1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][a-zA-Z$._][a-zA-Z$._0-9]*``'. Identifiers which require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves.
83 #. Unnamed values are represented as an unsigned numeric value with
84 their prefix. For example, ``%12``, ``@2``, ``%44``.
85 #. Constants, which are described in the section Constants_ below.
87 LLVM requires that values start with a prefix for two reasons: Compilers
88 don't need to worry about name clashes with reserved words, and the set
89 of reserved words may be expanded in the future without penalty.
90 Additionally, unnamed identifiers allow a compiler to quickly come up
91 with a temporary variable without having to avoid symbol table
94 Reserved words in LLVM are very similar to reserved words in other
95 languages. There are keywords for different opcodes ('``add``',
96 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
97 '``i32``', etc...), and others. These reserved words cannot conflict
98 with variable names, because none of them start with a prefix character
101 Here is an example of LLVM code to multiply the integer variable
108 %result = mul i32 %X, 8
110 After strength reduction:
114 %result = shl i32 %X, 3
120 %0 = add i32 %X, %X ; yields {i32}:%0
121 %1 = add i32 %0, %0 ; yields {i32}:%1
122 %result = add i32 %1, %1
124 This last way of multiplying ``%X`` by 8 illustrates several important
125 lexical features of LLVM:
127 #. Comments are delimited with a '``;``' and go until the end of line.
128 #. Unnamed temporaries are created when the result of a computation is
129 not assigned to a named value.
130 #. Unnamed temporaries are numbered sequentially (using a per-function
131 incrementing counter, starting with 0).
133 It also shows a convention that we follow in this document. When
134 demonstrating instructions, we will follow an instruction with a comment
135 that defines the type and name of value produced.
143 LLVM programs are composed of ``Module``'s, each of which is a
144 translation unit of the input programs. Each module consists of
145 functions, global variables, and symbol table entries. Modules may be
146 combined together with the LLVM linker, which merges function (and
147 global variable) definitions, resolves forward declarations, and merges
148 symbol table entries. Here is an example of the "hello world" module:
152 ; Declare the string constant as a global constant.
153 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
155 ; External declaration of the puts function
156 declare i32 @puts(i8* nocapture) nounwind
158 ; Definition of main function
159 define i32 @main() { ; i32()*
160 ; Convert [13 x i8]* to i8 *...
161 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
163 ; Call puts function to write out the string to stdout.
164 call i32 @puts(i8* %cast210)
169 !1 = metadata !{i32 42}
172 This example is made up of a :ref:`global variable <globalvars>` named
173 "``.str``", an external declaration of the "``puts``" function, a
174 :ref:`function definition <functionstructure>` for "``main``" and
175 :ref:`named metadata <namedmetadatastructure>` "``foo``".
177 In general, a module is made up of a list of global values (where both
178 functions and global variables are global values). Global values are
179 represented by a pointer to a memory location (in this case, a pointer
180 to an array of char, and a pointer to a function), and have one of the
181 following :ref:`linkage types <linkage>`.
188 All Global Variables and Functions have one of the following types of
192 Global values with "``private``" linkage are only directly
193 accessible by objects in the current module. In particular, linking
194 code into a module with an private global value may cause the
195 private to be renamed as necessary to avoid collisions. Because the
196 symbol is private to the module, all references can be updated. This
197 doesn't show up in any symbol table in the object file.
199 Similar to ``private``, but the symbol is passed through the
200 assembler and evaluated by the linker. Unlike normal strong symbols,
201 they are removed by the linker from the final linked image
202 (executable or dynamic library).
203 ``linker_private_weak``
204 Similar to "``linker_private``", but the symbol is weak. Note that
205 ``linker_private_weak`` symbols are subject to coalescing by the
206 linker. The symbols are removed by the linker from the final linked
207 image (executable or dynamic library).
209 Similar to private, but the value shows as a local symbol
210 (``STB_LOCAL`` in the case of ELF) in the object file. This
211 corresponds to the notion of the '``static``' keyword in C.
212 ``available_externally``
213 Globals with "``available_externally``" linkage are never emitted
214 into the object file corresponding to the LLVM module. They exist to
215 allow inlining and other optimizations to take place given knowledge
216 of the definition of the global, which is known to be somewhere
217 outside the module. Globals with ``available_externally`` linkage
218 are allowed to be discarded at will, and are otherwise the same as
219 ``linkonce_odr``. This linkage type is only allowed on definitions,
222 Globals with "``linkonce``" linkage are merged with other globals of
223 the same name when linkage occurs. This can be used to implement
224 some forms of inline functions, templates, or other code which must
225 be generated in each translation unit that uses it, but where the
226 body may be overridden with a more definitive definition later.
227 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
228 that ``linkonce`` linkage does not actually allow the optimizer to
229 inline the body of this function into callers because it doesn't
230 know if this definition of the function is the definitive definition
231 within the program or whether it will be overridden by a stronger
232 definition. To enable inlining and other optimizations, use
233 "``linkonce_odr``" linkage.
235 "``weak``" linkage has the same merging semantics as ``linkonce``
236 linkage, except that unreferenced globals with ``weak`` linkage may
237 not be discarded. This is used for globals that are declared "weak"
240 "``common``" linkage is most similar to "``weak``" linkage, but they
241 are used for tentative definitions in C, such as "``int X;``" at
242 global scope. Symbols with "``common``" linkage are merged in the
243 same way as ``weak symbols``, and they may not be deleted if
244 unreferenced. ``common`` symbols may not have an explicit section,
245 must have a zero initializer, and may not be marked
246 ':ref:`constant <globalvars>`'. Functions and aliases may not have
249 .. _linkage_appending:
252 "``appending``" linkage may only be applied to global variables of
253 pointer to array type. When two global variables with appending
254 linkage are linked together, the two global arrays are appended
255 together. This is the LLVM, typesafe, equivalent of having the
256 system linker append together "sections" with identical names when
259 The semantics of this linkage follow the ELF object file model: the
260 symbol is weak until linked, if not linked, the symbol becomes null
261 instead of being an undefined reference.
262 ``linkonce_odr``, ``weak_odr``
263 Some languages allow differing globals to be merged, such as two
264 functions with different semantics. Other languages, such as
265 ``C++``, ensure that only equivalent globals are ever merged (the
266 "one definition rule" --- "ODR"). Such languages can use the
267 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
268 global will only be merged with equivalent globals. These linkage
269 types are otherwise the same as their non-``odr`` versions.
270 ``linkonce_odr_auto_hide``
271 Similar to "``linkonce_odr``", but nothing in the translation unit
272 takes the address of this definition. For instance, functions that
273 had an inline definition, but the compiler decided not to inline it.
274 ``linkonce_odr_auto_hide`` may have only ``default`` visibility. The
275 symbols are removed by the linker from the final linked image
276 (executable or dynamic library).
278 If none of the above identifiers are used, the global is externally
279 visible, meaning that it participates in linkage and can be used to
280 resolve external symbol references.
282 The next two types of linkage are targeted for Microsoft Windows
283 platform only. They are designed to support importing (exporting)
284 symbols from (to) DLLs (Dynamic Link Libraries).
287 "``dllimport``" linkage causes the compiler to reference a function
288 or variable via a global pointer to a pointer that is set up by the
289 DLL exporting the symbol. On Microsoft Windows targets, the pointer
290 name is formed by combining ``__imp_`` and the function or variable
293 "``dllexport``" linkage causes the compiler to provide a global
294 pointer to a pointer in a DLL, so that it can be referenced with the
295 ``dllimport`` attribute. On Microsoft Windows targets, the pointer
296 name is formed by combining ``__imp_`` and the function or variable
299 For example, since the "``.LC0``" variable is defined to be internal, if
300 another module defined a "``.LC0``" variable and was linked with this
301 one, one of the two would be renamed, preventing a collision. Since
302 "``main``" and "``puts``" are external (i.e., lacking any linkage
303 declarations), they are accessible outside of the current module.
305 It is illegal for a function *declaration* to have any linkage type
306 other than ``external``, ``dllimport`` or ``extern_weak``.
308 Aliases can have only ``external``, ``internal``, ``weak`` or
309 ``weak_odr`` linkages.
316 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
317 :ref:`invokes <i_invoke>` can all have an optional calling convention
318 specified for the call. The calling convention of any pair of dynamic
319 caller/callee must match, or the behavior of the program is undefined.
320 The following calling conventions are supported by LLVM, and more may be
323 "``ccc``" - The C calling convention
324 This calling convention (the default if no other calling convention
325 is specified) matches the target C calling conventions. This calling
326 convention supports varargs function calls and tolerates some
327 mismatch in the declared prototype and implemented declaration of
328 the function (as does normal C).
329 "``fastcc``" - The fast calling convention
330 This calling convention attempts to make calls as fast as possible
331 (e.g. by passing things in registers). This calling convention
332 allows the target to use whatever tricks it wants to produce fast
333 code for the target, without having to conform to an externally
334 specified ABI (Application Binary Interface). `Tail calls can only
335 be optimized when this, the GHC or the HiPE convention is
336 used. <CodeGenerator.html#id80>`_ This calling convention does not
337 support varargs and requires the prototype of all callees to exactly
338 match the prototype of the function definition.
339 "``coldcc``" - The cold calling convention
340 This calling convention attempts to make code in the caller as
341 efficient as possible under the assumption that the call is not
342 commonly executed. As such, these calls often preserve all registers
343 so that the call does not break any live ranges in the caller side.
344 This calling convention does not support varargs and requires the
345 prototype of all callees to exactly match the prototype of the
347 "``cc 10``" - GHC convention
348 This calling convention has been implemented specifically for use by
349 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
350 It passes everything in registers, going to extremes to achieve this
351 by disabling callee save registers. This calling convention should
352 not be used lightly but only for specific situations such as an
353 alternative to the *register pinning* performance technique often
354 used when implementing functional programming languages. At the
355 moment only X86 supports this convention and it has the following
358 - On *X86-32* only supports up to 4 bit type parameters. No
359 floating point types are supported.
360 - On *X86-64* only supports up to 10 bit type parameters and 6
361 floating point parameters.
363 This calling convention supports `tail call
364 optimization <CodeGenerator.html#id80>`_ but requires both the
365 caller and callee are using it.
366 "``cc 11``" - The HiPE calling convention
367 This calling convention has been implemented specifically for use by
368 the `High-Performance Erlang
369 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
370 native code compiler of the `Ericsson's Open Source Erlang/OTP
371 system <http://www.erlang.org/download.shtml>`_. It uses more
372 registers for argument passing than the ordinary C calling
373 convention and defines no callee-saved registers. The calling
374 convention properly supports `tail call
375 optimization <CodeGenerator.html#id80>`_ but requires that both the
376 caller and the callee use it. It uses a *register pinning*
377 mechanism, similar to GHC's convention, for keeping frequently
378 accessed runtime components pinned to specific hardware registers.
379 At the moment only X86 supports this convention (both 32 and 64
381 "``cc <n>``" - Numbered convention
382 Any calling convention may be specified by number, allowing
383 target-specific calling conventions to be used. Target specific
384 calling conventions start at 64.
386 More calling conventions can be added/defined on an as-needed basis, to
387 support Pascal conventions or any other well-known target-independent
390 .. _visibilitystyles:
395 All Global Variables and Functions have one of the following visibility
398 "``default``" - Default style
399 On targets that use the ELF object file format, default visibility
400 means that the declaration is visible to other modules and, in
401 shared libraries, means that the declared entity may be overridden.
402 On Darwin, default visibility means that the declaration is visible
403 to other modules. Default visibility corresponds to "external
404 linkage" in the language.
405 "``hidden``" - Hidden style
406 Two declarations of an object with hidden visibility refer to the
407 same object if they are in the same shared object. Usually, hidden
408 visibility indicates that the symbol will not be placed into the
409 dynamic symbol table, so no other module (executable or shared
410 library) can reference it directly.
411 "``protected``" - Protected style
412 On ELF, protected visibility indicates that the symbol will be
413 placed in the dynamic symbol table, but that references within the
414 defining module will bind to the local symbol. That is, the symbol
415 cannot be overridden by another module.
422 LLVM IR allows you to specify name aliases for certain types. This can
423 make it easier to read the IR and make the IR more condensed
424 (particularly when recursive types are involved). An example of a name
429 %mytype = type { %mytype*, i32 }
431 You may give a name to any :ref:`type <typesystem>` except
432 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
433 expected with the syntax "%mytype".
435 Note that type names are aliases for the structural type that they
436 indicate, and that you can therefore specify multiple names for the same
437 type. This often leads to confusing behavior when dumping out a .ll
438 file. Since LLVM IR uses structural typing, the name is not part of the
439 type. When printing out LLVM IR, the printer will pick *one name* to
440 render all types of a particular shape. This means that if you have code
441 where two different source types end up having the same LLVM type, that
442 the dumper will sometimes print the "wrong" or unexpected type. This is
443 an important design point and isn't going to change.
450 Global variables define regions of memory allocated at compilation time
451 instead of run-time. Global variables may optionally be initialized, may
452 have an explicit section to be placed in, and may have an optional
453 explicit alignment specified.
455 A variable may be defined as ``thread_local``, which means that it will
456 not be shared by threads (each thread will have a separated copy of the
457 variable). Not all targets support thread-local variables. Optionally, a
458 TLS model may be specified:
461 For variables that are only used within the current shared library.
463 For variables in modules that will not be loaded dynamically.
465 For variables defined in the executable and only used within it.
467 The models correspond to the ELF TLS models; see `ELF Handling For
468 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
469 more information on under which circumstances the different models may
470 be used. The target may choose a different TLS model if the specified
471 model is not supported, or if a better choice of model can be made.
473 A variable may be defined as a global ``constant``, which indicates that
474 the contents of the variable will **never** be modified (enabling better
475 optimization, allowing the global data to be placed in the read-only
476 section of an executable, etc). Note that variables that need runtime
477 initialization cannot be marked ``constant`` as there is a store to the
480 LLVM explicitly allows *declarations* of global variables to be marked
481 constant, even if the final definition of the global is not. This
482 capability can be used to enable slightly better optimization of the
483 program, but requires the language definition to guarantee that
484 optimizations based on the 'constantness' are valid for the translation
485 units that do not include the definition.
487 As SSA values, global variables define pointer values that are in scope
488 (i.e. they dominate) all basic blocks in the program. Global variables
489 always define a pointer to their "content" type because they describe a
490 region of memory, and all memory objects in LLVM are accessed through
493 Global variables can be marked with ``unnamed_addr`` which indicates
494 that the address is not significant, only the content. Constants marked
495 like this can be merged with other constants if they have the same
496 initializer. Note that a constant with significant address *can* be
497 merged with a ``unnamed_addr`` constant, the result being a constant
498 whose address is significant.
500 A global variable may be declared to reside in a target-specific
501 numbered address space. For targets that support them, address spaces
502 may affect how optimizations are performed and/or what target
503 instructions are used to access the variable. The default address space
504 is zero. The address space qualifier must precede any other attributes.
506 LLVM allows an explicit section to be specified for globals. If the
507 target supports it, it will emit globals to the section specified.
509 By default, global initializers are optimized by assuming that global
510 variables defined within the module are not modified from their
511 initial values before the start of the global initializer. This is
512 true even for variables potentially accessible from outside the
513 module, including those with external linkage or appearing in
514 ``@llvm.used``. This assumption may be suppressed by marking the
515 variable with ``externally_initialized``.
517 An explicit alignment may be specified for a global, which must be a
518 power of 2. If not present, or if the alignment is set to zero, the
519 alignment of the global is set by the target to whatever it feels
520 convenient. If an explicit alignment is specified, the global is forced
521 to have exactly that alignment. Targets and optimizers are not allowed
522 to over-align the global if the global has an assigned section. In this
523 case, the extra alignment could be observable: for example, code could
524 assume that the globals are densely packed in their section and try to
525 iterate over them as an array, alignment padding would break this
528 For example, the following defines a global in a numbered address space
529 with an initializer, section, and alignment:
533 @G = addrspace(5) constant float 1.0, section "foo", align 4
535 The following example defines a thread-local global with the
536 ``initialexec`` TLS model:
540 @G = thread_local(initialexec) global i32 0, align 4
542 .. _functionstructure:
547 LLVM function definitions consist of the "``define``" keyword, an
548 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
549 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
550 an optional ``unnamed_addr`` attribute, a return type, an optional
551 :ref:`parameter attribute <paramattrs>` for the return type, a function
552 name, a (possibly empty) argument list (each with optional :ref:`parameter
553 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
554 an optional section, an optional alignment, an optional :ref:`garbage
555 collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
556 curly brace, a list of basic blocks, and a closing curly brace.
558 LLVM function declarations consist of the "``declare``" keyword, an
559 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
560 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
561 an optional ``unnamed_addr`` attribute, a return type, an optional
562 :ref:`parameter attribute <paramattrs>` for the return type, a function
563 name, a possibly empty list of arguments, an optional alignment, an optional
564 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
566 A function definition contains a list of basic blocks, forming the CFG
567 (Control Flow Graph) for the function. Each basic block may optionally
568 start with a label (giving the basic block a symbol table entry),
569 contains a list of instructions, and ends with a
570 :ref:`terminator <terminators>` instruction (such as a branch or function
571 return). If explicit label is not provided, a block is assigned an
572 implicit numbered label, using a next value from the same counter as used
573 for unnamed temporaries (:ref:`see above<identifiers>`). For example, if a
574 function entry block does not have explicit label, it will be assigned
575 label "%0", then first unnamed temporary in that block will be "%1", etc.
577 The first basic block in a function is special in two ways: it is
578 immediately executed on entrance to the function, and it is not allowed
579 to have predecessor basic blocks (i.e. there can not be any branches to
580 the entry block of a function). Because the block can have no
581 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
583 LLVM allows an explicit section to be specified for functions. If the
584 target supports it, it will emit functions to the section specified.
586 An explicit alignment may be specified for a function. If not present,
587 or if the alignment is set to zero, the alignment of the function is set
588 by the target to whatever it feels convenient. If an explicit alignment
589 is specified, the function is forced to have at least that much
590 alignment. All alignments must be a power of 2.
592 If the ``unnamed_addr`` attribute is given, the address is know to not
593 be significant and two identical functions can be merged.
597 define [linkage] [visibility]
599 <ResultType> @<FunctionName> ([argument list])
600 [fn Attrs] [section "name"] [align N]
601 [gc] [prefix Constant] { ... }
608 Aliases act as "second name" for the aliasee value (which can be either
609 function, global variable, another alias or bitcast of global value).
610 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
611 :ref:`visibility style <visibility>`.
615 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
617 .. _namedmetadatastructure:
622 Named metadata is a collection of metadata. :ref:`Metadata
623 nodes <metadata>` (but not metadata strings) are the only valid
624 operands for a named metadata.
628 ; Some unnamed metadata nodes, which are referenced by the named metadata.
629 !0 = metadata !{metadata !"zero"}
630 !1 = metadata !{metadata !"one"}
631 !2 = metadata !{metadata !"two"}
633 !name = !{!0, !1, !2}
640 The return type and each parameter of a function type may have a set of
641 *parameter attributes* associated with them. Parameter attributes are
642 used to communicate additional information about the result or
643 parameters of a function. Parameter attributes are considered to be part
644 of the function, not of the function type, so functions with different
645 parameter attributes can have the same function type.
647 Parameter attributes are simple keywords that follow the type specified.
648 If multiple parameter attributes are needed, they are space separated.
653 declare i32 @printf(i8* noalias nocapture, ...)
654 declare i32 @atoi(i8 zeroext)
655 declare signext i8 @returns_signed_char()
657 Note that any attributes for the function result (``nounwind``,
658 ``readonly``) come immediately after the argument list.
660 Currently, only the following parameter attributes are defined:
663 This indicates to the code generator that the parameter or return
664 value should be zero-extended to the extent required by the target's
665 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
666 the caller (for a parameter) or the callee (for a return value).
668 This indicates to the code generator that the parameter or return
669 value should be sign-extended to the extent required by the target's
670 ABI (which is usually 32-bits) by the caller (for a parameter) or
671 the callee (for a return value).
673 This indicates that this parameter or return value should be treated
674 in a special target-dependent fashion during while emitting code for
675 a function call or return (usually, by putting it in a register as
676 opposed to memory, though some targets use it to distinguish between
677 two different kinds of registers). Use of this attribute is
680 This indicates that the pointer parameter should really be passed by
681 value to the function. The attribute implies that a hidden copy of
682 the pointee is made between the caller and the callee, so the callee
683 is unable to modify the value in the caller. This attribute is only
684 valid on LLVM pointer arguments. It is generally used to pass
685 structs and arrays by value, but is also valid on pointers to
686 scalars. The copy is considered to belong to the caller not the
687 callee (for example, ``readonly`` functions should not write to
688 ``byval`` parameters). This is not a valid attribute for return
691 The byval attribute also supports specifying an alignment with the
692 align attribute. It indicates the alignment of the stack slot to
693 form and the known alignment of the pointer specified to the call
694 site. If the alignment is not specified, then the code generator
695 makes a target-specific assumption.
698 This indicates that the pointer parameter specifies the address of a
699 structure that is the return value of the function in the source
700 program. This pointer must be guaranteed by the caller to be valid:
701 loads and stores to the structure may be assumed by the callee
702 not to trap and to be properly aligned. This may only be applied to
703 the first parameter. This is not a valid attribute for return
706 This indicates that pointer values :ref:`based <pointeraliasing>` on
707 the argument or return value do not alias pointer values which are
708 not *based* on it, ignoring certain "irrelevant" dependencies. For a
709 call to the parent function, dependencies between memory references
710 from before or after the call and from those during the call are
711 "irrelevant" to the ``noalias`` keyword for the arguments and return
712 value used in that call. The caller shares the responsibility with
713 the callee for ensuring that these requirements are met. For further
714 details, please see the discussion of the NoAlias response in `alias
715 analysis <AliasAnalysis.html#MustMayNo>`_.
717 Note that this definition of ``noalias`` is intentionally similar
718 to the definition of ``restrict`` in C99 for function arguments,
719 though it is slightly weaker.
721 For function return values, C99's ``restrict`` is not meaningful,
722 while LLVM's ``noalias`` is.
724 This indicates that the callee does not make any copies of the
725 pointer that outlive the callee itself. This is not a valid
726 attribute for return values.
731 This indicates that the pointer parameter can be excised using the
732 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
733 attribute for return values and can only be applied to one parameter.
736 This indicates that the function always returns the argument as its return
737 value. This is an optimization hint to the code generator when generating
738 the caller, allowing tail call optimization and omission of register saves
739 and restores in some cases; it is not checked or enforced when generating
740 the callee. The parameter and the function return type must be valid
741 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
742 valid attribute for return values and can only be applied to one parameter.
746 Garbage Collector Names
747 -----------------------
749 Each function may specify a garbage collector name, which is simply a
754 define void @f() gc "name" { ... }
756 The compiler declares the supported values of *name*. Specifying a
757 collector which will cause the compiler to alter its output in order to
758 support the named garbage collection algorithm.
765 Prefix data is data associated with a function which the code generator
766 will emit immediately before the function body. The purpose of this feature
767 is to allow frontends to associate language-specific runtime metadata with
768 specific functions and make it available through the function pointer while
769 still allowing the function pointer to be called. To access the data for a
770 given function, a program may bitcast the function pointer to a pointer to
771 the constant's type. This implies that the IR symbol points to the start
774 To maintain the semantics of ordinary function calls, the prefix data must
775 have a particular format. Specifically, it must begin with a sequence of
776 bytes which decode to a sequence of machine instructions, valid for the
777 module's target, which transfer control to the point immediately succeeding
778 the prefix data, without performing any other visible action. This allows
779 the inliner and other passes to reason about the semantics of the function
780 definition without needing to reason about the prefix data. Obviously this
781 makes the format of the prefix data highly target dependent.
783 Prefix data is laid out as if it were an initializer for a global variable
784 of the prefix data's type. No padding is automatically placed between the
785 prefix data and the function body. If padding is required, it must be part
788 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
789 which encodes the ``nop`` instruction:
793 define void @f() prefix i8 144 { ... }
795 Generally prefix data can be formed by encoding a relative branch instruction
796 which skips the metadata, as in this example of valid prefix data for the
797 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
801 %0 = type <{ i8, i8, i8* }>
803 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
805 A function may have prefix data but no body. This has similar semantics
806 to the ``available_externally`` linkage in that the data may be used by the
807 optimizers but will not be emitted in the object file.
814 Attribute groups are groups of attributes that are referenced by objects within
815 the IR. They are important for keeping ``.ll`` files readable, because a lot of
816 functions will use the same set of attributes. In the degenerative case of a
817 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
818 group will capture the important command line flags used to build that file.
820 An attribute group is a module-level object. To use an attribute group, an
821 object references the attribute group's ID (e.g. ``#37``). An object may refer
822 to more than one attribute group. In that situation, the attributes from the
823 different groups are merged.
