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 The linkgage must be one of ``private``, ``linker_private``,
618 ``linker_private_weak``, ``internal``, ``linkonce``, ``weak``,
619 ``linkonce_odr``, ``weak_odr``, ``linkonce_odr_auto_hide``, ``external``. Note
620 that some system linkers might not correctly handle dropping a weak symbol that
621 is aliased by a non weak alias.
623 .. _namedmetadatastructure:
628 Named metadata is a collection of metadata. :ref:`Metadata
629 nodes <metadata>` (but not metadata strings) are the only valid
630 operands for a named metadata.
634 ; Some unnamed metadata nodes, which are referenced by the named metadata.
635 !0 = metadata !{metadata !"zero"}
636 !1 = metadata !{metadata !"one"}
637 !2 = metadata !{metadata !"two"}
639 !name = !{!0, !1, !2}
646 The return type and each parameter of a function type may have a set of
647 *parameter attributes* associated with them. Parameter attributes are
648 used to communicate additional information about the result or
649 parameters of a function. Parameter attributes are considered to be part
650 of the function, not of the function type, so functions with different
651 parameter attributes can have the same function type.
653 Parameter attributes are simple keywords that follow the type specified.
654 If multiple parameter attributes are needed, they are space separated.
659 declare i32 @printf(i8* noalias nocapture, ...)
660 declare i32 @atoi(i8 zeroext)
661 declare signext i8 @returns_signed_char()
663 Note that any attributes for the function result (``nounwind``,
664 ``readonly``) come immediately after the argument list.
666 Currently, only the following parameter attributes are defined:
669 This indicates to the code generator that the parameter or return
670 value should be zero-extended to the extent required by the target's
671 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
672 the caller (for a parameter) or the callee (for a return value).
674 This indicates to the code generator that the parameter or return
675 value should be sign-extended to the extent required by the target's
676 ABI (which is usually 32-bits) by the caller (for a parameter) or
677 the callee (for a return value).
679 This indicates that this parameter or return value should be treated
680 in a special target-dependent fashion during while emitting code for
681 a function call or return (usually, by putting it in a register as
682 opposed to memory, though some targets use it to distinguish between
683 two different kinds of registers). Use of this attribute is
686 This indicates that the pointer parameter should really be passed by
687 value to the function. The attribute implies that a hidden copy of
688 the pointee is made between the caller and the callee, so the callee
689 is unable to modify the value in the caller. This attribute is only
690 valid on LLVM pointer arguments. It is generally used to pass
691 structs and arrays by value, but is also valid on pointers to
692 scalars. The copy is considered to belong to the caller not the
693 callee (for example, ``readonly`` functions should not write to
694 ``byval`` parameters). This is not a valid attribute for return
697 The byval attribute also supports specifying an alignment with the
698 align attribute. It indicates the alignment of the stack slot to
699 form and the known alignment of the pointer specified to the call
700 site. If the alignment is not specified, then the code generator
701 makes a target-specific assumption.
704 This indicates that the pointer parameter specifies the address of a
705 structure that is the return value of the function in the source
706 program. This pointer must be guaranteed by the caller to be valid:
707 loads and stores to the structure may be assumed by the callee
708 not to trap and to be properly aligned. This may only be applied to
709 the first parameter. This is not a valid attribute for return
712 This indicates that pointer values :ref:`based <pointeraliasing>` on
713 the argument or return value do not alias pointer values which are
714 not *based* on it, ignoring certain "irrelevant" dependencies. For a
715 call to the parent function, dependencies between memory references
716 from before or after the call and from those during the call are
717 "irrelevant" to the ``noalias`` keyword for the arguments and return
718 value used in that call. The caller shares the responsibility with
719 the callee for ensuring that these requirements are met. For further
720 details, please see the discussion of the NoAlias response in `alias
721 analysis <AliasAnalysis.html#MustMayNo>`_.
723 Note that this definition of ``noalias`` is intentionally similar
724 to the definition of ``restrict`` in C99 for function arguments,
725 though it is slightly weaker.
727 For function return values, C99's ``restrict`` is not meaningful,
728 while LLVM's ``noalias`` is.
730 This indicates that the callee does not make any copies of the
731 pointer that outlive the callee itself. This is not a valid
732 attribute for return values.
737 This indicates that the pointer parameter can be excised using the
738 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
739 attribute for return values and can only be applied to one parameter.
742 This indicates that the function always returns the argument as its return
743 value. This is an optimization hint to the code generator when generating
744 the caller, allowing tail call optimization and omission of register saves
745 and restores in some cases; it is not checked or enforced when generating
746 the callee. The parameter and the function return type must be valid
747 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
748 valid attribute for return values and can only be applied to one parameter.
752 Garbage Collector Names
753 -----------------------
755 Each function may specify a garbage collector name, which is simply a
760 define void @f() gc "name" { ... }
762 The compiler declares the supported values of *name*. Specifying a
763 collector which will cause the compiler to alter its output in order to
764 support the named garbage collection algorithm.
771 Prefix data is data associated with a function which the code generator
772 will emit immediately before the function body. The purpose of this feature
773 is to allow frontends to associate language-specific runtime metadata with
774 specific functions and make it available through the function pointer while
775 still allowing the function pointer to be called. To access the data for a
776 given function, a program may bitcast the function pointer to a pointer to
777 the constant's type. This implies that the IR symbol points to the start
780 To maintain the semantics of ordinary function calls, the prefix data must
781 have a particular format. Specifically, it must begin with a sequence of
782 bytes which decode to a sequence of machine instructions, valid for the
783 module's target, which transfer control to the point immediately succeeding
784 the prefix data, without performing any other visible action. This allows
785 the inliner and other passes to reason about the semantics of the function
786 definition without needing to reason about the prefix data. Obviously this
787 makes the format of the prefix data highly target dependent.
789 Prefix data is laid out as if it were an initializer for a global variable
790 of the prefix data's type. No padding is automatically placed between the
791 prefix data and the function body. If padding is required, it must be part
794 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
795 which encodes the ``nop`` instruction:
799 define void @f() prefix i8 144 { ... }
801 Generally prefix data can be formed by encoding a relative branch instruction
802 which skips the metadata, as in this example of valid prefix data for the
803 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
807 %0 = type <{ i8, i8, i8* }>
809 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
811 A function may have prefix data but no body. This has similar semantics
812 to the ``available_externally`` linkage in that the data may be used by the
813 optimizers but will not be emitted in the object file.
820 Attribute groups are groups of attributes that are referenced by objects within
821 the IR. They are important for keeping ``.ll`` files readable, because a lot of
822 functions will use the same set of attributes. In the degenerative case of a
823 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
824 group will capture the important command line flags used to build that file.
826 An attribute group is a module-level object. To use an attribute group, an
827 object references the attribute group's ID (e.g. ``#37``). An object may refer
828 to more than one attribute group. In that situation, the attributes from the
829 different groups are merged.
831 Here is an example of attribute groups for a function that should always be
832 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
836 ; Target-independent attributes:
837 attributes #0 = { alwaysinline alignstack=4 }
839 ; Target-dependent attributes:
840 attributes #1 = { "no-sse" }
842 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
843 define void @f() #0 #1 { ... }
850 Function attributes are set to communicate additional information about
851 a function. Function attributes are considered to be part of the
852 function, not of the function type, so functions with different function
853 attributes can have the same function type.
855 Function attributes are simple keywords that follow the type specified.
856 If multiple attributes are needed, they are space separated. For
861 define void @f() noinline { ... }
862 define void @f() alwaysinline { ... }
863 define void @f() alwaysinline optsize { ... }
864 define void @f() optsize { ... }
867 This attribute indicates that, when emitting the prologue and
868 epilogue, the backend should forcibly align the stack pointer.
869 Specify the desired alignment, which must be a power of two, in
872 This attribute indicates that the inliner should attempt to inline
873 this function into callers whenever possible, ignoring any active
874 inlining size threshold for this caller.
876 This indicates that the callee function at a call site should be
877 recognized as a built-in function, even though the function's declaration
878 uses the ``nobuiltin`` attribute. This is only valid at call sites for
879 direct calls to functions which are declared with the ``nobuiltin``
882 This attribute indicates that this function is rarely called. When
883 computing edge weights, basic blocks post-dominated by a cold
884 function call are also considered to be cold; and, thus, given low
887 This attribute indicates that the source code contained a hint that
888 inlining this function is desirable (such as the "inline" keyword in
889 C/C++). It is just a hint; it imposes no requirements on the
892 This attribute suggests that optimization passes and code generator
893 passes make choices that keep the code size of this function as small
894 as possible and perform optimizations that may sacrifice runtime
895 performance in order to minimize the size of the generated code.
897 This attribute disables prologue / epilogue emission for the
898 function. This can have very system-specific consequences.
900 This indicates that the callee function at a call site is not recognized as
901 a built-in function. LLVM will retain the original call and not replace it
902 with equivalent code based on the semantics of the built-in function, unless
903 the call site uses the ``builtin`` attribute. This is valid at call sites
904 and on function declarations and definitions.
906 This attribute indicates that calls to the function cannot be
907 duplicated. A call to a ``noduplicate`` function may be moved
908 within its parent function, but may not be duplicated within
911 A function containing a ``noduplicate`` call may still
912 be an inlining candidate, provided that the call is not
913 duplicated by inlining. That implies that the function has
914 internal linkage and only has one call site, so the original
915 call is dead after inlining.
917 This attributes disables implicit floating point instructions.
919 This attribute indicates that the inliner should never inline this
920 function in any situation. This attribute may not be used together
921 with the ``alwaysinline`` attribute.
923 This attribute suppresses lazy symbol binding for the function. This
924 may make calls to the function faster, at the cost of extra program
925 startup time if the function is not called during program startup.
927 This attribute indicates that the code generator should not use a
928 red zone, even if the target-specific ABI normally permits it.
930 This function attribute indicates that the function never returns
931 normally. This produces undefined behavior at runtime if the
932 function ever does dynamically return.
934 This function attribute indicates that the function never returns
935 with an unwind or exceptional control flow. If the function does
936 unwind, its runtime behavior is undefined.
938 This function attribute indicates that the function is not optimized
939 by any optimization or code generator passes with the
940 exception of interprocedural optimization passes.
941 This attribute cannot be used together with the ``alwaysinline``
942 attribute; this attribute is also incompatible
943 with the ``minsize`` attribute and the ``optsize`` attribute.
945 The inliner should never inline this function in any situation.
946 Only functions with the ``alwaysinline`` attribute are valid
947 candidates for inlining inside the body of this function.
949 This attribute suggests that optimization passes and code generator
950 passes make choices that keep the code size of this function low,
951 and otherwise do optimizations specifically to reduce code size as
952 long as they do not significantly impact runtime performance.
954 On a function, this attribute indicates that the function computes its
955 result (or decides to unwind an exception) based strictly on its arguments,
956 without dereferencing any pointer arguments or otherwise accessing
957 any mutable state (e.g. memory, control registers, etc) visible to
958 caller functions. It does not write through any pointer arguments
959 (including ``byval`` arguments) and never changes any state visible
960 to callers. This means that it cannot unwind exceptions by calling
961 the ``C++`` exception throwing methods.
963 On an argument, this attribute indicates that the function does not
964 dereference that pointer argument, even though it may read or write the
965 memory that the pointer points to if accessed through other pointers.
967 On a function, this attribute indicates that the function does not write
968 through any pointer arguments (including ``byval`` arguments) or otherwise
969 modify any state (e.g. memory, control registers, etc) visible to
970 caller functions. It may dereference pointer arguments and read
971 state that may be set in the caller. A readonly function always
972 returns the same value (or unwinds an exception identically) when
973 called with the same set of arguments and global state. It cannot
974 unwind an exception by calling the ``C++`` exception throwing
977 On an argument, this attribute indicates that the function does not write
978 through this pointer argument, even though it may write to the memory that
979 the pointer points to.
981 This attribute indicates that this function can return twice. The C
982 ``setjmp`` is an example of such a function. The compiler disables
983 some optimizations (like tail calls) in the caller of these
986 This attribute indicates that AddressSanitizer checks
987 (dynamic address safety analysis) are enabled for this function.
989 This attribute indicates that MemorySanitizer checks (dynamic detection
990 of accesses to uninitialized memory) are enabled for this function.
992 This attribute indicates that ThreadSanitizer checks
993 (dynamic thread safety analysis) are enabled for this function.
995 This attribute indicates that the function should emit a stack
996 smashing protector. It is in the form of a "canary" --- a random value
997 placed on the stack before the local variables that's checked upon
998 return from the function to see if it has been overwritten. A
999 heuristic is used to determine if a function needs stack protectors
1000 or not. The heuristic used will enable protectors for functions with:
1002 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1003 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1004 - Calls to alloca() with variable sizes or constant sizes greater than
1005 ``ssp-buffer-size``.
1007 If a function that has an ``ssp`` attribute is inlined into a
1008 function that doesn't have an ``ssp`` attribute, then the resulting
1009 function will have an ``ssp`` attribute.
1011 This attribute indicates that the function should *always* emit a
1012 stack smashing protector. This overrides the ``ssp`` function
1015 If a function that has an ``sspreq`` attribute is inlined into a
1016 function that doesn't have an ``sspreq`` attribute or which has an
1017 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1018 an ``sspreq`` attribute.
1020 This attribute indicates that the function should emit a stack smashing
1021 protector. This attribute causes a strong heuristic to be used when
1022 determining if a function needs stack protectors. The strong heuristic
1023 will enable protectors for functions with:
1025 - Arrays of any size and type
1026 - Aggregates containing an array of any size and type.
1027 - Calls to alloca().
1028 - Local variables that have had their address taken.
1030 This overrides the ``ssp`` function attribute.
1032 If a function that has an ``sspstrong`` attribute is inlined into a
1033 function that doesn't have an ``sspstrong`` attribute, then the
1034 resulting function will have an ``sspstrong`` attribute.
1036 This attribute indicates that the ABI being targeted requires that
1037 an unwind table entry be produce for this function even if we can
1038 show that no exceptions passes by it. This is normally the case for
1039 the ELF x86-64 abi, but it can be disabled for some compilation
1044 Module-Level Inline Assembly
1045 ----------------------------
1047 Modules may contain "module-level inline asm" blocks, which corresponds
1048 to the GCC "file scope inline asm" blocks. These blocks are internally
1049 concatenated by LLVM and treated as a single unit, but may be separated
1050 in the ``.ll`` file if desired. The syntax is very simple:
1052 .. code-block:: llvm
1054 module asm "inline asm code goes here"
1055 module asm "more can go here"
1057 The strings can contain any character by escaping non-printable
1058 characters. The escape sequence used is simply "\\xx" where "xx" is the
1059 two digit hex code for the number.
1061 The inline asm code is simply printed to the machine code .s file when
1062 assembly code is generated.
1064 .. _langref_datalayout:
1069 A module may specify a target specific data layout string that specifies
1070 how data is to be laid out in memory. The syntax for the data layout is
1073 .. code-block:: llvm
1075 target datalayout = "layout specification"
1077 The *layout specification* consists of a list of specifications
1078 separated by the minus sign character ('-'). Each specification starts
1079 with a letter and may include other information after the letter to
1080 define some aspect of the data layout. The specifications accepted are
1084 Specifies that the target lays out data in big-endian form. That is,
1085 the bits with the most significance have the lowest address
1088 Specifies that the target lays out data in little-endian form. That
1089 is, the bits with the least significance have the lowest address
1092 Specifies the natural alignment of the stack in bits. Alignment
1093 promotion of stack variables is limited to the natural stack
1094 alignment to avoid dynamic stack realignment. The stack alignment
1095 must be a multiple of 8-bits. If omitted, the natural stack
1096 alignment defaults to "unspecified", which does not prevent any
1097 alignment promotions.
1098 ``p[n]:<size>:<abi>:<pref>``
1099 This specifies the *size* of a pointer and its ``<abi>`` and
1100 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1101 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1102 preceding ``:`` should be omitted too. The address space, ``n`` is
1103 optional, and if not specified, denotes the default address space 0.
1104 The value of ``n`` must be in the range [1,2^23).
1105 ``i<size>:<abi>:<pref>``
1106 This specifies the alignment for an integer type of a given bit
1107 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1108 ``v<size>:<abi>:<pref>``
1109 This specifies the alignment for a vector type of a given bit
1111 ``f<size>:<abi>:<pref>``
1112 This specifies the alignment for a floating point type of a given bit
1113 ``<size>``. Only values of ``<size>`` that are supported by the target
1114 will work. 32 (float) and 64 (double) are supported on all targets; 80
1115 or 128 (different flavors of long double) are also supported on some
1117 ``a<size>:<abi>:<pref>``
1118 This specifies the alignment for an aggregate type of a given bit
1120 ``s<size>:<abi>:<pref>``
1121 This specifies the alignment for a stack object of a given bit
1123 ``n<size1>:<size2>:<size3>...``
1124 This specifies a set of native integer widths for the target CPU in
1125 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1126 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1127 this set are considered to support most general arithmetic operations
1130 When constructing the data layout for a given target, LLVM starts with a
1131 default set of specifications which are then (possibly) overridden by
1132 the specifications in the ``datalayout`` keyword. The default
1133 specifications are given in this list:
1135 - ``E`` - big endian
1136 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1137 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1138 same as the default address space.
1139 - ``S0`` - natural stack alignment is unspecified
1140 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1141 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1142 - ``i16:16:16`` - i16 is 16-bit aligned
1143 - ``i32:32:32`` - i32 is 32-bit aligned
1144 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1145 alignment of 64-bits
1146 - ``f16:16:16`` - half is 16-bit aligned
1147 - ``f32:32:32`` - float is 32-bit aligned
1148 - ``f64:64:64`` - double is 64-bit aligned
1149 - ``f128:128:128`` - quad is 128-bit aligned
1150 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1151 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1152 - ``a0:0:64`` - aggregates are 64-bit aligned
1154 When LLVM is determining the alignment for a given type, it uses the
1157 #. If the type sought is an exact match for one of the specifications,
1158 that specification is used.
1159 #. If no match is found, and the type sought is an integer type, then
1160 the smallest integer type that is larger than the bitwidth of the
1161 sought type is used. If none of the specifications are larger than
1162 the bitwidth then the largest integer type is used. For example,
1163 given the default specifications above, the i7 type will use the
1164 alignment of i8 (next largest) while both i65 and i256 will use the
1165 alignment of i64 (largest specified).
1166 #. If no match is found, and the type sought is a vector type, then the
1167 largest vector type that is smaller than the sought vector type will
1168 be used as a fall back. This happens because <128 x double> can be
1169 implemented in terms of 64 <2 x double>, for example.
1171 The function of the data layout string may not be what you expect.
1172 Notably, this is not a specification from the frontend of what alignment
1173 the code generator should use.
1175 Instead, if specified, the target data layout is required to match what
1176 the ultimate *code generator* expects. This string is used by the
1177 mid-level optimizers to improve code, and this only works if it matches
1178 what the ultimate code generator uses. If you would like to generate IR
1179 that does not embed this target-specific detail into the IR, then you
1180 don't have to specify the string. This will disable some optimizations
1181 that require precise layout information, but this also prevents those
1182 optimizations from introducing target specificity into the IR.
1184 .. _pointeraliasing:
1186 Pointer Aliasing Rules
1187 ----------------------
1189 Any memory access must be done through a pointer value associated with
1190 an address range of the memory access, otherwise the behavior is
1191 undefined. Pointer values are associated with address ranges according
1192 to the following rules:
1194 - A pointer value is associated with the addresses associated with any
1195 value it is *based* on.
1196 - An address of a global variable is associated with the address range
1197 of the variable's storage.
1198 - The result value of an allocation instruction is associated with the
1199 address range of the allocated storage.
1200 - A null pointer in the default address-space is associated with no
1202 - An integer constant other than zero or a pointer value returned from
1203 a function not defined within LLVM may be associated with address
1204 ranges allocated through mechanisms other than those provided by
1205 LLVM. Such ranges shall not overlap with any ranges of addresses
1206 allocated by mechanisms provided by LLVM.
1208 A pointer value is *based* on another pointer value according to the
1211 - A pointer value formed from a ``getelementptr`` operation is *based*
1212 on the first operand of the ``getelementptr``.
1213 - The result value of a ``bitcast`` is *based* on the operand of the
1215 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1216 values that contribute (directly or indirectly) to the computation of
1217 the pointer's value.
1218 - The "*based* on" relationship is transitive.
1220 Note that this definition of *"based"* is intentionally similar to the
1221 definition of *"based"* in C99, though it is slightly weaker.
1223 LLVM IR does not associate types with memory. The result type of a
1224 ``load`` merely indicates the size and alignment of the memory from
1225 which to load, as well as the interpretation of the value. The first
1226 operand type of a ``store`` similarly only indicates the size and
1227 alignment of the store.
1229 Consequently, type-based alias analysis, aka TBAA, aka
1230 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1231 :ref:`Metadata <metadata>` may be used to encode additional information
1232 which specialized optimization passes may use to implement type-based
1237 Volatile Memory Accesses
1238 ------------------------
1240 Certain memory accesses, such as :ref:`load <i_load>`'s,
1241 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1242 marked ``volatile``. The optimizers must not change the number of
1243 volatile operations or change their order of execution relative to other
1244 volatile operations. The optimizers *may* change the order of volatile
1245 operations relative to non-volatile operations. This is not Java's
1246 "volatile" and has no cross-thread synchronization behavior.
1248 IR-level volatile loads and stores cannot safely be optimized into
1249 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1250 flagged volatile. Likewise, the backend should never split or merge
1251 target-legal volatile load/store instructions.
1253 .. admonition:: Rationale
1255 Platforms may rely on volatile loads and stores of natively supported
1256 data width to be executed as single instruction. For example, in C
1257 this holds for an l-value of volatile primitive type with native
1258 hardware support, but not necessarily for aggregate types. The
1259 frontend upholds these expectations, which are intentionally
1260 unspecified in the IR. The rules above ensure that IR transformation
1261 do not violate the frontend's contract with the language.
1265 Memory Model for Concurrent Operations
1266 --------------------------------------
1268 The LLVM IR does not define any way to start parallel threads of
1269 execution or to register signal handlers. Nonetheless, there are
1270 platform-specific ways to create them, and we define LLVM IR's behavior
1271 in their presence. This model is inspired by the C++0x memory model.
1273 For a more informal introduction to this model, see the :doc:`Atomics`.
1275 We define a *happens-before* partial order as the least partial order
1278 - Is a superset of single-thread program order, and
1279 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1280 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1281 techniques, like pthread locks, thread creation, thread joining,
1282 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1283 Constraints <ordering>`).
