^^^^
6623 <resultval> = va_arg <va_list*> <arglist>, <argty>
6628 The '``va_arg``' instruction is used to access arguments passed through
6629 the "variable argument" area of a function call. It is used to implement
6630 the ``va_arg`` macro in C.
6635 This instruction takes a ``va_list*`` value and the type of the
6636 argument. It returns a value of the specified argument type and
6637 increments the ``va_list`` to point to the next argument. The actual
6638 type of ``va_list`` is target specific.
6643 The '``va_arg``' instruction loads an argument of the specified type
6644 from the specified ``va_list`` and causes the ``va_list`` to point to
6645 the next argument. For more information, see the variable argument
6646 handling :ref:`Intrinsic Functions <int_varargs>`.
6648 It is legal for this instruction to be called in a function which does
6649 not take a variable number of arguments, for example, the ``vfprintf``
6652 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6653 function <intrinsics>` because it takes a type as an argument.
6658 See the :ref:`variable argument processing <int_varargs>` section.
6660 Note that the code generator does not yet fully support va\_arg on many
6661 targets. Also, it does not currently support va\_arg with aggregate
6662 types on any target.
6666 '``landingpad``' Instruction
6667 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6674 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6675 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6677 <clause> := catch <type> <value>
6678 <clause> := filter <array constant type> <array constant>
6683 The '``landingpad``' instruction is used by `LLVM's exception handling
6684 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6685 is a landing pad --- one where the exception lands, and corresponds to the
6686 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6687 defines values supplied by the personality function (``pers_fn``) upon
6688 re-entry to the function. The ``resultval`` has the type ``resultty``.
6693 This instruction takes a ``pers_fn`` value. This is the personality
6694 function associated with the unwinding mechanism. The optional
6695 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6697 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6698 contains the global variable representing the "type" that may be caught
6699 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6700 clause takes an array constant as its argument. Use
6701 "``[0 x i8**] undef``" for a filter which cannot throw. The
6702 '``landingpad``' instruction must contain *at least* one ``clause`` or
6703 the ``cleanup`` flag.
6708 The '``landingpad``' instruction defines the values which are set by the
6709 personality function (``pers_fn``) upon re-entry to the function, and
6710 therefore the "result type" of the ``landingpad`` instruction. As with
6711 calling conventions, how the personality function results are
6712 represented in LLVM IR is target specific.
6714 The clauses are applied in order from top to bottom. If two
6715 ``landingpad`` instructions are merged together through inlining, the
6716 clauses from the calling function are appended to the list of clauses.
6717 When the call stack is being unwound due to an exception being thrown,
6718 the exception is compared against each ``clause`` in turn. If it doesn't
6719 match any of the clauses, and the ``cleanup`` flag is not set, then
6720 unwinding continues further up the call stack.
6722 The ``landingpad`` instruction has several restrictions:
6724 - A landing pad block is a basic block which is the unwind destination
6725 of an '``invoke``' instruction.
6726 - A landing pad block must have a '``landingpad``' instruction as its
6727 first non-PHI instruction.
6728 - There can be only one '``landingpad``' instruction within the landing
6730 - A basic block that is not a landing pad block may not include a
6731 '``landingpad``' instruction.
6732 - All '``landingpad``' instructions in a function must have the same
6733 personality function.
6738 .. code-block:: llvm
6740 ;; A landing pad which can catch an integer.
6741 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6743 ;; A landing pad that is a cleanup.
6744 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6746 ;; A landing pad which can catch an integer and can only throw a double.
6747 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6749 filter [1 x i8**] [@_ZTId]
6756 LLVM supports the notion of an "intrinsic function". These functions
6757 have well known names and semantics and are required to follow certain
6758 restrictions. Overall, these intrinsics represent an extension mechanism
6759 for the LLVM language that does not require changing all of the
6760 transformations in LLVM when adding to the language (or the bitcode
6761 reader/writer, the parser, etc...).
6763 Intrinsic function names must all start with an "``llvm.``" prefix. This
6764 prefix is reserved in LLVM for intrinsic names; thus, function names may
6765 not begin with this prefix. Intrinsic functions must always be external
6766 functions: you cannot define the body of intrinsic functions. Intrinsic
6767 functions may only be used in call or invoke instructions: it is illegal
6768 to take the address of an intrinsic function. Additionally, because
6769 intrinsic functions are part of the LLVM language, it is required if any
6770 are added that they be documented here.
6772 Some intrinsic functions can be overloaded, i.e., the intrinsic
6773 represents a family of functions that perform the same operation but on
6774 different data types. Because LLVM can represent over 8 million
6775 different integer types, overloading is used commonly to allow an
6776 intrinsic function to operate on any integer type. One or more of the
6777 argument types or the result type can be overloaded to accept any
6778 integer type. Argument types may also be defined as exactly matching a
6779 previous argument's type or the result type. This allows an intrinsic
6780 function which accepts multiple arguments, but needs all of them to be
6781 of the same type, to only be overloaded with respect to a single
6782 argument or the result.
6784 Overloaded intrinsics will have the names of its overloaded argument
6785 types encoded into its function name, each preceded by a period. Only
6786 those types which are overloaded result in a name suffix. Arguments
6787 whose type is matched against another type do not. For example, the
6788 ``llvm.ctpop`` function can take an integer of any width and returns an
6789 integer of exactly the same integer width. This leads to a family of
6790 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6791 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6792 overloaded, and only one type suffix is required. Because the argument's
6793 type is matched against the return type, it does not require its own
6796 To learn how to add an intrinsic function, please see the `Extending
6797 LLVM Guide <ExtendingLLVM.html>`_.
6801 Variable Argument Handling Intrinsics
6802 -------------------------------------
6804 Variable argument support is defined in LLVM with the
6805 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6806 functions. These functions are related to the similarly named macros
6807 defined in the ``<stdarg.h>`` header file.
6809 All of these functions operate on arguments that use a target-specific
6810 value type "``va_list``". The LLVM assembly language reference manual
6811 does not define what this type is, so all transformations should be
6812 prepared to handle these functions regardless of the type used.
6814 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6815 variable argument handling intrinsic functions are used.
6817 .. code-block:: llvm
6819 define i32 @test(i32 %X, ...) {
6820 ; Initialize variable argument processing
6822 %ap2 = bitcast i8** %ap to i8*
6823 call void @llvm.va_start(i8* %ap2)
6825 ; Read a single integer argument
6826 %tmp = va_arg i8** %ap, i32
6828 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6830 %aq2 = bitcast i8** %aq to i8*
6831 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6832 call void @llvm.va_end(i8* %aq2)
6834 ; Stop processing of arguments.
6835 call void @llvm.va_end(i8* %ap2)
6839 declare void @llvm.va_start(i8*)
6840 declare void @llvm.va_copy(i8*, i8*)
6841 declare void @llvm.va_end(i8*)
6845 '``llvm.va_start``' Intrinsic
6846 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6853 declare void @llvm.va_start(i8* <arglist>)
6858 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6859 subsequent use by ``va_arg``.
