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 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
784 which encodes the ``nop`` instruction:
788 define void @f() prefix i8 144 { ... }
790 Generally prefix data can be formed by encoding a relative branch instruction
791 which skips the metadata, as in this example of valid prefix data for the
792 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
796 %0 = type <{ i8, i8, i8* }>
798 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
800 A function may have prefix data but no body. This has similar semantics
801 to the ``available_externally`` linkage in that the data may be used by the
802 optimizers but will not be emitted in the object file.
809 Attribute groups are groups of attributes that are referenced by objects within
810 the IR. They are important for keeping ``.ll`` files readable, because a lot of
811 functions will use the same set of attributes. In the degenerative case of a
812 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
813 group will capture the important command line flags used to build that file.
815 An attribute group is a module-level object. To use an attribute group, an
816 object references the attribute group's ID (e.g. ``#37``). An object may refer
817 to more than one attribute group. In that situation, the attributes from the
818 different groups are merged.
820 Here is an example of attribute groups for a function that should always be
821 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
825 ; Target-independent attributes:
826 attributes #0 = { alwaysinline alignstack=4 }
828 ; Target-dependent attributes:
829 attributes #1 = { "no-sse" }
831 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
832 define void @f() #0 #1 { ... }
839 Function attributes are set to communicate additional information about
840 a function. Function attributes are considered to be part of the
841 function, not of the function type, so functions with different function
842 attributes can have the same function type.
844 Function attributes are simple keywords that follow the type specified.
845 If multiple attributes are needed, they are space separated. For
850 define void @f() noinline { ... }
851 define void @f() alwaysinline { ... }
852 define void @f() alwaysinline optsize { ... }
853 define void @f() optsize { ... }
856 This attribute indicates that, when emitting the prologue and
857 epilogue, the backend should forcibly align the stack pointer.
858 Specify the desired alignment, which must be a power of two, in
861 This attribute indicates that the inliner should attempt to inline
862 this function into callers whenever possible, ignoring any active
863 inlining size threshold for this caller.
865 This indicates that the callee function at a call site should be
866 recognized as a built-in function, even though the function's declaration
867 uses the ``nobuiltin`` attribute. This is only valid at call sites for
868 direct calls to functions which are declared with the ``nobuiltin``
871 This attribute indicates that this function is rarely called. When
872 computing edge weights, basic blocks post-dominated by a cold
873 function call are also considered to be cold; and, thus, given low
876 This attribute indicates that the source code contained a hint that
877 inlining this function is desirable (such as the "inline" keyword in
878 C/C++). It is just a hint; it imposes no requirements on the
881 This attribute suggests that optimization passes and code generator
882 passes make choices that keep the code size of this function as small
883 as possible and perform optimizations that may sacrifice runtime
884 performance in order to minimize the size of the generated code.
886 This attribute disables prologue / epilogue emission for the
887 function. This can have very system-specific consequences.
889 This indicates that the callee function at a call site is not recognized as
890 a built-in function. LLVM will retain the original call and not replace it
891 with equivalent code based on the semantics of the built-in function, unless
892 the call site uses the ``builtin`` attribute. This is valid at call sites
893 and on function declarations and definitions.
895 This attribute indicates that calls to the function cannot be
896 duplicated. A call to a ``noduplicate`` function may be moved
897 within its parent function, but may not be duplicated within
900 A function containing a ``noduplicate`` call may still
901 be an inlining candidate, provided that the call is not
902 duplicated by inlining. That implies that the function has
903 internal linkage and only has one call site, so the original
904 call is dead after inlining.
906 This attributes disables implicit floating point instructions.
908 This attribute indicates that the inliner should never inline this
909 function in any situation. This attribute may not be used together
910 with the ``alwaysinline`` attribute.
912 This attribute suppresses lazy symbol binding for the function. This
913 may make calls to the function faster, at the cost of extra program
914 startup time if the function is not called during program startup.
916 This attribute indicates that the code generator should not use a
917 red zone, even if the target-specific ABI normally permits it.
919 This function attribute indicates that the function never returns
920 normally. This produces undefined behavior at runtime if the
921 function ever does dynamically return.
923 This function attribute indicates that the function never returns
924 with an unwind or exceptional control flow. If the function does
925 unwind, its runtime behavior is undefined.
927 This function attribute indicates that the function is not optimized
928 by any optimization or code generator passes with the
929 exception of interprocedural optimization passes.
930 This attribute cannot be used together with the ``alwaysinline``
931 attribute; this attribute is also incompatible
932 with the ``minsize`` attribute and the ``optsize`` attribute.
934 The inliner should never inline this function in any situation.
935 Only functions with the ``alwaysinline`` attribute are valid
936 candidates for inlining inside the body of this function.
938 This attribute suggests that optimization passes and code generator
939 passes make choices that keep the code size of this function low,
940 and otherwise do optimizations specifically to reduce code size as
941 long as they do not significantly impact runtime performance.
943 On a function, this attribute indicates that the function computes its
944 result (or decides to unwind an exception) based strictly on its arguments,
945 without dereferencing any pointer arguments or otherwise accessing
946 any mutable state (e.g. memory, control registers, etc) visible to
947 caller functions. It does not write through any pointer arguments
948 (including ``byval`` arguments) and never changes any state visible
949 to callers. This means that it cannot unwind exceptions by calling
950 the ``C++`` exception throwing methods.
952 On an argument, this attribute indicates that the function does not
953 dereference that pointer argument, even though it may read or write the
954 memory that the pointer points to if accessed through other pointers.
956 On a function, this attribute indicates that the function does not write
957 through any pointer arguments (including ``byval`` arguments) or otherwise
958 modify any state (e.g. memory, control registers, etc) visible to
959 caller functions. It may dereference pointer arguments and read
960 state that may be set in the caller. A readonly function always
961 returns the same value (or unwinds an exception identically) when
962 called with the same set of arguments and global state. It cannot
963 unwind an exception by calling the ``C++`` exception throwing
966 On an argument, this attribute indicates that the function does not write
967 through this pointer argument, even though it may write to the memory that
968 the pointer points to.
970 This attribute indicates that this function can return twice. The C
971 ``setjmp`` is an example of such a function. The compiler disables
972 some optimizations (like tail calls) in the caller of these
975 This attribute indicates that AddressSanitizer checks
976 (dynamic address safety analysis) are enabled for this function.
978 This attribute indicates that MemorySanitizer checks (dynamic detection
979 of accesses to uninitialized memory) are enabled for this function.
981 This attribute indicates that ThreadSanitizer checks
982 (dynamic thread safety analysis) are enabled for this function.
984 This attribute indicates that the function should emit a stack
985 smashing protector. It is in the form of a "canary" --- a random value
986 placed on the stack before the local variables that's checked upon
987 return from the function to see if it has been overwritten. A
988 heuristic is used to determine if a function needs stack protectors
989 or not. The heuristic used will enable protectors for functions with:
991 - Character arrays larger than ``ssp-buffer-size`` (default 8).
992 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
993 - Calls to alloca() with variable sizes or constant sizes greater than
996 If a function that has an ``ssp`` attribute is inlined into a
997 function that doesn't have an ``ssp`` attribute, then the resulting
998 function will have an ``ssp`` attribute.
1000 This attribute indicates that the function should *always* emit a
1001 stack smashing protector. This overrides the ``ssp`` function
1004 If a function that has an ``sspreq`` attribute is inlined into a
1005 function that doesn't have an ``sspreq`` attribute or which has an
1006 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1007 an ``sspreq`` attribute.
1009 This attribute indicates that the function should emit a stack smashing
1010 protector. This attribute causes a strong heuristic to be used when
1011 determining if a function needs stack protectors. The strong heuristic
1012 will enable protectors for functions with:
1014 - Arrays of any size and type
1015 - Aggregates containing an array of any size and type.
1016 - Calls to alloca().
1017 - Local variables that have had their address taken.
1019 This overrides the ``ssp`` function attribute.
1021 If a function that has an ``sspstrong`` attribute is inlined into a
1022 function that doesn't have an ``sspstrong`` attribute, then the
1023 resulting function will have an ``sspstrong`` attribute.
1025 This attribute indicates that the ABI being targeted requires that
1026 an unwind table entry be produce for this function even if we can
1027 show that no exceptions passes by it. This is normally the case for
1028 the ELF x86-64 abi, but it can be disabled for some compilation
1033 Module-Level Inline Assembly
1034 ----------------------------
1036 Modules may contain "module-level inline asm" blocks, which corresponds
1037 to the GCC "file scope inline asm" blocks. These blocks are internally
1038 concatenated by LLVM and treated as a single unit, but may be separated
1039 in the ``.ll`` file if desired. The syntax is very simple:
1041 .. code-block:: llvm
1043 module asm "inline asm code goes here"
1044 module asm "more can go here"
1046 The strings can contain any character by escaping non-printable
1047 characters. The escape sequence used is simply "\\xx" where "xx" is the
1048 two digit hex code for the number.
1050 The inline asm code is simply printed to the machine code .s file when
1051 assembly code is generated.
1053 .. _langref_datalayout:
1058 A module may specify a target specific data layout string that specifies
1059 how data is to be laid out in memory. The syntax for the data layout is
1062 .. code-block:: llvm
1064 target datalayout = "layout specification"
1066 The *layout specification* consists of a list of specifications
1067 separated by the minus sign character ('-'). Each specification starts
1068 with a letter and may include other information after the letter to
1069 define some aspect of the data layout. The specifications accepted are
1073 Specifies that the target lays out data in big-endian form. That is,
1074 the bits with the most significance have the lowest address
1077 Specifies that the target lays out data in little-endian form. That
1078 is, the bits with the least significance have the lowest address
1081 Specifies the natural alignment of the stack in bits. Alignment
1082 promotion of stack variables is limited to the natural stack
1083 alignment to avoid dynamic stack realignment. The stack alignment
1084 must be a multiple of 8-bits. If omitted, the natural stack
1085 alignment defaults to "unspecified", which does not prevent any
1086 alignment promotions.
1087 ``p[n]:<size>:<abi>:<pref>``
1088 This specifies the *size* of a pointer and its ``<abi>`` and
1089 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1090 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1091 preceding ``:`` should be omitted too. The address space, ``n`` is
1092 optional, and if not specified, denotes the default address space 0.
1093 The value of ``n`` must be in the range [1,2^23).
1094 ``i<size>:<abi>:<pref>``
1095 This specifies the alignment for an integer type of a given bit
1096 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1097 ``v<size>:<abi>:<pref>``
1098 This specifies the alignment for a vector type of a given bit
1100 ``f<size>:<abi>:<pref>``
1101 This specifies the alignment for a floating point type of a given bit
1102 ``<size>``. Only values of ``<size>`` that are supported by the target
1103 will work. 32 (float) and 64 (double) are supported on all targets; 80
1104 or 128 (different flavors of long double) are also supported on some
1106 ``a<size>:<abi>:<pref>``
1107 This specifies the alignment for an aggregate type of a given bit
1109 ``s<size>:<abi>:<pref>``
1110 This specifies the alignment for a stack object of a given bit
1112 ``n<size1>:<size2>:<size3>...``
1113 This specifies a set of native integer widths for the target CPU in
1114 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1115 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1116 this set are considered to support most general arithmetic operations
1119 When constructing the data layout for a given target, LLVM starts with a
1120 default set of specifications which are then (possibly) overridden by
1121 the specifications in the ``datalayout`` keyword. The default
1122 specifications are given in this list:
1124 - ``E`` - big endian
1125 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1126 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1127 same as the default address space.
1128 - ``S0`` - natural stack alignment is unspecified
1129 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1130 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1131 - ``i16:16:16`` - i16 is 16-bit aligned
1132 - ``i32:32:32`` - i32 is 32-bit aligned
1133 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1134 alignment of 64-bits
1135 - ``f16:16:16`` - half is 16-bit aligned
1136 - ``f32:32:32`` - float is 32-bit aligned
1137 - ``f64:64:64`` - double is 64-bit aligned
1138 - ``f128:128:128`` - quad is 128-bit aligned
1139 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1140 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1141 - ``a0:0:64`` - aggregates are 64-bit aligned
1143 When LLVM is determining the alignment for a given type, it uses the
1146 #. If the type sought is an exact match for one of the specifications,
1147 that specification is used.
1148 #. If no match is found, and the type sought is an integer type, then
1149 the smallest integer type that is larger than the bitwidth of the
1150 sought type is used. If none of the specifications are larger than
1151 the bitwidth then the largest integer type is used. For example,
1152 given the default specifications above, the i7 type will use the
1153 alignment of i8 (next largest) while both i65 and i256 will use the
1154 alignment of i64 (largest specified).
1155 #. If no match is found, and the type sought is a vector type, then the
1156 largest vector type that is smaller than the sought vector type will
1157 be used as a fall back. This happens because <128 x double> can be
1158 implemented in terms of 64 <2 x double>, for example.
1160 The function of the data layout string may not be what you expect.
1161 Notably, this is not a specification from the frontend of what alignment
1162 the code generator should use.
1164 Instead, if specified, the target data layout is required to match what
1165 the ultimate *code generator* expects. This string is used by the
1166 mid-level optimizers to improve code, and this only works if it matches
1167 what the ultimate code generator uses. If you would like to generate IR
1168 that does not embed this target-specific detail into the IR, then you
1169 don't have to specify the string. This will disable some optimizations
1170 that require precise layout information, but this also prevents those
1171 optimizations from introducing target specificity into the IR.
1173 .. _pointeraliasing:
1175 Pointer Aliasing Rules
1176 ----------------------
1178 Any memory access must be done through a pointer value associated with
1179 an address range of the memory access, otherwise the behavior is
1180 undefined. Pointer values are associated with address ranges according
1181 to the following rules:
1183 - A pointer value is associated with the addresses associated with any
1184 value it is *based* on.
1185 - An address of a global variable is associated with the address range
1186 of the variable's storage.
1187 - The result value of an allocation instruction is associated with the
1188 address range of the allocated storage.
1189 - A null pointer in the default address-space is associated with no
1191 - An integer constant other than zero or a pointer value returned from
1192 a function not defined within LLVM may be associated with address
1193 ranges allocated through mechanisms other than those provided by
1194 LLVM. Such ranges shall not overlap with any ranges of addresses
1195 allocated by mechanisms provided by LLVM.
1197 A pointer value is *based* on another pointer value according to the
1200 - A pointer value formed from a ``getelementptr`` operation is *based*
1201 on the first operand of the ``getelementptr``.
1202 - The result value of a ``bitcast`` is *based* on the operand of the
1204 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1205 values that contribute (directly or indirectly) to the computation of
1206 the pointer's value.
1207 - The "*based* on" relationship is transitive.
1209 Note that this definition of *"based"* is intentionally similar to the
1210 definition of *"based"* in C99, though it is slightly weaker.
1212 LLVM IR does not associate types with memory. The result type of a
1213 ``load`` merely indicates the size and alignment of the memory from
1214 which to load, as well as the interpretation of the value. The first
1215 operand type of a ``store`` similarly only indicates the size and
1216 alignment of the store.
1218 Consequently, type-based alias analysis, aka TBAA, aka
1219 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1220 :ref:`Metadata <metadata>` may be used to encode additional information
1221 which specialized optimization passes may use to implement type-based
1226 Volatile Memory Accesses
1227 ------------------------
1229 Certain memory accesses, such as :ref:`load <i_load>`'s,
1230 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1231 marked ``volatile``. The optimizers must not change the number of
1232 volatile operations or change their order of execution relative to other
1233 volatile operations. The optimizers *may* change the order of volatile
1234 operations relative to non-volatile operations. This is not Java's
1235 "volatile" and has no cross-thread synchronization behavior.
1237 IR-level volatile loads and stores cannot safely be optimized into
1238 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1239 flagged volatile. Likewise, the backend should never split or merge
1240 target-legal volatile load/store instructions.
1242 .. admonition:: Rationale
1244 Platforms may rely on volatile loads and stores of natively supported
1245 data width to be executed as single instruction. For example, in C
1246 this holds for an l-value of volatile primitive type with native
1247 hardware support, but not necessarily for aggregate types. The
1248 frontend upholds these expectations, which are intentionally
1249 unspecified in the IR. The rules above ensure that IR transformation
1250 do not violate the frontend's contract with the language.
1254 Memory Model for Concurrent Operations
1255 --------------------------------------
1257 The LLVM IR does not define any way to start parallel threads of
1258 execution or to register signal handlers. Nonetheless, there are
1259 platform-specific ways to create them, and we define LLVM IR's behavior
1260 in their presence. This model is inspired by the C++0x memory model.
1262 For a more informal introduction to this model, see the :doc:`Atomics`.
1264 We define a *happens-before* partial order as the least partial order
1267 - Is a superset of single-thread program order, and
1268 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1269 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1270 techniques, like pthread locks, thread creation, thread joining,
1271 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1272 Constraints <ordering>`).
1274 Note that program order does not introduce *happens-before* edges
1275 between a thread and signals executing inside that thread.
1277 Every (defined) read operation (load instructions, memcpy, atomic
1278 loads/read-modify-writes, etc.) R reads a series of bytes written by
1279 (defined) write operations (store instructions, atomic
1280 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1281 section, initialized globals are considered to have a write of the
1282 initializer which is atomic and happens before any other read or write
1283 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1284 may see any write to the same byte, except:
1286 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1287 write\ :sub:`2` happens before R\ :sub:`byte`, then
1288 R\ :sub:`byte` does not see write\ :sub:`1`.
1289 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1290 R\ :sub:`byte` does not see write\ :sub:`3`.
1292 Given that definition, R\ :sub:`byte` is defined as follows:
1294 - If R is volatile, the result is target-dependent. (Volatile is
1295 supposed to give guarantees which can support ``sig_atomic_t`` in
1296 C/C++, and may be used for accesses to addresses which do not behave
1297 like normal memory. It does not generally provide cross-thread
1299 - Otherwise, if there is no write to the same byte that happens before
1300 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1301 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1302 R\ :sub:`byte` returns the value written by that write.
1303 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1304 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1305 Memory Ordering Constraints <ordering>` section for additional
1306 constraints on how the choice is made.
1307 - Otherwise R\ :sub:`byte` returns ``undef``.
1309 R returns the value composed of the series of bytes it read. This
1310 implies that some bytes within the value may be ``undef`` **without**
1311 the entire value being ``undef``. Note that this only defines the
1312 semantics of the operation; it doesn't mean that targets will emit more
1313 than one instruction to read the series of bytes.
1315 Note that in cases where none of the atomic intrinsics are used, this
1316 model places only one restriction on IR transformations on top of what
1317 is required for single-threaded execution: introducing a store to a byte
1318 which might not otherwise be stored is not allowed in general.
1319 (Specifically, in the case where another thread might write to and read
1320 from an address, introducing a store can change a load that may see
1321 exactly one write into a load that may see multiple writes.)
1325 Atomic Memory Ordering Constraints
1326 ----------------------------------
1328 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1329 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1330 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1331 an ordering parameter that determines which other atomic instructions on
1332 the same address they *synchronize with*. These semantics are borrowed
1333 from Java and C++0x, but are somewhat more colloquial. If these
1334 descriptions aren't precise enough, check those specs (see spec
1335 references in the :doc:`atomics guide <Atomics>`).
1336 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1337 differently since they don't take an address. See that instruction's
1338 documentation for details.
1340 For a simpler introduction to the ordering constraints, see the
1344 The set of values that can be read is governed by the happens-before
1345 partial order. A value cannot be read unless some operation wrote
1346 it. This is intended to provide a guarantee strong enough to model
1347 Java's non-volatile shared variables. This ordering cannot be
1348 specified for read-modify-write operations; it is not strong enough
1349 to make them atomic in any interesting way.
1351 In addition to the guarantees of ``unordered``, there is a single
1352 total order for modifications by ``monotonic`` operations on each
1353 address. All modification orders must be compatible with the
1354 happens-before order. There is no guarantee that the modification
1355 orders can be combined to a global total order for the whole program
1356 (and this often will not be possible). The read in an atomic
1357 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1358 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1359 order immediately before the value it writes. If one atomic read
1360 happens before another atomic read of the same address, the later
1361 read must see the same value or a later value in the address's
1362 modification order. This disallows reordering of ``monotonic`` (or
1363 stronger) operations on the same address. If an address is written
1364 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1365 read that address repeatedly, the other threads must eventually see
1366 the write. This corresponds to the C++0x/C1x
1367 ``memory_order_relaxed``.
1369 In addition to the guarantees of ``monotonic``, a
1370 *synchronizes-with* edge may be formed with a ``release`` operation.
1371 This is intended to model C++'s ``memory_order_acquire``.
1373 In addition to the guarantees of ``monotonic``, if this operation
1374 writes a value which is subsequently read by an ``acquire``
1375 operation, it *synchronizes-with* that operation. (This isn't a
1376 complete description; see the C++0x definition of a release
1377 sequence.) This corresponds to the C++0x/C1x
1378 ``memory_order_release``.
1379 ``acq_rel`` (acquire+release)
1380 Acts as both an ``acquire`` and ``release`` operation on its
1381 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1382 ``seq_cst`` (sequentially consistent)
1383 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1384 operation which only reads, ``release`` for an operation which only
1385 writes), there is a global total order on all
1386 sequentially-consistent operations on all addresses, which is
1387 consistent with the *happens-before* partial order and with the
1388 modification orders of all the affected addresses. Each
1389 sequentially-consistent read sees the last preceding write to the
1390 same address in this global order. This corresponds to the C++0x/C1x
1391 ``memory_order_seq_cst`` and Java volatile.
