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
132 It also shows a convention that we follow in this document. When
133 demonstrating instructions, we will follow an instruction with a comment
134 that defines the type and name of value produced.
142 LLVM programs are composed of ``Module``'s, each of which is a
143 translation unit of the input programs. Each module consists of
144 functions, global variables, and symbol table entries. Modules may be
145 combined together with the LLVM linker, which merges function (and
146 global variable) definitions, resolves forward declarations, and merges
147 symbol table entries. Here is an example of the "hello world" module:
151 ; Declare the string constant as a global constant.
152 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
154 ; External declaration of the puts function
155 declare i32 @puts(i8* nocapture) nounwind
157 ; Definition of main function
158 define i32 @main() { ; i32()*
159 ; Convert [13 x i8]* to i8 *...
160 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
162 ; Call puts function to write out the string to stdout.
163 call i32 @puts(i8* %cast210)
168 !1 = metadata !{i32 42}
171 This example is made up of a :ref:`global variable <globalvars>` named
172 "``.str``", an external declaration of the "``puts``" function, a
173 :ref:`function definition <functionstructure>` for "``main``" and
174 :ref:`named metadata <namedmetadatastructure>` "``foo``".
176 In general, a module is made up of a list of global values (where both
177 functions and global variables are global values). Global values are
178 represented by a pointer to a memory location (in this case, a pointer
179 to an array of char, and a pointer to a function), and have one of the
180 following :ref:`linkage types <linkage>`.
187 All Global Variables and Functions have one of the following types of
191 Global values with "``private``" linkage are only directly
192 accessible by objects in the current module. In particular, linking
193 code into a module with an private global value may cause the
194 private to be renamed as necessary to avoid collisions. Because the
195 symbol is private to the module, all references can be updated. This
196 doesn't show up in any symbol table in the object file.
198 Similar to ``private``, but the symbol is passed through the
199 assembler and evaluated by the linker. Unlike normal strong symbols,
200 they are removed by the linker from the final linked image
201 (executable or dynamic library).
202 ``linker_private_weak``
203 Similar to "``linker_private``", but the symbol is weak. Note that
204 ``linker_private_weak`` symbols are subject to coalescing by the
205 linker. The symbols are removed by the linker from the final linked
206 image (executable or dynamic library).
208 Similar to private, but the value shows as a local symbol
209 (``STB_LOCAL`` in the case of ELF) in the object file. This
210 corresponds to the notion of the '``static``' keyword in C.
211 ``available_externally``
212 Globals with "``available_externally``" linkage are never emitted
213 into the object file corresponding to the LLVM module. They exist to
214 allow inlining and other optimizations to take place given knowledge
215 of the definition of the global, which is known to be somewhere
216 outside the module. Globals with ``available_externally`` linkage
217 are allowed to be discarded at will, and are otherwise the same as
218 ``linkonce_odr``. This linkage type is only allowed on definitions,
221 Globals with "``linkonce``" linkage are merged with other globals of
222 the same name when linkage occurs. This can be used to implement
223 some forms of inline functions, templates, or other code which must
224 be generated in each translation unit that uses it, but where the
225 body may be overridden with a more definitive definition later.
226 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
227 that ``linkonce`` linkage does not actually allow the optimizer to
228 inline the body of this function into callers because it doesn't
229 know if this definition of the function is the definitive definition
230 within the program or whether it will be overridden by a stronger
231 definition. To enable inlining and other optimizations, use
232 "``linkonce_odr``" linkage.
234 "``weak``" linkage has the same merging semantics as ``linkonce``
235 linkage, except that unreferenced globals with ``weak`` linkage may
236 not be discarded. This is used for globals that are declared "weak"
239 "``common``" linkage is most similar to "``weak``" linkage, but they
240 are used for tentative definitions in C, such as "``int X;``" at
241 global scope. Symbols with "``common``" linkage are merged in the
242 same way as ``weak symbols``, and they may not be deleted if
243 unreferenced. ``common`` symbols may not have an explicit section,
244 must have a zero initializer, and may not be marked
245 ':ref:`constant <globalvars>`'. Functions and aliases may not have
248 .. _linkage_appending:
251 "``appending``" linkage may only be applied to global variables of
252 pointer to array type. When two global variables with appending
253 linkage are linked together, the two global arrays are appended
254 together. This is the LLVM, typesafe, equivalent of having the
255 system linker append together "sections" with identical names when
258 The semantics of this linkage follow the ELF object file model: the
259 symbol is weak until linked, if not linked, the symbol becomes null
260 instead of being an undefined reference.
261 ``linkonce_odr``, ``weak_odr``
262 Some languages allow differing globals to be merged, such as two
263 functions with different semantics. Other languages, such as
264 ``C++``, ensure that only equivalent globals are ever merged (the
265 "one definition rule" --- "ODR"). Such languages can use the
266 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
267 global will only be merged with equivalent globals. These linkage
268 types are otherwise the same as their non-``odr`` versions.
269 ``linkonce_odr_auto_hide``
270 Similar to "``linkonce_odr``", but nothing in the translation unit
271 takes the address of this definition. For instance, functions that
272 had an inline definition, but the compiler decided not to inline it.
273 ``linkonce_odr_auto_hide`` may have only ``default`` visibility. The
274 symbols are removed by the linker from the final linked image
275 (executable or dynamic library).
277 If none of the above identifiers are used, the global is externally
278 visible, meaning that it participates in linkage and can be used to
279 resolve external symbol references.
281 The next two types of linkage are targeted for Microsoft Windows
282 platform only. They are designed to support importing (exporting)
283 symbols from (to) DLLs (Dynamic Link Libraries).
286 "``dllimport``" linkage causes the compiler to reference a function
287 or variable via a global pointer to a pointer that is set up by the
288 DLL exporting the symbol. On Microsoft Windows targets, the pointer
289 name is formed by combining ``__imp_`` and the function or variable
292 "``dllexport``" linkage causes the compiler to provide a global
293 pointer to a pointer in a DLL, so that it can be referenced with the
294 ``dllimport`` attribute. On Microsoft Windows targets, the pointer
295 name is formed by combining ``__imp_`` and the function or variable
298 For example, since the "``.LC0``" variable is defined to be internal, if
299 another module defined a "``.LC0``" variable and was linked with this
300 one, one of the two would be renamed, preventing a collision. Since
301 "``main``" and "``puts``" are external (i.e., lacking any linkage
302 declarations), they are accessible outside of the current module.
304 It is illegal for a function *declaration* to have any linkage type
305 other than ``external``, ``dllimport`` or ``extern_weak``.
307 Aliases can have only ``external``, ``internal``, ``weak`` or
308 ``weak_odr`` linkages.
315 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
316 :ref:`invokes <i_invoke>` can all have an optional calling convention
317 specified for the call. The calling convention of any pair of dynamic
318 caller/callee must match, or the behavior of the program is undefined.
319 The following calling conventions are supported by LLVM, and more may be
322 "``ccc``" - The C calling convention
323 This calling convention (the default if no other calling convention
324 is specified) matches the target C calling conventions. This calling
325 convention supports varargs function calls and tolerates some
326 mismatch in the declared prototype and implemented declaration of
327 the function (as does normal C).
328 "``fastcc``" - The fast calling convention
329 This calling convention attempts to make calls as fast as possible
330 (e.g. by passing things in registers). This calling convention
331 allows the target to use whatever tricks it wants to produce fast
332 code for the target, without having to conform to an externally
333 specified ABI (Application Binary Interface). `Tail calls can only
334 be optimized when this, the GHC or the HiPE convention is
335 used. <CodeGenerator.html#id80>`_ This calling convention does not
336 support varargs and requires the prototype of all callees to exactly
337 match the prototype of the function definition.
338 "``coldcc``" - The cold calling convention
339 This calling convention attempts to make code in the caller as
340 efficient as possible under the assumption that the call is not
341 commonly executed. As such, these calls often preserve all registers
342 so that the call does not break any live ranges in the caller side.
343 This calling convention does not support varargs and requires the
344 prototype of all callees to exactly match the prototype of the
346 "``cc 10``" - GHC convention
347 This calling convention has been implemented specifically for use by
348 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
349 It passes everything in registers, going to extremes to achieve this
350 by disabling callee save registers. This calling convention should
351 not be used lightly but only for specific situations such as an
352 alternative to the *register pinning* performance technique often
353 used when implementing functional programming languages. At the
354 moment only X86 supports this convention and it has the following
357 - On *X86-32* only supports up to 4 bit type parameters. No
358 floating point types are supported.
359 - On *X86-64* only supports up to 10 bit type parameters and 6
360 floating point parameters.
362 This calling convention supports `tail call
363 optimization <CodeGenerator.html#id80>`_ but requires both the
364 caller and callee are using it.
365 "``cc 11``" - The HiPE calling convention
366 This calling convention has been implemented specifically for use by
367 the `High-Performance Erlang
368 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
369 native code compiler of the `Ericsson's Open Source Erlang/OTP
370 system <http://www.erlang.org/download.shtml>`_. It uses more
371 registers for argument passing than the ordinary C calling
372 convention and defines no callee-saved registers. The calling
373 convention properly supports `tail call
374 optimization <CodeGenerator.html#id80>`_ but requires that both the
375 caller and the callee use it. It uses a *register pinning*
376 mechanism, similar to GHC's convention, for keeping frequently
377 accessed runtime components pinned to specific hardware registers.
378 At the moment only X86 supports this convention (both 32 and 64
380 "``cc <n>``" - Numbered convention
381 Any calling convention may be specified by number, allowing
382 target-specific calling conventions to be used. Target specific
383 calling conventions start at 64.
385 More calling conventions can be added/defined on an as-needed basis, to
386 support Pascal conventions or any other well-known target-independent
392 All Global Variables and Functions have one of the following visibility
395 "``default``" - Default style
396 On targets that use the ELF object file format, default visibility
397 means that the declaration is visible to other modules and, in
398 shared libraries, means that the declared entity may be overridden.
399 On Darwin, default visibility means that the declaration is visible
400 to other modules. Default visibility corresponds to "external
401 linkage" in the language.
402 "``hidden``" - Hidden style
403 Two declarations of an object with hidden visibility refer to the
404 same object if they are in the same shared object. Usually, hidden
405 visibility indicates that the symbol will not be placed into the
406 dynamic symbol table, so no other module (executable or shared
407 library) can reference it directly.
408 "``protected``" - Protected style
409 On ELF, protected visibility indicates that the symbol will be
410 placed in the dynamic symbol table, but that references within the
411 defining module will bind to the local symbol. That is, the symbol
412 cannot be overridden by another module.
417 LLVM IR allows you to specify name aliases for certain types. This can
418 make it easier to read the IR and make the IR more condensed
419 (particularly when recursive types are involved). An example of a name
424 %mytype = type { %mytype*, i32 }
426 You may give a name to any :ref:`type <typesystem>` except
427 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
428 expected with the syntax "%mytype".
430 Note that type names are aliases for the structural type that they
431 indicate, and that you can therefore specify multiple names for the same
432 type. This often leads to confusing behavior when dumping out a .ll
433 file. Since LLVM IR uses structural typing, the name is not part of the
434 type. When printing out LLVM IR, the printer will pick *one name* to
435 render all types of a particular shape. This means that if you have code
436 where two different source types end up having the same LLVM type, that
437 the dumper will sometimes print the "wrong" or unexpected type. This is
438 an important design point and isn't going to change.
445 Global variables define regions of memory allocated at compilation time
446 instead of run-time. Global variables may optionally be initialized, may
447 have an explicit section to be placed in, and may have an optional
448 explicit alignment specified.
450 A variable may be defined as ``thread_local``, which means that it will
451 not be shared by threads (each thread will have a separated copy of the
452 variable). Not all targets support thread-local variables. Optionally, a
453 TLS model may be specified:
456 For variables that are only used within the current shared library.
458 For variables in modules that will not be loaded dynamically.
460 For variables defined in the executable and only used within it.
462 The models correspond to the ELF TLS models; see `ELF Handling For
463 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
464 more information on under which circumstances the different models may
465 be used. The target may choose a different TLS model if the specified
466 model is not supported, or if a better choice of model can be made.
468 A variable may be defined as a global ``constant``, which indicates that
469 the contents of the variable will **never** be modified (enabling better
470 optimization, allowing the global data to be placed in the read-only
471 section of an executable, etc). Note that variables that need runtime
472 initialization cannot be marked ``constant`` as there is a store to the
475 LLVM explicitly allows *declarations* of global variables to be marked
476 constant, even if the final definition of the global is not. This
477 capability can be used to enable slightly better optimization of the
478 program, but requires the language definition to guarantee that
479 optimizations based on the 'constantness' are valid for the translation
480 units that do not include the definition.
482 As SSA values, global variables define pointer values that are in scope
483 (i.e. they dominate) all basic blocks in the program. Global variables
484 always define a pointer to their "content" type because they describe a
485 region of memory, and all memory objects in LLVM are accessed through
488 Global variables can be marked with ``unnamed_addr`` which indicates
489 that the address is not significant, only the content. Constants marked
490 like this can be merged with other constants if they have the same
491 initializer. Note that a constant with significant address *can* be
492 merged with a ``unnamed_addr`` constant, the result being a constant
493 whose address is significant.
495 A global variable may be declared to reside in a target-specific
496 numbered address space. For targets that support them, address spaces
497 may affect how optimizations are performed and/or what target
498 instructions are used to access the variable. The default address space
499 is zero. The address space qualifier must precede any other attributes.
501 LLVM allows an explicit section to be specified for globals. If the
502 target supports it, it will emit globals to the section specified.
504 By default, global initializers are optimized by assuming that global
505 variables defined within the module are not modified from their
506 initial values before the start of the global initializer. This is
507 true even for variables potentially accessible from outside the
508 module, including those with external linkage or appearing in
509 ``@llvm.used``. This assumption may be suppressed by marking the
510 variable with ``externally_initialized``.