6864 The argument is a pointer to a ``va_list`` element to initialize.
6869 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6870 available in C. In a target-dependent way, it initializes the
6871 ``va_list`` element to which the argument points, so that the next call
6872 to ``va_arg`` will produce the first variable argument passed to the
6873 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6874 to know the last argument of the function as the compiler can figure
6877 '``llvm.va_end``' Intrinsic
6878 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6885 declare void @llvm.va_end(i8* <arglist>)
6890 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6891 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6896 The argument is a pointer to a ``va_list`` to destroy.
6901 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6902 available in C. In a target-dependent way, it destroys the ``va_list``
6903 element to which the argument points. Calls to
6904 :ref:`llvm.va_start <int_va_start>` and
6905 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6910 '``llvm.va_copy``' Intrinsic
6911 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6918 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6923 The '``llvm.va_copy``' intrinsic copies the current argument position
6924 from the source argument list to the destination argument list.
6929 The first argument is a pointer to a ``va_list`` element to initialize.
6930 The second argument is a pointer to a ``va_list`` element to copy from.
6935 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6936 available in C. In a target-dependent way, it copies the source
6937 ``va_list`` element into the destination ``va_list`` element. This
6938 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6939 arbitrarily complex and require, for example, memory allocation.
6941 Accurate Garbage Collection Intrinsics
6942 --------------------------------------
6944 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6945 (GC) requires the implementation and generation of these intrinsics.
6946 These intrinsics allow identification of :ref:`GC roots on the
6947 stack <int_gcroot>`, as well as garbage collector implementations that
6948 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6949 Front-ends for type-safe garbage collected languages should generate
6950 these intrinsics to make use of the LLVM garbage collectors. For more
6951 details, see `Accurate Garbage Collection with
6952 LLVM <GarbageCollection.html>`_.
6954 The garbage collection intrinsics only operate on objects in the generic
6955 address space (address space zero).
6959 '``llvm.gcroot``' Intrinsic
6960 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6967 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6972 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6973 the code generator, and allows some metadata to be associated with it.
6978 The first argument specifies the address of a stack object that contains
6979 the root pointer. The second pointer (which must be either a constant or
6980 a global value address) contains the meta-data to be associated with the
6986 At runtime, a call to this intrinsic stores a null pointer into the
6987 "ptrloc" location. At compile-time, the code generator generates
6988 information to allow the runtime to find the pointer at GC safe points.
6989 The '``llvm.gcroot``' intrinsic may only be used in a function which
6990 :ref:`specifies a GC algorithm <gc>`.
6994 '``llvm.gcread``' Intrinsic
6995 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7002 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
7007 The '``llvm.gcread``' intrinsic identifies reads of references from heap
7008 locations, allowing garbage collector implementations that require read
7014 The second argument is the address to read from, which should be an
7015 address allocated from the garbage collector. The first object is a
7016 pointer to the start of the referenced object, if needed by the language
7017 runtime (otherwise null).
7022 The '``llvm.gcread``' intrinsic has the same semantics as a load
7023 instruction, but may be replaced with substantially more complex code by
7024 the garbage collector runtime, as needed. The '``llvm.gcread``'
7025 intrinsic may only be used in a function which :ref:`specifies a GC
7030 '``llvm.gcwrite``' Intrinsic
7031 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7038 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
7043 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
7044 locations, allowing garbage collector implementations that require write
7045 barriers (such as generational or reference counting collectors).
7050 The first argument is the reference to store, the second is the start of
7051 the object to store it to, and the third is the address of the field of
7052 Obj to store to. If the runtime does not require a pointer to the
7053 object, Obj may be null.
7058 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
7059 instruction, but may be replaced with substantially more complex code by
7060 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
7061 intrinsic may only be used in a function which :ref:`specifies a GC
7064 Code Generator Intrinsics
7065 -------------------------
7067 These intrinsics are provided by LLVM to expose special features that
7068 may only be implemented with code generator support.
7070 '``llvm.returnaddress``' Intrinsic
7071 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7078 declare i8 *@llvm.returnaddress(i32 <level>)
7083 The '``llvm.returnaddress``' intrinsic attempts to compute a
7084 target-specific value indicating the return address of the current
7085 function or one of its callers.
7090 The argument to this intrinsic indicates which function to return the
7091 address for. Zero indicates the calling function, one indicates its
7092 caller, etc. The argument is **required** to be a constant integer
7098 The '``llvm.returnaddress``' intrinsic either returns a pointer
7099 indicating the return address of the specified call frame, or zero if it
7100 cannot be identified. The value returned by this intrinsic is likely to
7101 be incorrect or 0 for arguments other than zero, so it should only be
7102 used for debugging purposes.
7104 Note that calling this intrinsic does not prevent function inlining or
7105 other aggressive transformations, so the value returned may not be that
7106 of the obvious source-language caller.
7108 '``llvm.frameaddress``' Intrinsic
7109 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7116 declare i8* @llvm.frameaddress(i32 <level>)
7121 The '``llvm.frameaddress``' intrinsic attempts to return the
7122 target-specific frame pointer value for the specified stack frame.
7127 The argument to this intrinsic indicates which function to return the
7128 frame pointer for. Zero indicates the calling function, one indicates
7129 its caller, etc. The argument is **required** to be a constant integer
7135 The '``llvm.frameaddress``' intrinsic either returns a pointer
7136 indicating the frame address of the specified call frame, or zero if it
7137 cannot be identified. The value returned by this intrinsic is likely to
7138 be incorrect or 0 for arguments other than zero, so it should only be
7139 used for debugging purposes.
7141 Note that calling this intrinsic does not prevent function inlining or
7142 other aggressive transformations, so the value returned may not be that
7143 of the obvious source-language caller.
7145 .. _int_read_register:
7146 .. _int_write_register:
7148 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
7149 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7156 declare i32 @llvm.read_register.i32(metadata)
7157 declare i64 @llvm.read_register.i64(metadata)
7158 declare void @llvm.write_register.i32(metadata, i32 @value)
7159 declare void @llvm.write_register.i64(metadata, i64 @value)
7160 !0 = metadata !{metadata !"sp\00"}
7165 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
7166 provides access to the named register. The register must be valid on
7167 the architecture being compiled to. The type needs to be compatible
7168 with the register being read.
7173 The '``llvm.read_register``' intrinsic returns the current value of the
7174 register, where possible. The '``llvm.write_register``' intrinsic sets
7175 the current value of the register, where possible.
7177 This is useful to implement named register global variables that need
7178 to always be mapped to a specific register, as is common practice on
7179 bare-metal programs including OS kernels.
7181 The compiler doesn't check for register availability or use of the used
7182 register in surrounding code, including inline assembly. Because of that,
7183 allocatable registers are not supported.