825 Here is an example of attribute groups for a function that should always be
826 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
830 ; Target-independent attributes:
831 attributes #0 = { alwaysinline alignstack=4 }
833 ; Target-dependent attributes:
834 attributes #1 = { "no-sse" }
836 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
837 define void @f() #0 #1 { ... }
844 Function attributes are set to communicate additional information about
845 a function. Function attributes are considered to be part of the
846 function, not of the function type, so functions with different function
847 attributes can have the same function type.
849 Function attributes are simple keywords that follow the type specified.
850 If multiple attributes are needed, they are space separated. For
855 define void @f() noinline { ... }
856 define void @f() alwaysinline { ... }
857 define void @f() alwaysinline optsize { ... }
858 define void @f() optsize { ... }
861 This attribute indicates that, when emitting the prologue and
862 epilogue, the backend should forcibly align the stack pointer.
863 Specify the desired alignment, which must be a power of two, in
866 This attribute indicates that the inliner should attempt to inline
867 this function into callers whenever possible, ignoring any active
868 inlining size threshold for this caller.
870 This indicates that the callee function at a call site should be
871 recognized as a built-in function, even though the function's declaration
872 uses the ``nobuiltin`` attribute. This is only valid at call sites for
873 direct calls to functions which are declared with the ``nobuiltin``
876 This attribute indicates that this function is rarely called. When
877 computing edge weights, basic blocks post-dominated by a cold
878 function call are also considered to be cold; and, thus, given low
881 This attribute indicates that the source code contained a hint that
882 inlining this function is desirable (such as the "inline" keyword in
883 C/C++). It is just a hint; it imposes no requirements on the
886 This attribute suggests that optimization passes and code generator
887 passes make choices that keep the code size of this function as small
888 as possible and perform optimizations that may sacrifice runtime
889 performance in order to minimize the size of the generated code.
891 This attribute disables prologue / epilogue emission for the
892 function. This can have very system-specific consequences.
894 This indicates that the callee function at a call site is not recognized as
895 a built-in function. LLVM will retain the original call and not replace it
896 with equivalent code based on the semantics of the built-in function, unless
897 the call site uses the ``builtin`` attribute. This is valid at call sites
898 and on function declarations and definitions.
900 This attribute indicates that calls to the function cannot be
901 duplicated. A call to a ``noduplicate`` function may be moved
902 within its parent function, but may not be duplicated within
905 A function containing a ``noduplicate`` call may still
906 be an inlining candidate, provided that the call is not
907 duplicated by inlining. That implies that the function has
908 internal linkage and only has one call site, so the original
909 call is dead after inlining.
911 This attributes disables implicit floating point instructions.
913 This attribute indicates that the inliner should never inline this
914 function in any situation. This attribute may not be used together
915 with the ``alwaysinline`` attribute.
917 This attribute suppresses lazy symbol binding for the function. This
918 may make calls to the function faster, at the cost of extra program
919 startup time if the function is not called during program startup.
921 This attribute indicates that the code generator should not use a
922 red zone, even if the target-specific ABI normally permits it.
924 This function attribute indicates that the function never returns
925 normally. This produces undefined behavior at runtime if the
926 function ever does dynamically return.
928 This function attribute indicates that the function never returns
929 with an unwind or exceptional control flow. If the function does
930 unwind, its runtime behavior is undefined.
932 This function attribute indicates that the function is not optimized
933 by any optimization or code generator passes with the
934 exception of interprocedural optimization passes.
935 This attribute cannot be used together with the ``alwaysinline``
936 attribute; this attribute is also incompatible
937 with the ``minsize`` attribute and the ``optsize`` attribute.
939 The inliner should never inline this function in any situation.
940 Only functions with the ``alwaysinline`` attribute are valid
941 candidates for inlining inside the body of this function.
943 This attribute suggests that optimization passes and code generator
944 passes make choices that keep the code size of this function low,
945 and otherwise do optimizations specifically to reduce code size as
946 long as they do not significantly impact runtime performance.
948 On a function, this attribute indicates that the function computes its
949 result (or decides to unwind an exception) based strictly on its arguments,
950 without dereferencing any pointer arguments or otherwise accessing
951 any mutable state (e.g. memory, control registers, etc) visible to
952 caller functions. It does not write through any pointer arguments
953 (including ``byval`` arguments) and never changes any state visible
954 to callers. This means that it cannot unwind exceptions by calling
955 the ``C++`` exception throwing methods.
957 On an argument, this attribute indicates that the function does not
958 dereference that pointer argument, even though it may read or write the
959 memory that the pointer points to if accessed through other pointers.
961 On a function, this attribute indicates that the function does not write
962 through any pointer arguments (including ``byval`` arguments) or otherwise
963 modify any state (e.g. memory, control registers, etc) visible to
964 caller functions. It may dereference pointer arguments and read
965 state that may be set in the caller. A readonly function always
966 returns the same value (or unwinds an exception identically) when
967 called with the same set of arguments and global state. It cannot
968 unwind an exception by calling the ``C++`` exception throwing
971 On an argument, this attribute indicates that the function does not write
972 through this pointer argument, even though it may write to the memory that
973 the pointer points to.
975 This attribute indicates that this function can return twice. The C
976 ``setjmp`` is an example of such a function. The compiler disables
977 some optimizations (like tail calls) in the caller of these
980 This attribute indicates that AddressSanitizer checks
981 (dynamic address safety analysis) are enabled for this function.
983 This attribute indicates that MemorySanitizer checks (dynamic detection
984 of accesses to uninitialized memory) are enabled for this function.
986 This attribute indicates that ThreadSanitizer checks
987 (dynamic thread safety analysis) are enabled for this function.
989 This attribute indicates that the function should emit a stack
990 smashing protector. It is in the form of a "canary" --- a random value
991 placed on the stack before the local variables that's checked upon
992 return from the function to see if it has been overwritten. A
993 heuristic is used to determine if a function needs stack protectors
994 or not. The heuristic used will enable protectors for functions with:
996 - Character arrays larger than ``ssp-buffer-size`` (default 8).
997 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
998 - Calls to alloca() with variable sizes or constant sizes greater than
1001 If a function that has an ``ssp`` attribute is inlined into a
1002 function that doesn't have an ``ssp`` attribute, then the resulting
1003 function will have an ``ssp`` attribute.
1005 This attribute indicates that the function should *always* emit a
1006 stack smashing protector. This overrides the ``ssp`` function
1009 If a function that has an ``sspreq`` attribute is inlined into a
1010 function that doesn't have an ``sspreq`` attribute or which has an
1011 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1012 an ``sspreq`` attribute.
1014 This attribute indicates that the function should emit a stack smashing
1015 protector. This attribute causes a strong heuristic to be used when
1016 determining if a function needs stack protectors. The strong heuristic
1017 will enable protectors for functions with:
1019 - Arrays of any size and type
1020 - Aggregates containing an array of any size and type.
1021 - Calls to alloca().
1022 - Local variables that have had their address taken.
1024 This overrides the ``ssp`` function attribute.
1026 If a function that has an ``sspstrong`` attribute is inlined into a
1027 function that doesn't have an ``sspstrong`` attribute, then the
1028 resulting function will have an ``sspstrong`` attribute.
1030 This attribute indicates that the ABI being targeted requires that
1031 an unwind table entry be produce for this function even if we can
1032 show that no exceptions passes by it. This is normally the case for
1033 the ELF x86-64 abi, but it can be disabled for some compilation
1038 Module-Level Inline Assembly
1039 ----------------------------
1041 Modules may contain "module-level inline asm" blocks, which corresponds
1042 to the GCC "file scope inline asm" blocks. These blocks are internally
1043 concatenated by LLVM and treated as a single unit, but may be separated
1044 in the ``.ll`` file if desired. The syntax is very simple:
1046 .. code-block:: llvm
1048 module asm "inline asm code goes here"
1049 module asm "more can go here"
1051 The strings can contain any character by escaping non-printable
1052 characters. The escape sequence used is simply "\\xx" where "xx" is the
1053 two digit hex code for the number.
1055 The inline asm code is simply printed to the machine code .s file when
1056 assembly code is generated.
1058 .. _langref_datalayout:
1063 A module may specify a target specific data layout string that specifies
1064 how data is to be laid out in memory. The syntax for the data layout is
1067 .. code-block:: llvm
1069 target datalayout = "layout specification"
1071 The *layout specification* consists of a list of specifications
1072 separated by the minus sign character ('-'). Each specification starts
1073 with a letter and may include other information after the letter to
1074 define some aspect of the data layout. The specifications accepted are
1078 Specifies that the target lays out data in big-endian form. That is,
1079 the bits with the most significance have the lowest address
1082 Specifies that the target lays out data in little-endian form. That
1083 is, the bits with the least significance have the lowest address
1086 Specifies the natural alignment of the stack in bits. Alignment
1087 promotion of stack variables is limited to the natural stack
1088 alignment to avoid dynamic stack realignment. The stack alignment
1089 must be a multiple of 8-bits. If omitted, the natural stack
1090 alignment defaults to "unspecified", which does not prevent any
1091 alignment promotions.
1092 ``p[n]:<size>:<abi>:<pref>``
1093 This specifies the *size* of a pointer and its ``<abi>`` and
1094 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1095 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1096 preceding ``:`` should be omitted too. The address space, ``n`` is
1097 optional, and if not specified, denotes the default address space 0.
1098 The value of ``n`` must be in the range [1,2^23).
1099 ``i<size>:<abi>:<pref>``
1100 This specifies the alignment for an integer type of a given bit
1101 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1102 ``v<size>:<abi>:<pref>``
1103 This specifies the alignment for a vector type of a given bit
1105 ``f<size>:<abi>:<pref>``
1106 This specifies the alignment for a floating point type of a given bit
1107 ``<size>``. Only values of ``<size>`` that are supported by the target
1108 will work. 32 (float) and 64 (double) are supported on all targets; 80
1109 or 128 (different flavors of long double) are also supported on some
1111 ``a<size>:<abi>:<pref>``
1112 This specifies the alignment for an aggregate type of a given bit
1114 ``s<size>:<abi>:<pref>``
1115 This specifies the alignment for a stack object of a given bit
1117 ``n<size1>:<size2>:<size3>...``
1118 This specifies a set of native integer widths for the target CPU in
1119 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1120 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1121 this set are considered to support most general arithmetic operations
1124 When constructing the data layout for a given target, LLVM starts with a
1125 default set of specifications which are then (possibly) overridden by
1126 the specifications in the ``datalayout`` keyword. The default
1127 specifications are given in this list:
1129 - ``E`` - big endian
1130 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1131 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1132 same as the default address space.
1133 - ``S0`` - natural stack alignment is unspecified
1134 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1135 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1136 - ``i16:16:16`` - i16 is 16-bit aligned
1137 - ``i32:32:32`` - i32 is 32-bit aligned
1138 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1139 alignment of 64-bits
1140 - ``f16:16:16`` - half is 16-bit aligned
1141 - ``f32:32:32`` - float is 32-bit aligned
1142 - ``f64:64:64`` - double is 64-bit aligned
1143 - ``f128:128:128`` - quad is 128-bit aligned
1144 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1145 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1146 - ``a0:0:64`` - aggregates are 64-bit aligned
1148 When LLVM is determining the alignment for a given type, it uses the
1151 #. If the type sought is an exact match for one of the specifications,
1152 that specification is used.
1153 #. If no match is found, and the type sought is an integer type, then
1154 the smallest integer type that is larger than the bitwidth of the
1155 sought type is used. If none of the specifications are larger than
1156 the bitwidth then the largest integer type is used. For example,
1157 given the default specifications above, the i7 type will use the
1158 alignment of i8 (next largest) while both i65 and i256 will use the
1159 alignment of i64 (largest specified).
1160 #. If no match is found, and the type sought is a vector type, then the
1161 largest vector type that is smaller than the sought vector type will
1162 be used as a fall back. This happens because <128 x double> can be
1163 implemented in terms of 64 <2 x double>, for example.
1165 The function of the data layout string may not be what you expect.
1166 Notably, this is not a specification from the frontend of what alignment
1167 the code generator should use.
1169 Instead, if specified, the target data layout is required to match what
1170 the ultimate *code generator* expects. This string is used by the
1171 mid-level optimizers to improve code, and this only works if it matches
1172 what the ultimate code generator uses. If you would like to generate IR
1173 that does not embed this target-specific detail into the IR, then you
1174 don't have to specify the string. This will disable some optimizations
1175 that require precise layout information, but this also prevents those
1176 optimizations from introducing target specificity into the IR.
1178 .. _pointeraliasing:
1180 Pointer Aliasing Rules
1181 ----------------------
1183 Any memory access must be done through a pointer value associated with
1184 an address range of the memory access, otherwise the behavior is
1185 undefined. Pointer values are associated with address ranges according
1186 to the following rules:
1188 - A pointer value is associated with the addresses associated with any
1189 value it is *based* on.
1190 - An address of a global variable is associated with the address range
1191 of the variable's storage.
1192 - The result value of an allocation instruction is associated with the
1193 address range of the allocated storage.
1194 - A null pointer in the default address-space is associated with no
1196 - An integer constant other than zero or a pointer value returned from
1197 a function not defined within LLVM may be associated with address
1198 ranges allocated through mechanisms other than those provided by
1199 LLVM. Such ranges shall not overlap with any ranges of addresses
1200 allocated by mechanisms provided by LLVM.
1202 A pointer value is *based* on another pointer value according to the
1205 - A pointer value formed from a ``getelementptr`` operation is *based*
1206 on the first operand of the ``getelementptr``.
1207 - The result value of a ``bitcast`` is *based* on the operand of the
1209 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1210 values that contribute (directly or indirectly) to the computation of
1211 the pointer's value.
1212 - The "*based* on" relationship is transitive.
1214 Note that this definition of *"based"* is intentionally similar to the
1215 definition of *"based"* in C99, though it is slightly weaker.
1217 LLVM IR does not associate types with memory. The result type of a
1218 ``load`` merely indicates the size and alignment of the memory from
1219 which to load, as well as the interpretation of the value. The first
1220 operand type of a ``store`` similarly only indicates the size and
1221 alignment of the store.
1223 Consequently, type-based alias analysis, aka TBAA, aka
1224 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1225 :ref:`Metadata <metadata>` may be used to encode additional information
1226 which specialized optimization passes may use to implement type-based
1231 Volatile Memory Accesses
1232 ------------------------
1234 Certain memory accesses, such as :ref:`load <i_load>`'s,
1235 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1236 marked ``volatile``. The optimizers must not change the number of
1237 volatile operations or change their order of execution relative to other
1238 volatile operations. The optimizers *may* change the order of volatile
1239 operations relative to non-volatile operations. This is not Java's
1240 "volatile" and has no cross-thread synchronization behavior.
1242 IR-level volatile loads and stores cannot safely be optimized into
1243 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1244 flagged volatile. Likewise, the backend should never split or merge
1245 target-legal volatile load/store instructions.
1247 .. admonition:: Rationale
1249 Platforms may rely on volatile loads and stores of natively supported
1250 data width to be executed as single instruction. For example, in C
1251 this holds for an l-value of volatile primitive type with native
1252 hardware support, but not necessarily for aggregate types. The
1253 frontend upholds these expectations, which are intentionally
1254 unspecified in the IR. The rules above ensure that IR transformation
1255 do not violate the frontend's contract with the language.
1259 Memory Model for Concurrent Operations
1260 --------------------------------------
1262 The LLVM IR does not define any way to start parallel threads of
1263 execution or to register signal handlers. Nonetheless, there are
1264 platform-specific ways to create them, and we define LLVM IR's behavior
1265 in their presence. This model is inspired by the C++0x memory model.
1267 For a more informal introduction to this model, see the :doc:`Atomics`.
1269 We define a *happens-before* partial order as the least partial order
1272 - Is a superset of single-thread program order, and
1273 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1274 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1275 techniques, like pthread locks, thread creation, thread joining,
1276 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1277 Constraints <ordering>`).
1279 Note that program order does not introduce *happens-before* edges
1280 between a thread and signals executing inside that thread.
1282 Every (defined) read operation (load instructions, memcpy, atomic
1283 loads/read-modify-writes, etc.) R reads a series of bytes written by
1284 (defined) write operations (store instructions, atomic
1285 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1286 section, initialized globals are considered to have a write of the
1287 initializer which is atomic and happens before any other read or write
1288 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1289 may see any write to the same byte, except:
1291 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1292 write\ :sub:`2` happens before R\ :sub:`byte`, then
1293 R\ :sub:`byte` does not see write\ :sub:`1`.
1294 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1295 R\ :sub:`byte` does not see write\ :sub:`3`.
1297 Given that definition, R\ :sub:`byte` is defined as follows:
1299 - If R is volatile, the result is target-dependent. (Volatile is
1300 supposed to give guarantees which can support ``sig_atomic_t`` in
1301 C/C++, and may be used for accesses to addresses which do not behave
1302 like normal memory. It does not generally provide cross-thread
1304 - Otherwise, if there is no write to the same byte that happens before
1305 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1306 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1307 R\ :sub:`byte` returns the value written by that write.
1308 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1309 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1310 Memory Ordering Constraints <ordering>` section for additional
1311 constraints on how the choice is made.
1312 - Otherwise R\ :sub:`byte` returns ``undef``.
1314 R returns the value composed of the series of bytes it read. This
1315 implies that some bytes within the value may be ``undef`` **without**
1316 the entire value being ``undef``. Note that this only defines the
1317 semantics of the operation; it doesn't mean that targets will emit more
1318 than one instruction to read the series of bytes.
1320 Note that in cases where none of the atomic intrinsics are used, this
1321 model places only one restriction on IR transformations on top of what
1322 is required for single-threaded execution: introducing a store to a byte
1323 which might not otherwise be stored is not allowed in general.
1324 (Specifically, in the case where another thread might write to and read
1325 from an address, introducing a store can change a load that may see
1326 exactly one write into a load that may see multiple writes.)
1330 Atomic Memory Ordering Constraints
1331 ----------------------------------
1333 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1334 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1335 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1336 an ordering parameter that determines which other atomic instructions on
1337 the same address they *synchronize with*. These semantics are borrowed
1338 from Java and C++0x, but are somewhat more colloquial. If these
1339 descriptions aren't precise enough, check those specs (see spec
1340 references in the :doc:`atomics guide <Atomics>`).
1341 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1342 differently since they don't take an address. See that instruction's
1343 documentation for details.
1345 For a simpler introduction to the ordering constraints, see the
1349 The set of values that can be read is governed by the happens-before
1350 partial order. A value cannot be read unless some operation wrote
1351 it. This is intended to provide a guarantee strong enough to model
1352 Java's non-volatile shared variables. This ordering cannot be
1353 specified for read-modify-write operations; it is not strong enough
1354 to make them atomic in any interesting way.
1356 In addition to the guarantees of ``unordered``, there is a single
1357 total order for modifications by ``monotonic`` operations on each
1358 address. All modification orders must be compatible with the
1359 happens-before order. There is no guarantee that the modification
1360 orders can be combined to a global total order for the whole program
1361 (and this often will not be possible). The read in an atomic
1362 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1363 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1364 order immediately before the value it writes. If one atomic read
1365 happens before another atomic read of the same address, the later
1366 read must see the same value or a later value in the address's
1367 modification order. This disallows reordering of ``monotonic`` (or
1368 stronger) operations on the same address. If an address is written
1369 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1370 read that address repeatedly, the other threads must eventually see
1371 the write. This corresponds to the C++0x/C1x
1372 ``memory_order_relaxed``.
1374 In addition to the guarantees of ``monotonic``, a
1375 *synchronizes-with* edge may be formed with a ``release`` operation.
1376 This is intended to model C++'s ``memory_order_acquire``.
1378 In addition to the guarantees of ``monotonic``, if this operation
1379 writes a value which is subsequently read by an ``acquire``
1380 operation, it *synchronizes-with* that operation. (This isn't a
1381 complete description; see the C++0x definition of a release
1382 sequence.) This corresponds to the C++0x/C1x
1383 ``memory_order_release``.
1384 ``acq_rel`` (acquire+release)
1385 Acts as both an ``acquire`` and ``release`` operation on its
1386 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1387 ``seq_cst`` (sequentially consistent)
1388 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1389 operation which only reads, ``release`` for an operation which only
1390 writes), there is a global total order on all
1391 sequentially-consistent operations on all addresses, which is
1392 consistent with the *happens-before* partial order and with the
1393 modification orders of all the affected addresses. Each
1394 sequentially-consistent read sees the last preceding write to the
1395 same address in this global order. This corresponds to the C++0x/C1x
1396 ``memory_order_seq_cst`` and Java volatile.
1400 If an atomic operation is marked ``singlethread``, it only *synchronizes
1401 with* or participates in modification and seq\_cst total orderings with
1402 other operations running in the same thread (for example, in signal
1410 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1411 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1412 :ref:`frem <i_frem>`) have the following flags that can set to enable
1413 otherwise unsafe floating point operations
1416 No NaNs - Allow optimizations to assume the arguments and result are not
1417 NaN. Such optimizations are required to retain defined behavior over
1418 NaNs, but the value of the result is undefined.
1421 No Infs - Allow optimizations to assume the arguments and result are not
1422 +/-Inf. Such optimizations are required to retain defined behavior over
1423 +/-Inf, but the value of the result is undefined.
1426 No Signed Zeros - Allow optimizations to treat the sign of a zero
1427 argument or result as insignificant.
1430 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1431 argument rather than perform division.
1434 Fast - Allow algebraically equivalent transformations that may
1435 dramatically change results in floating point (e.g. reassociate). This
1436 flag implies all the others.
1443 The LLVM type system is one of the most important features of the
1444 intermediate representation. Being typed enables a number of
1445 optimizations to be performed on the intermediate representation
1446 directly, without having to do extra analyses on the side before the
1447 transformation. A strong type system makes it easier to read the
1448 generated code and enables novel analyses and transformations that are
1449 not feasible to perform on normal three address code representations.
1451 .. _typeclassifications:
1453 Type Classifications
1454 --------------------
1456 The types fall into a few useful classifications:
1465 * - :ref:`integer <t_integer>`
1466 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1469 * - :ref:`floating point <t_floating>`
1470 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1478 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1479 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1480 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1481 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1483 * - :ref:`primitive <t_primitive>`
1484 - :ref:`label <t_label>`,
1485 :ref:`void <t_void>`,
1486 :ref:`integer <t_integer>`,
1487 :ref:`floating point <t_floating>`,
1488 :ref:`x86mmx <t_x86mmx>`,
1489 :ref:`metadata <t_metadata>`.
1491 * - :ref:`derived <t_derived>`
1492 - :ref:`array <t_array>`,
1493 :ref:`function <t_function>`,
1494 :ref:`pointer <t_pointer>`,
1495 :ref:`structure <t_struct>`,
1496 :ref:`vector <t_vector>`,
1497 :ref:`opaque <t_opaque>`.
1499 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1500 Values of these types are the only ones which can be produced by
1508 The primitive types are the fundamental building blocks of the LLVM
1519 The integer type is a very simple type that simply specifies an
1520 arbitrary bit width for the integer type desired. Any bit width from 1
1521 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1530 The number of bits the integer will occupy is specified by the ``N``
1536 +----------------+------------------------------------------------+
1537 | ``i1`` | a single-bit integer. |
1538 +----------------+------------------------------------------------+
1539 | ``i32`` | a 32-bit integer. |
1540 +----------------+------------------------------------------------+
1541 | ``i1942652`` | a really big integer of over 1 million bits. |
1542 +----------------+------------------------------------------------+
1546 Floating Point Types
1547 ^^^^^^^^^^^^^^^^^^^^
1556 - 16-bit floating point value
1559 - 32-bit floating point value
1562 - 64-bit floating point value
1565 - 128-bit floating point value (112-bit mantissa)
1568 - 80-bit floating point value (X87)
1571 - 128-bit floating point value (two 64-bits)
1581 The x86mmx type represents a value held in an MMX register on an x86
1582 machine. The operations allowed on it are quite limited: parameters and
1583 return values, load and store, and bitcast. User-specified MMX
1584 instructions are represented as intrinsic or asm calls with arguments
1585 and/or results of this type. There are no arrays, vectors or constants
1603 The void type does not represent any value and has no size.
1620 The label type represents code labels.
1637 The metadata type represents embedded metadata. No derived types may be
1638 created from metadata except for :ref:`function <t_function>` arguments.
1652 The real power in LLVM comes from the derived types in the system. This
1653 is what allows a programmer to represent arrays, functions, pointers,
1654 and other useful types. Each of these types contain one or more element
1655 types which may be a primitive type, or another derived type. For
1656 example, it is possible to have a two dimensional array, using an array
1657 as the element type of another array.