1395 If an atomic operation is marked ``singlethread``, it only *synchronizes
1396 with* or participates in modification and seq\_cst total orderings with
1397 other operations running in the same thread (for example, in signal
1405 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1406 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1407 :ref:`frem <i_frem>`) have the following flags that can set to enable
1408 otherwise unsafe floating point operations
1411 No NaNs - Allow optimizations to assume the arguments and result are not
1412 NaN. Such optimizations are required to retain defined behavior over
1413 NaNs, but the value of the result is undefined.
1416 No Infs - Allow optimizations to assume the arguments and result are not
1417 +/-Inf. Such optimizations are required to retain defined behavior over
1418 +/-Inf, but the value of the result is undefined.
1421 No Signed Zeros - Allow optimizations to treat the sign of a zero
1422 argument or result as insignificant.
1425 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1426 argument rather than perform division.
1429 Fast - Allow algebraically equivalent transformations that may
1430 dramatically change results in floating point (e.g. reassociate). This
1431 flag implies all the others.
1438 The LLVM type system is one of the most important features of the
1439 intermediate representation. Being typed enables a number of
1440 optimizations to be performed on the intermediate representation
1441 directly, without having to do extra analyses on the side before the
1442 transformation. A strong type system makes it easier to read the
1443 generated code and enables novel analyses and transformations that are
1444 not feasible to perform on normal three address code representations.
1446 .. _typeclassifications:
1448 Type Classifications
1449 --------------------
1451 The types fall into a few useful classifications:
1460 * - :ref:`integer <t_integer>`
1461 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1464 * - :ref:`floating point <t_floating>`
1465 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1473 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1474 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1475 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1476 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1478 * - :ref:`primitive <t_primitive>`
1479 - :ref:`label <t_label>`,
1480 :ref:`void <t_void>`,
1481 :ref:`integer <t_integer>`,
1482 :ref:`floating point <t_floating>`,
1483 :ref:`x86mmx <t_x86mmx>`,
1484 :ref:`metadata <t_metadata>`.
1486 * - :ref:`derived <t_derived>`
1487 - :ref:`array <t_array>`,
1488 :ref:`function <t_function>`,
1489 :ref:`pointer <t_pointer>`,
1490 :ref:`structure <t_struct>`,
1491 :ref:`vector <t_vector>`,
1492 :ref:`opaque <t_opaque>`.
1494 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1495 Values of these types are the only ones which can be produced by
1503 The primitive types are the fundamental building blocks of the LLVM
1514 The integer type is a very simple type that simply specifies an
1515 arbitrary bit width for the integer type desired. Any bit width from 1
1516 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1525 The number of bits the integer will occupy is specified by the ``N``
1531 +----------------+------------------------------------------------+
1532 | ``i1`` | a single-bit integer. |
1533 +----------------+------------------------------------------------+
1534 | ``i32`` | a 32-bit integer. |
1535 +----------------+------------------------------------------------+
1536 | ``i1942652`` | a really big integer of over 1 million bits. |
1537 +----------------+------------------------------------------------+
1541 Floating Point Types
1542 ^^^^^^^^^^^^^^^^^^^^
1551 - 16-bit floating point value
1554 - 32-bit floating point value
1557 - 64-bit floating point value
1560 - 128-bit floating point value (112-bit mantissa)
1563 - 80-bit floating point value (X87)
1566 - 128-bit floating point value (two 64-bits)
1576 The x86mmx type represents a value held in an MMX register on an x86
1577 machine. The operations allowed on it are quite limited: parameters and
1578 return values, load and store, and bitcast. User-specified MMX
1579 instructions are represented as intrinsic or asm calls with arguments
1580 and/or results of this type. There are no arrays, vectors or constants
1598 The void type does not represent any value and has no size.
1615 The label type represents code labels.
1632 The metadata type represents embedded metadata. No derived types may be
1633 created from metadata except for :ref:`function <t_function>` arguments.
1647 The real power in LLVM comes from the derived types in the system. This
1648 is what allows a programmer to represent arrays, functions, pointers,
1649 and other useful types. Each of these types contain one or more element
1650 types which may be a primitive type, or another derived type. For
1651 example, it is possible to have a two dimensional array, using an array
1652 as the element type of another array.
1659 Aggregate Types are a subset of derived types that can contain multiple
1660 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1661 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1672 The array type is a very simple derived type that arranges elements
1673 sequentially in memory. The array type requires a size (number of
1674 elements) and an underlying data type.
1681 [<# elements> x <elementtype>]
1683 The number of elements is a constant integer value; ``elementtype`` may
1684 be any type with a size.
1689 +------------------+--------------------------------------+
1690 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1691 +------------------+--------------------------------------+
1692 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1693 +------------------+--------------------------------------+
1694 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1695 +------------------+--------------------------------------+
1697 Here are some examples of multidimensional arrays:
1699 +-----------------------------+----------------------------------------------------------+
1700 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1701 +-----------------------------+----------------------------------------------------------+
1702 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1703 +-----------------------------+----------------------------------------------------------+
1704 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1705 +-----------------------------+----------------------------------------------------------+
1707 There is no restriction on indexing beyond the end of the array implied
1708 by a static type (though there are restrictions on indexing beyond the
1709 bounds of an allocated object in some cases). This means that
1710 single-dimension 'variable sized array' addressing can be implemented in
1711 LLVM with a zero length array type. An implementation of 'pascal style
1712 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1723 The function type can be thought of as a function signature. It consists
1724 of a return type and a list of formal parameter types. The return type
1725 of a function type is a first class type or a void type.
1732 <returntype> (<parameter list>)
1734 ...where '``<parameter list>``' is a comma-separated list of type
1735 specifiers. Optionally, the parameter list may include a type ``...``,
1736 which indicates that the function takes a variable number of arguments.
1737 Variable argument functions can access their arguments with the
1738 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1739 '``<returntype>``' is any type except :ref:`label <t_label>`.
1744 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1745 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1746 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1747 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1748 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1749 | ``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. |
1750 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1751 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1752 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1762 The structure type is used to represent a collection of data members
1763 together in memory. The elements of a structure may be any type that has
1766 Structures in memory are accessed using '``load``' and '``store``' by
1767 getting a pointer to a field with the '``getelementptr``' instruction.
1768 Structures in registers are accessed using the '``extractvalue``' and
1769 '``insertvalue``' instructions.
1771 Structures may optionally be "packed" structures, which indicate that
1772 the alignment of the struct is one byte, and that there is no padding
1773 between the elements. In non-packed structs, padding between field types
1774 is inserted as defined by the DataLayout string in the module, which is
1775 required to match what the underlying code generator expects.
1777 Structures can either be "literal" or "identified". A literal structure
1778 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1779 identified types are always defined at the top level with a name.
1780 Literal types are uniqued by their contents and can never be recursive
1781 or opaque since there is no way to write one. Identified types can be
1782 recursive, can be opaqued, and are never uniqued.
1789 %T1 = type { <type list> } ; Identified normal struct type
1790 %T2 = type <{ <type list> }> ; Identified packed struct type
1795 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1796 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1797 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1798 | ``{ 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``. |
1799 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1800 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1801 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1805 Opaque Structure Types
1806 ^^^^^^^^^^^^^^^^^^^^^^
1811 Opaque structure types are used to represent named structure types that
1812 do not have a body specified. This corresponds (for example) to the C
1813 notion of a forward declared structure.
1826 +--------------+-------------------+
1827 | ``opaque`` | An opaque type. |
1828 +--------------+-------------------+
1838 The pointer type is used to specify memory locations. Pointers are
1839 commonly used to reference objects in memory.
1841 Pointer types may have an optional address space attribute defining the
1842 numbered address space where the pointed-to object resides. The default
1843 address space is number zero. The semantics of non-zero address spaces
1844 are target-specific.
1846 Note that LLVM does not permit pointers to void (``void*``) nor does it
1847 permit pointers to labels (``label*``). Use ``i8*`` instead.
1859 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1860 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1861 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1862 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1863 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1864 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1865 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1875 A vector type is a simple derived type that represents a vector of
1876 elements. Vector types are used when multiple primitive data are
1877 operated in parallel using a single instruction (SIMD). A vector type
1878 requires a size (number of elements) and an underlying primitive data
1879 type. Vector types are considered :ref:`first class <t_firstclass>`.
1886 < <# elements> x <elementtype> >
1888 The number of elements is a constant integer value larger than 0;
1889 elementtype may be any integer or floating point type, or a pointer to
1890 these types. Vectors of size zero are not allowed.
1895 +-------------------+--------------------------------------------------+
1896 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1897 +-------------------+--------------------------------------------------+
1898 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1899 +-------------------+--------------------------------------------------+
1900 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1901 +-------------------+--------------------------------------------------+
1902 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1903 +-------------------+--------------------------------------------------+
1908 LLVM has several different basic types of constants. This section
1909 describes them all and their syntax.
1914 **Boolean constants**
1915 The two strings '``true``' and '``false``' are both valid constants
1917 **Integer constants**
1918 Standard integers (such as '4') are constants of the
1919 :ref:`integer <t_integer>` type. Negative numbers may be used with
1921 **Floating point constants**
1922 Floating point constants use standard decimal notation (e.g.
1923 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1924 hexadecimal notation (see below). The assembler requires the exact
1925 decimal value of a floating-point constant. For example, the
1926 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1927 decimal in binary. Floating point constants must have a :ref:`floating
1928 point <t_floating>` type.
1929 **Null pointer constants**
1930 The identifier '``null``' is recognized as a null pointer constant
1931 and must be of :ref:`pointer type <t_pointer>`.
1933 The one non-intuitive notation for constants is the hexadecimal form of
1934 floating point constants. For example, the form
1935 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1936 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1937 constants are required (and the only time that they are generated by the
1938 disassembler) is when a floating point constant must be emitted but it
1939 cannot be represented as a decimal floating point number in a reasonable
1940 number of digits. For example, NaN's, infinities, and other special
1941 values are represented in their IEEE hexadecimal format so that assembly
1942 and disassembly do not cause any bits to change in the constants.
1944 When using the hexadecimal form, constants of types half, float, and
1945 double are represented using the 16-digit form shown above (which
1946 matches the IEEE754 representation for double); half and float values
1947 must, however, be exactly representable as IEEE 754 half and single
1948 precision, respectively. Hexadecimal format is always used for long
1949 double, and there are three forms of long double. The 80-bit format used
1950 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1951 128-bit format used by PowerPC (two adjacent doubles) is represented by
1952 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1953 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1954 will only work if they match the long double format on your target.
1955 The IEEE 16-bit format (half precision) is represented by ``0xH``
1956 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1957 (sign bit at the left).
1959 There are no constants of type x86mmx.
1961 .. _complexconstants:
1966 Complex constants are a (potentially recursive) combination of simple
1967 constants and smaller complex constants.
1969 **Structure constants**
1970 Structure constants are represented with notation similar to
1971 structure type definitions (a comma separated list of elements,
1972 surrounded by braces (``{}``)). For example:
1973 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1974 "``@G = external global i32``". Structure constants must have
1975 :ref:`structure type <t_struct>`, and the number and types of elements
1976 must match those specified by the type.
1978 Array constants are represented with notation similar to array type
1979 definitions (a comma separated list of elements, surrounded by
1980 square brackets (``[]``)). For example:
1981 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1982 :ref:`array type <t_array>`, and the number and types of elements must
1983 match those specified by the type.
1984 **Vector constants**
1985 Vector constants are represented with notation similar to vector
1986 type definitions (a comma separated list of elements, surrounded by
1987 less-than/greater-than's (``<>``)). For example:
1988 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1989 must have :ref:`vector type <t_vector>`, and the number and types of
1990 elements must match those specified by the type.
1991 **Zero initialization**
1992 The string '``zeroinitializer``' can be used to zero initialize a
1993 value to zero of *any* type, including scalar and
1994 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1995 having to print large zero initializers (e.g. for large arrays) and
1996 is always exactly equivalent to using explicit zero initializers.
1998 A metadata node is a structure-like constant with :ref:`metadata
1999 type <t_metadata>`. For example:
2000 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2001 constants that are meant to be interpreted as part of the
2002 instruction stream, metadata is a place to attach additional
2003 information such as debug info.
2005 Global Variable and Function Addresses
2006 --------------------------------------
2008 The addresses of :ref:`global variables <globalvars>` and
2009 :ref:`functions <functionstructure>` are always implicitly valid
2010 (link-time) constants. These constants are explicitly referenced when
2011 the :ref:`identifier for the global <identifiers>` is used and always have
2012 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2015 .. code-block:: llvm
2019 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2026 The string '``undef``' can be used anywhere a constant is expected, and
2027 indicates that the user of the value may receive an unspecified
2028 bit-pattern. Undefined values may be of any type (other than '``label``'
2029 or '``void``') and be used anywhere a constant is permitted.
2031 Undefined values are useful because they indicate to the compiler that
2032 the program is well defined no matter what value is used. This gives the
2033 compiler more freedom to optimize. Here are some examples of
2034 (potentially surprising) transformations that are valid (in pseudo IR):
2036 .. code-block:: llvm
2046 This is safe because all of the output bits are affected by the undef
2047 bits. Any output bit can have a zero or one depending on the input bits.
2049 .. code-block:: llvm
2060 These logical operations have bits that are not always affected by the
2061 input. For example, if ``%X`` has a zero bit, then the output of the
2062 '``and``' operation will always be a zero for that bit, no matter what
2063 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2064 optimize or assume that the result of the '``and``' is '``undef``'.
2065 However, it is safe to assume that all bits of the '``undef``' could be
2066 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2067 all the bits of the '``undef``' operand to the '``or``' could be set,
2068 allowing the '``or``' to be folded to -1.
2070 .. code-block:: llvm
2072 %A = select undef, %X, %Y
2073 %B = select undef, 42, %Y
2074 %C = select %X, %Y, undef
2084 This set of examples shows that undefined '``select``' (and conditional
2085 branch) conditions can go *either way*, but they have to come from one
2086 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2087 both known to have a clear low bit, then ``%A`` would have to have a
2088 cleared low bit. However, in the ``%C`` example, the optimizer is
2089 allowed to assume that the '``undef``' operand could be the same as
2090 ``%Y``, allowing the whole '``select``' to be eliminated.
2092 .. code-block:: llvm
2094 %A = xor undef, undef
2111 This example points out that two '``undef``' operands are not
2112 necessarily the same. This can be surprising to people (and also matches
2113 C semantics) where they assume that "``X^X``" is always zero, even if
2114 ``X`` is undefined. This isn't true for a number of reasons, but the
2115 short answer is that an '``undef``' "variable" can arbitrarily change
2116 its value over its "live range". This is true because the variable
2117 doesn't actually *have a live range*. Instead, the value is logically
2118 read from arbitrary registers that happen to be around when needed, so
2119 the value is not necessarily consistent over time. In fact, ``%A`` and
2120 ``%C`` need to have the same semantics or the core LLVM "replace all
2121 uses with" concept would not hold.
2123 .. code-block:: llvm
2131 These examples show the crucial difference between an *undefined value*
2132 and *undefined behavior*. An undefined value (like '``undef``') is
2133 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2134 operation can be constant folded to '``undef``', because the '``undef``'
2135 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2136 However, in the second example, we can make a more aggressive
2137 assumption: because the ``undef`` is allowed to be an arbitrary value,
2138 we are allowed to assume that it could be zero. Since a divide by zero
2139 has *undefined behavior*, we are allowed to assume that the operation
2140 does not execute at all. This allows us to delete the divide and all
2141 code after it. Because the undefined operation "can't happen", the
2142 optimizer can assume that it occurs in dead code.
2144 .. code-block:: llvm
2146 a: store undef -> %X
2147 b: store %X -> undef
2152 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2153 value can be assumed to not have any effect; we can assume that the
2154 value is overwritten with bits that happen to match what was already
2155 there. However, a store *to* an undefined location could clobber
2156 arbitrary memory, therefore, it has undefined behavior.
2163 Poison values are similar to :ref:`undef values <undefvalues>`, however
2164 they also represent the fact that an instruction or constant expression
2165 which cannot evoke side effects has nevertheless detected a condition
2166 which results in undefined behavior.
2168 There is currently no way of representing a poison value in the IR; they
2169 only exist when produced by operations such as :ref:`add <i_add>` with
2172 Poison value behavior is defined in terms of value *dependence*:
2174 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2175 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2176 their dynamic predecessor basic block.
2177 - Function arguments depend on the corresponding actual argument values
2178 in the dynamic callers of their functions.
2179 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2180 instructions that dynamically transfer control back to them.
2181 - :ref:`Invoke <i_invoke>` instructions depend on the
2182 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2183 call instructions that dynamically transfer control back to them.
2184 - Non-volatile loads and stores depend on the most recent stores to all
2185 of the referenced memory addresses, following the order in the IR
2186 (including loads and stores implied by intrinsics such as
2187 :ref:`@llvm.memcpy <int_memcpy>`.)
2188 - An instruction with externally visible side effects depends on the
2189 most recent preceding instruction with externally visible side
2190 effects, following the order in the IR. (This includes :ref:`volatile
2191 operations <volatile>`.)
2192 - An instruction *control-depends* on a :ref:`terminator
2193 instruction <terminators>` if the terminator instruction has
2194 multiple successors and the instruction is always executed when
2195 control transfers to one of the successors, and may not be executed
2196 when control is transferred to another.
2197 - Additionally, an instruction also *control-depends* on a terminator
2198 instruction if the set of instructions it otherwise depends on would
2199 be different if the terminator had transferred control to a different
2201 - Dependence is transitive.
2203 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2204 with the additional affect that any instruction which has a *dependence*
2205 on a poison value has undefined behavior.
2207 Here are some examples:
2209 .. code-block:: llvm
2212 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2213 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2214 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2215 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2217 store i32 %poison, i32* @g ; Poison value stored to memory.
2218 %poison2 = load i32* @g ; Poison value loaded back from memory.
2220 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2222 %narrowaddr = bitcast i32* @g to i16*
2223 %wideaddr = bitcast i32* @g to i64*
2224 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2225 %poison4 = load i64* %wideaddr ; Returns a poison value.
2227 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2228 br i1 %cmp, label %true, label %end ; Branch to either destination.
2231 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2232 ; it has undefined behavior.
2236 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2237 ; Both edges into this PHI are
2238 ; control-dependent on %cmp, so this
2239 ; always results in a poison value.
2241 store volatile i32 0, i32* @g ; This would depend on the store in %true
2242 ; if %cmp is true, or the store in %entry
2243 ; otherwise, so this is undefined behavior.
2245 br i1 %cmp, label %second_true, label %second_end
2246 ; The same branch again, but this time the
2247 ; true block doesn't have side effects.
2254 store volatile i32 0, i32* @g ; This time, the instruction always depends
2255 ; on the store in %end. Also, it is
2256 ; control-equivalent to %end, so this is
2257 ; well-defined (ignoring earlier undefined
2258 ; behavior in this example).
2262 Addresses of Basic Blocks
2263 -------------------------
2265 ``blockaddress(@function, %block)``
2267 The '``blockaddress``' constant computes the address of the specified
2268 basic block in the specified function, and always has an ``i8*`` type.
2269 Taking the address of the entry block is illegal.
2271 This value only has defined behavior when used as an operand to the
2272 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2273 against null. Pointer equality tests between labels addresses results in
2274 undefined behavior --- though, again, comparison against null is ok, and
2275 no label is equal to the null pointer. This may be passed around as an
2276 opaque pointer sized value as long as the bits are not inspected. This
2277 allows ``ptrtoint`` and arithmetic to be performed on these values so
2278 long as the original value is reconstituted before the ``indirectbr``
2281 Finally, some targets may provide defined semantics when using the value
2282 as the operand to an inline assembly, but that is target specific.
2286 Constant Expressions
2287 --------------------
2289 Constant expressions are used to allow expressions involving other
2290 constants to be used as constants. Constant expressions may be of any
2291 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2292 that does not have side effects (e.g. load and call are not supported).
2293 The following is the syntax for constant expressions:
2295 ``trunc (CST to TYPE)``
2296 Truncate a constant to another type. The bit size of CST must be
2297 larger than the bit size of TYPE. Both types must be integers.
2298 ``zext (CST to TYPE)``
2299 Zero extend a constant to another type. The bit size of CST must be
2300 smaller than the bit size of TYPE. Both types must be integers.
2301 ``sext (CST to TYPE)``
2302 Sign extend a constant to another type. The bit size of CST must be
2303 smaller than the bit size of TYPE. Both types must be integers.
2304 ``fptrunc (CST to TYPE)``
2305 Truncate a floating point constant to another floating point type.
2306 The size of CST must be larger than the size of TYPE. Both types
2307 must be floating point.
2308 ``fpext (CST to TYPE)``
2309 Floating point extend a constant to another type. The size of CST
2310 must be smaller or equal to the size of TYPE. Both types must be
2312 ``fptoui (CST to TYPE)``
2313 Convert a floating point constant to the corresponding unsigned
2314 integer constant. TYPE must be a scalar or vector integer type. CST
2315 must be of scalar or vector floating point type. Both CST and TYPE
2316 must be scalars, or vectors of the same number of elements. If the
2317 value won't fit in the integer type, the results are undefined.
2318 ``fptosi (CST to TYPE)``
2319 Convert a floating point constant to the corresponding signed
2320 integer constant. TYPE must be a scalar or vector integer type. CST
2321 must be of scalar or vector floating point type. Both CST and TYPE
2322 must be scalars, or vectors of the same number of elements. If the
2323 value won't fit in the integer type, the results are undefined.
2324 ``uitofp (CST to TYPE)``
2325 Convert an unsigned integer constant to the corresponding floating
2326 point constant. TYPE must be a scalar or vector floating point type.
2327 CST must be of scalar or vector integer type. Both CST and TYPE must
2328 be scalars, or vectors of the same number of elements. If the value
2329 won't fit in the floating point type, the results are undefined.
2330 ``sitofp (CST to TYPE)``
2331 Convert a signed integer constant to the corresponding floating
2332 point constant. TYPE must be a scalar or vector floating point type.
2333 CST must be of scalar or vector integer type. Both CST and TYPE must
2334 be scalars, or vectors of the same number of elements. If the value
2335 won't fit in the floating point type, the results are undefined.