7185 Warning: So far it only works with the stack pointer on selected
7186 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
7187 work is needed to support other registers and even more so, allocatable
7192 '``llvm.stacksave``' Intrinsic
7193 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7200 declare i8* @llvm.stacksave()
7205 The '``llvm.stacksave``' intrinsic is used to remember the current state
7206 of the function stack, for use with
7207 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
7208 implementing language features like scoped automatic variable sized
7214 This intrinsic returns a opaque pointer value that can be passed to
7215 :ref:`llvm.stackrestore <int_stackrestore>`. When an
7216 ``llvm.stackrestore`` intrinsic is executed with a value saved from
7217 ``llvm.stacksave``, it effectively restores the state of the stack to
7218 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
7219 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
7220 were allocated after the ``llvm.stacksave`` was executed.
7222 .. _int_stackrestore:
7224 '``llvm.stackrestore``' Intrinsic
7225 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7232 declare void @llvm.stackrestore(i8* %ptr)
7237 The '``llvm.stackrestore``' intrinsic is used to restore the state of
7238 the function stack to the state it was in when the corresponding
7239 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
7240 useful for implementing language features like scoped automatic variable
7241 sized arrays in C99.
7246 See the description for :ref:`llvm.stacksave <int_stacksave>`.
7248 '``llvm.prefetch``' Intrinsic
7249 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7256 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7261 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
7262 insert a prefetch instruction if supported; otherwise, it is a noop.
7263 Prefetches have no effect on the behavior of the program but can change
7264 its performance characteristics.
7269 ``address`` is the address to be prefetched, ``rw`` is the specifier
7270 determining if the fetch should be for a read (0) or write (1), and
7271 ``locality`` is a temporal locality specifier ranging from (0) - no
7272 locality, to (3) - extremely local keep in cache. The ``cache type``
7273 specifies whether the prefetch is performed on the data (1) or
7274 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
7275 arguments must be constant integers.
7280 This intrinsic does not modify the behavior of the program. In
7281 particular, prefetches cannot trap and do not produce a value. On
7282 targets that support this intrinsic, the prefetch can provide hints to
7283 the processor cache for better performance.
7285 '``llvm.pcmarker``' Intrinsic
7286 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7293 declare void @llvm.pcmarker(i32 <id>)
7298 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7299 Counter (PC) in a region of code to simulators and other tools. The
7300 method is target specific, but it is expected that the marker will use
7301 exported symbols to transmit the PC of the marker. The marker makes no
7302 guarantees that it will remain with any specific instruction after
7303 optimizations. It is possible that the presence of a marker will inhibit
7304 optimizations. The intended use is to be inserted after optimizations to
7305 allow correlations of simulation runs.
7310 ``id`` is a numerical id identifying the marker.
7315 This intrinsic does not modify the behavior of the program. Backends
7316 that do not support this intrinsic may ignore it.
7318 '``llvm.readcyclecounter``' Intrinsic
7319 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7326 declare i64 @llvm.readcyclecounter()
7331 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
7332 counter register (or similar low latency, high accuracy clocks) on those
7333 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7334 should map to RPCC. As the backing counters overflow quickly (on the
7335 order of 9 seconds on alpha), this should only be used for small
7341 When directly supported, reading the cycle counter should not modify any
7342 memory. Implementations are allowed to either return a application
7343 specific value or a system wide value. On backends without support, this
7344 is lowered to a constant 0.
7346 Note that runtime support may be conditional on the privilege-level code is
7347 running at and the host platform.
7349 '``llvm.clear_cache``' Intrinsic
7350 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7357 declare void @llvm.clear_cache(i8*, i8*)
7362 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
7363 in the specified range to the execution unit of the processor. On
7364 targets with non-unified instruction and data cache, the implementation
7365 flushes the instruction cache.
7370 On platforms with coherent instruction and data caches (e.g. x86), this
7371 intrinsic is a nop. On platforms with non-coherent instruction and data
7372 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
7373 instructions or a system call, if cache flushing requires special
7376 The default behavior is to emit a call to ``__clear_cache`` from the run
7379 This instrinsic does *not* empty the instruction pipeline. Modifications
7380 of the current function are outside the scope of the intrinsic.
7382 Standard C Library Intrinsics
7383 -----------------------------
7385 LLVM provides intrinsics for a few important standard C library
7386 functions. These intrinsics allow source-language front-ends to pass
7387 information about the alignment of the pointer arguments to the code
7388 generator, providing opportunity for more efficient code generation.
7392 '``llvm.memcpy``' Intrinsic
7393 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7398 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
7399 integer bit width and for different address spaces. Not all targets
7400 support all bit widths however.
7404 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7405 i32 <len>, i32 <align>, i1 <isvolatile>)
7406 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7407 i64 <len>, i32 <align>, i1 <isvolatile>)
7412 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7413 source location to the destination location.
7415 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
7416 intrinsics do not return a value, takes extra alignment/isvolatile
7417 arguments and the pointers can be in specified address spaces.
7422 The first argument is a pointer to the destination, the second is a
7423 pointer to the source. The third argument is an integer argument
7424 specifying the number of bytes to copy, the fourth argument is the
7425 alignment of the source and destination locations, and the fifth is a
7426 boolean indicating a volatile access.
7428 If the call to this intrinsic has an alignment value that is not 0 or 1,
7429 then the caller guarantees that both the source and destination pointers
7430 are aligned to that boundary.
7432 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
7433 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7434 very cleanly specified and it is unwise to depend on it.
7439 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7440 source location to the destination location, which are not allowed to
7441 overlap. It copies "len" bytes of memory over. If the argument is known
7442 to be aligned to some boundary, this can be specified as the fourth
7443 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
7445 '``llvm.memmove``' Intrinsic
7446 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7451 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7452 bit width and for different address space. Not all targets support all
7457 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7458 i32 <len>, i32 <align>, i1 <isvolatile>)
7459 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7460 i64 <len>, i32 <align>, i1 <isvolatile>)
7465 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7466 source location to the destination location. It is similar to the
7467 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7470 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7471 intrinsics do not return a value, takes extra alignment/isvolatile
7472 arguments and the pointers can be in specified address spaces.
7477 The first argument is a pointer to the destination, the second is a
7478 pointer to the source. The third argument is an integer argument
7479 specifying the number of bytes to copy, the fourth argument is the
7480 alignment of the source and destination locations, and the fifth is a
7481 boolean indicating a volatile access.
7483 If the call to this intrinsic has an alignment value that is not 0 or 1,
7484 then the caller guarantees that the source and destination pointers are
7485 aligned to that boundary.
7487 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7488 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7489 not very cleanly specified and it is unwise to depend on it.
7494 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7495 source location to the destination location, which may overlap. It
7496 copies "len" bytes of memory over. If the argument is known to be
7497 aligned to some boundary, this can be specified as the fourth argument,
7498 otherwise it should be set to 0 or 1 (both meaning no alignment).
7500 '``llvm.memset.*``' Intrinsics
7501 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7506 This is an overloaded intrinsic. You can use llvm.memset on any integer
7507 bit width and for different address spaces. However, not all targets
7508 support all bit widths.