512 An explicit alignment may be specified for a global, which must be a
513 power of 2. If not present, or if the alignment is set to zero, the
514 alignment of the global is set by the target to whatever it feels
515 convenient. If an explicit alignment is specified, the global is forced
516 to have exactly that alignment. Targets and optimizers are not allowed
517 to over-align the global if the global has an assigned section. In this
518 case, the extra alignment could be observable: for example, code could
519 assume that the globals are densely packed in their section and try to
520 iterate over them as an array, alignment padding would break this
523 For example, the following defines a global in a numbered address space
524 with an initializer, section, and alignment:
528 @G = addrspace(5) constant float 1.0, section "foo", align 4
530 The following example defines a thread-local global with the
531 ``initialexec`` TLS model:
535 @G = thread_local(initialexec) global i32 0, align 4
537 .. _functionstructure:
542 LLVM function definitions consist of the "``define``" keyword, an
543 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
544 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
545 an optional ``unnamed_addr`` attribute, a return type, an optional
546 :ref:`parameter attribute <paramattrs>` for the return type, a function
547 name, a (possibly empty) argument list (each with optional :ref:`parameter
548 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
549 an optional section, an optional alignment, an optional :ref:`garbage
550 collector name <gc>`, an opening curly brace, a list of basic blocks,
551 and a closing curly brace.
553 LLVM function declarations consist of the "``declare``" keyword, an
554 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
555 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
556 an optional ``unnamed_addr`` attribute, a return type, an optional
557 :ref:`parameter attribute <paramattrs>` for the return type, a function
558 name, a possibly empty list of arguments, an optional alignment, and an
559 optional :ref:`garbage collector name <gc>`.
561 A function definition contains a list of basic blocks, forming the CFG
562 (Control Flow Graph) for the function. Each basic block may optionally
563 start with a label (giving the basic block a symbol table entry),
564 contains a list of instructions, and ends with a
565 :ref:`terminator <terminators>` instruction (such as a branch or function
568 The first basic block in a function is special in two ways: it is
569 immediately executed on entrance to the function, and it is not allowed
570 to have predecessor basic blocks (i.e. there can not be any branches to
571 the entry block of a function). Because the block can have no
572 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
574 LLVM allows an explicit section to be specified for functions. If the
575 target supports it, it will emit functions to the section specified.
577 An explicit alignment may be specified for a function. If not present,
578 or if the alignment is set to zero, the alignment of the function is set
579 by the target to whatever it feels convenient. If an explicit alignment
580 is specified, the function is forced to have at least that much
581 alignment. All alignments must be a power of 2.
583 If the ``unnamed_addr`` attribute is given, the address is know to not
584 be significant and two identical functions can be merged.
588 define [linkage] [visibility]
590 <ResultType> @<FunctionName> ([argument list])
591 [fn Attrs] [section "name"] [align N]
597 Aliases act as "second name" for the aliasee value (which can be either
598 function, global variable, another alias or bitcast of global value).
599 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
600 :ref:`visibility style <visibility>`.
604 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
606 .. _namedmetadatastructure:
611 Named metadata is a collection of metadata. :ref:`Metadata
612 nodes <metadata>` (but not metadata strings) are the only valid
613 operands for a named metadata.
617 ; Some unnamed metadata nodes, which are referenced by the named metadata.
618 !0 = metadata !{metadata !"zero"}
619 !1 = metadata !{metadata !"one"}
620 !2 = metadata !{metadata !"two"}
622 !name = !{!0, !1, !2}
629 The return type and each parameter of a function type may have a set of
630 *parameter attributes* associated with them. Parameter attributes are
631 used to communicate additional information about the result or
632 parameters of a function. Parameter attributes are considered to be part
633 of the function, not of the function type, so functions with different
634 parameter attributes can have the same function type.
636 Parameter attributes are simple keywords that follow the type specified.
637 If multiple parameter attributes are needed, they are space separated.
642 declare i32 @printf(i8* noalias nocapture, ...)
643 declare i32 @atoi(i8 zeroext)
644 declare signext i8 @returns_signed_char()
646 Note that any attributes for the function result (``nounwind``,
647 ``readonly``) come immediately after the argument list.
649 Currently, only the following parameter attributes are defined:
652 This indicates to the code generator that the parameter or return
653 value should be zero-extended to the extent required by the target's
654 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
655 the caller (for a parameter) or the callee (for a return value).
657 This indicates to the code generator that the parameter or return
658 value should be sign-extended to the extent required by the target's
659 ABI (which is usually 32-bits) by the caller (for a parameter) or
660 the callee (for a return value).
662 This indicates that this parameter or return value should be treated
663 in a special target-dependent fashion during while emitting code for
664 a function call or return (usually, by putting it in a register as
665 opposed to memory, though some targets use it to distinguish between
666 two different kinds of registers). Use of this attribute is
669 This indicates that the pointer parameter should really be passed by
670 value to the function. The attribute implies that a hidden copy of
671 the pointee is made between the caller and the callee, so the callee
672 is unable to modify the value in the caller. This attribute is only
673 valid on LLVM pointer arguments. It is generally used to pass
674 structs and arrays by value, but is also valid on pointers to
675 scalars. The copy is considered to belong to the caller not the
676 callee (for example, ``readonly`` functions should not write to
677 ``byval`` parameters). This is not a valid attribute for return
680 The byval attribute also supports specifying an alignment with the
681 align attribute. It indicates the alignment of the stack slot to
682 form and the known alignment of the pointer specified to the call
683 site. If the alignment is not specified, then the code generator
684 makes a target-specific assumption.
687 This indicates that the pointer parameter specifies the address of a
688 structure that is the return value of the function in the source
689 program. This pointer must be guaranteed by the caller to be valid:
690 loads and stores to the structure may be assumed by the callee
691 not to trap and to be properly aligned. This may only be applied to
692 the first parameter. This is not a valid attribute for return
695 This indicates that pointer values `*based* <pointeraliasing>` on
696 the argument or return value do not alias pointer values which are
697 not *based* on it, ignoring certain "irrelevant" dependencies. For a
698 call to the parent function, dependencies between memory references
699 from before or after the call and from those during the call are
700 "irrelevant" to the ``noalias`` keyword for the arguments and return
701 value used in that call. The caller shares the responsibility with
702 the callee for ensuring that these requirements are met. For further
703 details, please see the discussion of the NoAlias response in `alias
704 analysis <AliasAnalysis.html#MustMayNo>`_.
706 Note that this definition of ``noalias`` is intentionally similar
707 to the definition of ``restrict`` in C99 for function arguments,
708 though it is slightly weaker.
710 For function return values, C99's ``restrict`` is not meaningful,
711 while LLVM's ``noalias`` is.
713 This indicates that the callee does not make any copies of the
714 pointer that outlive the callee itself. This is not a valid
715 attribute for return values.
720 This indicates that the pointer parameter can be excised using the
721 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
722 attribute for return values.
726 Garbage Collector Names
727 -----------------------
729 Each function may specify a garbage collector name, which is simply a
734 define void @f() gc "name" { ... }
736 The compiler declares the supported values of *name*. Specifying a
737 collector which will cause the compiler to alter its output in order to
738 support the named garbage collection algorithm.
745 Function attributes are set to communicate additional information about
746 a function. Function attributes are considered to be part of the
747 function, not of the function type, so functions with different function
748 attributes can have the same function type.
750 Function attributes are simple keywords that follow the type specified.
751 If multiple attributes are needed, they are space separated. For
756 define void @f() noinline { ... }
757 define void @f() alwaysinline { ... }
758 define void @f() alwaysinline optsize { ... }
759 define void @f() optsize { ... }
762 This attribute indicates that the address safety analysis is enabled
765 This attribute indicates that, when emitting the prologue and
766 epilogue, the backend should forcibly align the stack pointer.
767 Specify the desired alignment, which must be a power of two, in
770 This attribute indicates that the inliner should attempt to inline
771 this function into callers whenever possible, ignoring any active
772 inlining size threshold for this caller.
774 This attribute suppresses lazy symbol binding for the function. This
775 may make calls to the function faster, at the cost of extra program
776 startup time if the function is not called during program startup.
778 This attribute indicates that the source code contained a hint that
779 inlining this function is desirable (such as the "inline" keyword in
780 C/C++). It is just a hint; it imposes no requirements on the
783 This attribute disables prologue / epilogue emission for the
784 function. This can have very system-specific consequences.
786 This attributes disables implicit floating point instructions.
788 This attribute indicates that the inliner should never inline this
789 function in any situation. This attribute may not be used together
790 with the ``alwaysinline`` attribute.
792 This attribute indicates that the code generator should not use a
793 red zone, even if the target-specific ABI normally permits it.
795 This function attribute indicates that the function never returns
796 normally. This produces undefined behavior at runtime if the
797 function ever does dynamically return.
799 This function attribute indicates that the function never returns
800 with an unwind or exceptional control flow. If the function does
801 unwind, its runtime behavior is undefined.
803 This attribute suggests that optimization passes and code generator
804 passes make choices that keep the code size of this function low,
805 and otherwise do optimizations specifically to reduce code size.
807 This attribute indicates that the function computes its result (or
808 decides to unwind an exception) based strictly on its arguments,
809 without dereferencing any pointer arguments or otherwise accessing
810 any mutable state (e.g. memory, control registers, etc) visible to
811 caller functions. It does not write through any pointer arguments
812 (including ``byval`` arguments) and never changes any state visible
813 to callers. This means that it cannot unwind exceptions by calling
814 the ``C++`` exception throwing methods.
816 This attribute indicates that the function does not write through
817 any pointer arguments (including ``byval`` arguments) or otherwise
818 modify any state (e.g. memory, control registers, etc) visible to
819 caller functions. It may dereference pointer arguments and read
820 state that may be set in the caller. A readonly function always
821 returns the same value (or unwinds an exception identically) when
822 called with the same set of arguments and global state. It cannot
823 unwind an exception by calling the ``C++`` exception throwing
826 This attribute indicates that this function can return twice. The C
827 ``setjmp`` is an example of such a function. The compiler disables
828 some optimizations (like tail calls) in the caller of these
831 This attribute indicates that the function should emit a stack
832 smashing protector. It is in the form of a "canary" --- a random value
833 placed on the stack before the local variables that's checked upon
834 return from the function to see if it has been overwritten. A
835 heuristic is used to determine if a function needs stack protectors
836 or not. The heuristic used will enable protectors for functions with:
838 - Character arrays larger than ``ssp-buffer-size`` (default 8).
839 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
840 - Calls to alloca() with variable sizes or constant sizes greater than
843 If a function that has an ``ssp`` attribute is inlined into a
844 function that doesn't have an ``ssp`` attribute, then the resulting
845 function will have an ``ssp`` attribute.
847 This attribute indicates that the function should *always* emit a
848 stack smashing protector. This overrides the ``ssp`` function
851 If a function that has an ``sspreq`` attribute is inlined into a
852 function that doesn't have an ``sspreq`` attribute or which has an
853 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
854 an ``sspreq`` attribute.
856 This attribute indicates that the function should emit a stack smashing
857 protector. This attribute causes a strong heuristic to be used when
858 determining if a function needs stack protectors. The strong heuristic
859 will enable protectors for functions with:
861 - Arrays of any size and type
862 - Aggregates containing an array of any size and type.
864 - Local variables that have had their address taken.
866 This overrides the ``ssp`` function attribute.
868 If a function that has an ``sspstrong`` attribute is inlined into a
869 function that doesn't have an ``sspstrong`` attribute, then the
870 resulting function will have an ``sspstrong`` attribute.
872 This attribute indicates that the ABI being targeted requires that
873 an unwind table entry be produce for this function even if we can
874 show that no exceptions passes by it. This is normally the case for
875 the ELF x86-64 abi, but it can be disabled for some compilation
878 This attribute indicates that calls to the function cannot be
879 duplicated. A call to a ``noduplicate`` function may be moved
880 within its parent function, but may not be duplicated within
883 A function containing a ``noduplicate`` call may still
884 be an inlining candidate, provided that the call is not
885 duplicated by inlining. That implies that the function has
886 internal linkage and only has one call site, so the original
887 call is dead after inlining.
891 Module-Level Inline Assembly
892 ----------------------------
894 Modules may contain "module-level inline asm" blocks, which corresponds
895 to the GCC "file scope inline asm" blocks. These blocks are internally
896 concatenated by LLVM and treated as a single unit, but may be separated
897 in the ``.ll`` file if desired. The syntax is very simple:
901 module asm "inline asm code goes here"
902 module asm "more can go here"
904 The strings can contain any character by escaping non-printable
905 characters. The escape sequence used is simply "\\xx" where "xx" is the
906 two digit hex code for the number.
908 The inline asm code is simply printed to the machine code .s file when
909 assembly code is generated.
914 A module may specify a target specific data layout string that specifies
915 how data is to be laid out in memory. The syntax for the data layout is
920 target datalayout = "layout specification"
922 The *layout specification* consists of a list of specifications
923 separated by the minus sign character ('-'). Each specification starts
924 with a letter and may include other information after the letter to
925 define some aspect of the data layout. The specifications accepted are
929 Specifies that the target lays out data in big-endian form. That is,
930 the bits with the most significance have the lowest address
933 Specifies that the target lays out data in little-endian form. That
934 is, the bits with the least significance have the lowest address
937 Specifies the natural alignment of the stack in bits. Alignment
938 promotion of stack variables is limited to the natural stack
939 alignment to avoid dynamic stack realignment. The stack alignment
940 must be a multiple of 8-bits. If omitted, the natural stack
941 alignment defaults to "unspecified", which does not prevent any
942 alignment promotions.
943 ``p[n]:<size>:<abi>:<pref>``
944 This specifies the *size* of a pointer and its ``<abi>`` and
945 ``<pref>``\erred alignments for address space ``n``. All sizes are in
946 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
947 preceding ``:`` should be omitted too. The address space, ``n`` is
948 optional, and if not specified, denotes the default address space 0.