1285 Note that program order does not introduce *happens-before* edges
1286 between a thread and signals executing inside that thread.
1288 Every (defined) read operation (load instructions, memcpy, atomic
1289 loads/read-modify-writes, etc.) R reads a series of bytes written by
1290 (defined) write operations (store instructions, atomic
1291 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1292 section, initialized globals are considered to have a write of the
1293 initializer which is atomic and happens before any other read or write
1294 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1295 may see any write to the same byte, except:
1297 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1298 write\ :sub:`2` happens before R\ :sub:`byte`, then
1299 R\ :sub:`byte` does not see write\ :sub:`1`.
1300 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1301 R\ :sub:`byte` does not see write\ :sub:`3`.
1303 Given that definition, R\ :sub:`byte` is defined as follows:
1305 - If R is volatile, the result is target-dependent. (Volatile is
1306 supposed to give guarantees which can support ``sig_atomic_t`` in
1307 C/C++, and may be used for accesses to addresses which do not behave
1308 like normal memory. It does not generally provide cross-thread
1310 - Otherwise, if there is no write to the same byte that happens before
1311 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1312 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1313 R\ :sub:`byte` returns the value written by that write.
1314 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1315 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1316 Memory Ordering Constraints <ordering>` section for additional
1317 constraints on how the choice is made.
1318 - Otherwise R\ :sub:`byte` returns ``undef``.
1320 R returns the value composed of the series of bytes it read. This
1321 implies that some bytes within the value may be ``undef`` **without**
1322 the entire value being ``undef``. Note that this only defines the
1323 semantics of the operation; it doesn't mean that targets will emit more
1324 than one instruction to read the series of bytes.
1326 Note that in cases where none of the atomic intrinsics are used, this
1327 model places only one restriction on IR transformations on top of what
1328 is required for single-threaded execution: introducing a store to a byte
1329 which might not otherwise be stored is not allowed in general.
1330 (Specifically, in the case where another thread might write to and read
1331 from an address, introducing a store can change a load that may see
1332 exactly one write into a load that may see multiple writes.)
1336 Atomic Memory Ordering Constraints
1337 ----------------------------------
1339 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1340 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1341 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1342 an ordering parameter that determines which other atomic instructions on
1343 the same address they *synchronize with*. These semantics are borrowed
1344 from Java and C++0x, but are somewhat more colloquial. If these
1345 descriptions aren't precise enough, check those specs (see spec
1346 references in the :doc:`atomics guide <Atomics>`).
1347 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1348 differently since they don't take an address. See that instruction's
1349 documentation for details.
1351 For a simpler introduction to the ordering constraints, see the
1355 The set of values that can be read is governed by the happens-before
1356 partial order. A value cannot be read unless some operation wrote
1357 it. This is intended to provide a guarantee strong enough to model
1358 Java's non-volatile shared variables. This ordering cannot be
1359 specified for read-modify-write operations; it is not strong enough
1360 to make them atomic in any interesting way.
1362 In addition to the guarantees of ``unordered``, there is a single
1363 total order for modifications by ``monotonic`` operations on each
1364 address. All modification orders must be compatible with the
1365 happens-before order. There is no guarantee that the modification
1366 orders can be combined to a global total order for the whole program
1367 (and this often will not be possible). The read in an atomic
1368 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1369 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1370 order immediately before the value it writes. If one atomic read
1371 happens before another atomic read of the same address, the later
1372 read must see the same value or a later value in the address's
1373 modification order. This disallows reordering of ``monotonic`` (or
1374 stronger) operations on the same address. If an address is written
1375 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1376 read that address repeatedly, the other threads must eventually see
1377 the write. This corresponds to the C++0x/C1x
1378 ``memory_order_relaxed``.
1380 In addition to the guarantees of ``monotonic``, a
1381 *synchronizes-with* edge may be formed with a ``release`` operation.
1382 This is intended to model C++'s ``memory_order_acquire``.
1384 In addition to the guarantees of ``monotonic``, if this operation
1385 writes a value which is subsequently read by an ``acquire``
1386 operation, it *synchronizes-with* that operation. (This isn't a
1387 complete description; see the C++0x definition of a release
1388 sequence.) This corresponds to the C++0x/C1x
1389 ``memory_order_release``.
1390 ``acq_rel`` (acquire+release)
1391 Acts as both an ``acquire`` and ``release`` operation on its
1392 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1393 ``seq_cst`` (sequentially consistent)
1394 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1395 operation which only reads, ``release`` for an operation which only
1396 writes), there is a global total order on all
1397 sequentially-consistent operations on all addresses, which is
1398 consistent with the *happens-before* partial order and with the
1399 modification orders of all the affected addresses. Each
1400 sequentially-consistent read sees the last preceding write to the
1401 same address in this global order. This corresponds to the C++0x/C1x
1402 ``memory_order_seq_cst`` and Java volatile.
1406 If an atomic operation is marked ``singlethread``, it only *synchronizes
1407 with* or participates in modification and seq\_cst total orderings with
1408 other operations running in the same thread (for example, in signal
1416 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1417 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1418 :ref:`frem <i_frem>`) have the following flags that can set to enable
1419 otherwise unsafe floating point operations
1422 No NaNs - Allow optimizations to assume the arguments and result are not
1423 NaN. Such optimizations are required to retain defined behavior over
1424 NaNs, but the value of the result is undefined.
1427 No Infs - Allow optimizations to assume the arguments and result are not
1428 +/-Inf. Such optimizations are required to retain defined behavior over
1429 +/-Inf, but the value of the result is undefined.
1432 No Signed Zeros - Allow optimizations to treat the sign of a zero
1433 argument or result as insignificant.
1436 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1437 argument rather than perform division.
1440 Fast - Allow algebraically equivalent transformations that may
1441 dramatically change results in floating point (e.g. reassociate). This
1442 flag implies all the others.
1449 The LLVM type system is one of the most important features of the
1450 intermediate representation. Being typed enables a number of
1451 optimizations to be performed on the intermediate representation
1452 directly, without having to do extra analyses on the side before the
1453 transformation. A strong type system makes it easier to read the
1454 generated code and enables novel analyses and transformations that are
1455 not feasible to perform on normal three address code representations.
1457 .. _typeclassifications:
1459 Type Classifications
1460 --------------------
1462 The types fall into a few useful classifications:
1471 * - :ref:`integer <t_integer>`
1472 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1475 * - :ref:`floating point <t_floating>`
1476 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1484 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1485 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1486 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1487 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1489 * - :ref:`primitive <t_primitive>`
1490 - :ref:`label <t_label>`,
1491 :ref:`void <t_void>`,
1492 :ref:`integer <t_integer>`,
1493 :ref:`floating point <t_floating>`,
1494 :ref:`x86mmx <t_x86mmx>`,
1495 :ref:`metadata <t_metadata>`.
1497 * - :ref:`derived <t_derived>`
1498 - :ref:`array <t_array>`,
1499 :ref:`function <t_function>`,
1500 :ref:`pointer <t_pointer>`,
1501 :ref:`structure <t_struct>`,
1502 :ref:`vector <t_vector>`,
1503 :ref:`opaque <t_opaque>`.
1505 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1506 Values of these types are the only ones which can be produced by
1514 The primitive types are the fundamental building blocks of the LLVM
1525 The integer type is a very simple type that simply specifies an
1526 arbitrary bit width for the integer type desired. Any bit width from 1
1527 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1536 The number of bits the integer will occupy is specified by the ``N``
1542 +----------------+------------------------------------------------+
1543 | ``i1`` | a single-bit integer. |
1544 +----------------+------------------------------------------------+
1545 | ``i32`` | a 32-bit integer. |
1546 +----------------+------------------------------------------------+
1547 | ``i1942652`` | a really big integer of over 1 million bits. |
1548 +----------------+------------------------------------------------+
1552 Floating Point Types
1553 ^^^^^^^^^^^^^^^^^^^^
1562 - 16-bit floating point value
1565 - 32-bit floating point value
1568 - 64-bit floating point value
1571 - 128-bit floating point value (112-bit mantissa)
1574 - 80-bit floating point value (X87)
1577 - 128-bit floating point value (two 64-bits)
1587 The x86mmx type represents a value held in an MMX register on an x86
1588 machine. The operations allowed on it are quite limited: parameters and
1589 return values, load and store, and bitcast. User-specified MMX
1590 instructions are represented as intrinsic or asm calls with arguments
1591 and/or results of this type. There are no arrays, vectors or constants
1609 The void type does not represent any value and has no size.
1626 The label type represents code labels.
1643 The metadata type represents embedded metadata. No derived types may be
1644 created from metadata except for :ref:`function <t_function>` arguments.
1658 The real power in LLVM comes from the derived types in the system. This
1659 is what allows a programmer to represent arrays, functions, pointers,
1660 and other useful types. Each of these types contain one or more element
1661 types which may be a primitive type, or another derived type. For
1662 example, it is possible to have a two dimensional array, using an array
1663 as the element type of another array.
1670 Aggregate Types are a subset of derived types that can contain multiple
1671 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1672 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1683 The array type is a very simple derived type that arranges elements
1684 sequentially in memory. The array type requires a size (number of
1685 elements) and an underlying data type.
1692 [<# elements> x <elementtype>]
1694 The number of elements is a constant integer value; ``elementtype`` may
1695 be any type with a size.
1700 +------------------+--------------------------------------+
1701 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1702 +------------------+--------------------------------------+
1703 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1704 +------------------+--------------------------------------+
1705 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1706 +------------------+--------------------------------------+
1708 Here are some examples of multidimensional arrays:
1710 +-----------------------------+----------------------------------------------------------+
1711 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1712 +-----------------------------+----------------------------------------------------------+
1713 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1714 +-----------------------------+----------------------------------------------------------+
1715 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1716 +-----------------------------+----------------------------------------------------------+
1718 There is no restriction on indexing beyond the end of the array implied
1719 by a static type (though there are restrictions on indexing beyond the
1720 bounds of an allocated object in some cases). This means that
1721 single-dimension 'variable sized array' addressing can be implemented in
1722 LLVM with a zero length array type. An implementation of 'pascal style
1723 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1734 The function type can be thought of as a function signature. It consists
1735 of a return type and a list of formal parameter types. The return type
1736 of a function type is a first class type or a void type.
1743 <returntype> (<parameter list>)
1745 ...where '``<parameter list>``' is a comma-separated list of type
1746 specifiers. Optionally, the parameter list may include a type ``...``,
1747 which indicates that the function takes a variable number of arguments.
1748 Variable argument functions can access their arguments with the
1749 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1750 '``<returntype>``' is any type except :ref:`label <t_label>`.
1755 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1756 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1757 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1758 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1759 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1760 | ``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. |
1761 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1762 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1763 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1773 The structure type is used to represent a collection of data members
1774 together in memory. The elements of a structure may be any type that has
1777 Structures in memory are accessed using '``load``' and '``store``' by
1778 getting a pointer to a field with the '``getelementptr``' instruction.
1779 Structures in registers are accessed using the '``extractvalue``' and
1780 '``insertvalue``' instructions.
1782 Structures may optionally be "packed" structures, which indicate that
1783 the alignment of the struct is one byte, and that there is no padding
1784 between the elements. In non-packed structs, padding between field types
1785 is inserted as defined by the DataLayout string in the module, which is
1786 required to match what the underlying code generator expects.
1788 Structures can either be "literal" or "identified". A literal structure
1789 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1790 identified types are always defined at the top level with a name.
1791 Literal types are uniqued by their contents and can never be recursive
1792 or opaque since there is no way to write one. Identified types can be
1793 recursive, can be opaqued, and are never uniqued.
1800 %T1 = type { <type list> } ; Identified normal struct type
1801 %T2 = type <{ <type list> }> ; Identified packed struct type
1806 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1807 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1808 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1809 | ``{ 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``. |
1810 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1811 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1812 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1816 Opaque Structure Types
1817 ^^^^^^^^^^^^^^^^^^^^^^
1822 Opaque structure types are used to represent named structure types that
1823 do not have a body specified. This corresponds (for example) to the C
1824 notion of a forward declared structure.
1837 +--------------+-------------------+
1838 | ``opaque`` | An opaque type. |
1839 +--------------+-------------------+
1849 The pointer type is used to specify memory locations. Pointers are
1850 commonly used to reference objects in memory.
1852 Pointer types may have an optional address space attribute defining the
1853 numbered address space where the pointed-to object resides. The default
1854 address space is number zero. The semantics of non-zero address spaces
1855 are target-specific.
1857 Note that LLVM does not permit pointers to void (``void*``) nor does it
1858 permit pointers to labels (``label*``). Use ``i8*`` instead.
1870 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1871 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1872 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1873 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1874 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1875 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1876 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1886 A vector type is a simple derived type that represents a vector of
1887 elements. Vector types are used when multiple primitive data are
1888 operated in parallel using a single instruction (SIMD). A vector type
1889 requires a size (number of elements) and an underlying primitive data
1890 type. Vector types are considered :ref:`first class <t_firstclass>`.
1897 < <# elements> x <elementtype> >
1899 The number of elements is a constant integer value larger than 0;
1900 elementtype may be any integer or floating point type, or a pointer to
1901 these types. Vectors of size zero are not allowed.
1906 +-------------------+--------------------------------------------------+
1907 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1908 +-------------------+--------------------------------------------------+
1909 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1910 +-------------------+--------------------------------------------------+
1911 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1912 +-------------------+--------------------------------------------------+
1913 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1914 +-------------------+--------------------------------------------------+
1919 LLVM has several different basic types of constants. This section
1920 describes them all and their syntax.
1925 **Boolean constants**
1926 The two strings '``true``' and '``false``' are both valid constants
1928 **Integer constants**
1929 Standard integers (such as '4') are constants of the
1930 :ref:`integer <t_integer>` type. Negative numbers may be used with
1932 **Floating point constants**
1933 Floating point constants use standard decimal notation (e.g.
1934 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1935 hexadecimal notation (see below). The assembler requires the exact
1936 decimal value of a floating-point constant. For example, the
1937 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1938 decimal in binary. Floating point constants must have a :ref:`floating
1939 point <t_floating>` type.
1940 **Null pointer constants**
1941 The identifier '``null``' is recognized as a null pointer constant
1942 and must be of :ref:`pointer type <t_pointer>`.
1944 The one non-intuitive notation for constants is the hexadecimal form of
1945 floating point constants. For example, the form
1946 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1947 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1948 constants are required (and the only time that they are generated by the
1949 disassembler) is when a floating point constant must be emitted but it
1950 cannot be represented as a decimal floating point number in a reasonable
1951 number of digits. For example, NaN's, infinities, and other special
1952 values are represented in their IEEE hexadecimal format so that assembly
1953 and disassembly do not cause any bits to change in the constants.
1955 When using the hexadecimal form, constants of types half, float, and
1956 double are represented using the 16-digit form shown above (which
1957 matches the IEEE754 representation for double); half and float values
1958 must, however, be exactly representable as IEEE 754 half and single
1959 precision, respectively. Hexadecimal format is always used for long
1960 double, and there are three forms of long double. The 80-bit format used
1961 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1962 128-bit format used by PowerPC (two adjacent doubles) is represented by
1963 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1964 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1965 will only work if they match the long double format on your target.
1966 The IEEE 16-bit format (half precision) is represented by ``0xH``
1967 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1968 (sign bit at the left).
1970 There are no constants of type x86mmx.
1972 .. _complexconstants:
1977 Complex constants are a (potentially recursive) combination of simple
1978 constants and smaller complex constants.
1980 **Structure constants**
1981 Structure constants are represented with notation similar to
1982 structure type definitions (a comma separated list of elements,
1983 surrounded by braces (``{}``)). For example:
1984 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1985 "``@G = external global i32``". Structure constants must have
1986 :ref:`structure type <t_struct>`, and the number and types of elements
1987 must match those specified by the type.
1989 Array constants are represented with notation similar to array type
1990 definitions (a comma separated list of elements, surrounded by
1991 square brackets (``[]``)). For example:
1992 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1993 :ref:`array type <t_array>`, and the number and types of elements must
1994 match those specified by the type.
1995 **Vector constants**
1996 Vector constants are represented with notation similar to vector
1997 type definitions (a comma separated list of elements, surrounded by
1998 less-than/greater-than's (``<>``)). For example:
1999 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2000 must have :ref:`vector type <t_vector>`, and the number and types of
2001 elements must match those specified by the type.
2002 **Zero initialization**
2003 The string '``zeroinitializer``' can be used to zero initialize a
2004 value to zero of *any* type, including scalar and
2005 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2006 having to print large zero initializers (e.g. for large arrays) and
2007 is always exactly equivalent to using explicit zero initializers.
2009 A metadata node is a structure-like constant with :ref:`metadata
2010 type <t_metadata>`. For example:
2011 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2012 constants that are meant to be interpreted as part of the
2013 instruction stream, metadata is a place to attach additional
2014 information such as debug info.
2016 Global Variable and Function Addresses
2017 --------------------------------------
2019 The addresses of :ref:`global variables <globalvars>` and
2020 :ref:`functions <functionstructure>` are always implicitly valid
2021 (link-time) constants. These constants are explicitly referenced when
2022 the :ref:`identifier for the global <identifiers>` is used and always have
2023 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2026 .. code-block:: llvm
2030 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2037 The string '``undef``' can be used anywhere a constant is expected, and
2038 indicates that the user of the value may receive an unspecified
2039 bit-pattern. Undefined values may be of any type (other than '``label``'
2040 or '``void``') and be used anywhere a constant is permitted.
2042 Undefined values are useful because they indicate to the compiler that
2043 the program is well defined no matter what value is used. This gives the
2044 compiler more freedom to optimize. Here are some examples of
2045 (potentially surprising) transformations that are valid (in pseudo IR):
2047 .. code-block:: llvm
2057 This is safe because all of the output bits are affected by the undef
2058 bits. Any output bit can have a zero or one depending on the input bits.
2060 .. code-block:: llvm
2071 These logical operations have bits that are not always affected by the
2072 input. For example, if ``%X`` has a zero bit, then the output of the
2073 '``and``' operation will always be a zero for that bit, no matter what
2074 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2075 optimize or assume that the result of the '``and``' is '``undef``'.
2076 However, it is safe to assume that all bits of the '``undef``' could be
2077 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2078 all the bits of the '``undef``' operand to the '``or``' could be set,
2079 allowing the '``or``' to be folded to -1.
2081 .. code-block:: llvm
2083 %A = select undef, %X, %Y
2084 %B = select undef, 42, %Y
2085 %C = select %X, %Y, undef
2095 This set of examples shows that undefined '``select``' (and conditional
2096 branch) conditions can go *either way*, but they have to come from one
2097 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2098 both known to have a clear low bit, then ``%A`` would have to have a
2099 cleared low bit. However, in the ``%C`` example, the optimizer is
2100 allowed to assume that the '``undef``' operand could be the same as
2101 ``%Y``, allowing the whole '``select``' to be eliminated.
2103 .. code-block:: llvm
2105 %A = xor undef, undef
2122 This example points out that two '``undef``' operands are not
2123 necessarily the same. This can be surprising to people (and also matches
2124 C semantics) where they assume that "``X^X``" is always zero, even if
2125 ``X`` is undefined. This isn't true for a number of reasons, but the
2126 short answer is that an '``undef``' "variable" can arbitrarily change
2127 its value over its "live range". This is true because the variable
2128 doesn't actually *have a live range*. Instead, the value is logically
2129 read from arbitrary registers that happen to be around when needed, so
2130 the value is not necessarily consistent over time. In fact, ``%A`` and
2131 ``%C`` need to have the same semantics or the core LLVM "replace all
2132 uses with" concept would not hold.
2134 .. code-block:: llvm
2142 These examples show the crucial difference between an *undefined value*
2143 and *undefined behavior*. An undefined value (like '``undef``') is
2144 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2145 operation can be constant folded to '``undef``', because the '``undef``'
2146 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2147 However, in the second example, we can make a more aggressive
2148 assumption: because the ``undef`` is allowed to be an arbitrary value,
2149 we are allowed to assume that it could be zero. Since a divide by zero
2150 has *undefined behavior*, we are allowed to assume that the operation
2151 does not execute at all. This allows us to delete the divide and all
2152 code after it. Because the undefined operation "can't happen", the
2153 optimizer can assume that it occurs in dead code.
2155 .. code-block:: llvm
2157 a: store undef -> %X
2158 b: store %X -> undef
2163 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2164 value can be assumed to not have any effect; we can assume that the
2165 value is overwritten with bits that happen to match what was already
2166 there. However, a store *to* an undefined location could clobber
2167 arbitrary memory, therefore, it has undefined behavior.
2174 Poison values are similar to :ref:`undef values <undefvalues>`, however
2175 they also represent the fact that an instruction or constant expression
2176 which cannot evoke side effects has nevertheless detected a condition
2177 which results in undefined behavior.
2179 There is currently no way of representing a poison value in the IR; they
2180 only exist when produced by operations such as :ref:`add <i_add>` with
2183 Poison value behavior is defined in terms of value *dependence*:
2185 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2186 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2187 their dynamic predecessor basic block.
2188 - Function arguments depend on the corresponding actual argument values
2189 in the dynamic callers of their functions.
2190 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2191 instructions that dynamically transfer control back to them.
2192 - :ref:`Invoke <i_invoke>` instructions depend on the
2193 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2194 call instructions that dynamically transfer control back to them.
2195 - Non-volatile loads and stores depend on the most recent stores to all
2196 of the referenced memory addresses, following the order in the IR
2197 (including loads and stores implied by intrinsics such as
2198 :ref:`@llvm.memcpy <int_memcpy>`.)
2199 - An instruction with externally visible side effects depends on the
2200 most recent preceding instruction with externally visible side
2201 effects, following the order in the IR. (This includes :ref:`volatile
2202 operations <volatile>`.)
2203 - An instruction *control-depends* on a :ref:`terminator
2204 instruction <terminators>` if the terminator instruction has
2205 multiple successors and the instruction is always executed when
2206 control transfers to one of the successors, and may not be executed
2207 when control is transferred to another.
2208 - Additionally, an instruction also *control-depends* on a terminator
2209 instruction if the set of instructions it otherwise depends on would
2210 be different if the terminator had transferred control to a different
2212 - Dependence is transitive.
2214 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2215 with the additional affect that any instruction which has a *dependence*
2216 on a poison value has undefined behavior.
2218 Here are some examples:
2220 .. code-block:: llvm
2223 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2224 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2225 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2226 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2228 store i32 %poison, i32* @g ; Poison value stored to memory.
2229 %poison2 = load i32* @g ; Poison value loaded back from memory.