2336 ``ptrtoint (CST to TYPE)``
2337 Convert a pointer typed constant to the corresponding integer
2338 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2339 pointer type. The ``CST`` value is zero extended, truncated, or
2340 unchanged to make it fit in ``TYPE``.
2341 ``inttoptr (CST to TYPE)``
2342 Convert an integer constant to a pointer constant. TYPE must be a
2343 pointer type. CST must be of integer type. The CST value is zero
2344 extended, truncated, or unchanged to make it fit in a pointer size.
2345 This one is *really* dangerous!
2346 ``bitcast (CST to TYPE)``
2347 Convert a constant, CST, to another TYPE. The constraints of the
2348 operands are the same as those for the :ref:`bitcast
2349 instruction <i_bitcast>`.
2350 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2351 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2352 constants. As with the :ref:`getelementptr <i_getelementptr>`
2353 instruction, the index list may have zero or more indexes, which are
2354 required to make sense for the type of "CSTPTR".
2355 ``select (COND, VAL1, VAL2)``
2356 Perform the :ref:`select operation <i_select>` on constants.
2357 ``icmp COND (VAL1, VAL2)``
2358 Performs the :ref:`icmp operation <i_icmp>` on constants.
2359 ``fcmp COND (VAL1, VAL2)``
2360 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2361 ``extractelement (VAL, IDX)``
2362 Perform the :ref:`extractelement operation <i_extractelement>` on
2364 ``insertelement (VAL, ELT, IDX)``
2365 Perform the :ref:`insertelement operation <i_insertelement>` on
2367 ``shufflevector (VEC1, VEC2, IDXMASK)``
2368 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2370 ``extractvalue (VAL, IDX0, IDX1, ...)``
2371 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2372 constants. The index list is interpreted in a similar manner as
2373 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2374 least one index value must be specified.
2375 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2376 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2377 The index list is interpreted in a similar manner as indices in a
2378 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2379 value must be specified.
2380 ``OPCODE (LHS, RHS)``
2381 Perform the specified operation of the LHS and RHS constants. OPCODE
2382 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2383 binary <bitwiseops>` operations. The constraints on operands are
2384 the same as those for the corresponding instruction (e.g. no bitwise
2385 operations on floating point values are allowed).
2392 Inline Assembler Expressions
2393 ----------------------------
2395 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2396 Inline Assembly <moduleasm>`) through the use of a special value. This
2397 value represents the inline assembler as a string (containing the
2398 instructions to emit), a list of operand constraints (stored as a
2399 string), a flag that indicates whether or not the inline asm expression
2400 has side effects, and a flag indicating whether the function containing
2401 the asm needs to align its stack conservatively. An example inline
2402 assembler expression is:
2404 .. code-block:: llvm
2406 i32 (i32) asm "bswap $0", "=r,r"
2408 Inline assembler expressions may **only** be used as the callee operand
2409 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2410 Thus, typically we have:
2412 .. code-block:: llvm
2414 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2416 Inline asms with side effects not visible in the constraint list must be
2417 marked as having side effects. This is done through the use of the
2418 '``sideeffect``' keyword, like so:
2420 .. code-block:: llvm
2422 call void asm sideeffect "eieio", ""()
2424 In some cases inline asms will contain code that will not work unless
2425 the stack is aligned in some way, such as calls or SSE instructions on
2426 x86, yet will not contain code that does that alignment within the asm.
2427 The compiler should make conservative assumptions about what the asm
2428 might contain and should generate its usual stack alignment code in the
2429 prologue if the '``alignstack``' keyword is present:
2431 .. code-block:: llvm
2433 call void asm alignstack "eieio", ""()
2435 Inline asms also support using non-standard assembly dialects. The
2436 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2437 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2438 the only supported dialects. An example is:
2440 .. code-block:: llvm
2442 call void asm inteldialect "eieio", ""()
2444 If multiple keywords appear the '``sideeffect``' keyword must come
2445 first, the '``alignstack``' keyword second and the '``inteldialect``'
2451 The call instructions that wrap inline asm nodes may have a
2452 "``!srcloc``" MDNode attached to it that contains a list of constant
2453 integers. If present, the code generator will use the integer as the
2454 location cookie value when report errors through the ``LLVMContext``
2455 error reporting mechanisms. This allows a front-end to correlate backend
2456 errors that occur with inline asm back to the source code that produced
2459 .. code-block:: llvm
2461 call void asm sideeffect "something bad", ""(), !srcloc !42
2463 !42 = !{ i32 1234567 }
2465 It is up to the front-end to make sense of the magic numbers it places
2466 in the IR. If the MDNode contains multiple constants, the code generator
2467 will use the one that corresponds to the line of the asm that the error
2472 Metadata Nodes and Metadata Strings
2473 -----------------------------------
2475 LLVM IR allows metadata to be attached to instructions in the program
2476 that can convey extra information about the code to the optimizers and
2477 code generator. One example application of metadata is source-level
2478 debug information. There are two metadata primitives: strings and nodes.
2479 All metadata has the ``metadata`` type and is identified in syntax by a
2480 preceding exclamation point ('``!``').
2482 A metadata string is a string surrounded by double quotes. It can
2483 contain any character by escaping non-printable characters with
2484 "``\xx``" where "``xx``" is the two digit hex code. For example:
2487 Metadata nodes are represented with notation similar to structure
2488 constants (a comma separated list of elements, surrounded by braces and
2489 preceded by an exclamation point). Metadata nodes can have any values as
2490 their operand. For example:
2492 .. code-block:: llvm
2494 !{ metadata !"test\00", i32 10}
2496 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2497 metadata nodes, which can be looked up in the module symbol table. For
2500 .. code-block:: llvm
2502 !foo = metadata !{!4, !3}
2504 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2505 function is using two metadata arguments:
2507 .. code-block:: llvm
2509 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2511 Metadata can be attached with an instruction. Here metadata ``!21`` is
2512 attached to the ``add`` instruction using the ``!dbg`` identifier:
2514 .. code-block:: llvm
2516 %indvar.next = add i64 %indvar, 1, !dbg !21
2518 More information about specific metadata nodes recognized by the
2519 optimizers and code generator is found below.
2524 In LLVM IR, memory does not have types, so LLVM's own type system is not
2525 suitable for doing TBAA. Instead, metadata is added to the IR to
2526 describe a type system of a higher level language. This can be used to
2527 implement typical C/C++ TBAA, but it can also be used to implement
2528 custom alias analysis behavior for other languages.
2530 The current metadata format is very simple. TBAA metadata nodes have up
2531 to three fields, e.g.:
2533 .. code-block:: llvm
2535 !0 = metadata !{ metadata !"an example type tree" }
2536 !1 = metadata !{ metadata !"int", metadata !0 }
2537 !2 = metadata !{ metadata !"float", metadata !0 }
2538 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2540 The first field is an identity field. It can be any value, usually a
2541 metadata string, which uniquely identifies the type. The most important
2542 name in the tree is the name of the root node. Two trees with different
2543 root node names are entirely disjoint, even if they have leaves with
2546 The second field identifies the type's parent node in the tree, or is
2547 null or omitted for a root node. A type is considered to alias all of
2548 its descendants and all of its ancestors in the tree. Also, a type is
2549 considered to alias all types in other trees, so that bitcode produced
2550 from multiple front-ends is handled conservatively.
2552 If the third field is present, it's an integer which if equal to 1
2553 indicates that the type is "constant" (meaning
2554 ``pointsToConstantMemory`` should return true; see `other useful
2555 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2557 '``tbaa.struct``' Metadata
2558 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2560 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2561 aggregate assignment operations in C and similar languages, however it
2562 is defined to copy a contiguous region of memory, which is more than
2563 strictly necessary for aggregate types which contain holes due to
2564 padding. Also, it doesn't contain any TBAA information about the fields
2567 ``!tbaa.struct`` metadata can describe which memory subregions in a
2568 memcpy are padding and what the TBAA tags of the struct are.
2570 The current metadata format is very simple. ``!tbaa.struct`` metadata
2571 nodes are a list of operands which are in conceptual groups of three.
2572 For each group of three, the first operand gives the byte offset of a
2573 field in bytes, the second gives its size in bytes, and the third gives
2576 .. code-block:: llvm
2578 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2580 This describes a struct with two fields. The first is at offset 0 bytes
2581 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2582 and has size 4 bytes and has tbaa tag !2.
2584 Note that the fields need not be contiguous. In this example, there is a
2585 4 byte gap between the two fields. This gap represents padding which
2586 does not carry useful data and need not be preserved.
2588 '``fpmath``' Metadata
2589 ^^^^^^^^^^^^^^^^^^^^^
2591 ``fpmath`` metadata may be attached to any instruction of floating point
2592 type. It can be used to express the maximum acceptable error in the
2593 result of that instruction, in ULPs, thus potentially allowing the
2594 compiler to use a more efficient but less accurate method of computing
2595 it. ULP is defined as follows:
2597 If ``x`` is a real number that lies between two finite consecutive
2598 floating-point numbers ``a`` and ``b``, without being equal to one
2599 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2600 distance between the two non-equal finite floating-point numbers
2601 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2603 The metadata node shall consist of a single positive floating point
2604 number representing the maximum relative error, for example:
2606 .. code-block:: llvm
2608 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2610 '``range``' Metadata
2611 ^^^^^^^^^^^^^^^^^^^^
2613 ``range`` metadata may be attached only to loads of integer types. It
2614 expresses the possible ranges the loaded value is in. The ranges are
2615 represented with a flattened list of integers. The loaded value is known
2616 to be in the union of the ranges defined by each consecutive pair. Each
2617 pair has the following properties:
2619 - The type must match the type loaded by the instruction.
2620 - The pair ``a,b`` represents the range ``[a,b)``.
2621 - Both ``a`` and ``b`` are constants.
2622 - The range is allowed to wrap.
2623 - The range should not represent the full or empty set. That is,
2626 In addition, the pairs must be in signed order of the lower bound and
2627 they must be non-contiguous.
2631 .. code-block:: llvm
2633 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2634 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2635 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2636 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2638 !0 = metadata !{ i8 0, i8 2 }
2639 !1 = metadata !{ i8 255, i8 2 }
2640 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2641 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2646 It is sometimes useful to attach information to loop constructs. Currently,
2647 loop metadata is implemented as metadata attached to the branch instruction
2648 in the loop latch block. This type of metadata refer to a metadata node that is
2649 guaranteed to be separate for each loop. The loop identifier metadata is
2650 specified with the name ``llvm.loop``.
2652 The loop identifier metadata is implemented using a metadata that refers to
2653 itself to avoid merging it with any other identifier metadata, e.g.,
2654 during module linkage or function inlining. That is, each loop should refer
2655 to their own identification metadata even if they reside in separate functions.
2656 The following example contains loop identifier metadata for two separate loop
2659 .. code-block:: llvm
2661 !0 = metadata !{ metadata !0 }
2662 !1 = metadata !{ metadata !1 }
2664 The loop identifier metadata can be used to specify additional per-loop
2665 metadata. Any operands after the first operand can be treated as user-defined
2666 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2667 by the loop vectorizer to indicate how many times to unroll the loop:
2669 .. code-block:: llvm
2671 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2673 !0 = metadata !{ metadata !0, metadata !1 }
2674 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2679 Metadata types used to annotate memory accesses with information helpful
2680 for optimizations are prefixed with ``llvm.mem``.
2682 '``llvm.mem.parallel_loop_access``' Metadata
2683 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2685 For a loop to be parallel, in addition to using
2686 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2687 also all of the memory accessing instructions in the loop body need to be
2688 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2689 is at least one memory accessing instruction not marked with the metadata,
2690 the loop must be considered a sequential loop. This causes parallel loops to be
2691 converted to sequential loops due to optimization passes that are unaware of
2692 the parallel semantics and that insert new memory instructions to the loop
2695 Example of a loop that is considered parallel due to its correct use of
2696 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2697 metadata types that refer to the same loop identifier metadata.
2699 .. code-block:: llvm
2703 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2705 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2707 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2711 !0 = metadata !{ metadata !0 }
2713 It is also possible to have nested parallel loops. In that case the
2714 memory accesses refer to a list of loop identifier metadata nodes instead of
2715 the loop identifier metadata node directly:
2717 .. code-block:: llvm
2724 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2726 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2728 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2732 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2734 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2736 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2738 outer.for.end: ; preds = %for.body
2740 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2741 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2742 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2744 '``llvm.vectorizer``'
2745 ^^^^^^^^^^^^^^^^^^^^^
2747 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2748 vectorization parameters such as vectorization factor and unroll factor.
2750 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2751 loop identification metadata.
2753 '``llvm.vectorizer.unroll``' Metadata
2754 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2756 This metadata instructs the loop vectorizer to unroll the specified
2757 loop exactly ``N`` times.
2759 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2760 operand is an integer specifying the unroll factor. For example:
2762 .. code-block:: llvm
2764 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2766 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2769 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2770 determined automatically.
2772 '``llvm.vectorizer.width``' Metadata
2773 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2775 This metadata sets the target width of the vectorizer to ``N``. Without
2776 this metadata, the vectorizer will choose a width automatically.
2777 Regardless of this metadata, the vectorizer will only vectorize loops if
2778 it believes it is valid to do so.
2780 The first operand is the string ``llvm.vectorizer.width`` and the second
2781 operand is an integer specifying the width. For example:
2783 .. code-block:: llvm
2785 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2787 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2790 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2793 Module Flags Metadata
2794 =====================
2796 Information about the module as a whole is difficult to convey to LLVM's
2797 subsystems. The LLVM IR isn't sufficient to transmit this information.
2798 The ``llvm.module.flags`` named metadata exists in order to facilitate
2799 this. These flags are in the form of key / value pairs --- much like a
2800 dictionary --- making it easy for any subsystem who cares about a flag to
2803 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2804 Each triplet has the following form:
2806 - The first element is a *behavior* flag, which specifies the behavior
2807 when two (or more) modules are merged together, and it encounters two
2808 (or more) metadata with the same ID. The supported behaviors are
2810 - The second element is a metadata string that is a unique ID for the
2811 metadata. Each module may only have one flag entry for each unique ID (not
2812 including entries with the **Require** behavior).
2813 - The third element is the value of the flag.
2815 When two (or more) modules are merged together, the resulting
2816 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2817 each unique metadata ID string, there will be exactly one entry in the merged
2818 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2819 be determined by the merge behavior flag, as described below. The only exception
2820 is that entries with the *Require* behavior are always preserved.
2822 The following behaviors are supported:
2833 Emits an error if two values disagree, otherwise the resulting value
2834 is that of the operands.
2838 Emits a warning if two values disagree. The result value will be the
2839 operand for the flag from the first module being linked.
2843 Adds a requirement that another module flag be present and have a
2844 specified value after linking is performed. The value must be a
2845 metadata pair, where the first element of the pair is the ID of the
2846 module flag to be restricted, and the second element of the pair is
2847 the value the module flag should be restricted to. This behavior can
2848 be used to restrict the allowable results (via triggering of an
2849 error) of linking IDs with the **Override** behavior.
2853 Uses the specified value, regardless of the behavior or value of the
2854 other module. If both modules specify **Override**, but the values
2855 differ, an error will be emitted.
2859 Appends the two values, which are required to be metadata nodes.
2863 Appends the two values, which are required to be metadata
2864 nodes. However, duplicate entries in the second list are dropped
2865 during the append operation.
2867 It is an error for a particular unique flag ID to have multiple behaviors,
2868 except in the case of **Require** (which adds restrictions on another metadata
2869 value) or **Override**.
2871 An example of module flags:
2873 .. code-block:: llvm
2875 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2876 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2877 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2878 !3 = metadata !{ i32 3, metadata !"qux",
2880 metadata !"foo", i32 1
2883 !llvm.module.flags = !{ !0, !1, !2, !3 }
2885 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2886 if two or more ``!"foo"`` flags are seen is to emit an error if their
2887 values are not equal.
2889 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2890 behavior if two or more ``!"bar"`` flags are seen is to use the value
2893 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2894 behavior if two or more ``!"qux"`` flags are seen is to emit a
2895 warning if their values are not equal.
2897 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2901 metadata !{ metadata !"foo", i32 1 }
2903 The behavior is to emit an error if the ``llvm.module.flags`` does not
2904 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2907 Objective-C Garbage Collection Module Flags Metadata
2908 ----------------------------------------------------
2910 On the Mach-O platform, Objective-C stores metadata about garbage
2911 collection in a special section called "image info". The metadata
2912 consists of a version number and a bitmask specifying what types of
2913 garbage collection are supported (if any) by the file. If two or more
2914 modules are linked together their garbage collection metadata needs to
2915 be merged rather than appended together.
2917 The Objective-C garbage collection module flags metadata consists of the
2918 following key-value pairs:
2927 * - ``Objective-C Version``
2928 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2930 * - ``Objective-C Image Info Version``
2931 - **[Required]** --- The version of the image info section. Currently
2934 * - ``Objective-C Image Info Section``
2935 - **[Required]** --- The section to place the metadata. Valid values are
2936 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2937 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2938 Objective-C ABI version 2.
2940 * - ``Objective-C Garbage Collection``
2941 - **[Required]** --- Specifies whether garbage collection is supported or
2942 not. Valid values are 0, for no garbage collection, and 2, for garbage
2943 collection supported.
2945 * - ``Objective-C GC Only``
2946 - **[Optional]** --- Specifies that only garbage collection is supported.
2947 If present, its value must be 6. This flag requires that the
2948 ``Objective-C Garbage Collection`` flag have the value 2.
2950 Some important flag interactions:
2952 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2953 merged with a module with ``Objective-C Garbage Collection`` set to
2954 2, then the resulting module has the
2955 ``Objective-C Garbage Collection`` flag set to 0.
2956 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2957 merged with a module with ``Objective-C GC Only`` set to 6.
2959 Automatic Linker Flags Module Flags Metadata
2960 --------------------------------------------
2962 Some targets support embedding flags to the linker inside individual object
2963 files. Typically this is used in conjunction with language extensions which
2964 allow source files to explicitly declare the libraries they depend on, and have
2965 these automatically be transmitted to the linker via object files.
2967 These flags are encoded in the IR using metadata in the module flags section,
2968 using the ``Linker Options`` key. The merge behavior for this flag is required
2969 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2970 node which should be a list of other metadata nodes, each of which should be a
2971 list of metadata strings defining linker options.
2973 For example, the following metadata section specifies two separate sets of
2974 linker options, presumably to link against ``libz`` and the ``Cocoa``
2977 !0 = metadata !{ i32 6, metadata !"Linker Options",
2979 metadata !{ metadata !"-lz" },
2980 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2981 !llvm.module.flags = !{ !0 }
2983 The metadata encoding as lists of lists of options, as opposed to a collapsed
2984 list of options, is chosen so that the IR encoding can use multiple option
2985 strings to specify e.g., a single library, while still having that specifier be
2986 preserved as an atomic element that can be recognized by a target specific
2987 assembly writer or object file emitter.
2989 Each individual option is required to be either a valid option for the target's
2990 linker, or an option that is reserved by the target specific assembly writer or
2991 object file emitter. No other aspect of these options is defined by the IR.
2993 .. _intrinsicglobalvariables:
2995 Intrinsic Global Variables
2996 ==========================
2998 LLVM has a number of "magic" global variables that contain data that
2999 affect code generation or other IR semantics. These are documented here.
3000 All globals of this sort should have a section specified as
3001 "``llvm.metadata``". This section and all globals that start with
3002 "``llvm.``" are reserved for use by LLVM.
3006 The '``llvm.used``' Global Variable
3007 -----------------------------------
3009 The ``@llvm.used`` global is an array which has
3010 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3011 pointers to named global variables, functions and aliases which may optionally
3012 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3015 .. code-block:: llvm
3020 @llvm.used = appending global [2 x i8*] [
3022 i8* bitcast (i32* @Y to i8*)
3023 ], section "llvm.metadata"
3025 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3026 and linker are required to treat the symbol as if there is a reference to the
3027 symbol that it cannot see (which is why they have to be named). For example, if
3028 a variable has internal linkage and no references other than that from the
3029 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3030 references from inline asms and other things the compiler cannot "see", and
3031 corresponds to "``attribute((used))``" in GNU C.
3033 On some targets, the code generator must emit a directive to the
3034 assembler or object file to prevent the assembler and linker from
3035 molesting the symbol.
3037 .. _gv_llvmcompilerused:
3039 The '``llvm.compiler.used``' Global Variable
3040 --------------------------------------------
3042 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3043 directive, except that it only prevents the compiler from touching the
3044 symbol. On targets that support it, this allows an intelligent linker to
3045 optimize references to the symbol without being impeded as it would be
3048 This is a rare construct that should only be used in rare circumstances,
3049 and should not be exposed to source languages.
3051 .. _gv_llvmglobalctors:
3053 The '``llvm.global_ctors``' Global Variable
3054 -------------------------------------------
3056 .. code-block:: llvm
3058 %0 = type { i32, void ()* }
3059 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3061 The ``@llvm.global_ctors`` array contains a list of constructor
3062 functions and associated priorities. The functions referenced by this
3063 array will be called in ascending order of priority (i.e. lowest first)
3064 when the module is loaded. The order of functions with the same priority
3067 .. _llvmglobaldtors:
3069 The '``llvm.global_dtors``' Global Variable
3070 -------------------------------------------
3072 .. code-block:: llvm
3074 %0 = type { i32, void ()* }
3075 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3077 The ``@llvm.global_dtors`` array contains a list of destructor functions
3078 and associated priorities. The functions referenced by this array will
3079 be called in descending order of priority (i.e. highest first) when the
3080 module is loaded. The order of functions with the same priority is not
3083 Instruction Reference
3084 =====================
3086 The LLVM instruction set consists of several different classifications
3087 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3088 instructions <binaryops>`, :ref:`bitwise binary
3089 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3090 :ref:`other instructions <otherops>`.