1664 Aggregate Types are a subset of derived types that can contain multiple
1665 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1666 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1677 The array type is a very simple derived type that arranges elements
1678 sequentially in memory. The array type requires a size (number of
1679 elements) and an underlying data type.
1686 [<# elements> x <elementtype>]
1688 The number of elements is a constant integer value; ``elementtype`` may
1689 be any type with a size.
1694 +------------------+--------------------------------------+
1695 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1696 +------------------+--------------------------------------+
1697 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1698 +------------------+--------------------------------------+
1699 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1700 +------------------+--------------------------------------+
1702 Here are some examples of multidimensional arrays:
1704 +-----------------------------+----------------------------------------------------------+
1705 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1706 +-----------------------------+----------------------------------------------------------+
1707 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1708 +-----------------------------+----------------------------------------------------------+
1709 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1710 +-----------------------------+----------------------------------------------------------+
1712 There is no restriction on indexing beyond the end of the array implied
1713 by a static type (though there are restrictions on indexing beyond the
1714 bounds of an allocated object in some cases). This means that
1715 single-dimension 'variable sized array' addressing can be implemented in
1716 LLVM with a zero length array type. An implementation of 'pascal style
1717 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1728 The function type can be thought of as a function signature. It consists
1729 of a return type and a list of formal parameter types. The return type
1730 of a function type is a first class type or a void type.
1737 <returntype> (<parameter list>)
1739 ...where '``<parameter list>``' is a comma-separated list of type
1740 specifiers. Optionally, the parameter list may include a type ``...``,
1741 which indicates that the function takes a variable number of arguments.
1742 Variable argument functions can access their arguments with the
1743 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1744 '``<returntype>``' is any type except :ref:`label <t_label>`.
1749 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1750 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1751 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1752 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1753 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1754 | ``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. |
1755 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1756 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1757 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1767 The structure type is used to represent a collection of data members
1768 together in memory. The elements of a structure may be any type that has
1771 Structures in memory are accessed using '``load``' and '``store``' by
1772 getting a pointer to a field with the '``getelementptr``' instruction.
1773 Structures in registers are accessed using the '``extractvalue``' and
1774 '``insertvalue``' instructions.
1776 Structures may optionally be "packed" structures, which indicate that
1777 the alignment of the struct is one byte, and that there is no padding
1778 between the elements. In non-packed structs, padding between field types
1779 is inserted as defined by the DataLayout string in the module, which is
1780 required to match what the underlying code generator expects.
1782 Structures can either be "literal" or "identified". A literal structure
1783 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1784 identified types are always defined at the top level with a name.
1785 Literal types are uniqued by their contents and can never be recursive
1786 or opaque since there is no way to write one. Identified types can be
1787 recursive, can be opaqued, and are never uniqued.
1794 %T1 = type { <type list> } ; Identified normal struct type
1795 %T2 = type <{ <type list> }> ; Identified packed struct type
1800 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1801 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1802 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1803 | ``{ 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``. |
1804 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1805 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1806 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1810 Opaque Structure Types
1811 ^^^^^^^^^^^^^^^^^^^^^^
1816 Opaque structure types are used to represent named structure types that
1817 do not have a body specified. This corresponds (for example) to the C
1818 notion of a forward declared structure.
1831 +--------------+-------------------+
1832 | ``opaque`` | An opaque type. |
1833 +--------------+-------------------+
1843 The pointer type is used to specify memory locations. Pointers are
1844 commonly used to reference objects in memory.
1846 Pointer types may have an optional address space attribute defining the
1847 numbered address space where the pointed-to object resides. The default
1848 address space is number zero. The semantics of non-zero address spaces
1849 are target-specific.
1851 Note that LLVM does not permit pointers to void (``void*``) nor does it
1852 permit pointers to labels (``label*``). Use ``i8*`` instead.
1864 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1865 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1866 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1867 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1868 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1869 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1870 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1880 A vector type is a simple derived type that represents a vector of
1881 elements. Vector types are used when multiple primitive data are
1882 operated in parallel using a single instruction (SIMD). A vector type
1883 requires a size (number of elements) and an underlying primitive data
1884 type. Vector types are considered :ref:`first class <t_firstclass>`.
1891 < <# elements> x <elementtype> >
1893 The number of elements is a constant integer value larger than 0;
1894 elementtype may be any integer or floating point type, or a pointer to
1895 these types. Vectors of size zero are not allowed.
1900 +-------------------+--------------------------------------------------+
1901 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1902 +-------------------+--------------------------------------------------+
1903 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1904 +-------------------+--------------------------------------------------+
1905 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1906 +-------------------+--------------------------------------------------+
1907 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1908 +-------------------+--------------------------------------------------+
1913 LLVM has several different basic types of constants. This section
1914 describes them all and their syntax.
1919 **Boolean constants**
1920 The two strings '``true``' and '``false``' are both valid constants
1922 **Integer constants**
1923 Standard integers (such as '4') are constants of the
1924 :ref:`integer <t_integer>` type. Negative numbers may be used with
1926 **Floating point constants**
1927 Floating point constants use standard decimal notation (e.g.
1928 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1929 hexadecimal notation (see below). The assembler requires the exact
1930 decimal value of a floating-point constant. For example, the
1931 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1932 decimal in binary. Floating point constants must have a :ref:`floating
1933 point <t_floating>` type.
1934 **Null pointer constants**
1935 The identifier '``null``' is recognized as a null pointer constant
1936 and must be of :ref:`pointer type <t_pointer>`.
1938 The one non-intuitive notation for constants is the hexadecimal form of
1939 floating point constants. For example, the form
1940 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1941 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1942 constants are required (and the only time that they are generated by the
1943 disassembler) is when a floating point constant must be emitted but it
1944 cannot be represented as a decimal floating point number in a reasonable
1945 number of digits. For example, NaN's, infinities, and other special
1946 values are represented in their IEEE hexadecimal format so that assembly
1947 and disassembly do not cause any bits to change in the constants.
1949 When using the hexadecimal form, constants of types half, float, and
1950 double are represented using the 16-digit form shown above (which
1951 matches the IEEE754 representation for double); half and float values
1952 must, however, be exactly representable as IEEE 754 half and single
1953 precision, respectively. Hexadecimal format is always used for long
1954 double, and there are three forms of long double. The 80-bit format used
1955 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1956 128-bit format used by PowerPC (two adjacent doubles) is represented by
1957 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1958 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1959 will only work if they match the long double format on your target.
1960 The IEEE 16-bit format (half precision) is represented by ``0xH``
1961 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1962 (sign bit at the left).
1964 There are no constants of type x86mmx.
1966 .. _complexconstants:
1971 Complex constants are a (potentially recursive) combination of simple
1972 constants and smaller complex constants.
1974 **Structure constants**
1975 Structure constants are represented with notation similar to
1976 structure type definitions (a comma separated list of elements,
1977 surrounded by braces (``{}``)). For example:
1978 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1979 "``@G = external global i32``". Structure constants must have
1980 :ref:`structure type <t_struct>`, and the number and types of elements
1981 must match those specified by the type.
1983 Array constants are represented with notation similar to array type
1984 definitions (a comma separated list of elements, surrounded by
1985 square brackets (``[]``)). For example:
1986 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1987 :ref:`array type <t_array>`, and the number and types of elements must
1988 match those specified by the type.
1989 **Vector constants**
1990 Vector constants are represented with notation similar to vector
1991 type definitions (a comma separated list of elements, surrounded by
1992 less-than/greater-than's (``<>``)). For example:
1993 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1994 must have :ref:`vector type <t_vector>`, and the number and types of
1995 elements must match those specified by the type.
1996 **Zero initialization**
1997 The string '``zeroinitializer``' can be used to zero initialize a
1998 value to zero of *any* type, including scalar and
1999 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2000 having to print large zero initializers (e.g. for large arrays) and
2001 is always exactly equivalent to using explicit zero initializers.
2003 A metadata node is a structure-like constant with :ref:`metadata
2004 type <t_metadata>`. For example:
2005 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2006 constants that are meant to be interpreted as part of the
2007 instruction stream, metadata is a place to attach additional
2008 information such as debug info.
2010 Global Variable and Function Addresses
2011 --------------------------------------
2013 The addresses of :ref:`global variables <globalvars>` and
2014 :ref:`functions <functionstructure>` are always implicitly valid
2015 (link-time) constants. These constants are explicitly referenced when
2016 the :ref:`identifier for the global <identifiers>` is used and always have
2017 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2020 .. code-block:: llvm
2024 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2031 The string '``undef``' can be used anywhere a constant is expected, and
2032 indicates that the user of the value may receive an unspecified
2033 bit-pattern. Undefined values may be of any type (other than '``label``'
2034 or '``void``') and be used anywhere a constant is permitted.
2036 Undefined values are useful because they indicate to the compiler that
2037 the program is well defined no matter what value is used. This gives the
2038 compiler more freedom to optimize. Here are some examples of
2039 (potentially surprising) transformations that are valid (in pseudo IR):
2041 .. code-block:: llvm
2051 This is safe because all of the output bits are affected by the undef
2052 bits. Any output bit can have a zero or one depending on the input bits.
2054 .. code-block:: llvm
2065 These logical operations have bits that are not always affected by the
2066 input. For example, if ``%X`` has a zero bit, then the output of the
2067 '``and``' operation will always be a zero for that bit, no matter what
2068 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2069 optimize or assume that the result of the '``and``' is '``undef``'.
2070 However, it is safe to assume that all bits of the '``undef``' could be
2071 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2072 all the bits of the '``undef``' operand to the '``or``' could be set,
2073 allowing the '``or``' to be folded to -1.
2075 .. code-block:: llvm
2077 %A = select undef, %X, %Y
2078 %B = select undef, 42, %Y
2079 %C = select %X, %Y, undef
2089 This set of examples shows that undefined '``select``' (and conditional
2090 branch) conditions can go *either way*, but they have to come from one
2091 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2092 both known to have a clear low bit, then ``%A`` would have to have a
2093 cleared low bit. However, in the ``%C`` example, the optimizer is
2094 allowed to assume that the '``undef``' operand could be the same as
2095 ``%Y``, allowing the whole '``select``' to be eliminated.
2097 .. code-block:: llvm
2099 %A = xor undef, undef
2116 This example points out that two '``undef``' operands are not
2117 necessarily the same. This can be surprising to people (and also matches
2118 C semantics) where they assume that "``X^X``" is always zero, even if
2119 ``X`` is undefined. This isn't true for a number of reasons, but the
2120 short answer is that an '``undef``' "variable" can arbitrarily change
2121 its value over its "live range". This is true because the variable
2122 doesn't actually *have a live range*. Instead, the value is logically
2123 read from arbitrary registers that happen to be around when needed, so
2124 the value is not necessarily consistent over time. In fact, ``%A`` and
2125 ``%C`` need to have the same semantics or the core LLVM "replace all
2126 uses with" concept would not hold.
2128 .. code-block:: llvm
2136 These examples show the crucial difference between an *undefined value*
2137 and *undefined behavior*. An undefined value (like '``undef``') is
2138 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2139 operation can be constant folded to '``undef``', because the '``undef``'
2140 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2141 However, in the second example, we can make a more aggressive
2142 assumption: because the ``undef`` is allowed to be an arbitrary value,
2143 we are allowed to assume that it could be zero. Since a divide by zero
2144 has *undefined behavior*, we are allowed to assume that the operation
2145 does not execute at all. This allows us to delete the divide and all
2146 code after it. Because the undefined operation "can't happen", the
2147 optimizer can assume that it occurs in dead code.
2149 .. code-block:: llvm
2151 a: store undef -> %X
2152 b: store %X -> undef
2157 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2158 value can be assumed to not have any effect; we can assume that the
2159 value is overwritten with bits that happen to match what was already
2160 there. However, a store *to* an undefined location could clobber
2161 arbitrary memory, therefore, it has undefined behavior.
2168 Poison values are similar to :ref:`undef values <undefvalues>`, however
2169 they also represent the fact that an instruction or constant expression
2170 which cannot evoke side effects has nevertheless detected a condition
2171 which results in undefined behavior.
2173 There is currently no way of representing a poison value in the IR; they
2174 only exist when produced by operations such as :ref:`add <i_add>` with
2177 Poison value behavior is defined in terms of value *dependence*:
2179 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2180 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2181 their dynamic predecessor basic block.
2182 - Function arguments depend on the corresponding actual argument values
2183 in the dynamic callers of their functions.
2184 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2185 instructions that dynamically transfer control back to them.
2186 - :ref:`Invoke <i_invoke>` instructions depend on the
2187 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2188 call instructions that dynamically transfer control back to them.
2189 - Non-volatile loads and stores depend on the most recent stores to all
2190 of the referenced memory addresses, following the order in the IR
2191 (including loads and stores implied by intrinsics such as
2192 :ref:`@llvm.memcpy <int_memcpy>`.)
2193 - An instruction with externally visible side effects depends on the
2194 most recent preceding instruction with externally visible side
2195 effects, following the order in the IR. (This includes :ref:`volatile
2196 operations <volatile>`.)
2197 - An instruction *control-depends* on a :ref:`terminator
2198 instruction <terminators>` if the terminator instruction has
2199 multiple successors and the instruction is always executed when
2200 control transfers to one of the successors, and may not be executed
2201 when control is transferred to another.
2202 - Additionally, an instruction also *control-depends* on a terminator
2203 instruction if the set of instructions it otherwise depends on would
2204 be different if the terminator had transferred control to a different
2206 - Dependence is transitive.
2208 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2209 with the additional affect that any instruction which has a *dependence*
2210 on a poison value has undefined behavior.
2212 Here are some examples:
2214 .. code-block:: llvm
2217 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2218 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2219 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2220 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2222 store i32 %poison, i32* @g ; Poison value stored to memory.
2223 %poison2 = load i32* @g ; Poison value loaded back from memory.
2225 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2227 %narrowaddr = bitcast i32* @g to i16*
2228 %wideaddr = bitcast i32* @g to i64*
2229 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2230 %poison4 = load i64* %wideaddr ; Returns a poison value.
2232 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2233 br i1 %cmp, label %true, label %end ; Branch to either destination.
2236 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2237 ; it has undefined behavior.
2241 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2242 ; Both edges into this PHI are
2243 ; control-dependent on %cmp, so this
2244 ; always results in a poison value.
2246 store volatile i32 0, i32* @g ; This would depend on the store in %true
2247 ; if %cmp is true, or the store in %entry
2248 ; otherwise, so this is undefined behavior.
2250 br i1 %cmp, label %second_true, label %second_end
2251 ; The same branch again, but this time the
2252 ; true block doesn't have side effects.
2259 store volatile i32 0, i32* @g ; This time, the instruction always depends
2260 ; on the store in %end. Also, it is
2261 ; control-equivalent to %end, so this is
2262 ; well-defined (ignoring earlier undefined
2263 ; behavior in this example).
2267 Addresses of Basic Blocks
2268 -------------------------
2270 ``blockaddress(@function, %block)``
2272 The '``blockaddress``' constant computes the address of the specified
2273 basic block in the specified function, and always has an ``i8*`` type.
2274 Taking the address of the entry block is illegal.
2276 This value only has defined behavior when used as an operand to the
2277 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2278 against null. Pointer equality tests between labels addresses results in
2279 undefined behavior --- though, again, comparison against null is ok, and
2280 no label is equal to the null pointer. This may be passed around as an
2281 opaque pointer sized value as long as the bits are not inspected. This
2282 allows ``ptrtoint`` and arithmetic to be performed on these values so
2283 long as the original value is reconstituted before the ``indirectbr``
2286 Finally, some targets may provide defined semantics when using the value
2287 as the operand to an inline assembly, but that is target specific.
2291 Constant Expressions
2292 --------------------
2294 Constant expressions are used to allow expressions involving other
2295 constants to be used as constants. Constant expressions may be of any
2296 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2297 that does not have side effects (e.g. load and call are not supported).
2298 The following is the syntax for constant expressions:
2300 ``trunc (CST to TYPE)``
2301 Truncate a constant to another type. The bit size of CST must be
2302 larger than the bit size of TYPE. Both types must be integers.
2303 ``zext (CST to TYPE)``
2304 Zero extend a constant to another type. The bit size of CST must be
2305 smaller than the bit size of TYPE. Both types must be integers.
2306 ``sext (CST to TYPE)``
2307 Sign extend a constant to another type. The bit size of CST must be
2308 smaller than the bit size of TYPE. Both types must be integers.
2309 ``fptrunc (CST to TYPE)``
2310 Truncate a floating point constant to another floating point type.
2311 The size of CST must be larger than the size of TYPE. Both types
2312 must be floating point.
2313 ``fpext (CST to TYPE)``
2314 Floating point extend a constant to another type. The size of CST
2315 must be smaller or equal to the size of TYPE. Both types must be
2317 ``fptoui (CST to TYPE)``
2318 Convert a floating point constant to the corresponding unsigned
2319 integer constant. TYPE must be a scalar or vector integer type. CST
2320 must be of scalar or vector floating point type. Both CST and TYPE
2321 must be scalars, or vectors of the same number of elements. If the
2322 value won't fit in the integer type, the results are undefined.
2323 ``fptosi (CST to TYPE)``
2324 Convert a floating point constant to the corresponding signed
2325 integer constant. TYPE must be a scalar or vector integer type. CST
2326 must be of scalar or vector floating point type. Both CST and TYPE
2327 must be scalars, or vectors of the same number of elements. If the
2328 value won't fit in the integer type, the results are undefined.
2329 ``uitofp (CST to TYPE)``
2330 Convert an unsigned integer constant to the corresponding floating
2331 point constant. TYPE must be a scalar or vector floating point type.
2332 CST must be of scalar or vector integer type. Both CST and TYPE must
2333 be scalars, or vectors of the same number of elements. If the value
2334 won't fit in the floating point type, the results are undefined.
2335 ``sitofp (CST to TYPE)``
2336 Convert a signed integer constant to the corresponding floating
2337 point constant. TYPE must be a scalar or vector floating point type.
2338 CST must be of scalar or vector integer type. Both CST and TYPE must
2339 be scalars, or vectors of the same number of elements. If the value
2340 won't fit in the floating point type, the results are undefined.
2341 ``ptrtoint (CST to TYPE)``
2342 Convert a pointer typed constant to the corresponding integer
2343 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2344 pointer type. The ``CST`` value is zero extended, truncated, or
2345 unchanged to make it fit in ``TYPE``.
2346 ``inttoptr (CST to TYPE)``
2347 Convert an integer constant to a pointer constant. TYPE must be a
2348 pointer type. CST must be of integer type. The CST value is zero
2349 extended, truncated, or unchanged to make it fit in a pointer size.
2350 This one is *really* dangerous!
2351 ``bitcast (CST to TYPE)``
2352 Convert a constant, CST, to another TYPE. The constraints of the
2353 operands are the same as those for the :ref:`bitcast
2354 instruction <i_bitcast>`.
2355 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2356 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2357 constants. As with the :ref:`getelementptr <i_getelementptr>`
2358 instruction, the index list may have zero or more indexes, which are
2359 required to make sense for the type of "CSTPTR".
2360 ``select (COND, VAL1, VAL2)``
2361 Perform the :ref:`select operation <i_select>` on constants.
2362 ``icmp COND (VAL1, VAL2)``
2363 Performs the :ref:`icmp operation <i_icmp>` on constants.
2364 ``fcmp COND (VAL1, VAL2)``
2365 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2366 ``extractelement (VAL, IDX)``
2367 Perform the :ref:`extractelement operation <i_extractelement>` on
2369 ``insertelement (VAL, ELT, IDX)``
2370 Perform the :ref:`insertelement operation <i_insertelement>` on
2372 ``shufflevector (VEC1, VEC2, IDXMASK)``
2373 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2375 ``extractvalue (VAL, IDX0, IDX1, ...)``
2376 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2377 constants. The index list is interpreted in a similar manner as
2378 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2379 least one index value must be specified.
2380 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2381 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2382 The index list is interpreted in a similar manner as indices in a
2383 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2384 value must be specified.
2385 ``OPCODE (LHS, RHS)``
2386 Perform the specified operation of the LHS and RHS constants. OPCODE
2387 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2388 binary <bitwiseops>` operations. The constraints on operands are
2389 the same as those for the corresponding instruction (e.g. no bitwise
2390 operations on floating point values are allowed).
2397 Inline Assembler Expressions
2398 ----------------------------
2400 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2401 Inline Assembly <moduleasm>`) through the use of a special value. This
2402 value represents the inline assembler as a string (containing the
2403 instructions to emit), a list of operand constraints (stored as a
2404 string), a flag that indicates whether or not the inline asm expression
2405 has side effects, and a flag indicating whether the function containing
2406 the asm needs to align its stack conservatively. An example inline
2407 assembler expression is:
2409 .. code-block:: llvm
2411 i32 (i32) asm "bswap $0", "=r,r"
2413 Inline assembler expressions may **only** be used as the callee operand
2414 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2415 Thus, typically we have:
2417 .. code-block:: llvm
2419 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2421 Inline asms with side effects not visible in the constraint list must be
2422 marked as having side effects. This is done through the use of the
2423 '``sideeffect``' keyword, like so:
2425 .. code-block:: llvm
2427 call void asm sideeffect "eieio", ""()
2429 In some cases inline asms will contain code that will not work unless
2430 the stack is aligned in some way, such as calls or SSE instructions on
2431 x86, yet will not contain code that does that alignment within the asm.
2432 The compiler should make conservative assumptions about what the asm
2433 might contain and should generate its usual stack alignment code in the
2434 prologue if the '``alignstack``' keyword is present:
2436 .. code-block:: llvm
2438 call void asm alignstack "eieio", ""()
2440 Inline asms also support using non-standard assembly dialects. The
2441 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2442 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2443 the only supported dialects. An example is:
2445 .. code-block:: llvm
2447 call void asm inteldialect "eieio", ""()
2449 If multiple keywords appear the '``sideeffect``' keyword must come
2450 first, the '``alignstack``' keyword second and the '``inteldialect``'
2456 The call instructions that wrap inline asm nodes may have a
2457 "``!srcloc``" MDNode attached to it that contains a list of constant
2458 integers. If present, the code generator will use the integer as the
2459 location cookie value when report errors through the ``LLVMContext``
2460 error reporting mechanisms. This allows a front-end to correlate backend
2461 errors that occur with inline asm back to the source code that produced
2464 .. code-block:: llvm
2466 call void asm sideeffect "something bad", ""(), !srcloc !42
2468 !42 = !{ i32 1234567 }
2470 It is up to the front-end to make sense of the magic numbers it places
2471 in the IR. If the MDNode contains multiple constants, the code generator
2472 will use the one that corresponds to the line of the asm that the error
2477 Metadata Nodes and Metadata Strings
2478 -----------------------------------
2480 LLVM IR allows metadata to be attached to instructions in the program
2481 that can convey extra information about the code to the optimizers and
2482 code generator. One example application of metadata is source-level
2483 debug information. There are two metadata primitives: strings and nodes.
2484 All metadata has the ``metadata`` type and is identified in syntax by a
2485 preceding exclamation point ('``!``').
2487 A metadata string is a string surrounded by double quotes. It can
2488 contain any character by escaping non-printable characters with
2489 "``\xx``" where "``xx``" is the two digit hex code. For example:
2492 Metadata nodes are represented with notation similar to structure
2493 constants (a comma separated list of elements, surrounded by braces and
2494 preceded by an exclamation point). Metadata nodes can have any values as
2495 their operand. For example:
2497 .. code-block:: llvm
2499 !{ metadata !"test\00", i32 10}
2501 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2502 metadata nodes, which can be looked up in the module symbol table. For
2505 .. code-block:: llvm
2507 !foo = metadata !{!4, !3}
2509 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2510 function is using two metadata arguments:
2512 .. code-block:: llvm
2514 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2516 Metadata can be attached with an instruction. Here metadata ``!21`` is
2517 attached to the ``add`` instruction using the ``!dbg`` identifier:
2519 .. code-block:: llvm
2521 %indvar.next = add i64 %indvar, 1, !dbg !21
2523 More information about specific metadata nodes recognized by the
2524 optimizers and code generator is found below.
2529 In LLVM IR, memory does not have types, so LLVM's own type system is not
2530 suitable for doing TBAA. Instead, metadata is added to the IR to
2531 describe a type system of a higher level language. This can be used to
2532 implement typical C/C++ TBAA, but it can also be used to implement
2533 custom alias analysis behavior for other languages.
2535 The current metadata format is very simple. TBAA metadata nodes have up
2536 to three fields, e.g.:
2538 .. code-block:: llvm
2540 !0 = metadata !{ metadata !"an example type tree" }
2541 !1 = metadata !{ metadata !"int", metadata !0 }
2542 !2 = metadata !{ metadata !"float", metadata !0 }
2543 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2545 The first field is an identity field. It can be any value, usually a
2546 metadata string, which uniquely identifies the type. The most important
2547 name in the tree is the name of the root node. Two trees with different
2548 root node names are entirely disjoint, even if they have leaves with
2551 The second field identifies the type's parent node in the tree, or is
2552 null or omitted for a root node. A type is considered to alias all of
2553 its descendants and all of its ancestors in the tree. Also, a type is
2554 considered to alias all types in other trees, so that bitcode produced
2555 from multiple front-ends is handled conservatively.