2231 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2233 %narrowaddr = bitcast i32* @g to i16*
2234 %wideaddr = bitcast i32* @g to i64*
2235 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2236 %poison4 = load i64* %wideaddr ; Returns a poison value.
2238 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2239 br i1 %cmp, label %true, label %end ; Branch to either destination.
2242 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2243 ; it has undefined behavior.
2247 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2248 ; Both edges into this PHI are
2249 ; control-dependent on %cmp, so this
2250 ; always results in a poison value.
2252 store volatile i32 0, i32* @g ; This would depend on the store in %true
2253 ; if %cmp is true, or the store in %entry
2254 ; otherwise, so this is undefined behavior.
2256 br i1 %cmp, label %second_true, label %second_end
2257 ; The same branch again, but this time the
2258 ; true block doesn't have side effects.
2265 store volatile i32 0, i32* @g ; This time, the instruction always depends
2266 ; on the store in %end. Also, it is
2267 ; control-equivalent to %end, so this is
2268 ; well-defined (ignoring earlier undefined
2269 ; behavior in this example).
2273 Addresses of Basic Blocks
2274 -------------------------
2276 ``blockaddress(@function, %block)``
2278 The '``blockaddress``' constant computes the address of the specified
2279 basic block in the specified function, and always has an ``i8*`` type.
2280 Taking the address of the entry block is illegal.
2282 This value only has defined behavior when used as an operand to the
2283 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2284 against null. Pointer equality tests between labels addresses results in
2285 undefined behavior --- though, again, comparison against null is ok, and
2286 no label is equal to the null pointer. This may be passed around as an
2287 opaque pointer sized value as long as the bits are not inspected. This
2288 allows ``ptrtoint`` and arithmetic to be performed on these values so
2289 long as the original value is reconstituted before the ``indirectbr``
2292 Finally, some targets may provide defined semantics when using the value
2293 as the operand to an inline assembly, but that is target specific.
2297 Constant Expressions
2298 --------------------
2300 Constant expressions are used to allow expressions involving other
2301 constants to be used as constants. Constant expressions may be of any
2302 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2303 that does not have side effects (e.g. load and call are not supported).
2304 The following is the syntax for constant expressions:
2306 ``trunc (CST to TYPE)``
2307 Truncate a constant to another type. The bit size of CST must be
2308 larger than the bit size of TYPE. Both types must be integers.
2309 ``zext (CST to TYPE)``
2310 Zero extend a constant to another type. The bit size of CST must be
2311 smaller than the bit size of TYPE. Both types must be integers.
2312 ``sext (CST to TYPE)``
2313 Sign extend a constant to another type. The bit size of CST must be
2314 smaller than the bit size of TYPE. Both types must be integers.
2315 ``fptrunc (CST to TYPE)``
2316 Truncate a floating point constant to another floating point type.
2317 The size of CST must be larger than the size of TYPE. Both types
2318 must be floating point.
2319 ``fpext (CST to TYPE)``
2320 Floating point extend a constant to another type. The size of CST
2321 must be smaller or equal to the size of TYPE. Both types must be
2323 ``fptoui (CST to TYPE)``
2324 Convert a floating point constant to the corresponding unsigned
2325 integer constant. TYPE must be a scalar or vector integer type. CST
2326 must be of scalar or vector floating point type. Both CST and TYPE
2327 must be scalars, or vectors of the same number of elements. If the
2328 value won't fit in the integer type, the results are undefined.
2329 ``fptosi (CST to TYPE)``
2330 Convert a floating point constant to the corresponding signed
2331 integer constant. TYPE must be a scalar or vector integer type. CST
2332 must be of scalar or vector floating point type. Both CST and TYPE
2333 must be scalars, or vectors of the same number of elements. If the
2334 value won't fit in the integer type, the results are undefined.
2335 ``uitofp (CST to TYPE)``
2336 Convert an unsigned integer constant to the corresponding floating
2337 point constant. TYPE must be a scalar or vector floating point type.
2338 CST must be of scalar or vector integer type. Both CST and TYPE must
2339 be scalars, or vectors of the same number of elements. If the value
2340 won't fit in the floating point type, the results are undefined.
2341 ``sitofp (CST to TYPE)``
2342 Convert a signed integer constant to the corresponding floating
2343 point constant. TYPE must be a scalar or vector floating point type.
2344 CST must be of scalar or vector integer type. Both CST and TYPE must
2345 be scalars, or vectors of the same number of elements. If the value
2346 won't fit in the floating point type, the results are undefined.
2347 ``ptrtoint (CST to TYPE)``
2348 Convert a pointer typed constant to the corresponding integer
2349 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2350 pointer type. The ``CST`` value is zero extended, truncated, or
2351 unchanged to make it fit in ``TYPE``.
2352 ``inttoptr (CST to TYPE)``
2353 Convert an integer constant to a pointer constant. TYPE must be a
2354 pointer type. CST must be of integer type. The CST value is zero
2355 extended, truncated, or unchanged to make it fit in a pointer size.
2356 This one is *really* dangerous!
2357 ``bitcast (CST to TYPE)``
2358 Convert a constant, CST, to another TYPE. The constraints of the
2359 operands are the same as those for the :ref:`bitcast
2360 instruction <i_bitcast>`.
2361 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2362 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2363 constants. As with the :ref:`getelementptr <i_getelementptr>`
2364 instruction, the index list may have zero or more indexes, which are
2365 required to make sense for the type of "CSTPTR".
2366 ``select (COND, VAL1, VAL2)``
2367 Perform the :ref:`select operation <i_select>` on constants.
2368 ``icmp COND (VAL1, VAL2)``
2369 Performs the :ref:`icmp operation <i_icmp>` on constants.
2370 ``fcmp COND (VAL1, VAL2)``
2371 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2372 ``extractelement (VAL, IDX)``
2373 Perform the :ref:`extractelement operation <i_extractelement>` on
2375 ``insertelement (VAL, ELT, IDX)``
2376 Perform the :ref:`insertelement operation <i_insertelement>` on
2378 ``shufflevector (VEC1, VEC2, IDXMASK)``
2379 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2381 ``extractvalue (VAL, IDX0, IDX1, ...)``
2382 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2383 constants. The index list is interpreted in a similar manner as
2384 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2385 least one index value must be specified.
2386 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2387 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2388 The index list is interpreted in a similar manner as indices in a
2389 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2390 value must be specified.
2391 ``OPCODE (LHS, RHS)``
2392 Perform the specified operation of the LHS and RHS constants. OPCODE
2393 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2394 binary <bitwiseops>` operations. The constraints on operands are
2395 the same as those for the corresponding instruction (e.g. no bitwise
2396 operations on floating point values are allowed).
2403 Inline Assembler Expressions
2404 ----------------------------
2406 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2407 Inline Assembly <moduleasm>`) through the use of a special value. This
2408 value represents the inline assembler as a string (containing the
2409 instructions to emit), a list of operand constraints (stored as a
2410 string), a flag that indicates whether or not the inline asm expression
2411 has side effects, and a flag indicating whether the function containing
2412 the asm needs to align its stack conservatively. An example inline
2413 assembler expression is:
2415 .. code-block:: llvm
2417 i32 (i32) asm "bswap $0", "=r,r"
2419 Inline assembler expressions may **only** be used as the callee operand
2420 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2421 Thus, typically we have:
2423 .. code-block:: llvm
2425 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2427 Inline asms with side effects not visible in the constraint list must be
2428 marked as having side effects. This is done through the use of the
2429 '``sideeffect``' keyword, like so:
2431 .. code-block:: llvm
2433 call void asm sideeffect "eieio", ""()
2435 In some cases inline asms will contain code that will not work unless
2436 the stack is aligned in some way, such as calls or SSE instructions on
2437 x86, yet will not contain code that does that alignment within the asm.
2438 The compiler should make conservative assumptions about what the asm
2439 might contain and should generate its usual stack alignment code in the
2440 prologue if the '``alignstack``' keyword is present:
2442 .. code-block:: llvm
2444 call void asm alignstack "eieio", ""()
2446 Inline asms also support using non-standard assembly dialects. The
2447 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2448 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2449 the only supported dialects. An example is:
2451 .. code-block:: llvm
2453 call void asm inteldialect "eieio", ""()
2455 If multiple keywords appear the '``sideeffect``' keyword must come
2456 first, the '``alignstack``' keyword second and the '``inteldialect``'
2462 The call instructions that wrap inline asm nodes may have a
2463 "``!srcloc``" MDNode attached to it that contains a list of constant
2464 integers. If present, the code generator will use the integer as the
2465 location cookie value when report errors through the ``LLVMContext``
2466 error reporting mechanisms. This allows a front-end to correlate backend
2467 errors that occur with inline asm back to the source code that produced
2470 .. code-block:: llvm
2472 call void asm sideeffect "something bad", ""(), !srcloc !42
2474 !42 = !{ i32 1234567 }
2476 It is up to the front-end to make sense of the magic numbers it places
2477 in the IR. If the MDNode contains multiple constants, the code generator
2478 will use the one that corresponds to the line of the asm that the error
2483 Metadata Nodes and Metadata Strings
2484 -----------------------------------
2486 LLVM IR allows metadata to be attached to instructions in the program
2487 that can convey extra information about the code to the optimizers and
2488 code generator. One example application of metadata is source-level
2489 debug information. There are two metadata primitives: strings and nodes.
2490 All metadata has the ``metadata`` type and is identified in syntax by a
2491 preceding exclamation point ('``!``').
2493 A metadata string is a string surrounded by double quotes. It can
2494 contain any character by escaping non-printable characters with
2495 "``\xx``" where "``xx``" is the two digit hex code. For example:
2498 Metadata nodes are represented with notation similar to structure
2499 constants (a comma separated list of elements, surrounded by braces and
2500 preceded by an exclamation point). Metadata nodes can have any values as
2501 their operand. For example:
2503 .. code-block:: llvm
2505 !{ metadata !"test\00", i32 10}
2507 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2508 metadata nodes, which can be looked up in the module symbol table. For
2511 .. code-block:: llvm
2513 !foo = metadata !{!4, !3}
2515 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2516 function is using two metadata arguments:
2518 .. code-block:: llvm
2520 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2522 Metadata can be attached with an instruction. Here metadata ``!21`` is
2523 attached to the ``add`` instruction using the ``!dbg`` identifier:
2525 .. code-block:: llvm
2527 %indvar.next = add i64 %indvar, 1, !dbg !21
2529 More information about specific metadata nodes recognized by the
2530 optimizers and code generator is found below.
2535 In LLVM IR, memory does not have types, so LLVM's own type system is not
2536 suitable for doing TBAA. Instead, metadata is added to the IR to
2537 describe a type system of a higher level language. This can be used to
2538 implement typical C/C++ TBAA, but it can also be used to implement
2539 custom alias analysis behavior for other languages.
2541 The current metadata format is very simple. TBAA metadata nodes have up
2542 to three fields, e.g.:
2544 .. code-block:: llvm
2546 !0 = metadata !{ metadata !"an example type tree" }
2547 !1 = metadata !{ metadata !"int", metadata !0 }
2548 !2 = metadata !{ metadata !"float", metadata !0 }
2549 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2551 The first field is an identity field. It can be any value, usually a
2552 metadata string, which uniquely identifies the type. The most important
2553 name in the tree is the name of the root node. Two trees with different
2554 root node names are entirely disjoint, even if they have leaves with
2557 The second field identifies the type's parent node in the tree, or is
2558 null or omitted for a root node. A type is considered to alias all of
2559 its descendants and all of its ancestors in the tree. Also, a type is
2560 considered to alias all types in other trees, so that bitcode produced
2561 from multiple front-ends is handled conservatively.
2563 If the third field is present, it's an integer which if equal to 1
2564 indicates that the type is "constant" (meaning
2565 ``pointsToConstantMemory`` should return true; see `other useful
2566 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2568 '``tbaa.struct``' Metadata
2569 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2571 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2572 aggregate assignment operations in C and similar languages, however it
2573 is defined to copy a contiguous region of memory, which is more than
2574 strictly necessary for aggregate types which contain holes due to
2575 padding. Also, it doesn't contain any TBAA information about the fields
2578 ``!tbaa.struct`` metadata can describe which memory subregions in a
2579 memcpy are padding and what the TBAA tags of the struct are.
2581 The current metadata format is very simple. ``!tbaa.struct`` metadata
2582 nodes are a list of operands which are in conceptual groups of three.
2583 For each group of three, the first operand gives the byte offset of a
2584 field in bytes, the second gives its size in bytes, and the third gives
2587 .. code-block:: llvm
2589 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2591 This describes a struct with two fields. The first is at offset 0 bytes
2592 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2593 and has size 4 bytes and has tbaa tag !2.
2595 Note that the fields need not be contiguous. In this example, there is a
2596 4 byte gap between the two fields. This gap represents padding which
2597 does not carry useful data and need not be preserved.
2599 '``fpmath``' Metadata
2600 ^^^^^^^^^^^^^^^^^^^^^
2602 ``fpmath`` metadata may be attached to any instruction of floating point
2603 type. It can be used to express the maximum acceptable error in the
2604 result of that instruction, in ULPs, thus potentially allowing the
2605 compiler to use a more efficient but less accurate method of computing
2606 it. ULP is defined as follows:
2608 If ``x`` is a real number that lies between two finite consecutive
2609 floating-point numbers ``a`` and ``b``, without being equal to one
2610 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2611 distance between the two non-equal finite floating-point numbers
2612 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2614 The metadata node shall consist of a single positive floating point
2615 number representing the maximum relative error, for example:
2617 .. code-block:: llvm
2619 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2621 '``range``' Metadata
2622 ^^^^^^^^^^^^^^^^^^^^
2624 ``range`` metadata may be attached only to loads of integer types. It
2625 expresses the possible ranges the loaded value is in. The ranges are
2626 represented with a flattened list of integers. The loaded value is known
2627 to be in the union of the ranges defined by each consecutive pair. Each
2628 pair has the following properties:
2630 - The type must match the type loaded by the instruction.
2631 - The pair ``a,b`` represents the range ``[a,b)``.
2632 - Both ``a`` and ``b`` are constants.
2633 - The range is allowed to wrap.
2634 - The range should not represent the full or empty set. That is,
2637 In addition, the pairs must be in signed order of the lower bound and
2638 they must be non-contiguous.
2642 .. code-block:: llvm
2644 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2645 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2646 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2647 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2649 !0 = metadata !{ i8 0, i8 2 }
2650 !1 = metadata !{ i8 255, i8 2 }
2651 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2652 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2657 It is sometimes useful to attach information to loop constructs. Currently,
2658 loop metadata is implemented as metadata attached to the branch instruction
2659 in the loop latch block. This type of metadata refer to a metadata node that is
2660 guaranteed to be separate for each loop. The loop identifier metadata is
2661 specified with the name ``llvm.loop``.
2663 The loop identifier metadata is implemented using a metadata that refers to
2664 itself to avoid merging it with any other identifier metadata, e.g.,
2665 during module linkage or function inlining. That is, each loop should refer
2666 to their own identification metadata even if they reside in separate functions.
2667 The following example contains loop identifier metadata for two separate loop
2670 .. code-block:: llvm
2672 !0 = metadata !{ metadata !0 }
2673 !1 = metadata !{ metadata !1 }
2675 The loop identifier metadata can be used to specify additional per-loop
2676 metadata. Any operands after the first operand can be treated as user-defined
2677 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2678 by the loop vectorizer to indicate how many times to unroll the loop:
2680 .. code-block:: llvm
2682 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2684 !0 = metadata !{ metadata !0, metadata !1 }
2685 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2690 Metadata types used to annotate memory accesses with information helpful
2691 for optimizations are prefixed with ``llvm.mem``.
2693 '``llvm.mem.parallel_loop_access``' Metadata
2694 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2696 For a loop to be parallel, in addition to using
2697 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2698 also all of the memory accessing instructions in the loop body need to be
2699 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2700 is at least one memory accessing instruction not marked with the metadata,
2701 the loop must be considered a sequential loop. This causes parallel loops to be
2702 converted to sequential loops due to optimization passes that are unaware of
2703 the parallel semantics and that insert new memory instructions to the loop
2706 Example of a loop that is considered parallel due to its correct use of
2707 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2708 metadata types that refer to the same loop identifier metadata.
2710 .. code-block:: llvm
2714 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2716 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2718 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2722 !0 = metadata !{ metadata !0 }
2724 It is also possible to have nested parallel loops. In that case the
2725 memory accesses refer to a list of loop identifier metadata nodes instead of
2726 the loop identifier metadata node directly:
2728 .. code-block:: llvm
2735 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2737 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2739 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2743 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2745 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2747 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2749 outer.for.end: ; preds = %for.body
2751 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2752 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2753 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2755 '``llvm.vectorizer``'
2756 ^^^^^^^^^^^^^^^^^^^^^
2758 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2759 vectorization parameters such as vectorization factor and unroll factor.
2761 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2762 loop identification metadata.
2764 '``llvm.vectorizer.unroll``' Metadata
2765 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2767 This metadata instructs the loop vectorizer to unroll the specified
2768 loop exactly ``N`` times.
2770 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2771 operand is an integer specifying the unroll factor. For example:
2773 .. code-block:: llvm
2775 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2777 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2780 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2781 determined automatically.
2783 '``llvm.vectorizer.width``' Metadata
2784 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2786 This metadata sets the target width of the vectorizer to ``N``. Without
2787 this metadata, the vectorizer will choose a width automatically.
2788 Regardless of this metadata, the vectorizer will only vectorize loops if
2789 it believes it is valid to do so.
2791 The first operand is the string ``llvm.vectorizer.width`` and the second
2792 operand is an integer specifying the width. For example:
2794 .. code-block:: llvm
2796 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2798 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2801 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2804 Module Flags Metadata
2805 =====================
2807 Information about the module as a whole is difficult to convey to LLVM's
2808 subsystems. The LLVM IR isn't sufficient to transmit this information.
2809 The ``llvm.module.flags`` named metadata exists in order to facilitate
2810 this. These flags are in the form of key / value pairs --- much like a
2811 dictionary --- making it easy for any subsystem who cares about a flag to
2814 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2815 Each triplet has the following form:
2817 - The first element is a *behavior* flag, which specifies the behavior
2818 when two (or more) modules are merged together, and it encounters two
2819 (or more) metadata with the same ID. The supported behaviors are
2821 - The second element is a metadata string that is a unique ID for the
2822 metadata. Each module may only have one flag entry for each unique ID (not
2823 including entries with the **Require** behavior).
2824 - The third element is the value of the flag.
2826 When two (or more) modules are merged together, the resulting
2827 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2828 each unique metadata ID string, there will be exactly one entry in the merged
2829 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2830 be determined by the merge behavior flag, as described below. The only exception
2831 is that entries with the *Require* behavior are always preserved.
2833 The following behaviors are supported:
2844 Emits an error if two values disagree, otherwise the resulting value
2845 is that of the operands.
2849 Emits a warning if two values disagree. The result value will be the
2850 operand for the flag from the first module being linked.
2854 Adds a requirement that another module flag be present and have a
2855 specified value after linking is performed. The value must be a
2856 metadata pair, where the first element of the pair is the ID of the
2857 module flag to be restricted, and the second element of the pair is
2858 the value the module flag should be restricted to. This behavior can
2859 be used to restrict the allowable results (via triggering of an
2860 error) of linking IDs with the **Override** behavior.
2864 Uses the specified value, regardless of the behavior or value of the
2865 other module. If both modules specify **Override**, but the values
2866 differ, an error will be emitted.
2870 Appends the two values, which are required to be metadata nodes.
2874 Appends the two values, which are required to be metadata
2875 nodes. However, duplicate entries in the second list are dropped
2876 during the append operation.
2878 It is an error for a particular unique flag ID to have multiple behaviors,
2879 except in the case of **Require** (which adds restrictions on another metadata
2880 value) or **Override**.
2882 An example of module flags:
2884 .. code-block:: llvm
2886 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2887 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2888 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2889 !3 = metadata !{ i32 3, metadata !"qux",
2891 metadata !"foo", i32 1
2894 !llvm.module.flags = !{ !0, !1, !2, !3 }
2896 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2897 if two or more ``!"foo"`` flags are seen is to emit an error if their
2898 values are not equal.
2900 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2901 behavior if two or more ``!"bar"`` flags are seen is to use the value
2904 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2905 behavior if two or more ``!"qux"`` flags are seen is to emit a
2906 warning if their values are not equal.
2908 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2912 metadata !{ metadata !"foo", i32 1 }
2914 The behavior is to emit an error if the ``llvm.module.flags`` does not
2915 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2918 Objective-C Garbage Collection Module Flags Metadata
2919 ----------------------------------------------------
2921 On the Mach-O platform, Objective-C stores metadata about garbage
2922 collection in a special section called "image info". The metadata
2923 consists of a version number and a bitmask specifying what types of
2924 garbage collection are supported (if any) by the file. If two or more
2925 modules are linked together their garbage collection metadata needs to
2926 be merged rather than appended together.
2928 The Objective-C garbage collection module flags metadata consists of the
2929 following key-value pairs:
2938 * - ``Objective-C Version``
2939 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2941 * - ``Objective-C Image Info Version``
2942 - **[Required]** --- The version of the image info section. Currently
2945 * - ``Objective-C Image Info Section``
2946 - **[Required]** --- The section to place the metadata. Valid values are
2947 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2948 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2949 Objective-C ABI version 2.
2951 * - ``Objective-C Garbage Collection``
2952 - **[Required]** --- Specifies whether garbage collection is supported or
2953 not. Valid values are 0, for no garbage collection, and 2, for garbage
2954 collection supported.
2956 * - ``Objective-C GC Only``
2957 - **[Optional]** --- Specifies that only garbage collection is supported.
2958 If present, its value must be 6. This flag requires that the
2959 ``Objective-C Garbage Collection`` flag have the value 2.
2961 Some important flag interactions:
2963 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2964 merged with a module with ``Objective-C Garbage Collection`` set to
2965 2, then the resulting module has the
2966 ``Objective-C Garbage Collection`` flag set to 0.
2967 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2968 merged with a module with ``Objective-C GC Only`` set to 6.
2970 Automatic Linker Flags Module Flags Metadata
2971 --------------------------------------------
2973 Some targets support embedding flags to the linker inside individual object
2974 files. Typically this is used in conjunction with language extensions which
2975 allow source files to explicitly declare the libraries they depend on, and have
2976 these automatically be transmitted to the linker via object files.