3094 Terminator Instructions
3095 -----------------------
3097 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3098 program ends with a "Terminator" instruction, which indicates which
3099 block should be executed after the current block is finished. These
3100 terminator instructions typically yield a '``void``' value: they produce
3101 control flow, not values (the one exception being the
3102 ':ref:`invoke <i_invoke>`' instruction).
3104 The terminator instructions are: ':ref:`ret <i_ret>`',
3105 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3106 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3107 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3111 '``ret``' Instruction
3112 ^^^^^^^^^^^^^^^^^^^^^
3119 ret <type> <value> ; Return a value from a non-void function
3120 ret void ; Return from void function
3125 The '``ret``' instruction is used to return control flow (and optionally
3126 a value) from a function back to the caller.
3128 There are two forms of the '``ret``' instruction: one that returns a
3129 value and then causes control flow, and one that just causes control
3135 The '``ret``' instruction optionally accepts a single argument, the
3136 return value. The type of the return value must be a ':ref:`first
3137 class <t_firstclass>`' type.
3139 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3140 return type and contains a '``ret``' instruction with no return value or
3141 a return value with a type that does not match its type, or if it has a
3142 void return type and contains a '``ret``' instruction with a return
3148 When the '``ret``' instruction is executed, control flow returns back to
3149 the calling function's context. If the caller is a
3150 ":ref:`call <i_call>`" instruction, execution continues at the
3151 instruction after the call. If the caller was an
3152 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3153 beginning of the "normal" destination block. If the instruction returns
3154 a value, that value shall set the call or invoke instruction's return
3160 .. code-block:: llvm
3162 ret i32 5 ; Return an integer value of 5
3163 ret void ; Return from a void function
3164 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3168 '``br``' Instruction
3169 ^^^^^^^^^^^^^^^^^^^^
3176 br i1 <cond>, label <iftrue>, label <iffalse>
3177 br label <dest> ; Unconditional branch
3182 The '``br``' instruction is used to cause control flow to transfer to a
3183 different basic block in the current function. There are two forms of
3184 this instruction, corresponding to a conditional branch and an
3185 unconditional branch.
3190 The conditional branch form of the '``br``' instruction takes a single
3191 '``i1``' value and two '``label``' values. The unconditional form of the
3192 '``br``' instruction takes a single '``label``' value as a target.
3197 Upon execution of a conditional '``br``' instruction, the '``i1``'
3198 argument is evaluated. If the value is ``true``, control flows to the
3199 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3200 to the '``iffalse``' ``label`` argument.
3205 .. code-block:: llvm
3208 %cond = icmp eq i32 %a, %b
3209 br i1 %cond, label %IfEqual, label %IfUnequal
3217 '``switch``' Instruction
3218 ^^^^^^^^^^^^^^^^^^^^^^^^
3225 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3230 The '``switch``' instruction is used to transfer control flow to one of
3231 several different places. It is a generalization of the '``br``'
3232 instruction, allowing a branch to occur to one of many possible
3238 The '``switch``' instruction uses three parameters: an integer
3239 comparison value '``value``', a default '``label``' destination, and an
3240 array of pairs of comparison value constants and '``label``'s. The table
3241 is not allowed to contain duplicate constant entries.
3246 The ``switch`` instruction specifies a table of values and destinations.
3247 When the '``switch``' instruction is executed, this table is searched
3248 for the given value. If the value is found, control flow is transferred
3249 to the corresponding destination; otherwise, control flow is transferred
3250 to the default destination.
3255 Depending on properties of the target machine and the particular
3256 ``switch`` instruction, this instruction may be code generated in
3257 different ways. For example, it could be generated as a series of
3258 chained conditional branches or with a lookup table.
3263 .. code-block:: llvm
3265 ; Emulate a conditional br instruction
3266 %Val = zext i1 %value to i32
3267 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3269 ; Emulate an unconditional br instruction
3270 switch i32 0, label %dest [ ]
3272 ; Implement a jump table:
3273 switch i32 %val, label %otherwise [ i32 0, label %onzero
3275 i32 2, label %ontwo ]
3279 '``indirectbr``' Instruction
3280 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3287 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3292 The '``indirectbr``' instruction implements an indirect branch to a
3293 label within the current function, whose address is specified by
3294 "``address``". Address must be derived from a
3295 :ref:`blockaddress <blockaddress>` constant.
3300 The '``address``' argument is the address of the label to jump to. The
3301 rest of the arguments indicate the full set of possible destinations
3302 that the address may point to. Blocks are allowed to occur multiple
3303 times in the destination list, though this isn't particularly useful.
3305 This destination list is required so that dataflow analysis has an
3306 accurate understanding of the CFG.
3311 Control transfers to the block specified in the address argument. All
3312 possible destination blocks must be listed in the label list, otherwise
3313 this instruction has undefined behavior. This implies that jumps to
3314 labels defined in other functions have undefined behavior as well.
3319 This is typically implemented with a jump through a register.
3324 .. code-block:: llvm
3326 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3330 '``invoke``' Instruction
3331 ^^^^^^^^^^^^^^^^^^^^^^^^
3338 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3339 to label <normal label> unwind label <exception label>
3344 The '``invoke``' instruction causes control to transfer to a specified
3345 function, with the possibility of control flow transfer to either the
3346 '``normal``' label or the '``exception``' label. If the callee function
3347 returns with the "``ret``" instruction, control flow will return to the
3348 "normal" label. If the callee (or any indirect callees) returns via the
3349 ":ref:`resume <i_resume>`" instruction or other exception handling
3350 mechanism, control is interrupted and continued at the dynamically
3351 nearest "exception" label.
3353 The '``exception``' label is a `landing
3354 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3355 '``exception``' label is required to have the
3356 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3357 information about the behavior of the program after unwinding happens,
3358 as its first non-PHI instruction. The restrictions on the
3359 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3360 instruction, so that the important information contained within the
3361 "``landingpad``" instruction can't be lost through normal code motion.
3366 This instruction requires several arguments:
3368 #. The optional "cconv" marker indicates which :ref:`calling
3369 convention <callingconv>` the call should use. If none is
3370 specified, the call defaults to using C calling conventions.
3371 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3372 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3374 #. '``ptr to function ty``': shall be the signature of the pointer to
3375 function value being invoked. In most cases, this is a direct
3376 function invocation, but indirect ``invoke``'s are just as possible,
3377 branching off an arbitrary pointer to function value.
3378 #. '``function ptr val``': An LLVM value containing a pointer to a
3379 function to be invoked.
3380 #. '``function args``': argument list whose types match the function
3381 signature argument types and parameter attributes. All arguments must
3382 be of :ref:`first class <t_firstclass>` type. If the function signature
3383 indicates the function accepts a variable number of arguments, the
3384 extra arguments can be specified.
3385 #. '``normal label``': the label reached when the called function
3386 executes a '``ret``' instruction.
3387 #. '``exception label``': the label reached when a callee returns via
3388 the :ref:`resume <i_resume>` instruction or other exception handling
3390 #. The optional :ref:`function attributes <fnattrs>` list. Only
3391 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3392 attributes are valid here.
3397 This instruction is designed to operate as a standard '``call``'
3398 instruction in most regards. The primary difference is that it
3399 establishes an association with a label, which is used by the runtime
3400 library to unwind the stack.
3402 This instruction is used in languages with destructors to ensure that
3403 proper cleanup is performed in the case of either a ``longjmp`` or a
3404 thrown exception. Additionally, this is important for implementation of
3405 '``catch``' clauses in high-level languages that support them.
3407 For the purposes of the SSA form, the definition of the value returned
3408 by the '``invoke``' instruction is deemed to occur on the edge from the
3409 current block to the "normal" label. If the callee unwinds then no
3410 return value is available.
3415 .. code-block:: llvm
3417 %retval = invoke i32 @Test(i32 15) to label %Continue
3418 unwind label %TestCleanup ; {i32}:retval set
3419 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3420 unwind label %TestCleanup ; {i32}:retval set
3424 '``resume``' Instruction
3425 ^^^^^^^^^^^^^^^^^^^^^^^^
3432 resume <type> <value>
3437 The '``resume``' instruction is a terminator instruction that has no
3443 The '``resume``' instruction requires one argument, which must have the
3444 same type as the result of any '``landingpad``' instruction in the same
3450 The '``resume``' instruction resumes propagation of an existing
3451 (in-flight) exception whose unwinding was interrupted with a
3452 :ref:`landingpad <i_landingpad>` instruction.
3457 .. code-block:: llvm
3459 resume { i8*, i32 } %exn
3463 '``unreachable``' Instruction
3464 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3476 The '``unreachable``' instruction has no defined semantics. This
3477 instruction is used to inform the optimizer that a particular portion of
3478 the code is not reachable. This can be used to indicate that the code
3479 after a no-return function cannot be reached, and other facts.
3484 The '``unreachable``' instruction has no defined semantics.
3491 Binary operators are used to do most of the computation in a program.
3492 They require two operands of the same type, execute an operation on
3493 them, and produce a single value. The operands might represent multiple
3494 data, as is the case with the :ref:`vector <t_vector>` data type. The
3495 result value has the same type as its operands.
3497 There are several different binary operators:
3501 '``add``' Instruction
3502 ^^^^^^^^^^^^^^^^^^^^^
3509 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3510 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3511 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3512 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3517 The '``add``' instruction returns the sum of its two operands.
3522 The two arguments to the '``add``' instruction must be
3523 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3524 arguments must have identical types.
3529 The value produced is the integer sum of the two operands.
3531 If the sum has unsigned overflow, the result returned is the
3532 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3535 Because LLVM integers use a two's complement representation, this
3536 instruction is appropriate for both signed and unsigned integers.
3538 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3539 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3540 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3541 unsigned and/or signed overflow, respectively, occurs.
3546 .. code-block:: llvm
3548 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3552 '``fadd``' Instruction
3553 ^^^^^^^^^^^^^^^^^^^^^^
3560 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3565 The '``fadd``' instruction returns the sum of its two operands.
3570 The two arguments to the '``fadd``' instruction must be :ref:`floating
3571 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3572 Both arguments must have identical types.
3577 The value produced is the floating point sum of the two operands. This
3578 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3579 which are optimization hints to enable otherwise unsafe floating point
3585 .. code-block:: llvm
3587 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3589 '``sub``' Instruction
3590 ^^^^^^^^^^^^^^^^^^^^^
3597 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3598 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3599 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3600 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3605 The '``sub``' instruction returns the difference of its two operands.
3607 Note that the '``sub``' instruction is used to represent the '``neg``'
3608 instruction present in most other intermediate representations.
3613 The two arguments to the '``sub``' instruction must be
3614 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3615 arguments must have identical types.
3620 The value produced is the integer difference of the two operands.
3622 If the difference has unsigned overflow, the result returned is the
3623 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3626 Because LLVM integers use a two's complement representation, this
3627 instruction is appropriate for both signed and unsigned integers.
3629 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3630 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3631 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3632 unsigned and/or signed overflow, respectively, occurs.
3637 .. code-block:: llvm
3639 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3640 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3644 '``fsub``' Instruction
3645 ^^^^^^^^^^^^^^^^^^^^^^
3652 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3657 The '``fsub``' instruction returns the difference of its two operands.
3659 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3660 instruction present in most other intermediate representations.
3665 The two arguments to the '``fsub``' instruction must be :ref:`floating
3666 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3667 Both arguments must have identical types.
3672 The value produced is the floating point difference of the two operands.
3673 This instruction can also take any number of :ref:`fast-math
3674 flags <fastmath>`, which are optimization hints to enable otherwise
3675 unsafe floating point optimizations:
3680 .. code-block:: llvm
3682 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3683 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3685 '``mul``' Instruction
3686 ^^^^^^^^^^^^^^^^^^^^^
3693 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3694 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3695 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3696 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3701 The '``mul``' instruction returns the product of its two operands.
3706 The two arguments to the '``mul``' instruction must be
3707 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3708 arguments must have identical types.
3713 The value produced is the integer product of the two operands.
3715 If the result of the multiplication has unsigned overflow, the result
3716 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3717 bit width of the result.
3719 Because LLVM integers use a two's complement representation, and the
3720 result is the same width as the operands, this instruction returns the
3721 correct result for both signed and unsigned integers. If a full product
3722 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3723 sign-extended or zero-extended as appropriate to the width of the full
3726 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3727 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3728 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3729 unsigned and/or signed overflow, respectively, occurs.
3734 .. code-block:: llvm
3736 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3740 '``fmul``' Instruction
3741 ^^^^^^^^^^^^^^^^^^^^^^
3748 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3753 The '``fmul``' instruction returns the product of its two operands.
3758 The two arguments to the '``fmul``' instruction must be :ref:`floating
3759 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3760 Both arguments must have identical types.
3765 The value produced is the floating point product of the two operands.
3766 This instruction can also take any number of :ref:`fast-math
3767 flags <fastmath>`, which are optimization hints to enable otherwise
3768 unsafe floating point optimizations:
3773 .. code-block:: llvm
3775 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3777 '``udiv``' Instruction
3778 ^^^^^^^^^^^^^^^^^^^^^^
3785 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3786 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3791 The '``udiv``' instruction returns the quotient of its two operands.
3796 The two arguments to the '``udiv``' instruction must be
3797 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3798 arguments must have identical types.
3803 The value produced is the unsigned integer quotient of the two operands.
3805 Note that unsigned integer division and signed integer division are
3806 distinct operations; for signed integer division, use '``sdiv``'.
3808 Division by zero leads to undefined behavior.
3810 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3811 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3812 such, "((a udiv exact b) mul b) == a").
3817 .. code-block:: llvm
3819 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3821 '``sdiv``' Instruction
3822 ^^^^^^^^^^^^^^^^^^^^^^
3829 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3830 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3835 The '``sdiv``' instruction returns the quotient of its two operands.
3840 The two arguments to the '``sdiv``' instruction must be
3841 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3842 arguments must have identical types.
3847 The value produced is the signed integer quotient of the two operands
3848 rounded towards zero.
3850 Note that signed integer division and unsigned integer division are
3851 distinct operations; for unsigned integer division, use '``udiv``'.
3853 Division by zero leads to undefined behavior. Overflow also leads to
3854 undefined behavior; this is a rare case, but can occur, for example, by
3855 doing a 32-bit division of -2147483648 by -1.
3857 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3858 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3863 .. code-block:: llvm
3865 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3869 '``fdiv``' Instruction
3870 ^^^^^^^^^^^^^^^^^^^^^^
3877 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3882 The '``fdiv``' instruction returns the quotient of its two operands.
3887 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3888 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3889 Both arguments must have identical types.
3894 The value produced is the floating point quotient of the two operands.
3895 This instruction can also take any number of :ref:`fast-math
3896 flags <fastmath>`, which are optimization hints to enable otherwise
3897 unsafe floating point optimizations:
3902 .. code-block:: llvm
3904 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3906 '``urem``' Instruction
3907 ^^^^^^^^^^^^^^^^^^^^^^
3914 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3919 The '``urem``' instruction returns the remainder from the unsigned
3920 division of its two arguments.
3925 The two arguments to the '``urem``' instruction must be
3926 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3927 arguments must have identical types.
3932 This instruction returns the unsigned integer *remainder* of a division.
3933 This instruction always performs an unsigned division to get the
3936 Note that unsigned integer remainder and signed integer remainder are
3937 distinct operations; for signed integer remainder, use '``srem``'.
3939 Taking the remainder of a division by zero leads to undefined behavior.
3944 .. code-block:: llvm
3946 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3948 '``srem``' Instruction
3949 ^^^^^^^^^^^^^^^^^^^^^^
3956 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3961 The '``srem``' instruction returns the remainder from the signed
3962 division of its two operands. This instruction can also take
3963 :ref:`vector <t_vector>` versions of the values in which case the elements
3969 The two arguments to the '``srem``' instruction must be
3970 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3971 arguments must have identical types.
3976 This instruction returns the *remainder* of a division (where the result
3977 is either zero or has the same sign as the dividend, ``op1``), not the
3978 *modulo* operator (where the result is either zero or has the same sign
3979 as the divisor, ``op2``) of a value. For more information about the
3980 difference, see `The Math
3981 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3982 table of how this is implemented in various languages, please see
3984 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3986 Note that signed integer remainder and unsigned integer remainder are
3987 distinct operations; for unsigned integer remainder, use '``urem``'.
3989 Taking the remainder of a division by zero leads to undefined behavior.
3990 Overflow also leads to undefined behavior; this is a rare case, but can
3991 occur, for example, by taking the remainder of a 32-bit division of
3992 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3993 rule lets srem be implemented using instructions that return both the
3994 result of the division and the remainder.)
3999 .. code-block:: llvm
4001 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4005 '``frem``' Instruction
4006 ^^^^^^^^^^^^^^^^^^^^^^
4013 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4018 The '``frem``' instruction returns the remainder from the division of
4024 The two arguments to the '``frem``' instruction must be :ref:`floating
4025 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4026 Both arguments must have identical types.
4031 This instruction returns the *remainder* of a division. The remainder
4032 has the same sign as the dividend. This instruction can also take any
4033 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4034 to enable otherwise unsafe floating point optimizations:
4039 .. code-block:: llvm
4041 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4045 Bitwise Binary Operations
4046 -------------------------
4048 Bitwise binary operators are used to do various forms of bit-twiddling
4049 in a program. They are generally very efficient instructions and can
4050 commonly be strength reduced from other instructions. They require two
4051 operands of the same type, execute an operation on them, and produce a
4052 single value. The resulting value is the same type as its operands.
4054 '``shl``' Instruction
4055 ^^^^^^^^^^^^^^^^^^^^^
4062 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4063 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4064 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4065 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4070 The '``shl``' instruction returns the first operand shifted to the left
4071 a specified number of bits.
4076 Both arguments to the '``shl``' instruction must be the same
4077 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4078 '``op2``' is treated as an unsigned value.
4083 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4084 where ``n`` is the width of the result. If ``op2`` is (statically or
4085 dynamically) negative or equal to or larger than the number of bits in
4086 ``op1``, the result is undefined. If the arguments are vectors, each
4087 vector element of ``op1`` is shifted by the corresponding shift amount
4090 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4091 value <poisonvalues>` if it shifts out any non-zero bits. If the
4092 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4093 value <poisonvalues>` if it shifts out any bits that disagree with the
4094 resultant sign bit. As such, NUW/NSW have the same semantics as they
4095 would if the shift were expressed as a mul instruction with the same
4096 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4101 .. code-block:: llvm
4103 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4104 <result> = shl i32 4, 2 ; yields {i32}: 16
4105 <result> = shl i32 1, 10 ; yields {i32}: 1024
4106 <result> = shl i32 1, 32 ; undefined
4107 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4109 '``lshr``' Instruction
4110 ^^^^^^^^^^^^^^^^^^^^^^
4117 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4118 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4123 The '``lshr``' instruction (logical shift right) returns the first
4124 operand shifted to the right a specified number of bits with zero fill.
4129 Both arguments to the '``lshr``' instruction must be the same
4130 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4131 '``op2``' is treated as an unsigned value.
4136 This instruction always performs a logical shift right operation. The
4137 most significant bits of the result will be filled with zero bits after
4138 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4139 than the number of bits in ``op1``, the result is undefined. If the
4140 arguments are vectors, each vector element of ``op1`` is shifted by the
4141 corresponding shift amount in ``op2``.
4143 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4144 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4150 .. code-block:: llvm
4152 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4153 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4154 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4155 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4156 <result> = lshr i32 1, 32 ; undefined
4157 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4159 '``ashr``' Instruction
4160 ^^^^^^^^^^^^^^^^^^^^^^
4167 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4168 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4173 The '``ashr``' instruction (arithmetic shift right) returns the first
4174 operand shifted to the right a specified number of bits with sign
4180 Both arguments to the '``ashr``' instruction must be the same
4181 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4182 '``op2``' is treated as an unsigned value.
4187 This instruction always performs an arithmetic shift right operation,
4188 The most significant bits of the result will be filled with the sign bit
4189 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4190 than the number of bits in ``op1``, the result is undefined. If the
4191 arguments are vectors, each vector element of ``op1`` is shifted by the
4192 corresponding shift amount in ``op2``.
4194 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4195 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4201 .. code-block:: llvm
4203 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4204 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4205 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4206 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4207 <result> = ashr i32 1, 32 ; undefined
4208 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4210 '``and``' Instruction
4211 ^^^^^^^^^^^^^^^^^^^^^
4218 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4223 The '``and``' instruction returns the bitwise logical and of its two
4229 The two arguments to the '``and``' instruction must be
4230 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4231 arguments must have identical types.
4236 The truth table used for the '``and``' instruction is:
4253 .. code-block:: llvm
4255 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4256 <result> = and i32 15, 40 ; yields {i32}:result = 8
4257 <result> = and i32 4, 8 ; yields {i32}:result = 0
4259 '``or``' Instruction
4260 ^^^^^^^^^^^^^^^^^^^^
4267 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4272 The '``or``' instruction returns the bitwise logical inclusive or of its
4278 The two arguments to the '``or``' instruction must be
4279 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4280 arguments must have identical types.
4285 The truth table used for the '``or``' instruction is:
4304 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4305 <result> = or i32 15, 40 ; yields {i32}:result = 47
4306 <result> = or i32 4, 8 ; yields {i32}:result = 12
4308 '``xor``' Instruction
4309 ^^^^^^^^^^^^^^^^^^^^^
4316 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4321 The '``xor``' instruction returns the bitwise logical exclusive or of
4322 its two operands. The ``xor`` is used to implement the "one's
4323 complement" operation, which is the "~" operator in C.
4328 The two arguments to the '``xor``' instruction must be
4329 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4330 arguments must have identical types.
4335 The truth table used for the '``xor``' instruction is:
4352 .. code-block:: llvm
4354 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4355 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4356 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4357 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4362 LLVM supports several instructions to represent vector operations in a
4363 target-independent manner. These instructions cover the element-access
4364 and vector-specific operations needed to process vectors effectively.