2557 If the third field is present, it's an integer which if equal to 1
2558 indicates that the type is "constant" (meaning
2559 ``pointsToConstantMemory`` should return true; see `other useful
2560 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2562 '``tbaa.struct``' Metadata
2563 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2565 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2566 aggregate assignment operations in C and similar languages, however it
2567 is defined to copy a contiguous region of memory, which is more than
2568 strictly necessary for aggregate types which contain holes due to
2569 padding. Also, it doesn't contain any TBAA information about the fields
2572 ``!tbaa.struct`` metadata can describe which memory subregions in a
2573 memcpy are padding and what the TBAA tags of the struct are.
2575 The current metadata format is very simple. ``!tbaa.struct`` metadata
2576 nodes are a list of operands which are in conceptual groups of three.
2577 For each group of three, the first operand gives the byte offset of a
2578 field in bytes, the second gives its size in bytes, and the third gives
2581 .. code-block:: llvm
2583 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2585 This describes a struct with two fields. The first is at offset 0 bytes
2586 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2587 and has size 4 bytes and has tbaa tag !2.
2589 Note that the fields need not be contiguous. In this example, there is a
2590 4 byte gap between the two fields. This gap represents padding which
2591 does not carry useful data and need not be preserved.
2593 '``fpmath``' Metadata
2594 ^^^^^^^^^^^^^^^^^^^^^
2596 ``fpmath`` metadata may be attached to any instruction of floating point
2597 type. It can be used to express the maximum acceptable error in the
2598 result of that instruction, in ULPs, thus potentially allowing the
2599 compiler to use a more efficient but less accurate method of computing
2600 it. ULP is defined as follows:
2602 If ``x`` is a real number that lies between two finite consecutive
2603 floating-point numbers ``a`` and ``b``, without being equal to one
2604 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2605 distance between the two non-equal finite floating-point numbers
2606 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2608 The metadata node shall consist of a single positive floating point
2609 number representing the maximum relative error, for example:
2611 .. code-block:: llvm
2613 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2615 '``range``' Metadata
2616 ^^^^^^^^^^^^^^^^^^^^
2618 ``range`` metadata may be attached only to loads of integer types. It
2619 expresses the possible ranges the loaded value is in. The ranges are
2620 represented with a flattened list of integers. The loaded value is known
2621 to be in the union of the ranges defined by each consecutive pair. Each
2622 pair has the following properties:
2624 - The type must match the type loaded by the instruction.
2625 - The pair ``a,b`` represents the range ``[a,b)``.
2626 - Both ``a`` and ``b`` are constants.
2627 - The range is allowed to wrap.
2628 - The range should not represent the full or empty set. That is,
2631 In addition, the pairs must be in signed order of the lower bound and
2632 they must be non-contiguous.
2636 .. code-block:: llvm
2638 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2639 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2640 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2641 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2643 !0 = metadata !{ i8 0, i8 2 }
2644 !1 = metadata !{ i8 255, i8 2 }
2645 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2646 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2651 It is sometimes useful to attach information to loop constructs. Currently,
2652 loop metadata is implemented as metadata attached to the branch instruction
2653 in the loop latch block. This type of metadata refer to a metadata node that is
2654 guaranteed to be separate for each loop. The loop identifier metadata is
2655 specified with the name ``llvm.loop``.
2657 The loop identifier metadata is implemented using a metadata that refers to
2658 itself to avoid merging it with any other identifier metadata, e.g.,
2659 during module linkage or function inlining. That is, each loop should refer
2660 to their own identification metadata even if they reside in separate functions.
2661 The following example contains loop identifier metadata for two separate loop
2664 .. code-block:: llvm
2666 !0 = metadata !{ metadata !0 }
2667 !1 = metadata !{ metadata !1 }
2669 The loop identifier metadata can be used to specify additional per-loop
2670 metadata. Any operands after the first operand can be treated as user-defined
2671 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2672 by the loop vectorizer to indicate how many times to unroll the loop:
2674 .. code-block:: llvm
2676 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2678 !0 = metadata !{ metadata !0, metadata !1 }
2679 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2684 Metadata types used to annotate memory accesses with information helpful
2685 for optimizations are prefixed with ``llvm.mem``.
2687 '``llvm.mem.parallel_loop_access``' Metadata
2688 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2690 For a loop to be parallel, in addition to using
2691 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2692 also all of the memory accessing instructions in the loop body need to be
2693 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2694 is at least one memory accessing instruction not marked with the metadata,
2695 the loop must be considered a sequential loop. This causes parallel loops to be
2696 converted to sequential loops due to optimization passes that are unaware of
2697 the parallel semantics and that insert new memory instructions to the loop
2700 Example of a loop that is considered parallel due to its correct use of
2701 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2702 metadata types that refer to the same loop identifier metadata.
2704 .. code-block:: llvm
2708 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2710 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2712 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2716 !0 = metadata !{ metadata !0 }
2718 It is also possible to have nested parallel loops. In that case the
2719 memory accesses refer to a list of loop identifier metadata nodes instead of
2720 the loop identifier metadata node directly:
2722 .. code-block:: llvm
2729 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2731 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2733 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2737 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2739 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2741 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2743 outer.for.end: ; preds = %for.body
2745 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2746 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2747 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2749 '``llvm.vectorizer``'
2750 ^^^^^^^^^^^^^^^^^^^^^
2752 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2753 vectorization parameters such as vectorization factor and unroll factor.
2755 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2756 loop identification metadata.
2758 '``llvm.vectorizer.unroll``' Metadata
2759 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2761 This metadata instructs the loop vectorizer to unroll the specified
2762 loop exactly ``N`` times.
2764 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2765 operand is an integer specifying the unroll factor. For example:
2767 .. code-block:: llvm
2769 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2771 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2774 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2775 determined automatically.
2777 '``llvm.vectorizer.width``' Metadata
2778 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2780 This metadata sets the target width of the vectorizer to ``N``. Without
2781 this metadata, the vectorizer will choose a width automatically.
2782 Regardless of this metadata, the vectorizer will only vectorize loops if
2783 it believes it is valid to do so.
2785 The first operand is the string ``llvm.vectorizer.width`` and the second
2786 operand is an integer specifying the width. For example:
2788 .. code-block:: llvm
2790 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2792 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2795 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2798 Module Flags Metadata
2799 =====================
2801 Information about the module as a whole is difficult to convey to LLVM's
2802 subsystems. The LLVM IR isn't sufficient to transmit this information.
2803 The ``llvm.module.flags`` named metadata exists in order to facilitate
2804 this. These flags are in the form of key / value pairs --- much like a
2805 dictionary --- making it easy for any subsystem who cares about a flag to
2808 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2809 Each triplet has the following form:
2811 - The first element is a *behavior* flag, which specifies the behavior
2812 when two (or more) modules are merged together, and it encounters two
2813 (or more) metadata with the same ID. The supported behaviors are
2815 - The second element is a metadata string that is a unique ID for the
2816 metadata. Each module may only have one flag entry for each unique ID (not
2817 including entries with the **Require** behavior).
2818 - The third element is the value of the flag.
2820 When two (or more) modules are merged together, the resulting
2821 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2822 each unique metadata ID string, there will be exactly one entry in the merged
2823 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2824 be determined by the merge behavior flag, as described below. The only exception
2825 is that entries with the *Require* behavior are always preserved.
2827 The following behaviors are supported:
2838 Emits an error if two values disagree, otherwise the resulting value
2839 is that of the operands.
2843 Emits a warning if two values disagree. The result value will be the
2844 operand for the flag from the first module being linked.
2848 Adds a requirement that another module flag be present and have a
2849 specified value after linking is performed. The value must be a
2850 metadata pair, where the first element of the pair is the ID of the
2851 module flag to be restricted, and the second element of the pair is
2852 the value the module flag should be restricted to. This behavior can
2853 be used to restrict the allowable results (via triggering of an
2854 error) of linking IDs with the **Override** behavior.
2858 Uses the specified value, regardless of the behavior or value of the
2859 other module. If both modules specify **Override**, but the values
2860 differ, an error will be emitted.
2864 Appends the two values, which are required to be metadata nodes.
2868 Appends the two values, which are required to be metadata
2869 nodes. However, duplicate entries in the second list are dropped
2870 during the append operation.
2872 It is an error for a particular unique flag ID to have multiple behaviors,
2873 except in the case of **Require** (which adds restrictions on another metadata
2874 value) or **Override**.
2876 An example of module flags:
2878 .. code-block:: llvm
2880 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2881 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2882 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2883 !3 = metadata !{ i32 3, metadata !"qux",
2885 metadata !"foo", i32 1
2888 !llvm.module.flags = !{ !0, !1, !2, !3 }
2890 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2891 if two or more ``!"foo"`` flags are seen is to emit an error if their
2892 values are not equal.
2894 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2895 behavior if two or more ``!"bar"`` flags are seen is to use the value
2898 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2899 behavior if two or more ``!"qux"`` flags are seen is to emit a
2900 warning if their values are not equal.
2902 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2906 metadata !{ metadata !"foo", i32 1 }
2908 The behavior is to emit an error if the ``llvm.module.flags`` does not
2909 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2912 Objective-C Garbage Collection Module Flags Metadata
2913 ----------------------------------------------------
2915 On the Mach-O platform, Objective-C stores metadata about garbage
2916 collection in a special section called "image info". The metadata
2917 consists of a version number and a bitmask specifying what types of
2918 garbage collection are supported (if any) by the file. If two or more
2919 modules are linked together their garbage collection metadata needs to
2920 be merged rather than appended together.
2922 The Objective-C garbage collection module flags metadata consists of the
2923 following key-value pairs:
2932 * - ``Objective-C Version``
2933 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2935 * - ``Objective-C Image Info Version``
2936 - **[Required]** --- The version of the image info section. Currently
2939 * - ``Objective-C Image Info Section``
2940 - **[Required]** --- The section to place the metadata. Valid values are
2941 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2942 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2943 Objective-C ABI version 2.
2945 * - ``Objective-C Garbage Collection``
2946 - **[Required]** --- Specifies whether garbage collection is supported or
2947 not. Valid values are 0, for no garbage collection, and 2, for garbage
2948 collection supported.
2950 * - ``Objective-C GC Only``
2951 - **[Optional]** --- Specifies that only garbage collection is supported.
2952 If present, its value must be 6. This flag requires that the
2953 ``Objective-C Garbage Collection`` flag have the value 2.
2955 Some important flag interactions:
2957 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2958 merged with a module with ``Objective-C Garbage Collection`` set to
2959 2, then the resulting module has the
2960 ``Objective-C Garbage Collection`` flag set to 0.
2961 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2962 merged with a module with ``Objective-C GC Only`` set to 6.
2964 Automatic Linker Flags Module Flags Metadata
2965 --------------------------------------------
2967 Some targets support embedding flags to the linker inside individual object
2968 files. Typically this is used in conjunction with language extensions which
2969 allow source files to explicitly declare the libraries they depend on, and have
2970 these automatically be transmitted to the linker via object files.
2972 These flags are encoded in the IR using metadata in the module flags section,
2973 using the ``Linker Options`` key. The merge behavior for this flag is required
2974 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2975 node which should be a list of other metadata nodes, each of which should be a
2976 list of metadata strings defining linker options.
2978 For example, the following metadata section specifies two separate sets of
2979 linker options, presumably to link against ``libz`` and the ``Cocoa``
2982 !0 = metadata !{ i32 6, metadata !"Linker Options",
2984 metadata !{ metadata !"-lz" },
2985 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2986 !llvm.module.flags = !{ !0 }
2988 The metadata encoding as lists of lists of options, as opposed to a collapsed
2989 list of options, is chosen so that the IR encoding can use multiple option
2990 strings to specify e.g., a single library, while still having that specifier be
2991 preserved as an atomic element that can be recognized by a target specific
2992 assembly writer or object file emitter.
2994 Each individual option is required to be either a valid option for the target's
2995 linker, or an option that is reserved by the target specific assembly writer or
2996 object file emitter. No other aspect of these options is defined by the IR.
2998 .. _intrinsicglobalvariables:
3000 Intrinsic Global Variables
3001 ==========================
3003 LLVM has a number of "magic" global variables that contain data that
3004 affect code generation or other IR semantics. These are documented here.
3005 All globals of this sort should have a section specified as
3006 "``llvm.metadata``". This section and all globals that start with
3007 "``llvm.``" are reserved for use by LLVM.
3011 The '``llvm.used``' Global Variable
3012 -----------------------------------
3014 The ``@llvm.used`` global is an array which has
3015 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3016 pointers to named global variables, functions and aliases which may optionally
3017 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3020 .. code-block:: llvm
3025 @llvm.used = appending global [2 x i8*] [
3027 i8* bitcast (i32* @Y to i8*)
3028 ], section "llvm.metadata"
3030 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3031 and linker are required to treat the symbol as if there is a reference to the
3032 symbol that it cannot see (which is why they have to be named). For example, if
3033 a variable has internal linkage and no references other than that from the
3034 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3035 references from inline asms and other things the compiler cannot "see", and
3036 corresponds to "``attribute((used))``" in GNU C.
3038 On some targets, the code generator must emit a directive to the
3039 assembler or object file to prevent the assembler and linker from
3040 molesting the symbol.
3042 .. _gv_llvmcompilerused:
3044 The '``llvm.compiler.used``' Global Variable
3045 --------------------------------------------
3047 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3048 directive, except that it only prevents the compiler from touching the
3049 symbol. On targets that support it, this allows an intelligent linker to
3050 optimize references to the symbol without being impeded as it would be
3053 This is a rare construct that should only be used in rare circumstances,
3054 and should not be exposed to source languages.
3056 .. _gv_llvmglobalctors:
3058 The '``llvm.global_ctors``' Global Variable
3059 -------------------------------------------
3061 .. code-block:: llvm
3063 %0 = type { i32, void ()* }
3064 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3066 The ``@llvm.global_ctors`` array contains a list of constructor
3067 functions and associated priorities. The functions referenced by this
3068 array will be called in ascending order of priority (i.e. lowest first)
3069 when the module is loaded. The order of functions with the same priority
3072 .. _llvmglobaldtors:
3074 The '``llvm.global_dtors``' Global Variable
3075 -------------------------------------------
3077 .. code-block:: llvm
3079 %0 = type { i32, void ()* }
3080 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3082 The ``@llvm.global_dtors`` array contains a list of destructor functions
3083 and associated priorities. The functions referenced by this array will
3084 be called in descending order of priority (i.e. highest first) when the
3085 module is loaded. The order of functions with the same priority is not
3088 Instruction Reference
3089 =====================
3091 The LLVM instruction set consists of several different classifications
3092 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3093 instructions <binaryops>`, :ref:`bitwise binary
3094 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3095 :ref:`other instructions <otherops>`.
3099 Terminator Instructions
3100 -----------------------
3102 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3103 program ends with a "Terminator" instruction, which indicates which
3104 block should be executed after the current block is finished. These
3105 terminator instructions typically yield a '``void``' value: they produce
3106 control flow, not values (the one exception being the
3107 ':ref:`invoke <i_invoke>`' instruction).
3109 The terminator instructions are: ':ref:`ret <i_ret>`',
3110 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3111 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3112 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3116 '``ret``' Instruction
3117 ^^^^^^^^^^^^^^^^^^^^^
3124 ret <type> <value> ; Return a value from a non-void function
3125 ret void ; Return from void function
3130 The '``ret``' instruction is used to return control flow (and optionally
3131 a value) from a function back to the caller.
3133 There are two forms of the '``ret``' instruction: one that returns a
3134 value and then causes control flow, and one that just causes control
3140 The '``ret``' instruction optionally accepts a single argument, the
3141 return value. The type of the return value must be a ':ref:`first
3142 class <t_firstclass>`' type.
3144 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3145 return type and contains a '``ret``' instruction with no return value or
3146 a return value with a type that does not match its type, or if it has a
3147 void return type and contains a '``ret``' instruction with a return
3153 When the '``ret``' instruction is executed, control flow returns back to
3154 the calling function's context. If the caller is a
3155 ":ref:`call <i_call>`" instruction, execution continues at the
3156 instruction after the call. If the caller was an
3157 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3158 beginning of the "normal" destination block. If the instruction returns
3159 a value, that value shall set the call or invoke instruction's return
3165 .. code-block:: llvm
3167 ret i32 5 ; Return an integer value of 5
3168 ret void ; Return from a void function
3169 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3173 '``br``' Instruction
3174 ^^^^^^^^^^^^^^^^^^^^
3181 br i1 <cond>, label <iftrue>, label <iffalse>
3182 br label <dest> ; Unconditional branch
3187 The '``br``' instruction is used to cause control flow to transfer to a
3188 different basic block in the current function. There are two forms of
3189 this instruction, corresponding to a conditional branch and an
3190 unconditional branch.
3195 The conditional branch form of the '``br``' instruction takes a single
3196 '``i1``' value and two '``label``' values. The unconditional form of the
3197 '``br``' instruction takes a single '``label``' value as a target.
3202 Upon execution of a conditional '``br``' instruction, the '``i1``'
3203 argument is evaluated. If the value is ``true``, control flows to the
3204 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3205 to the '``iffalse``' ``label`` argument.
3210 .. code-block:: llvm
3213 %cond = icmp eq i32 %a, %b
3214 br i1 %cond, label %IfEqual, label %IfUnequal
3222 '``switch``' Instruction
3223 ^^^^^^^^^^^^^^^^^^^^^^^^
3230 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3235 The '``switch``' instruction is used to transfer control flow to one of
3236 several different places. It is a generalization of the '``br``'
3237 instruction, allowing a branch to occur to one of many possible
3243 The '``switch``' instruction uses three parameters: an integer
3244 comparison value '``value``', a default '``label``' destination, and an
3245 array of pairs of comparison value constants and '``label``'s. The table
3246 is not allowed to contain duplicate constant entries.
3251 The ``switch`` instruction specifies a table of values and destinations.
3252 When the '``switch``' instruction is executed, this table is searched
3253 for the given value. If the value is found, control flow is transferred
3254 to the corresponding destination; otherwise, control flow is transferred
3255 to the default destination.
3260 Depending on properties of the target machine and the particular
3261 ``switch`` instruction, this instruction may be code generated in
3262 different ways. For example, it could be generated as a series of
3263 chained conditional branches or with a lookup table.
3268 .. code-block:: llvm
3270 ; Emulate a conditional br instruction
3271 %Val = zext i1 %value to i32
3272 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3274 ; Emulate an unconditional br instruction
3275 switch i32 0, label %dest [ ]
3277 ; Implement a jump table:
3278 switch i32 %val, label %otherwise [ i32 0, label %onzero
3280 i32 2, label %ontwo ]
3284 '``indirectbr``' Instruction
3285 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3292 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3297 The '``indirectbr``' instruction implements an indirect branch to a
3298 label within the current function, whose address is specified by
3299 "``address``". Address must be derived from a
3300 :ref:`blockaddress <blockaddress>` constant.
3305 The '``address``' argument is the address of the label to jump to. The
3306 rest of the arguments indicate the full set of possible destinations
3307 that the address may point to. Blocks are allowed to occur multiple
3308 times in the destination list, though this isn't particularly useful.
3310 This destination list is required so that dataflow analysis has an
3311 accurate understanding of the CFG.
3316 Control transfers to the block specified in the address argument. All
3317 possible destination blocks must be listed in the label list, otherwise
3318 this instruction has undefined behavior. This implies that jumps to
3319 labels defined in other functions have undefined behavior as well.
3324 This is typically implemented with a jump through a register.
3329 .. code-block:: llvm
3331 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3335 '``invoke``' Instruction
3336 ^^^^^^^^^^^^^^^^^^^^^^^^
3343 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3344 to label <normal label> unwind label <exception label>
3349 The '``invoke``' instruction causes control to transfer to a specified
3350 function, with the possibility of control flow transfer to either the
3351 '``normal``' label or the '``exception``' label. If the callee function
3352 returns with the "``ret``" instruction, control flow will return to the
3353 "normal" label. If the callee (or any indirect callees) returns via the
3354 ":ref:`resume <i_resume>`" instruction or other exception handling
3355 mechanism, control is interrupted and continued at the dynamically
3356 nearest "exception" label.
3358 The '``exception``' label is a `landing
3359 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3360 '``exception``' label is required to have the
3361 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3362 information about the behavior of the program after unwinding happens,
3363 as its first non-PHI instruction. The restrictions on the
3364 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3365 instruction, so that the important information contained within the
3366 "``landingpad``" instruction can't be lost through normal code motion.
3371 This instruction requires several arguments:
3373 #. The optional "cconv" marker indicates which :ref:`calling
3374 convention <callingconv>` the call should use. If none is
3375 specified, the call defaults to using C calling conventions.
3376 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3377 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3379 #. '``ptr to function ty``': shall be the signature of the pointer to
3380 function value being invoked. In most cases, this is a direct
3381 function invocation, but indirect ``invoke``'s are just as possible,
3382 branching off an arbitrary pointer to function value.
3383 #. '``function ptr val``': An LLVM value containing a pointer to a
3384 function to be invoked.
3385 #. '``function args``': argument list whose types match the function
3386 signature argument types and parameter attributes. All arguments must
3387 be of :ref:`first class <t_firstclass>` type. If the function signature
3388 indicates the function accepts a variable number of arguments, the
3389 extra arguments can be specified.
3390 #. '``normal label``': the label reached when the called function
3391 executes a '``ret``' instruction.
3392 #. '``exception label``': the label reached when a callee returns via
3393 the :ref:`resume <i_resume>` instruction or other exception handling
3395 #. The optional :ref:`function attributes <fnattrs>` list. Only
3396 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3397 attributes are valid here.
3402 This instruction is designed to operate as a standard '``call``'
3403 instruction in most regards. The primary difference is that it
3404 establishes an association with a label, which is used by the runtime
3405 library to unwind the stack.
3407 This instruction is used in languages with destructors to ensure that
3408 proper cleanup is performed in the case of either a ``longjmp`` or a
3409 thrown exception. Additionally, this is important for implementation of
3410 '``catch``' clauses in high-level languages that support them.
3412 For the purposes of the SSA form, the definition of the value returned
3413 by the '``invoke``' instruction is deemed to occur on the edge from the
3414 current block to the "normal" label. If the callee unwinds then no
3415 return value is available.
3420 .. code-block:: llvm
3422 %retval = invoke i32 @Test(i32 15) to label %Continue
3423 unwind label %TestCleanup ; {i32}:retval set
3424 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3425 unwind label %TestCleanup ; {i32}:retval set
3429 '``resume``' Instruction
3430 ^^^^^^^^^^^^^^^^^^^^^^^^
3437 resume <type> <value>
3442 The '``resume``' instruction is a terminator instruction that has no
3448 The '``resume``' instruction requires one argument, which must have the
3449 same type as the result of any '``landingpad``' instruction in the same
3455 The '``resume``' instruction resumes propagation of an existing
3456 (in-flight) exception whose unwinding was interrupted with a
3457 :ref:`landingpad <i_landingpad>` instruction.
3462 .. code-block:: llvm
3464 resume { i8*, i32 } %exn
3468 '``unreachable``' Instruction
3469 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3481 The '``unreachable``' instruction has no defined semantics. This
3482 instruction is used to inform the optimizer that a particular portion of
3483 the code is not reachable. This can be used to indicate that the code
3484 after a no-return function cannot be reached, and other facts.
3489 The '``unreachable``' instruction has no defined semantics.
3496 Binary operators are used to do most of the computation in a program.
3497 They require two operands of the same type, execute an operation on
3498 them, and produce a single value. The operands might represent multiple
3499 data, as is the case with the :ref:`vector <t_vector>` data type. The
3500 result value has the same type as its operands.
3502 There are several different binary operators:
3506 '``add``' Instruction
3507 ^^^^^^^^^^^^^^^^^^^^^
3514 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3515 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3516 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3517 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3522 The '``add``' instruction returns the sum of its two operands.
3527 The two arguments to the '``add``' instruction must be
3528 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3529 arguments must have identical types.
3534 The value produced is the integer sum of the two operands.
3536 If the sum has unsigned overflow, the result returned is the
3537 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3540 Because LLVM integers use a two's complement representation, this
3541 instruction is appropriate for both signed and unsigned integers.
3543 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3544 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3545 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3546 unsigned and/or signed overflow, respectively, occurs.
3551 .. code-block:: llvm
3553 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3557 '``fadd``' Instruction
3558 ^^^^^^^^^^^^^^^^^^^^^^
3565 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3570 The '``fadd``' instruction returns the sum of its two operands.