949 The value of ``n`` must be in the range [1,2^23).
950 ``i<size>:<abi>:<pref>``
951 This specifies the alignment for an integer type of a given bit
952 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
953 ``v<size>:<abi>:<pref>``
954 This specifies the alignment for a vector type of a given bit
956 ``f<size>:<abi>:<pref>``
957 This specifies the alignment for a floating point type of a given bit
958 ``<size>``. Only values of ``<size>`` that are supported by the target
959 will work. 32 (float) and 64 (double) are supported on all targets; 80
960 or 128 (different flavors of long double) are also supported on some
962 ``a<size>:<abi>:<pref>``
963 This specifies the alignment for an aggregate type of a given bit
965 ``s<size>:<abi>:<pref>``
966 This specifies the alignment for a stack object of a given bit
968 ``n<size1>:<size2>:<size3>...``
969 This specifies a set of native integer widths for the target CPU in
970 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
971 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
972 this set are considered to support most general arithmetic operations
975 When constructing the data layout for a given target, LLVM starts with a
976 default set of specifications which are then (possibly) overridden by
977 the specifications in the ``datalayout`` keyword. The default
978 specifications are given in this list:
981 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment
982 - ``S0`` - natural stack alignment is unspecified
983 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
984 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
985 - ``i16:16:16`` - i16 is 16-bit aligned
986 - ``i32:32:32`` - i32 is 32-bit aligned
987 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
989 - ``f16:16:16`` - half is 16-bit aligned
990 - ``f32:32:32`` - float is 32-bit aligned
991 - ``f64:64:64`` - double is 64-bit aligned
992 - ``f128:128:128`` - quad is 128-bit aligned
993 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
994 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
995 - ``a0:0:64`` - aggregates are 64-bit aligned
997 When LLVM is determining the alignment for a given type, it uses the
1000 #. If the type sought is an exact match for one of the specifications,
1001 that specification is used.
1002 #. If no match is found, and the type sought is an integer type, then
1003 the smallest integer type that is larger than the bitwidth of the
1004 sought type is used. If none of the specifications are larger than
1005 the bitwidth then the largest integer type is used. For example,
1006 given the default specifications above, the i7 type will use the
1007 alignment of i8 (next largest) while both i65 and i256 will use the
1008 alignment of i64 (largest specified).
1009 #. If no match is found, and the type sought is a vector type, then the
1010 largest vector type that is smaller than the sought vector type will
1011 be used as a fall back. This happens because <128 x double> can be
1012 implemented in terms of 64 <2 x double>, for example.
1014 The function of the data layout string may not be what you expect.
1015 Notably, this is not a specification from the frontend of what alignment
1016 the code generator should use.
1018 Instead, if specified, the target data layout is required to match what
1019 the ultimate *code generator* expects. This string is used by the
1020 mid-level optimizers to improve code, and this only works if it matches
1021 what the ultimate code generator uses. If you would like to generate IR
1022 that does not embed this target-specific detail into the IR, then you
1023 don't have to specify the string. This will disable some optimizations
1024 that require precise layout information, but this also prevents those
1025 optimizations from introducing target specificity into the IR.
1027 .. _pointeraliasing:
1029 Pointer Aliasing Rules
1030 ----------------------
1032 Any memory access must be done through a pointer value associated with
1033 an address range of the memory access, otherwise the behavior is
1034 undefined. Pointer values are associated with address ranges according
1035 to the following rules:
1037 - A pointer value is associated with the addresses associated with any
1038 value it is *based* on.
1039 - An address of a global variable is associated with the address range
1040 of the variable's storage.
1041 - The result value of an allocation instruction is associated with the
1042 address range of the allocated storage.
1043 - A null pointer in the default address-space is associated with no
1045 - An integer constant other than zero or a pointer value returned from
1046 a function not defined within LLVM may be associated with address
1047 ranges allocated through mechanisms other than those provided by
1048 LLVM. Such ranges shall not overlap with any ranges of addresses
1049 allocated by mechanisms provided by LLVM.
1051 A pointer value is *based* on another pointer value according to the
1054 - A pointer value formed from a ``getelementptr`` operation is *based*
1055 on the first operand of the ``getelementptr``.
1056 - The result value of a ``bitcast`` is *based* on the operand of the
1058 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1059 values that contribute (directly or indirectly) to the computation of
1060 the pointer's value.
1061 - The "*based* on" relationship is transitive.
1063 Note that this definition of *"based"* is intentionally similar to the
1064 definition of *"based"* in C99, though it is slightly weaker.
1066 LLVM IR does not associate types with memory. The result type of a
1067 ``load`` merely indicates the size and alignment of the memory from
1068 which to load, as well as the interpretation of the value. The first
1069 operand type of a ``store`` similarly only indicates the size and
1070 alignment of the store.
1072 Consequently, type-based alias analysis, aka TBAA, aka
1073 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1074 :ref:`Metadata <metadata>` may be used to encode additional information
1075 which specialized optimization passes may use to implement type-based
1080 Volatile Memory Accesses
1081 ------------------------
1083 Certain memory accesses, such as :ref:`load <i_load>`'s,
1084 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1085 marked ``volatile``. The optimizers must not change the number of
1086 volatile operations or change their order of execution relative to other
1087 volatile operations. The optimizers *may* change the order of volatile
1088 operations relative to non-volatile operations. This is not Java's
1089 "volatile" and has no cross-thread synchronization behavior.
1091 IR-level volatile loads and stores cannot safely be optimized into
1092 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1093 flagged volatile. Likewise, the backend should never split or merge
1094 target-legal volatile load/store instructions.
1096 .. admonition:: Rationale
1098 Platforms may rely on volatile loads and stores of natively supported
1099 data width to be executed as single instruction. For example, in C
1100 this holds for an l-value of volatile primitive type with native
1101 hardware support, but not necessarily for aggregate types. The
1102 frontend upholds these expectations, which are intentionally
1103 unspecified in the IR. The rules above ensure that IR transformation
1104 do not violate the frontend's contract with the language.
1108 Memory Model for Concurrent Operations
1109 --------------------------------------
1111 The LLVM IR does not define any way to start parallel threads of
1112 execution or to register signal handlers. Nonetheless, there are
1113 platform-specific ways to create them, and we define LLVM IR's behavior
1114 in their presence. This model is inspired by the C++0x memory model.
1116 For a more informal introduction to this model, see the :doc:`Atomics`.
1118 We define a *happens-before* partial order as the least partial order
1121 - Is a superset of single-thread program order, and
1122 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1123 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1124 techniques, like pthread locks, thread creation, thread joining,
1125 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1126 Constraints <ordering>`).
1128 Note that program order does not introduce *happens-before* edges
1129 between a thread and signals executing inside that thread.
1131 Every (defined) read operation (load instructions, memcpy, atomic
1132 loads/read-modify-writes, etc.) R reads a series of bytes written by
1133 (defined) write operations (store instructions, atomic
1134 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1135 section, initialized globals are considered to have a write of the
1136 initializer which is atomic and happens before any other read or write
1137 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1138 may see any write to the same byte, except:
1140 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1141 write\ :sub:`2` happens before R\ :sub:`byte`, then
1142 R\ :sub:`byte` does not see write\ :sub:`1`.
1143 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1144 R\ :sub:`byte` does not see write\ :sub:`3`.
1146 Given that definition, R\ :sub:`byte` is defined as follows:
1148 - If R is volatile, the result is target-dependent. (Volatile is
1149 supposed to give guarantees which can support ``sig_atomic_t`` in
1150 C/C++, and may be used for accesses to addresses which do not behave
1151 like normal memory. It does not generally provide cross-thread
1153 - Otherwise, if there is no write to the same byte that happens before
1154 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1155 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1156 R\ :sub:`byte` returns the value written by that write.
1157 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1158 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1159 Memory Ordering Constraints <ordering>` section for additional
1160 constraints on how the choice is made.
1161 - Otherwise R\ :sub:`byte` returns ``undef``.
1163 R returns the value composed of the series of bytes it read. This
1164 implies that some bytes within the value may be ``undef`` **without**
1165 the entire value being ``undef``. Note that this only defines the
1166 semantics of the operation; it doesn't mean that targets will emit more
1167 than one instruction to read the series of bytes.
1169 Note that in cases where none of the atomic intrinsics are used, this
1170 model places only one restriction on IR transformations on top of what
1171 is required for single-threaded execution: introducing a store to a byte
1172 which might not otherwise be stored is not allowed in general.
1173 (Specifically, in the case where another thread might write to and read
1174 from an address, introducing a store can change a load that may see
1175 exactly one write into a load that may see multiple writes.)
1179 Atomic Memory Ordering Constraints
1180 ----------------------------------
1182 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1183 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1184 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1185 an ordering parameter that determines which other atomic instructions on
1186 the same address they *synchronize with*. These semantics are borrowed
1187 from Java and C++0x, but are somewhat more colloquial. If these
1188 descriptions aren't precise enough, check those specs (see spec
1189 references in the :doc:`atomics guide <Atomics>`).
1190 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1191 differently since they don't take an address. See that instruction's
1192 documentation for details.
1194 For a simpler introduction to the ordering constraints, see the
1198 The set of values that can be read is governed by the happens-before
1199 partial order. A value cannot be read unless some operation wrote
1200 it. This is intended to provide a guarantee strong enough to model
1201 Java's non-volatile shared variables. This ordering cannot be
1202 specified for read-modify-write operations; it is not strong enough
1203 to make them atomic in any interesting way.
1205 In addition to the guarantees of ``unordered``, there is a single
1206 total order for modifications by ``monotonic`` operations on each
1207 address. All modification orders must be compatible with the
1208 happens-before order. There is no guarantee that the modification
1209 orders can be combined to a global total order for the whole program
1210 (and this often will not be possible). The read in an atomic
1211 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1212 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1213 order immediately before the value it writes. If one atomic read
1214 happens before another atomic read of the same address, the later
1215 read must see the same value or a later value in the address's
1216 modification order. This disallows reordering of ``monotonic`` (or
1217 stronger) operations on the same address. If an address is written
1218 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1219 read that address repeatedly, the other threads must eventually see
1220 the write. This corresponds to the C++0x/C1x
1221 ``memory_order_relaxed``.
1223 In addition to the guarantees of ``monotonic``, a
1224 *synchronizes-with* edge may be formed with a ``release`` operation.
1225 This is intended to model C++'s ``memory_order_acquire``.
1227 In addition to the guarantees of ``monotonic``, if this operation
1228 writes a value which is subsequently read by an ``acquire``
1229 operation, it *synchronizes-with* that operation. (This isn't a
1230 complete description; see the C++0x definition of a release
1231 sequence.) This corresponds to the C++0x/C1x
1232 ``memory_order_release``.
1233 ``acq_rel`` (acquire+release)
1234 Acts as both an ``acquire`` and ``release`` operation on its
1235 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1236 ``seq_cst`` (sequentially consistent)
1237 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1238 operation which only reads, ``release`` for an operation which only
1239 writes), there is a global total order on all
1240 sequentially-consistent operations on all addresses, which is
1241 consistent with the *happens-before* partial order and with the
1242 modification orders of all the affected addresses. Each
1243 sequentially-consistent read sees the last preceding write to the
1244 same address in this global order. This corresponds to the C++0x/C1x
1245 ``memory_order_seq_cst`` and Java volatile.
1249 If an atomic operation is marked ``singlethread``, it only *synchronizes
1250 with* or participates in modification and seq\_cst total orderings with
1251 other operations running in the same thread (for example, in signal
1259 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1260 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1261 :ref:`frem <i_frem>`) have the following flags that can set to enable
1262 otherwise unsafe floating point operations
1265 No NaNs - Allow optimizations to assume the arguments and result are not
1266 NaN. Such optimizations are required to retain defined behavior over
1267 NaNs, but the value of the result is undefined.
1270 No Infs - Allow optimizations to assume the arguments and result are not
1271 +/-Inf. Such optimizations are required to retain defined behavior over
1272 +/-Inf, but the value of the result is undefined.
1275 No Signed Zeros - Allow optimizations to treat the sign of a zero
1276 argument or result as insignificant.
1279 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1280 argument rather than perform division.
1283 Fast - Allow algebraically equivalent transformations that may
1284 dramatically change results in floating point (e.g. reassociate). This
1285 flag implies all the others.
1292 The LLVM type system is one of the most important features of the
1293 intermediate representation. Being typed enables a number of
1294 optimizations to be performed on the intermediate representation
1295 directly, without having to do extra analyses on the side before the
1296 transformation. A strong type system makes it easier to read the
1297 generated code and enables novel analyses and transformations that are
1298 not feasible to perform on normal three address code representations.
1300 Type Classifications
1301 --------------------
1303 The types fall into a few useful classifications:
1312 * - :ref:`integer <t_integer>`
1313 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1316 * - :ref:`floating point <t_floating>`
1317 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1325 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1326 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1327 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1328 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1330 * - :ref:`primitive <t_primitive>`
1331 - :ref:`label <t_label>`,
1332 :ref:`void <t_void>`,
1333 :ref:`integer <t_integer>`,
1334 :ref:`floating point <t_floating>`,
1335 :ref:`x86mmx <t_x86mmx>`,
1336 :ref:`metadata <t_metadata>`.
1338 * - :ref:`derived <t_derived>`
1339 - :ref:`array <t_array>`,
1340 :ref:`function <t_function>`,
1341 :ref:`pointer <t_pointer>`,
1342 :ref:`structure <t_struct>`,
1343 :ref:`vector <t_vector>`,
1344 :ref:`opaque <t_opaque>`.