4365 While LLVM does directly support these vector operations, many
4366 sophisticated algorithms will want to use target-specific intrinsics to
4367 take full advantage of a specific target.
4369 .. _i_extractelement:
4371 '``extractelement``' Instruction
4372 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4379 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4384 The '``extractelement``' instruction extracts a single scalar element
4385 from a vector at a specified index.
4390 The first operand of an '``extractelement``' instruction is a value of
4391 :ref:`vector <t_vector>` type. The second operand is an index indicating
4392 the position from which to extract the element. The index may be a
4398 The result is a scalar of the same type as the element type of ``val``.
4399 Its value is the value at position ``idx`` of ``val``. If ``idx``
4400 exceeds the length of ``val``, the results are undefined.
4405 .. code-block:: llvm
4407 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4409 .. _i_insertelement:
4411 '``insertelement``' Instruction
4412 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4419 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4424 The '``insertelement``' instruction inserts a scalar element into a
4425 vector at a specified index.
4430 The first operand of an '``insertelement``' instruction is a value of
4431 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4432 type must equal the element type of the first operand. The third operand
4433 is an index indicating the position at which to insert the value. The
4434 index may be a variable.
4439 The result is a vector of the same type as ``val``. Its element values
4440 are those of ``val`` except at position ``idx``, where it gets the value
4441 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4447 .. code-block:: llvm
4449 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4451 .. _i_shufflevector:
4453 '``shufflevector``' Instruction
4454 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4461 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4466 The '``shufflevector``' instruction constructs a permutation of elements
4467 from two input vectors, returning a vector with the same element type as
4468 the input and length that is the same as the shuffle mask.
4473 The first two operands of a '``shufflevector``' instruction are vectors
4474 with the same type. The third argument is a shuffle mask whose element
4475 type is always 'i32'. The result of the instruction is a vector whose
4476 length is the same as the shuffle mask and whose element type is the
4477 same as the element type of the first two operands.
4479 The shuffle mask operand is required to be a constant vector with either
4480 constant integer or undef values.
4485 The elements of the two input vectors are numbered from left to right
4486 across both of the vectors. The shuffle mask operand specifies, for each
4487 element of the result vector, which element of the two input vectors the
4488 result element gets. The element selector may be undef (meaning "don't
4489 care") and the second operand may be undef if performing a shuffle from
4495 .. code-block:: llvm
4497 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4498 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4499 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4500 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4501 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4502 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4503 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4504 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4506 Aggregate Operations
4507 --------------------
4509 LLVM supports several instructions for working with
4510 :ref:`aggregate <t_aggregate>` values.
4514 '``extractvalue``' Instruction
4515 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4522 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4527 The '``extractvalue``' instruction extracts the value of a member field
4528 from an :ref:`aggregate <t_aggregate>` value.
4533 The first operand of an '``extractvalue``' instruction is a value of
4534 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4535 constant indices to specify which value to extract in a similar manner
4536 as indices in a '``getelementptr``' instruction.
4538 The major differences to ``getelementptr`` indexing are:
4540 - Since the value being indexed is not a pointer, the first index is
4541 omitted and assumed to be zero.
4542 - At least one index must be specified.
4543 - Not only struct indices but also array indices must be in bounds.
4548 The result is the value at the position in the aggregate specified by
4554 .. code-block:: llvm
4556 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4560 '``insertvalue``' Instruction
4561 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4568 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4573 The '``insertvalue``' instruction inserts a value into a member field in
4574 an :ref:`aggregate <t_aggregate>` value.
4579 The first operand of an '``insertvalue``' instruction is a value of
4580 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4581 a first-class value to insert. The following operands are constant
4582 indices indicating the position at which to insert the value in a
4583 similar manner as indices in a '``extractvalue``' instruction. The value
4584 to insert must have the same type as the value identified by the
4590 The result is an aggregate of the same type as ``val``. Its value is
4591 that of ``val`` except that the value at the position specified by the
4592 indices is that of ``elt``.
4597 .. code-block:: llvm
4599 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4600 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4601 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4605 Memory Access and Addressing Operations
4606 ---------------------------------------
4608 A key design point of an SSA-based representation is how it represents
4609 memory. In LLVM, no memory locations are in SSA form, which makes things
4610 very simple. This section describes how to read, write, and allocate
4615 '``alloca``' Instruction
4616 ^^^^^^^^^^^^^^^^^^^^^^^^
4623 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4628 The '``alloca``' instruction allocates memory on the stack frame of the
4629 currently executing function, to be automatically released when this
4630 function returns to its caller. The object is always allocated in the
4631 generic address space (address space zero).
4636 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4637 bytes of memory on the runtime stack, returning a pointer of the
4638 appropriate type to the program. If "NumElements" is specified, it is
4639 the number of elements allocated, otherwise "NumElements" is defaulted
4640 to be one. If a constant alignment is specified, the value result of the
4641 allocation is guaranteed to be aligned to at least that boundary. If not
4642 specified, or if zero, the target can choose to align the allocation on
4643 any convenient boundary compatible with the type.
4645 '``type``' may be any sized type.
4650 Memory is allocated; a pointer is returned. The operation is undefined
4651 if there is insufficient stack space for the allocation. '``alloca``'d
4652 memory is automatically released when the function returns. The
4653 '``alloca``' instruction is commonly used to represent automatic
4654 variables that must have an address available. When the function returns
4655 (either with the ``ret`` or ``resume`` instructions), the memory is
4656 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4657 The order in which memory is allocated (ie., which way the stack grows)
4663 .. code-block:: llvm
4665 %ptr = alloca i32 ; yields {i32*}:ptr
4666 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4667 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4668 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4672 '``load``' Instruction
4673 ^^^^^^^^^^^^^^^^^^^^^^
4680 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4681 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4682 !<index> = !{ i32 1 }
4687 The '``load``' instruction is used to read from memory.
4692 The argument to the ``load`` instruction specifies the memory address
4693 from which to load. The pointer must point to a :ref:`first
4694 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4695 then the optimizer is not allowed to modify the number or order of
4696 execution of this ``load`` with other :ref:`volatile
4697 operations <volatile>`.
4699 If the ``load`` is marked as ``atomic``, it takes an extra
4700 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4701 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4702 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4703 when they may see multiple atomic stores. The type of the pointee must
4704 be an integer type whose bit width is a power of two greater than or
4705 equal to eight and less than or equal to a target-specific size limit.
4706 ``align`` must be explicitly specified on atomic loads, and the load has
4707 undefined behavior if the alignment is not set to a value which is at
4708 least the size in bytes of the pointee. ``!nontemporal`` does not have
4709 any defined semantics for atomic loads.
4711 The optional constant ``align`` argument specifies the alignment of the
4712 operation (that is, the alignment of the memory address). A value of 0
4713 or an omitted ``align`` argument means that the operation has the ABI
4714 alignment for the target. It is the responsibility of the code emitter
4715 to ensure that the alignment information is correct. Overestimating the
4716 alignment results in undefined behavior. Underestimating the alignment
4717 may produce less efficient code. An alignment of 1 is always safe.
4719 The optional ``!nontemporal`` metadata must reference a single
4720 metadata name ``<index>`` corresponding to a metadata node with one
4721 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4722 metadata on the instruction tells the optimizer and code generator
4723 that this load is not expected to be reused in the cache. The code
4724 generator may select special instructions to save cache bandwidth, such
4725 as the ``MOVNT`` instruction on x86.
4727 The optional ``!invariant.load`` metadata must reference a single
4728 metadata name ``<index>`` corresponding to a metadata node with no
4729 entries. The existence of the ``!invariant.load`` metadata on the
4730 instruction tells the optimizer and code generator that this load
4731 address points to memory which does not change value during program
4732 execution. The optimizer may then move this load around, for example, by
4733 hoisting it out of loops using loop invariant code motion.
4738 The location of memory pointed to is loaded. If the value being loaded
4739 is of scalar type then the number of bytes read does not exceed the
4740 minimum number of bytes needed to hold all bits of the type. For
4741 example, loading an ``i24`` reads at most three bytes. When loading a
4742 value of a type like ``i20`` with a size that is not an integral number
4743 of bytes, the result is undefined if the value was not originally
4744 written using a store of the same type.
4749 .. code-block:: llvm
4751 %ptr = alloca i32 ; yields {i32*}:ptr
4752 store i32 3, i32* %ptr ; yields {void}
4753 %val = load i32* %ptr ; yields {i32}:val = i32 3
4757 '``store``' Instruction
4758 ^^^^^^^^^^^^^^^^^^^^^^^
4765 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4766 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4771 The '``store``' instruction is used to write to memory.
4776 There are two arguments to the ``store`` instruction: a value to store
4777 and an address at which to store it. The type of the ``<pointer>``
4778 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4779 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4780 then the optimizer is not allowed to modify the number or order of
4781 execution of this ``store`` with other :ref:`volatile
4782 operations <volatile>`.
4784 If the ``store`` is marked as ``atomic``, it takes an extra
4785 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4786 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4787 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4788 when they may see multiple atomic stores. The type of the pointee must
4789 be an integer type whose bit width is a power of two greater than or
4790 equal to eight and less than or equal to a target-specific size limit.
4791 ``align`` must be explicitly specified on atomic stores, and the store
4792 has undefined behavior if the alignment is not set to a value which is
4793 at least the size in bytes of the pointee. ``!nontemporal`` does not
4794 have any defined semantics for atomic stores.
4796 The optional constant ``align`` argument specifies the alignment of the
4797 operation (that is, the alignment of the memory address). A value of 0
4798 or an omitted ``align`` argument means that the operation has the ABI
4799 alignment for the target. It is the responsibility of the code emitter
4800 to ensure that the alignment information is correct. Overestimating the
4801 alignment results in undefined behavior. Underestimating the
4802 alignment may produce less efficient code. An alignment of 1 is always
4805 The optional ``!nontemporal`` metadata must reference a single metadata
4806 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4807 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4808 tells the optimizer and code generator that this load is not expected to
4809 be reused in the cache. The code generator may select special
4810 instructions to save cache bandwidth, such as the MOVNT instruction on
4816 The contents of memory are updated to contain ``<value>`` at the
4817 location specified by the ``<pointer>`` operand. If ``<value>`` is
4818 of scalar type then the number of bytes written does not exceed the
4819 minimum number of bytes needed to hold all bits of the type. For
4820 example, storing an ``i24`` writes at most three bytes. When writing a
4821 value of a type like ``i20`` with a size that is not an integral number
4822 of bytes, it is unspecified what happens to the extra bits that do not
4823 belong to the type, but they will typically be overwritten.
4828 .. code-block:: llvm
4830 %ptr = alloca i32 ; yields {i32*}:ptr
4831 store i32 3, i32* %ptr ; yields {void}
4832 %val = load i32* %ptr ; yields {i32}:val = i32 3
4836 '``fence``' Instruction
4837 ^^^^^^^^^^^^^^^^^^^^^^^
4844 fence [singlethread] <ordering> ; yields {void}
4849 The '``fence``' instruction is used to introduce happens-before edges
4855 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4856 defines what *synchronizes-with* edges they add. They can only be given
4857 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4862 A fence A which has (at least) ``release`` ordering semantics
4863 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4864 semantics if and only if there exist atomic operations X and Y, both
4865 operating on some atomic object M, such that A is sequenced before X, X
4866 modifies M (either directly or through some side effect of a sequence
4867 headed by X), Y is sequenced before B, and Y observes M. This provides a
4868 *happens-before* dependency between A and B. Rather than an explicit
4869 ``fence``, one (but not both) of the atomic operations X or Y might
4870 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4871 still *synchronize-with* the explicit ``fence`` and establish the
4872 *happens-before* edge.
4874 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4875 ``acquire`` and ``release`` semantics specified above, participates in
4876 the global program order of other ``seq_cst`` operations and/or fences.
4878 The optional ":ref:`singlethread <singlethread>`" argument specifies
4879 that the fence only synchronizes with other fences in the same thread.
4880 (This is useful for interacting with signal handlers.)
4885 .. code-block:: llvm
4887 fence acquire ; yields {void}
4888 fence singlethread seq_cst ; yields {void}
4892 '``cmpxchg``' Instruction
4893 ^^^^^^^^^^^^^^^^^^^^^^^^^
4900 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4905 The '``cmpxchg``' instruction is used to atomically modify memory. It
4906 loads a value in memory and compares it to a given value. If they are
4907 equal, it stores a new value into the memory.
4912 There are three arguments to the '``cmpxchg``' instruction: an address
4913 to operate on, a value to compare to the value currently be at that
4914 address, and a new value to place at that address if the compared values
4915 are equal. The type of '<cmp>' must be an integer type whose bit width
4916 is a power of two greater than or equal to eight and less than or equal
4917 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4918 type, and the type of '<pointer>' must be a pointer to that type. If the
4919 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4920 to modify the number or order of execution of this ``cmpxchg`` with
4921 other :ref:`volatile operations <volatile>`.
4923 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4924 synchronizes with other atomic operations.
4926 The optional "``singlethread``" argument declares that the ``cmpxchg``
4927 is only atomic with respect to code (usually signal handlers) running in
4928 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4929 respect to all other code in the system.
4931 The pointer passed into cmpxchg must have alignment greater than or
4932 equal to the size in memory of the operand.
4937 The contents of memory at the location specified by the '``<pointer>``'
4938 operand is read and compared to '``<cmp>``'; if the read value is the
4939 equal, '``<new>``' is written. The original value at the location is
4942 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4943 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4944 atomic load with an ordering parameter determined by dropping any
4945 ``release`` part of the ``cmpxchg``'s ordering.
4950 .. code-block:: llvm
4953 %orig = atomic load i32* %ptr unordered ; yields {i32}
4957 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4958 %squared = mul i32 %cmp, %cmp
4959 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4960 %success = icmp eq i32 %cmp, %old
4961 br i1 %success, label %done, label %loop
4968 '``atomicrmw``' Instruction
4969 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4976 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4981 The '``atomicrmw``' instruction is used to atomically modify memory.
4986 There are three arguments to the '``atomicrmw``' instruction: an
4987 operation to apply, an address whose value to modify, an argument to the
4988 operation. The operation must be one of the following keywords:
5002 The type of '<value>' must be an integer type whose bit width is a power
5003 of two greater than or equal to eight and less than or equal to a
5004 target-specific size limit. The type of the '``<pointer>``' operand must
5005 be a pointer to that type. If the ``atomicrmw`` is marked as
5006 ``volatile``, then the optimizer is not allowed to modify the number or
5007 order of execution of this ``atomicrmw`` with other :ref:`volatile
5008 operations <volatile>`.
5013 The contents of memory at the location specified by the '``<pointer>``'
5014 operand are atomically read, modified, and written back. The original
5015 value at the location is returned. The modification is specified by the
5018 - xchg: ``*ptr = val``
5019 - add: ``*ptr = *ptr + val``
5020 - sub: ``*ptr = *ptr - val``
5021 - and: ``*ptr = *ptr & val``
5022 - nand: ``*ptr = ~(*ptr & val)``
5023 - or: ``*ptr = *ptr | val``
5024 - xor: ``*ptr = *ptr ^ val``
5025 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5026 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5027 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5029 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5035 .. code-block:: llvm
5037 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5039 .. _i_getelementptr:
5041 '``getelementptr``' Instruction
5042 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5049 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5050 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5051 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5056 The '``getelementptr``' instruction is used to get the address of a
5057 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5058 address calculation only and does not access memory.
5063 The first argument is always a pointer or a vector of pointers, and
5064 forms the basis of the calculation. The remaining arguments are indices
5065 that indicate which of the elements of the aggregate object are indexed.
5066 The interpretation of each index is dependent on the type being indexed
5067 into. The first index always indexes the pointer value given as the
5068 first argument, the second index indexes a value of the type pointed to
5069 (not necessarily the value directly pointed to, since the first index
5070 can be non-zero), etc. The first type indexed into must be a pointer
5071 value, subsequent types can be arrays, vectors, and structs. Note that
5072 subsequent types being indexed into can never be pointers, since that
5073 would require loading the pointer before continuing calculation.
5075 The type of each index argument depends on the type it is indexing into.
5076 When indexing into a (optionally packed) structure, only ``i32`` integer
5077 **constants** are allowed (when using a vector of indices they must all
5078 be the **same** ``i32`` integer constant). When indexing into an array,
5079 pointer or vector, integers of any width are allowed, and they are not
5080 required to be constant. These integers are treated as signed values
5083 For example, let's consider a C code fragment and how it gets compiled
5099 int *foo(struct ST *s) {
5100 return &s[1].Z.B[5][13];
5103 The LLVM code generated by Clang is:
5105 .. code-block:: llvm
5107 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5108 %struct.ST = type { i32, double, %struct.RT }
5110 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5112 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5119 In the example above, the first index is indexing into the
5120 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5121 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5122 indexes into the third element of the structure, yielding a
5123 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5124 structure. The third index indexes into the second element of the
5125 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5126 dimensions of the array are subscripted into, yielding an '``i32``'
5127 type. The '``getelementptr``' instruction returns a pointer to this
5128 element, thus computing a value of '``i32*``' type.
5130 Note that it is perfectly legal to index partially through a structure,
5131 returning a pointer to an inner element. Because of this, the LLVM code
5132 for the given testcase is equivalent to:
5134 .. code-block:: llvm
5136 define i32* @foo(%struct.ST* %s) {
5137 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5138 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5139 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5140 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5141 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5145 If the ``inbounds`` keyword is present, the result value of the
5146 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5147 pointer is not an *in bounds* address of an allocated object, or if any
5148 of the addresses that would be formed by successive addition of the
5149 offsets implied by the indices to the base address with infinitely
5150 precise signed arithmetic are not an *in bounds* address of that
5151 allocated object. The *in bounds* addresses for an allocated object are
5152 all the addresses that point into the object, plus the address one byte
5153 past the end. In cases where the base is a vector of pointers the
5154 ``inbounds`` keyword applies to each of the computations element-wise.
5156 If the ``inbounds`` keyword is not present, the offsets are added to the
5157 base address with silently-wrapping two's complement arithmetic. If the
5158 offsets have a different width from the pointer, they are sign-extended
5159 or truncated to the width of the pointer. The result value of the
5160 ``getelementptr`` may be outside the object pointed to by the base
5161 pointer. The result value may not necessarily be used to access memory
5162 though, even if it happens to point into allocated storage. See the
5163 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5166 The getelementptr instruction is often confusing. For some more insight
5167 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5172 .. code-block:: llvm
5174 ; yields [12 x i8]*:aptr
5175 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5177 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5179 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5181 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5183 In cases where the pointer argument is a vector of pointers, each index
5184 must be a vector with the same number of elements. For example:
5186 .. code-block:: llvm
5188 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5190 Conversion Operations
5191 ---------------------
5193 The instructions in this category are the conversion instructions
5194 (casting) which all take a single operand and a type. They perform
5195 various bit conversions on the operand.
5197 '``trunc .. to``' Instruction
5198 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5205 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5210 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5215 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5216 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5217 of the same number of integers. The bit size of the ``value`` must be
5218 larger than the bit size of the destination type, ``ty2``. Equal sized
5219 types are not allowed.
5224 The '``trunc``' instruction truncates the high order bits in ``value``
5225 and converts the remaining bits to ``ty2``. Since the source size must
5226 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5227 It will always truncate bits.
5232 .. code-block:: llvm
5234 %X = trunc i32 257 to i8 ; yields i8:1
5235 %Y = trunc i32 123 to i1 ; yields i1:true
5236 %Z = trunc i32 122 to i1 ; yields i1:false
5237 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5239 '``zext .. to``' Instruction
5240 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5247 <result> = zext <ty> <value> to <ty2> ; yields ty2
5252 The '``zext``' instruction zero extends its operand to type ``ty2``.
5257 The '``zext``' instruction takes a value to cast, and a type to cast it
5258 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5259 the same number of integers. The bit size of the ``value`` must be
5260 smaller than the bit size of the destination type, ``ty2``.
5265 The ``zext`` fills the high order bits of the ``value`` with zero bits
5266 until it reaches the size of the destination type, ``ty2``.
5268 When zero extending from i1, the result will always be either 0 or 1.
5273 .. code-block:: llvm
5275 %X = zext i32 257 to i64 ; yields i64:257
5276 %Y = zext i1 true to i32 ; yields i32:1
5277 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5279 '``sext .. to``' Instruction
5280 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5287 <result> = sext <ty> <value> to <ty2> ; yields ty2
5292 The '``sext``' sign extends ``value`` to the type ``ty2``.
5297 The '``sext``' instruction takes a value to cast, and a type to cast it
5298 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5299 the same number of integers. The bit size of the ``value`` must be
5300 smaller than the bit size of the destination type, ``ty2``.
5305 The '``sext``' instruction performs a sign extension by copying the sign
5306 bit (highest order bit) of the ``value`` until it reaches the bit size
5307 of the type ``ty2``.
5309 When sign extending from i1, the extension always results in -1 or 0.
5314 .. code-block:: llvm
5316 %X = sext i8 -1 to i16 ; yields i16 :65535
5317 %Y = sext i1 true to i32 ; yields i32:-1
5318 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5320 '``fptrunc .. to``' Instruction
5321 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5328 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5333 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5338 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5339 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5340 The size of ``value`` must be larger than the size of ``ty2``. This
5341 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5346 The '``fptrunc``' instruction truncates a ``value`` from a larger
5347 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5348 point <t_floating>` type. If the value cannot fit within the
5349 destination type, ``ty2``, then the results are undefined.
5354 .. code-block:: llvm
5356 %X = fptrunc double 123.0 to float ; yields float:123.0
5357 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5359 '``fpext .. to``' Instruction
5360 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5367 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5372 The '``fpext``' extends a floating point ``value`` to a larger floating
5378 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5379 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5380 to. The source type must be smaller than the destination type.