3575 The two arguments to the '``fadd``' instruction must be :ref:`floating
3576 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3577 Both arguments must have identical types.
3582 The value produced is the floating point sum of the two operands. This
3583 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3584 which are optimization hints to enable otherwise unsafe floating point
3590 .. code-block:: llvm
3592 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3594 '``sub``' Instruction
3595 ^^^^^^^^^^^^^^^^^^^^^
3602 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3603 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3604 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3605 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3610 The '``sub``' instruction returns the difference of its two operands.
3612 Note that the '``sub``' instruction is used to represent the '``neg``'
3613 instruction present in most other intermediate representations.
3618 The two arguments to the '``sub``' instruction must be
3619 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3620 arguments must have identical types.
3625 The value produced is the integer difference of the two operands.
3627 If the difference has unsigned overflow, the result returned is the
3628 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3631 Because LLVM integers use a two's complement representation, this
3632 instruction is appropriate for both signed and unsigned integers.
3634 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3635 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3636 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3637 unsigned and/or signed overflow, respectively, occurs.
3642 .. code-block:: llvm
3644 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3645 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3649 '``fsub``' Instruction
3650 ^^^^^^^^^^^^^^^^^^^^^^
3657 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3662 The '``fsub``' instruction returns the difference of its two operands.
3664 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3665 instruction present in most other intermediate representations.
3670 The two arguments to the '``fsub``' instruction must be :ref:`floating
3671 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3672 Both arguments must have identical types.
3677 The value produced is the floating point difference of the two operands.
3678 This instruction can also take any number of :ref:`fast-math
3679 flags <fastmath>`, which are optimization hints to enable otherwise
3680 unsafe floating point optimizations:
3685 .. code-block:: llvm
3687 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3688 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3690 '``mul``' Instruction
3691 ^^^^^^^^^^^^^^^^^^^^^
3698 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3699 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3700 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3701 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3706 The '``mul``' instruction returns the product of its two operands.
3711 The two arguments to the '``mul``' instruction must be
3712 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3713 arguments must have identical types.
3718 The value produced is the integer product of the two operands.
3720 If the result of the multiplication has unsigned overflow, the result
3721 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3722 bit width of the result.
3724 Because LLVM integers use a two's complement representation, and the
3725 result is the same width as the operands, this instruction returns the
3726 correct result for both signed and unsigned integers. If a full product
3727 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3728 sign-extended or zero-extended as appropriate to the width of the full
3731 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3732 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3733 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3734 unsigned and/or signed overflow, respectively, occurs.
3739 .. code-block:: llvm
3741 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3745 '``fmul``' Instruction
3746 ^^^^^^^^^^^^^^^^^^^^^^
3753 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3758 The '``fmul``' instruction returns the product of its two operands.
3763 The two arguments to the '``fmul``' instruction must be :ref:`floating
3764 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3765 Both arguments must have identical types.
3770 The value produced is the floating point product of the two operands.
3771 This instruction can also take any number of :ref:`fast-math
3772 flags <fastmath>`, which are optimization hints to enable otherwise
3773 unsafe floating point optimizations:
3778 .. code-block:: llvm
3780 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3782 '``udiv``' Instruction
3783 ^^^^^^^^^^^^^^^^^^^^^^
3790 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3791 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3796 The '``udiv``' instruction returns the quotient of its two operands.
3801 The two arguments to the '``udiv``' instruction must be
3802 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3803 arguments must have identical types.
3808 The value produced is the unsigned integer quotient of the two operands.
3810 Note that unsigned integer division and signed integer division are
3811 distinct operations; for signed integer division, use '``sdiv``'.
3813 Division by zero leads to undefined behavior.
3815 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3816 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3817 such, "((a udiv exact b) mul b) == a").
3822 .. code-block:: llvm
3824 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3826 '``sdiv``' Instruction
3827 ^^^^^^^^^^^^^^^^^^^^^^
3834 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3835 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3840 The '``sdiv``' instruction returns the quotient of its two operands.
3845 The two arguments to the '``sdiv``' instruction must be
3846 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3847 arguments must have identical types.
3852 The value produced is the signed integer quotient of the two operands
3853 rounded towards zero.
3855 Note that signed integer division and unsigned integer division are
3856 distinct operations; for unsigned integer division, use '``udiv``'.
3858 Division by zero leads to undefined behavior. Overflow also leads to
3859 undefined behavior; this is a rare case, but can occur, for example, by
3860 doing a 32-bit division of -2147483648 by -1.
3862 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3863 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3868 .. code-block:: llvm
3870 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3874 '``fdiv``' Instruction
3875 ^^^^^^^^^^^^^^^^^^^^^^
3882 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3887 The '``fdiv``' instruction returns the quotient of its two operands.
3892 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3893 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3894 Both arguments must have identical types.
3899 The value produced is the floating point quotient of the two operands.
3900 This instruction can also take any number of :ref:`fast-math
3901 flags <fastmath>`, which are optimization hints to enable otherwise
3902 unsafe floating point optimizations:
3907 .. code-block:: llvm
3909 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3911 '``urem``' Instruction
3912 ^^^^^^^^^^^^^^^^^^^^^^
3919 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3924 The '``urem``' instruction returns the remainder from the unsigned
3925 division of its two arguments.
3930 The two arguments to the '``urem``' instruction must be
3931 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3932 arguments must have identical types.
3937 This instruction returns the unsigned integer *remainder* of a division.
3938 This instruction always performs an unsigned division to get the
3941 Note that unsigned integer remainder and signed integer remainder are
3942 distinct operations; for signed integer remainder, use '``srem``'.
3944 Taking the remainder of a division by zero leads to undefined behavior.
3949 .. code-block:: llvm
3951 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3953 '``srem``' Instruction
3954 ^^^^^^^^^^^^^^^^^^^^^^
3961 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3966 The '``srem``' instruction returns the remainder from the signed
3967 division of its two operands. This instruction can also take
3968 :ref:`vector <t_vector>` versions of the values in which case the elements
3974 The two arguments to the '``srem``' instruction must be
3975 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3976 arguments must have identical types.
3981 This instruction returns the *remainder* of a division (where the result
3982 is either zero or has the same sign as the dividend, ``op1``), not the
3983 *modulo* operator (where the result is either zero or has the same sign
3984 as the divisor, ``op2``) of a value. For more information about the
3985 difference, see `The Math
3986 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3987 table of how this is implemented in various languages, please see
3989 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3991 Note that signed integer remainder and unsigned integer remainder are
3992 distinct operations; for unsigned integer remainder, use '``urem``'.
3994 Taking the remainder of a division by zero leads to undefined behavior.
3995 Overflow also leads to undefined behavior; this is a rare case, but can
3996 occur, for example, by taking the remainder of a 32-bit division of
3997 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3998 rule lets srem be implemented using instructions that return both the
3999 result of the division and the remainder.)
4004 .. code-block:: llvm
4006 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4010 '``frem``' Instruction
4011 ^^^^^^^^^^^^^^^^^^^^^^
4018 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4023 The '``frem``' instruction returns the remainder from the division of
4029 The two arguments to the '``frem``' instruction must be :ref:`floating
4030 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4031 Both arguments must have identical types.
4036 This instruction returns the *remainder* of a division. The remainder
4037 has the same sign as the dividend. This instruction can also take any
4038 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4039 to enable otherwise unsafe floating point optimizations:
4044 .. code-block:: llvm
4046 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4050 Bitwise Binary Operations
4051 -------------------------
4053 Bitwise binary operators are used to do various forms of bit-twiddling
4054 in a program. They are generally very efficient instructions and can
4055 commonly be strength reduced from other instructions. They require two
4056 operands of the same type, execute an operation on them, and produce a
4057 single value. The resulting value is the same type as its operands.
4059 '``shl``' Instruction
4060 ^^^^^^^^^^^^^^^^^^^^^
4067 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4068 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4069 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4070 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4075 The '``shl``' instruction returns the first operand shifted to the left
4076 a specified number of bits.
4081 Both arguments to the '``shl``' instruction must be the same
4082 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4083 '``op2``' is treated as an unsigned value.
4088 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4089 where ``n`` is the width of the result. If ``op2`` is (statically or
4090 dynamically) negative or equal to or larger than the number of bits in
4091 ``op1``, the result is undefined. If the arguments are vectors, each
4092 vector element of ``op1`` is shifted by the corresponding shift amount
4095 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4096 value <poisonvalues>` if it shifts out any non-zero bits. If the
4097 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4098 value <poisonvalues>` if it shifts out any bits that disagree with the
4099 resultant sign bit. As such, NUW/NSW have the same semantics as they
4100 would if the shift were expressed as a mul instruction with the same
4101 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4106 .. code-block:: llvm
4108 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4109 <result> = shl i32 4, 2 ; yields {i32}: 16
4110 <result> = shl i32 1, 10 ; yields {i32}: 1024
4111 <result> = shl i32 1, 32 ; undefined
4112 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4114 '``lshr``' Instruction
4115 ^^^^^^^^^^^^^^^^^^^^^^
4122 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4123 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4128 The '``lshr``' instruction (logical shift right) returns the first
4129 operand shifted to the right a specified number of bits with zero fill.
4134 Both arguments to the '``lshr``' instruction must be the same
4135 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4136 '``op2``' is treated as an unsigned value.
4141 This instruction always performs a logical shift right operation. The
4142 most significant bits of the result will be filled with zero bits after
4143 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4144 than the number of bits in ``op1``, the result is undefined. If the
4145 arguments are vectors, each vector element of ``op1`` is shifted by the
4146 corresponding shift amount in ``op2``.
4148 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4149 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4155 .. code-block:: llvm
4157 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4158 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4159 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4160 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4161 <result> = lshr i32 1, 32 ; undefined
4162 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4164 '``ashr``' Instruction
4165 ^^^^^^^^^^^^^^^^^^^^^^
4172 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4173 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4178 The '``ashr``' instruction (arithmetic shift right) returns the first
4179 operand shifted to the right a specified number of bits with sign
4185 Both arguments to the '``ashr``' instruction must be the same
4186 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4187 '``op2``' is treated as an unsigned value.
4192 This instruction always performs an arithmetic shift right operation,
4193 The most significant bits of the result will be filled with the sign bit
4194 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4195 than the number of bits in ``op1``, the result is undefined. If the
4196 arguments are vectors, each vector element of ``op1`` is shifted by the
4197 corresponding shift amount in ``op2``.
4199 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4200 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4206 .. code-block:: llvm
4208 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4209 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4210 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4211 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4212 <result> = ashr i32 1, 32 ; undefined
4213 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4215 '``and``' Instruction
4216 ^^^^^^^^^^^^^^^^^^^^^
4223 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4228 The '``and``' instruction returns the bitwise logical and of its two
4234 The two arguments to the '``and``' instruction must be
4235 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4236 arguments must have identical types.
4241 The truth table used for the '``and``' instruction is:
4258 .. code-block:: llvm
4260 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4261 <result> = and i32 15, 40 ; yields {i32}:result = 8
4262 <result> = and i32 4, 8 ; yields {i32}:result = 0
4264 '``or``' Instruction
4265 ^^^^^^^^^^^^^^^^^^^^
4272 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4277 The '``or``' instruction returns the bitwise logical inclusive or of its
4283 The two arguments to the '``or``' instruction must be
4284 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4285 arguments must have identical types.
4290 The truth table used for the '``or``' instruction is:
4309 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4310 <result> = or i32 15, 40 ; yields {i32}:result = 47
4311 <result> = or i32 4, 8 ; yields {i32}:result = 12
4313 '``xor``' Instruction
4314 ^^^^^^^^^^^^^^^^^^^^^
4321 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4326 The '``xor``' instruction returns the bitwise logical exclusive or of
4327 its two operands. The ``xor`` is used to implement the "one's
4328 complement" operation, which is the "~" operator in C.
4333 The two arguments to the '``xor``' instruction must be
4334 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4335 arguments must have identical types.
4340 The truth table used for the '``xor``' instruction is:
4357 .. code-block:: llvm
4359 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4360 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4361 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4362 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4367 LLVM supports several instructions to represent vector operations in a
4368 target-independent manner. These instructions cover the element-access
4369 and vector-specific operations needed to process vectors effectively.
4370 While LLVM does directly support these vector operations, many
4371 sophisticated algorithms will want to use target-specific intrinsics to
4372 take full advantage of a specific target.
4374 .. _i_extractelement:
4376 '``extractelement``' Instruction
4377 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4384 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4389 The '``extractelement``' instruction extracts a single scalar element
4390 from a vector at a specified index.
4395 The first operand of an '``extractelement``' instruction is a value of
4396 :ref:`vector <t_vector>` type. The second operand is an index indicating
4397 the position from which to extract the element. The index may be a
4403 The result is a scalar of the same type as the element type of ``val``.
4404 Its value is the value at position ``idx`` of ``val``. If ``idx``
4405 exceeds the length of ``val``, the results are undefined.
4410 .. code-block:: llvm
4412 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4414 .. _i_insertelement:
4416 '``insertelement``' Instruction
4417 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4424 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4429 The '``insertelement``' instruction inserts a scalar element into a
4430 vector at a specified index.
4435 The first operand of an '``insertelement``' instruction is a value of
4436 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4437 type must equal the element type of the first operand. The third operand
4438 is an index indicating the position at which to insert the value. The
4439 index may be a variable.
4444 The result is a vector of the same type as ``val``. Its element values
4445 are those of ``val`` except at position ``idx``, where it gets the value
4446 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4452 .. code-block:: llvm
4454 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4456 .. _i_shufflevector:
4458 '``shufflevector``' Instruction
4459 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4466 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4471 The '``shufflevector``' instruction constructs a permutation of elements
4472 from two input vectors, returning a vector with the same element type as
4473 the input and length that is the same as the shuffle mask.
4478 The first two operands of a '``shufflevector``' instruction are vectors
4479 with the same type. The third argument is a shuffle mask whose element
4480 type is always 'i32'. The result of the instruction is a vector whose
4481 length is the same as the shuffle mask and whose element type is the
4482 same as the element type of the first two operands.
4484 The shuffle mask operand is required to be a constant vector with either
4485 constant integer or undef values.
4490 The elements of the two input vectors are numbered from left to right
4491 across both of the vectors. The shuffle mask operand specifies, for each
4492 element of the result vector, which element of the two input vectors the
4493 result element gets. The element selector may be undef (meaning "don't
4494 care") and the second operand may be undef if performing a shuffle from
4500 .. code-block:: llvm
4502 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4503 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4504 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4505 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4506 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4507 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4508 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4509 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4511 Aggregate Operations
4512 --------------------
4514 LLVM supports several instructions for working with
4515 :ref:`aggregate <t_aggregate>` values.
4519 '``extractvalue``' Instruction
4520 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4527 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4532 The '``extractvalue``' instruction extracts the value of a member field
4533 from an :ref:`aggregate <t_aggregate>` value.
4538 The first operand of an '``extractvalue``' instruction is a value of
4539 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4540 constant indices to specify which value to extract in a similar manner
4541 as indices in a '``getelementptr``' instruction.
4543 The major differences to ``getelementptr`` indexing are:
4545 - Since the value being indexed is not a pointer, the first index is
4546 omitted and assumed to be zero.
4547 - At least one index must be specified.
4548 - Not only struct indices but also array indices must be in bounds.
4553 The result is the value at the position in the aggregate specified by
4559 .. code-block:: llvm
4561 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4565 '``insertvalue``' Instruction
4566 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4573 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4578 The '``insertvalue``' instruction inserts a value into a member field in
4579 an :ref:`aggregate <t_aggregate>` value.
4584 The first operand of an '``insertvalue``' instruction is a value of
4585 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4586 a first-class value to insert. The following operands are constant
4587 indices indicating the position at which to insert the value in a
4588 similar manner as indices in a '``extractvalue``' instruction. The value
4589 to insert must have the same type as the value identified by the
4595 The result is an aggregate of the same type as ``val``. Its value is
4596 that of ``val`` except that the value at the position specified by the
4597 indices is that of ``elt``.
4602 .. code-block:: llvm
4604 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4605 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4606 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4610 Memory Access and Addressing Operations
4611 ---------------------------------------
4613 A key design point of an SSA-based representation is how it represents
4614 memory. In LLVM, no memory locations are in SSA form, which makes things
4615 very simple. This section describes how to read, write, and allocate
4620 '``alloca``' Instruction
4621 ^^^^^^^^^^^^^^^^^^^^^^^^
4628 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4633 The '``alloca``' instruction allocates memory on the stack frame of the
4634 currently executing function, to be automatically released when this
4635 function returns to its caller. The object is always allocated in the
4636 generic address space (address space zero).
4641 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4642 bytes of memory on the runtime stack, returning a pointer of the
4643 appropriate type to the program. If "NumElements" is specified, it is
4644 the number of elements allocated, otherwise "NumElements" is defaulted
4645 to be one. If a constant alignment is specified, the value result of the
4646 allocation is guaranteed to be aligned to at least that boundary. If not
4647 specified, or if zero, the target can choose to align the allocation on
4648 any convenient boundary compatible with the type.
4650 '``type``' may be any sized type.
4655 Memory is allocated; a pointer is returned. The operation is undefined
4656 if there is insufficient stack space for the allocation. '``alloca``'d
4657 memory is automatically released when the function returns. The
4658 '``alloca``' instruction is commonly used to represent automatic
4659 variables that must have an address available. When the function returns
4660 (either with the ``ret`` or ``resume`` instructions), the memory is
4661 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4662 The order in which memory is allocated (ie., which way the stack grows)
4668 .. code-block:: llvm
4670 %ptr = alloca i32 ; yields {i32*}:ptr
4671 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4672 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4673 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4677 '``load``' Instruction
4678 ^^^^^^^^^^^^^^^^^^^^^^
4685 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4686 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4687 !<index> = !{ i32 1 }
4692 The '``load``' instruction is used to read from memory.
4697 The argument to the ``load`` instruction specifies the memory address
4698 from which to load. The pointer must point to a :ref:`first
4699 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4700 then the optimizer is not allowed to modify the number or order of
4701 execution of this ``load`` with other :ref:`volatile
4702 operations <volatile>`.
4704 If the ``load`` is marked as ``atomic``, it takes an extra
4705 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4706 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4707 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4708 when they may see multiple atomic stores. The type of the pointee must
4709 be an integer type whose bit width is a power of two greater than or
4710 equal to eight and less than or equal to a target-specific size limit.
4711 ``align`` must be explicitly specified on atomic loads, and the load has
4712 undefined behavior if the alignment is not set to a value which is at
4713 least the size in bytes of the pointee. ``!nontemporal`` does not have
4714 any defined semantics for atomic loads.
4716 The optional constant ``align`` argument specifies the alignment of the
4717 operation (that is, the alignment of the memory address). A value of 0
4718 or an omitted ``align`` argument means that the operation has the ABI
4719 alignment for the target. It is the responsibility of the code emitter
4720 to ensure that the alignment information is correct. Overestimating the
4721 alignment results in undefined behavior. Underestimating the alignment
4722 may produce less efficient code. An alignment of 1 is always safe.
4724 The optional ``!nontemporal`` metadata must reference a single
4725 metadata name ``<index>`` corresponding to a metadata node with one
4726 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4727 metadata on the instruction tells the optimizer and code generator
4728 that this load is not expected to be reused in the cache. The code
4729 generator may select special instructions to save cache bandwidth, such
4730 as the ``MOVNT`` instruction on x86.
4732 The optional ``!invariant.load`` metadata must reference a single
4733 metadata name ``<index>`` corresponding to a metadata node with no
4734 entries. The existence of the ``!invariant.load`` metadata on the
4735 instruction tells the optimizer and code generator that this load
4736 address points to memory which does not change value during program
4737 execution. The optimizer may then move this load around, for example, by
4738 hoisting it out of loops using loop invariant code motion.
4743 The location of memory pointed to is loaded. If the value being loaded
4744 is of scalar type then the number of bytes read does not exceed the
4745 minimum number of bytes needed to hold all bits of the type. For
4746 example, loading an ``i24`` reads at most three bytes. When loading a
4747 value of a type like ``i20`` with a size that is not an integral number
4748 of bytes, the result is undefined if the value was not originally
4749 written using a store of the same type.
4754 .. code-block:: llvm
4756 %ptr = alloca i32 ; yields {i32*}:ptr
4757 store i32 3, i32* %ptr ; yields {void}
4758 %val = load i32* %ptr ; yields {i32}:val = i32 3
4762 '``store``' Instruction
4763 ^^^^^^^^^^^^^^^^^^^^^^^
4770 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4771 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4776 The '``store``' instruction is used to write to memory.
4781 There are two arguments to the ``store`` instruction: a value to store
4782 and an address at which to store it. The type of the ``<pointer>``
4783 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4784 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4785 then the optimizer is not allowed to modify the number or order of
4786 execution of this ``store`` with other :ref:`volatile
4787 operations <volatile>`.
4789 If the ``store`` is marked as ``atomic``, it takes an extra
4790 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4791 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4792 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4793 when they may see multiple atomic stores. The type of the pointee must
4794 be an integer type whose bit width is a power of two greater than or
4795 equal to eight and less than or equal to a target-specific size limit.
4796 ``align`` must be explicitly specified on atomic stores, and the store
4797 has undefined behavior if the alignment is not set to a value which is
4798 at least the size in bytes of the pointee. ``!nontemporal`` does not
4799 have any defined semantics for atomic stores.
4801 The optional constant ``align`` argument specifies the alignment of the
4802 operation (that is, the alignment of the memory address). A value of 0
4803 or an omitted ``align`` argument means that the operation has the ABI
4804 alignment for the target. It is the responsibility of the code emitter
4805 to ensure that the alignment information is correct. Overestimating the
4806 alignment results in undefined behavior. Underestimating the
4807 alignment may produce less efficient code. An alignment of 1 is always
4810 The optional ``!nontemporal`` metadata must reference a single metadata
4811 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4812 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4813 tells the optimizer and code generator that this load is not expected to
4814 be reused in the cache. The code generator may select special
4815 instructions to save cache bandwidth, such as the MOVNT instruction on
4821 The contents of memory are updated to contain ``<value>`` at the
4822 location specified by the ``<pointer>`` operand. If ``<value>`` is
4823 of scalar type then the number of bytes written does not exceed the
4824 minimum number of bytes needed to hold all bits of the type. For
4825 example, storing an ``i24`` writes at most three bytes. When writing a
4826 value of a type like ``i20`` with a size that is not an integral number
4827 of bytes, it is unspecified what happens to the extra bits that do not
4828 belong to the type, but they will typically be overwritten.
4833 .. code-block:: llvm
4835 %ptr = alloca i32 ; yields {i32*}:ptr
4836 store i32 3, i32* %ptr ; yields {void}
4837 %val = load i32* %ptr ; yields {i32}:val = i32 3
4841 '``fence``' Instruction
4842 ^^^^^^^^^^^^^^^^^^^^^^^
4849 fence [singlethread] <ordering> ; yields {void}
4854 The '``fence``' instruction is used to introduce happens-before edges
4860 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4861 defines what *synchronizes-with* edges they add. They can only be given
4862 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4867 A fence A which has (at least) ``release`` ordering semantics
4868 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4869 semantics if and only if there exist atomic operations X and Y, both
4870 operating on some atomic object M, such that A is sequenced before X, X
4871 modifies M (either directly or through some side effect of a sequence
4872 headed by X), Y is sequenced before B, and Y observes M. This provides a
4873 *happens-before* dependency between A and B. Rather than an explicit
4874 ``fence``, one (but not both) of the atomic operations X or Y might
4875 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4876 still *synchronize-with* the explicit ``fence`` and establish the
4877 *happens-before* edge.
4879 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4880 ``acquire`` and ``release`` semantics specified above, participates in
4881 the global program order of other ``seq_cst`` operations and/or fences.
4883 The optional ":ref:`singlethread <singlethread>`" argument specifies
4884 that the fence only synchronizes with other fences in the same thread.
4885 (This is useful for interacting with signal handlers.)
4890 .. code-block:: llvm
4892 fence acquire ; yields {void}
4893 fence singlethread seq_cst ; yields {void}
4897 '``cmpxchg``' Instruction
4898 ^^^^^^^^^^^^^^^^^^^^^^^^^
4905 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4910 The '``cmpxchg``' instruction is used to atomically modify memory. It
4911 loads a value in memory and compares it to a given value. If they are
4912 equal, it stores a new value into the memory.
4917 There are three arguments to the '``cmpxchg``' instruction: an address
4918 to operate on, a value to compare to the value currently be at that
4919 address, and a new value to place at that address if the compared values
4920 are equal. The type of '<cmp>' must be an integer type whose bit width
4921 is a power of two greater than or equal to eight and less than or equal
4922 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4923 type, and the type of '<pointer>' must be a pointer to that type. If the
4924 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4925 to modify the number or order of execution of this ``cmpxchg`` with
4926 other :ref:`volatile operations <volatile>`.