2978 These flags are encoded in the IR using metadata in the module flags section,
2979 using the ``Linker Options`` key. The merge behavior for this flag is required
2980 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2981 node which should be a list of other metadata nodes, each of which should be a
2982 list of metadata strings defining linker options.
2984 For example, the following metadata section specifies two separate sets of
2985 linker options, presumably to link against ``libz`` and the ``Cocoa``
2988 !0 = metadata !{ i32 6, metadata !"Linker Options",
2990 metadata !{ metadata !"-lz" },
2991 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2992 !llvm.module.flags = !{ !0 }
2994 The metadata encoding as lists of lists of options, as opposed to a collapsed
2995 list of options, is chosen so that the IR encoding can use multiple option
2996 strings to specify e.g., a single library, while still having that specifier be
2997 preserved as an atomic element that can be recognized by a target specific
2998 assembly writer or object file emitter.
3000 Each individual option is required to be either a valid option for the target's
3001 linker, or an option that is reserved by the target specific assembly writer or
3002 object file emitter. No other aspect of these options is defined by the IR.
3004 .. _intrinsicglobalvariables:
3006 Intrinsic Global Variables
3007 ==========================
3009 LLVM has a number of "magic" global variables that contain data that
3010 affect code generation or other IR semantics. These are documented here.
3011 All globals of this sort should have a section specified as
3012 "``llvm.metadata``". This section and all globals that start with
3013 "``llvm.``" are reserved for use by LLVM.
3017 The '``llvm.used``' Global Variable
3018 -----------------------------------
3020 The ``@llvm.used`` global is an array which has
3021 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3022 pointers to named global variables, functions and aliases which may optionally
3023 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3026 .. code-block:: llvm
3031 @llvm.used = appending global [2 x i8*] [
3033 i8* bitcast (i32* @Y to i8*)
3034 ], section "llvm.metadata"
3036 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3037 and linker are required to treat the symbol as if there is a reference to the
3038 symbol that it cannot see (which is why they have to be named). For example, if
3039 a variable has internal linkage and no references other than that from the
3040 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3041 references from inline asms and other things the compiler cannot "see", and
3042 corresponds to "``attribute((used))``" in GNU C.
3044 On some targets, the code generator must emit a directive to the
3045 assembler or object file to prevent the assembler and linker from
3046 molesting the symbol.
3048 .. _gv_llvmcompilerused:
3050 The '``llvm.compiler.used``' Global Variable
3051 --------------------------------------------
3053 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3054 directive, except that it only prevents the compiler from touching the
3055 symbol. On targets that support it, this allows an intelligent linker to
3056 optimize references to the symbol without being impeded as it would be
3059 This is a rare construct that should only be used in rare circumstances,
3060 and should not be exposed to source languages.
3062 .. _gv_llvmglobalctors:
3064 The '``llvm.global_ctors``' Global Variable
3065 -------------------------------------------
3067 .. code-block:: llvm
3069 %0 = type { i32, void ()* }
3070 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3072 The ``@llvm.global_ctors`` array contains a list of constructor
3073 functions and associated priorities. The functions referenced by this
3074 array will be called in ascending order of priority (i.e. lowest first)
3075 when the module is loaded. The order of functions with the same priority
3078 .. _llvmglobaldtors:
3080 The '``llvm.global_dtors``' Global Variable
3081 -------------------------------------------
3083 .. code-block:: llvm
3085 %0 = type { i32, void ()* }
3086 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3088 The ``@llvm.global_dtors`` array contains a list of destructor functions
3089 and associated priorities. The functions referenced by this array will
3090 be called in descending order of priority (i.e. highest first) when the
3091 module is loaded. The order of functions with the same priority is not
3094 Instruction Reference
3095 =====================
3097 The LLVM instruction set consists of several different classifications
3098 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3099 instructions <binaryops>`, :ref:`bitwise binary
3100 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3101 :ref:`other instructions <otherops>`.
3105 Terminator Instructions
3106 -----------------------
3108 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3109 program ends with a "Terminator" instruction, which indicates which
3110 block should be executed after the current block is finished. These
3111 terminator instructions typically yield a '``void``' value: they produce
3112 control flow, not values (the one exception being the
3113 ':ref:`invoke <i_invoke>`' instruction).
3115 The terminator instructions are: ':ref:`ret <i_ret>`',
3116 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3117 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3118 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3122 '``ret``' Instruction
3123 ^^^^^^^^^^^^^^^^^^^^^
3130 ret <type> <value> ; Return a value from a non-void function
3131 ret void ; Return from void function
3136 The '``ret``' instruction is used to return control flow (and optionally
3137 a value) from a function back to the caller.
3139 There are two forms of the '``ret``' instruction: one that returns a
3140 value and then causes control flow, and one that just causes control
3146 The '``ret``' instruction optionally accepts a single argument, the
3147 return value. The type of the return value must be a ':ref:`first
3148 class <t_firstclass>`' type.
3150 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3151 return type and contains a '``ret``' instruction with no return value or
3152 a return value with a type that does not match its type, or if it has a
3153 void return type and contains a '``ret``' instruction with a return
3159 When the '``ret``' instruction is executed, control flow returns back to
3160 the calling function's context. If the caller is a
3161 ":ref:`call <i_call>`" instruction, execution continues at the
3162 instruction after the call. If the caller was an
3163 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3164 beginning of the "normal" destination block. If the instruction returns
3165 a value, that value shall set the call or invoke instruction's return
3171 .. code-block:: llvm
3173 ret i32 5 ; Return an integer value of 5
3174 ret void ; Return from a void function
3175 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3179 '``br``' Instruction
3180 ^^^^^^^^^^^^^^^^^^^^
3187 br i1 <cond>, label <iftrue>, label <iffalse>
3188 br label <dest> ; Unconditional branch
3193 The '``br``' instruction is used to cause control flow to transfer to a
3194 different basic block in the current function. There are two forms of
3195 this instruction, corresponding to a conditional branch and an
3196 unconditional branch.
3201 The conditional branch form of the '``br``' instruction takes a single
3202 '``i1``' value and two '``label``' values. The unconditional form of the
3203 '``br``' instruction takes a single '``label``' value as a target.
3208 Upon execution of a conditional '``br``' instruction, the '``i1``'
3209 argument is evaluated. If the value is ``true``, control flows to the
3210 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3211 to the '``iffalse``' ``label`` argument.
3216 .. code-block:: llvm
3219 %cond = icmp eq i32 %a, %b
3220 br i1 %cond, label %IfEqual, label %IfUnequal
3228 '``switch``' Instruction
3229 ^^^^^^^^^^^^^^^^^^^^^^^^
3236 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3241 The '``switch``' instruction is used to transfer control flow to one of
3242 several different places. It is a generalization of the '``br``'
3243 instruction, allowing a branch to occur to one of many possible
3249 The '``switch``' instruction uses three parameters: an integer
3250 comparison value '``value``', a default '``label``' destination, and an
3251 array of pairs of comparison value constants and '``label``'s. The table
3252 is not allowed to contain duplicate constant entries.
3257 The ``switch`` instruction specifies a table of values and destinations.
3258 When the '``switch``' instruction is executed, this table is searched
3259 for the given value. If the value is found, control flow is transferred
3260 to the corresponding destination; otherwise, control flow is transferred
3261 to the default destination.
3266 Depending on properties of the target machine and the particular
3267 ``switch`` instruction, this instruction may be code generated in
3268 different ways. For example, it could be generated as a series of
3269 chained conditional branches or with a lookup table.
3274 .. code-block:: llvm
3276 ; Emulate a conditional br instruction
3277 %Val = zext i1 %value to i32
3278 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3280 ; Emulate an unconditional br instruction
3281 switch i32 0, label %dest [ ]
3283 ; Implement a jump table:
3284 switch i32 %val, label %otherwise [ i32 0, label %onzero
3286 i32 2, label %ontwo ]
3290 '``indirectbr``' Instruction
3291 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3298 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3303 The '``indirectbr``' instruction implements an indirect branch to a
3304 label within the current function, whose address is specified by
3305 "``address``". Address must be derived from a
3306 :ref:`blockaddress <blockaddress>` constant.
3311 The '``address``' argument is the address of the label to jump to. The
3312 rest of the arguments indicate the full set of possible destinations
3313 that the address may point to. Blocks are allowed to occur multiple
3314 times in the destination list, though this isn't particularly useful.
3316 This destination list is required so that dataflow analysis has an
3317 accurate understanding of the CFG.
3322 Control transfers to the block specified in the address argument. All
3323 possible destination blocks must be listed in the label list, otherwise
3324 this instruction has undefined behavior. This implies that jumps to
3325 labels defined in other functions have undefined behavior as well.
3330 This is typically implemented with a jump through a register.
3335 .. code-block:: llvm
3337 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3341 '``invoke``' Instruction
3342 ^^^^^^^^^^^^^^^^^^^^^^^^
3349 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3350 to label <normal label> unwind label <exception label>
3355 The '``invoke``' instruction causes control to transfer to a specified
3356 function, with the possibility of control flow transfer to either the
3357 '``normal``' label or the '``exception``' label. If the callee function
3358 returns with the "``ret``" instruction, control flow will return to the
3359 "normal" label. If the callee (or any indirect callees) returns via the
3360 ":ref:`resume <i_resume>`" instruction or other exception handling
3361 mechanism, control is interrupted and continued at the dynamically
3362 nearest "exception" label.
3364 The '``exception``' label is a `landing
3365 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3366 '``exception``' label is required to have the
3367 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3368 information about the behavior of the program after unwinding happens,
3369 as its first non-PHI instruction. The restrictions on the
3370 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3371 instruction, so that the important information contained within the
3372 "``landingpad``" instruction can't be lost through normal code motion.
3377 This instruction requires several arguments:
3379 #. The optional "cconv" marker indicates which :ref:`calling
3380 convention <callingconv>` the call should use. If none is
3381 specified, the call defaults to using C calling conventions.
3382 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3383 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3385 #. '``ptr to function ty``': shall be the signature of the pointer to
3386 function value being invoked. In most cases, this is a direct
3387 function invocation, but indirect ``invoke``'s are just as possible,
3388 branching off an arbitrary pointer to function value.
3389 #. '``function ptr val``': An LLVM value containing a pointer to a
3390 function to be invoked.
3391 #. '``function args``': argument list whose types match the function
3392 signature argument types and parameter attributes. All arguments must
3393 be of :ref:`first class <t_firstclass>` type. If the function signature
3394 indicates the function accepts a variable number of arguments, the
3395 extra arguments can be specified.
3396 #. '``normal label``': the label reached when the called function
3397 executes a '``ret``' instruction.
3398 #. '``exception label``': the label reached when a callee returns via
3399 the :ref:`resume <i_resume>` instruction or other exception handling
3401 #. The optional :ref:`function attributes <fnattrs>` list. Only
3402 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3403 attributes are valid here.
3408 This instruction is designed to operate as a standard '``call``'
3409 instruction in most regards. The primary difference is that it
3410 establishes an association with a label, which is used by the runtime
3411 library to unwind the stack.
3413 This instruction is used in languages with destructors to ensure that
3414 proper cleanup is performed in the case of either a ``longjmp`` or a
3415 thrown exception. Additionally, this is important for implementation of
3416 '``catch``' clauses in high-level languages that support them.
3418 For the purposes of the SSA form, the definition of the value returned
3419 by the '``invoke``' instruction is deemed to occur on the edge from the
3420 current block to the "normal" label. If the callee unwinds then no
3421 return value is available.
3426 .. code-block:: llvm
3428 %retval = invoke i32 @Test(i32 15) to label %Continue
3429 unwind label %TestCleanup ; {i32}:retval set
3430 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3431 unwind label %TestCleanup ; {i32}:retval set
3435 '``resume``' Instruction
3436 ^^^^^^^^^^^^^^^^^^^^^^^^
3443 resume <type> <value>
3448 The '``resume``' instruction is a terminator instruction that has no
3454 The '``resume``' instruction requires one argument, which must have the
3455 same type as the result of any '``landingpad``' instruction in the same
3461 The '``resume``' instruction resumes propagation of an existing
3462 (in-flight) exception whose unwinding was interrupted with a
3463 :ref:`landingpad <i_landingpad>` instruction.
3468 .. code-block:: llvm
3470 resume { i8*, i32 } %exn
3474 '``unreachable``' Instruction
3475 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3487 The '``unreachable``' instruction has no defined semantics. This
3488 instruction is used to inform the optimizer that a particular portion of
3489 the code is not reachable. This can be used to indicate that the code
3490 after a no-return function cannot be reached, and other facts.
3495 The '``unreachable``' instruction has no defined semantics.
3502 Binary operators are used to do most of the computation in a program.
3503 They require two operands of the same type, execute an operation on
3504 them, and produce a single value. The operands might represent multiple
3505 data, as is the case with the :ref:`vector <t_vector>` data type. The
3506 result value has the same type as its operands.
3508 There are several different binary operators:
3512 '``add``' Instruction
3513 ^^^^^^^^^^^^^^^^^^^^^
3520 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3521 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3522 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3523 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3528 The '``add``' instruction returns the sum of its two operands.
3533 The two arguments to the '``add``' instruction must be
3534 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3535 arguments must have identical types.
3540 The value produced is the integer sum of the two operands.
3542 If the sum has unsigned overflow, the result returned is the
3543 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3546 Because LLVM integers use a two's complement representation, this
3547 instruction is appropriate for both signed and unsigned integers.
3549 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3550 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3551 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3552 unsigned and/or signed overflow, respectively, occurs.
3557 .. code-block:: llvm
3559 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3563 '``fadd``' Instruction
3564 ^^^^^^^^^^^^^^^^^^^^^^
3571 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3576 The '``fadd``' instruction returns the sum of its two operands.
3581 The two arguments to the '``fadd``' instruction must be :ref:`floating
3582 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3583 Both arguments must have identical types.
3588 The value produced is the floating point sum of the two operands. This
3589 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3590 which are optimization hints to enable otherwise unsafe floating point
3596 .. code-block:: llvm
3598 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3600 '``sub``' Instruction
3601 ^^^^^^^^^^^^^^^^^^^^^
3608 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3609 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3610 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3611 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3616 The '``sub``' instruction returns the difference of its two operands.
3618 Note that the '``sub``' instruction is used to represent the '``neg``'
3619 instruction present in most other intermediate representations.
3624 The two arguments to the '``sub``' instruction must be
3625 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3626 arguments must have identical types.
3631 The value produced is the integer difference of the two operands.
3633 If the difference has unsigned overflow, the result returned is the
3634 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3637 Because LLVM integers use a two's complement representation, this
3638 instruction is appropriate for both signed and unsigned integers.
3640 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3641 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3642 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3643 unsigned and/or signed overflow, respectively, occurs.
3648 .. code-block:: llvm
3650 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3651 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3655 '``fsub``' Instruction
3656 ^^^^^^^^^^^^^^^^^^^^^^
3663 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3668 The '``fsub``' instruction returns the difference of its two operands.
3670 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3671 instruction present in most other intermediate representations.
3676 The two arguments to the '``fsub``' instruction must be :ref:`floating
3677 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3678 Both arguments must have identical types.
3683 The value produced is the floating point difference of the two operands.
3684 This instruction can also take any number of :ref:`fast-math
3685 flags <fastmath>`, which are optimization hints to enable otherwise
3686 unsafe floating point optimizations:
3691 .. code-block:: llvm
3693 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3694 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3696 '``mul``' Instruction
3697 ^^^^^^^^^^^^^^^^^^^^^
3704 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3705 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3706 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3707 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3712 The '``mul``' instruction returns the product of its two operands.
3717 The two arguments to the '``mul``' instruction must be
3718 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3719 arguments must have identical types.
3724 The value produced is the integer product of the two operands.
3726 If the result of the multiplication has unsigned overflow, the result
3727 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3728 bit width of the result.
3730 Because LLVM integers use a two's complement representation, and the
3731 result is the same width as the operands, this instruction returns the
3732 correct result for both signed and unsigned integers. If a full product
3733 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3734 sign-extended or zero-extended as appropriate to the width of the full
3737 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3738 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3739 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3740 unsigned and/or signed overflow, respectively, occurs.
3745 .. code-block:: llvm
3747 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3751 '``fmul``' Instruction
3752 ^^^^^^^^^^^^^^^^^^^^^^
3759 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3764 The '``fmul``' instruction returns the product of its two operands.
3769 The two arguments to the '``fmul``' instruction must be :ref:`floating
3770 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3771 Both arguments must have identical types.
3776 The value produced is the floating point product of the two operands.
3777 This instruction can also take any number of :ref:`fast-math
3778 flags <fastmath>`, which are optimization hints to enable otherwise
3779 unsafe floating point optimizations:
3784 .. code-block:: llvm
3786 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3788 '``udiv``' Instruction
3789 ^^^^^^^^^^^^^^^^^^^^^^
3796 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3797 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3802 The '``udiv``' instruction returns the quotient of its two operands.
3807 The two arguments to the '``udiv``' instruction must be
3808 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3809 arguments must have identical types.
3814 The value produced is the unsigned integer quotient of the two operands.
3816 Note that unsigned integer division and signed integer division are
3817 distinct operations; for signed integer division, use '``sdiv``'.
3819 Division by zero leads to undefined behavior.
3821 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3822 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3823 such, "((a udiv exact b) mul b) == a").
3828 .. code-block:: llvm
3830 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3832 '``sdiv``' Instruction
3833 ^^^^^^^^^^^^^^^^^^^^^^
3840 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3841 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3846 The '``sdiv``' instruction returns the quotient of its two operands.
3851 The two arguments to the '``sdiv``' instruction must be
3852 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3853 arguments must have identical types.
3858 The value produced is the signed integer quotient of the two operands
3859 rounded towards zero.
3861 Note that signed integer division and unsigned integer division are
3862 distinct operations; for unsigned integer division, use '``udiv``'.
3864 Division by zero leads to undefined behavior. Overflow also leads to
3865 undefined behavior; this is a rare case, but can occur, for example, by
3866 doing a 32-bit division of -2147483648 by -1.
3868 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3869 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3874 .. code-block:: llvm
3876 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3880 '``fdiv``' Instruction
3881 ^^^^^^^^^^^^^^^^^^^^^^
3888 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3893 The '``fdiv``' instruction returns the quotient of its two operands.
3898 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3899 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3900 Both arguments must have identical types.
3905 The value produced is the floating point quotient of the two operands.
3906 This instruction can also take any number of :ref:`fast-math
3907 flags <fastmath>`, which are optimization hints to enable otherwise
3908 unsafe floating point optimizations:
3913 .. code-block:: llvm
3915 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3917 '``urem``' Instruction
3918 ^^^^^^^^^^^^^^^^^^^^^^
3925 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3930 The '``urem``' instruction returns the remainder from the unsigned
3931 division of its two arguments.
3936 The two arguments to the '``urem``' instruction must be
3937 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3938 arguments must have identical types.
3943 This instruction returns the unsigned integer *remainder* of a division.
3944 This instruction always performs an unsigned division to get the
3947 Note that unsigned integer remainder and signed integer remainder are
3948 distinct operations; for signed integer remainder, use '``srem``'.
3950 Taking the remainder of a division by zero leads to undefined behavior.
3955 .. code-block:: llvm
3957 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3959 '``srem``' Instruction
3960 ^^^^^^^^^^^^^^^^^^^^^^
3967 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3972 The '``srem``' instruction returns the remainder from the signed
3973 division of its two operands. This instruction can also take
3974 :ref:`vector <t_vector>` versions of the values in which case the elements
3980 The two arguments to the '``srem``' instruction must be
3981 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3982 arguments must have identical types.
3987 This instruction returns the *remainder* of a division (where the result
3988 is either zero or has the same sign as the dividend, ``op1``), not the
3989 *modulo* operator (where the result is either zero or has the same sign
3990 as the divisor, ``op2``) of a value. For more information about the
3991 difference, see `The Math
3992 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3993 table of how this is implemented in various languages, please see
3995 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3997 Note that signed integer remainder and unsigned integer remainder are
3998 distinct operations; for unsigned integer remainder, use '``urem``'.
4000 Taking the remainder of a division by zero leads to undefined behavior.
4001 Overflow also leads to undefined behavior; this is a rare case, but can
4002 occur, for example, by taking the remainder of a 32-bit division of
4003 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4004 rule lets srem be implemented using instructions that return both the
4005 result of the division and the remainder.)
4010 .. code-block:: llvm
4012 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4016 '``frem``' Instruction
4017 ^^^^^^^^^^^^^^^^^^^^^^
4024 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4029 The '``frem``' instruction returns the remainder from the division of
4035 The two arguments to the '``frem``' instruction must be :ref:`floating
4036 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4037 Both arguments must have identical types.
4042 This instruction returns the *remainder* of a division. The remainder
4043 has the same sign as the dividend. This instruction can also take any
4044 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4045 to enable otherwise unsafe floating point optimizations:
4050 .. code-block:: llvm
4052 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4056 Bitwise Binary Operations
4057 -------------------------
4059 Bitwise binary operators are used to do various forms of bit-twiddling
4060 in a program. They are generally very efficient instructions and can
4061 commonly be strength reduced from other instructions. They require two
4062 operands of the same type, execute an operation on them, and produce a
4063 single value. The resulting value is the same type as its operands.
4065 '``shl``' Instruction
4066 ^^^^^^^^^^^^^^^^^^^^^
4073 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4074 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4075 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4076 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4081 The '``shl``' instruction returns the first operand shifted to the left
4082 a specified number of bits.
4087 Both arguments to the '``shl``' instruction must be the same
4088 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4089 '``op2``' is treated as an unsigned value.
4094 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4095 where ``n`` is the width of the result. If ``op2`` is (statically or
4096 dynamically) negative or equal to or larger than the number of bits in
4097 ``op1``, the result is undefined. If the arguments are vectors, each
4098 vector element of ``op1`` is shifted by the corresponding shift amount
4101 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4102 value <poisonvalues>` if it shifts out any non-zero bits. If the
4103 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4104 value <poisonvalues>` if it shifts out any bits that disagree with the
4105 resultant sign bit. As such, NUW/NSW have the same semantics as they
4106 would if the shift were expressed as a mul instruction with the same
4107 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4112 .. code-block:: llvm
4114 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4115 <result> = shl i32 4, 2 ; yields {i32}: 16
4116 <result> = shl i32 1, 10 ; yields {i32}: 1024
4117 <result> = shl i32 1, 32 ; undefined
4118 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4120 '``lshr``' Instruction
4121 ^^^^^^^^^^^^^^^^^^^^^^
4128 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4129 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4134 The '``lshr``' instruction (logical shift right) returns the first
4135 operand shifted to the right a specified number of bits with zero fill.