1346 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1347 Values of these types are the only ones which can be produced by
1355 The primitive types are the fundamental building blocks of the LLVM
1366 The integer type is a very simple type that simply specifies an
1367 arbitrary bit width for the integer type desired. Any bit width from 1
1368 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1377 The number of bits the integer will occupy is specified by the ``N``
1383 +----------------+------------------------------------------------+
1384 | ``i1`` | a single-bit integer. |
1385 +----------------+------------------------------------------------+
1386 | ``i32`` | a 32-bit integer. |
1387 +----------------+------------------------------------------------+
1388 | ``i1942652`` | a really big integer of over 1 million bits. |
1389 +----------------+------------------------------------------------+
1393 Floating Point Types
1394 ^^^^^^^^^^^^^^^^^^^^
1403 - 16-bit floating point value
1406 - 32-bit floating point value
1409 - 64-bit floating point value
1412 - 128-bit floating point value (112-bit mantissa)
1415 - 80-bit floating point value (X87)
1418 - 128-bit floating point value (two 64-bits)
1428 The x86mmx type represents a value held in an MMX register on an x86
1429 machine. The operations allowed on it are quite limited: parameters and
1430 return values, load and store, and bitcast. User-specified MMX
1431 instructions are represented as intrinsic or asm calls with arguments
1432 and/or results of this type. There are no arrays, vectors or constants
1450 The void type does not represent any value and has no size.
1467 The label type represents code labels.
1484 The metadata type represents embedded metadata. No derived types may be
1485 created from metadata except for :ref:`function <t_function>` arguments.
1499 The real power in LLVM comes from the derived types in the system. This
1500 is what allows a programmer to represent arrays, functions, pointers,
1501 and other useful types. Each of these types contain one or more element
1502 types which may be a primitive type, or another derived type. For
1503 example, it is possible to have a two dimensional array, using an array
1504 as the element type of another array.
1511 Aggregate Types are a subset of derived types that can contain multiple
1512 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1513 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1524 The array type is a very simple derived type that arranges elements
1525 sequentially in memory. The array type requires a size (number of
1526 elements) and an underlying data type.
1533 [<# elements> x <elementtype>]
1535 The number of elements is a constant integer value; ``elementtype`` may
1536 be any type with a size.
1541 +------------------+--------------------------------------+
1542 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1543 +------------------+--------------------------------------+
1544 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1545 +------------------+--------------------------------------+
1546 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1547 +------------------+--------------------------------------+
1549 Here are some examples of multidimensional arrays:
1551 +-----------------------------+----------------------------------------------------------+
1552 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1553 +-----------------------------+----------------------------------------------------------+
1554 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1555 +-----------------------------+----------------------------------------------------------+
1556 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1557 +-----------------------------+----------------------------------------------------------+
1559 There is no restriction on indexing beyond the end of the array implied
1560 by a static type (though there are restrictions on indexing beyond the
1561 bounds of an allocated object in some cases). This means that
1562 single-dimension 'variable sized array' addressing can be implemented in
1563 LLVM with a zero length array type. An implementation of 'pascal style
1564 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1575 The function type can be thought of as a function signature. It consists
1576 of a return type and a list of formal parameter types. The return type
1577 of a function type is a first class type or a void type.
1584 <returntype> (<parameter list>)
1586 ...where '``<parameter list>``' is a comma-separated list of type
1587 specifiers. Optionally, the parameter list may include a type ``...``,
1588 which indicates that the function takes a variable number of arguments.
1589 Variable argument functions can access their arguments with the
1590 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1591 '``<returntype>``' is any type except :ref:`label <t_label>`.
1596 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1597 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1598 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1599 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1600 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1601 | ``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. |
1602 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1603 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1604 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1614 The structure type is used to represent a collection of data members
1615 together in memory. The elements of a structure may be any type that has
1618 Structures in memory are accessed using '``load``' and '``store``' by
1619 getting a pointer to a field with the '``getelementptr``' instruction.
1620 Structures in registers are accessed using the '``extractvalue``' and
1621 '``insertvalue``' instructions.
1623 Structures may optionally be "packed" structures, which indicate that
1624 the alignment of the struct is one byte, and that there is no padding
1625 between the elements. In non-packed structs, padding between field types
1626 is inserted as defined by the DataLayout string in the module, which is
1627 required to match what the underlying code generator expects.
1629 Structures can either be "literal" or "identified". A literal structure
1630 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1631 identified types are always defined at the top level with a name.
1632 Literal types are uniqued by their contents and can never be recursive
1633 or opaque since there is no way to write one. Identified types can be
1634 recursive, can be opaqued, and are never uniqued.
1641 %T1 = type { <type list> } ; Identified normal struct type
1642 %T2 = type <{ <type list> }> ; Identified packed struct type
1647 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1648 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1649 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1650 | ``{ 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``. |
1651 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1652 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1653 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1657 Opaque Structure Types
1658 ^^^^^^^^^^^^^^^^^^^^^^
1663 Opaque structure types are used to represent named structure types that
1664 do not have a body specified. This corresponds (for example) to the C
1665 notion of a forward declared structure.
1678 +--------------+-------------------+
1679 | ``opaque`` | An opaque type. |
1680 +--------------+-------------------+
1690 The pointer type is used to specify memory locations. Pointers are
1691 commonly used to reference objects in memory.
1693 Pointer types may have an optional address space attribute defining the
1694 numbered address space where the pointed-to object resides. The default
1695 address space is number zero. The semantics of non-zero address spaces
1696 are target-specific.
1698 Note that LLVM does not permit pointers to void (``void*``) nor does it
1699 permit pointers to labels (``label*``). Use ``i8*`` instead.
1711 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1712 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1713 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1714 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1715 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1716 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1717 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1727 A vector type is a simple derived type that represents a vector of
1728 elements. Vector types are used when multiple primitive data are
1729 operated in parallel using a single instruction (SIMD). A vector type
1730 requires a size (number of elements) and an underlying primitive data
1731 type. Vector types are considered :ref:`first class <t_firstclass>`.
1738 < <# elements> x <elementtype> >
1740 The number of elements is a constant integer value larger than 0;
1741 elementtype may be any integer or floating point type, or a pointer to
1742 these types. Vectors of size zero are not allowed.
1747 +-------------------+--------------------------------------------------+
1748 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1749 +-------------------+--------------------------------------------------+
1750 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1751 +-------------------+--------------------------------------------------+
1752 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1753 +-------------------+--------------------------------------------------+
1754 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1755 +-------------------+--------------------------------------------------+
1760 LLVM has several different basic types of constants. This section
1761 describes them all and their syntax.
1766 **Boolean constants**
1767 The two strings '``true``' and '``false``' are both valid constants
1769 **Integer constants**
1770 Standard integers (such as '4') are constants of the
1771 :ref:`integer <t_integer>` type. Negative numbers may be used with
1773 **Floating point constants**
1774 Floating point constants use standard decimal notation (e.g.
1775 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1776 hexadecimal notation (see below). The assembler requires the exact
1777 decimal value of a floating-point constant. For example, the
1778 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1779 decimal in binary. Floating point constants must have a :ref:`floating
1780 point <t_floating>` type.
1781 **Null pointer constants**
1782 The identifier '``null``' is recognized as a null pointer constant
1783 and must be of :ref:`pointer type <t_pointer>`.
1785 The one non-intuitive notation for constants is the hexadecimal form of
1786 floating point constants. For example, the form
1787 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1788 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1789 constants are required (and the only time that they are generated by the
1790 disassembler) is when a floating point constant must be emitted but it
1791 cannot be represented as a decimal floating point number in a reasonable
1792 number of digits. For example, NaN's, infinities, and other special
1793 values are represented in their IEEE hexadecimal format so that assembly
1794 and disassembly do not cause any bits to change in the constants.
1796 When using the hexadecimal form, constants of types half, float, and
1797 double are represented using the 16-digit form shown above (which
1798 matches the IEEE754 representation for double); half and float values
1799 must, however, be exactly representable as IEEE 754 half and single
1800 precision, respectively. Hexadecimal format is always used for long
1801 double, and there are three forms of long double. The 80-bit format used
1802 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1803 128-bit format used by PowerPC (two adjacent doubles) is represented by
1804 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1805 represented by ``0xL`` followed by 32 hexadecimal digits; no currently
1806 supported target uses this format. Long doubles will only work if they
1807 match the long double format on your target. The IEEE 16-bit format
1808 (half precision) is represented by ``0xH`` followed by 4 hexadecimal
1809 digits. All hexadecimal formats are big-endian (sign bit at the left).
1811 There are no constants of type x86mmx.
1816 Complex constants are a (potentially recursive) combination of simple
1817 constants and smaller complex constants.
1819 **Structure constants**
1820 Structure constants are represented with notation similar to
1821 structure type definitions (a comma separated list of elements,
1822 surrounded by braces (``{}``)). For example:
1823 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1824 "``@G = external global i32``". Structure constants must have
1825 :ref:`structure type <t_struct>`, and the number and types of elements
1826 must match those specified by the type.
1828 Array constants are represented with notation similar to array type
1829 definitions (a comma separated list of elements, surrounded by
1830 square brackets (``[]``)). For example:
1831 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1832 :ref:`array type <t_array>`, and the number and types of elements must
1833 match those specified by the type.
1834 **Vector constants**
1835 Vector constants are represented with notation similar to vector
1836 type definitions (a comma separated list of elements, surrounded by
1837 less-than/greater-than's (``<>``)). For example:
1838 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1839 must have :ref:`vector type <t_vector>`, and the number and types of
1840 elements must match those specified by the type.
1841 **Zero initialization**
1842 The string '``zeroinitializer``' can be used to zero initialize a
1843 value to zero of *any* type, including scalar and
1844 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1845 having to print large zero initializers (e.g. for large arrays) and
1846 is always exactly equivalent to using explicit zero initializers.
1848 A metadata node is a structure-like constant with :ref:`metadata
1849 type <t_metadata>`. For example:
1850 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1851 constants that are meant to be interpreted as part of the
1852 instruction stream, metadata is a place to attach additional
1853 information such as debug info.
1855 Global Variable and Function Addresses
1856 --------------------------------------
1858 The addresses of :ref:`global variables <globalvars>` and
1859 :ref:`functions <functionstructure>` are always implicitly valid
1860 (link-time) constants. These constants are explicitly referenced when
1861 the :ref:`identifier for the global <identifiers>` is used and always have
1862 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1865 .. code-block:: llvm
1869 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1876 The string '``undef``' can be used anywhere a constant is expected, and
1877 indicates that the user of the value may receive an unspecified
1878 bit-pattern. Undefined values may be of any type (other than '``label``'
1879 or '``void``') and be used anywhere a constant is permitted.
1881 Undefined values are useful because they indicate to the compiler that
1882 the program is well defined no matter what value is used. This gives the
1883 compiler more freedom to optimize. Here are some examples of
1884 (potentially surprising) transformations that are valid (in pseudo IR):
1886 .. code-block:: llvm
1896 This is safe because all of the output bits are affected by the undef
1897 bits. Any output bit can have a zero or one depending on the input bits.
1899 .. code-block:: llvm
1910 These logical operations have bits that are not always affected by the
1911 input. For example, if ``%X`` has a zero bit, then the output of the
1912 '``and``' operation will always be a zero for that bit, no matter what
1913 the corresponding bit from the '``undef``' is. As such, it is unsafe to
1914 optimize or assume that the result of the '``and``' is '``undef``'.
1915 However, it is safe to assume that all bits of the '``undef``' could be
1916 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
1917 all the bits of the '``undef``' operand to the '``or``' could be set,
1918 allowing the '``or``' to be folded to -1.
1920 .. code-block:: llvm
1922 %A = select undef, %X, %Y
1923 %B = select undef, 42, %Y
1924 %C = select %X, %Y, undef
1934 This set of examples shows that undefined '``select``' (and conditional
1935 branch) conditions can go *either way*, but they have to come from one
1936 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
1937 both known to have a clear low bit, then ``%A`` would have to have a
1938 cleared low bit. However, in the ``%C`` example, the optimizer is
1939 allowed to assume that the '``undef``' operand could be the same as
1940 ``%Y``, allowing the whole '``select``' to be eliminated.
1942 .. code-block:: llvm
1944 %A = xor undef, undef
1961 This example points out that two '``undef``' operands are not
1962 necessarily the same. This can be surprising to people (and also matches
1963 C semantics) where they assume that "``X^X``" is always zero, even if
1964 ``X`` is undefined. This isn't true for a number of reasons, but the
1965 short answer is that an '``undef``' "variable" can arbitrarily change
1966 its value over its "live range". This is true because the variable
1967 doesn't actually *have a live range*. Instead, the value is logically
1968 read from arbitrary registers that happen to be around when needed, so
1969 the value is not necessarily consistent over time. In fact, ``%A`` and
1970 ``%C`` need to have the same semantics or the core LLVM "replace all
1971 uses with" concept would not hold.
1973 .. code-block:: llvm
1981 These examples show the crucial difference between an *undefined value*
1982 and *undefined behavior*. An undefined value (like '``undef``') is
1983 allowed to have an arbitrary bit-pattern. This means that the ``%A``
1984 operation can be constant folded to '``undef``', because the '``undef``'
1985 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
1986 However, in the second example, we can make a more aggressive
1987 assumption: because the ``undef`` is allowed to be an arbitrary value,
1988 we are allowed to assume that it could be zero. Since a divide by zero
1989 has *undefined behavior*, we are allowed to assume that the operation
1990 does not execute at all. This allows us to delete the divide and all
1991 code after it. Because the undefined operation "can't happen", the
1992 optimizer can assume that it occurs in dead code.
1994 .. code-block:: llvm
1996 a: store undef -> %X
1997 b: store %X -> undef
2002 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2003 value can be assumed to not have any effect; we can assume that the
2004 value is overwritten with bits that happen to match what was already
2005 there. However, a store *to* an undefined location could clobber
2006 arbitrary memory, therefore, it has undefined behavior.
2013 Poison values are similar to :ref:`undef values <undefvalues>`, however
2014 they also represent the fact that an instruction or constant expression
2015 which cannot evoke side effects has nevertheless detected a condition
2016 which results in undefined behavior.
2018 There is currently no way of representing a poison value in the IR; they
2019 only exist when produced by operations such as :ref:`add <i_add>` with
2022 Poison value behavior is defined in terms of value *dependence*:
2024 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2025 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2026 their dynamic predecessor basic block.
2027 - Function arguments depend on the corresponding actual argument values
2028 in the dynamic callers of their functions.
2029 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2030 instructions that dynamically transfer control back to them.