7512 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7513 i32 <len>, i32 <align>, i1 <isvolatile>)
7514 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7515 i64 <len>, i32 <align>, i1 <isvolatile>)
7520 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7521 particular byte value.
7523 Note that, unlike the standard libc function, the ``llvm.memset``
7524 intrinsic does not return a value and takes extra alignment/volatile
7525 arguments. Also, the destination can be in an arbitrary address space.
7530 The first argument is a pointer to the destination to fill, the second
7531 is the byte value with which to fill it, the third argument is an
7532 integer argument specifying the number of bytes to fill, and the fourth
7533 argument is the known alignment of the destination location.
7535 If the call to this intrinsic has an alignment value that is not 0 or 1,
7536 then the caller guarantees that the destination pointer is aligned to
7539 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7540 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7541 very cleanly specified and it is unwise to depend on it.
7546 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7547 at the destination location. If the argument is known to be aligned to
7548 some boundary, this can be specified as the fourth argument, otherwise
7549 it should be set to 0 or 1 (both meaning no alignment).
7551 '``llvm.sqrt.*``' Intrinsic
7552 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7557 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7558 floating point or vector of floating point type. Not all targets support
7563 declare float @llvm.sqrt.f32(float %Val)
7564 declare double @llvm.sqrt.f64(double %Val)
7565 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7566 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7567 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7572 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7573 returning the same value as the libm '``sqrt``' functions would. Unlike
7574 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7575 negative numbers other than -0.0 (which allows for better optimization,
7576 because there is no need to worry about errno being set).
7577 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7582 The argument and return value are floating point numbers of the same
7588 This function returns the sqrt of the specified operand if it is a
7589 nonnegative floating point number.
7591 '``llvm.powi.*``' Intrinsic
7592 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7597 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7598 floating point or vector of floating point type. Not all targets support
7603 declare float @llvm.powi.f32(float %Val, i32 %power)
7604 declare double @llvm.powi.f64(double %Val, i32 %power)
7605 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7606 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7607 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7612 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7613 specified (positive or negative) power. The order of evaluation of
7614 multiplications is not defined. When a vector of floating point type is
7615 used, the second argument remains a scalar integer value.
7620 The second argument is an integer power, and the first is a value to
7621 raise to that power.
7626 This function returns the first value raised to the second power with an
7627 unspecified sequence of rounding operations.
7629 '``llvm.sin.*``' Intrinsic
7630 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7635 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7636 floating point or vector of floating point type. Not all targets support
7641 declare float @llvm.sin.f32(float %Val)
7642 declare double @llvm.sin.f64(double %Val)
7643 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7644 declare fp128 @llvm.sin.f128(fp128 %Val)
7645 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7650 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7655 The argument and return value are floating point numbers of the same
7661 This function returns the sine of the specified operand, returning the
7662 same values as the libm ``sin`` functions would, and handles error
7663 conditions in the same way.
7665 '``llvm.cos.*``' Intrinsic
7666 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7671 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7672 floating point or vector of floating point type. Not all targets support
7677 declare float @llvm.cos.f32(float %Val)
7678 declare double @llvm.cos.f64(double %Val)
7679 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7680 declare fp128 @llvm.cos.f128(fp128 %Val)
7681 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7686 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7691 The argument and return value are floating point numbers of the same
7697 This function returns the cosine of the specified operand, returning the
7698 same values as the libm ``cos`` functions would, and handles error
7699 conditions in the same way.
7701 '``llvm.pow.*``' Intrinsic
7702 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7707 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7708 floating point or vector of floating point type. Not all targets support
7713 declare float @llvm.pow.f32(float %Val, float %Power)
7714 declare double @llvm.pow.f64(double %Val, double %Power)
7715 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7716 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7717 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7722 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7723 specified (positive or negative) power.
7728 The second argument is a floating point power, and the first is a value
7729 to raise to that power.
7734 This function returns the first value raised to the second power,
7735 returning the same values as the libm ``pow`` functions would, and
7736 handles error conditions in the same way.
7738 '``llvm.exp.*``' Intrinsic
7739 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7744 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7745 floating point or vector of floating point type. Not all targets support
7750 declare float @llvm.exp.f32(float %Val)
7751 declare double @llvm.exp.f64(double %Val)
7752 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7753 declare fp128 @llvm.exp.f128(fp128 %Val)
7754 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7759 The '``llvm.exp.*``' intrinsics perform the exp function.
7764 The argument and return value are floating point numbers of the same
7770 This function returns the same values as the libm ``exp`` functions
7771 would, and handles error conditions in the same way.
7773 '``llvm.exp2.*``' Intrinsic
7774 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7779 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7780 floating point or vector of floating point type. Not all targets support
7785 declare float @llvm.exp2.f32(float %Val)
7786 declare double @llvm.exp2.f64(double %Val)
7787 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7788 declare fp128 @llvm.exp2.f128(fp128 %Val)
7789 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7794 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7799 The argument and return value are floating point numbers of the same
7805 This function returns the same values as the libm ``exp2`` functions
7806 would, and handles error conditions in the same way.
7808 '``llvm.log.*``' Intrinsic
7809 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7814 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7815 floating point or vector of floating point type. Not all targets support
7820 declare float @llvm.log.f32(float %Val)
7821 declare double @llvm.log.f64(double %Val)
7822 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7823 declare fp128 @llvm.log.f128(fp128 %Val)
7824 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7829 The '``llvm.log.*``' intrinsics perform the log function.
7834 The argument and return value are floating point numbers of the same
7840 This function returns the same values as the libm ``log`` functions
7841 would, and handles error conditions in the same way.
7843 '``llvm.log10.*``' Intrinsic
7844 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7849 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7850 floating point or vector of floating point type. Not all targets support
7855 declare float @llvm.log10.f32(float %Val)
7856 declare double @llvm.log10.f64(double %Val)
7857 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7858 declare fp128 @llvm.log10.f128(fp128 %Val)
7859 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7864 The '``llvm.log10.*``' intrinsics perform the log10 function.
7869 The argument and return value are floating point numbers of the same
7875 This function returns the same values as the libm ``log10`` functions
7876 would, and handles error conditions in the same way.
7878 '``llvm.log2.*``' Intrinsic
7879 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7884 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7885 floating point or vector of floating point type. Not all targets support
7890 declare float @llvm.log2.f32(float %Val)
7891 declare double @llvm.log2.f64(double %Val)
7892 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7893 declare fp128 @llvm.log2.f128(fp128 %Val)
7894 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7899 The '``llvm.log2.*``' intrinsics perform the log2 function.
7904 The argument and return value are floating point numbers of the same
7910 This function returns the same values as the libm ``log2`` functions
7911 would, and handles error conditions in the same way.