4140 Both arguments to the '``lshr``' instruction must be the same
4141 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4142 '``op2``' is treated as an unsigned value.
4147 This instruction always performs a logical shift right operation. The
4148 most significant bits of the result will be filled with zero bits after
4149 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4150 than the number of bits in ``op1``, the result is undefined. If the
4151 arguments are vectors, each vector element of ``op1`` is shifted by the
4152 corresponding shift amount in ``op2``.
4154 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4155 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4161 .. code-block:: llvm
4163 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4164 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4165 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4166 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4167 <result> = lshr i32 1, 32 ; undefined
4168 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4170 '``ashr``' Instruction
4171 ^^^^^^^^^^^^^^^^^^^^^^
4178 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4179 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4184 The '``ashr``' instruction (arithmetic shift right) returns the first
4185 operand shifted to the right a specified number of bits with sign
4191 Both arguments to the '``ashr``' instruction must be the same
4192 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4193 '``op2``' is treated as an unsigned value.
4198 This instruction always performs an arithmetic shift right operation,
4199 The most significant bits of the result will be filled with the sign bit
4200 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4201 than the number of bits in ``op1``, the result is undefined. If the
4202 arguments are vectors, each vector element of ``op1`` is shifted by the
4203 corresponding shift amount in ``op2``.
4205 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4206 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4212 .. code-block:: llvm
4214 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4215 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4216 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4217 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4218 <result> = ashr i32 1, 32 ; undefined
4219 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4221 '``and``' Instruction
4222 ^^^^^^^^^^^^^^^^^^^^^
4229 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4234 The '``and``' instruction returns the bitwise logical and of its two
4240 The two arguments to the '``and``' instruction must be
4241 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4242 arguments must have identical types.
4247 The truth table used for the '``and``' instruction is:
4264 .. code-block:: llvm
4266 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4267 <result> = and i32 15, 40 ; yields {i32}:result = 8
4268 <result> = and i32 4, 8 ; yields {i32}:result = 0
4270 '``or``' Instruction
4271 ^^^^^^^^^^^^^^^^^^^^
4278 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4283 The '``or``' instruction returns the bitwise logical inclusive or of its
4289 The two arguments to the '``or``' instruction must be
4290 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4291 arguments must have identical types.
4296 The truth table used for the '``or``' instruction is:
4315 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4316 <result> = or i32 15, 40 ; yields {i32}:result = 47
4317 <result> = or i32 4, 8 ; yields {i32}:result = 12
4319 '``xor``' Instruction
4320 ^^^^^^^^^^^^^^^^^^^^^
4327 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4332 The '``xor``' instruction returns the bitwise logical exclusive or of
4333 its two operands. The ``xor`` is used to implement the "one's
4334 complement" operation, which is the "~" operator in C.
4339 The two arguments to the '``xor``' instruction must be
4340 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4341 arguments must have identical types.
4346 The truth table used for the '``xor``' instruction is:
4363 .. code-block:: llvm
4365 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4366 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4367 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4368 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4373 LLVM supports several instructions to represent vector operations in a
4374 target-independent manner. These instructions cover the element-access
4375 and vector-specific operations needed to process vectors effectively.
4376 While LLVM does directly support these vector operations, many
4377 sophisticated algorithms will want to use target-specific intrinsics to
4378 take full advantage of a specific target.
4380 .. _i_extractelement:
4382 '``extractelement``' Instruction
4383 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4390 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4395 The '``extractelement``' instruction extracts a single scalar element
4396 from a vector at a specified index.
4401 The first operand of an '``extractelement``' instruction is a value of
4402 :ref:`vector <t_vector>` type. The second operand is an index indicating
4403 the position from which to extract the element. The index may be a
4409 The result is a scalar of the same type as the element type of ``val``.
4410 Its value is the value at position ``idx`` of ``val``. If ``idx``
4411 exceeds the length of ``val``, the results are undefined.
4416 .. code-block:: llvm
4418 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4420 .. _i_insertelement:
4422 '``insertelement``' Instruction
4423 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4430 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4435 The '``insertelement``' instruction inserts a scalar element into a
4436 vector at a specified index.
4441 The first operand of an '``insertelement``' instruction is a value of
4442 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4443 type must equal the element type of the first operand. The third operand
4444 is an index indicating the position at which to insert the value. The
4445 index may be a variable.
4450 The result is a vector of the same type as ``val``. Its element values
4451 are those of ``val`` except at position ``idx``, where it gets the value
4452 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4458 .. code-block:: llvm
4460 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4462 .. _i_shufflevector:
4464 '``shufflevector``' Instruction
4465 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4472 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4477 The '``shufflevector``' instruction constructs a permutation of elements
4478 from two input vectors, returning a vector with the same element type as
4479 the input and length that is the same as the shuffle mask.
4484 The first two operands of a '``shufflevector``' instruction are vectors
4485 with the same type. The third argument is a shuffle mask whose element
4486 type is always 'i32'. The result of the instruction is a vector whose
4487 length is the same as the shuffle mask and whose element type is the
4488 same as the element type of the first two operands.
4490 The shuffle mask operand is required to be a constant vector with either
4491 constant integer or undef values.
4496 The elements of the two input vectors are numbered from left to right
4497 across both of the vectors. The shuffle mask operand specifies, for each
4498 element of the result vector, which element of the two input vectors the
4499 result element gets. The element selector may be undef (meaning "don't
4500 care") and the second operand may be undef if performing a shuffle from
4506 .. code-block:: llvm
4508 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4509 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4510 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4511 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4512 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4513 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4514 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4515 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4517 Aggregate Operations
4518 --------------------
4520 LLVM supports several instructions for working with
4521 :ref:`aggregate <t_aggregate>` values.
4525 '``extractvalue``' Instruction
4526 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4533 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4538 The '``extractvalue``' instruction extracts the value of a member field
4539 from an :ref:`aggregate <t_aggregate>` value.
4544 The first operand of an '``extractvalue``' instruction is a value of
4545 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4546 constant indices to specify which value to extract in a similar manner
4547 as indices in a '``getelementptr``' instruction.
4549 The major differences to ``getelementptr`` indexing are:
4551 - Since the value being indexed is not a pointer, the first index is
4552 omitted and assumed to be zero.
4553 - At least one index must be specified.
4554 - Not only struct indices but also array indices must be in bounds.
4559 The result is the value at the position in the aggregate specified by
4565 .. code-block:: llvm
4567 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4571 '``insertvalue``' Instruction
4572 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4579 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4584 The '``insertvalue``' instruction inserts a value into a member field in
4585 an :ref:`aggregate <t_aggregate>` value.
4590 The first operand of an '``insertvalue``' instruction is a value of
4591 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4592 a first-class value to insert. The following operands are constant
4593 indices indicating the position at which to insert the value in a
4594 similar manner as indices in a '``extractvalue``' instruction. The value
4595 to insert must have the same type as the value identified by the
4601 The result is an aggregate of the same type as ``val``. Its value is
4602 that of ``val`` except that the value at the position specified by the
4603 indices is that of ``elt``.
4608 .. code-block:: llvm
4610 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4611 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4612 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4616 Memory Access and Addressing Operations
4617 ---------------------------------------
4619 A key design point of an SSA-based representation is how it represents
4620 memory. In LLVM, no memory locations are in SSA form, which makes things
4621 very simple. This section describes how to read, write, and allocate
4626 '``alloca``' Instruction
4627 ^^^^^^^^^^^^^^^^^^^^^^^^
4634 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4639 The '``alloca``' instruction allocates memory on the stack frame of the
4640 currently executing function, to be automatically released when this
4641 function returns to its caller. The object is always allocated in the
4642 generic address space (address space zero).
4647 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4648 bytes of memory on the runtime stack, returning a pointer of the
4649 appropriate type to the program. If "NumElements" is specified, it is
4650 the number of elements allocated, otherwise "NumElements" is defaulted
4651 to be one. If a constant alignment is specified, the value result of the
4652 allocation is guaranteed to be aligned to at least that boundary. If not
4653 specified, or if zero, the target can choose to align the allocation on
4654 any convenient boundary compatible with the type.
4656 '``type``' may be any sized type.
4661 Memory is allocated; a pointer is returned. The operation is undefined
4662 if there is insufficient stack space for the allocation. '``alloca``'d
4663 memory is automatically released when the function returns. The
4664 '``alloca``' instruction is commonly used to represent automatic
4665 variables that must have an address available. When the function returns
4666 (either with the ``ret`` or ``resume`` instructions), the memory is
4667 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4668 The order in which memory is allocated (ie., which way the stack grows)
4674 .. code-block:: llvm
4676 %ptr = alloca i32 ; yields {i32*}:ptr
4677 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4678 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4679 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4683 '``load``' Instruction
4684 ^^^^^^^^^^^^^^^^^^^^^^
4691 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4692 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4693 !<index> = !{ i32 1 }
4698 The '``load``' instruction is used to read from memory.
4703 The argument to the ``load`` instruction specifies the memory address
4704 from which to load. The pointer must point to a :ref:`first
4705 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4706 then the optimizer is not allowed to modify the number or order of
4707 execution of this ``load`` with other :ref:`volatile
4708 operations <volatile>`.
4710 If the ``load`` is marked as ``atomic``, it takes an extra
4711 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4712 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4713 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4714 when they may see multiple atomic stores. The type of the pointee must
4715 be an integer type whose bit width is a power of two greater than or
4716 equal to eight and less than or equal to a target-specific size limit.
4717 ``align`` must be explicitly specified on atomic loads, and the load has
4718 undefined behavior if the alignment is not set to a value which is at
4719 least the size in bytes of the pointee. ``!nontemporal`` does not have
4720 any defined semantics for atomic loads.
4722 The optional constant ``align`` argument specifies the alignment of the
4723 operation (that is, the alignment of the memory address). A value of 0
4724 or an omitted ``align`` argument means that the operation has the ABI
4725 alignment for the target. It is the responsibility of the code emitter
4726 to ensure that the alignment information is correct. Overestimating the
4727 alignment results in undefined behavior. Underestimating the alignment
4728 may produce less efficient code. An alignment of 1 is always safe.
4730 The optional ``!nontemporal`` metadata must reference a single
4731 metadata name ``<index>`` corresponding to a metadata node with one
4732 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4733 metadata on the instruction tells the optimizer and code generator
4734 that this load is not expected to be reused in the cache. The code
4735 generator may select special instructions to save cache bandwidth, such
4736 as the ``MOVNT`` instruction on x86.
4738 The optional ``!invariant.load`` metadata must reference a single
4739 metadata name ``<index>`` corresponding to a metadata node with no
4740 entries. The existence of the ``!invariant.load`` metadata on the
4741 instruction tells the optimizer and code generator that this load
4742 address points to memory which does not change value during program
4743 execution. The optimizer may then move this load around, for example, by
4744 hoisting it out of loops using loop invariant code motion.
4749 The location of memory pointed to is loaded. If the value being loaded
4750 is of scalar type then the number of bytes read does not exceed the
4751 minimum number of bytes needed to hold all bits of the type. For
4752 example, loading an ``i24`` reads at most three bytes. When loading a
4753 value of a type like ``i20`` with a size that is not an integral number
4754 of bytes, the result is undefined if the value was not originally
4755 written using a store of the same type.
4760 .. code-block:: llvm
4762 %ptr = alloca i32 ; yields {i32*}:ptr
4763 store i32 3, i32* %ptr ; yields {void}
4764 %val = load i32* %ptr ; yields {i32}:val = i32 3
4768 '``store``' Instruction
4769 ^^^^^^^^^^^^^^^^^^^^^^^
4776 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4777 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4782 The '``store``' instruction is used to write to memory.
4787 There are two arguments to the ``store`` instruction: a value to store
4788 and an address at which to store it. The type of the ``<pointer>``
4789 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4790 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4791 then the optimizer is not allowed to modify the number or order of
4792 execution of this ``store`` with other :ref:`volatile
4793 operations <volatile>`.
4795 If the ``store`` is marked as ``atomic``, it takes an extra
4796 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4797 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4798 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4799 when they may see multiple atomic stores. The type of the pointee must
4800 be an integer type whose bit width is a power of two greater than or
4801 equal to eight and less than or equal to a target-specific size limit.
4802 ``align`` must be explicitly specified on atomic stores, and the store
4803 has undefined behavior if the alignment is not set to a value which is
4804 at least the size in bytes of the pointee. ``!nontemporal`` does not
4805 have any defined semantics for atomic stores.
4807 The optional constant ``align`` argument specifies the alignment of the
4808 operation (that is, the alignment of the memory address). A value of 0
4809 or an omitted ``align`` argument means that the operation has the ABI
4810 alignment for the target. It is the responsibility of the code emitter
4811 to ensure that the alignment information is correct. Overestimating the
4812 alignment results in undefined behavior. Underestimating the
4813 alignment may produce less efficient code. An alignment of 1 is always
4816 The optional ``!nontemporal`` metadata must reference a single metadata
4817 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4818 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4819 tells the optimizer and code generator that this load is not expected to
4820 be reused in the cache. The code generator may select special
4821 instructions to save cache bandwidth, such as the MOVNT instruction on
4827 The contents of memory are updated to contain ``<value>`` at the
4828 location specified by the ``<pointer>`` operand. If ``<value>`` is
4829 of scalar type then the number of bytes written does not exceed the
4830 minimum number of bytes needed to hold all bits of the type. For
4831 example, storing an ``i24`` writes at most three bytes. When writing a
4832 value of a type like ``i20`` with a size that is not an integral number
4833 of bytes, it is unspecified what happens to the extra bits that do not
4834 belong to the type, but they will typically be overwritten.
4839 .. code-block:: llvm
4841 %ptr = alloca i32 ; yields {i32*}:ptr
4842 store i32 3, i32* %ptr ; yields {void}
4843 %val = load i32* %ptr ; yields {i32}:val = i32 3
4847 '``fence``' Instruction
4848 ^^^^^^^^^^^^^^^^^^^^^^^
4855 fence [singlethread] <ordering> ; yields {void}
4860 The '``fence``' instruction is used to introduce happens-before edges
4866 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4867 defines what *synchronizes-with* edges they add. They can only be given
4868 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4873 A fence A which has (at least) ``release`` ordering semantics
4874 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4875 semantics if and only if there exist atomic operations X and Y, both
4876 operating on some atomic object M, such that A is sequenced before X, X
4877 modifies M (either directly or through some side effect of a sequence
4878 headed by X), Y is sequenced before B, and Y observes M. This provides a
4879 *happens-before* dependency between A and B. Rather than an explicit
4880 ``fence``, one (but not both) of the atomic operations X or Y might
4881 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4882 still *synchronize-with* the explicit ``fence`` and establish the
4883 *happens-before* edge.
4885 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4886 ``acquire`` and ``release`` semantics specified above, participates in
4887 the global program order of other ``seq_cst`` operations and/or fences.
4889 The optional ":ref:`singlethread <singlethread>`" argument specifies
4890 that the fence only synchronizes with other fences in the same thread.
4891 (This is useful for interacting with signal handlers.)
4896 .. code-block:: llvm
4898 fence acquire ; yields {void}
4899 fence singlethread seq_cst ; yields {void}
4903 '``cmpxchg``' Instruction
4904 ^^^^^^^^^^^^^^^^^^^^^^^^^
4911 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4916 The '``cmpxchg``' instruction is used to atomically modify memory. It
4917 loads a value in memory and compares it to a given value. If they are
4918 equal, it stores a new value into the memory.
4923 There are three arguments to the '``cmpxchg``' instruction: an address
4924 to operate on, a value to compare to the value currently be at that
4925 address, and a new value to place at that address if the compared values
4926 are equal. The type of '<cmp>' must be an integer type whose bit width
4927 is a power of two greater than or equal to eight and less than or equal
4928 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4929 type, and the type of '<pointer>' must be a pointer to that type. If the
4930 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4931 to modify the number or order of execution of this ``cmpxchg`` with
4932 other :ref:`volatile operations <volatile>`.
4934 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4935 synchronizes with other atomic operations.
4937 The optional "``singlethread``" argument declares that the ``cmpxchg``
4938 is only atomic with respect to code (usually signal handlers) running in
4939 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4940 respect to all other code in the system.
4942 The pointer passed into cmpxchg must have alignment greater than or
4943 equal to the size in memory of the operand.
4948 The contents of memory at the location specified by the '``<pointer>``'
4949 operand is read and compared to '``<cmp>``'; if the read value is the
4950 equal, '``<new>``' is written. The original value at the location is
4953 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4954 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4955 atomic load with an ordering parameter determined by dropping any
4956 ``release`` part of the ``cmpxchg``'s ordering.
4961 .. code-block:: llvm
4964 %orig = atomic load i32* %ptr unordered ; yields {i32}
4968 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4969 %squared = mul i32 %cmp, %cmp
4970 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4971 %success = icmp eq i32 %cmp, %old
4972 br i1 %success, label %done, label %loop
4979 '``atomicrmw``' Instruction
4980 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4987 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4992 The '``atomicrmw``' instruction is used to atomically modify memory.
4997 There are three arguments to the '``atomicrmw``' instruction: an
4998 operation to apply, an address whose value to modify, an argument to the
4999 operation. The operation must be one of the following keywords:
5013 The type of '<value>' must be an integer type whose bit width is a power
5014 of two greater than or equal to eight and less than or equal to a
5015 target-specific size limit. The type of the '``<pointer>``' operand must
5016 be a pointer to that type. If the ``atomicrmw`` is marked as
5017 ``volatile``, then the optimizer is not allowed to modify the number or
5018 order of execution of this ``atomicrmw`` with other :ref:`volatile
5019 operations <volatile>`.
5024 The contents of memory at the location specified by the '``<pointer>``'
5025 operand are atomically read, modified, and written back. The original
5026 value at the location is returned. The modification is specified by the
5029 - xchg: ``*ptr = val``
5030 - add: ``*ptr = *ptr + val``
5031 - sub: ``*ptr = *ptr - val``
5032 - and: ``*ptr = *ptr & val``
5033 - nand: ``*ptr = ~(*ptr & val)``
5034 - or: ``*ptr = *ptr | val``
5035 - xor: ``*ptr = *ptr ^ val``
5036 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5037 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5038 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5040 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5046 .. code-block:: llvm
5048 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5050 .. _i_getelementptr:
5052 '``getelementptr``' Instruction
5053 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5060 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5061 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5062 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5067 The '``getelementptr``' instruction is used to get the address of a
5068 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5069 address calculation only and does not access memory.
5074 The first argument is always a pointer or a vector of pointers, and
5075 forms the basis of the calculation. The remaining arguments are indices
5076 that indicate which of the elements of the aggregate object are indexed.
5077 The interpretation of each index is dependent on the type being indexed
5078 into. The first index always indexes the pointer value given as the
5079 first argument, the second index indexes a value of the type pointed to
5080 (not necessarily the value directly pointed to, since the first index
5081 can be non-zero), etc. The first type indexed into must be a pointer
5082 value, subsequent types can be arrays, vectors, and structs. Note that
5083 subsequent types being indexed into can never be pointers, since that
5084 would require loading the pointer before continuing calculation.
5086 The type of each index argument depends on the type it is indexing into.
5087 When indexing into a (optionally packed) structure, only ``i32`` integer
5088 **constants** are allowed (when using a vector of indices they must all
5089 be the **same** ``i32`` integer constant). When indexing into an array,
5090 pointer or vector, integers of any width are allowed, and they are not
5091 required to be constant. These integers are treated as signed values
5094 For example, let's consider a C code fragment and how it gets compiled
5110 int *foo(struct ST *s) {
5111 return &s[1].Z.B[5][13];
5114 The LLVM code generated by Clang is:
5116 .. code-block:: llvm
5118 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5119 %struct.ST = type { i32, double, %struct.RT }
5121 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5123 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5130 In the example above, the first index is indexing into the
5131 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5132 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5133 indexes into the third element of the structure, yielding a
5134 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5135 structure. The third index indexes into the second element of the
5136 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5137 dimensions of the array are subscripted into, yielding an '``i32``'
5138 type. The '``getelementptr``' instruction returns a pointer to this
5139 element, thus computing a value of '``i32*``' type.
5141 Note that it is perfectly legal to index partially through a structure,
5142 returning a pointer to an inner element. Because of this, the LLVM code
5143 for the given testcase is equivalent to:
5145 .. code-block:: llvm
5147 define i32* @foo(%struct.ST* %s) {
5148 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5149 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5150 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5151 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5152 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5156 If the ``inbounds`` keyword is present, the result value of the
5157 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5158 pointer is not an *in bounds* address of an allocated object, or if any
5159 of the addresses that would be formed by successive addition of the
5160 offsets implied by the indices to the base address with infinitely
5161 precise signed arithmetic are not an *in bounds* address of that
5162 allocated object. The *in bounds* addresses for an allocated object are
5163 all the addresses that point into the object, plus the address one byte
5164 past the end. In cases where the base is a vector of pointers the
5165 ``inbounds`` keyword applies to each of the computations element-wise.
5167 If the ``inbounds`` keyword is not present, the offsets are added to the
5168 base address with silently-wrapping two's complement arithmetic. If the
5169 offsets have a different width from the pointer, they are sign-extended
5170 or truncated to the width of the pointer. The result value of the
5171 ``getelementptr`` may be outside the object pointed to by the base
5172 pointer. The result value may not necessarily be used to access memory
5173 though, even if it happens to point into allocated storage. See the
5174 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5177 The getelementptr instruction is often confusing. For some more insight
5178 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5183 .. code-block:: llvm
5185 ; yields [12 x i8]*:aptr
5186 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5188 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5190 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5192 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5194 In cases where the pointer argument is a vector of pointers, each index
5195 must be a vector with the same number of elements. For example:
5197 .. code-block:: llvm
5199 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5201 Conversion Operations
5202 ---------------------
5204 The instructions in this category are the conversion instructions
5205 (casting) which all take a single operand and a type. They perform
5206 various bit conversions on the operand.
5208 '``trunc .. to``' Instruction
5209 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5216 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5221 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5226 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5227 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5228 of the same number of integers. The bit size of the ``value`` must be
5229 larger than the bit size of the destination type, ``ty2``. Equal sized
5230 types are not allowed.
5235 The '``trunc``' instruction truncates the high order bits in ``value``
5236 and converts the remaining bits to ``ty2``. Since the source size must
5237 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5238 It will always truncate bits.