2031 - :ref:`Invoke <i_invoke>` instructions depend on the
2032 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2033 call instructions that dynamically transfer control back to them.
2034 - Non-volatile loads and stores depend on the most recent stores to all
2035 of the referenced memory addresses, following the order in the IR
2036 (including loads and stores implied by intrinsics such as
2037 :ref:`@llvm.memcpy <int_memcpy>`.)
2038 - An instruction with externally visible side effects depends on the
2039 most recent preceding instruction with externally visible side
2040 effects, following the order in the IR. (This includes :ref:`volatile
2041 operations <volatile>`.)
2042 - An instruction *control-depends* on a :ref:`terminator
2043 instruction <terminators>` if the terminator instruction has
2044 multiple successors and the instruction is always executed when
2045 control transfers to one of the successors, and may not be executed
2046 when control is transferred to another.
2047 - Additionally, an instruction also *control-depends* on a terminator
2048 instruction if the set of instructions it otherwise depends on would
2049 be different if the terminator had transferred control to a different
2051 - Dependence is transitive.
2053 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2054 with the additional affect that any instruction which has a *dependence*
2055 on a poison value has undefined behavior.
2057 Here are some examples:
2059 .. code-block:: llvm
2062 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2063 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2064 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2065 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2067 store i32 %poison, i32* @g ; Poison value stored to memory.
2068 %poison2 = load i32* @g ; Poison value loaded back from memory.
2070 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2072 %narrowaddr = bitcast i32* @g to i16*
2073 %wideaddr = bitcast i32* @g to i64*
2074 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2075 %poison4 = load i64* %wideaddr ; Returns a poison value.
2077 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2078 br i1 %cmp, label %true, label %end ; Branch to either destination.
2081 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2082 ; it has undefined behavior.
2086 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2087 ; Both edges into this PHI are
2088 ; control-dependent on %cmp, so this
2089 ; always results in a poison value.
2091 store volatile i32 0, i32* @g ; This would depend on the store in %true
2092 ; if %cmp is true, or the store in %entry
2093 ; otherwise, so this is undefined behavior.
2095 br i1 %cmp, label %second_true, label %second_end
2096 ; The same branch again, but this time the
2097 ; true block doesn't have side effects.
2104 store volatile i32 0, i32* @g ; This time, the instruction always depends
2105 ; on the store in %end. Also, it is
2106 ; control-equivalent to %end, so this is
2107 ; well-defined (ignoring earlier undefined
2108 ; behavior in this example).
2112 Addresses of Basic Blocks
2113 -------------------------
2115 ``blockaddress(@function, %block)``
2117 The '``blockaddress``' constant computes the address of the specified
2118 basic block in the specified function, and always has an ``i8*`` type.
2119 Taking the address of the entry block is illegal.
2121 This value only has defined behavior when used as an operand to the
2122 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2123 against null. Pointer equality tests between labels addresses results in
2124 undefined behavior --- though, again, comparison against null is ok, and
2125 no label is equal to the null pointer. This may be passed around as an
2126 opaque pointer sized value as long as the bits are not inspected. This
2127 allows ``ptrtoint`` and arithmetic to be performed on these values so
2128 long as the original value is reconstituted before the ``indirectbr``
2131 Finally, some targets may provide defined semantics when using the value
2132 as the operand to an inline assembly, but that is target specific.
2134 Constant Expressions
2135 --------------------
2137 Constant expressions are used to allow expressions involving other
2138 constants to be used as constants. Constant expressions may be of any
2139 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2140 that does not have side effects (e.g. load and call are not supported).
2141 The following is the syntax for constant expressions:
2143 ``trunc (CST to TYPE)``
2144 Truncate a constant to another type. The bit size of CST must be
2145 larger than the bit size of TYPE. Both types must be integers.
2146 ``zext (CST to TYPE)``
2147 Zero extend a constant to another type. The bit size of CST must be
2148 smaller than the bit size of TYPE. Both types must be integers.
2149 ``sext (CST to TYPE)``
2150 Sign extend a constant to another type. The bit size of CST must be
2151 smaller than the bit size of TYPE. Both types must be integers.
2152 ``fptrunc (CST to TYPE)``
2153 Truncate a floating point constant to another floating point type.
2154 The size of CST must be larger than the size of TYPE. Both types
2155 must be floating point.
2156 ``fpext (CST to TYPE)``
2157 Floating point extend a constant to another type. The size of CST
2158 must be smaller or equal to the size of TYPE. Both types must be
2160 ``fptoui (CST to TYPE)``
2161 Convert a floating point constant to the corresponding unsigned
2162 integer constant. TYPE must be a scalar or vector integer type. CST
2163 must be of scalar or vector floating point type. Both CST and TYPE
2164 must be scalars, or vectors of the same number of elements. If the
2165 value won't fit in the integer type, the results are undefined.
2166 ``fptosi (CST to TYPE)``
2167 Convert a floating point constant to the corresponding signed
2168 integer constant. TYPE must be a scalar or vector integer type. CST
2169 must be of scalar or vector floating point type. Both CST and TYPE
2170 must be scalars, or vectors of the same number of elements. If the
2171 value won't fit in the integer type, the results are undefined.
2172 ``uitofp (CST to TYPE)``
2173 Convert an unsigned integer constant to the corresponding floating
2174 point constant. TYPE must be a scalar or vector floating point type.
2175 CST must be of scalar or vector integer type. Both CST and TYPE must
2176 be scalars, or vectors of the same number of elements. If the value
2177 won't fit in the floating point type, the results are undefined.
2178 ``sitofp (CST to TYPE)``
2179 Convert a signed integer constant to the corresponding floating
2180 point constant. TYPE must be a scalar or vector floating point type.
2181 CST must be of scalar or vector integer type. Both CST and TYPE must
2182 be scalars, or vectors of the same number of elements. If the value
2183 won't fit in the floating point type, the results are undefined.
2184 ``ptrtoint (CST to TYPE)``
2185 Convert a pointer typed constant to the corresponding integer
2186 constant ``TYPE`` must be an integer type. ``CST`` must be of
2187 pointer type. The ``CST`` value is zero extended, truncated, or
2188 unchanged to make it fit in ``TYPE``.
2189 ``inttoptr (CST to TYPE)``
2190 Convert an integer constant to a pointer constant. TYPE must be a
2191 pointer type. CST must be of integer type. The CST value is zero
2192 extended, truncated, or unchanged to make it fit in a pointer size.
2193 This one is *really* dangerous!
2194 ``bitcast (CST to TYPE)``
2195 Convert a constant, CST, to another TYPE. The constraints of the
2196 operands are the same as those for the :ref:`bitcast
2197 instruction <i_bitcast>`.
2198 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2199 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2200 constants. As with the :ref:`getelementptr <i_getelementptr>`
2201 instruction, the index list may have zero or more indexes, which are
2202 required to make sense for the type of "CSTPTR".
2203 ``select (COND, VAL1, VAL2)``
2204 Perform the :ref:`select operation <i_select>` on constants.
2205 ``icmp COND (VAL1, VAL2)``
2206 Performs the :ref:`icmp operation <i_icmp>` on constants.
2207 ``fcmp COND (VAL1, VAL2)``
2208 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2209 ``extractelement (VAL, IDX)``
2210 Perform the :ref:`extractelement operation <i_extractelement>` on
2212 ``insertelement (VAL, ELT, IDX)``
2213 Perform the :ref:`insertelement operation <i_insertelement>` on
2215 ``shufflevector (VEC1, VEC2, IDXMASK)``
2216 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2218 ``extractvalue (VAL, IDX0, IDX1, ...)``
2219 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2220 constants. The index list is interpreted in a similar manner as
2221 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2222 least one index value must be specified.
2223 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2224 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2225 The index list is interpreted in a similar manner as indices in a
2226 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2227 value must be specified.
2228 ``OPCODE (LHS, RHS)``
2229 Perform the specified operation of the LHS and RHS constants. OPCODE
2230 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2231 binary <bitwiseops>` operations. The constraints on operands are
2232 the same as those for the corresponding instruction (e.g. no bitwise
2233 operations on floating point values are allowed).
2238 Inline Assembler Expressions
2239 ----------------------------
2241 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2242 Inline Assembly <moduleasm>`) through the use of a special value. This
2243 value represents the inline assembler as a string (containing the
2244 instructions to emit), a list of operand constraints (stored as a
2245 string), a flag that indicates whether or not the inline asm expression
2246 has side effects, and a flag indicating whether the function containing
2247 the asm needs to align its stack conservatively. An example inline
2248 assembler expression is:
2250 .. code-block:: llvm
2252 i32 (i32) asm "bswap $0", "=r,r"
2254 Inline assembler expressions may **only** be used as the callee operand
2255 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2256 Thus, typically we have:
2258 .. code-block:: llvm
2260 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2262 Inline asms with side effects not visible in the constraint list must be
2263 marked as having side effects. This is done through the use of the
2264 '``sideeffect``' keyword, like so:
2266 .. code-block:: llvm
2268 call void asm sideeffect "eieio", ""()
2270 In some cases inline asms will contain code that will not work unless
2271 the stack is aligned in some way, such as calls or SSE instructions on
2272 x86, yet will not contain code that does that alignment within the asm.
2273 The compiler should make conservative assumptions about what the asm
2274 might contain and should generate its usual stack alignment code in the
2275 prologue if the '``alignstack``' keyword is present:
2277 .. code-block:: llvm
2279 call void asm alignstack "eieio", ""()
2281 Inline asms also support using non-standard assembly dialects. The
2282 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2283 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2284 the only supported dialects. An example is:
2286 .. code-block:: llvm
2288 call void asm inteldialect "eieio", ""()
2290 If multiple keywords appear the '``sideeffect``' keyword must come
2291 first, the '``alignstack``' keyword second and the '``inteldialect``'
2297 The call instructions that wrap inline asm nodes may have a
2298 "``!srcloc``" MDNode attached to it that contains a list of constant
2299 integers. If present, the code generator will use the integer as the
2300 location cookie value when report errors through the ``LLVMContext``
2301 error reporting mechanisms. This allows a front-end to correlate backend
2302 errors that occur with inline asm back to the source code that produced
2305 .. code-block:: llvm
2307 call void asm sideeffect "something bad", ""(), !srcloc !42
2309 !42 = !{ i32 1234567 }
2311 It is up to the front-end to make sense of the magic numbers it places
2312 in the IR. If the MDNode contains multiple constants, the code generator
2313 will use the one that corresponds to the line of the asm that the error
2318 Metadata Nodes and Metadata Strings
2319 -----------------------------------
2321 LLVM IR allows metadata to be attached to instructions in the program
2322 that can convey extra information about the code to the optimizers and
2323 code generator. One example application of metadata is source-level
2324 debug information. There are two metadata primitives: strings and nodes.
2325 All metadata has the ``metadata`` type and is identified in syntax by a
2326 preceding exclamation point ('``!``').
2328 A metadata string is a string surrounded by double quotes. It can
2329 contain any character by escaping non-printable characters with
2330 "``\xx``" where "``xx``" is the two digit hex code. For example:
2333 Metadata nodes are represented with notation similar to structure
2334 constants (a comma separated list of elements, surrounded by braces and
2335 preceded by an exclamation point). Metadata nodes can have any values as
2336 their operand. For example:
2338 .. code-block:: llvm
2340 !{ metadata !"test\00", i32 10}
2342 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2343 metadata nodes, which can be looked up in the module symbol table. For
2346 .. code-block:: llvm
2348 !foo = metadata !{!4, !3}
2350 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2351 function is using two metadata arguments:
2353 .. code-block:: llvm
2355 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2357 Metadata can be attached with an instruction. Here metadata ``!21`` is
2358 attached to the ``add`` instruction using the ``!dbg`` identifier:
2360 .. code-block:: llvm
2362 %indvar.next = add i64 %indvar, 1, !dbg !21
2364 More information about specific metadata nodes recognized by the
2365 optimizers and code generator is found below.
2370 In LLVM IR, memory does not have types, so LLVM's own type system is not
2371 suitable for doing TBAA. Instead, metadata is added to the IR to
2372 describe a type system of a higher level language. This can be used to
2373 implement typical C/C++ TBAA, but it can also be used to implement
2374 custom alias analysis behavior for other languages.
2376 The current metadata format is very simple. TBAA metadata nodes have up
2377 to three fields, e.g.:
2379 .. code-block:: llvm
2381 !0 = metadata !{ metadata !"an example type tree" }
2382 !1 = metadata !{ metadata !"int", metadata !0 }
2383 !2 = metadata !{ metadata !"float", metadata !0 }
2384 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2386 The first field is an identity field. It can be any value, usually a
2387 metadata string, which uniquely identifies the type. The most important
2388 name in the tree is the name of the root node. Two trees with different
2389 root node names are entirely disjoint, even if they have leaves with
2392 The second field identifies the type's parent node in the tree, or is
2393 null or omitted for a root node. A type is considered to alias all of
2394 its descendants and all of its ancestors in the tree. Also, a type is
2395 considered to alias all types in other trees, so that bitcode produced
2396 from multiple front-ends is handled conservatively.
2398 If the third field is present, it's an integer which if equal to 1
2399 indicates that the type is "constant" (meaning
2400 ``pointsToConstantMemory`` should return true; see `other useful
2401 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2403 '``tbaa.struct``' Metadata
2404 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2406 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2407 aggregate assignment operations in C and similar languages, however it
2408 is defined to copy a contiguous region of memory, which is more than
2409 strictly necessary for aggregate types which contain holes due to
2410 padding. Also, it doesn't contain any TBAA information about the fields
2413 ``!tbaa.struct`` metadata can describe which memory subregions in a
2414 memcpy are padding and what the TBAA tags of the struct are.
2416 The current metadata format is very simple. ``!tbaa.struct`` metadata
2417 nodes are a list of operands which are in conceptual groups of three.
2418 For each group of three, the first operand gives the byte offset of a
2419 field in bytes, the second gives its size in bytes, and the third gives
2422 .. code-block:: llvm
2424 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2426 This describes a struct with two fields. The first is at offset 0 bytes
2427 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2428 and has size 4 bytes and has tbaa tag !2.