7913 '``llvm.fma.*``' Intrinsic
7914 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7919 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7920 floating point or vector of floating point type. Not all targets support
7925 declare float @llvm.fma.f32(float %a, float %b, float %c)
7926 declare double @llvm.fma.f64(double %a, double %b, double %c)
7927 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7928 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7929 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7934 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7940 The argument and return value are floating point numbers of the same
7946 This function returns the same values as the libm ``fma`` functions
7947 would, and does not set errno.
7949 '``llvm.fabs.*``' Intrinsic
7950 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7955 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7956 floating point or vector of floating point type. Not all targets support
7961 declare float @llvm.fabs.f32(float %Val)
7962 declare double @llvm.fabs.f64(double %Val)
7963 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7964 declare fp128 @llvm.fabs.f128(fp128 %Val)
7965 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7970 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7976 The argument and return value are floating point numbers of the same
7982 This function returns the same values as the libm ``fabs`` functions
7983 would, and handles error conditions in the same way.
7985 '``llvm.copysign.*``' Intrinsic
7986 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7991 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7992 floating point or vector of floating point type. Not all targets support
7997 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7998 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7999 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
8000 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
8001 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
8006 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
8007 first operand and the sign of the second operand.
8012 The arguments and return value are floating point numbers of the same
8018 This function returns the same values as the libm ``copysign``
8019 functions would, and handles error conditions in the same way.
8021 '``llvm.floor.*``' Intrinsic
8022 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8027 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
8028 floating point or vector of floating point type. Not all targets support
8033 declare float @llvm.floor.f32(float %Val)
8034 declare double @llvm.floor.f64(double %Val)
8035 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
8036 declare fp128 @llvm.floor.f128(fp128 %Val)
8037 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
8042 The '``llvm.floor.*``' intrinsics return the floor of the operand.
8047 The argument and return value are floating point numbers of the same
8053 This function returns the same values as the libm ``floor`` functions
8054 would, and handles error conditions in the same way.
8056 '``llvm.ceil.*``' Intrinsic
8057 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8062 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
8063 floating point or vector of floating point type. Not all targets support
8068 declare float @llvm.ceil.f32(float %Val)
8069 declare double @llvm.ceil.f64(double %Val)
8070 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
8071 declare fp128 @llvm.ceil.f128(fp128 %Val)
8072 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
8077 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
8082 The argument and return value are floating point numbers of the same
8088 This function returns the same values as the libm ``ceil`` functions
8089 would, and handles error conditions in the same way.
8091 '``llvm.trunc.*``' Intrinsic
8092 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8097 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
8098 floating point or vector of floating point type. Not all targets support
8103 declare float @llvm.trunc.f32(float %Val)
8104 declare double @llvm.trunc.f64(double %Val)
8105 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
8106 declare fp128 @llvm.trunc.f128(fp128 %Val)
8107 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
8112 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
8113 nearest integer not larger in magnitude than the operand.
8118 The argument and return value are floating point numbers of the same
8124 This function returns the same values as the libm ``trunc`` functions
8125 would, and handles error conditions in the same way.
8127 '``llvm.rint.*``' Intrinsic
8128 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8133 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
8134 floating point or vector of floating point type. Not all targets support
8139 declare float @llvm.rint.f32(float %Val)
8140 declare double @llvm.rint.f64(double %Val)
8141 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
8142 declare fp128 @llvm.rint.f128(fp128 %Val)
8143 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
8148 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
8149 nearest integer. It may raise an inexact floating-point exception if the
8150 operand isn't an integer.
8155 The argument and return value are floating point numbers of the same
8161 This function returns the same values as the libm ``rint`` functions
8162 would, and handles error conditions in the same way.
8164 '``llvm.nearbyint.*``' Intrinsic
8165 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8170 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
8171 floating point or vector of floating point type. Not all targets support
8176 declare float @llvm.nearbyint.f32(float %Val)
8177 declare double @llvm.nearbyint.f64(double %Val)
8178 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
8179 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
8180 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
8185 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
8191 The argument and return value are floating point numbers of the same
8197 This function returns the same values as the libm ``nearbyint``
8198 functions would, and handles error conditions in the same way.
8200 '``llvm.round.*``' Intrinsic
8201 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8206 This is an overloaded intrinsic. You can use ``llvm.round`` on any
8207 floating point or vector of floating point type. Not all targets support
8212 declare float @llvm.round.f32(float %Val)
8213 declare double @llvm.round.f64(double %Val)
8214 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
8215 declare fp128 @llvm.round.f128(fp128 %Val)
8216 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
8221 The '``llvm.round.*``' intrinsics returns the operand rounded to the
8227 The argument and return value are floating point numbers of the same
8233 This function returns the same values as the libm ``round``
8234 functions would, and handles error conditions in the same way.
8236 Bit Manipulation Intrinsics
8237 ---------------------------
8239 LLVM provides intrinsics for a few important bit manipulation
8240 operations. These allow efficient code generation for some algorithms.
8242 '``llvm.bswap.*``' Intrinsics
8243 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8248 This is an overloaded intrinsic function. You can use bswap on any
8249 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
8253 declare i16 @llvm.bswap.i16(i16 <id>)
8254 declare i32 @llvm.bswap.i32(i32 <id>)
8255 declare i64 @llvm.bswap.i64(i64 <id>)
8260 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
8261 values with an even number of bytes (positive multiple of 16 bits).
8262 These are useful for performing operations on data that is not in the
8263 target's native byte order.
8268 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
8269 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
8270 intrinsic returns an i32 value that has the four bytes of the input i32
8271 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
8272 returned i32 will have its bytes in 3, 2, 1, 0 order. The
8273 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
8274 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
8277 '``llvm.ctpop.*``' Intrinsic
8278 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8283 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
8284 bit width, or on any vector with integer elements. Not all targets
8285 support all bit widths or vector types, however.
8289 declare i8 @llvm.ctpop.i8(i8 <src>)
8290 declare i16 @llvm.ctpop.i16(i16 <src>)
8291 declare i32 @llvm.ctpop.i32(i32 <src>)
8292 declare i64 @llvm.ctpop.i64(i64 <src>)
8293 declare i256 @llvm.ctpop.i256(i256 <src>)
8294 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
8299 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
8305 The only argument is the value to be counted. The argument may be of any
8306 integer type, or a vector with integer elements. The return type must
8307 match the argument type.
8312 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
8313 each element of a vector.
8315 '``llvm.ctlz.*``' Intrinsic
8316 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8321 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
8322 integer bit width, or any vector whose elements are integers. Not all
8323 targets support all bit widths or vector types, however.
8327 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
8328 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
8329 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
8330 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
8331 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
8332 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8337 The '``llvm.ctlz``' family of intrinsic functions counts the number of
8338 leading zeros in a variable.
8343 The first argument is the value to be counted. This argument may be of
8344 any integer type, or a vectory with integer element type. The return
8345 type must match the first argument type.
8347 The second argument must be a constant and is a flag to indicate whether
8348 the intrinsic should ensure that a zero as the first argument produces a
8349 defined result. Historically some architectures did not provide a
8350 defined result for zero values as efficiently, and many algorithms are
8351 now predicated on avoiding zero-value inputs.