5243 .. code-block:: llvm
5245 %X = trunc i32 257 to i8 ; yields i8:1
5246 %Y = trunc i32 123 to i1 ; yields i1:true
5247 %Z = trunc i32 122 to i1 ; yields i1:false
5248 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5250 '``zext .. to``' Instruction
5251 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5258 <result> = zext <ty> <value> to <ty2> ; yields ty2
5263 The '``zext``' instruction zero extends its operand to type ``ty2``.
5268 The '``zext``' instruction takes a value to cast, and a type to cast it
5269 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5270 the same number of integers. The bit size of the ``value`` must be
5271 smaller than the bit size of the destination type, ``ty2``.
5276 The ``zext`` fills the high order bits of the ``value`` with zero bits
5277 until it reaches the size of the destination type, ``ty2``.
5279 When zero extending from i1, the result will always be either 0 or 1.
5284 .. code-block:: llvm
5286 %X = zext i32 257 to i64 ; yields i64:257
5287 %Y = zext i1 true to i32 ; yields i32:1
5288 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5290 '``sext .. to``' Instruction
5291 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5298 <result> = sext <ty> <value> to <ty2> ; yields ty2
5303 The '``sext``' sign extends ``value`` to the type ``ty2``.
5308 The '``sext``' instruction takes a value to cast, and a type to cast it
5309 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5310 the same number of integers. The bit size of the ``value`` must be
5311 smaller than the bit size of the destination type, ``ty2``.
5316 The '``sext``' instruction performs a sign extension by copying the sign
5317 bit (highest order bit) of the ``value`` until it reaches the bit size
5318 of the type ``ty2``.
5320 When sign extending from i1, the extension always results in -1 or 0.
5325 .. code-block:: llvm
5327 %X = sext i8 -1 to i16 ; yields i16 :65535
5328 %Y = sext i1 true to i32 ; yields i32:-1
5329 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5331 '``fptrunc .. to``' Instruction
5332 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5339 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5344 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5349 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5350 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5351 The size of ``value`` must be larger than the size of ``ty2``. This
5352 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5357 The '``fptrunc``' instruction truncates a ``value`` from a larger
5358 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5359 point <t_floating>` type. If the value cannot fit within the
5360 destination type, ``ty2``, then the results are undefined.
5365 .. code-block:: llvm
5367 %X = fptrunc double 123.0 to float ; yields float:123.0
5368 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5370 '``fpext .. to``' Instruction
5371 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5378 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5383 The '``fpext``' extends a floating point ``value`` to a larger floating
5389 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5390 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5391 to. The source type must be smaller than the destination type.
5396 The '``fpext``' instruction extends the ``value`` from a smaller
5397 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5398 point <t_floating>` type. The ``fpext`` cannot be used to make a
5399 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5400 *no-op cast* for a floating point cast.
5405 .. code-block:: llvm
5407 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5408 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5410 '``fptoui .. to``' Instruction
5411 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5418 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5423 The '``fptoui``' converts a floating point ``value`` to its unsigned
5424 integer equivalent of type ``ty2``.
5429 The '``fptoui``' instruction takes a value to cast, which must be a
5430 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5431 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5432 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5433 type with the same number of elements as ``ty``
5438 The '``fptoui``' instruction converts its :ref:`floating
5439 point <t_floating>` operand into the nearest (rounding towards zero)
5440 unsigned integer value. If the value cannot fit in ``ty2``, the results
5446 .. code-block:: llvm
5448 %X = fptoui double 123.0 to i32 ; yields i32:123
5449 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5450 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5452 '``fptosi .. to``' Instruction
5453 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5460 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5465 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5466 ``value`` to type ``ty2``.
5471 The '``fptosi``' instruction takes a value to cast, which must be a
5472 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5473 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5474 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5475 type with the same number of elements as ``ty``
5480 The '``fptosi``' instruction converts its :ref:`floating
5481 point <t_floating>` operand into the nearest (rounding towards zero)
5482 signed integer value. If the value cannot fit in ``ty2``, the results
5488 .. code-block:: llvm
5490 %X = fptosi double -123.0 to i32 ; yields i32:-123
5491 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5492 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5494 '``uitofp .. to``' Instruction
5495 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5502 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5507 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5508 and converts that value to the ``ty2`` type.
5513 The '``uitofp``' instruction takes a value to cast, which must be a
5514 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5515 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5516 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5517 type with the same number of elements as ``ty``
5522 The '``uitofp``' instruction interprets its operand as an unsigned
5523 integer quantity and converts it to the corresponding floating point
5524 value. If the value cannot fit in the floating point value, the results
5530 .. code-block:: llvm
5532 %X = uitofp i32 257 to float ; yields float:257.0
5533 %Y = uitofp i8 -1 to double ; yields double:255.0
5535 '``sitofp .. to``' Instruction
5536 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5543 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5548 The '``sitofp``' instruction regards ``value`` as a signed integer and
5549 converts that value to the ``ty2`` type.
5554 The '``sitofp``' instruction takes a value to cast, which must be a
5555 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5556 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5557 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5558 type with the same number of elements as ``ty``
5563 The '``sitofp``' instruction interprets its operand as a signed integer
5564 quantity and converts it to the corresponding floating point value. If
5565 the value cannot fit in the floating point value, the results are
5571 .. code-block:: llvm
5573 %X = sitofp i32 257 to float ; yields float:257.0
5574 %Y = sitofp i8 -1 to double ; yields double:-1.0
5578 '``ptrtoint .. to``' Instruction
5579 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5586 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5591 The '``ptrtoint``' instruction converts the pointer or a vector of
5592 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5597 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5598 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5599 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5600 a vector of integers type.
5605 The '``ptrtoint``' instruction converts ``value`` to integer type
5606 ``ty2`` by interpreting the pointer value as an integer and either
5607 truncating or zero extending that value to the size of the integer type.
5608 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5609 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5610 the same size, then nothing is done (*no-op cast*) other than a type
5616 .. code-block:: llvm
5618 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5619 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5620 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5624 '``inttoptr .. to``' Instruction
5625 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5632 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5637 The '``inttoptr``' instruction converts an integer ``value`` to a
5638 pointer type, ``ty2``.
5643 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5644 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5650 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5651 applying either a zero extension or a truncation depending on the size
5652 of the integer ``value``. If ``value`` is larger than the size of a
5653 pointer then a truncation is done. If ``value`` is smaller than the size
5654 of a pointer then a zero extension is done. If they are the same size,
5655 nothing is done (*no-op cast*).
5660 .. code-block:: llvm
5662 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5663 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5664 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5665 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5669 '``bitcast .. to``' Instruction
5670 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5677 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5682 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5688 The '``bitcast``' instruction takes a value to cast, which must be a
5689 non-aggregate first class value, and a type to cast it to, which must
5690 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5691 bit sizes of ``value`` and the destination type, ``ty2``, must be
5692 identical. If the source type is a pointer, the destination type must
5693 also be a pointer of the same size. This instruction supports bitwise
5694 conversion of vectors to integers and to vectors of other types (as
5695 long as they have the same size).
5700 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5701 is always a *no-op cast* because no bits change with this
5702 conversion. The conversion is done as if the ``value`` had been stored
5703 to memory and read back as type ``ty2``. Pointer (or vector of
5704 pointers) types may only be converted to other pointer (or vector of
5705 pointers) types with this instruction if the pointer sizes are
5706 equal. To convert pointers to other types, use the :ref:`inttoptr
5707 <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5712 .. code-block:: llvm
5714 %X = bitcast i8 255 to i8 ; yields i8 :-1
5715 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5716 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5717 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5724 The instructions in this category are the "miscellaneous" instructions,
5725 which defy better classification.
5729 '``icmp``' Instruction
5730 ^^^^^^^^^^^^^^^^^^^^^^
5737 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5742 The '``icmp``' instruction returns a boolean value or a vector of
5743 boolean values based on comparison of its two integer, integer vector,
5744 pointer, or pointer vector operands.
5749 The '``icmp``' instruction takes three operands. The first operand is
5750 the condition code indicating the kind of comparison to perform. It is
5751 not a value, just a keyword. The possible condition code are:
5754 #. ``ne``: not equal
5755 #. ``ugt``: unsigned greater than
5756 #. ``uge``: unsigned greater or equal
5757 #. ``ult``: unsigned less than
5758 #. ``ule``: unsigned less or equal
5759 #. ``sgt``: signed greater than
5760 #. ``sge``: signed greater or equal
5761 #. ``slt``: signed less than
5762 #. ``sle``: signed less or equal
5764 The remaining two arguments must be :ref:`integer <t_integer>` or
5765 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5766 must also be identical types.
5771 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5772 code given as ``cond``. The comparison performed always yields either an
5773 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5775 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5776 otherwise. No sign interpretation is necessary or performed.
5777 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5778 otherwise. No sign interpretation is necessary or performed.
5779 #. ``ugt``: interprets the operands as unsigned values and yields
5780 ``true`` if ``op1`` is greater than ``op2``.
5781 #. ``uge``: interprets the operands as unsigned values and yields
5782 ``true`` if ``op1`` is greater than or equal to ``op2``.
5783 #. ``ult``: interprets the operands as unsigned values and yields
5784 ``true`` if ``op1`` is less than ``op2``.
5785 #. ``ule``: interprets the operands as unsigned values and yields
5786 ``true`` if ``op1`` is less than or equal to ``op2``.
5787 #. ``sgt``: interprets the operands as signed values and yields ``true``
5788 if ``op1`` is greater than ``op2``.
5789 #. ``sge``: interprets the operands as signed values and yields ``true``
5790 if ``op1`` is greater than or equal to ``op2``.
5791 #. ``slt``: interprets the operands as signed values and yields ``true``
5792 if ``op1`` is less than ``op2``.
5793 #. ``sle``: interprets the operands as signed values and yields ``true``
5794 if ``op1`` is less than or equal to ``op2``.
5796 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5797 are compared as if they were integers.
5799 If the operands are integer vectors, then they are compared element by
5800 element. The result is an ``i1`` vector with the same number of elements
5801 as the values being compared. Otherwise, the result is an ``i1``.
5806 .. code-block:: llvm
5808 <result> = icmp eq i32 4, 5 ; yields: result=false
5809 <result> = icmp ne float* %X, %X ; yields: result=false
5810 <result> = icmp ult i16 4, 5 ; yields: result=true
5811 <result> = icmp sgt i16 4, 5 ; yields: result=false
5812 <result> = icmp ule i16 -4, 5 ; yields: result=false
5813 <result> = icmp sge i16 4, 5 ; yields: result=false
5815 Note that the code generator does not yet support vector types with the
5816 ``icmp`` instruction.
5820 '``fcmp``' Instruction
5821 ^^^^^^^^^^^^^^^^^^^^^^
5828 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5833 The '``fcmp``' instruction returns a boolean value or vector of boolean
5834 values based on comparison of its operands.
5836 If the operands are floating point scalars, then the result type is a
5837 boolean (:ref:`i1 <t_integer>`).
5839 If the operands are floating point vectors, then the result type is a
5840 vector of boolean with the same number of elements as the operands being
5846 The '``fcmp``' instruction takes three operands. The first operand is
5847 the condition code indicating the kind of comparison to perform. It is
5848 not a value, just a keyword. The possible condition code are:
5850 #. ``false``: no comparison, always returns false
5851 #. ``oeq``: ordered and equal
5852 #. ``ogt``: ordered and greater than
5853 #. ``oge``: ordered and greater than or equal
5854 #. ``olt``: ordered and less than
5855 #. ``ole``: ordered and less than or equal
5856 #. ``one``: ordered and not equal
5857 #. ``ord``: ordered (no nans)
5858 #. ``ueq``: unordered or equal
5859 #. ``ugt``: unordered or greater than
5860 #. ``uge``: unordered or greater than or equal
5861 #. ``ult``: unordered or less than
5862 #. ``ule``: unordered or less than or equal
5863 #. ``une``: unordered or not equal
5864 #. ``uno``: unordered (either nans)
5865 #. ``true``: no comparison, always returns true
5867 *Ordered* means that neither operand is a QNAN while *unordered* means
5868 that either operand may be a QNAN.
5870 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5871 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5872 type. They must have identical types.
5877 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5878 condition code given as ``cond``. If the operands are vectors, then the
5879 vectors are compared element by element. Each comparison performed
5880 always yields an :ref:`i1 <t_integer>` result, as follows:
5882 #. ``false``: always yields ``false``, regardless of operands.
5883 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5884 is equal to ``op2``.
5885 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5886 is greater than ``op2``.
5887 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5888 is greater than or equal to ``op2``.
5889 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5890 is less than ``op2``.
5891 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5892 is less than or equal to ``op2``.
5893 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5894 is not equal to ``op2``.
5895 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5896 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5898 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5899 greater than ``op2``.
5900 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5901 greater than or equal to ``op2``.
5902 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5904 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5905 less than or equal to ``op2``.
5906 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5907 not equal to ``op2``.
5908 #. ``uno``: yields ``true`` if either operand is a QNAN.
5909 #. ``true``: always yields ``true``, regardless of operands.
5914 .. code-block:: llvm
5916 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5917 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5918 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5919 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5921 Note that the code generator does not yet support vector types with the
5922 ``fcmp`` instruction.
5926 '``phi``' Instruction
5927 ^^^^^^^^^^^^^^^^^^^^^
5934 <result> = phi <ty> [ <val0>, <label0>], ...
5939 The '``phi``' instruction is used to implement the φ node in the SSA
5940 graph representing the function.
5945 The type of the incoming values is specified with the first type field.
5946 After this, the '``phi``' instruction takes a list of pairs as
5947 arguments, with one pair for each predecessor basic block of the current
5948 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5949 the value arguments to the PHI node. Only labels may be used as the
5952 There must be no non-phi instructions between the start of a basic block
5953 and the PHI instructions: i.e. PHI instructions must be first in a basic
5956 For the purposes of the SSA form, the use of each incoming value is
5957 deemed to occur on the edge from the corresponding predecessor block to
5958 the current block (but after any definition of an '``invoke``'
5959 instruction's return value on the same edge).
5964 At runtime, the '``phi``' instruction logically takes on the value
5965 specified by the pair corresponding to the predecessor basic block that
5966 executed just prior to the current block.
5971 .. code-block:: llvm
5973 Loop: ; Infinite loop that counts from 0 on up...
5974 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5975 %nextindvar = add i32 %indvar, 1
5980 '``select``' Instruction
5981 ^^^^^^^^^^^^^^^^^^^^^^^^
5988 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5990 selty is either i1 or {<N x i1>}
5995 The '``select``' instruction is used to choose one value based on a
5996 condition, without branching.
6001 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6002 values indicating the condition, and two values of the same :ref:`first
6003 class <t_firstclass>` type. If the val1/val2 are vectors and the
6004 condition is a scalar, then entire vectors are selected, not individual
6010 If the condition is an i1 and it evaluates to 1, the instruction returns
6011 the first value argument; otherwise, it returns the second value
6014 If the condition is a vector of i1, then the value arguments must be
6015 vectors of the same size, and the selection is done element by element.
6020 .. code-block:: llvm
6022 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6026 '``call``' Instruction
6027 ^^^^^^^^^^^^^^^^^^^^^^
6034 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6039 The '``call``' instruction represents a simple function call.
6044 This instruction requires several arguments:
6046 #. The optional "tail" marker indicates that the callee function does
6047 not access any allocas or varargs in the caller. Note that calls may
6048 be marked "tail" even if they do not occur before a
6049 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6050 function call is eligible for tail call optimization, but `might not
6051 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6052 The code generator may optimize calls marked "tail" with either 1)
6053 automatic `sibling call
6054 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6055 callee have matching signatures, or 2) forced tail call optimization
6056 when the following extra requirements are met:
6058 - Caller and callee both have the calling convention ``fastcc``.
6059 - The call is in tail position (ret immediately follows call and ret
6060 uses value of call or is void).
6061 - Option ``-tailcallopt`` is enabled, or
6062 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6063 - `Platform specific constraints are
6064 met. <CodeGenerator.html#tailcallopt>`_
6066 #. The optional "cconv" marker indicates which :ref:`calling
6067 convention <callingconv>` the call should use. If none is
6068 specified, the call defaults to using C calling conventions. The
6069 calling convention of the call must match the calling convention of
6070 the target function, or else the behavior is undefined.
6071 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6072 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6074 #. '``ty``': the type of the call instruction itself which is also the
6075 type of the return value. Functions that return no value are marked
6077 #. '``fnty``': shall be the signature of the pointer to function value
6078 being invoked. The argument types must match the types implied by
6079 this signature. This type can be omitted if the function is not
6080 varargs and if the function type does not return a pointer to a
6082 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6083 be invoked. In most cases, this is a direct function invocation, but
6084 indirect ``call``'s are just as possible, calling an arbitrary pointer
6086 #. '``function args``': argument list whose types match the function
6087 signature argument types and parameter attributes. All arguments must
6088 be of :ref:`first class <t_firstclass>` type. If the function signature
6089 indicates the function accepts a variable number of arguments, the
6090 extra arguments can be specified.
6091 #. The optional :ref:`function attributes <fnattrs>` list. Only
6092 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6093 attributes are valid here.
6098 The '``call``' instruction is used to cause control flow to transfer to
6099 a specified function, with its incoming arguments bound to the specified
6100 values. Upon a '``ret``' instruction in the called function, control
6101 flow continues with the instruction after the function call, and the
6102 return value of the function is bound to the result argument.
6107 .. code-block:: llvm
6109 %retval = call i32 @test(i32 %argc)
6110 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6111 %X = tail call i32 @foo() ; yields i32
6112 %Y = tail call fastcc i32 @foo() ; yields i32
6113 call void %foo(i8 97 signext)
6115 %struct.A = type { i32, i8 }
6116 %r = call %struct.A @foo() ; yields { 32, i8 }
6117 %gr = extractvalue %struct.A %r, 0 ; yields i32
6118 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6119 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6120 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6122 llvm treats calls to some functions with names and arguments that match
6123 the standard C99 library as being the C99 library functions, and may
6124 perform optimizations or generate code for them under that assumption.
6125 This is something we'd like to change in the future to provide better
6126 support for freestanding environments and non-C-based languages.
6130 '``va_arg``' Instruction
6131 ^^^^^^^^^^^^^^^^^^^^^^^^
6138 <resultval> = va_arg <va_list*> <arglist>, <argty>
6143 The '``va_arg``' instruction is used to access arguments passed through
6144 the "variable argument" area of a function call. It is used to implement
6145 the ``va_arg`` macro in C.
6150 This instruction takes a ``va_list*`` value and the type of the
6151 argument. It returns a value of the specified argument type and
6152 increments the ``va_list`` to point to the next argument. The actual
6153 type of ``va_list`` is target specific.
6158 The '``va_arg``' instruction loads an argument of the specified type
6159 from the specified ``va_list`` and causes the ``va_list`` to point to
6160 the next argument. For more information, see the variable argument
6161 handling :ref:`Intrinsic Functions <int_varargs>`.
6163 It is legal for this instruction to be called in a function which does
6164 not take a variable number of arguments, for example, the ``vfprintf``
6167 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6168 function <intrinsics>` because it takes a type as an argument.
6173 See the :ref:`variable argument processing <int_varargs>` section.
6175 Note that the code generator does not yet fully support va\_arg on many
6176 targets. Also, it does not currently support va\_arg with aggregate
6177 types on any target.
6181 '``landingpad``' Instruction
6182 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6189 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6190 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6192 <clause> := catch <type> <value>
6193 <clause> := filter <array constant type> <array constant>
6198 The '``landingpad``' instruction is used by `LLVM's exception handling
6199 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6200 is a landing pad --- one where the exception lands, and corresponds to the
6201 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6202 defines values supplied by the personality function (``pers_fn``) upon
6203 re-entry to the function. The ``resultval`` has the type ``resultty``.
6208 This instruction takes a ``pers_fn`` value. This is the personality
6209 function associated with the unwinding mechanism. The optional
6210 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6212 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6213 contains the global variable representing the "type" that may be caught
6214 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6215 clause takes an array constant as its argument. Use
6216 "``[0 x i8**] undef``" for a filter which cannot throw. The
6217 '``landingpad``' instruction must contain *at least* one ``clause`` or
6218 the ``cleanup`` flag.
6223 The '``landingpad``' instruction defines the values which are set by the
6224 personality function (``pers_fn``) upon re-entry to the function, and
6225 therefore the "result type" of the ``landingpad`` instruction. As with
6226 calling conventions, how the personality function results are
6227 represented in LLVM IR is target specific.
6229 The clauses are applied in order from top to bottom. If two
6230 ``landingpad`` instructions are merged together through inlining, the
6231 clauses from the calling function are appended to the list of clauses.
6232 When the call stack is being unwound due to an exception being thrown,
6233 the exception is compared against each ``clause`` in turn. If it doesn't
6234 match any of the clauses, and the ``cleanup`` flag is not set, then
6235 unwinding continues further up the call stack.
6237 The ``landingpad`` instruction has several restrictions:
6239 - A landing pad block is a basic block which is the unwind destination
6240 of an '``invoke``' instruction.
6241 - A landing pad block must have a '``landingpad``' instruction as its
6242 first non-PHI instruction.
6243 - There can be only one '``landingpad``' instruction within the landing
6245 - A basic block that is not a landing pad block may not include a
6246 '``landingpad``' instruction.
6247 - All '``landingpad``' instructions in a function must have the same
6248 personality function.
6253 .. code-block:: llvm
6255 ;; A landing pad which can catch an integer.
6256 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6258 ;; A landing pad that is a cleanup.
6259 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6261 ;; A landing pad which can catch an integer and can only throw a double.
6262 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6264 filter [1 x i8**] [@_ZTId]
6271 LLVM supports the notion of an "intrinsic function". These functions
6272 have well known names and semantics and are required to follow certain
6273 restrictions. Overall, these intrinsics represent an extension mechanism
6274 for the LLVM language that does not require changing all of the
6275 transformations in LLVM when adding to the language (or the bitcode
6276 reader/writer, the parser, etc...).
6278 Intrinsic function names must all start with an "``llvm.``" prefix. This
6279 prefix is reserved in LLVM for intrinsic names; thus, function names may
6280 not begin with this prefix. Intrinsic functions must always be external
6281 functions: you cannot define the body of intrinsic functions. Intrinsic
6282 functions may only be used in call or invoke instructions: it is illegal
6283 to take the address of an intrinsic function. Additionally, because
6284 intrinsic functions are part of the LLVM language, it is required if any
6285 are added that they be documented here.
6287 Some intrinsic functions can be overloaded, i.e., the intrinsic
6288 represents a family of functions that perform the same operation but on
6289 different data types. Because LLVM can represent over 8 million
6290 different integer types, overloading is used commonly to allow an
6291 intrinsic function to operate on any integer type. One or more of the
6292 argument types or the result type can be overloaded to accept any
6293 integer type. Argument types may also be defined as exactly matching a
6294 previous argument's type or the result type. This allows an intrinsic
6295 function which accepts multiple arguments, but needs all of them to be
6296 of the same type, to only be overloaded with respect to a single
6297 argument or the result.