2430 Note that the fields need not be contiguous. In this example, there is a
2431 4 byte gap between the two fields. This gap represents padding which
2432 does not carry useful data and need not be preserved.
2434 '``fpmath``' Metadata
2435 ^^^^^^^^^^^^^^^^^^^^^
2437 ``fpmath`` metadata may be attached to any instruction of floating point
2438 type. It can be used to express the maximum acceptable error in the
2439 result of that instruction, in ULPs, thus potentially allowing the
2440 compiler to use a more efficient but less accurate method of computing
2441 it. ULP is defined as follows:
2443 If ``x`` is a real number that lies between two finite consecutive
2444 floating-point numbers ``a`` and ``b``, without being equal to one
2445 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2446 distance between the two non-equal finite floating-point numbers
2447 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2449 The metadata node shall consist of a single positive floating point
2450 number representing the maximum relative error, for example:
2452 .. code-block:: llvm
2454 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2456 '``range``' Metadata
2457 ^^^^^^^^^^^^^^^^^^^^
2459 ``range`` metadata may be attached only to loads of integer types. It
2460 expresses the possible ranges the loaded value is in. The ranges are
2461 represented with a flattened list of integers. The loaded value is known
2462 to be in the union of the ranges defined by each consecutive pair. Each
2463 pair has the following properties:
2465 - The type must match the type loaded by the instruction.
2466 - The pair ``a,b`` represents the range ``[a,b)``.
2467 - Both ``a`` and ``b`` are constants.
2468 - The range is allowed to wrap.
2469 - The range should not represent the full or empty set. That is,
2472 In addition, the pairs must be in signed order of the lower bound and
2473 they must be non-contiguous.
2477 .. code-block:: llvm
2479 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2480 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2481 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2482 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2484 !0 = metadata !{ i8 0, i8 2 }
2485 !1 = metadata !{ i8 255, i8 2 }
2486 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2487 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2489 Module Flags Metadata
2490 =====================
2492 Information about the module as a whole is difficult to convey to LLVM's
2493 subsystems. The LLVM IR isn't sufficient to transmit this information.
2494 The ``llvm.module.flags`` named metadata exists in order to facilitate
2495 this. These flags are in the form of key / value pairs --- much like a
2496 dictionary --- making it easy for any subsystem who cares about a flag to
2499 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2500 Each triplet has the following form:
2502 - The first element is a *behavior* flag, which specifies the behavior
2503 when two (or more) modules are merged together, and it encounters two
2504 (or more) metadata with the same ID. The supported behaviors are
2506 - The second element is a metadata string that is a unique ID for the
2507 metadata. Each module may only have one flag entry for each unique ID (not
2508 including entries with the **Require** behavior).
2509 - The third element is the value of the flag.
2511 When two (or more) modules are merged together, the resulting
2512 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2513 each unique metadata ID string, there will be exactly one entry in the merged
2514 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2515 be determined by the merge behavior flag, as described below. The only exception
2516 is that entries with the *Require* behavior are always preserved.
2518 The following behaviors are supported:
2529 Emits an error if two values disagree, otherwise the resulting value
2530 is that of the operands.
2534 Emits a warning if two values disagree. The result value will be the
2535 operand for the flag from the first module being linked.
2539 Adds a requirement that another module flag be present and have a
2540 specified value after linking is performed. The value must be a
2541 metadata pair, where the first element of the pair is the ID of the
2542 module flag to be restricted, and the second element of the pair is
2543 the value the module flag should be restricted to. This behavior can
2544 be used to restrict the allowable results (via triggering of an
2545 error) of linking IDs with the **Override** behavior.
2549 Uses the specified value, regardless of the behavior or value of the
2550 other module. If both modules specify **Override**, but the values
2551 differ, an error will be emitted.
2555 Appends the two values, which are required to be metadata nodes.
2559 Appends the two values, which are required to be metadata
2560 nodes. However, duplicate entries in the second list are dropped
2561 during the append operation.
2563 It is an error for a particular unique flag ID to have multiple behaviors,
2564 except in the case of **Require** (which adds restrictions on another metadata
2565 value) or **Override**.
2567 An example of module flags:
2569 .. code-block:: llvm
2571 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2572 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2573 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2574 !3 = metadata !{ i32 3, metadata !"qux",
2576 metadata !"foo", i32 1
2579 !llvm.module.flags = !{ !0, !1, !2, !3 }
2581 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2582 if two or more ``!"foo"`` flags are seen is to emit an error if their
2583 values are not equal.
2585 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2586 behavior if two or more ``!"bar"`` flags are seen is to use the value
2589 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2590 behavior if two or more ``!"qux"`` flags are seen is to emit a
2591 warning if their values are not equal.
2593 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2597 metadata !{ metadata !"foo", i32 1 }
2599 The behavior is to emit an error if the ``llvm.module.flags`` does not
2600 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2603 Objective-C Garbage Collection Module Flags Metadata
2604 ----------------------------------------------------
2606 On the Mach-O platform, Objective-C stores metadata about garbage
2607 collection in a special section called "image info". The metadata
2608 consists of a version number and a bitmask specifying what types of
2609 garbage collection are supported (if any) by the file. If two or more
2610 modules are linked together their garbage collection metadata needs to
2611 be merged rather than appended together.
2613 The Objective-C garbage collection module flags metadata consists of the
2614 following key-value pairs:
2623 * - ``Objective-C Version``
2624 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2626 * - ``Objective-C Image Info Version``
2627 - **[Required]** --- The version of the image info section. Currently
2630 * - ``Objective-C Image Info Section``
2631 - **[Required]** --- The section to place the metadata. Valid values are
2632 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2633 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2634 Objective-C ABI version 2.
2636 * - ``Objective-C Garbage Collection``
2637 - **[Required]** --- Specifies whether garbage collection is supported or
2638 not. Valid values are 0, for no garbage collection, and 2, for garbage
2639 collection supported.
2641 * - ``Objective-C GC Only``
2642 - **[Optional]** --- Specifies that only garbage collection is supported.
2643 If present, its value must be 6. This flag requires that the
2644 ``Objective-C Garbage Collection`` flag have the value 2.
2646 Some important flag interactions:
2648 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2649 merged with a module with ``Objective-C Garbage Collection`` set to
2650 2, then the resulting module has the
2651 ``Objective-C Garbage Collection`` flag set to 0.
2652 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2653 merged with a module with ``Objective-C GC Only`` set to 6.
2655 Automatic Linker Flags Module Flags Metadata
2656 --------------------------------------------
2658 Some targets support embedding flags to the linker inside individual object
2659 files. Typically this is used in conjunction with language extensions which
2660 allow source files to explicitly declare the libraries they depend on, and have
2661 these automatically be transmitted to the linker via object files.
2663 These flags are encoded in the IR using metadata in the module flags section,
2664 using the ``Linker Options`` key. The merge behavior for this flag is required
2665 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2666 node which should be a list of other metadata nodes, each of which should be a
2667 list of metadata strings defining linker options.
2669 For example, the following metadata section specifies two separate sets of
2670 linker options, presumably to link against ``libz`` and the ``Cocoa``
2673 !0 = metadata !{ i32 6, metadata !"Linker Options",
2675 metadata !{ metadata !"-lz" },
2676 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2677 !llvm.module.flags = !{ !0 }
2679 The metadata encoding as lists of lists of options, as opposed to a collapsed
2680 list of options, is chosen so that the IR encoding can use multiple option
2681 strings to specify e.g., a single library, while still having that specifier be
2682 preserved as an atomic element that can be recognized by a target specific
2683 assembly writer or object file emitter.
2685 Each individual option is required to be either a valid option for the target's
2686 linker, or an option that is reserved by the target specific assembly writer or
2687 object file emitter. No other aspect of these options is defined by the IR.
2689 Intrinsic Global Variables
2690 ==========================
2692 LLVM has a number of "magic" global variables that contain data that
2693 affect code generation or other IR semantics. These are documented here.
2694 All globals of this sort should have a section specified as
2695 "``llvm.metadata``". This section and all globals that start with
2696 "``llvm.``" are reserved for use by LLVM.
2698 The '``llvm.used``' Global Variable
2699 -----------------------------------
2701 The ``@llvm.used`` global is an array with i8\* element type which has
2702 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2703 pointers to global variables and functions which may optionally have a
2704 pointer cast formed of bitcast or getelementptr. For example, a legal
2707 .. code-block:: llvm
2712 @llvm.used = appending global [2 x i8*] [
2714 i8* bitcast (i32* @Y to i8*)
2715 ], section "llvm.metadata"
2717 If a global variable appears in the ``@llvm.used`` list, then the
2718 compiler, assembler, and linker are required to treat the symbol as if
2719 there is a reference to the global that it cannot see. For example, if a
2720 variable has internal linkage and no references other than that from the
2721 ``@llvm.used`` list, it cannot be deleted. This is commonly used to
2722 represent references from inline asms and other things the compiler
2723 cannot "see", and corresponds to "``attribute((used))``" in GNU C.
2725 On some targets, the code generator must emit a directive to the
2726 assembler or object file to prevent the assembler and linker from
2727 molesting the symbol.
2729 The '``llvm.compiler.used``' Global Variable
2730 --------------------------------------------
2732 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2733 directive, except that it only prevents the compiler from touching the
2734 symbol. On targets that support it, this allows an intelligent linker to
2735 optimize references to the symbol without being impeded as it would be
2738 This is a rare construct that should only be used in rare circumstances,
2739 and should not be exposed to source languages.
2741 The '``llvm.global_ctors``' Global Variable
2742 -------------------------------------------
2744 .. code-block:: llvm
2746 %0 = type { i32, void ()* }
2747 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2749 The ``@llvm.global_ctors`` array contains a list of constructor
2750 functions and associated priorities. The functions referenced by this
2751 array will be called in ascending order of priority (i.e. lowest first)
2752 when the module is loaded. The order of functions with the same priority
2755 The '``llvm.global_dtors``' Global Variable
2756 -------------------------------------------
2758 .. code-block:: llvm
2760 %0 = type { i32, void ()* }
2761 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2763 The ``@llvm.global_dtors`` array contains a list of destructor functions
2764 and associated priorities. The functions referenced by this array will
2765 be called in descending order of priority (i.e. highest first) when the
2766 module is loaded. The order of functions with the same priority is not
2769 Instruction Reference
2770 =====================
2772 The LLVM instruction set consists of several different classifications
2773 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
2774 instructions <binaryops>`, :ref:`bitwise binary
2775 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
2776 :ref:`other instructions <otherops>`.
2780 Terminator Instructions
2781 -----------------------
2783 As mentioned :ref:`previously <functionstructure>`, every basic block in a
2784 program ends with a "Terminator" instruction, which indicates which
2785 block should be executed after the current block is finished. These
2786 terminator instructions typically yield a '``void``' value: they produce
2787 control flow, not values (the one exception being the
2788 ':ref:`invoke <i_invoke>`' instruction).
2790 The terminator instructions are: ':ref:`ret <i_ret>`',
2791 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
2792 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
2793 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
2797 '``ret``' Instruction
2798 ^^^^^^^^^^^^^^^^^^^^^
2805 ret <type> <value> ; Return a value from a non-void function
2806 ret void ; Return from void function
2811 The '``ret``' instruction is used to return control flow (and optionally
2812 a value) from a function back to the caller.
2814 There are two forms of the '``ret``' instruction: one that returns a
2815 value and then causes control flow, and one that just causes control
2821 The '``ret``' instruction optionally accepts a single argument, the
2822 return value. The type of the return value must be a ':ref:`first
2823 class <t_firstclass>`' type.
2825 A function is not :ref:`well formed <wellformed>` if it it has a non-void
2826 return type and contains a '``ret``' instruction with no return value or
2827 a return value with a type that does not match its type, or if it has a
2828 void return type and contains a '``ret``' instruction with a return
2834 When the '``ret``' instruction is executed, control flow returns back to
2835 the calling function's context. If the caller is a
2836 ":ref:`call <i_call>`" instruction, execution continues at the
2837 instruction after the call. If the caller was an
2838 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
2839 beginning of the "normal" destination block. If the instruction returns
2840 a value, that value shall set the call or invoke instruction's return
2846 .. code-block:: llvm
2848 ret i32 5 ; Return an integer value of 5
2849 ret void ; Return from a void function
2850 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
2854 '``br``' Instruction
2855 ^^^^^^^^^^^^^^^^^^^^
2862 br i1 <cond>, label <iftrue>, label <iffalse>
2863 br label <dest> ; Unconditional branch
2868 The '``br``' instruction is used to cause control flow to transfer to a
2869 different basic block in the current function. There are two forms of
2870 this instruction, corresponding to a conditional branch and an
2871 unconditional branch.
2876 The conditional branch form of the '``br``' instruction takes a single
2877 '``i1``' value and two '``label``' values. The unconditional form of the
2878 '``br``' instruction takes a single '``label``' value as a target.
2883 Upon execution of a conditional '``br``' instruction, the '``i1``'
2884 argument is evaluated. If the value is ``true``, control flows to the
2885 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
2886 to the '``iffalse``' ``label`` argument.
2891 .. code-block:: llvm
2894 %cond = icmp eq i32 %a, %b
2895 br i1 %cond, label %IfEqual, label %IfUnequal
2903 '``switch``' Instruction
2904 ^^^^^^^^^^^^^^^^^^^^^^^^
2911 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2916 The '``switch``' instruction is used to transfer control flow to one of
2917 several different places. It is a generalization of the '``br``'
2918 instruction, allowing a branch to occur to one of many possible
2924 The '``switch``' instruction uses three parameters: an integer
2925 comparison value '``value``', a default '``label``' destination, and an
2926 array of pairs of comparison value constants and '``label``'s. The table
2927 is not allowed to contain duplicate constant entries.
2932 The ``switch`` instruction specifies a table of values and destinations.
2933 When the '``switch``' instruction is executed, this table is searched
2934 for the given value. If the value is found, control flow is transferred
2935 to the corresponding destination; otherwise, control flow is transferred
2936 to the default destination.