8356 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
8357 zeros in a variable, or within each element of the vector. If
8358 ``src == 0`` then the result is the size in bits of the type of ``src``
8359 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8360 ``llvm.ctlz(i32 2) = 30``.
8362 '``llvm.cttz.*``' Intrinsic
8363 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8368 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
8369 integer bit width, or any vector of integer elements. Not all targets
8370 support all bit widths or vector types, however.
8374 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
8375 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
8376 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
8377 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
8378 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
8379 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8384 The '``llvm.cttz``' family of intrinsic functions counts the number of
8390 The first argument is the value to be counted. This argument may be of
8391 any integer type, or a vectory with integer element type. The return
8392 type must match the first argument type.
8394 The second argument must be a constant and is a flag to indicate whether
8395 the intrinsic should ensure that a zero as the first argument produces a
8396 defined result. Historically some architectures did not provide a
8397 defined result for zero values as efficiently, and many algorithms are
8398 now predicated on avoiding zero-value inputs.
8403 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
8404 zeros in a variable, or within each element of a vector. If ``src == 0``
8405 then the result is the size in bits of the type of ``src`` if
8406 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8407 ``llvm.cttz(2) = 1``.
8409 Arithmetic with Overflow Intrinsics
8410 -----------------------------------
8412 LLVM provides intrinsics for some arithmetic with overflow operations.
8414 '``llvm.sadd.with.overflow.*``' Intrinsics
8415 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8420 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
8421 on any integer bit width.
8425 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8426 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8427 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8432 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8433 a signed addition of the two arguments, and indicate whether an overflow
8434 occurred during the signed summation.
8439 The arguments (%a and %b) and the first element of the result structure
8440 may be of integer types of any bit width, but they must have the same
8441 bit width. The second element of the result structure must be of type
8442 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8448 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8449 a signed addition of the two variables. They return a structure --- the
8450 first element of which is the signed summation, and the second element
8451 of which is a bit specifying if the signed summation resulted in an
8457 .. code-block:: llvm
8459 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8460 %sum = extractvalue {i32, i1} %res, 0
8461 %obit = extractvalue {i32, i1} %res, 1
8462 br i1 %obit, label %overflow, label %normal
8464 '``llvm.uadd.with.overflow.*``' Intrinsics
8465 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8470 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8471 on any integer bit width.
8475 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8476 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8477 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8482 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8483 an unsigned addition of the two arguments, and indicate whether a carry
8484 occurred during the unsigned summation.
8489 The arguments (%a and %b) and the first element of the result structure
8490 may be of integer types of any bit width, but they must have the same
8491 bit width. The second element of the result structure must be of type
8492 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8498 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8499 an unsigned addition of the two arguments. They return a structure --- the
8500 first element of which is the sum, and the second element of which is a
8501 bit specifying if the unsigned summation resulted in a carry.
8506 .. code-block:: llvm
8508 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8509 %sum = extractvalue {i32, i1} %res, 0
8510 %obit = extractvalue {i32, i1} %res, 1
8511 br i1 %obit, label %carry, label %normal
8513 '``llvm.ssub.with.overflow.*``' Intrinsics
8514 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8519 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8520 on any integer bit width.
8524 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8525 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8526 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8531 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8532 a signed subtraction of the two arguments, and indicate whether an
8533 overflow occurred during the signed subtraction.
8538 The arguments (%a and %b) and the first element of the result structure
8539 may be of integer types of any bit width, but they must have the same
8540 bit width. The second element of the result structure must be of type
8541 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8547 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8548 a signed subtraction of the two arguments. They return a structure --- the
8549 first element of which is the subtraction, and the second element of
8550 which is a bit specifying if the signed subtraction resulted in an
8556 .. code-block:: llvm
8558 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8559 %sum = extractvalue {i32, i1} %res, 0
8560 %obit = extractvalue {i32, i1} %res, 1
8561 br i1 %obit, label %overflow, label %normal
8563 '``llvm.usub.with.overflow.*``' Intrinsics
8564 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8569 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8570 on any integer bit width.
8574 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8575 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8576 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8581 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8582 an unsigned subtraction of the two arguments, and indicate whether an
8583 overflow occurred during the unsigned subtraction.
8588 The arguments (%a and %b) and the first element of the result structure
8589 may be of integer types of any bit width, but they must have the same
8590 bit width. The second element of the result structure must be of type
8591 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8597 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8598 an unsigned subtraction of the two arguments. They return a structure ---
8599 the first element of which is the subtraction, and the second element of
8600 which is a bit specifying if the unsigned subtraction resulted in an
8606 .. code-block:: llvm
8608 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8609 %sum = extractvalue {i32, i1} %res, 0
8610 %obit = extractvalue {i32, i1} %res, 1
8611 br i1 %obit, label %overflow, label %normal
8613 '``llvm.smul.with.overflow.*``' Intrinsics
8614 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8619 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8620 on any integer bit width.
8624 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8625 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8626 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8631 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8632 a signed multiplication of the two arguments, and indicate whether an
8633 overflow occurred during the signed multiplication.
8638 The arguments (%a and %b) and the first element of the result structure
8639 may be of integer types of any bit width, but they must have the same
8640 bit width. The second element of the result structure must be of type
8641 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8647 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8648 a signed multiplication of the two arguments. They return a structure ---
8649 the first element of which is the multiplication, and the second element
8650 of which is a bit specifying if the signed multiplication resulted in an
8656 .. code-block:: llvm
8658 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8659 %sum = extractvalue {i32, i1} %res, 0
8660 %obit = extractvalue {i32, i1} %res, 1
8661 br i1 %obit, label %overflow, label %normal
8663 '``llvm.umul.with.overflow.*``' Intrinsics
8664 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8669 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8670 on any integer bit width.
8674 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8675 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8676 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8681 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8682 a unsigned multiplication of the two arguments, and indicate whether an
8683 overflow occurred during the unsigned multiplication.
8688 The arguments (%a and %b) and the first element of the result structure
8689 may be of integer types of any bit width, but they must have the same
8690 bit width. The second element of the result structure must be of type
8691 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8697 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8698 an unsigned multiplication of the two arguments. They return a structure ---
8699 the first element of which is the multiplication, and the second
8700 element of which is a bit specifying if the unsigned multiplication
8701 resulted in an overflow.
8706 .. code-block:: llvm
8708 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8709 %sum = extractvalue {i32, i1} %res, 0
8710 %obit = extractvalue {i32, i1} %res, 1
8711 br i1 %obit, label %overflow, label %normal
8713 Specialised Arithmetic Intrinsics
8714 ---------------------------------
8716 '``llvm.fmuladd.*``' Intrinsic
8717 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8724 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8725 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8730 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8731 expressions that can be fused if the code generator determines that (a) the
8732 target instruction set has support for a fused operation, and (b) that the
8733 fused operation is more efficient than the equivalent, separate pair of mul
8734 and add instructions.