5385 The '``fpext``' instruction extends the ``value`` from a smaller
5386 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5387 point <t_floating>` type. The ``fpext`` cannot be used to make a
5388 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5389 *no-op cast* for a floating point cast.
5394 .. code-block:: llvm
5396 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5397 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5399 '``fptoui .. to``' Instruction
5400 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5407 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5412 The '``fptoui``' converts a floating point ``value`` to its unsigned
5413 integer equivalent of type ``ty2``.
5418 The '``fptoui``' instruction takes a value to cast, which must be a
5419 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5420 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5421 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5422 type with the same number of elements as ``ty``
5427 The '``fptoui``' instruction converts its :ref:`floating
5428 point <t_floating>` operand into the nearest (rounding towards zero)
5429 unsigned integer value. If the value cannot fit in ``ty2``, the results
5435 .. code-block:: llvm
5437 %X = fptoui double 123.0 to i32 ; yields i32:123
5438 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5439 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5441 '``fptosi .. to``' Instruction
5442 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5449 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5454 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5455 ``value`` to type ``ty2``.
5460 The '``fptosi``' instruction takes a value to cast, which must be a
5461 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5462 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5463 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5464 type with the same number of elements as ``ty``
5469 The '``fptosi``' instruction converts its :ref:`floating
5470 point <t_floating>` operand into the nearest (rounding towards zero)
5471 signed integer value. If the value cannot fit in ``ty2``, the results
5477 .. code-block:: llvm
5479 %X = fptosi double -123.0 to i32 ; yields i32:-123
5480 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5481 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5483 '``uitofp .. to``' Instruction
5484 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5491 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5496 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5497 and converts that value to the ``ty2`` type.
5502 The '``uitofp``' instruction takes a value to cast, which must be a
5503 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5504 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5505 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5506 type with the same number of elements as ``ty``
5511 The '``uitofp``' instruction interprets its operand as an unsigned
5512 integer quantity and converts it to the corresponding floating point
5513 value. If the value cannot fit in the floating point value, the results
5519 .. code-block:: llvm
5521 %X = uitofp i32 257 to float ; yields float:257.0
5522 %Y = uitofp i8 -1 to double ; yields double:255.0
5524 '``sitofp .. to``' Instruction
5525 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5532 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5537 The '``sitofp``' instruction regards ``value`` as a signed integer and
5538 converts that value to the ``ty2`` type.
5543 The '``sitofp``' instruction takes a value to cast, which must be a
5544 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5545 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5546 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5547 type with the same number of elements as ``ty``
5552 The '``sitofp``' instruction interprets its operand as a signed integer
5553 quantity and converts it to the corresponding floating point value. If
5554 the value cannot fit in the floating point value, the results are
5560 .. code-block:: llvm
5562 %X = sitofp i32 257 to float ; yields float:257.0
5563 %Y = sitofp i8 -1 to double ; yields double:-1.0
5567 '``ptrtoint .. to``' Instruction
5568 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5575 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5580 The '``ptrtoint``' instruction converts the pointer or a vector of
5581 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5586 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5587 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5588 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5589 a vector of integers type.
5594 The '``ptrtoint``' instruction converts ``value`` to integer type
5595 ``ty2`` by interpreting the pointer value as an integer and either
5596 truncating or zero extending that value to the size of the integer type.
5597 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5598 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5599 the same size, then nothing is done (*no-op cast*) other than a type
5605 .. code-block:: llvm
5607 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5608 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5609 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5613 '``inttoptr .. to``' Instruction
5614 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5621 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5626 The '``inttoptr``' instruction converts an integer ``value`` to a
5627 pointer type, ``ty2``.
5632 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5633 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5639 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5640 applying either a zero extension or a truncation depending on the size
5641 of the integer ``value``. If ``value`` is larger than the size of a
5642 pointer then a truncation is done. If ``value`` is smaller than the size
5643 of a pointer then a zero extension is done. If they are the same size,
5644 nothing is done (*no-op cast*).
5649 .. code-block:: llvm
5651 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5652 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5653 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5654 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5658 '``bitcast .. to``' Instruction
5659 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5666 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5671 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5677 The '``bitcast``' instruction takes a value to cast, which must be a
5678 non-aggregate first class value, and a type to cast it to, which must
5679 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5680 bit sizes of ``value`` and the destination type, ``ty2``, must be
5681 identical. If the source type is a pointer, the destination type must
5682 also be a pointer of the same size. This instruction supports bitwise
5683 conversion of vectors to integers and to vectors of other types (as
5684 long as they have the same size).
5689 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5690 is always a *no-op cast* because no bits change with this
5691 conversion. The conversion is done as if the ``value`` had been stored
5692 to memory and read back as type ``ty2``. Pointer (or vector of
5693 pointers) types may only be converted to other pointer (or vector of
5694 pointers) types with this instruction if the pointer sizes are
5695 equal. To convert pointers to other types, use the :ref:`inttoptr
5696 <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5701 .. code-block:: llvm
5703 %X = bitcast i8 255 to i8 ; yields i8 :-1
5704 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5705 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5706 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5713 The instructions in this category are the "miscellaneous" instructions,
5714 which defy better classification.
5718 '``icmp``' Instruction
5719 ^^^^^^^^^^^^^^^^^^^^^^
5726 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5731 The '``icmp``' instruction returns a boolean value or a vector of
5732 boolean values based on comparison of its two integer, integer vector,
5733 pointer, or pointer vector operands.
5738 The '``icmp``' instruction takes three operands. The first operand is
5739 the condition code indicating the kind of comparison to perform. It is
5740 not a value, just a keyword. The possible condition code are:
5743 #. ``ne``: not equal
5744 #. ``ugt``: unsigned greater than
5745 #. ``uge``: unsigned greater or equal
5746 #. ``ult``: unsigned less than
5747 #. ``ule``: unsigned less or equal
5748 #. ``sgt``: signed greater than
5749 #. ``sge``: signed greater or equal
5750 #. ``slt``: signed less than
5751 #. ``sle``: signed less or equal
5753 The remaining two arguments must be :ref:`integer <t_integer>` or
5754 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5755 must also be identical types.
5760 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5761 code given as ``cond``. The comparison performed always yields either an
5762 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5764 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5765 otherwise. No sign interpretation is necessary or performed.
5766 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5767 otherwise. No sign interpretation is necessary or performed.
5768 #. ``ugt``: interprets the operands as unsigned values and yields
5769 ``true`` if ``op1`` is greater than ``op2``.
5770 #. ``uge``: interprets the operands as unsigned values and yields
5771 ``true`` if ``op1`` is greater than or equal to ``op2``.
5772 #. ``ult``: interprets the operands as unsigned values and yields
5773 ``true`` if ``op1`` is less than ``op2``.
5774 #. ``ule``: interprets the operands as unsigned values and yields
5775 ``true`` if ``op1`` is less than or equal to ``op2``.
5776 #. ``sgt``: interprets the operands as signed values and yields ``true``
5777 if ``op1`` is greater than ``op2``.
5778 #. ``sge``: interprets the operands as signed values and yields ``true``
5779 if ``op1`` is greater than or equal to ``op2``.
5780 #. ``slt``: interprets the operands as signed values and yields ``true``
5781 if ``op1`` is less than ``op2``.
5782 #. ``sle``: interprets the operands as signed values and yields ``true``
5783 if ``op1`` is less than or equal to ``op2``.
5785 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5786 are compared as if they were integers.
5788 If the operands are integer vectors, then they are compared element by
5789 element. The result is an ``i1`` vector with the same number of elements
5790 as the values being compared. Otherwise, the result is an ``i1``.
5795 .. code-block:: llvm
5797 <result> = icmp eq i32 4, 5 ; yields: result=false
5798 <result> = icmp ne float* %X, %X ; yields: result=false
5799 <result> = icmp ult i16 4, 5 ; yields: result=true
5800 <result> = icmp sgt i16 4, 5 ; yields: result=false
5801 <result> = icmp ule i16 -4, 5 ; yields: result=false
5802 <result> = icmp sge i16 4, 5 ; yields: result=false
5804 Note that the code generator does not yet support vector types with the
5805 ``icmp`` instruction.
5809 '``fcmp``' Instruction
5810 ^^^^^^^^^^^^^^^^^^^^^^
5817 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5822 The '``fcmp``' instruction returns a boolean value or vector of boolean
5823 values based on comparison of its operands.
5825 If the operands are floating point scalars, then the result type is a
5826 boolean (:ref:`i1 <t_integer>`).
5828 If the operands are floating point vectors, then the result type is a
5829 vector of boolean with the same number of elements as the operands being
5835 The '``fcmp``' instruction takes three operands. The first operand is
5836 the condition code indicating the kind of comparison to perform. It is
5837 not a value, just a keyword. The possible condition code are:
5839 #. ``false``: no comparison, always returns false
5840 #. ``oeq``: ordered and equal
5841 #. ``ogt``: ordered and greater than
5842 #. ``oge``: ordered and greater than or equal
5843 #. ``olt``: ordered and less than
5844 #. ``ole``: ordered and less than or equal
5845 #. ``one``: ordered and not equal
5846 #. ``ord``: ordered (no nans)
5847 #. ``ueq``: unordered or equal
5848 #. ``ugt``: unordered or greater than
5849 #. ``uge``: unordered or greater than or equal
5850 #. ``ult``: unordered or less than
5851 #. ``ule``: unordered or less than or equal
5852 #. ``une``: unordered or not equal
5853 #. ``uno``: unordered (either nans)
5854 #. ``true``: no comparison, always returns true
5856 *Ordered* means that neither operand is a QNAN while *unordered* means
5857 that either operand may be a QNAN.
5859 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5860 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5861 type. They must have identical types.
5866 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5867 condition code given as ``cond``. If the operands are vectors, then the
5868 vectors are compared element by element. Each comparison performed
5869 always yields an :ref:`i1 <t_integer>` result, as follows:
5871 #. ``false``: always yields ``false``, regardless of operands.
5872 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5873 is equal to ``op2``.
5874 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5875 is greater than ``op2``.
5876 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5877 is greater than or equal to ``op2``.
5878 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5879 is less than ``op2``.
5880 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5881 is less than or equal to ``op2``.
5882 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5883 is not equal to ``op2``.
5884 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5885 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5887 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5888 greater than ``op2``.
5889 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5890 greater than or equal to ``op2``.
5891 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5893 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5894 less than or equal to ``op2``.
5895 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5896 not equal to ``op2``.
5897 #. ``uno``: yields ``true`` if either operand is a QNAN.
5898 #. ``true``: always yields ``true``, regardless of operands.
5903 .. code-block:: llvm
5905 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5906 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5907 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5908 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5910 Note that the code generator does not yet support vector types with the
5911 ``fcmp`` instruction.
5915 '``phi``' Instruction
5916 ^^^^^^^^^^^^^^^^^^^^^
5923 <result> = phi <ty> [ <val0>, <label0>], ...
5928 The '``phi``' instruction is used to implement the φ node in the SSA
5929 graph representing the function.
5934 The type of the incoming values is specified with the first type field.
5935 After this, the '``phi``' instruction takes a list of pairs as
5936 arguments, with one pair for each predecessor basic block of the current
5937 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5938 the value arguments to the PHI node. Only labels may be used as the
5941 There must be no non-phi instructions between the start of a basic block
5942 and the PHI instructions: i.e. PHI instructions must be first in a basic
5945 For the purposes of the SSA form, the use of each incoming value is
5946 deemed to occur on the edge from the corresponding predecessor block to
5947 the current block (but after any definition of an '``invoke``'
5948 instruction's return value on the same edge).
5953 At runtime, the '``phi``' instruction logically takes on the value
5954 specified by the pair corresponding to the predecessor basic block that
5955 executed just prior to the current block.
5960 .. code-block:: llvm
5962 Loop: ; Infinite loop that counts from 0 on up...
5963 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5964 %nextindvar = add i32 %indvar, 1
5969 '``select``' Instruction
5970 ^^^^^^^^^^^^^^^^^^^^^^^^
5977 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5979 selty is either i1 or {<N x i1>}
5984 The '``select``' instruction is used to choose one value based on a
5985 condition, without branching.
5990 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5991 values indicating the condition, and two values of the same :ref:`first
5992 class <t_firstclass>` type. If the val1/val2 are vectors and the
5993 condition is a scalar, then entire vectors are selected, not individual
5999 If the condition is an i1 and it evaluates to 1, the instruction returns
6000 the first value argument; otherwise, it returns the second value
6003 If the condition is a vector of i1, then the value arguments must be
6004 vectors of the same size, and the selection is done element by element.
6009 .. code-block:: llvm
6011 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6015 '``call``' Instruction
6016 ^^^^^^^^^^^^^^^^^^^^^^
6023 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6028 The '``call``' instruction represents a simple function call.
6033 This instruction requires several arguments:
6035 #. The optional "tail" marker indicates that the callee function does
6036 not access any allocas or varargs in the caller. Note that calls may
6037 be marked "tail" even if they do not occur before a
6038 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6039 function call is eligible for tail call optimization, but `might not
6040 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6041 The code generator may optimize calls marked "tail" with either 1)
6042 automatic `sibling call
6043 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6044 callee have matching signatures, or 2) forced tail call optimization
6045 when the following extra requirements are met:
6047 - Caller and callee both have the calling convention ``fastcc``.
6048 - The call is in tail position (ret immediately follows call and ret
6049 uses value of call or is void).
6050 - Option ``-tailcallopt`` is enabled, or
6051 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6052 - `Platform specific constraints are
6053 met. <CodeGenerator.html#tailcallopt>`_
6055 #. The optional "cconv" marker indicates which :ref:`calling
6056 convention <callingconv>` the call should use. If none is
6057 specified, the call defaults to using C calling conventions. The
6058 calling convention of the call must match the calling convention of
6059 the target function, or else the behavior is undefined.
6060 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6061 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6063 #. '``ty``': the type of the call instruction itself which is also the
6064 type of the return value. Functions that return no value are marked
6066 #. '``fnty``': shall be the signature of the pointer to function value
6067 being invoked. The argument types must match the types implied by
6068 this signature. This type can be omitted if the function is not
6069 varargs and if the function type does not return a pointer to a
6071 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6072 be invoked. In most cases, this is a direct function invocation, but
6073 indirect ``call``'s are just as possible, calling an arbitrary pointer
6075 #. '``function args``': argument list whose types match the function
6076 signature argument types and parameter attributes. All arguments must
6077 be of :ref:`first class <t_firstclass>` type. If the function signature
6078 indicates the function accepts a variable number of arguments, the
6079 extra arguments can be specified.
6080 #. The optional :ref:`function attributes <fnattrs>` list. Only
6081 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6082 attributes are valid here.
6087 The '``call``' instruction is used to cause control flow to transfer to
6088 a specified function, with its incoming arguments bound to the specified
6089 values. Upon a '``ret``' instruction in the called function, control
6090 flow continues with the instruction after the function call, and the
6091 return value of the function is bound to the result argument.
6096 .. code-block:: llvm
6098 %retval = call i32 @test(i32 %argc)
6099 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6100 %X = tail call i32 @foo() ; yields i32
6101 %Y = tail call fastcc i32 @foo() ; yields i32
6102 call void %foo(i8 97 signext)
6104 %struct.A = type { i32, i8 }
6105 %r = call %struct.A @foo() ; yields { 32, i8 }
6106 %gr = extractvalue %struct.A %r, 0 ; yields i32
6107 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6108 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6109 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6111 llvm treats calls to some functions with names and arguments that match
6112 the standard C99 library as being the C99 library functions, and may
6113 perform optimizations or generate code for them under that assumption.
6114 This is something we'd like to change in the future to provide better
6115 support for freestanding environments and non-C-based languages.
6119 '``va_arg``' Instruction
6120 ^^^^^^^^^^^^^^^^^^^^^^^^
6127 <resultval> = va_arg <va_list*> <arglist>, <argty>
6132 The '``va_arg``' instruction is used to access arguments passed through
6133 the "variable argument" area of a function call. It is used to implement
6134 the ``va_arg`` macro in C.
6139 This instruction takes a ``va_list*`` value and the type of the
6140 argument. It returns a value of the specified argument type and
6141 increments the ``va_list`` to point to the next argument. The actual
6142 type of ``va_list`` is target specific.
6147 The '``va_arg``' instruction loads an argument of the specified type
6148 from the specified ``va_list`` and causes the ``va_list`` to point to
6149 the next argument. For more information, see the variable argument
6150 handling :ref:`Intrinsic Functions <int_varargs>`.
6152 It is legal for this instruction to be called in a function which does
6153 not take a variable number of arguments, for example, the ``vfprintf``
6156 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6157 function <intrinsics>` because it takes a type as an argument.
6162 See the :ref:`variable argument processing <int_varargs>` section.
6164 Note that the code generator does not yet fully support va\_arg on many
6165 targets. Also, it does not currently support va\_arg with aggregate
6166 types on any target.
6170 '``landingpad``' Instruction
6171 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6178 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6179 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6181 <clause> := catch <type> <value>
6182 <clause> := filter <array constant type> <array constant>
6187 The '``landingpad``' instruction is used by `LLVM's exception handling
6188 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6189 is a landing pad --- one where the exception lands, and corresponds to the
6190 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6191 defines values supplied by the personality function (``pers_fn``) upon
6192 re-entry to the function. The ``resultval`` has the type ``resultty``.
6197 This instruction takes a ``pers_fn`` value. This is the personality
6198 function associated with the unwinding mechanism. The optional
6199 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6201 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6202 contains the global variable representing the "type" that may be caught
6203 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6204 clause takes an array constant as its argument. Use
6205 "``[0 x i8**] undef``" for a filter which cannot throw. The
6206 '``landingpad``' instruction must contain *at least* one ``clause`` or
6207 the ``cleanup`` flag.
6212 The '``landingpad``' instruction defines the values which are set by the
6213 personality function (``pers_fn``) upon re-entry to the function, and
6214 therefore the "result type" of the ``landingpad`` instruction. As with
6215 calling conventions, how the personality function results are
6216 represented in LLVM IR is target specific.
6218 The clauses are applied in order from top to bottom. If two
6219 ``landingpad`` instructions are merged together through inlining, the
6220 clauses from the calling function are appended to the list of clauses.
6221 When the call stack is being unwound due to an exception being thrown,
6222 the exception is compared against each ``clause`` in turn. If it doesn't
6223 match any of the clauses, and the ``cleanup`` flag is not set, then
6224 unwinding continues further up the call stack.
6226 The ``landingpad`` instruction has several restrictions:
6228 - A landing pad block is a basic block which is the unwind destination
6229 of an '``invoke``' instruction.
6230 - A landing pad block must have a '``landingpad``' instruction as its
6231 first non-PHI instruction.
6232 - There can be only one '``landingpad``' instruction within the landing
6234 - A basic block that is not a landing pad block may not include a
6235 '``landingpad``' instruction.
6236 - All '``landingpad``' instructions in a function must have the same
6237 personality function.
6242 .. code-block:: llvm
6244 ;; A landing pad which can catch an integer.
6245 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6247 ;; A landing pad that is a cleanup.
6248 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6250 ;; A landing pad which can catch an integer and can only throw a double.
6251 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6253 filter [1 x i8**] [@_ZTId]
6260 LLVM supports the notion of an "intrinsic function". These functions
6261 have well known names and semantics and are required to follow certain
6262 restrictions. Overall, these intrinsics represent an extension mechanism
6263 for the LLVM language that does not require changing all of the
6264 transformations in LLVM when adding to the language (or the bitcode
6265 reader/writer, the parser, etc...).
6267 Intrinsic function names must all start with an "``llvm.``" prefix. This
6268 prefix is reserved in LLVM for intrinsic names; thus, function names may
6269 not begin with this prefix. Intrinsic functions must always be external
6270 functions: you cannot define the body of intrinsic functions. Intrinsic
6271 functions may only be used in call or invoke instructions: it is illegal
6272 to take the address of an intrinsic function. Additionally, because
6273 intrinsic functions are part of the LLVM language, it is required if any
6274 are added that they be documented here.
6276 Some intrinsic functions can be overloaded, i.e., the intrinsic
6277 represents a family of functions that perform the same operation but on
6278 different data types. Because LLVM can represent over 8 million
6279 different integer types, overloading is used commonly to allow an
6280 intrinsic function to operate on any integer type. One or more of the
6281 argument types or the result type can be overloaded to accept any
6282 integer type. Argument types may also be defined as exactly matching a
6283 previous argument's type or the result type. This allows an intrinsic
6284 function which accepts multiple arguments, but needs all of them to be
6285 of the same type, to only be overloaded with respect to a single
6286 argument or the result.
6288 Overloaded intrinsics will have the names of its overloaded argument
6289 types encoded into its function name, each preceded by a period. Only
6290 those types which are overloaded result in a name suffix. Arguments
6291 whose type is matched against another type do not. For example, the
6292 ``llvm.ctpop`` function can take an integer of any width and returns an
6293 integer of exactly the same integer width. This leads to a family of
6294 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6295 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6296 overloaded, and only one type suffix is required. Because the argument's
6297 type is matched against the return type, it does not require its own
6300 To learn how to add an intrinsic function, please see the `Extending
6301 LLVM Guide <ExtendingLLVM.html>`_.
6305 Variable Argument Handling Intrinsics
6306 -------------------------------------
6308 Variable argument support is defined in LLVM with the
6309 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6310 functions. These functions are related to the similarly named macros
6311 defined in the ``<stdarg.h>`` header file.
6313 All of these functions operate on arguments that use a target-specific
6314 value type "``va_list``". The LLVM assembly language reference manual
6315 does not define what this type is, so all transformations should be
6316 prepared to handle these functions regardless of the type used.
6318 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6319 variable argument handling intrinsic functions are used.