2941 Depending on properties of the target machine and the particular
2942 ``switch`` instruction, this instruction may be code generated in
2943 different ways. For example, it could be generated as a series of
2944 chained conditional branches or with a lookup table.
2949 .. code-block:: llvm
2951 ; Emulate a conditional br instruction
2952 %Val = zext i1 %value to i32
2953 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2955 ; Emulate an unconditional br instruction
2956 switch i32 0, label %dest [ ]
2958 ; Implement a jump table:
2959 switch i32 %val, label %otherwise [ i32 0, label %onzero
2961 i32 2, label %ontwo ]
2965 '``indirectbr``' Instruction
2966 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2973 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
2978 The '``indirectbr``' instruction implements an indirect branch to a
2979 label within the current function, whose address is specified by
2980 "``address``". Address must be derived from a
2981 :ref:`blockaddress <blockaddress>` constant.
2986 The '``address``' argument is the address of the label to jump to. The
2987 rest of the arguments indicate the full set of possible destinations
2988 that the address may point to. Blocks are allowed to occur multiple
2989 times in the destination list, though this isn't particularly useful.
2991 This destination list is required so that dataflow analysis has an
2992 accurate understanding of the CFG.
2997 Control transfers to the block specified in the address argument. All
2998 possible destination blocks must be listed in the label list, otherwise
2999 this instruction has undefined behavior. This implies that jumps to
3000 labels defined in other functions have undefined behavior as well.
3005 This is typically implemented with a jump through a register.
3010 .. code-block:: llvm
3012 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3016 '``invoke``' Instruction
3017 ^^^^^^^^^^^^^^^^^^^^^^^^
3024 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3025 to label <normal label> unwind label <exception label>
3030 The '``invoke``' instruction causes control to transfer to a specified
3031 function, with the possibility of control flow transfer to either the
3032 '``normal``' label or the '``exception``' label. If the callee function
3033 returns with the "``ret``" instruction, control flow will return to the
3034 "normal" label. If the callee (or any indirect callees) returns via the
3035 ":ref:`resume <i_resume>`" instruction or other exception handling
3036 mechanism, control is interrupted and continued at the dynamically
3037 nearest "exception" label.
3039 The '``exception``' label is a `landing
3040 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3041 '``exception``' label is required to have the
3042 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3043 information about the behavior of the program after unwinding happens,
3044 as its first non-PHI instruction. The restrictions on the
3045 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3046 instruction, so that the important information contained within the
3047 "``landingpad``" instruction can't be lost through normal code motion.
3052 This instruction requires several arguments:
3054 #. The optional "cconv" marker indicates which :ref:`calling
3055 convention <callingconv>` the call should use. If none is
3056 specified, the call defaults to using C calling conventions.
3057 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3058 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3060 #. '``ptr to function ty``': shall be the signature of the pointer to
3061 function value being invoked. In most cases, this is a direct
3062 function invocation, but indirect ``invoke``'s are just as possible,
3063 branching off an arbitrary pointer to function value.
3064 #. '``function ptr val``': An LLVM value containing a pointer to a
3065 function to be invoked.
3066 #. '``function args``': argument list whose types match the function
3067 signature argument types and parameter attributes. All arguments must
3068 be of :ref:`first class <t_firstclass>` type. If the function signature
3069 indicates the function accepts a variable number of arguments, the
3070 extra arguments can be specified.
3071 #. '``normal label``': the label reached when the called function
3072 executes a '``ret``' instruction.
3073 #. '``exception label``': the label reached when a callee returns via
3074 the :ref:`resume <i_resume>` instruction or other exception handling
3076 #. The optional :ref:`function attributes <fnattrs>` list. Only
3077 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3078 attributes are valid here.
3083 This instruction is designed to operate as a standard '``call``'
3084 instruction in most regards. The primary difference is that it
3085 establishes an association with a label, which is used by the runtime
3086 library to unwind the stack.
3088 This instruction is used in languages with destructors to ensure that
3089 proper cleanup is performed in the case of either a ``longjmp`` or a
3090 thrown exception. Additionally, this is important for implementation of
3091 '``catch``' clauses in high-level languages that support them.
3093 For the purposes of the SSA form, the definition of the value returned
3094 by the '``invoke``' instruction is deemed to occur on the edge from the
3095 current block to the "normal" label. If the callee unwinds then no
3096 return value is available.
3101 .. code-block:: llvm
3103 %retval = invoke i32 @Test(i32 15) to label %Continue
3104 unwind label %TestCleanup ; {i32}:retval set
3105 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3106 unwind label %TestCleanup ; {i32}:retval set
3110 '``resume``' Instruction
3111 ^^^^^^^^^^^^^^^^^^^^^^^^
3118 resume <type> <value>
3123 The '``resume``' instruction is a terminator instruction that has no
3129 The '``resume``' instruction requires one argument, which must have the
3130 same type as the result of any '``landingpad``' instruction in the same
3136 The '``resume``' instruction resumes propagation of an existing
3137 (in-flight) exception whose unwinding was interrupted with a
3138 :ref:`landingpad <i_landingpad>` instruction.
3143 .. code-block:: llvm
3145 resume { i8*, i32 } %exn
3149 '``unreachable``' Instruction
3150 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3162 The '``unreachable``' instruction has no defined semantics. This
3163 instruction is used to inform the optimizer that a particular portion of
3164 the code is not reachable. This can be used to indicate that the code
3165 after a no-return function cannot be reached, and other facts.
3170 The '``unreachable``' instruction has no defined semantics.
3177 Binary operators are used to do most of the computation in a program.
3178 They require two operands of the same type, execute an operation on
3179 them, and produce a single value. The operands might represent multiple
3180 data, as is the case with the :ref:`vector <t_vector>` data type. The
3181 result value has the same type as its operands.
3183 There are several different binary operators:
3187 '``add``' Instruction
3188 ^^^^^^^^^^^^^^^^^^^^^
3195 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3196 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3197 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3198 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3203 The '``add``' instruction returns the sum of its two operands.
3208 The two arguments to the '``add``' instruction must be
3209 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3210 arguments must have identical types.
3215 The value produced is the integer sum of the two operands.
3217 If the sum has unsigned overflow, the result returned is the
3218 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3221 Because LLVM integers use a two's complement representation, this
3222 instruction is appropriate for both signed and unsigned integers.
3224 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3225 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3226 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3227 unsigned and/or signed overflow, respectively, occurs.
3232 .. code-block:: llvm
3234 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3238 '``fadd``' Instruction
3239 ^^^^^^^^^^^^^^^^^^^^^^
3246 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3251 The '``fadd``' instruction returns the sum of its two operands.
3256 The two arguments to the '``fadd``' instruction must be :ref:`floating
3257 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3258 Both arguments must have identical types.
3263 The value produced is the floating point sum of the two operands. This
3264 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3265 which are optimization hints to enable otherwise unsafe floating point
3271 .. code-block:: llvm
3273 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3275 '``sub``' Instruction
3276 ^^^^^^^^^^^^^^^^^^^^^
3283 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3284 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3285 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3286 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3291 The '``sub``' instruction returns the difference of its two operands.
3293 Note that the '``sub``' instruction is used to represent the '``neg``'
3294 instruction present in most other intermediate representations.
3299 The two arguments to the '``sub``' instruction must be
3300 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3301 arguments must have identical types.
3306 The value produced is the integer difference of the two operands.
3308 If the difference has unsigned overflow, the result returned is the
3309 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3312 Because LLVM integers use a two's complement representation, this
3313 instruction is appropriate for both signed and unsigned integers.
3315 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3316 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3317 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3318 unsigned and/or signed overflow, respectively, occurs.
3323 .. code-block:: llvm
3325 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3326 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3330 '``fsub``' Instruction
3331 ^^^^^^^^^^^^^^^^^^^^^^
3338 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3343 The '``fsub``' instruction returns the difference of its two operands.
3345 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3346 instruction present in most other intermediate representations.
3351 The two arguments to the '``fsub``' instruction must be :ref:`floating
3352 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3353 Both arguments must have identical types.
3358 The value produced is the floating point difference of the two operands.
3359 This instruction can also take any number of :ref:`fast-math
3360 flags <fastmath>`, which are optimization hints to enable otherwise
3361 unsafe floating point optimizations:
3366 .. code-block:: llvm
3368 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3369 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3371 '``mul``' Instruction
3372 ^^^^^^^^^^^^^^^^^^^^^
3379 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3380 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3381 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3382 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3387 The '``mul``' instruction returns the product of its two operands.
3392 The two arguments to the '``mul``' instruction must be
3393 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3394 arguments must have identical types.
3399 The value produced is the integer product of the two operands.
3401 If the result of the multiplication has unsigned overflow, the result
3402 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3403 bit width of the result.
3405 Because LLVM integers use a two's complement representation, and the
3406 result is the same width as the operands, this instruction returns the
3407 correct result for both signed and unsigned integers. If a full product
3408 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3409 sign-extended or zero-extended as appropriate to the width of the full
3412 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3413 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3414 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3415 unsigned and/or signed overflow, respectively, occurs.
3420 .. code-block:: llvm
3422 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3426 '``fmul``' Instruction
3427 ^^^^^^^^^^^^^^^^^^^^^^
3434 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3439 The '``fmul``' instruction returns the product of its two operands.
3444 The two arguments to the '``fmul``' instruction must be :ref:`floating
3445 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3446 Both arguments must have identical types.
3451 The value produced is the floating point product of the two operands.
3452 This instruction can also take any number of :ref:`fast-math
3453 flags <fastmath>`, which are optimization hints to enable otherwise
3454 unsafe floating point optimizations:
3459 .. code-block:: llvm
3461 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3463 '``udiv``' Instruction
3464 ^^^^^^^^^^^^^^^^^^^^^^
3471 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3472 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3477 The '``udiv``' instruction returns the quotient of its two operands.
3482 The two arguments to the '``udiv``' instruction must be
3483 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3484 arguments must have identical types.
3489 The value produced is the unsigned integer quotient of the two operands.
3491 Note that unsigned integer division and signed integer division are
3492 distinct operations; for signed integer division, use '``sdiv``'.
3494 Division by zero leads to undefined behavior.
3496 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3497 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3498 such, "((a udiv exact b) mul b) == a").
3503 .. code-block:: llvm
3505 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3507 '``sdiv``' Instruction
3508 ^^^^^^^^^^^^^^^^^^^^^^
3515 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3516 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3521 The '``sdiv``' instruction returns the quotient of its two operands.
3526 The two arguments to the '``sdiv``' instruction must be
3527 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3528 arguments must have identical types.
3533 The value produced is the signed integer quotient of the two operands
3534 rounded towards zero.
3536 Note that signed integer division and unsigned integer division are
3537 distinct operations; for unsigned integer division, use '``udiv``'.
3539 Division by zero leads to undefined behavior. Overflow also leads to
3540 undefined behavior; this is a rare case, but can occur, for example, by
3541 doing a 32-bit division of -2147483648 by -1.
3543 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3544 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3549 .. code-block:: llvm
3551 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3555 '``fdiv``' Instruction
3556 ^^^^^^^^^^^^^^^^^^^^^^
3563 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3568 The '``fdiv``' instruction returns the quotient of its two operands.
3573 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3574 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3575 Both arguments must have identical types.
3580 The value produced is the floating point quotient of the two operands.
3581 This instruction can also take any number of :ref:`fast-math
3582 flags <fastmath>`, which are optimization hints to enable otherwise
3583 unsafe floating point optimizations:
3588 .. code-block:: llvm
3590 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3592 '``urem``' Instruction
3593 ^^^^^^^^^^^^^^^^^^^^^^
3600 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3605 The '``urem``' instruction returns the remainder from the unsigned
3606 division of its two arguments.
3611 The two arguments to the '``urem``' instruction must be
3612 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3613 arguments must have identical types.
3618 This instruction returns the unsigned integer *remainder* of a division.
3619 This instruction always performs an unsigned division to get the
3622 Note that unsigned integer remainder and signed integer remainder are
3623 distinct operations; for signed integer remainder, use '``srem``'.
3625 Taking the remainder of a division by zero leads to undefined behavior.
3630 .. code-block:: llvm
3632 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3634 '``srem``' Instruction
3635 ^^^^^^^^^^^^^^^^^^^^^^
3642 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3647 The '``srem``' instruction returns the remainder from the signed
3648 division of its two operands. This instruction can also take
3649 :ref:`vector <t_vector>` versions of the values in which case the elements
3655 The two arguments to the '``srem``' instruction must be
3656 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3657 arguments must have identical types.
3662 This instruction returns the *remainder* of a division (where the result
3663 is either zero or has the same sign as the dividend, ``op1``), not the
3664 *modulo* operator (where the result is either zero or has the same sign
3665 as the divisor, ``op2``) of a value. For more information about the
3666 difference, see `The Math
3667 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3668 table of how this is implemented in various languages, please see
3670 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3672 Note that signed integer remainder and unsigned integer remainder are
3673 distinct operations; for unsigned integer remainder, use '``urem``'.
3675 Taking the remainder of a division by zero leads to undefined behavior.
3676 Overflow also leads to undefined behavior; this is a rare case, but can
3677 occur, for example, by taking the remainder of a 32-bit division of
3678 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3679 rule lets srem be implemented using instructions that return both the
3680 result of the division and the remainder.)
3685 .. code-block:: llvm
3687 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3691 '``frem``' Instruction
3692 ^^^^^^^^^^^^^^^^^^^^^^
3699 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3704 The '``frem``' instruction returns the remainder from the division of
3710 The two arguments to the '``frem``' instruction must be :ref:`floating
3711 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3712 Both arguments must have identical types.