6299 Overloaded intrinsics will have the names of its overloaded argument
6300 types encoded into its function name, each preceded by a period. Only
6301 those types which are overloaded result in a name suffix. Arguments
6302 whose type is matched against another type do not. For example, the
6303 ``llvm.ctpop`` function can take an integer of any width and returns an
6304 integer of exactly the same integer width. This leads to a family of
6305 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6306 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6307 overloaded, and only one type suffix is required. Because the argument's
6308 type is matched against the return type, it does not require its own
6311 To learn how to add an intrinsic function, please see the `Extending
6312 LLVM Guide <ExtendingLLVM.html>`_.
6316 Variable Argument Handling Intrinsics
6317 -------------------------------------
6319 Variable argument support is defined in LLVM with the
6320 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6321 functions. These functions are related to the similarly named macros
6322 defined in the ``<stdarg.h>`` header file.
6324 All of these functions operate on arguments that use a target-specific
6325 value type "``va_list``". The LLVM assembly language reference manual
6326 does not define what this type is, so all transformations should be
6327 prepared to handle these functions regardless of the type used.
6329 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6330 variable argument handling intrinsic functions are used.
6332 .. code-block:: llvm
6334 define i32 @test(i32 %X, ...) {
6335 ; Initialize variable argument processing
6337 %ap2 = bitcast i8** %ap to i8*
6338 call void @llvm.va_start(i8* %ap2)
6340 ; Read a single integer argument
6341 %tmp = va_arg i8** %ap, i32
6343 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6345 %aq2 = bitcast i8** %aq to i8*
6346 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6347 call void @llvm.va_end(i8* %aq2)
6349 ; Stop processing of arguments.
6350 call void @llvm.va_end(i8* %ap2)
6354 declare void @llvm.va_start(i8*)
6355 declare void @llvm.va_copy(i8*, i8*)
6356 declare void @llvm.va_end(i8*)
6360 '``llvm.va_start``' Intrinsic
6361 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6368 declare void @llvm.va_start(i8* <arglist>)
6373 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6374 subsequent use by ``va_arg``.
6379 The argument is a pointer to a ``va_list`` element to initialize.
6384 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6385 available in C. In a target-dependent way, it initializes the
6386 ``va_list`` element to which the argument points, so that the next call
6387 to ``va_arg`` will produce the first variable argument passed to the
6388 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6389 to know the last argument of the function as the compiler can figure
6392 '``llvm.va_end``' Intrinsic
6393 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6400 declare void @llvm.va_end(i8* <arglist>)
6405 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6406 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6411 The argument is a pointer to a ``va_list`` to destroy.
6416 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6417 available in C. In a target-dependent way, it destroys the ``va_list``
6418 element to which the argument points. Calls to
6419 :ref:`llvm.va_start <int_va_start>` and
6420 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6425 '``llvm.va_copy``' Intrinsic
6426 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6433 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6438 The '``llvm.va_copy``' intrinsic copies the current argument position
6439 from the source argument list to the destination argument list.
6444 The first argument is a pointer to a ``va_list`` element to initialize.
6445 The second argument is a pointer to a ``va_list`` element to copy from.
6450 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6451 available in C. In a target-dependent way, it copies the source
6452 ``va_list`` element into the destination ``va_list`` element. This
6453 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6454 arbitrarily complex and require, for example, memory allocation.
6456 Accurate Garbage Collection Intrinsics
6457 --------------------------------------
6459 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6460 (GC) requires the implementation and generation of these intrinsics.
6461 These intrinsics allow identification of :ref:`GC roots on the
6462 stack <int_gcroot>`, as well as garbage collector implementations that
6463 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6464 Front-ends for type-safe garbage collected languages should generate
6465 these intrinsics to make use of the LLVM garbage collectors. For more
6466 details, see `Accurate Garbage Collection with
6467 LLVM <GarbageCollection.html>`_.
6469 The garbage collection intrinsics only operate on objects in the generic
6470 address space (address space zero).
6474 '``llvm.gcroot``' Intrinsic
6475 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6482 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6487 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6488 the code generator, and allows some metadata to be associated with it.
6493 The first argument specifies the address of a stack object that contains
6494 the root pointer. The second pointer (which must be either a constant or
6495 a global value address) contains the meta-data to be associated with the
6501 At runtime, a call to this intrinsic stores a null pointer into the
6502 "ptrloc" location. At compile-time, the code generator generates
6503 information to allow the runtime to find the pointer at GC safe points.
6504 The '``llvm.gcroot``' intrinsic may only be used in a function which
6505 :ref:`specifies a GC algorithm <gc>`.
6509 '``llvm.gcread``' Intrinsic
6510 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6517 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6522 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6523 locations, allowing garbage collector implementations that require read
6529 The second argument is the address to read from, which should be an
6530 address allocated from the garbage collector. The first object is a
6531 pointer to the start of the referenced object, if needed by the language
6532 runtime (otherwise null).
6537 The '``llvm.gcread``' intrinsic has the same semantics as a load
6538 instruction, but may be replaced with substantially more complex code by
6539 the garbage collector runtime, as needed. The '``llvm.gcread``'
6540 intrinsic may only be used in a function which :ref:`specifies a GC
6545 '``llvm.gcwrite``' Intrinsic
6546 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6553 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6558 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6559 locations, allowing garbage collector implementations that require write
6560 barriers (such as generational or reference counting collectors).
6565 The first argument is the reference to store, the second is the start of
6566 the object to store it to, and the third is the address of the field of
6567 Obj to store to. If the runtime does not require a pointer to the
6568 object, Obj may be null.
6573 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6574 instruction, but may be replaced with substantially more complex code by
6575 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6576 intrinsic may only be used in a function which :ref:`specifies a GC
6579 Code Generator Intrinsics
6580 -------------------------
6582 These intrinsics are provided by LLVM to expose special features that
6583 may only be implemented with code generator support.
6585 '``llvm.returnaddress``' Intrinsic
6586 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6593 declare i8 *@llvm.returnaddress(i32 <level>)
6598 The '``llvm.returnaddress``' intrinsic attempts to compute a
6599 target-specific value indicating the return address of the current
6600 function or one of its callers.
6605 The argument to this intrinsic indicates which function to return the
6606 address for. Zero indicates the calling function, one indicates its
6607 caller, etc. The argument is **required** to be a constant integer
6613 The '``llvm.returnaddress``' intrinsic either returns a pointer
6614 indicating the return address of the specified call frame, or zero if it
6615 cannot be identified. The value returned by this intrinsic is likely to
6616 be incorrect or 0 for arguments other than zero, so it should only be
6617 used for debugging purposes.
6619 Note that calling this intrinsic does not prevent function inlining or
6620 other aggressive transformations, so the value returned may not be that
6621 of the obvious source-language caller.
6623 '``llvm.frameaddress``' Intrinsic
6624 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6631 declare i8* @llvm.frameaddress(i32 <level>)
6636 The '``llvm.frameaddress``' intrinsic attempts to return the
6637 target-specific frame pointer value for the specified stack frame.
6642 The argument to this intrinsic indicates which function to return the
6643 frame pointer for. Zero indicates the calling function, one indicates
6644 its caller, etc. The argument is **required** to be a constant integer
6650 The '``llvm.frameaddress``' intrinsic either returns a pointer
6651 indicating the frame address of the specified call frame, or zero if it
6652 cannot be identified. The value returned by this intrinsic is likely to
6653 be incorrect or 0 for arguments other than zero, so it should only be
6654 used for debugging purposes.
6656 Note that calling this intrinsic does not prevent function inlining or
6657 other aggressive transformations, so the value returned may not be that
6658 of the obvious source-language caller.
6662 '``llvm.stacksave``' Intrinsic
6663 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6670 declare i8* @llvm.stacksave()
6675 The '``llvm.stacksave``' intrinsic is used to remember the current state
6676 of the function stack, for use with
6677 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6678 implementing language features like scoped automatic variable sized
6684 This intrinsic returns a opaque pointer value that can be passed to
6685 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6686 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6687 ``llvm.stacksave``, it effectively restores the state of the stack to
6688 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6689 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6690 were allocated after the ``llvm.stacksave`` was executed.
6692 .. _int_stackrestore:
6694 '``llvm.stackrestore``' Intrinsic
6695 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6702 declare void @llvm.stackrestore(i8* %ptr)
6707 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6708 the function stack to the state it was in when the corresponding
6709 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6710 useful for implementing language features like scoped automatic variable
6711 sized arrays in C99.
6716 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6718 '``llvm.prefetch``' Intrinsic
6719 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6726 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6731 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6732 insert a prefetch instruction if supported; otherwise, it is a noop.
6733 Prefetches have no effect on the behavior of the program but can change
6734 its performance characteristics.
6739 ``address`` is the address to be prefetched, ``rw`` is the specifier
6740 determining if the fetch should be for a read (0) or write (1), and
6741 ``locality`` is a temporal locality specifier ranging from (0) - no
6742 locality, to (3) - extremely local keep in cache. The ``cache type``
6743 specifies whether the prefetch is performed on the data (1) or
6744 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6745 arguments must be constant integers.
6750 This intrinsic does not modify the behavior of the program. In
6751 particular, prefetches cannot trap and do not produce a value. On
6752 targets that support this intrinsic, the prefetch can provide hints to
6753 the processor cache for better performance.
6755 '``llvm.pcmarker``' Intrinsic
6756 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6763 declare void @llvm.pcmarker(i32 <id>)
6768 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6769 Counter (PC) in a region of code to simulators and other tools. The
6770 method is target specific, but it is expected that the marker will use
6771 exported symbols to transmit the PC of the marker. The marker makes no
6772 guarantees that it will remain with any specific instruction after
6773 optimizations. It is possible that the presence of a marker will inhibit
6774 optimizations. The intended use is to be inserted after optimizations to
6775 allow correlations of simulation runs.
6780 ``id`` is a numerical id identifying the marker.
6785 This intrinsic does not modify the behavior of the program. Backends
6786 that do not support this intrinsic may ignore it.
6788 '``llvm.readcyclecounter``' Intrinsic
6789 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6796 declare i64 @llvm.readcyclecounter()
6801 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6802 counter register (or similar low latency, high accuracy clocks) on those
6803 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6804 should map to RPCC. As the backing counters overflow quickly (on the
6805 order of 9 seconds on alpha), this should only be used for small
6811 When directly supported, reading the cycle counter should not modify any
6812 memory. Implementations are allowed to either return a application
6813 specific value or a system wide value. On backends without support, this
6814 is lowered to a constant 0.
6816 Note that runtime support may be conditional on the privilege-level code is
6817 running at and the host platform.
6819 Standard C Library Intrinsics
6820 -----------------------------
6822 LLVM provides intrinsics for a few important standard C library
6823 functions. These intrinsics allow source-language front-ends to pass
6824 information about the alignment of the pointer arguments to the code
6825 generator, providing opportunity for more efficient code generation.
6829 '``llvm.memcpy``' Intrinsic
6830 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6835 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6836 integer bit width and for different address spaces. Not all targets
6837 support all bit widths however.
6841 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6842 i32 <len>, i32 <align>, i1 <isvolatile>)
6843 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6844 i64 <len>, i32 <align>, i1 <isvolatile>)
6849 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6850 source location to the destination location.
6852 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6853 intrinsics do not return a value, takes extra alignment/isvolatile
6854 arguments and the pointers can be in specified address spaces.
6859 The first argument is a pointer to the destination, the second is a
6860 pointer to the source. The third argument is an integer argument
6861 specifying the number of bytes to copy, the fourth argument is the
6862 alignment of the source and destination locations, and the fifth is a
6863 boolean indicating a volatile access.
6865 If the call to this intrinsic has an alignment value that is not 0 or 1,
6866 then the caller guarantees that both the source and destination pointers
6867 are aligned to that boundary.
6869 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6870 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6871 very cleanly specified and it is unwise to depend on it.
6876 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6877 source location to the destination location, which are not allowed to
6878 overlap. It copies "len" bytes of memory over. If the argument is known
6879 to be aligned to some boundary, this can be specified as the fourth
6880 argument, otherwise it should be set to 0 or 1.
6882 '``llvm.memmove``' Intrinsic
6883 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6888 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6889 bit width and for different address space. Not all targets support all
6894 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6895 i32 <len>, i32 <align>, i1 <isvolatile>)
6896 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6897 i64 <len>, i32 <align>, i1 <isvolatile>)
6902 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6903 source location to the destination location. It is similar to the
6904 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6907 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6908 intrinsics do not return a value, takes extra alignment/isvolatile
6909 arguments and the pointers can be in specified address spaces.
6914 The first argument is a pointer to the destination, the second is a
6915 pointer to the source. The third argument is an integer argument
6916 specifying the number of bytes to copy, the fourth argument is the
6917 alignment of the source and destination locations, and the fifth is a
6918 boolean indicating a volatile access.
6920 If the call to this intrinsic has an alignment value that is not 0 or 1,
6921 then the caller guarantees that the source and destination pointers are
6922 aligned to that boundary.
6924 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6925 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6926 not very cleanly specified and it is unwise to depend on it.
6931 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6932 source location to the destination location, which may overlap. It
6933 copies "len" bytes of memory over. If the argument is known to be
6934 aligned to some boundary, this can be specified as the fourth argument,
6935 otherwise it should be set to 0 or 1.
6937 '``llvm.memset.*``' Intrinsics
6938 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6943 This is an overloaded intrinsic. You can use llvm.memset on any integer
6944 bit width and for different address spaces. However, not all targets
6945 support all bit widths.
6949 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6950 i32 <len>, i32 <align>, i1 <isvolatile>)
6951 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6952 i64 <len>, i32 <align>, i1 <isvolatile>)
6957 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6958 particular byte value.
6960 Note that, unlike the standard libc function, the ``llvm.memset``
6961 intrinsic does not return a value and takes extra alignment/volatile
6962 arguments. Also, the destination can be in an arbitrary address space.
6967 The first argument is a pointer to the destination to fill, the second
6968 is the byte value with which to fill it, the third argument is an
6969 integer argument specifying the number of bytes to fill, and the fourth
6970 argument is the known alignment of the destination location.
6972 If the call to this intrinsic has an alignment value that is not 0 or 1,
6973 then the caller guarantees that the destination pointer is aligned to
6976 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6977 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6978 very cleanly specified and it is unwise to depend on it.
6983 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6984 at the destination location. If the argument is known to be aligned to
6985 some boundary, this can be specified as the fourth argument, otherwise
6986 it should be set to 0 or 1.
6988 '``llvm.sqrt.*``' Intrinsic
6989 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6994 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6995 floating point or vector of floating point type. Not all targets support
7000 declare float @llvm.sqrt.f32(float %Val)
7001 declare double @llvm.sqrt.f64(double %Val)
7002 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7003 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7004 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7009 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7010 returning the same value as the libm '``sqrt``' functions would. Unlike
7011 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7012 negative numbers other than -0.0 (which allows for better optimization,
7013 because there is no need to worry about errno being set).
7014 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7019 The argument and return value are floating point numbers of the same
7025 This function returns the sqrt of the specified operand if it is a
7026 nonnegative floating point number.
7028 '``llvm.powi.*``' Intrinsic
7029 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7034 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7035 floating point or vector of floating point type. Not all targets support
7040 declare float @llvm.powi.f32(float %Val, i32 %power)
7041 declare double @llvm.powi.f64(double %Val, i32 %power)
7042 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7043 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7044 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7049 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7050 specified (positive or negative) power. The order of evaluation of
7051 multiplications is not defined. When a vector of floating point type is
7052 used, the second argument remains a scalar integer value.
7057 The second argument is an integer power, and the first is a value to
7058 raise to that power.
7063 This function returns the first value raised to the second power with an
7064 unspecified sequence of rounding operations.
7066 '``llvm.sin.*``' Intrinsic
7067 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7072 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7073 floating point or vector of floating point type. Not all targets support
7078 declare float @llvm.sin.f32(float %Val)
7079 declare double @llvm.sin.f64(double %Val)
7080 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7081 declare fp128 @llvm.sin.f128(fp128 %Val)
7082 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7087 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7092 The argument and return value are floating point numbers of the same
7098 This function returns the sine of the specified operand, returning the
7099 same values as the libm ``sin`` functions would, and handles error
7100 conditions in the same way.
7102 '``llvm.cos.*``' Intrinsic
7103 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7108 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7109 floating point or vector of floating point type. Not all targets support
7114 declare float @llvm.cos.f32(float %Val)
7115 declare double @llvm.cos.f64(double %Val)
7116 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7117 declare fp128 @llvm.cos.f128(fp128 %Val)
7118 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7123 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7128 The argument and return value are floating point numbers of the same
7134 This function returns the cosine of the specified operand, returning the
7135 same values as the libm ``cos`` functions would, and handles error
7136 conditions in the same way.
7138 '``llvm.pow.*``' Intrinsic
7139 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7144 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7145 floating point or vector of floating point type. Not all targets support
7150 declare float @llvm.pow.f32(float %Val, float %Power)
7151 declare double @llvm.pow.f64(double %Val, double %Power)
7152 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7153 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7154 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7159 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7160 specified (positive or negative) power.
7165 The second argument is a floating point power, and the first is a value
7166 to raise to that power.
7171 This function returns the first value raised to the second power,
7172 returning the same values as the libm ``pow`` functions would, and
7173 handles error conditions in the same way.
7175 '``llvm.exp.*``' Intrinsic
7176 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7181 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7182 floating point or vector of floating point type. Not all targets support
7187 declare float @llvm.exp.f32(float %Val)
7188 declare double @llvm.exp.f64(double %Val)
7189 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7190 declare fp128 @llvm.exp.f128(fp128 %Val)
7191 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7196 The '``llvm.exp.*``' intrinsics perform the exp function.
7201 The argument and return value are floating point numbers of the same
7207 This function returns the same values as the libm ``exp`` functions
7208 would, and handles error conditions in the same way.
7210 '``llvm.exp2.*``' Intrinsic
7211 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7216 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7217 floating point or vector of floating point type. Not all targets support
7222 declare float @llvm.exp2.f32(float %Val)
7223 declare double @llvm.exp2.f64(double %Val)
7224 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7225 declare fp128 @llvm.exp2.f128(fp128 %Val)
7226 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7231 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7236 The argument and return value are floating point numbers of the same
7242 This function returns the same values as the libm ``exp2`` functions
7243 would, and handles error conditions in the same way.
7245 '``llvm.log.*``' Intrinsic
7246 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7251 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7252 floating point or vector of floating point type. Not all targets support
7257 declare float @llvm.log.f32(float %Val)
7258 declare double @llvm.log.f64(double %Val)
7259 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7260 declare fp128 @llvm.log.f128(fp128 %Val)
7261 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7266 The '``llvm.log.*``' intrinsics perform the log function.
7271 The argument and return value are floating point numbers of the same
7277 This function returns the same values as the libm ``log`` functions
7278 would, and handles error conditions in the same way.
7280 '``llvm.log10.*``' Intrinsic
7281 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7286 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7287 floating point or vector of floating point type. Not all targets support
7292 declare float @llvm.log10.f32(float %Val)
7293 declare double @llvm.log10.f64(double %Val)
7294 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7295 declare fp128 @llvm.log10.f128(fp128 %Val)
7296 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7301 The '``llvm.log10.*``' intrinsics perform the log10 function.
7306 The argument and return value are floating point numbers of the same
7312 This function returns the same values as the libm ``log10`` functions
7313 would, and handles error conditions in the same way.
7315 '``llvm.log2.*``' Intrinsic
7316 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7321 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7322 floating point or vector of floating point type. Not all targets support
7327 declare float @llvm.log2.f32(float %Val)
7328 declare double @llvm.log2.f64(double %Val)
7329 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7330 declare fp128 @llvm.log2.f128(fp128 %Val)
7331 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7336 The '``llvm.log2.*``' intrinsics perform the log2 function.
7341 The argument and return value are floating point numbers of the same
7347 This function returns the same values as the libm ``log2`` functions
7348 would, and handles error conditions in the same way.
7350 '``llvm.fma.*``' Intrinsic
7351 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7356 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7357 floating point or vector of floating point type. Not all targets support
7362 declare float @llvm.fma.f32(float %a, float %b, float %c)
7363 declare double @llvm.fma.f64(double %a, double %b, double %c)
7364 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7365 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7366 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7371 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7377 The argument and return value are floating point numbers of the same
7383 This function returns the same values as the libm ``fma`` functions
7386 '``llvm.fabs.*``' Intrinsic
7387 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7392 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7393 floating point or vector of floating point type. Not all targets support
7398 declare float @llvm.fabs.f32(float %Val)
7399 declare double @llvm.fabs.f64(double %Val)
7400 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7401 declare fp128 @llvm.fabs.f128(fp128 %Val)
7402 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7407 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7413 The argument and return value are floating point numbers of the same
7419 This function returns the same values as the libm ``fabs`` functions
7420 would, and handles error conditions in the same way.
7422 '``llvm.copysign.*``' Intrinsic
7423 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7428 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7429 floating point or vector of floating point type. Not all targets support
7434 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7435 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7436 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7437 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7438 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7443 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7444 first operand and the sign of the second operand.
7449 The arguments and return value are floating point numbers of the same
7455 This function returns the same values as the libm ``copysign``
7456 functions would, and handles error conditions in the same way.
7458 '``llvm.floor.*``' Intrinsic
7459 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7464 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7465 floating point or vector of floating point type. Not all targets support
7470 declare float @llvm.floor.f32(float %Val)
7471 declare double @llvm.floor.f64(double %Val)
7472 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7473 declare fp128 @llvm.floor.f128(fp128 %Val)
7474 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7479 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7484 The argument and return value are floating point numbers of the same
7490 This function returns the same values as the libm ``floor`` functions
7491 would, and handles error conditions in the same way.
7493 '``llvm.ceil.*``' Intrinsic
7494 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7499 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7500 floating point or vector of floating point type. Not all targets support
7505 declare float @llvm.ceil.f32(float %Val)
7506 declare double @llvm.ceil.f64(double %Val)
7507 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7508 declare fp128 @llvm.ceil.f128(fp128 %Val)
7509 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7514 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7519 The argument and return value are floating point numbers of the same
7525 This function returns the same values as the libm ``ceil`` functions
7526 would, and handles error conditions in the same way.