6321 .. code-block:: llvm
6323 define i32 @test(i32 %X, ...) {
6324 ; Initialize variable argument processing
6326 %ap2 = bitcast i8** %ap to i8*
6327 call void @llvm.va_start(i8* %ap2)
6329 ; Read a single integer argument
6330 %tmp = va_arg i8** %ap, i32
6332 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6334 %aq2 = bitcast i8** %aq to i8*
6335 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6336 call void @llvm.va_end(i8* %aq2)
6338 ; Stop processing of arguments.
6339 call void @llvm.va_end(i8* %ap2)
6343 declare void @llvm.va_start(i8*)
6344 declare void @llvm.va_copy(i8*, i8*)
6345 declare void @llvm.va_end(i8*)
6349 '``llvm.va_start``' Intrinsic
6350 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6357 declare void @llvm.va_start(i8* <arglist>)
6362 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6363 subsequent use by ``va_arg``.
6368 The argument is a pointer to a ``va_list`` element to initialize.
6373 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6374 available in C. In a target-dependent way, it initializes the
6375 ``va_list`` element to which the argument points, so that the next call
6376 to ``va_arg`` will produce the first variable argument passed to the
6377 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6378 to know the last argument of the function as the compiler can figure
6381 '``llvm.va_end``' Intrinsic
6382 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6389 declare void @llvm.va_end(i8* <arglist>)
6394 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6395 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6400 The argument is a pointer to a ``va_list`` to destroy.
6405 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6406 available in C. In a target-dependent way, it destroys the ``va_list``
6407 element to which the argument points. Calls to
6408 :ref:`llvm.va_start <int_va_start>` and
6409 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6414 '``llvm.va_copy``' Intrinsic
6415 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6422 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6427 The '``llvm.va_copy``' intrinsic copies the current argument position
6428 from the source argument list to the destination argument list.
6433 The first argument is a pointer to a ``va_list`` element to initialize.
6434 The second argument is a pointer to a ``va_list`` element to copy from.
6439 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6440 available in C. In a target-dependent way, it copies the source
6441 ``va_list`` element into the destination ``va_list`` element. This
6442 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6443 arbitrarily complex and require, for example, memory allocation.
6445 Accurate Garbage Collection Intrinsics
6446 --------------------------------------
6448 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6449 (GC) requires the implementation and generation of these intrinsics.
6450 These intrinsics allow identification of :ref:`GC roots on the
6451 stack <int_gcroot>`, as well as garbage collector implementations that
6452 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6453 Front-ends for type-safe garbage collected languages should generate
6454 these intrinsics to make use of the LLVM garbage collectors. For more
6455 details, see `Accurate Garbage Collection with
6456 LLVM <GarbageCollection.html>`_.
6458 The garbage collection intrinsics only operate on objects in the generic
6459 address space (address space zero).
6463 '``llvm.gcroot``' Intrinsic
6464 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6471 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6476 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6477 the code generator, and allows some metadata to be associated with it.
6482 The first argument specifies the address of a stack object that contains
6483 the root pointer. The second pointer (which must be either a constant or
6484 a global value address) contains the meta-data to be associated with the
6490 At runtime, a call to this intrinsic stores a null pointer into the
6491 "ptrloc" location. At compile-time, the code generator generates
6492 information to allow the runtime to find the pointer at GC safe points.
6493 The '``llvm.gcroot``' intrinsic may only be used in a function which
6494 :ref:`specifies a GC algorithm <gc>`.
6498 '``llvm.gcread``' Intrinsic
6499 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6506 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6511 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6512 locations, allowing garbage collector implementations that require read
6518 The second argument is the address to read from, which should be an
6519 address allocated from the garbage collector. The first object is a
6520 pointer to the start of the referenced object, if needed by the language
6521 runtime (otherwise null).
6526 The '``llvm.gcread``' intrinsic has the same semantics as a load
6527 instruction, but may be replaced with substantially more complex code by
6528 the garbage collector runtime, as needed. The '``llvm.gcread``'
6529 intrinsic may only be used in a function which :ref:`specifies a GC
6534 '``llvm.gcwrite``' Intrinsic
6535 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6542 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6547 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6548 locations, allowing garbage collector implementations that require write
6549 barriers (such as generational or reference counting collectors).
6554 The first argument is the reference to store, the second is the start of
6555 the object to store it to, and the third is the address of the field of
6556 Obj to store to. If the runtime does not require a pointer to the
6557 object, Obj may be null.
6562 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6563 instruction, but may be replaced with substantially more complex code by
6564 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6565 intrinsic may only be used in a function which :ref:`specifies a GC
6568 Code Generator Intrinsics
6569 -------------------------
6571 These intrinsics are provided by LLVM to expose special features that
6572 may only be implemented with code generator support.
6574 '``llvm.returnaddress``' Intrinsic
6575 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6582 declare i8 *@llvm.returnaddress(i32 <level>)
6587 The '``llvm.returnaddress``' intrinsic attempts to compute a
6588 target-specific value indicating the return address of the current
6589 function or one of its callers.
6594 The argument to this intrinsic indicates which function to return the
6595 address for. Zero indicates the calling function, one indicates its
6596 caller, etc. The argument is **required** to be a constant integer
6602 The '``llvm.returnaddress``' intrinsic either returns a pointer
6603 indicating the return address of the specified call frame, or zero if it
6604 cannot be identified. The value returned by this intrinsic is likely to
6605 be incorrect or 0 for arguments other than zero, so it should only be
6606 used for debugging purposes.
6608 Note that calling this intrinsic does not prevent function inlining or
6609 other aggressive transformations, so the value returned may not be that
6610 of the obvious source-language caller.
6612 '``llvm.frameaddress``' Intrinsic
6613 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6620 declare i8* @llvm.frameaddress(i32 <level>)
6625 The '``llvm.frameaddress``' intrinsic attempts to return the
6626 target-specific frame pointer value for the specified stack frame.
6631 The argument to this intrinsic indicates which function to return the
6632 frame pointer for. Zero indicates the calling function, one indicates
6633 its caller, etc. The argument is **required** to be a constant integer
6639 The '``llvm.frameaddress``' intrinsic either returns a pointer
6640 indicating the frame address of the specified call frame, or zero if it
6641 cannot be identified. The value returned by this intrinsic is likely to
6642 be incorrect or 0 for arguments other than zero, so it should only be
6643 used for debugging purposes.
6645 Note that calling this intrinsic does not prevent function inlining or
6646 other aggressive transformations, so the value returned may not be that
6647 of the obvious source-language caller.
6651 '``llvm.stacksave``' Intrinsic
6652 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6659 declare i8* @llvm.stacksave()
6664 The '``llvm.stacksave``' intrinsic is used to remember the current state
6665 of the function stack, for use with
6666 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6667 implementing language features like scoped automatic variable sized
6673 This intrinsic returns a opaque pointer value that can be passed to
6674 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6675 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6676 ``llvm.stacksave``, it effectively restores the state of the stack to
6677 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6678 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6679 were allocated after the ``llvm.stacksave`` was executed.
6681 .. _int_stackrestore:
6683 '``llvm.stackrestore``' Intrinsic
6684 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6691 declare void @llvm.stackrestore(i8* %ptr)
6696 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6697 the function stack to the state it was in when the corresponding
6698 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6699 useful for implementing language features like scoped automatic variable
6700 sized arrays in C99.
6705 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6707 '``llvm.prefetch``' Intrinsic
6708 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6715 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6720 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6721 insert a prefetch instruction if supported; otherwise, it is a noop.
6722 Prefetches have no effect on the behavior of the program but can change
6723 its performance characteristics.
6728 ``address`` is the address to be prefetched, ``rw`` is the specifier
6729 determining if the fetch should be for a read (0) or write (1), and
6730 ``locality`` is a temporal locality specifier ranging from (0) - no
6731 locality, to (3) - extremely local keep in cache. The ``cache type``
6732 specifies whether the prefetch is performed on the data (1) or
6733 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6734 arguments must be constant integers.
6739 This intrinsic does not modify the behavior of the program. In
6740 particular, prefetches cannot trap and do not produce a value. On
6741 targets that support this intrinsic, the prefetch can provide hints to
6742 the processor cache for better performance.
6744 '``llvm.pcmarker``' Intrinsic
6745 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6752 declare void @llvm.pcmarker(i32 <id>)
6757 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6758 Counter (PC) in a region of code to simulators and other tools. The
6759 method is target specific, but it is expected that the marker will use
6760 exported symbols to transmit the PC of the marker. The marker makes no
6761 guarantees that it will remain with any specific instruction after
6762 optimizations. It is possible that the presence of a marker will inhibit
6763 optimizations. The intended use is to be inserted after optimizations to
6764 allow correlations of simulation runs.
6769 ``id`` is a numerical id identifying the marker.
6774 This intrinsic does not modify the behavior of the program. Backends
6775 that do not support this intrinsic may ignore it.
6777 '``llvm.readcyclecounter``' Intrinsic
6778 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6785 declare i64 @llvm.readcyclecounter()
6790 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6791 counter register (or similar low latency, high accuracy clocks) on those
6792 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6793 should map to RPCC. As the backing counters overflow quickly (on the
6794 order of 9 seconds on alpha), this should only be used for small
6800 When directly supported, reading the cycle counter should not modify any
6801 memory. Implementations are allowed to either return a application
6802 specific value or a system wide value. On backends without support, this
6803 is lowered to a constant 0.
6805 Note that runtime support may be conditional on the privilege-level code is
6806 running at and the host platform.
6808 Standard C Library Intrinsics
6809 -----------------------------
6811 LLVM provides intrinsics for a few important standard C library
6812 functions. These intrinsics allow source-language front-ends to pass
6813 information about the alignment of the pointer arguments to the code
6814 generator, providing opportunity for more efficient code generation.
6818 '``llvm.memcpy``' Intrinsic
6819 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6824 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6825 integer bit width and for different address spaces. Not all targets
6826 support all bit widths however.
6830 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6831 i32 <len>, i32 <align>, i1 <isvolatile>)
6832 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6833 i64 <len>, i32 <align>, i1 <isvolatile>)
6838 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6839 source location to the destination location.
6841 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6842 intrinsics do not return a value, takes extra alignment/isvolatile
6843 arguments and the pointers can be in specified address spaces.
6848 The first argument is a pointer to the destination, the second is a
6849 pointer to the source. The third argument is an integer argument
6850 specifying the number of bytes to copy, the fourth argument is the
6851 alignment of the source and destination locations, and the fifth is a
6852 boolean indicating a volatile access.
6854 If the call to this intrinsic has an alignment value that is not 0 or 1,
6855 then the caller guarantees that both the source and destination pointers
6856 are aligned to that boundary.
6858 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6859 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6860 very cleanly specified and it is unwise to depend on it.
6865 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6866 source location to the destination location, which are not allowed to
6867 overlap. It copies "len" bytes of memory over. If the argument is known
6868 to be aligned to some boundary, this can be specified as the fourth
6869 argument, otherwise it should be set to 0 or 1.
6871 '``llvm.memmove``' Intrinsic
6872 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6877 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6878 bit width and for different address space. Not all targets support all
6883 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6884 i32 <len>, i32 <align>, i1 <isvolatile>)
6885 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6886 i64 <len>, i32 <align>, i1 <isvolatile>)
6891 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6892 source location to the destination location. It is similar to the
6893 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6896 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6897 intrinsics do not return a value, takes extra alignment/isvolatile
6898 arguments and the pointers can be in specified address spaces.
6903 The first argument is a pointer to the destination, the second is a
6904 pointer to the source. The third argument is an integer argument
6905 specifying the number of bytes to copy, the fourth argument is the
6906 alignment of the source and destination locations, and the fifth is a
6907 boolean indicating a volatile access.
6909 If the call to this intrinsic has an alignment value that is not 0 or 1,
6910 then the caller guarantees that the source and destination pointers are
6911 aligned to that boundary.
6913 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6914 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6915 not very cleanly specified and it is unwise to depend on it.
6920 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6921 source location to the destination location, which may overlap. It
6922 copies "len" bytes of memory over. If the argument is known to be
6923 aligned to some boundary, this can be specified as the fourth argument,
6924 otherwise it should be set to 0 or 1.
6926 '``llvm.memset.*``' Intrinsics
6927 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6932 This is an overloaded intrinsic. You can use llvm.memset on any integer
6933 bit width and for different address spaces. However, not all targets
6934 support all bit widths.
6938 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6939 i32 <len>, i32 <align>, i1 <isvolatile>)
6940 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6941 i64 <len>, i32 <align>, i1 <isvolatile>)
6946 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6947 particular byte value.
6949 Note that, unlike the standard libc function, the ``llvm.memset``
6950 intrinsic does not return a value and takes extra alignment/volatile
6951 arguments. Also, the destination can be in an arbitrary address space.
6956 The first argument is a pointer to the destination to fill, the second
6957 is the byte value with which to fill it, the third argument is an
6958 integer argument specifying the number of bytes to fill, and the fourth
6959 argument is the known alignment of the destination location.
6961 If the call to this intrinsic has an alignment value that is not 0 or 1,
6962 then the caller guarantees that the destination pointer is aligned to
6965 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6966 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6967 very cleanly specified and it is unwise to depend on it.
6972 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6973 at the destination location. If the argument is known to be aligned to
6974 some boundary, this can be specified as the fourth argument, otherwise
6975 it should be set to 0 or 1.
6977 '``llvm.sqrt.*``' Intrinsic
6978 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6983 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6984 floating point or vector of floating point type. Not all targets support
6989 declare float @llvm.sqrt.f32(float %Val)
6990 declare double @llvm.sqrt.f64(double %Val)
6991 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6992 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6993 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6998 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6999 returning the same value as the libm '``sqrt``' functions would. Unlike
7000 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7001 negative numbers other than -0.0 (which allows for better optimization,
7002 because there is no need to worry about errno being set).
7003 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7008 The argument and return value are floating point numbers of the same
7014 This function returns the sqrt of the specified operand if it is a
7015 nonnegative floating point number.
7017 '``llvm.powi.*``' Intrinsic
7018 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7023 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7024 floating point or vector of floating point type. Not all targets support
7029 declare float @llvm.powi.f32(float %Val, i32 %power)
7030 declare double @llvm.powi.f64(double %Val, i32 %power)
7031 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7032 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7033 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7038 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7039 specified (positive or negative) power. The order of evaluation of
7040 multiplications is not defined. When a vector of floating point type is
7041 used, the second argument remains a scalar integer value.
7046 The second argument is an integer power, and the first is a value to
7047 raise to that power.
7052 This function returns the first value raised to the second power with an
7053 unspecified sequence of rounding operations.
7055 '``llvm.sin.*``' Intrinsic
7056 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7061 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7062 floating point or vector of floating point type. Not all targets support
7067 declare float @llvm.sin.f32(float %Val)
7068 declare double @llvm.sin.f64(double %Val)
7069 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7070 declare fp128 @llvm.sin.f128(fp128 %Val)
7071 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7076 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7081 The argument and return value are floating point numbers of the same
7087 This function returns the sine of the specified operand, returning the
7088 same values as the libm ``sin`` functions would, and handles error
7089 conditions in the same way.
7091 '``llvm.cos.*``' Intrinsic
7092 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7097 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7098 floating point or vector of floating point type. Not all targets support
7103 declare float @llvm.cos.f32(float %Val)
7104 declare double @llvm.cos.f64(double %Val)
7105 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7106 declare fp128 @llvm.cos.f128(fp128 %Val)
7107 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7112 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7117 The argument and return value are floating point numbers of the same
7123 This function returns the cosine of the specified operand, returning the
7124 same values as the libm ``cos`` functions would, and handles error
7125 conditions in the same way.
7127 '``llvm.pow.*``' Intrinsic
7128 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7133 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7134 floating point or vector of floating point type. Not all targets support
7139 declare float @llvm.pow.f32(float %Val, float %Power)
7140 declare double @llvm.pow.f64(double %Val, double %Power)
7141 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7142 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7143 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7148 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7149 specified (positive or negative) power.
7154 The second argument is a floating point power, and the first is a value
7155 to raise to that power.
7160 This function returns the first value raised to the second power,
7161 returning the same values as the libm ``pow`` functions would, and
7162 handles error conditions in the same way.
7164 '``llvm.exp.*``' Intrinsic
7165 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7170 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7171 floating point or vector of floating point type. Not all targets support
7176 declare float @llvm.exp.f32(float %Val)
7177 declare double @llvm.exp.f64(double %Val)
7178 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7179 declare fp128 @llvm.exp.f128(fp128 %Val)
7180 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7185 The '``llvm.exp.*``' intrinsics perform the exp function.
7190 The argument and return value are floating point numbers of the same
7196 This function returns the same values as the libm ``exp`` functions
7197 would, and handles error conditions in the same way.
7199 '``llvm.exp2.*``' Intrinsic
7200 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7205 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7206 floating point or vector of floating point type. Not all targets support
7211 declare float @llvm.exp2.f32(float %Val)
7212 declare double @llvm.exp2.f64(double %Val)
7213 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7214 declare fp128 @llvm.exp2.f128(fp128 %Val)
7215 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7220 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7225 The argument and return value are floating point numbers of the same
7231 This function returns the same values as the libm ``exp2`` functions
7232 would, and handles error conditions in the same way.
7234 '``llvm.log.*``' Intrinsic
7235 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7240 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7241 floating point or vector of floating point type. Not all targets support
7246 declare float @llvm.log.f32(float %Val)
7247 declare double @llvm.log.f64(double %Val)
7248 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7249 declare fp128 @llvm.log.f128(fp128 %Val)
7250 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7255 The '``llvm.log.*``' intrinsics perform the log function.
7260 The argument and return value are floating point numbers of the same
7266 This function returns the same values as the libm ``log`` functions
7267 would, and handles error conditions in the same way.
7269 '``llvm.log10.*``' Intrinsic
7270 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7275 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7276 floating point or vector of floating point type. Not all targets support
7281 declare float @llvm.log10.f32(float %Val)
7282 declare double @llvm.log10.f64(double %Val)
7283 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7284 declare fp128 @llvm.log10.f128(fp128 %Val)
7285 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7290 The '``llvm.log10.*``' intrinsics perform the log10 function.
7295 The argument and return value are floating point numbers of the same
7301 This function returns the same values as the libm ``log10`` functions
7302 would, and handles error conditions in the same way.
7304 '``llvm.log2.*``' Intrinsic
7305 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7310 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7311 floating point or vector of floating point type. Not all targets support
7316 declare float @llvm.log2.f32(float %Val)
7317 declare double @llvm.log2.f64(double %Val)
7318 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7319 declare fp128 @llvm.log2.f128(fp128 %Val)
7320 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7325 The '``llvm.log2.*``' intrinsics perform the log2 function.
7330 The argument and return value are floating point numbers of the same
7336 This function returns the same values as the libm ``log2`` functions
7337 would, and handles error conditions in the same way.
7339 '``llvm.fma.*``' Intrinsic
7340 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7345 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7346 floating point or vector of floating point type. Not all targets support
7351 declare float @llvm.fma.f32(float %a, float %b, float %c)
7352 declare double @llvm.fma.f64(double %a, double %b, double %c)
7353 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7354 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7355 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7360 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7366 The argument and return value are floating point numbers of the same
7372 This function returns the same values as the libm ``fma`` functions
7375 '``llvm.fabs.*``' Intrinsic
7376 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7381 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7382 floating point or vector of floating point type. Not all targets support
7387 declare float @llvm.fabs.f32(float %Val)
7388 declare double @llvm.fabs.f64(double %Val)
7389 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7390 declare fp128 @llvm.fabs.f128(fp128 %Val)
7391 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7396 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7402 The argument and return value are floating point numbers of the same
7408 This function returns the same values as the libm ``fabs`` functions
7409 would, and handles error conditions in the same way.
7411 '``llvm.copysign.*``' Intrinsic
7412 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7417 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7418 floating point or vector of floating point type. Not all targets support
7423 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7424 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7425 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7426 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7427 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7432 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7433 first operand and the sign of the second operand.
7438 The arguments and return value are floating point numbers of the same
7444 This function returns the same values as the libm ``copysign``
7445 functions would, and handles error conditions in the same way.
7447 '``llvm.floor.*``' Intrinsic
7448 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7453 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7454 floating point or vector of floating point type. Not all targets support
7459 declare float @llvm.floor.f32(float %Val)
7460 declare double @llvm.floor.f64(double %Val)
7461 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7462 declare fp128 @llvm.floor.f128(fp128 %Val)
7463 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7468 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7473 The argument and return value are floating point numbers of the same
7479 This function returns the same values as the libm ``floor`` functions
7480 would, and handles error conditions in the same way.
7482 '``llvm.ceil.*``' Intrinsic
7483 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7488 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7489 floating point or vector of floating point type. Not all targets support
7494 declare float @llvm.ceil.f32(float %Val)
7495 declare double @llvm.ceil.f64(double %Val)
7496 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7497 declare fp128 @llvm.ceil.f128(fp128 %Val)
7498 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7503 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7508 The argument and return value are floating point numbers of the same
7514 This function returns the same values as the libm ``ceil`` functions
7515 would, and handles error conditions in the same way.
7517 '``llvm.trunc.*``' Intrinsic
7518 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7523 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7524 floating point or vector of floating point type. Not all targets support
7529 declare float @llvm.trunc.f32(float %Val)
7530 declare double @llvm.trunc.f64(double %Val)
7531 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7532 declare fp128 @llvm.trunc.f128(fp128 %Val)
7533 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7538 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7539 nearest integer not larger in magnitude than the operand.
7544 The argument and return value are floating point numbers of the same
7550 This function returns the same values as the libm ``trunc`` functions
7551 would, and handles error conditions in the same way.
7553 '``llvm.rint.*``' Intrinsic
7554 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7559 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7560 floating point or vector of floating point type. Not all targets support
7565 declare float @llvm.rint.f32(float %Val)
7566 declare double @llvm.rint.f64(double %Val)
7567 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7568 declare fp128 @llvm.rint.f128(fp128 %Val)
7569 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7574 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7575 nearest integer. It may raise an inexact floating-point exception if the
7576 operand isn't an integer.