8739 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8740 multiplicands, a and b, and an addend c.
8749 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8751 is equivalent to the expression a \* b + c, except that rounding will
8752 not be performed between the multiplication and addition steps if the
8753 code generator fuses the operations. Fusion is not guaranteed, even if
8754 the target platform supports it. If a fused multiply-add is required the
8755 corresponding llvm.fma.\* intrinsic function should be used
8756 instead. This never sets errno, just as '``llvm.fma.*``'.
8761 .. code-block:: llvm
8763 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
8765 Half Precision Floating Point Intrinsics
8766 ----------------------------------------
8768 For most target platforms, half precision floating point is a
8769 storage-only format. This means that it is a dense encoding (in memory)
8770 but does not support computation in the format.
8772 This means that code must first load the half-precision floating point
8773 value as an i16, then convert it to float with
8774 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8775 then be performed on the float value (including extending to double
8776 etc). To store the value back to memory, it is first converted to float
8777 if needed, then converted to i16 with
8778 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8781 .. _int_convert_to_fp16:
8783 '``llvm.convert.to.fp16``' Intrinsic
8784 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8791 declare i16 @llvm.convert.to.fp16.f32(float %a)
8792 declare i16 @llvm.convert.to.fp16.f64(double %a)
8797 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
8798 conventional floating point type to half precision floating point format.
8803 The intrinsic function contains single argument - the value to be
8809 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
8810 conventional floating point format to half precision floating point format. The
8811 return value is an ``i16`` which contains the converted number.
8816 .. code-block:: llvm
8818 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
8819 store i16 %res, i16* @x, align 2
8821 .. _int_convert_from_fp16:
8823 '``llvm.convert.from.fp16``' Intrinsic
8824 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8831 declare float @llvm.convert.from.fp16.f32(i16 %a)
8832 declare double @llvm.convert.from.fp16.f64(i16 %a)
8837 The '``llvm.convert.from.fp16``' intrinsic function performs a
8838 conversion from half precision floating point format to single precision
8839 floating point format.
8844 The intrinsic function contains single argument - the value to be
8850 The '``llvm.convert.from.fp16``' intrinsic function performs a
8851 conversion from half single precision floating point format to single
8852 precision floating point format. The input half-float value is
8853 represented by an ``i16`` value.
8858 .. code-block:: llvm
8860 %a = load i16* @x, align 2
8861 %res = call float @llvm.convert.from.fp16(i16 %a)
8866 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8867 prefix), are described in the `LLVM Source Level
8868 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8871 Exception Handling Intrinsics
8872 -----------------------------
8874 The LLVM exception handling intrinsics (which all start with
8875 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8876 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8880 Trampoline Intrinsics
8881 ---------------------
8883 These intrinsics make it possible to excise one parameter, marked with
8884 the :ref:`nest <nest>` attribute, from a function. The result is a
8885 callable function pointer lacking the nest parameter - the caller does
8886 not need to provide a value for it. Instead, the value to use is stored
8887 in advance in a "trampoline", a block of memory usually allocated on the
8888 stack, which also contains code to splice the nest value into the
8889 argument list. This is used to implement the GCC nested function address
8892 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8893 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8894 It can be created as follows:
8896 .. code-block:: llvm
8898 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8899 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8900 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8901 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8902 %fp = bitcast i8* %p to i32 (i32, i32)*
8904 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8905 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8909 '``llvm.init.trampoline``' Intrinsic
8910 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8917 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8922 This fills the memory pointed to by ``tramp`` with executable code,
8923 turning it into a trampoline.
8928 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8929 pointers. The ``tramp`` argument must point to a sufficiently large and
8930 sufficiently aligned block of memory; this memory is written to by the
8931 intrinsic. Note that the size and the alignment are target-specific -
8932 LLVM currently provides no portable way of determining them, so a
8933 front-end that generates this intrinsic needs to have some
8934 target-specific knowledge. The ``func`` argument must hold a function
8935 bitcast to an ``i8*``.
8940 The block of memory pointed to by ``tramp`` is filled with target
8941 dependent code, turning it into a function. Then ``tramp`` needs to be
8942 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8943 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8944 function's signature is the same as that of ``func`` with any arguments
8945 marked with the ``nest`` attribute removed. At most one such ``nest``
8946 argument is allowed, and it must be of pointer type. Calling the new
8947 function is equivalent to calling ``func`` with the same argument list,
8948 but with ``nval`` used for the missing ``nest`` argument. If, after
8949 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8950 modified, then the effect of any later call to the returned function
8951 pointer is undefined.
8955 '``llvm.adjust.trampoline``' Intrinsic
8956 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8963 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8968 This performs any required machine-specific adjustment to the address of
8969 a trampoline (passed as ``tramp``).
8974 ``tramp`` must point to a block of memory which already has trampoline
8975 code filled in by a previous call to
8976 :ref:`llvm.init.trampoline <int_it>`.
8981 On some architectures the address of the code to be executed needs to be
8982 different than the address where the trampoline is actually stored. This
8983 intrinsic returns the executable address corresponding to ``tramp``
8984 after performing the required machine specific adjustments. The pointer
8985 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8990 This class of intrinsics provides information about the lifetime of
8991 memory objects and ranges where variables are immutable.
8995 '``llvm.lifetime.start``' Intrinsic
8996 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9003 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
9008 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
9014 The first argument is a constant integer representing the size of the
9015 object, or -1 if it is variable sized. The second argument is a pointer
9021 This intrinsic indicates that before this point in the code, the value
9022 of the memory pointed to by ``ptr`` is dead. This means that it is known
9023 to never be used and has an undefined value. A load from the pointer
9024 that precedes this intrinsic can be replaced with ``'undef'``.
9028 '``llvm.lifetime.end``' Intrinsic
9029 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9036 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
9041 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
9047 The first argument is a constant integer representing the size of the
9048 object, or -1 if it is variable sized. The second argument is a pointer
9054 This intrinsic indicates that after this point in the code, the value of
9055 the memory pointed to by ``ptr`` is dead. This means that it is known to
9056 never be used and has an undefined value. Any stores into the memory
9057 object following this intrinsic may be removed as dead.
9059 '``llvm.invariant.start``' Intrinsic
9060 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9067 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
9072 The '``llvm.invariant.start``' intrinsic specifies that the contents of
9073 a memory object will not change.
9078 The first argument is a constant integer representing the size of the
9079 object, or -1 if it is variable sized. The second argument is a pointer
9085 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
9086 the return value, the referenced memory location is constant and
9089 '``llvm.invariant.end``' Intrinsic
9090 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9097 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
9102 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
9103 memory object are mutable.
9108 The first argument is the matching ``llvm.invariant.start`` intrinsic.
9109 The second argument is a constant integer representing the size of the
9110 object, or -1 if it is variable sized and the third argument is a
9111 pointer to the object.