4928 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4929 synchronizes with other atomic operations.
4931 The optional "``singlethread``" argument declares that the ``cmpxchg``
4932 is only atomic with respect to code (usually signal handlers) running in
4933 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4934 respect to all other code in the system.
4936 The pointer passed into cmpxchg must have alignment greater than or
4937 equal to the size in memory of the operand.
4942 The contents of memory at the location specified by the '``<pointer>``'
4943 operand is read and compared to '``<cmp>``'; if the read value is the
4944 equal, '``<new>``' is written. The original value at the location is
4947 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4948 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4949 atomic load with an ordering parameter determined by dropping any
4950 ``release`` part of the ``cmpxchg``'s ordering.
4955 .. code-block:: llvm
4958 %orig = atomic load i32* %ptr unordered ; yields {i32}
4962 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4963 %squared = mul i32 %cmp, %cmp
4964 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4965 %success = icmp eq i32 %cmp, %old
4966 br i1 %success, label %done, label %loop
4973 '``atomicrmw``' Instruction
4974 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4981 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4986 The '``atomicrmw``' instruction is used to atomically modify memory.
4991 There are three arguments to the '``atomicrmw``' instruction: an
4992 operation to apply, an address whose value to modify, an argument to the
4993 operation. The operation must be one of the following keywords:
5007 The type of '<value>' must be an integer type whose bit width is a power
5008 of two greater than or equal to eight and less than or equal to a
5009 target-specific size limit. The type of the '``<pointer>``' operand must
5010 be a pointer to that type. If the ``atomicrmw`` is marked as
5011 ``volatile``, then the optimizer is not allowed to modify the number or
5012 order of execution of this ``atomicrmw`` with other :ref:`volatile
5013 operations <volatile>`.
5018 The contents of memory at the location specified by the '``<pointer>``'
5019 operand are atomically read, modified, and written back. The original
5020 value at the location is returned. The modification is specified by the
5023 - xchg: ``*ptr = val``
5024 - add: ``*ptr = *ptr + val``
5025 - sub: ``*ptr = *ptr - val``
5026 - and: ``*ptr = *ptr & val``
5027 - nand: ``*ptr = ~(*ptr & val)``
5028 - or: ``*ptr = *ptr | val``
5029 - xor: ``*ptr = *ptr ^ val``
5030 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5031 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5032 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5034 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5040 .. code-block:: llvm
5042 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5044 .. _i_getelementptr:
5046 '``getelementptr``' Instruction
5047 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5054 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5055 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5056 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5061 The '``getelementptr``' instruction is used to get the address of a
5062 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5063 address calculation only and does not access memory.
5068 The first argument is always a pointer or a vector of pointers, and
5069 forms the basis of the calculation. The remaining arguments are indices
5070 that indicate which of the elements of the aggregate object are indexed.
5071 The interpretation of each index is dependent on the type being indexed
5072 into. The first index always indexes the pointer value given as the
5073 first argument, the second index indexes a value of the type pointed to
5074 (not necessarily the value directly pointed to, since the first index
5075 can be non-zero), etc. The first type indexed into must be a pointer
5076 value, subsequent types can be arrays, vectors, and structs. Note that
5077 subsequent types being indexed into can never be pointers, since that
5078 would require loading the pointer before continuing calculation.
5080 The type of each index argument depends on the type it is indexing into.
5081 When indexing into a (optionally packed) structure, only ``i32`` integer
5082 **constants** are allowed (when using a vector of indices they must all
5083 be the **same** ``i32`` integer constant). When indexing into an array,
5084 pointer or vector, integers of any width are allowed, and they are not
5085 required to be constant. These integers are treated as signed values
5088 For example, let's consider a C code fragment and how it gets compiled
5104 int *foo(struct ST *s) {
5105 return &s[1].Z.B[5][13];
5108 The LLVM code generated by Clang is:
5110 .. code-block:: llvm
5112 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5113 %struct.ST = type { i32, double, %struct.RT }
5115 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5117 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5124 In the example above, the first index is indexing into the
5125 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5126 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5127 indexes into the third element of the structure, yielding a
5128 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5129 structure. The third index indexes into the second element of the
5130 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5131 dimensions of the array are subscripted into, yielding an '``i32``'
5132 type. The '``getelementptr``' instruction returns a pointer to this
5133 element, thus computing a value of '``i32*``' type.
5135 Note that it is perfectly legal to index partially through a structure,
5136 returning a pointer to an inner element. Because of this, the LLVM code
5137 for the given testcase is equivalent to:
5139 .. code-block:: llvm
5141 define i32* @foo(%struct.ST* %s) {
5142 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5143 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5144 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5145 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5146 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5150 If the ``inbounds`` keyword is present, the result value of the
5151 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5152 pointer is not an *in bounds* address of an allocated object, or if any
5153 of the addresses that would be formed by successive addition of the
5154 offsets implied by the indices to the base address with infinitely
5155 precise signed arithmetic are not an *in bounds* address of that
5156 allocated object. The *in bounds* addresses for an allocated object are
5157 all the addresses that point into the object, plus the address one byte
5158 past the end. In cases where the base is a vector of pointers the
5159 ``inbounds`` keyword applies to each of the computations element-wise.
5161 If the ``inbounds`` keyword is not present, the offsets are added to the
5162 base address with silently-wrapping two's complement arithmetic. If the
5163 offsets have a different width from the pointer, they are sign-extended
5164 or truncated to the width of the pointer. The result value of the
5165 ``getelementptr`` may be outside the object pointed to by the base
5166 pointer. The result value may not necessarily be used to access memory
5167 though, even if it happens to point into allocated storage. See the
5168 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5171 The getelementptr instruction is often confusing. For some more insight
5172 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5177 .. code-block:: llvm
5179 ; yields [12 x i8]*:aptr
5180 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5182 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5184 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5186 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5188 In cases where the pointer argument is a vector of pointers, each index
5189 must be a vector with the same number of elements. For example:
5191 .. code-block:: llvm
5193 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5195 Conversion Operations
5196 ---------------------
5198 The instructions in this category are the conversion instructions
5199 (casting) which all take a single operand and a type. They perform
5200 various bit conversions on the operand.
5202 '``trunc .. to``' Instruction
5203 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5210 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5215 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5220 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5221 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5222 of the same number of integers. The bit size of the ``value`` must be
5223 larger than the bit size of the destination type, ``ty2``. Equal sized
5224 types are not allowed.
5229 The '``trunc``' instruction truncates the high order bits in ``value``
5230 and converts the remaining bits to ``ty2``. Since the source size must
5231 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5232 It will always truncate bits.
5237 .. code-block:: llvm
5239 %X = trunc i32 257 to i8 ; yields i8:1
5240 %Y = trunc i32 123 to i1 ; yields i1:true
5241 %Z = trunc i32 122 to i1 ; yields i1:false
5242 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5244 '``zext .. to``' Instruction
5245 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5252 <result> = zext <ty> <value> to <ty2> ; yields ty2
5257 The '``zext``' instruction zero extends its operand to type ``ty2``.
5262 The '``zext``' instruction takes a value to cast, and a type to cast it
5263 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5264 the same number of integers. The bit size of the ``value`` must be
5265 smaller than the bit size of the destination type, ``ty2``.
5270 The ``zext`` fills the high order bits of the ``value`` with zero bits
5271 until it reaches the size of the destination type, ``ty2``.
5273 When zero extending from i1, the result will always be either 0 or 1.
5278 .. code-block:: llvm
5280 %X = zext i32 257 to i64 ; yields i64:257
5281 %Y = zext i1 true to i32 ; yields i32:1
5282 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5284 '``sext .. to``' Instruction
5285 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5292 <result> = sext <ty> <value> to <ty2> ; yields ty2
5297 The '``sext``' sign extends ``value`` to the type ``ty2``.
5302 The '``sext``' instruction takes a value to cast, and a type to cast it
5303 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5304 the same number of integers. The bit size of the ``value`` must be
5305 smaller than the bit size of the destination type, ``ty2``.
5310 The '``sext``' instruction performs a sign extension by copying the sign
5311 bit (highest order bit) of the ``value`` until it reaches the bit size
5312 of the type ``ty2``.
5314 When sign extending from i1, the extension always results in -1 or 0.
5319 .. code-block:: llvm
5321 %X = sext i8 -1 to i16 ; yields i16 :65535
5322 %Y = sext i1 true to i32 ; yields i32:-1
5323 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5325 '``fptrunc .. to``' Instruction
5326 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5333 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5338 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5343 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5344 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5345 The size of ``value`` must be larger than the size of ``ty2``. This
5346 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5351 The '``fptrunc``' instruction truncates a ``value`` from a larger
5352 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5353 point <t_floating>` type. If the value cannot fit within the
5354 destination type, ``ty2``, then the results are undefined.
5359 .. code-block:: llvm
5361 %X = fptrunc double 123.0 to float ; yields float:123.0
5362 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5364 '``fpext .. to``' Instruction
5365 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5372 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5377 The '``fpext``' extends a floating point ``value`` to a larger floating
5383 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5384 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5385 to. The source type must be smaller than the destination type.
5390 The '``fpext``' instruction extends the ``value`` from a smaller
5391 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5392 point <t_floating>` type. The ``fpext`` cannot be used to make a
5393 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5394 *no-op cast* for a floating point cast.
5399 .. code-block:: llvm
5401 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5402 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5404 '``fptoui .. to``' Instruction
5405 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5412 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5417 The '``fptoui``' converts a floating point ``value`` to its unsigned
5418 integer equivalent of type ``ty2``.
5423 The '``fptoui``' instruction takes a value to cast, which must be a
5424 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5425 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5426 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5427 type with the same number of elements as ``ty``
5432 The '``fptoui``' instruction converts its :ref:`floating
5433 point <t_floating>` operand into the nearest (rounding towards zero)
5434 unsigned integer value. If the value cannot fit in ``ty2``, the results
5440 .. code-block:: llvm
5442 %X = fptoui double 123.0 to i32 ; yields i32:123
5443 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5444 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5446 '``fptosi .. to``' Instruction
5447 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5454 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5459 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5460 ``value`` to type ``ty2``.
5465 The '``fptosi``' instruction takes a value to cast, which must be a
5466 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5467 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5468 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5469 type with the same number of elements as ``ty``
5474 The '``fptosi``' instruction converts its :ref:`floating
5475 point <t_floating>` operand into the nearest (rounding towards zero)
5476 signed integer value. If the value cannot fit in ``ty2``, the results
5482 .. code-block:: llvm
5484 %X = fptosi double -123.0 to i32 ; yields i32:-123
5485 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5486 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5488 '``uitofp .. to``' Instruction
5489 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5496 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5501 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5502 and converts that value to the ``ty2`` type.
5507 The '``uitofp``' instruction takes a value to cast, which must be a
5508 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5509 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5510 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5511 type with the same number of elements as ``ty``
5516 The '``uitofp``' instruction interprets its operand as an unsigned
5517 integer quantity and converts it to the corresponding floating point
5518 value. If the value cannot fit in the floating point value, the results
5524 .. code-block:: llvm
5526 %X = uitofp i32 257 to float ; yields float:257.0
5527 %Y = uitofp i8 -1 to double ; yields double:255.0
5529 '``sitofp .. to``' Instruction
5530 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5537 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5542 The '``sitofp``' instruction regards ``value`` as a signed integer and
5543 converts that value to the ``ty2`` type.
5548 The '``sitofp``' instruction takes a value to cast, which must be a
5549 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5550 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5551 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5552 type with the same number of elements as ``ty``
5557 The '``sitofp``' instruction interprets its operand as a signed integer
5558 quantity and converts it to the corresponding floating point value. If
5559 the value cannot fit in the floating point value, the results are
5565 .. code-block:: llvm
5567 %X = sitofp i32 257 to float ; yields float:257.0
5568 %Y = sitofp i8 -1 to double ; yields double:-1.0
5572 '``ptrtoint .. to``' Instruction
5573 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5580 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5585 The '``ptrtoint``' instruction converts the pointer or a vector of
5586 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5591 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5592 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5593 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5594 a vector of integers type.
5599 The '``ptrtoint``' instruction converts ``value`` to integer type
5600 ``ty2`` by interpreting the pointer value as an integer and either
5601 truncating or zero extending that value to the size of the integer type.
5602 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5603 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5604 the same size, then nothing is done (*no-op cast*) other than a type
5610 .. code-block:: llvm
5612 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5613 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5614 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5618 '``inttoptr .. to``' Instruction
5619 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5626 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5631 The '``inttoptr``' instruction converts an integer ``value`` to a
5632 pointer type, ``ty2``.
5637 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5638 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5644 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5645 applying either a zero extension or a truncation depending on the size
5646 of the integer ``value``. If ``value`` is larger than the size of a
5647 pointer then a truncation is done. If ``value`` is smaller than the size
5648 of a pointer then a zero extension is done. If they are the same size,
5649 nothing is done (*no-op cast*).
5654 .. code-block:: llvm
5656 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5657 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5658 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5659 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5663 '``bitcast .. to``' Instruction
5664 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5671 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5676 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5682 The '``bitcast``' instruction takes a value to cast, which must be a
5683 non-aggregate first class value, and a type to cast it to, which must
5684 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5685 bit sizes of ``value`` and the destination type, ``ty2``, must be
5686 identical. If the source type is a pointer, the destination type must
5687 also be a pointer of the same size. This instruction supports bitwise
5688 conversion of vectors to integers and to vectors of other types (as
5689 long as they have the same size).
5694 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5695 is always a *no-op cast* because no bits change with this
5696 conversion. The conversion is done as if the ``value`` had been stored
5697 to memory and read back as type ``ty2``. Pointer (or vector of
5698 pointers) types may only be converted to other pointer (or vector of
5699 pointers) types with this instruction if the pointer sizes are
5700 equal. To convert pointers to other types, use the :ref:`inttoptr
5701 <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5706 .. code-block:: llvm
5708 %X = bitcast i8 255 to i8 ; yields i8 :-1
5709 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5710 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5711 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5718 The instructions in this category are the "miscellaneous" instructions,
5719 which defy better classification.
5723 '``icmp``' Instruction
5724 ^^^^^^^^^^^^^^^^^^^^^^
5731 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5736 The '``icmp``' instruction returns a boolean value or a vector of
5737 boolean values based on comparison of its two integer, integer vector,
5738 pointer, or pointer vector operands.
5743 The '``icmp``' instruction takes three operands. The first operand is
5744 the condition code indicating the kind of comparison to perform. It is
5745 not a value, just a keyword. The possible condition code are:
5748 #. ``ne``: not equal
5749 #. ``ugt``: unsigned greater than
5750 #. ``uge``: unsigned greater or equal
5751 #. ``ult``: unsigned less than
5752 #. ``ule``: unsigned less or equal
5753 #. ``sgt``: signed greater than
5754 #. ``sge``: signed greater or equal
5755 #. ``slt``: signed less than
5756 #. ``sle``: signed less or equal
5758 The remaining two arguments must be :ref:`integer <t_integer>` or
5759 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5760 must also be identical types.
5765 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5766 code given as ``cond``. The comparison performed always yields either an
5767 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5769 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5770 otherwise. No sign interpretation is necessary or performed.
5771 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5772 otherwise. No sign interpretation is necessary or performed.
5773 #. ``ugt``: interprets the operands as unsigned values and yields
5774 ``true`` if ``op1`` is greater than ``op2``.
5775 #. ``uge``: interprets the operands as unsigned values and yields
5776 ``true`` if ``op1`` is greater than or equal to ``op2``.
5777 #. ``ult``: interprets the operands as unsigned values and yields
5778 ``true`` if ``op1`` is less than ``op2``.
5779 #. ``ule``: interprets the operands as unsigned values and yields
5780 ``true`` if ``op1`` is less than or equal to ``op2``.
5781 #. ``sgt``: interprets the operands as signed values and yields ``true``
5782 if ``op1`` is greater than ``op2``.
5783 #. ``sge``: interprets the operands as signed values and yields ``true``
5784 if ``op1`` is greater than or equal to ``op2``.
5785 #. ``slt``: interprets the operands as signed values and yields ``true``
5786 if ``op1`` is less than ``op2``.
5787 #. ``sle``: interprets the operands as signed values and yields ``true``
5788 if ``op1`` is less than or equal to ``op2``.
5790 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5791 are compared as if they were integers.
5793 If the operands are integer vectors, then they are compared element by
5794 element. The result is an ``i1`` vector with the same number of elements
5795 as the values being compared. Otherwise, the result is an ``i1``.
5800 .. code-block:: llvm
5802 <result> = icmp eq i32 4, 5 ; yields: result=false
5803 <result> = icmp ne float* %X, %X ; yields: result=false
5804 <result> = icmp ult i16 4, 5 ; yields: result=true
5805 <result> = icmp sgt i16 4, 5 ; yields: result=false
5806 <result> = icmp ule i16 -4, 5 ; yields: result=false
5807 <result> = icmp sge i16 4, 5 ; yields: result=false
5809 Note that the code generator does not yet support vector types with the
5810 ``icmp`` instruction.
5814 '``fcmp``' Instruction
5815 ^^^^^^^^^^^^^^^^^^^^^^
5822 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5827 The '``fcmp``' instruction returns a boolean value or vector of boolean
5828 values based on comparison of its operands.
5830 If the operands are floating point scalars, then the result type is a
5831 boolean (:ref:`i1 <t_integer>`).
5833 If the operands are floating point vectors, then the result type is a
5834 vector of boolean with the same number of elements as the operands being
5840 The '``fcmp``' instruction takes three operands. The first operand is
5841 the condition code indicating the kind of comparison to perform. It is
5842 not a value, just a keyword. The possible condition code are:
5844 #. ``false``: no comparison, always returns false
5845 #. ``oeq``: ordered and equal
5846 #. ``ogt``: ordered and greater than
5847 #. ``oge``: ordered and greater than or equal
5848 #. ``olt``: ordered and less than
5849 #. ``ole``: ordered and less than or equal
5850 #. ``one``: ordered and not equal
5851 #. ``ord``: ordered (no nans)
5852 #. ``ueq``: unordered or equal
5853 #. ``ugt``: unordered or greater than
5854 #. ``uge``: unordered or greater than or equal
5855 #. ``ult``: unordered or less than
5856 #. ``ule``: unordered or less than or equal
5857 #. ``une``: unordered or not equal
5858 #. ``uno``: unordered (either nans)
5859 #. ``true``: no comparison, always returns true
5861 *Ordered* means that neither operand is a QNAN while *unordered* means
5862 that either operand may be a QNAN.
5864 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5865 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5866 type. They must have identical types.
5871 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5872 condition code given as ``cond``. If the operands are vectors, then the
5873 vectors are compared element by element. Each comparison performed
5874 always yields an :ref:`i1 <t_integer>` result, as follows:
5876 #. ``false``: always yields ``false``, regardless of operands.
5877 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5878 is equal to ``op2``.
5879 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5880 is greater than ``op2``.
5881 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5882 is greater than or equal to ``op2``.
5883 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5884 is less than ``op2``.
5885 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5886 is less than or equal to ``op2``.
5887 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5888 is not equal to ``op2``.
5889 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5890 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5892 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5893 greater than ``op2``.
5894 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5895 greater than or equal to ``op2``.
5896 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5898 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5899 less than or equal to ``op2``.
5900 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5901 not equal to ``op2``.
5902 #. ``uno``: yields ``true`` if either operand is a QNAN.
5903 #. ``true``: always yields ``true``, regardless of operands.
5908 .. code-block:: llvm
5910 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5911 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5912 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5913 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5915 Note that the code generator does not yet support vector types with the
5916 ``fcmp`` instruction.
5920 '``phi``' Instruction
5921 ^^^^^^^^^^^^^^^^^^^^^
5928 <result> = phi <ty> [ <val0>, <label0>], ...
5933 The '``phi``' instruction is used to implement the φ node in the SSA
5934 graph representing the function.
5939 The type of the incoming values is specified with the first type field.
5940 After this, the '``phi``' instruction takes a list of pairs as
5941 arguments, with one pair for each predecessor basic block of the current
5942 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5943 the value arguments to the PHI node. Only labels may be used as the
5946 There must be no non-phi instructions between the start of a basic block
5947 and the PHI instructions: i.e. PHI instructions must be first in a basic
5950 For the purposes of the SSA form, the use of each incoming value is
5951 deemed to occur on the edge from the corresponding predecessor block to
5952 the current block (but after any definition of an '``invoke``'
5953 instruction's return value on the same edge).
5958 At runtime, the '``phi``' instruction logically takes on the value
5959 specified by the pair corresponding to the predecessor basic block that
5960 executed just prior to the current block.
5965 .. code-block:: llvm
5967 Loop: ; Infinite loop that counts from 0 on up...
5968 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5969 %nextindvar = add i32 %indvar, 1
5974 '``select``' Instruction
5975 ^^^^^^^^^^^^^^^^^^^^^^^^
5982 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5984 selty is either i1 or {<N x i1>}
5989 The '``select``' instruction is used to choose one value based on a
5990 condition, without branching.
5995 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5996 values indicating the condition, and two values of the same :ref:`first
5997 class <t_firstclass>` type. If the val1/val2 are vectors and the
5998 condition is a scalar, then entire vectors are selected, not individual
6004 If the condition is an i1 and it evaluates to 1, the instruction returns
6005 the first value argument; otherwise, it returns the second value
6008 If the condition is a vector of i1, then the value arguments must be
6009 vectors of the same size, and the selection is done element by element.
6014 .. code-block:: llvm
6016 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6020 '``call``' Instruction
6021 ^^^^^^^^^^^^^^^^^^^^^^
6028 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6033 The '``call``' instruction represents a simple function call.
6038 This instruction requires several arguments:
6040 #. The optional "tail" marker indicates that the callee function does
6041 not access any allocas or varargs in the caller. Note that calls may
6042 be marked "tail" even if they do not occur before a
6043 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6044 function call is eligible for tail call optimization, but `might not
6045 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6046 The code generator may optimize calls marked "tail" with either 1)
6047 automatic `sibling call
6048 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6049 callee have matching signatures, or 2) forced tail call optimization
6050 when the following extra requirements are met:
6052 - Caller and callee both have the calling convention ``fastcc``.
6053 - The call is in tail position (ret immediately follows call and ret
6054 uses value of call or is void).
6055 - Option ``-tailcallopt`` is enabled, or
6056 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6057 - `Platform specific constraints are
6058 met. <CodeGenerator.html#tailcallopt>`_
6060 #. The optional "cconv" marker indicates which :ref:`calling
6061 convention <callingconv>` the call should use. If none is
6062 specified, the call defaults to using C calling conventions. The
6063 calling convention of the call must match the calling convention of
6064 the target function, or else the behavior is undefined.
6065 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6066 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6068 #. '``ty``': the type of the call instruction itself which is also the
6069 type of the return value. Functions that return no value are marked
6071 #. '``fnty``': shall be the signature of the pointer to function value
6072 being invoked. The argument types must match the types implied by
6073 this signature. This type can be omitted if the function is not
6074 varargs and if the function type does not return a pointer to a
6076 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6077 be invoked. In most cases, this is a direct function invocation, but
6078 indirect ``call``'s are just as possible, calling an arbitrary pointer
6080 #. '``function args``': argument list whose types match the function
6081 signature argument types and parameter attributes. All arguments must
6082 be of :ref:`first class <t_firstclass>` type. If the function signature
6083 indicates the function accepts a variable number of arguments, the
6084 extra arguments can be specified.
6085 #. The optional :ref:`function attributes <fnattrs>` list. Only
6086 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6087 attributes are valid here.
6092 The '``call``' instruction is used to cause control flow to transfer to
6093 a specified function, with its incoming arguments bound to the specified
6094 values. Upon a '``ret``' instruction in the called function, control
6095 flow continues with the instruction after the function call, and the
6096 return value of the function is bound to the result argument.
6101 .. code-block:: llvm
6103 %retval = call i32 @test(i32 %argc)
6104 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6105 %X = tail call i32 @foo() ; yields i32
6106 %Y = tail call fastcc i32 @foo() ; yields i32
6107 call void %foo(i8 97 signext)
6109 %struct.A = type { i32, i8 }
6110 %r = call %struct.A @foo() ; yields { 32, i8 }
6111 %gr = extractvalue %struct.A %r, 0 ; yields i32
6112 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6113 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6114 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6116 llvm treats calls to some functions with names and arguments that match
6117 the standard C99 library as being the C99 library functions, and may
6118 perform optimizations or generate code for them under that assumption.
6119 This is something we'd like to change in the future to provide better
6120 support for freestanding environments and non-C-based languages.
6124 '``va_arg``' Instruction
6125 ^^^^^^^^^^^^^^^^^^^^^^^^
6132 <resultval> = va_arg <va_list*> <arglist>, <argty>
6137 The '``va_arg``' instruction is used to access arguments passed through
6138 the "variable argument" area of a function call. It is used to implement
6139 the ``va_arg`` macro in C.