7581 The argument and return value are floating point numbers of the same
7587 This function returns the same values as the libm ``rint`` functions
7588 would, and handles error conditions in the same way.
7590 '``llvm.nearbyint.*``' Intrinsic
7591 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7596 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7597 floating point or vector of floating point type. Not all targets support
7602 declare float @llvm.nearbyint.f32(float %Val)
7603 declare double @llvm.nearbyint.f64(double %Val)
7604 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7605 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7606 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7611 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7617 The argument and return value are floating point numbers of the same
7623 This function returns the same values as the libm ``nearbyint``
7624 functions would, and handles error conditions in the same way.
7626 '``llvm.round.*``' Intrinsic
7627 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7632 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7633 floating point or vector of floating point type. Not all targets support
7638 declare float @llvm.round.f32(float %Val)
7639 declare double @llvm.round.f64(double %Val)
7640 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7641 declare fp128 @llvm.round.f128(fp128 %Val)
7642 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7647 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7653 The argument and return value are floating point numbers of the same
7659 This function returns the same values as the libm ``round``
7660 functions would, and handles error conditions in the same way.
7662 Bit Manipulation Intrinsics
7663 ---------------------------
7665 LLVM provides intrinsics for a few important bit manipulation
7666 operations. These allow efficient code generation for some algorithms.
7668 '``llvm.bswap.*``' Intrinsics
7669 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7674 This is an overloaded intrinsic function. You can use bswap on any
7675 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7679 declare i16 @llvm.bswap.i16(i16 <id>)
7680 declare i32 @llvm.bswap.i32(i32 <id>)
7681 declare i64 @llvm.bswap.i64(i64 <id>)
7686 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7687 values with an even number of bytes (positive multiple of 16 bits).
7688 These are useful for performing operations on data that is not in the
7689 target's native byte order.
7694 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7695 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7696 intrinsic returns an i32 value that has the four bytes of the input i32
7697 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7698 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7699 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7700 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7703 '``llvm.ctpop.*``' Intrinsic
7704 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7709 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7710 bit width, or on any vector with integer elements. Not all targets
7711 support all bit widths or vector types, however.
7715 declare i8 @llvm.ctpop.i8(i8 <src>)
7716 declare i16 @llvm.ctpop.i16(i16 <src>)
7717 declare i32 @llvm.ctpop.i32(i32 <src>)
7718 declare i64 @llvm.ctpop.i64(i64 <src>)
7719 declare i256 @llvm.ctpop.i256(i256 <src>)
7720 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7725 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7731 The only argument is the value to be counted. The argument may be of any
7732 integer type, or a vector with integer elements. The return type must
7733 match the argument type.
7738 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7739 each element of a vector.
7741 '``llvm.ctlz.*``' Intrinsic
7742 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7747 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7748 integer bit width, or any vector whose elements are integers. Not all
7749 targets support all bit widths or vector types, however.
7753 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7754 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7755 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7756 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7757 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7758 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7763 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7764 leading zeros in a variable.
7769 The first argument is the value to be counted. This argument may be of
7770 any integer type, or a vectory with integer element type. The return
7771 type must match the first argument type.
7773 The second argument must be a constant and is a flag to indicate whether
7774 the intrinsic should ensure that a zero as the first argument produces a
7775 defined result. Historically some architectures did not provide a
7776 defined result for zero values as efficiently, and many algorithms are
7777 now predicated on avoiding zero-value inputs.
7782 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7783 zeros in a variable, or within each element of the vector. If
7784 ``src == 0`` then the result is the size in bits of the type of ``src``
7785 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7786 ``llvm.ctlz(i32 2) = 30``.
7788 '``llvm.cttz.*``' Intrinsic
7789 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7794 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7795 integer bit width, or any vector of integer elements. Not all targets
7796 support all bit widths or vector types, however.
7800 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7801 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7802 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7803 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7804 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7805 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7810 The '``llvm.cttz``' family of intrinsic functions counts the number of
7816 The first argument is the value to be counted. This argument may be of
7817 any integer type, or a vectory with integer element type. The return
7818 type must match the first argument type.
7820 The second argument must be a constant and is a flag to indicate whether
7821 the intrinsic should ensure that a zero as the first argument produces a
7822 defined result. Historically some architectures did not provide a
7823 defined result for zero values as efficiently, and many algorithms are
7824 now predicated on avoiding zero-value inputs.
7829 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7830 zeros in a variable, or within each element of a vector. If ``src == 0``
7831 then the result is the size in bits of the type of ``src`` if
7832 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7833 ``llvm.cttz(2) = 1``.
7835 Arithmetic with Overflow Intrinsics
7836 -----------------------------------
7838 LLVM provides intrinsics for some arithmetic with overflow operations.
7840 '``llvm.sadd.with.overflow.*``' Intrinsics
7841 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7846 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7847 on any integer bit width.
7851 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7852 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7853 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7858 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7859 a signed addition of the two arguments, and indicate whether an overflow
7860 occurred during the signed summation.
7865 The arguments (%a and %b) and the first element of the result structure
7866 may be of integer types of any bit width, but they must have the same
7867 bit width. The second element of the result structure must be of type
7868 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7874 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7875 a signed addition of the two variables. They return a structure --- the
7876 first element of which is the signed summation, and the second element
7877 of which is a bit specifying if the signed summation resulted in an
7883 .. code-block:: llvm
7885 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7886 %sum = extractvalue {i32, i1} %res, 0
7887 %obit = extractvalue {i32, i1} %res, 1
7888 br i1 %obit, label %overflow, label %normal
7890 '``llvm.uadd.with.overflow.*``' Intrinsics
7891 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7896 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7897 on any integer bit width.
7901 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7902 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7903 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7908 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7909 an unsigned addition of the two arguments, and indicate whether a carry
7910 occurred during the unsigned summation.
7915 The arguments (%a and %b) and the first element of the result structure
7916 may be of integer types of any bit width, but they must have the same
7917 bit width. The second element of the result structure must be of type
7918 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7924 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7925 an unsigned addition of the two arguments. They return a structure --- the
7926 first element of which is the sum, and the second element of which is a
7927 bit specifying if the unsigned summation resulted in a carry.
7932 .. code-block:: llvm
7934 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7935 %sum = extractvalue {i32, i1} %res, 0
7936 %obit = extractvalue {i32, i1} %res, 1
7937 br i1 %obit, label %carry, label %normal
7939 '``llvm.ssub.with.overflow.*``' Intrinsics
7940 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7945 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7946 on any integer bit width.
7950 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7951 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7952 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7957 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7958 a signed subtraction of the two arguments, and indicate whether an
7959 overflow occurred during the signed subtraction.
7964 The arguments (%a and %b) and the first element of the result structure
7965 may be of integer types of any bit width, but they must have the same
7966 bit width. The second element of the result structure must be of type
7967 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7973 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7974 a signed subtraction of the two arguments. They return a structure --- the
7975 first element of which is the subtraction, and the second element of
7976 which is a bit specifying if the signed subtraction resulted in an
7982 .. code-block:: llvm
7984 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7985 %sum = extractvalue {i32, i1} %res, 0
7986 %obit = extractvalue {i32, i1} %res, 1
7987 br i1 %obit, label %overflow, label %normal
7989 '``llvm.usub.with.overflow.*``' Intrinsics
7990 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7995 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7996 on any integer bit width.
8000 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8001 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8002 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8007 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8008 an unsigned subtraction of the two arguments, and indicate whether an
8009 overflow occurred during the unsigned subtraction.
8014 The arguments (%a and %b) and the first element of the result structure
8015 may be of integer types of any bit width, but they must have the same
8016 bit width. The second element of the result structure must be of type
8017 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8023 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8024 an unsigned subtraction of the two arguments. They return a structure ---
8025 the first element of which is the subtraction, and the second element of
8026 which is a bit specifying if the unsigned subtraction resulted in an
8032 .. code-block:: llvm
8034 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8035 %sum = extractvalue {i32, i1} %res, 0
8036 %obit = extractvalue {i32, i1} %res, 1
8037 br i1 %obit, label %overflow, label %normal
8039 '``llvm.smul.with.overflow.*``' Intrinsics
8040 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8045 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8046 on any integer bit width.
8050 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8051 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8052 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8057 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8058 a signed multiplication of the two arguments, and indicate whether an
8059 overflow occurred during the signed multiplication.
8064 The arguments (%a and %b) and the first element of the result structure
8065 may be of integer types of any bit width, but they must have the same
8066 bit width. The second element of the result structure must be of type
8067 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8073 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8074 a signed multiplication of the two arguments. They return a structure ---
8075 the first element of which is the multiplication, and the second element
8076 of which is a bit specifying if the signed multiplication resulted in an
8082 .. code-block:: llvm
8084 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8085 %sum = extractvalue {i32, i1} %res, 0
8086 %obit = extractvalue {i32, i1} %res, 1
8087 br i1 %obit, label %overflow, label %normal
8089 '``llvm.umul.with.overflow.*``' Intrinsics
8090 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8095 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8096 on any integer bit width.
8100 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8101 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8102 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8107 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8108 a unsigned multiplication of the two arguments, and indicate whether an
8109 overflow occurred during the unsigned multiplication.
8114 The arguments (%a and %b) and the first element of the result structure
8115 may be of integer types of any bit width, but they must have the same
8116 bit width. The second element of the result structure must be of type
8117 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8123 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8124 an unsigned multiplication of the two arguments. They return a structure ---
8125 the first element of which is the multiplication, and the second
8126 element of which is a bit specifying if the unsigned multiplication
8127 resulted in an overflow.
8132 .. code-block:: llvm
8134 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8135 %sum = extractvalue {i32, i1} %res, 0
8136 %obit = extractvalue {i32, i1} %res, 1
8137 br i1 %obit, label %overflow, label %normal
8139 Specialised Arithmetic Intrinsics
8140 ---------------------------------
8142 '``llvm.fmuladd.*``' Intrinsic
8143 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8150 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8151 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8156 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8157 expressions that can be fused if the code generator determines that (a) the
8158 target instruction set has support for a fused operation, and (b) that the
8159 fused operation is more efficient than the equivalent, separate pair of mul
8160 and add instructions.
8165 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8166 multiplicands, a and b, and an addend c.
8175 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8177 is equivalent to the expression a \* b + c, except that rounding will
8178 not be performed between the multiplication and addition steps if the
8179 code generator fuses the operations. Fusion is not guaranteed, even if
8180 the target platform supports it. If a fused multiply-add is required the
8181 corresponding llvm.fma.\* intrinsic function should be used instead.
8186 .. code-block:: llvm
8188 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8190 Half Precision Floating Point Intrinsics
8191 ----------------------------------------
8193 For most target platforms, half precision floating point is a
8194 storage-only format. This means that it is a dense encoding (in memory)
8195 but does not support computation in the format.
8197 This means that code must first load the half-precision floating point
8198 value as an i16, then convert it to float with
8199 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8200 then be performed on the float value (including extending to double
8201 etc). To store the value back to memory, it is first converted to float
8202 if needed, then converted to i16 with
8203 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8206 .. _int_convert_to_fp16:
8208 '``llvm.convert.to.fp16``' Intrinsic
8209 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8216 declare i16 @llvm.convert.to.fp16(f32 %a)
8221 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8222 from single precision floating point format to half precision floating
8228 The intrinsic function contains single argument - the value to be
8234 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8235 from single precision floating point format to half precision floating
8236 point format. The return value is an ``i16`` which contains the
8242 .. code-block:: llvm
8244 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8245 store i16 %res, i16* @x, align 2
8247 .. _int_convert_from_fp16:
8249 '``llvm.convert.from.fp16``' Intrinsic
8250 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8257 declare f32 @llvm.convert.from.fp16(i16 %a)
8262 The '``llvm.convert.from.fp16``' intrinsic function performs a
8263 conversion from half precision floating point format to single precision
8264 floating point format.
8269 The intrinsic function contains single argument - the value to be
8275 The '``llvm.convert.from.fp16``' intrinsic function performs a
8276 conversion from half single precision floating point format to single
8277 precision floating point format. The input half-float value is
8278 represented by an ``i16`` value.
8283 .. code-block:: llvm
8285 %a = load i16* @x, align 2
8286 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8291 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8292 prefix), are described in the `LLVM Source Level
8293 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8296 Exception Handling Intrinsics
8297 -----------------------------
8299 The LLVM exception handling intrinsics (which all start with
8300 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8301 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8305 Trampoline Intrinsics
8306 ---------------------
8308 These intrinsics make it possible to excise one parameter, marked with
8309 the :ref:`nest <nest>` attribute, from a function. The result is a
8310 callable function pointer lacking the nest parameter - the caller does
8311 not need to provide a value for it. Instead, the value to use is stored
8312 in advance in a "trampoline", a block of memory usually allocated on the
8313 stack, which also contains code to splice the nest value into the
8314 argument list. This is used to implement the GCC nested function address
8317 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8318 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8319 It can be created as follows:
8321 .. code-block:: llvm
8323 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8324 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8325 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8326 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8327 %fp = bitcast i8* %p to i32 (i32, i32)*
8329 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8330 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8334 '``llvm.init.trampoline``' Intrinsic
8335 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8342 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8347 This fills the memory pointed to by ``tramp`` with executable code,
8348 turning it into a trampoline.
8353 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8354 pointers. The ``tramp`` argument must point to a sufficiently large and
8355 sufficiently aligned block of memory; this memory is written to by the
8356 intrinsic. Note that the size and the alignment are target-specific -
8357 LLVM currently provides no portable way of determining them, so a
8358 front-end that generates this intrinsic needs to have some
8359 target-specific knowledge. The ``func`` argument must hold a function
8360 bitcast to an ``i8*``.
8365 The block of memory pointed to by ``tramp`` is filled with target
8366 dependent code, turning it into a function. Then ``tramp`` needs to be
8367 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8368 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8369 function's signature is the same as that of ``func`` with any arguments
8370 marked with the ``nest`` attribute removed. At most one such ``nest``
8371 argument is allowed, and it must be of pointer type. Calling the new
8372 function is equivalent to calling ``func`` with the same argument list,
8373 but with ``nval`` used for the missing ``nest`` argument. If, after
8374 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8375 modified, then the effect of any later call to the returned function
8376 pointer is undefined.
8380 '``llvm.adjust.trampoline``' Intrinsic
8381 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8388 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8393 This performs any required machine-specific adjustment to the address of
8394 a trampoline (passed as ``tramp``).
8399 ``tramp`` must point to a block of memory which already has trampoline
8400 code filled in by a previous call to
8401 :ref:`llvm.init.trampoline <int_it>`.
8406 On some architectures the address of the code to be executed needs to be
8407 different to the address where the trampoline is actually stored. This
8408 intrinsic returns the executable address corresponding to ``tramp``
8409 after performing the required machine specific adjustments. The pointer
8410 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8415 This class of intrinsics exists to information about the lifetime of
8416 memory objects and ranges where variables are immutable.
8418 '``llvm.lifetime.start``' Intrinsic
8419 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8426 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8431 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8437 The first argument is a constant integer representing the size of the
8438 object, or -1 if it is variable sized. The second argument is a pointer
8444 This intrinsic indicates that before this point in the code, the value
8445 of the memory pointed to by ``ptr`` is dead. This means that it is known
8446 to never be used and has an undefined value. A load from the pointer
8447 that precedes this intrinsic can be replaced with ``'undef'``.
8449 '``llvm.lifetime.end``' Intrinsic
8450 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8457 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8462 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8468 The first argument is a constant integer representing the size of the
8469 object, or -1 if it is variable sized. The second argument is a pointer
8475 This intrinsic indicates that after this point in the code, the value of
8476 the memory pointed to by ``ptr`` is dead. This means that it is known to
8477 never be used and has an undefined value. Any stores into the memory
8478 object following this intrinsic may be removed as dead.
8480 '``llvm.invariant.start``' Intrinsic
8481 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8488 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8493 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8494 a memory object will not change.
8499 The first argument is a constant integer representing the size of the
8500 object, or -1 if it is variable sized. The second argument is a pointer
8506 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8507 the return value, the referenced memory location is constant and
8510 '``llvm.invariant.end``' Intrinsic
8511 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8518 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8523 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8524 memory object are mutable.
8529 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8530 The second argument is a constant integer representing the size of the
8531 object, or -1 if it is variable sized and the third argument is a
8532 pointer to the object.
8537 This intrinsic indicates that the memory is mutable again.
8542 This class of intrinsics is designed to be generic and has no specific
8545 '``llvm.var.annotation``' Intrinsic
8546 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8553 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8558 The '``llvm.var.annotation``' intrinsic.
8563 The first argument is a pointer to a value, the second is a pointer to a
8564 global string, the third is a pointer to a global string which is the
8565 source file name, and the last argument is the line number.
8570 This intrinsic allows annotation of local variables with arbitrary
8571 strings. This can be useful for special purpose optimizations that want
8572 to look for these annotations. These have no other defined use; they are
8573 ignored by code generation and optimization.
8575 '``llvm.ptr.annotation.*``' Intrinsic
8576 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8581 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8582 pointer to an integer of any width. *NOTE* you must specify an address space for
8583 the pointer. The identifier for the default address space is the integer
8588 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8589 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8590 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8591 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8592 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8597 The '``llvm.ptr.annotation``' intrinsic.
8602 The first argument is a pointer to an integer value of arbitrary bitwidth
8603 (result of some expression), the second is a pointer to a global string, the
8604 third is a pointer to a global string which is the source file name, and the
8605 last argument is the line number. It returns the value of the first argument.
8610 This intrinsic allows annotation of a pointer to an integer with arbitrary
8611 strings. This can be useful for special purpose optimizations that want to look
8612 for these annotations. These have no other defined use; they are ignored by code
8613 generation and optimization.
8615 '``llvm.annotation.*``' Intrinsic
8616 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8621 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8622 any integer bit width.
8626 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8627 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8628 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8629 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8630 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8635 The '``llvm.annotation``' intrinsic.
8640 The first argument is an integer value (result of some expression), the
8641 second is a pointer to a global string, the third is a pointer to a
8642 global string which is the source file name, and the last argument is
8643 the line number. It returns the value of the first argument.
8648 This intrinsic allows annotations to be put on arbitrary expressions
8649 with arbitrary strings. This can be useful for special purpose
8650 optimizations that want to look for these annotations. These have no
8651 other defined use; they are ignored by code generation and optimization.
8653 '``llvm.trap``' Intrinsic
8654 ^^^^^^^^^^^^^^^^^^^^^^^^^
8661 declare void @llvm.trap() noreturn nounwind
8666 The '``llvm.trap``' intrinsic.
8676 This intrinsic is lowered to the target dependent trap instruction. If
8677 the target does not have a trap instruction, this intrinsic will be
8678 lowered to a call of the ``abort()`` function.
8680 '``llvm.debugtrap``' Intrinsic
8681 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8688 declare void @llvm.debugtrap() nounwind
8693 The '``llvm.debugtrap``' intrinsic.
8703 This intrinsic is lowered to code which is intended to cause an
8704 execution trap with the intention of requesting the attention of a
8707 '``llvm.stackprotector``' Intrinsic
8708 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8715 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8720 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8721 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8722 is placed on the stack before local variables.
8727 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8728 The first argument is the value loaded from the stack guard
8729 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8730 enough space to hold the value of the guard.
8735 This intrinsic causes the prologue/epilogue inserter to force the position of
8736 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8737 to ensure that if a local variable on the stack is overwritten, it will destroy
8738 the value of the guard. When the function exits, the guard on the stack is
8739 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8740 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8741 calling the ``__stack_chk_fail()`` function.
8743 '``llvm.stackprotectorcheck``' Intrinsic
8744 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8751 declare void @llvm.stackprotectorcheck(i8** <guard>)
8756 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8757 created stack protector and if they are not equal calls the
8758 ``__stack_chk_fail()`` function.
8763 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8764 the variable ``@__stack_chk_guard``.
8769 This intrinsic is provided to perform the stack protector check by comparing
8770 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8771 values do not match call the ``__stack_chk_fail()`` function.
8773 The reason to provide this as an IR level intrinsic instead of implementing it
8774 via other IR operations is that in order to perform this operation at the IR
8775 level without an intrinsic, one would need to create additional basic blocks to
8776 handle the success/failure cases. This makes it difficult to stop the stack
8777 protector check from disrupting sibling tail calls in Codegen. With this
8778 intrinsic, we are able to generate the stack protector basic blocks late in
8779 codegen after the tail call decision has occured.
8781 '``llvm.objectsize``' Intrinsic
8782 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8789 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8790 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8795 The ``llvm.objectsize`` intrinsic is designed to provide information to
8796 the optimizers to determine at compile time whether a) an operation
8797 (like memcpy) will overflow a buffer that corresponds to an object, or
8798 b) that a runtime check for overflow isn't necessary. An object in this
8799 context means an allocation of a specific class, structure, array, or
8805 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8806 argument is a pointer to or into the ``object``. The second argument is
8807 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8808 or -1 (if false) when the object size is unknown. The second argument
8809 only accepts constants.
8814 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8815 the size of the object concerned. If the size cannot be determined at
8816 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8817 on the ``min`` argument).
8819 '``llvm.expect``' Intrinsic
8820 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8827 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8828 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8833 The ``llvm.expect`` intrinsic provides information about expected (the
8834 most probable) value of ``val``, which can be used by optimizers.
8839 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8840 a value. The second argument is an expected value, this needs to be a
8841 constant value, variables are not allowed.
8846 This intrinsic is lowered to the ``val``.
8848 '``llvm.donothing``' Intrinsic
8849 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8856 declare void @llvm.donothing() nounwind readnone
8861 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8862 only intrinsic that can be called with an invoke instruction.
8872 This intrinsic does nothing, and it's removed by optimizers and ignored