7528 '``llvm.trunc.*``' Intrinsic
7529 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7534 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7535 floating point or vector of floating point type. Not all targets support
7540 declare float @llvm.trunc.f32(float %Val)
7541 declare double @llvm.trunc.f64(double %Val)
7542 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7543 declare fp128 @llvm.trunc.f128(fp128 %Val)
7544 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7549 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7550 nearest integer not larger in magnitude than the operand.
7555 The argument and return value are floating point numbers of the same
7561 This function returns the same values as the libm ``trunc`` functions
7562 would, and handles error conditions in the same way.
7564 '``llvm.rint.*``' Intrinsic
7565 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7570 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7571 floating point or vector of floating point type. Not all targets support
7576 declare float @llvm.rint.f32(float %Val)
7577 declare double @llvm.rint.f64(double %Val)
7578 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7579 declare fp128 @llvm.rint.f128(fp128 %Val)
7580 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7585 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7586 nearest integer. It may raise an inexact floating-point exception if the
7587 operand isn't an integer.
7592 The argument and return value are floating point numbers of the same
7598 This function returns the same values as the libm ``rint`` functions
7599 would, and handles error conditions in the same way.
7601 '``llvm.nearbyint.*``' Intrinsic
7602 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7607 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7608 floating point or vector of floating point type. Not all targets support
7613 declare float @llvm.nearbyint.f32(float %Val)
7614 declare double @llvm.nearbyint.f64(double %Val)
7615 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7616 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7617 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7622 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7628 The argument and return value are floating point numbers of the same
7634 This function returns the same values as the libm ``nearbyint``
7635 functions would, and handles error conditions in the same way.
7637 '``llvm.round.*``' Intrinsic
7638 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7643 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7644 floating point or vector of floating point type. Not all targets support
7649 declare float @llvm.round.f32(float %Val)
7650 declare double @llvm.round.f64(double %Val)
7651 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7652 declare fp128 @llvm.round.f128(fp128 %Val)
7653 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7658 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7664 The argument and return value are floating point numbers of the same
7670 This function returns the same values as the libm ``round``
7671 functions would, and handles error conditions in the same way.
7673 Bit Manipulation Intrinsics
7674 ---------------------------
7676 LLVM provides intrinsics for a few important bit manipulation
7677 operations. These allow efficient code generation for some algorithms.
7679 '``llvm.bswap.*``' Intrinsics
7680 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7685 This is an overloaded intrinsic function. You can use bswap on any
7686 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7690 declare i16 @llvm.bswap.i16(i16 <id>)
7691 declare i32 @llvm.bswap.i32(i32 <id>)
7692 declare i64 @llvm.bswap.i64(i64 <id>)
7697 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7698 values with an even number of bytes (positive multiple of 16 bits).
7699 These are useful for performing operations on data that is not in the
7700 target's native byte order.
7705 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7706 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7707 intrinsic returns an i32 value that has the four bytes of the input i32
7708 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7709 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7710 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7711 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7714 '``llvm.ctpop.*``' Intrinsic
7715 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7720 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7721 bit width, or on any vector with integer elements. Not all targets
7722 support all bit widths or vector types, however.
7726 declare i8 @llvm.ctpop.i8(i8 <src>)
7727 declare i16 @llvm.ctpop.i16(i16 <src>)
7728 declare i32 @llvm.ctpop.i32(i32 <src>)
7729 declare i64 @llvm.ctpop.i64(i64 <src>)
7730 declare i256 @llvm.ctpop.i256(i256 <src>)
7731 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7736 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7742 The only argument is the value to be counted. The argument may be of any
7743 integer type, or a vector with integer elements. The return type must
7744 match the argument type.
7749 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7750 each element of a vector.
7752 '``llvm.ctlz.*``' Intrinsic
7753 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7758 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7759 integer bit width, or any vector whose elements are integers. Not all
7760 targets support all bit widths or vector types, however.
7764 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7765 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7766 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7767 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7768 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7769 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7774 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7775 leading zeros in a variable.
7780 The first argument is the value to be counted. This argument may be of
7781 any integer type, or a vectory with integer element type. The return
7782 type must match the first argument type.
7784 The second argument must be a constant and is a flag to indicate whether
7785 the intrinsic should ensure that a zero as the first argument produces a
7786 defined result. Historically some architectures did not provide a
7787 defined result for zero values as efficiently, and many algorithms are
7788 now predicated on avoiding zero-value inputs.
7793 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7794 zeros in a variable, or within each element of the vector. If
7795 ``src == 0`` then the result is the size in bits of the type of ``src``
7796 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7797 ``llvm.ctlz(i32 2) = 30``.
7799 '``llvm.cttz.*``' Intrinsic
7800 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7805 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7806 integer bit width, or any vector of integer elements. Not all targets
7807 support all bit widths or vector types, however.
7811 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7812 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7813 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7814 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7815 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7816 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7821 The '``llvm.cttz``' family of intrinsic functions counts the number of
7827 The first argument is the value to be counted. This argument may be of
7828 any integer type, or a vectory with integer element type. The return
7829 type must match the first argument type.
7831 The second argument must be a constant and is a flag to indicate whether
7832 the intrinsic should ensure that a zero as the first argument produces a
7833 defined result. Historically some architectures did not provide a
7834 defined result for zero values as efficiently, and many algorithms are
7835 now predicated on avoiding zero-value inputs.
7840 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7841 zeros in a variable, or within each element of a vector. If ``src == 0``
7842 then the result is the size in bits of the type of ``src`` if
7843 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7844 ``llvm.cttz(2) = 1``.
7846 Arithmetic with Overflow Intrinsics
7847 -----------------------------------
7849 LLVM provides intrinsics for some arithmetic with overflow operations.
7851 '``llvm.sadd.with.overflow.*``' Intrinsics
7852 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7857 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7858 on any integer bit width.
7862 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7863 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7864 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7869 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7870 a signed addition of the two arguments, and indicate whether an overflow
7871 occurred during the signed summation.
7876 The arguments (%a and %b) and the first element of the result structure
7877 may be of integer types of any bit width, but they must have the same
7878 bit width. The second element of the result structure must be of type
7879 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7885 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7886 a signed addition of the two variables. They return a structure --- the
7887 first element of which is the signed summation, and the second element
7888 of which is a bit specifying if the signed summation resulted in an
7894 .. code-block:: llvm
7896 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7897 %sum = extractvalue {i32, i1} %res, 0
7898 %obit = extractvalue {i32, i1} %res, 1
7899 br i1 %obit, label %overflow, label %normal
7901 '``llvm.uadd.with.overflow.*``' Intrinsics
7902 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7907 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7908 on any integer bit width.
7912 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7913 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7914 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7919 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7920 an unsigned addition of the two arguments, and indicate whether a carry
7921 occurred during the unsigned summation.
7926 The arguments (%a and %b) and the first element of the result structure
7927 may be of integer types of any bit width, but they must have the same
7928 bit width. The second element of the result structure must be of type
7929 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7935 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7936 an unsigned addition of the two arguments. They return a structure --- the
7937 first element of which is the sum, and the second element of which is a
7938 bit specifying if the unsigned summation resulted in a carry.
7943 .. code-block:: llvm
7945 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7946 %sum = extractvalue {i32, i1} %res, 0
7947 %obit = extractvalue {i32, i1} %res, 1
7948 br i1 %obit, label %carry, label %normal
7950 '``llvm.ssub.with.overflow.*``' Intrinsics
7951 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7956 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7957 on any integer bit width.
7961 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7962 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7963 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7968 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7969 a signed subtraction of the two arguments, and indicate whether an
7970 overflow occurred during the signed subtraction.
7975 The arguments (%a and %b) and the first element of the result structure
7976 may be of integer types of any bit width, but they must have the same
7977 bit width. The second element of the result structure must be of type
7978 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7984 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7985 a signed subtraction of the two arguments. They return a structure --- the
7986 first element of which is the subtraction, and the second element of
7987 which is a bit specifying if the signed subtraction resulted in an
7993 .. code-block:: llvm
7995 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7996 %sum = extractvalue {i32, i1} %res, 0
7997 %obit = extractvalue {i32, i1} %res, 1
7998 br i1 %obit, label %overflow, label %normal
8000 '``llvm.usub.with.overflow.*``' Intrinsics
8001 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8006 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8007 on any integer bit width.
8011 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8012 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8013 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8018 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8019 an unsigned subtraction of the two arguments, and indicate whether an
8020 overflow occurred during the unsigned subtraction.
8025 The arguments (%a and %b) and the first element of the result structure
8026 may be of integer types of any bit width, but they must have the same
8027 bit width. The second element of the result structure must be of type
8028 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8034 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8035 an unsigned subtraction of the two arguments. They return a structure ---
8036 the first element of which is the subtraction, and the second element of
8037 which is a bit specifying if the unsigned subtraction resulted in an
8043 .. code-block:: llvm
8045 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8046 %sum = extractvalue {i32, i1} %res, 0
8047 %obit = extractvalue {i32, i1} %res, 1
8048 br i1 %obit, label %overflow, label %normal
8050 '``llvm.smul.with.overflow.*``' Intrinsics
8051 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8056 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8057 on any integer bit width.
8061 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8062 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8063 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8068 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8069 a signed multiplication of the two arguments, and indicate whether an
8070 overflow occurred during the signed multiplication.
8075 The arguments (%a and %b) and the first element of the result structure
8076 may be of integer types of any bit width, but they must have the same
8077 bit width. The second element of the result structure must be of type
8078 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8084 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8085 a signed multiplication of the two arguments. They return a structure ---
8086 the first element of which is the multiplication, and the second element
8087 of which is a bit specifying if the signed multiplication resulted in an
8093 .. code-block:: llvm
8095 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8096 %sum = extractvalue {i32, i1} %res, 0
8097 %obit = extractvalue {i32, i1} %res, 1
8098 br i1 %obit, label %overflow, label %normal
8100 '``llvm.umul.with.overflow.*``' Intrinsics
8101 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8106 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8107 on any integer bit width.
8111 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8112 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8113 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8118 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8119 a unsigned multiplication of the two arguments, and indicate whether an
8120 overflow occurred during the unsigned multiplication.
8125 The arguments (%a and %b) and the first element of the result structure
8126 may be of integer types of any bit width, but they must have the same
8127 bit width. The second element of the result structure must be of type
8128 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8134 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8135 an unsigned multiplication of the two arguments. They return a structure ---
8136 the first element of which is the multiplication, and the second
8137 element of which is a bit specifying if the unsigned multiplication
8138 resulted in an overflow.
8143 .. code-block:: llvm
8145 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8146 %sum = extractvalue {i32, i1} %res, 0
8147 %obit = extractvalue {i32, i1} %res, 1
8148 br i1 %obit, label %overflow, label %normal
8150 Specialised Arithmetic Intrinsics
8151 ---------------------------------
8153 '``llvm.fmuladd.*``' Intrinsic
8154 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8161 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8162 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8167 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8168 expressions that can be fused if the code generator determines that (a) the
8169 target instruction set has support for a fused operation, and (b) that the
8170 fused operation is more efficient than the equivalent, separate pair of mul
8171 and add instructions.
8176 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8177 multiplicands, a and b, and an addend c.
8186 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8188 is equivalent to the expression a \* b + c, except that rounding will
8189 not be performed between the multiplication and addition steps if the
8190 code generator fuses the operations. Fusion is not guaranteed, even if
8191 the target platform supports it. If a fused multiply-add is required the
8192 corresponding llvm.fma.\* intrinsic function should be used instead.
8197 .. code-block:: llvm
8199 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8201 Half Precision Floating Point Intrinsics
8202 ----------------------------------------
8204 For most target platforms, half precision floating point is a
8205 storage-only format. This means that it is a dense encoding (in memory)
8206 but does not support computation in the format.
8208 This means that code must first load the half-precision floating point
8209 value as an i16, then convert it to float with
8210 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8211 then be performed on the float value (including extending to double
8212 etc). To store the value back to memory, it is first converted to float
8213 if needed, then converted to i16 with
8214 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8217 .. _int_convert_to_fp16:
8219 '``llvm.convert.to.fp16``' Intrinsic
8220 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8227 declare i16 @llvm.convert.to.fp16(f32 %a)
8232 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8233 from single precision floating point format to half precision floating
8239 The intrinsic function contains single argument - the value to be
8245 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8246 from single precision floating point format to half precision floating
8247 point format. The return value is an ``i16`` which contains the
8253 .. code-block:: llvm
8255 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8256 store i16 %res, i16* @x, align 2
8258 .. _int_convert_from_fp16:
8260 '``llvm.convert.from.fp16``' Intrinsic
8261 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8268 declare f32 @llvm.convert.from.fp16(i16 %a)
8273 The '``llvm.convert.from.fp16``' intrinsic function performs a
8274 conversion from half precision floating point format to single precision
8275 floating point format.
8280 The intrinsic function contains single argument - the value to be
8286 The '``llvm.convert.from.fp16``' intrinsic function performs a
8287 conversion from half single precision floating point format to single
8288 precision floating point format. The input half-float value is
8289 represented by an ``i16`` value.
8294 .. code-block:: llvm
8296 %a = load i16* @x, align 2
8297 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8302 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8303 prefix), are described in the `LLVM Source Level
8304 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8307 Exception Handling Intrinsics
8308 -----------------------------
8310 The LLVM exception handling intrinsics (which all start with
8311 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8312 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8316 Trampoline Intrinsics
8317 ---------------------
8319 These intrinsics make it possible to excise one parameter, marked with
8320 the :ref:`nest <nest>` attribute, from a function. The result is a
8321 callable function pointer lacking the nest parameter - the caller does
8322 not need to provide a value for it. Instead, the value to use is stored
8323 in advance in a "trampoline", a block of memory usually allocated on the
8324 stack, which also contains code to splice the nest value into the
8325 argument list. This is used to implement the GCC nested function address
8328 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8329 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8330 It can be created as follows:
8332 .. code-block:: llvm
8334 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8335 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8336 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8337 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8338 %fp = bitcast i8* %p to i32 (i32, i32)*
8340 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8341 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8345 '``llvm.init.trampoline``' Intrinsic
8346 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8353 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8358 This fills the memory pointed to by ``tramp`` with executable code,
8359 turning it into a trampoline.
8364 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8365 pointers. The ``tramp`` argument must point to a sufficiently large and
8366 sufficiently aligned block of memory; this memory is written to by the
8367 intrinsic. Note that the size and the alignment are target-specific -
8368 LLVM currently provides no portable way of determining them, so a
8369 front-end that generates this intrinsic needs to have some
8370 target-specific knowledge. The ``func`` argument must hold a function
8371 bitcast to an ``i8*``.
8376 The block of memory pointed to by ``tramp`` is filled with target
8377 dependent code, turning it into a function. Then ``tramp`` needs to be
8378 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8379 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8380 function's signature is the same as that of ``func`` with any arguments
8381 marked with the ``nest`` attribute removed. At most one such ``nest``
8382 argument is allowed, and it must be of pointer type. Calling the new
8383 function is equivalent to calling ``func`` with the same argument list,
8384 but with ``nval`` used for the missing ``nest`` argument. If, after
8385 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8386 modified, then the effect of any later call to the returned function
8387 pointer is undefined.
8391 '``llvm.adjust.trampoline``' Intrinsic
8392 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8399 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8404 This performs any required machine-specific adjustment to the address of
8405 a trampoline (passed as ``tramp``).
8410 ``tramp`` must point to a block of memory which already has trampoline
8411 code filled in by a previous call to
8412 :ref:`llvm.init.trampoline <int_it>`.
8417 On some architectures the address of the code to be executed needs to be
8418 different to the address where the trampoline is actually stored. This
8419 intrinsic returns the executable address corresponding to ``tramp``
8420 after performing the required machine specific adjustments. The pointer
8421 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8426 This class of intrinsics exists to information about the lifetime of
8427 memory objects and ranges where variables are immutable.
8429 '``llvm.lifetime.start``' Intrinsic
8430 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8437 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8442 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8448 The first argument is a constant integer representing the size of the
8449 object, or -1 if it is variable sized. The second argument is a pointer
8455 This intrinsic indicates that before this point in the code, the value
8456 of the memory pointed to by ``ptr`` is dead. This means that it is known
8457 to never be used and has an undefined value. A load from the pointer
8458 that precedes this intrinsic can be replaced with ``'undef'``.
8460 '``llvm.lifetime.end``' Intrinsic
8461 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8468 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8473 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8479 The first argument is a constant integer representing the size of the
8480 object, or -1 if it is variable sized. The second argument is a pointer
8486 This intrinsic indicates that after this point in the code, the value of
8487 the memory pointed to by ``ptr`` is dead. This means that it is known to
8488 never be used and has an undefined value. Any stores into the memory
8489 object following this intrinsic may be removed as dead.
8491 '``llvm.invariant.start``' Intrinsic
8492 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8499 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8504 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8505 a memory object will not change.
8510 The first argument is a constant integer representing the size of the
8511 object, or -1 if it is variable sized. The second argument is a pointer
8517 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8518 the return value, the referenced memory location is constant and
8521 '``llvm.invariant.end``' Intrinsic
8522 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8529 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8534 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8535 memory object are mutable.
8540 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8541 The second argument is a constant integer representing the size of the
8542 object, or -1 if it is variable sized and the third argument is a
8543 pointer to the object.
8548 This intrinsic indicates that the memory is mutable again.
8553 This class of intrinsics is designed to be generic and has no specific
8556 '``llvm.var.annotation``' Intrinsic
8557 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8564 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8569 The '``llvm.var.annotation``' intrinsic.
8574 The first argument is a pointer to a value, the second is a pointer to a
8575 global string, the third is a pointer to a global string which is the
8576 source file name, and the last argument is the line number.
8581 This intrinsic allows annotation of local variables with arbitrary
8582 strings. This can be useful for special purpose optimizations that want
8583 to look for these annotations. These have no other defined use; they are
8584 ignored by code generation and optimization.
8586 '``llvm.ptr.annotation.*``' Intrinsic
8587 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8592 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8593 pointer to an integer of any width. *NOTE* you must specify an address space for
8594 the pointer. The identifier for the default address space is the integer
8599 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8600 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8601 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8602 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8603 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8608 The '``llvm.ptr.annotation``' intrinsic.
8613 The first argument is a pointer to an integer value of arbitrary bitwidth
8614 (result of some expression), the second is a pointer to a global string, the
8615 third is a pointer to a global string which is the source file name, and the
8616 last argument is the line number. It returns the value of the first argument.
8621 This intrinsic allows annotation of a pointer to an integer with arbitrary
8622 strings. This can be useful for special purpose optimizations that want to look
8623 for these annotations. These have no other defined use; they are ignored by code
8624 generation and optimization.
8626 '``llvm.annotation.*``' Intrinsic
8627 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8632 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8633 any integer bit width.
8637 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8638 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8639 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8640 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8641 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8646 The '``llvm.annotation``' intrinsic.
8651 The first argument is an integer value (result of some expression), the
8652 second is a pointer to a global string, the third is a pointer to a
8653 global string which is the source file name, and the last argument is
8654 the line number. It returns the value of the first argument.
8659 This intrinsic allows annotations to be put on arbitrary expressions
8660 with arbitrary strings. This can be useful for special purpose
8661 optimizations that want to look for these annotations. These have no
8662 other defined use; they are ignored by code generation and optimization.
8664 '``llvm.trap``' Intrinsic
8665 ^^^^^^^^^^^^^^^^^^^^^^^^^
8672 declare void @llvm.trap() noreturn nounwind
8677 The '``llvm.trap``' intrinsic.
8687 This intrinsic is lowered to the target dependent trap instruction. If
8688 the target does not have a trap instruction, this intrinsic will be
8689 lowered to a call of the ``abort()`` function.
8691 '``llvm.debugtrap``' Intrinsic
8692 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8699 declare void @llvm.debugtrap() nounwind
8704 The '``llvm.debugtrap``' intrinsic.
8714 This intrinsic is lowered to code which is intended to cause an
8715 execution trap with the intention of requesting the attention of a
8718 '``llvm.stackprotector``' Intrinsic
8719 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8726 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8731 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8732 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8733 is placed on the stack before local variables.
8738 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8739 The first argument is the value loaded from the stack guard
8740 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8741 enough space to hold the value of the guard.
8746 This intrinsic causes the prologue/epilogue inserter to force the position of
8747 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8748 to ensure that if a local variable on the stack is overwritten, it will destroy
8749 the value of the guard. When the function exits, the guard on the stack is
8750 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8751 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8752 calling the ``__stack_chk_fail()`` function.
8754 '``llvm.stackprotectorcheck``' Intrinsic
8755 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8762 declare void @llvm.stackprotectorcheck(i8** <guard>)
8767 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8768 created stack protector and if they are not equal calls the
8769 ``__stack_chk_fail()`` function.
8774 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8775 the variable ``@__stack_chk_guard``.
8780 This intrinsic is provided to perform the stack protector check by comparing
8781 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8782 values do not match call the ``__stack_chk_fail()`` function.
8784 The reason to provide this as an IR level intrinsic instead of implementing it
8785 via other IR operations is that in order to perform this operation at the IR
8786 level without an intrinsic, one would need to create additional basic blocks to
8787 handle the success/failure cases. This makes it difficult to stop the stack
8788 protector check from disrupting sibling tail calls in Codegen. With this
8789 intrinsic, we are able to generate the stack protector basic blocks late in
8790 codegen after the tail call decision has occured.
8792 '``llvm.objectsize``' Intrinsic
8793 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8800 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8801 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8806 The ``llvm.objectsize`` intrinsic is designed to provide information to
8807 the optimizers to determine at compile time whether a) an operation
8808 (like memcpy) will overflow a buffer that corresponds to an object, or
8809 b) that a runtime check for overflow isn't necessary. An object in this
8810 context means an allocation of a specific class, structure, array, or
8816 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8817 argument is a pointer to or into the ``object``. The second argument is
8818 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8819 or -1 (if false) when the object size is unknown. The second argument
8820 only accepts constants.
8825 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8826 the size of the object concerned. If the size cannot be determined at
8827 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8828 on the ``min`` argument).
8830 '``llvm.expect``' Intrinsic
8831 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8838 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8839 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8844 The ``llvm.expect`` intrinsic provides information about expected (the
8845 most probable) value of ``val``, which can be used by optimizers.
8850 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8851 a value. The second argument is an expected value, this needs to be a
8852 constant value, variables are not allowed.
8857 This intrinsic is lowered to the ``val``.
8859 '``llvm.donothing``' Intrinsic
8860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8867 declare void @llvm.donothing() nounwind readnone
8872 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8873 only intrinsic that can be called with an invoke instruction.
8883 This intrinsic does nothing, and it's removed by optimizers and ignored