6144 This instruction takes a ``va_list*`` value and the type of the
6145 argument. It returns a value of the specified argument type and
6146 increments the ``va_list`` to point to the next argument. The actual
6147 type of ``va_list`` is target specific.
6152 The '``va_arg``' instruction loads an argument of the specified type
6153 from the specified ``va_list`` and causes the ``va_list`` to point to
6154 the next argument. For more information, see the variable argument
6155 handling :ref:`Intrinsic Functions <int_varargs>`.
6157 It is legal for this instruction to be called in a function which does
6158 not take a variable number of arguments, for example, the ``vfprintf``
6161 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6162 function <intrinsics>` because it takes a type as an argument.
6167 See the :ref:`variable argument processing <int_varargs>` section.
6169 Note that the code generator does not yet fully support va\_arg on many
6170 targets. Also, it does not currently support va\_arg with aggregate
6171 types on any target.
6175 '``landingpad``' Instruction
6176 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6183 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6184 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6186 <clause> := catch <type> <value>
6187 <clause> := filter <array constant type> <array constant>
6192 The '``landingpad``' instruction is used by `LLVM's exception handling
6193 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6194 is a landing pad --- one where the exception lands, and corresponds to the
6195 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6196 defines values supplied by the personality function (``pers_fn``) upon
6197 re-entry to the function. The ``resultval`` has the type ``resultty``.
6202 This instruction takes a ``pers_fn`` value. This is the personality
6203 function associated with the unwinding mechanism. The optional
6204 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6206 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6207 contains the global variable representing the "type" that may be caught
6208 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6209 clause takes an array constant as its argument. Use
6210 "``[0 x i8**] undef``" for a filter which cannot throw. The
6211 '``landingpad``' instruction must contain *at least* one ``clause`` or
6212 the ``cleanup`` flag.
6217 The '``landingpad``' instruction defines the values which are set by the
6218 personality function (``pers_fn``) upon re-entry to the function, and
6219 therefore the "result type" of the ``landingpad`` instruction. As with
6220 calling conventions, how the personality function results are
6221 represented in LLVM IR is target specific.
6223 The clauses are applied in order from top to bottom. If two
6224 ``landingpad`` instructions are merged together through inlining, the
6225 clauses from the calling function are appended to the list of clauses.
6226 When the call stack is being unwound due to an exception being thrown,
6227 the exception is compared against each ``clause`` in turn. If it doesn't
6228 match any of the clauses, and the ``cleanup`` flag is not set, then
6229 unwinding continues further up the call stack.
6231 The ``landingpad`` instruction has several restrictions:
6233 - A landing pad block is a basic block which is the unwind destination
6234 of an '``invoke``' instruction.
6235 - A landing pad block must have a '``landingpad``' instruction as its
6236 first non-PHI instruction.
6237 - There can be only one '``landingpad``' instruction within the landing
6239 - A basic block that is not a landing pad block may not include a
6240 '``landingpad``' instruction.
6241 - All '``landingpad``' instructions in a function must have the same
6242 personality function.
6247 .. code-block:: llvm
6249 ;; A landing pad which can catch an integer.
6250 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6252 ;; A landing pad that is a cleanup.
6253 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6255 ;; A landing pad which can catch an integer and can only throw a double.
6256 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6258 filter [1 x i8**] [@_ZTId]
6265 LLVM supports the notion of an "intrinsic function". These functions
6266 have well known names and semantics and are required to follow certain
6267 restrictions. Overall, these intrinsics represent an extension mechanism
6268 for the LLVM language that does not require changing all of the
6269 transformations in LLVM when adding to the language (or the bitcode
6270 reader/writer, the parser, etc...).
6272 Intrinsic function names must all start with an "``llvm.``" prefix. This
6273 prefix is reserved in LLVM for intrinsic names; thus, function names may
6274 not begin with this prefix. Intrinsic functions must always be external
6275 functions: you cannot define the body of intrinsic functions. Intrinsic
6276 functions may only be used in call or invoke instructions: it is illegal
6277 to take the address of an intrinsic function. Additionally, because
6278 intrinsic functions are part of the LLVM language, it is required if any
6279 are added that they be documented here.
6281 Some intrinsic functions can be overloaded, i.e., the intrinsic
6282 represents a family of functions that perform the same operation but on
6283 different data types. Because LLVM can represent over 8 million
6284 different integer types, overloading is used commonly to allow an
6285 intrinsic function to operate on any integer type. One or more of the
6286 argument types or the result type can be overloaded to accept any
6287 integer type. Argument types may also be defined as exactly matching a
6288 previous argument's type or the result type. This allows an intrinsic
6289 function which accepts multiple arguments, but needs all of them to be
6290 of the same type, to only be overloaded with respect to a single
6291 argument or the result.
6293 Overloaded intrinsics will have the names of its overloaded argument
6294 types encoded into its function name, each preceded by a period. Only
6295 those types which are overloaded result in a name suffix. Arguments
6296 whose type is matched against another type do not. For example, the
6297 ``llvm.ctpop`` function can take an integer of any width and returns an
6298 integer of exactly the same integer width. This leads to a family of
6299 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6300 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6301 overloaded, and only one type suffix is required. Because the argument's
6302 type is matched against the return type, it does not require its own
6305 To learn how to add an intrinsic function, please see the `Extending
6306 LLVM Guide <ExtendingLLVM.html>`_.
6310 Variable Argument Handling Intrinsics
6311 -------------------------------------
6313 Variable argument support is defined in LLVM with the
6314 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6315 functions. These functions are related to the similarly named macros
6316 defined in the ``<stdarg.h>`` header file.
6318 All of these functions operate on arguments that use a target-specific
6319 value type "``va_list``". The LLVM assembly language reference manual
6320 does not define what this type is, so all transformations should be
6321 prepared to handle these functions regardless of the type used.
6323 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6324 variable argument handling intrinsic functions are used.
6326 .. code-block:: llvm
6328 define i32 @test(i32 %X, ...) {
6329 ; Initialize variable argument processing
6331 %ap2 = bitcast i8** %ap to i8*
6332 call void @llvm.va_start(i8* %ap2)
6334 ; Read a single integer argument
6335 %tmp = va_arg i8** %ap, i32
6337 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6339 %aq2 = bitcast i8** %aq to i8*
6340 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6341 call void @llvm.va_end(i8* %aq2)
6343 ; Stop processing of arguments.
6344 call void @llvm.va_end(i8* %ap2)
6348 declare void @llvm.va_start(i8*)
6349 declare void @llvm.va_copy(i8*, i8*)
6350 declare void @llvm.va_end(i8*)
6354 '``llvm.va_start``' Intrinsic
6355 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6362 declare void @llvm.va_start(i8* <arglist>)
6367 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6368 subsequent use by ``va_arg``.
6373 The argument is a pointer to a ``va_list`` element to initialize.
6378 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6379 available in C. In a target-dependent way, it initializes the
6380 ``va_list`` element to which the argument points, so that the next call
6381 to ``va_arg`` will produce the first variable argument passed to the
6382 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6383 to know the last argument of the function as the compiler can figure
6386 '``llvm.va_end``' Intrinsic
6387 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6394 declare void @llvm.va_end(i8* <arglist>)
6399 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6400 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6405 The argument is a pointer to a ``va_list`` to destroy.
6410 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6411 available in C. In a target-dependent way, it destroys the ``va_list``
6412 element to which the argument points. Calls to
6413 :ref:`llvm.va_start <int_va_start>` and
6414 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6419 '``llvm.va_copy``' Intrinsic
6420 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6427 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6432 The '``llvm.va_copy``' intrinsic copies the current argument position
6433 from the source argument list to the destination argument list.
6438 The first argument is a pointer to a ``va_list`` element to initialize.
6439 The second argument is a pointer to a ``va_list`` element to copy from.
6444 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6445 available in C. In a target-dependent way, it copies the source
6446 ``va_list`` element into the destination ``va_list`` element. This
6447 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6448 arbitrarily complex and require, for example, memory allocation.
6450 Accurate Garbage Collection Intrinsics
6451 --------------------------------------
6453 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6454 (GC) requires the implementation and generation of these intrinsics.
6455 These intrinsics allow identification of :ref:`GC roots on the
6456 stack <int_gcroot>`, as well as garbage collector implementations that
6457 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6458 Front-ends for type-safe garbage collected languages should generate
6459 these intrinsics to make use of the LLVM garbage collectors. For more
6460 details, see `Accurate Garbage Collection with
6461 LLVM <GarbageCollection.html>`_.
6463 The garbage collection intrinsics only operate on objects in the generic
6464 address space (address space zero).
6468 '``llvm.gcroot``' Intrinsic
6469 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6476 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6481 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6482 the code generator, and allows some metadata to be associated with it.
6487 The first argument specifies the address of a stack object that contains
6488 the root pointer. The second pointer (which must be either a constant or
6489 a global value address) contains the meta-data to be associated with the
6495 At runtime, a call to this intrinsic stores a null pointer into the
6496 "ptrloc" location. At compile-time, the code generator generates
6497 information to allow the runtime to find the pointer at GC safe points.
6498 The '``llvm.gcroot``' intrinsic may only be used in a function which
6499 :ref:`specifies a GC algorithm <gc>`.
6503 '``llvm.gcread``' Intrinsic
6504 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6511 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6516 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6517 locations, allowing garbage collector implementations that require read
6523 The second argument is the address to read from, which should be an
6524 address allocated from the garbage collector. The first object is a
6525 pointer to the start of the referenced object, if needed by the language
6526 runtime (otherwise null).
6531 The '``llvm.gcread``' intrinsic has the same semantics as a load
6532 instruction, but may be replaced with substantially more complex code by
6533 the garbage collector runtime, as needed. The '``llvm.gcread``'
6534 intrinsic may only be used in a function which :ref:`specifies a GC
6539 '``llvm.gcwrite``' Intrinsic
6540 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6547 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6552 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6553 locations, allowing garbage collector implementations that require write
6554 barriers (such as generational or reference counting collectors).
6559 The first argument is the reference to store, the second is the start of
6560 the object to store it to, and the third is the address of the field of
6561 Obj to store to. If the runtime does not require a pointer to the
6562 object, Obj may be null.
6567 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6568 instruction, but may be replaced with substantially more complex code by
6569 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6570 intrinsic may only be used in a function which :ref:`specifies a GC
6573 Code Generator Intrinsics
6574 -------------------------
6576 These intrinsics are provided by LLVM to expose special features that
6577 may only be implemented with code generator support.
6579 '``llvm.returnaddress``' Intrinsic
6580 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6587 declare i8 *@llvm.returnaddress(i32 <level>)
6592 The '``llvm.returnaddress``' intrinsic attempts to compute a
6593 target-specific value indicating the return address of the current
6594 function or one of its callers.
6599 The argument to this intrinsic indicates which function to return the
6600 address for. Zero indicates the calling function, one indicates its
6601 caller, etc. The argument is **required** to be a constant integer
6607 The '``llvm.returnaddress``' intrinsic either returns a pointer
6608 indicating the return address of the specified call frame, or zero if it
6609 cannot be identified. The value returned by this intrinsic is likely to
6610 be incorrect or 0 for arguments other than zero, so it should only be
6611 used for debugging purposes.
6613 Note that calling this intrinsic does not prevent function inlining or
6614 other aggressive transformations, so the value returned may not be that
6615 of the obvious source-language caller.
6617 '``llvm.frameaddress``' Intrinsic
6618 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6625 declare i8* @llvm.frameaddress(i32 <level>)
6630 The '``llvm.frameaddress``' intrinsic attempts to return the
6631 target-specific frame pointer value for the specified stack frame.
6636 The argument to this intrinsic indicates which function to return the
6637 frame pointer for. Zero indicates the calling function, one indicates
6638 its caller, etc. The argument is **required** to be a constant integer
6644 The '``llvm.frameaddress``' intrinsic either returns a pointer
6645 indicating the frame address of the specified call frame, or zero if it
6646 cannot be identified. The value returned by this intrinsic is likely to
6647 be incorrect or 0 for arguments other than zero, so it should only be
6648 used for debugging purposes.
6650 Note that calling this intrinsic does not prevent function inlining or
6651 other aggressive transformations, so the value returned may not be that
6652 of the obvious source-language caller.
6656 '``llvm.stacksave``' Intrinsic
6657 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6664 declare i8* @llvm.stacksave()
6669 The '``llvm.stacksave``' intrinsic is used to remember the current state
6670 of the function stack, for use with
6671 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6672 implementing language features like scoped automatic variable sized
6678 This intrinsic returns a opaque pointer value that can be passed to
6679 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6680 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6681 ``llvm.stacksave``, it effectively restores the state of the stack to
6682 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6683 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6684 were allocated after the ``llvm.stacksave`` was executed.
6686 .. _int_stackrestore:
6688 '``llvm.stackrestore``' Intrinsic
6689 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6696 declare void @llvm.stackrestore(i8* %ptr)
6701 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6702 the function stack to the state it was in when the corresponding
6703 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6704 useful for implementing language features like scoped automatic variable
6705 sized arrays in C99.
6710 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6712 '``llvm.prefetch``' Intrinsic
6713 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6720 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6725 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6726 insert a prefetch instruction if supported; otherwise, it is a noop.
6727 Prefetches have no effect on the behavior of the program but can change
6728 its performance characteristics.
6733 ``address`` is the address to be prefetched, ``rw`` is the specifier
6734 determining if the fetch should be for a read (0) or write (1), and
6735 ``locality`` is a temporal locality specifier ranging from (0) - no
6736 locality, to (3) - extremely local keep in cache. The ``cache type``
6737 specifies whether the prefetch is performed on the data (1) or
6738 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6739 arguments must be constant integers.
6744 This intrinsic does not modify the behavior of the program. In
6745 particular, prefetches cannot trap and do not produce a value. On
6746 targets that support this intrinsic, the prefetch can provide hints to
6747 the processor cache for better performance.
6749 '``llvm.pcmarker``' Intrinsic
6750 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6757 declare void @llvm.pcmarker(i32 <id>)
6762 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6763 Counter (PC) in a region of code to simulators and other tools. The
6764 method is target specific, but it is expected that the marker will use
6765 exported symbols to transmit the PC of the marker. The marker makes no
6766 guarantees that it will remain with any specific instruction after
6767 optimizations. It is possible that the presence of a marker will inhibit
6768 optimizations. The intended use is to be inserted after optimizations to
6769 allow correlations of simulation runs.
6774 ``id`` is a numerical id identifying the marker.
6779 This intrinsic does not modify the behavior of the program. Backends
6780 that do not support this intrinsic may ignore it.
6782 '``llvm.readcyclecounter``' Intrinsic
6783 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6790 declare i64 @llvm.readcyclecounter()
6795 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6796 counter register (or similar low latency, high accuracy clocks) on those
6797 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6798 should map to RPCC. As the backing counters overflow quickly (on the
6799 order of 9 seconds on alpha), this should only be used for small
6805 When directly supported, reading the cycle counter should not modify any
6806 memory. Implementations are allowed to either return a application
6807 specific value or a system wide value. On backends without support, this
6808 is lowered to a constant 0.
6810 Note that runtime support may be conditional on the privilege-level code is
6811 running at and the host platform.
6813 Standard C Library Intrinsics
6814 -----------------------------
6816 LLVM provides intrinsics for a few important standard C library
6817 functions. These intrinsics allow source-language front-ends to pass
6818 information about the alignment of the pointer arguments to the code
6819 generator, providing opportunity for more efficient code generation.
6823 '``llvm.memcpy``' Intrinsic
6824 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6829 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6830 integer bit width and for different address spaces. Not all targets
6831 support all bit widths however.
6835 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6836 i32 <len>, i32 <align>, i1 <isvolatile>)
6837 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6838 i64 <len>, i32 <align>, i1 <isvolatile>)
6843 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6844 source location to the destination location.
6846 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6847 intrinsics do not return a value, takes extra alignment/isvolatile
6848 arguments and the pointers can be in specified address spaces.
6853 The first argument is a pointer to the destination, the second is a
6854 pointer to the source. The third argument is an integer argument
6855 specifying the number of bytes to copy, the fourth argument is the
6856 alignment of the source and destination locations, and the fifth is a
6857 boolean indicating a volatile access.
6859 If the call to this intrinsic has an alignment value that is not 0 or 1,
6860 then the caller guarantees that both the source and destination pointers
6861 are aligned to that boundary.
6863 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6864 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6865 very cleanly specified and it is unwise to depend on it.
6870 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6871 source location to the destination location, which are not allowed to
6872 overlap. It copies "len" bytes of memory over. If the argument is known
6873 to be aligned to some boundary, this can be specified as the fourth
6874 argument, otherwise it should be set to 0 or 1.
6876 '``llvm.memmove``' Intrinsic
6877 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6882 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6883 bit width and for different address space. Not all targets support all
6888 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6889 i32 <len>, i32 <align>, i1 <isvolatile>)
6890 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6891 i64 <len>, i32 <align>, i1 <isvolatile>)
6896 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6897 source location to the destination location. It is similar to the
6898 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6901 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6902 intrinsics do not return a value, takes extra alignment/isvolatile
6903 arguments and the pointers can be in specified address spaces.
6908 The first argument is a pointer to the destination, the second is a
6909 pointer to the source. The third argument is an integer argument
6910 specifying the number of bytes to copy, the fourth argument is the
6911 alignment of the source and destination locations, and the fifth is a
6912 boolean indicating a volatile access.
6914 If the call to this intrinsic has an alignment value that is not 0 or 1,
6915 then the caller guarantees that the source and destination pointers are
6916 aligned to that boundary.
6918 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6919 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6920 not very cleanly specified and it is unwise to depend on it.
6925 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6926 source location to the destination location, which may overlap. It
6927 copies "len" bytes of memory over. If the argument is known to be
6928 aligned to some boundary, this can be specified as the fourth argument,
6929 otherwise it should be set to 0 or 1.
6931 '``llvm.memset.*``' Intrinsics
6932 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6937 This is an overloaded intrinsic. You can use llvm.memset on any integer
6938 bit width and for different address spaces. However, not all targets
6939 support all bit widths.
6943 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6944 i32 <len>, i32 <align>, i1 <isvolatile>)
6945 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6946 i64 <len>, i32 <align>, i1 <isvolatile>)
6951 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6952 particular byte value.
6954 Note that, unlike the standard libc function, the ``llvm.memset``
6955 intrinsic does not return a value and takes extra alignment/volatile
6956 arguments. Also, the destination can be in an arbitrary address space.
6961 The first argument is a pointer to the destination to fill, the second
6962 is the byte value with which to fill it, the third argument is an
6963 integer argument specifying the number of bytes to fill, and the fourth
6964 argument is the known alignment of the destination location.
6966 If the call to this intrinsic has an alignment value that is not 0 or 1,
6967 then the caller guarantees that the destination pointer is aligned to
6970 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6971 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6972 very cleanly specified and it is unwise to depend on it.
6977 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6978 at the destination location. If the argument is known to be aligned to
6979 some boundary, this can be specified as the fourth argument, otherwise
6980 it should be set to 0 or 1.
6982 '``llvm.sqrt.*``' Intrinsic
6983 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6988 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6989 floating point or vector of floating point type. Not all targets support
6994 declare float @llvm.sqrt.f32(float %Val)
6995 declare double @llvm.sqrt.f64(double %Val)
6996 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6997 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6998 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7003 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7004 returning the same value as the libm '``sqrt``' functions would. Unlike
7005 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7006 negative numbers other than -0.0 (which allows for better optimization,
7007 because there is no need to worry about errno being set).
7008 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7013 The argument and return value are floating point numbers of the same
7019 This function returns the sqrt of the specified operand if it is a
7020 nonnegative floating point number.
7022 '``llvm.powi.*``' Intrinsic
7023 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7028 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7029 floating point or vector of floating point type. Not all targets support
7034 declare float @llvm.powi.f32(float %Val, i32 %power)
7035 declare double @llvm.powi.f64(double %Val, i32 %power)
7036 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7037 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7038 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7043 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7044 specified (positive or negative) power. The order of evaluation of
7045 multiplications is not defined. When a vector of floating point type is
7046 used, the second argument remains a scalar integer value.
7051 The second argument is an integer power, and the first is a value to
7052 raise to that power.
7057 This function returns the first value raised to the second power with an
7058 unspecified sequence of rounding operations.
7060 '``llvm.sin.*``' Intrinsic
7061 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7066 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7067 floating point or vector of floating point type. Not all targets support
7072 declare float @llvm.sin.f32(float %Val)
7073 declare double @llvm.sin.f64(double %Val)
7074 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7075 declare fp128 @llvm.sin.f128(fp128 %Val)
7076 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7081 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7086 The argument and return value are floating point numbers of the same
7092 This function returns the sine of the specified operand, returning the
7093 same values as the libm ``sin`` functions would, and handles error
7094 conditions in the same way.
7096 '``llvm.cos.*``' Intrinsic
7097 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7102 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7103 floating point or vector of floating point type. Not all targets support
7108 declare float @llvm.cos.f32(float %Val)
7109 declare double @llvm.cos.f64(double %Val)
7110 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7111 declare fp128 @llvm.cos.f128(fp128 %Val)
7112 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7117 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7122 The argument and return value are floating point numbers of the same
7128 This function returns the cosine of the specified operand, returning the
7129 same values as the libm ``cos`` functions would, and handles error
7130 conditions in the same way.
7132 '``llvm.pow.*``' Intrinsic
7133 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7138 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7139 floating point or vector of floating point type. Not all targets support
7144 declare float @llvm.pow.f32(float %Val, float %Power)
7145 declare double @llvm.pow.f64(double %Val, double %Power)
7146 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7147 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7148 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7153 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7154 specified (positive or negative) power.
7159 The second argument is a floating point power, and the first is a value
7160 to raise to that power.
7165 This function returns the first value raised to the second power,
7166 returning the same values as the libm ``pow`` functions would, and
7167 handles error conditions in the same way.
7169 '``llvm.exp.*``' Intrinsic
7170 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7175 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7176 floating point or vector of floating point type. Not all targets support
7181 declare float @llvm.exp.f32(float %Val)
7182 declare double @llvm.exp.f64(double %Val)
7183 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7184 declare fp128 @llvm.exp.f128(fp128 %Val)
7185 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7190 The '``llvm.exp.*``' intrinsics perform the exp function.
7195 The argument and return value are floating point numbers of the same
7201 This function returns the same values as the libm ``exp`` functions
7202 would, and handles error conditions in the same way.
7204 '``llvm.exp2.*``' Intrinsic
7205 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7210 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7211 floating point or vector of floating point type. Not all targets support
7216 declare float @llvm.exp2.f32(float %Val)
7217 declare double @llvm.exp2.f64(double %Val)
7218 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7219 declare fp128 @llvm.exp2.f128(fp128 %Val)
7220 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7225 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7230 The argument and return value are floating point numbers of the same
7236 This function returns the same values as the libm ``exp2`` functions
7237 would, and handles error conditions in the same way.
7239 '``llvm.log.*``' Intrinsic
7240 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7245 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7246 floating point or vector of floating point type. Not all targets support
7251 declare float @llvm.log.f32(float %Val)
7252 declare double @llvm.log.f64(double %Val)
7253 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7254 declare fp128 @llvm.log.f128(fp128 %Val)
7255 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7260 The '``llvm.log.*``' intrinsics perform the log function.
7265 The argument and return value are floating point numbers of the same
7271 This function returns the same values as the libm ``log`` functions
7272 would, and handles error conditions in the same way.
7274 '``llvm.log10.*``' Intrinsic
7275 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7280 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7281 floating point or vector of floating point type. Not all targets support
7286 declare float @llvm.log10.f32(float %Val)
7287 declare double @llvm.log10.f64(double %Val)
7288 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7289 declare fp128 @llvm.log10.f128(fp128 %Val)
7290 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7295 The '``llvm.log10.*``' intrinsics perform the log10 function.
7300 The argument and return value are floating point numbers of the same
7306 This function returns the same values as the libm ``log10`` functions
7307 would, and handles error conditions in the same way.
7309 '``llvm.log2.*``' Intrinsic
7310 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7315 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7316 floating point or vector of floating point type. Not all targets support
7321 declare float @llvm.log2.f32(float %Val)
7322 declare double @llvm.log2.f64(double %Val)
7323 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7324 declare fp128 @llvm.log2.f128(fp128 %Val)
7325 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7330 The '``llvm.log2.*``' intrinsics perform the log2 function.
7335 The argument and return value are floating point numbers of the same
7341 This function returns the same values as the libm ``log2`` functions
7342 would, and handles error conditions in the same way.