9116 This intrinsic indicates that the memory is mutable again.
9121 This class of intrinsics is designed to be generic and has no specific
9124 '``llvm.var.annotation``' Intrinsic
9125 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9132 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9137 The '``llvm.var.annotation``' intrinsic.
9142 The first argument is a pointer to a value, the second is a pointer to a
9143 global string, the third is a pointer to a global string which is the
9144 source file name, and the last argument is the line number.
9149 This intrinsic allows annotation of local variables with arbitrary
9150 strings. This can be useful for special purpose optimizations that want
9151 to look for these annotations. These have no other defined use; they are
9152 ignored by code generation and optimization.
9154 '``llvm.ptr.annotation.*``' Intrinsic
9155 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9160 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
9161 pointer to an integer of any width. *NOTE* you must specify an address space for
9162 the pointer. The identifier for the default address space is the integer
9167 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9168 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
9169 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
9170 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
9171 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
9176 The '``llvm.ptr.annotation``' intrinsic.
9181 The first argument is a pointer to an integer value of arbitrary bitwidth
9182 (result of some expression), the second is a pointer to a global string, the
9183 third is a pointer to a global string which is the source file name, and the
9184 last argument is the line number. It returns the value of the first argument.
9189 This intrinsic allows annotation of a pointer to an integer with arbitrary
9190 strings. This can be useful for special purpose optimizations that want to look
9191 for these annotations. These have no other defined use; they are ignored by code
9192 generation and optimization.
9194 '``llvm.annotation.*``' Intrinsic
9195 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9200 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
9201 any integer bit width.
9205 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
9206 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
9207 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
9208 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
9209 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
9214 The '``llvm.annotation``' intrinsic.
9219 The first argument is an integer value (result of some expression), the
9220 second is a pointer to a global string, the third is a pointer to a
9221 global string which is the source file name, and the last argument is
9222 the line number. It returns the value of the first argument.
9227 This intrinsic allows annotations to be put on arbitrary expressions
9228 with arbitrary strings. This can be useful for special purpose
9229 optimizations that want to look for these annotations. These have no
9230 other defined use; they are ignored by code generation and optimization.
9232 '``llvm.trap``' Intrinsic
9233 ^^^^^^^^^^^^^^^^^^^^^^^^^
9240 declare void @llvm.trap() noreturn nounwind
9245 The '``llvm.trap``' intrinsic.
9255 This intrinsic is lowered to the target dependent trap instruction. If
9256 the target does not have a trap instruction, this intrinsic will be
9257 lowered to a call of the ``abort()`` function.
9259 '``llvm.debugtrap``' Intrinsic
9260 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9267 declare void @llvm.debugtrap() nounwind
9272 The '``llvm.debugtrap``' intrinsic.
9282 This intrinsic is lowered to code which is intended to cause an
9283 execution trap with the intention of requesting the attention of a
9286 '``llvm.stackprotector``' Intrinsic
9287 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9294 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
9299 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
9300 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
9301 is placed on the stack before local variables.
9306 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
9307 The first argument is the value loaded from the stack guard
9308 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
9309 enough space to hold the value of the guard.
9314 This intrinsic causes the prologue/epilogue inserter to force the position of
9315 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
9316 to ensure that if a local variable on the stack is overwritten, it will destroy
9317 the value of the guard. When the function exits, the guard on the stack is
9318 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
9319 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
9320 calling the ``__stack_chk_fail()`` function.
9322 '``llvm.stackprotectorcheck``' Intrinsic
9323 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9330 declare void @llvm.stackprotectorcheck(i8** <guard>)
9335 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
9336 created stack protector and if they are not equal calls the
9337 ``__stack_chk_fail()`` function.
9342 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
9343 the variable ``@__stack_chk_guard``.
9348 This intrinsic is provided to perform the stack protector check by comparing
9349 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
9350 values do not match call the ``__stack_chk_fail()`` function.
9352 The reason to provide this as an IR level intrinsic instead of implementing it
9353 via other IR operations is that in order to perform this operation at the IR
9354 level without an intrinsic, one would need to create additional basic blocks to
9355 handle the success/failure cases. This makes it difficult to stop the stack
9356 protector check from disrupting sibling tail calls in Codegen. With this
9357 intrinsic, we are able to generate the stack protector basic blocks late in
9358 codegen after the tail call decision has occurred.
9360 '``llvm.objectsize``' Intrinsic
9361 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9368 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
9369 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
9374 The ``llvm.objectsize`` intrinsic is designed to provide information to
9375 the optimizers to determine at compile time whether a) an operation
9376 (like memcpy) will overflow a buffer that corresponds to an object, or
9377 b) that a runtime check for overflow isn't necessary. An object in this
9378 context means an allocation of a specific class, structure, array, or
9384 The ``llvm.objectsize`` intrinsic takes two arguments. The first
9385 argument is a pointer to or into the ``object``. The second argument is
9386 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
9387 or -1 (if false) when the object size is unknown. The second argument
9388 only accepts constants.
9393 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
9394 the size of the object concerned. If the size cannot be determined at
9395 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
9396 on the ``min`` argument).
9398 '``llvm.expect``' Intrinsic
9399 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9404 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
9409 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
9410 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9411 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9416 The ``llvm.expect`` intrinsic provides information about expected (the
9417 most probable) value of ``val``, which can be used by optimizers.
9422 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
9423 a value. The second argument is an expected value, this needs to be a
9424 constant value, variables are not allowed.
9429 This intrinsic is lowered to the ``val``.
9431 '``llvm.assume``' Intrinsic
9432 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9439 declare void @llvm.assume(i1 %cond)
9444 The ``llvm.assume`` allows the optimizer to assume that the provided
9445 condition is true. This information can then be used in simplifying other parts
9451 The condition which the optimizer may assume is always true.
9456 The intrinsic allows the optimizer to assume that the provided condition is
9457 always true whenever the control flow reaches the intrinsic call. No code is
9458 generated for this intrinsic, and instructions that contribute only to the
9459 provided condition are not used for code generation. If the condition is
9460 violated during execution, the behavior is undefined.
9462 Please note that optimizer might limit the transformations performed on values
9463 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
9464 only used to form the intrinsic's input argument. This might prove undesirable
9465 if the extra information provided by the ``llvm.assume`` intrinsic does cause
9466 sufficient overall improvement in code quality. For this reason,
9467 ``llvm.assume`` should not be used to document basic mathematical invariants
9468 that the optimizer can otherwise deduce or facts that are of little use to the
9471 '``llvm.donothing``' Intrinsic
9472 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9479 declare void @llvm.donothing() nounwind readnone
9484 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
9485 only intrinsic that can be called with an invoke instruction.
9495 This intrinsic does nothing, and it's removed by optimizers and ignored
9498 Stack Map Intrinsics
9499 --------------------
9501 LLVM provides experimental intrinsics to support runtime patching
9502 mechanisms commonly desired in dynamic language JITs. These intrinsics
9503 are described in :doc:`StackMaps`.