7344 '``llvm.fma.*``' Intrinsic
7345 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7350 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7351 floating point or vector of floating point type. Not all targets support
7356 declare float @llvm.fma.f32(float %a, float %b, float %c)
7357 declare double @llvm.fma.f64(double %a, double %b, double %c)
7358 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7359 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7360 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7365 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7371 The argument and return value are floating point numbers of the same
7377 This function returns the same values as the libm ``fma`` functions
7380 '``llvm.fabs.*``' Intrinsic
7381 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7386 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7387 floating point or vector of floating point type. Not all targets support
7392 declare float @llvm.fabs.f32(float %Val)
7393 declare double @llvm.fabs.f64(double %Val)
7394 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7395 declare fp128 @llvm.fabs.f128(fp128 %Val)
7396 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7401 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7407 The argument and return value are floating point numbers of the same
7413 This function returns the same values as the libm ``fabs`` functions
7414 would, and handles error conditions in the same way.
7416 '``llvm.copysign.*``' Intrinsic
7417 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7422 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7423 floating point or vector of floating point type. Not all targets support
7428 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7429 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7430 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7431 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7432 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7437 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7438 first operand and the sign of the second operand.
7443 The arguments and return value are floating point numbers of the same
7449 This function returns the same values as the libm ``copysign``
7450 functions would, and handles error conditions in the same way.
7452 '``llvm.floor.*``' Intrinsic
7453 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7458 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7459 floating point or vector of floating point type. Not all targets support
7464 declare float @llvm.floor.f32(float %Val)
7465 declare double @llvm.floor.f64(double %Val)
7466 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7467 declare fp128 @llvm.floor.f128(fp128 %Val)
7468 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7473 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7478 The argument and return value are floating point numbers of the same
7484 This function returns the same values as the libm ``floor`` functions
7485 would, and handles error conditions in the same way.
7487 '``llvm.ceil.*``' Intrinsic
7488 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7493 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7494 floating point or vector of floating point type. Not all targets support
7499 declare float @llvm.ceil.f32(float %Val)
7500 declare double @llvm.ceil.f64(double %Val)
7501 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7502 declare fp128 @llvm.ceil.f128(fp128 %Val)
7503 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7508 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7513 The argument and return value are floating point numbers of the same
7519 This function returns the same values as the libm ``ceil`` functions
7520 would, and handles error conditions in the same way.
7522 '``llvm.trunc.*``' Intrinsic
7523 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7528 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7529 floating point or vector of floating point type. Not all targets support
7534 declare float @llvm.trunc.f32(float %Val)
7535 declare double @llvm.trunc.f64(double %Val)
7536 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7537 declare fp128 @llvm.trunc.f128(fp128 %Val)
7538 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7543 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7544 nearest integer not larger in magnitude than the operand.
7549 The argument and return value are floating point numbers of the same
7555 This function returns the same values as the libm ``trunc`` functions
7556 would, and handles error conditions in the same way.
7558 '``llvm.rint.*``' Intrinsic
7559 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7564 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7565 floating point or vector of floating point type. Not all targets support
7570 declare float @llvm.rint.f32(float %Val)
7571 declare double @llvm.rint.f64(double %Val)
7572 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7573 declare fp128 @llvm.rint.f128(fp128 %Val)
7574 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7579 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7580 nearest integer. It may raise an inexact floating-point exception if the
7581 operand isn't an integer.
7586 The argument and return value are floating point numbers of the same
7592 This function returns the same values as the libm ``rint`` functions
7593 would, and handles error conditions in the same way.
7595 '``llvm.nearbyint.*``' Intrinsic
7596 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7601 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7602 floating point or vector of floating point type. Not all targets support
7607 declare float @llvm.nearbyint.f32(float %Val)
7608 declare double @llvm.nearbyint.f64(double %Val)
7609 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7610 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7611 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7616 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7622 The argument and return value are floating point numbers of the same
7628 This function returns the same values as the libm ``nearbyint``
7629 functions would, and handles error conditions in the same way.
7631 '``llvm.round.*``' Intrinsic
7632 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7637 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7638 floating point or vector of floating point type. Not all targets support
7643 declare float @llvm.round.f32(float %Val)
7644 declare double @llvm.round.f64(double %Val)
7645 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7646 declare fp128 @llvm.round.f128(fp128 %Val)
7647 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7652 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7658 The argument and return value are floating point numbers of the same
7664 This function returns the same values as the libm ``round``
7665 functions would, and handles error conditions in the same way.
7667 Bit Manipulation Intrinsics
7668 ---------------------------
7670 LLVM provides intrinsics for a few important bit manipulation
7671 operations. These allow efficient code generation for some algorithms.
7673 '``llvm.bswap.*``' Intrinsics
7674 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7679 This is an overloaded intrinsic function. You can use bswap on any
7680 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7684 declare i16 @llvm.bswap.i16(i16 <id>)
7685 declare i32 @llvm.bswap.i32(i32 <id>)
7686 declare i64 @llvm.bswap.i64(i64 <id>)
7691 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7692 values with an even number of bytes (positive multiple of 16 bits).
7693 These are useful for performing operations on data that is not in the
7694 target's native byte order.
7699 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7700 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7701 intrinsic returns an i32 value that has the four bytes of the input i32
7702 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7703 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7704 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7705 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7708 '``llvm.ctpop.*``' Intrinsic
7709 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7714 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7715 bit width, or on any vector with integer elements. Not all targets
7716 support all bit widths or vector types, however.
7720 declare i8 @llvm.ctpop.i8(i8 <src>)
7721 declare i16 @llvm.ctpop.i16(i16 <src>)
7722 declare i32 @llvm.ctpop.i32(i32 <src>)
7723 declare i64 @llvm.ctpop.i64(i64 <src>)
7724 declare i256 @llvm.ctpop.i256(i256 <src>)
7725 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7730 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7736 The only argument is the value to be counted. The argument may be of any
7737 integer type, or a vector with integer elements. The return type must
7738 match the argument type.
7743 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7744 each element of a vector.
7746 '``llvm.ctlz.*``' Intrinsic
7747 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7752 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7753 integer bit width, or any vector whose elements are integers. Not all
7754 targets support all bit widths or vector types, however.
7758 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7759 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7760 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7761 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7762 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7763 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7768 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7769 leading zeros in a variable.
7774 The first argument is the value to be counted. This argument may be of
7775 any integer type, or a vectory with integer element type. The return
7776 type must match the first argument type.
7778 The second argument must be a constant and is a flag to indicate whether
7779 the intrinsic should ensure that a zero as the first argument produces a
7780 defined result. Historically some architectures did not provide a
7781 defined result for zero values as efficiently, and many algorithms are
7782 now predicated on avoiding zero-value inputs.
7787 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7788 zeros in a variable, or within each element of the vector. If
7789 ``src == 0`` then the result is the size in bits of the type of ``src``
7790 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7791 ``llvm.ctlz(i32 2) = 30``.
7793 '``llvm.cttz.*``' Intrinsic
7794 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7799 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7800 integer bit width, or any vector of integer elements. Not all targets
7801 support all bit widths or vector types, however.
7805 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7806 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7807 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7808 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7809 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7810 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7815 The '``llvm.cttz``' family of intrinsic functions counts the number of
7821 The first argument is the value to be counted. This argument may be of
7822 any integer type, or a vectory with integer element type. The return
7823 type must match the first argument type.
7825 The second argument must be a constant and is a flag to indicate whether
7826 the intrinsic should ensure that a zero as the first argument produces a
7827 defined result. Historically some architectures did not provide a
7828 defined result for zero values as efficiently, and many algorithms are
7829 now predicated on avoiding zero-value inputs.
7834 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7835 zeros in a variable, or within each element of a vector. If ``src == 0``
7836 then the result is the size in bits of the type of ``src`` if
7837 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7838 ``llvm.cttz(2) = 1``.
7840 Arithmetic with Overflow Intrinsics
7841 -----------------------------------
7843 LLVM provides intrinsics for some arithmetic with overflow operations.
7845 '``llvm.sadd.with.overflow.*``' Intrinsics
7846 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7851 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7852 on any integer bit width.
7856 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7857 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7858 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7863 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7864 a signed addition of the two arguments, and indicate whether an overflow
7865 occurred during the signed summation.
7870 The arguments (%a and %b) and the first element of the result structure
7871 may be of integer types of any bit width, but they must have the same
7872 bit width. The second element of the result structure must be of type
7873 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7879 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7880 a signed addition of the two variables. They return a structure --- the
7881 first element of which is the signed summation, and the second element
7882 of which is a bit specifying if the signed summation resulted in an
7888 .. code-block:: llvm
7890 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7891 %sum = extractvalue {i32, i1} %res, 0
7892 %obit = extractvalue {i32, i1} %res, 1
7893 br i1 %obit, label %overflow, label %normal
7895 '``llvm.uadd.with.overflow.*``' Intrinsics
7896 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7901 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7902 on any integer bit width.
7906 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7907 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7908 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7913 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7914 an unsigned addition of the two arguments, and indicate whether a carry
7915 occurred during the unsigned summation.
7920 The arguments (%a and %b) and the first element of the result structure
7921 may be of integer types of any bit width, but they must have the same
7922 bit width. The second element of the result structure must be of type
7923 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7929 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7930 an unsigned addition of the two arguments. They return a structure --- the
7931 first element of which is the sum, and the second element of which is a
7932 bit specifying if the unsigned summation resulted in a carry.
7937 .. code-block:: llvm
7939 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7940 %sum = extractvalue {i32, i1} %res, 0
7941 %obit = extractvalue {i32, i1} %res, 1
7942 br i1 %obit, label %carry, label %normal
7944 '``llvm.ssub.with.overflow.*``' Intrinsics
7945 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7950 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7951 on any integer bit width.
7955 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7956 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7957 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7962 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7963 a signed subtraction of the two arguments, and indicate whether an
7964 overflow occurred during the signed subtraction.
7969 The arguments (%a and %b) and the first element of the result structure
7970 may be of integer types of any bit width, but they must have the same
7971 bit width. The second element of the result structure must be of type
7972 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7978 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7979 a signed subtraction of the two arguments. They return a structure --- the
7980 first element of which is the subtraction, and the second element of
7981 which is a bit specifying if the signed subtraction resulted in an
7987 .. code-block:: llvm
7989 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7990 %sum = extractvalue {i32, i1} %res, 0
7991 %obit = extractvalue {i32, i1} %res, 1
7992 br i1 %obit, label %overflow, label %normal
7994 '``llvm.usub.with.overflow.*``' Intrinsics
7995 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8000 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8001 on any integer bit width.
8005 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8006 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8007 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8012 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8013 an unsigned subtraction of the two arguments, and indicate whether an
8014 overflow occurred during the unsigned subtraction.
8019 The arguments (%a and %b) and the first element of the result structure
8020 may be of integer types of any bit width, but they must have the same
8021 bit width. The second element of the result structure must be of type
8022 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8028 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8029 an unsigned subtraction of the two arguments. They return a structure ---
8030 the first element of which is the subtraction, and the second element of
8031 which is a bit specifying if the unsigned subtraction resulted in an
8037 .. code-block:: llvm
8039 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8040 %sum = extractvalue {i32, i1} %res, 0
8041 %obit = extractvalue {i32, i1} %res, 1
8042 br i1 %obit, label %overflow, label %normal
8044 '``llvm.smul.with.overflow.*``' Intrinsics
8045 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8050 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8051 on any integer bit width.
8055 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8056 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8057 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8062 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8063 a signed multiplication of the two arguments, and indicate whether an
8064 overflow occurred during the signed multiplication.
8069 The arguments (%a and %b) and the first element of the result structure
8070 may be of integer types of any bit width, but they must have the same
8071 bit width. The second element of the result structure must be of type
8072 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8078 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8079 a signed multiplication of the two arguments. They return a structure ---
8080 the first element of which is the multiplication, and the second element
8081 of which is a bit specifying if the signed multiplication resulted in an
8087 .. code-block:: llvm
8089 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8090 %sum = extractvalue {i32, i1} %res, 0
8091 %obit = extractvalue {i32, i1} %res, 1
8092 br i1 %obit, label %overflow, label %normal
8094 '``llvm.umul.with.overflow.*``' Intrinsics
8095 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8100 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8101 on any integer bit width.
8105 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8106 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8107 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8112 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8113 a unsigned multiplication of the two arguments, and indicate whether an
8114 overflow occurred during the unsigned multiplication.
8119 The arguments (%a and %b) and the first element of the result structure
8120 may be of integer types of any bit width, but they must have the same
8121 bit width. The second element of the result structure must be of type
8122 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8128 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8129 an unsigned multiplication of the two arguments. They return a structure ---
8130 the first element of which is the multiplication, and the second
8131 element of which is a bit specifying if the unsigned multiplication
8132 resulted in an overflow.
8137 .. code-block:: llvm
8139 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8140 %sum = extractvalue {i32, i1} %res, 0
8141 %obit = extractvalue {i32, i1} %res, 1
8142 br i1 %obit, label %overflow, label %normal
8144 Specialised Arithmetic Intrinsics
8145 ---------------------------------
8147 '``llvm.fmuladd.*``' Intrinsic
8148 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8155 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8156 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8161 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8162 expressions that can be fused if the code generator determines that (a) the
8163 target instruction set has support for a fused operation, and (b) that the
8164 fused operation is more efficient than the equivalent, separate pair of mul
8165 and add instructions.
8170 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8171 multiplicands, a and b, and an addend c.
8180 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8182 is equivalent to the expression a \* b + c, except that rounding will
8183 not be performed between the multiplication and addition steps if the
8184 code generator fuses the operations. Fusion is not guaranteed, even if
8185 the target platform supports it. If a fused multiply-add is required the
8186 corresponding llvm.fma.\* intrinsic function should be used instead.
8191 .. code-block:: llvm
8193 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8195 Half Precision Floating Point Intrinsics
8196 ----------------------------------------
8198 For most target platforms, half precision floating point is a
8199 storage-only format. This means that it is a dense encoding (in memory)
8200 but does not support computation in the format.
8202 This means that code must first load the half-precision floating point
8203 value as an i16, then convert it to float with
8204 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8205 then be performed on the float value (including extending to double
8206 etc). To store the value back to memory, it is first converted to float
8207 if needed, then converted to i16 with
8208 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8211 .. _int_convert_to_fp16:
8213 '``llvm.convert.to.fp16``' Intrinsic
8214 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8221 declare i16 @llvm.convert.to.fp16(f32 %a)
8226 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8227 from single precision floating point format to half precision floating
8233 The intrinsic function contains single argument - the value to be
8239 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8240 from single precision floating point format to half precision floating
8241 point format. The return value is an ``i16`` which contains the
8247 .. code-block:: llvm
8249 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8250 store i16 %res, i16* @x, align 2
8252 .. _int_convert_from_fp16:
8254 '``llvm.convert.from.fp16``' Intrinsic
8255 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8262 declare f32 @llvm.convert.from.fp16(i16 %a)
8267 The '``llvm.convert.from.fp16``' intrinsic function performs a
8268 conversion from half precision floating point format to single precision
8269 floating point format.
8274 The intrinsic function contains single argument - the value to be
8280 The '``llvm.convert.from.fp16``' intrinsic function performs a
8281 conversion from half single precision floating point format to single
8282 precision floating point format. The input half-float value is
8283 represented by an ``i16`` value.
8288 .. code-block:: llvm
8290 %a = load i16* @x, align 2
8291 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8296 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8297 prefix), are described in the `LLVM Source Level
8298 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8301 Exception Handling Intrinsics
8302 -----------------------------
8304 The LLVM exception handling intrinsics (which all start with
8305 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8306 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8310 Trampoline Intrinsics
8311 ---------------------
8313 These intrinsics make it possible to excise one parameter, marked with
8314 the :ref:`nest <nest>` attribute, from a function. The result is a
8315 callable function pointer lacking the nest parameter - the caller does
8316 not need to provide a value for it. Instead, the value to use is stored
8317 in advance in a "trampoline", a block of memory usually allocated on the
8318 stack, which also contains code to splice the nest value into the
8319 argument list. This is used to implement the GCC nested function address
8322 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8323 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8324 It can be created as follows:
8326 .. code-block:: llvm
8328 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8329 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8330 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8331 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8332 %fp = bitcast i8* %p to i32 (i32, i32)*
8334 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8335 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8339 '``llvm.init.trampoline``' Intrinsic
8340 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8347 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8352 This fills the memory pointed to by ``tramp`` with executable code,
8353 turning it into a trampoline.
8358 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8359 pointers. The ``tramp`` argument must point to a sufficiently large and
8360 sufficiently aligned block of memory; this memory is written to by the
8361 intrinsic. Note that the size and the alignment are target-specific -
8362 LLVM currently provides no portable way of determining them, so a
8363 front-end that generates this intrinsic needs to have some
8364 target-specific knowledge. The ``func`` argument must hold a function
8365 bitcast to an ``i8*``.
8370 The block of memory pointed to by ``tramp`` is filled with target
8371 dependent code, turning it into a function. Then ``tramp`` needs to be
8372 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8373 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8374 function's signature is the same as that of ``func`` with any arguments
8375 marked with the ``nest`` attribute removed. At most one such ``nest``
8376 argument is allowed, and it must be of pointer type. Calling the new
8377 function is equivalent to calling ``func`` with the same argument list,
8378 but with ``nval`` used for the missing ``nest`` argument. If, after
8379 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8380 modified, then the effect of any later call to the returned function
8381 pointer is undefined.
8385 '``llvm.adjust.trampoline``' Intrinsic
8386 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8393 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8398 This performs any required machine-specific adjustment to the address of
8399 a trampoline (passed as ``tramp``).
8404 ``tramp`` must point to a block of memory which already has trampoline
8405 code filled in by a previous call to
8406 :ref:`llvm.init.trampoline <int_it>`.
8411 On some architectures the address of the code to be executed needs to be
8412 different to the address where the trampoline is actually stored. This
8413 intrinsic returns the executable address corresponding to ``tramp``
8414 after performing the required machine specific adjustments. The pointer
8415 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8420 This class of intrinsics exists to information about the lifetime of
8421 memory objects and ranges where variables are immutable.
8423 '``llvm.lifetime.start``' Intrinsic
8424 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8431 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8436 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8442 The first argument is a constant integer representing the size of the
8443 object, or -1 if it is variable sized. The second argument is a pointer
8449 This intrinsic indicates that before this point in the code, the value
8450 of the memory pointed to by ``ptr`` is dead. This means that it is known
8451 to never be used and has an undefined value. A load from the pointer
8452 that precedes this intrinsic can be replaced with ``'undef'``.
8454 '``llvm.lifetime.end``' Intrinsic
8455 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8462 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8467 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8473 The first argument is a constant integer representing the size of the
8474 object, or -1 if it is variable sized. The second argument is a pointer
8480 This intrinsic indicates that after this point in the code, the value of
8481 the memory pointed to by ``ptr`` is dead. This means that it is known to
8482 never be used and has an undefined value. Any stores into the memory
8483 object following this intrinsic may be removed as dead.
8485 '``llvm.invariant.start``' Intrinsic
8486 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8493 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8498 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8499 a memory object will not change.
8504 The first argument is a constant integer representing the size of the
8505 object, or -1 if it is variable sized. The second argument is a pointer
8511 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8512 the return value, the referenced memory location is constant and
8515 '``llvm.invariant.end``' Intrinsic
8516 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8523 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8528 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8529 memory object are mutable.
8534 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8535 The second argument is a constant integer representing the size of the
8536 object, or -1 if it is variable sized and the third argument is a
8537 pointer to the object.
8542 This intrinsic indicates that the memory is mutable again.
8547 This class of intrinsics is designed to be generic and has no specific
8550 '``llvm.var.annotation``' Intrinsic
8551 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8558 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8563 The '``llvm.var.annotation``' intrinsic.
8568 The first argument is a pointer to a value, the second is a pointer to a
8569 global string, the third is a pointer to a global string which is the
8570 source file name, and the last argument is the line number.
8575 This intrinsic allows annotation of local variables with arbitrary
8576 strings. This can be useful for special purpose optimizations that want
8577 to look for these annotations. These have no other defined use; they are
8578 ignored by code generation and optimization.
8580 '``llvm.ptr.annotation.*``' Intrinsic
8581 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8586 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8587 pointer to an integer of any width. *NOTE* you must specify an address space for
8588 the pointer. The identifier for the default address space is the integer
8593 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8594 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8595 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8596 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8597 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8602 The '``llvm.ptr.annotation``' intrinsic.
8607 The first argument is a pointer to an integer value of arbitrary bitwidth
8608 (result of some expression), the second is a pointer to a global string, the
8609 third is a pointer to a global string which is the source file name, and the
8610 last argument is the line number. It returns the value of the first argument.
8615 This intrinsic allows annotation of a pointer to an integer with arbitrary
8616 strings. This can be useful for special purpose optimizations that want to look
8617 for these annotations. These have no other defined use; they are ignored by code
8618 generation and optimization.
8620 '``llvm.annotation.*``' Intrinsic
8621 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8626 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8627 any integer bit width.
8631 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8632 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8633 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8634 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8635 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8640 The '``llvm.annotation``' intrinsic.
8645 The first argument is an integer value (result of some expression), the
8646 second is a pointer to a global string, the third is a pointer to a
8647 global string which is the source file name, and the last argument is
8648 the line number. It returns the value of the first argument.
8653 This intrinsic allows annotations to be put on arbitrary expressions
8654 with arbitrary strings. This can be useful for special purpose
8655 optimizations that want to look for these annotations. These have no
8656 other defined use; they are ignored by code generation and optimization.
8658 '``llvm.trap``' Intrinsic
8659 ^^^^^^^^^^^^^^^^^^^^^^^^^
8666 declare void @llvm.trap() noreturn nounwind
8671 The '``llvm.trap``' intrinsic.
8681 This intrinsic is lowered to the target dependent trap instruction. If
8682 the target does not have a trap instruction, this intrinsic will be
8683 lowered to a call of the ``abort()`` function.
8685 '``llvm.debugtrap``' Intrinsic
8686 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8693 declare void @llvm.debugtrap() nounwind
8698 The '``llvm.debugtrap``' intrinsic.
8708 This intrinsic is lowered to code which is intended to cause an
8709 execution trap with the intention of requesting the attention of a
8712 '``llvm.stackprotector``' Intrinsic
8713 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8720 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8725 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8726 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8727 is placed on the stack before local variables.
8732 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8733 The first argument is the value loaded from the stack guard
8734 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8735 enough space to hold the value of the guard.
8740 This intrinsic causes the prologue/epilogue inserter to force the position of
8741 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8742 to ensure that if a local variable on the stack is overwritten, it will destroy
8743 the value of the guard. When the function exits, the guard on the stack is
8744 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8745 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8746 calling the ``__stack_chk_fail()`` function.
8748 '``llvm.stackprotectorcheck``' Intrinsic
8749 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8756 declare void @llvm.stackprotectorcheck(i8** <guard>)
8761 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8762 created stack protector and if they are not equal calls the
8763 ``__stack_chk_fail()`` function.
8768 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8769 the variable ``@__stack_chk_guard``.
8774 This intrinsic is provided to perform the stack protector check by comparing
8775 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8776 values do not match call the ``__stack_chk_fail()`` function.
8778 The reason to provide this as an IR level intrinsic instead of implementing it
8779 via other IR operations is that in order to perform this operation at the IR
8780 level without an intrinsic, one would need to create additional basic blocks to
8781 handle the success/failure cases. This makes it difficult to stop the stack
8782 protector check from disrupting sibling tail calls in Codegen. With this
8783 intrinsic, we are able to generate the stack protector basic blocks late in
8784 codegen after the tail call decision has occured.
8786 '``llvm.objectsize``' Intrinsic
8787 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8794 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8795 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8800 The ``llvm.objectsize`` intrinsic is designed to provide information to
8801 the optimizers to determine at compile time whether a) an operation
8802 (like memcpy) will overflow a buffer that corresponds to an object, or
8803 b) that a runtime check for overflow isn't necessary. An object in this
8804 context means an allocation of a specific class, structure, array, or
8810 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8811 argument is a pointer to or into the ``object``. The second argument is
8812 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8813 or -1 (if false) when the object size is unknown. The second argument
8814 only accepts constants.
8819 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8820 the size of the object concerned. If the size cannot be determined at
8821 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8822 on the ``min`` argument).
8824 '``llvm.expect``' Intrinsic
8825 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8832 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8833 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8838 The ``llvm.expect`` intrinsic provides information about expected (the
8839 most probable) value of ``val``, which can be used by optimizers.
8844 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8845 a value. The second argument is an expected value, this needs to be a
8846 constant value, variables are not allowed.
8851 This intrinsic is lowered to the ``val``.
8853 '``llvm.donothing``' Intrinsic
8854 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8861 declare void @llvm.donothing() nounwind readnone
8866 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8867 only intrinsic that can be called with an invoke instruction.
8877 This intrinsic does nothing, and it's removed by optimizers and ignored