3717 This instruction returns the *remainder* of a division. The remainder
3718 has the same sign as the dividend. This instruction can also take any
3719 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3720 to enable otherwise unsafe floating point optimizations:
3725 .. code-block:: llvm
3727 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3731 Bitwise Binary Operations
3732 -------------------------
3734 Bitwise binary operators are used to do various forms of bit-twiddling
3735 in a program. They are generally very efficient instructions and can
3736 commonly be strength reduced from other instructions. They require two
3737 operands of the same type, execute an operation on them, and produce a
3738 single value. The resulting value is the same type as its operands.
3740 '``shl``' Instruction
3741 ^^^^^^^^^^^^^^^^^^^^^
3748 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
3749 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
3750 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
3751 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3756 The '``shl``' instruction returns the first operand shifted to the left
3757 a specified number of bits.
3762 Both arguments to the '``shl``' instruction must be the same
3763 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3764 '``op2``' is treated as an unsigned value.
3769 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
3770 where ``n`` is the width of the result. If ``op2`` is (statically or
3771 dynamically) negative or equal to or larger than the number of bits in
3772 ``op1``, the result is undefined. If the arguments are vectors, each
3773 vector element of ``op1`` is shifted by the corresponding shift amount
3776 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
3777 value <poisonvalues>` if it shifts out any non-zero bits. If the
3778 ``nsw`` keyword is present, then the shift produces a :ref:`poison
3779 value <poisonvalues>` if it shifts out any bits that disagree with the
3780 resultant sign bit. As such, NUW/NSW have the same semantics as they
3781 would if the shift were expressed as a mul instruction with the same
3782 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
3787 .. code-block:: llvm
3789 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
3790 <result> = shl i32 4, 2 ; yields {i32}: 16
3791 <result> = shl i32 1, 10 ; yields {i32}: 1024
3792 <result> = shl i32 1, 32 ; undefined
3793 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
3795 '``lshr``' Instruction
3796 ^^^^^^^^^^^^^^^^^^^^^^
3803 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
3804 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
3809 The '``lshr``' instruction (logical shift right) returns the first
3810 operand shifted to the right a specified number of bits with zero fill.
3815 Both arguments to the '``lshr``' instruction must be the same
3816 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3817 '``op2``' is treated as an unsigned value.
3822 This instruction always performs a logical shift right operation. The
3823 most significant bits of the result will be filled with zero bits after
3824 the shift. If ``op2`` is (statically or dynamically) equal to or larger
3825 than the number of bits in ``op1``, the result is undefined. If the
3826 arguments are vectors, each vector element of ``op1`` is shifted by the
3827 corresponding shift amount in ``op2``.
3829 If the ``exact`` keyword is present, the result value of the ``lshr`` is
3830 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3836 .. code-block:: llvm
3838 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
3839 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
3840 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
3841 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7FFFFFFF
3842 <result> = lshr i32 1, 32 ; undefined
3843 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
3845 '``ashr``' Instruction
3846 ^^^^^^^^^^^^^^^^^^^^^^
3853 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
3854 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
3859 The '``ashr``' instruction (arithmetic shift right) returns the first
3860 operand shifted to the right a specified number of bits with sign
3866 Both arguments to the '``ashr``' instruction must be the same
3867 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3868 '``op2``' is treated as an unsigned value.
3873 This instruction always performs an arithmetic shift right operation,
3874 The most significant bits of the result will be filled with the sign bit
3875 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
3876 than the number of bits in ``op1``, the result is undefined. If the
3877 arguments are vectors, each vector element of ``op1`` is shifted by the
3878 corresponding shift amount in ``op2``.
3880 If the ``exact`` keyword is present, the result value of the ``ashr`` is
3881 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3887 .. code-block:: llvm
3889 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
3890 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
3891 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
3892 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
3893 <result> = ashr i32 1, 32 ; undefined
3894 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
3896 '``and``' Instruction
3897 ^^^^^^^^^^^^^^^^^^^^^
3904 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
3909 The '``and``' instruction returns the bitwise logical and of its two
3915 The two arguments to the '``and``' instruction must be
3916 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3917 arguments must have identical types.
3922 The truth table used for the '``and``' instruction is:
3939 .. code-block:: llvm
3941 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
3942 <result> = and i32 15, 40 ; yields {i32}:result = 8
3943 <result> = and i32 4, 8 ; yields {i32}:result = 0
3945 '``or``' Instruction
3946 ^^^^^^^^^^^^^^^^^^^^
3953 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
3958 The '``or``' instruction returns the bitwise logical inclusive or of its
3964 The two arguments to the '``or``' instruction must be
3965 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3966 arguments must have identical types.
3971 The truth table used for the '``or``' instruction is:
3990 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
3991 <result> = or i32 15, 40 ; yields {i32}:result = 47
3992 <result> = or i32 4, 8 ; yields {i32}:result = 12
3994 '``xor``' Instruction
3995 ^^^^^^^^^^^^^^^^^^^^^
4002 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4007 The '``xor``' instruction returns the bitwise logical exclusive or of
4008 its two operands. The ``xor`` is used to implement the "one's
4009 complement" operation, which is the "~" operator in C.
4014 The two arguments to the '``xor``' instruction must be
4015 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4016 arguments must have identical types.
4021 The truth table used for the '``xor``' instruction is:
4038 .. code-block:: llvm
4040 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4041 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4042 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4043 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4048 LLVM supports several instructions to represent vector operations in a
4049 target-independent manner. These instructions cover the element-access
4050 and vector-specific operations needed to process vectors effectively.
4051 While LLVM does directly support these vector operations, many
4052 sophisticated algorithms will want to use target-specific intrinsics to
4053 take full advantage of a specific target.
4055 .. _i_extractelement:
4057 '``extractelement``' Instruction
4058 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4065 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4070 The '``extractelement``' instruction extracts a single scalar element
4071 from a vector at a specified index.
4076 The first operand of an '``extractelement``' instruction is a value of
4077 :ref:`vector <t_vector>` type. The second operand is an index indicating
4078 the position from which to extract the element. The index may be a
4084 The result is a scalar of the same type as the element type of ``val``.
4085 Its value is the value at position ``idx`` of ``val``. If ``idx``
4086 exceeds the length of ``val``, the results are undefined.
4091 .. code-block:: llvm
4093 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4095 .. _i_insertelement:
4097 '``insertelement``' Instruction
4098 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4105 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4110 The '``insertelement``' instruction inserts a scalar element into a
4111 vector at a specified index.
4116 The first operand of an '``insertelement``' instruction is a value of
4117 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4118 type must equal the element type of the first operand. The third operand
4119 is an index indicating the position at which to insert the value. The
4120 index may be a variable.
4125 The result is a vector of the same type as ``val``. Its element values
4126 are those of ``val`` except at position ``idx``, where it gets the value
4127 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4133 .. code-block:: llvm
4135 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4137 .. _i_shufflevector:
4139 '``shufflevector``' Instruction
4140 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4147 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4152 The '``shufflevector``' instruction constructs a permutation of elements
4153 from two input vectors, returning a vector with the same element type as
4154 the input and length that is the same as the shuffle mask.
4159 The first two operands of a '``shufflevector``' instruction are vectors
4160 with the same type. The third argument is a shuffle mask whose element
4161 type is always 'i32'. The result of the instruction is a vector whose
4162 length is the same as the shuffle mask and whose element type is the
4163 same as the element type of the first two operands.
4165 The shuffle mask operand is required to be a constant vector with either
4166 constant integer or undef values.
4171 The elements of the two input vectors are numbered from left to right
4172 across both of the vectors. The shuffle mask operand specifies, for each
4173 element of the result vector, which element of the two input vectors the
4174 result element gets. The element selector may be undef (meaning "don't
4175 care") and the second operand may be undef if performing a shuffle from
4181 .. code-block:: llvm
4183 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4184 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4185 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4186 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4187 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4188 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4189 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4190 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4192 Aggregate Operations
4193 --------------------
4195 LLVM supports several instructions for working with
4196 :ref:`aggregate <t_aggregate>` values.
4200 '``extractvalue``' Instruction
4201 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4208 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4213 The '``extractvalue``' instruction extracts the value of a member field
4214 from an :ref:`aggregate <t_aggregate>` value.
4219 The first operand of an '``extractvalue``' instruction is a value of
4220 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4221 constant indices to specify which value to extract in a similar manner
4222 as indices in a '``getelementptr``' instruction.
4224 The major differences to ``getelementptr`` indexing are:
4226 - Since the value being indexed is not a pointer, the first index is
4227 omitted and assumed to be zero.
4228 - At least one index must be specified.
4229 - Not only struct indices but also array indices must be in bounds.
4234 The result is the value at the position in the aggregate specified by
4240 .. code-block:: llvm
4242 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4246 '``insertvalue``' Instruction
4247 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4254 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4259 The '``insertvalue``' instruction inserts a value into a member field in
4260 an :ref:`aggregate <t_aggregate>` value.
4265 The first operand of an '``insertvalue``' instruction is a value of
4266 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4267 a first-class value to insert. The following operands are constant
4268 indices indicating the position at which to insert the value in a
4269 similar manner as indices in a '``extractvalue``' instruction. The value
4270 to insert must have the same type as the value identified by the
4276 The result is an aggregate of the same type as ``val``. Its value is
4277 that of ``val`` except that the value at the position specified by the
4278 indices is that of ``elt``.
4283 .. code-block:: llvm
4285 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4286 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4287 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4291 Memory Access and Addressing Operations
4292 ---------------------------------------
4294 A key design point of an SSA-based representation is how it represents
4295 memory. In LLVM, no memory locations are in SSA form, which makes things
4296 very simple. This section describes how to read, write, and allocate
4301 '``alloca``' Instruction
4302 ^^^^^^^^^^^^^^^^^^^^^^^^
4309 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4314 The '``alloca``' instruction allocates memory on the stack frame of the
4315 currently executing function, to be automatically released when this
4316 function returns to its caller. The object is always allocated in the
4317 generic address space (address space zero).
4322 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4323 bytes of memory on the runtime stack, returning a pointer of the
4324 appropriate type to the program. If "NumElements" is specified, it is
4325 the number of elements allocated, otherwise "NumElements" is defaulted
4326 to be one. If a constant alignment is specified, the value result of the
4327 allocation is guaranteed to be aligned to at least that boundary. If not
4328 specified, or if zero, the target can choose to align the allocation on
4329 any convenient boundary compatible with the type.
4331 '``type``' may be any sized type.
4336 Memory is allocated; a pointer is returned. The operation is undefined
4337 if there is insufficient stack space for the allocation. '``alloca``'d
4338 memory is automatically released when the function returns. The
4339 '``alloca``' instruction is commonly used to represent automatic
4340 variables that must have an address available. When the function returns
4341 (either with the ``ret`` or ``resume`` instructions), the memory is
4342 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4343 The order in which memory is allocated (ie., which way the stack grows)
4349 .. code-block:: llvm
4351 %ptr = alloca i32 ; yields {i32*}:ptr
4352 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4353 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4354 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4358 '``load``' Instruction
4359 ^^^^^^^^^^^^^^^^^^^^^^
4366 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4367 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4368 !<index> = !{ i32 1 }
4373 The '``load``' instruction is used to read from memory.
4378 The argument to the '``load``' instruction specifies the memory address
4379 from which to load. The pointer must point to a :ref:`first
4380 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4381 then the optimizer is not allowed to modify the number or order of
4382 execution of this ``load`` with other :ref:`volatile
4383 operations <volatile>`.
4385 If the ``load`` is marked as ``atomic``, it takes an extra
4386 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4387 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4388 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4389 when they may see multiple atomic stores. The type of the pointee must
4390 be an integer type whose bit width is a power of two greater than or
4391 equal to eight and less than or equal to a target-specific size limit.
4392 ``align`` must be explicitly specified on atomic loads, and the load has
4393 undefined behavior if the alignment is not set to a value which is at
4394 least the size in bytes of the pointee. ``!nontemporal`` does not have
4395 any defined semantics for atomic loads.
4397 The optional constant ``align`` argument specifies the alignment of the
4398 operation (that is, the alignment of the memory address). A value of 0
4399 or an omitted ``align`` argument means that the operation has the abi
4400 alignment for the target. It is the responsibility of the code emitter
4401 to ensure that the alignment information is correct. Overestimating the
4402 alignment results in undefined behavior. Underestimating the alignment
4403 may produce less efficient code. An alignment of 1 is always safe.
4405 The optional ``!nontemporal`` metadata must reference a single
4406 metatadata name <index> corresponding to a metadata node with one
4407 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4408 metatadata on the instruction tells the optimizer and code generator
4409 that this load is not expected to be reused in the cache. The code
4410 generator may select special instructions to save cache bandwidth, such
4411 as the ``MOVNT`` instruction on x86.
4413 The optional ``!invariant.load`` metadata must reference a single
4414 metatadata name <index> corresponding to a metadata node with no
4415 entries. The existence of the ``!invariant.load`` metatadata on the
4416 instruction tells the optimizer and code generator that this load
4417 address points to memory which does not change value during program
4418 execution. The optimizer may then move this load around, for example, by
4419 hoisting it out of loops using loop invariant code motion.
4424 The location of memory pointed to is loaded. If the value being loaded
4425 is of scalar type then the number of bytes read does not exceed the
4426 minimum number of bytes needed to hold all bits of the type. For
4427 example, loading an ``i24`` reads at most three bytes. When loading a
4428 value of a type like ``i20`` with a size that is not an integral number
4429 of bytes, the result is undefined if the value was not originally
4430 written using a store of the same type.
4435 .. code-block:: llvm
4437 %ptr = alloca i32 ; yields {i32*}:ptr
4438 store i32 3, i32* %ptr ; yields {void}
4439 %val = load i32* %ptr ; yields {i32}:val = i32 3
4443 '``store``' Instruction
4444 ^^^^^^^^^^^^^^^^^^^^^^^
4451 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4452 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4457 The '``store``' instruction is used to write to memory.
4462 There are two arguments to the '``store``' instruction: a value to store
4463 and an address at which to store it. The type of the '``<pointer>``'
4464 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4465 the '``<value>``' operand. If the ``store`` is marked as ``volatile``,
4466 then the optimizer is not allowed to modify the number or order of
4467 execution of this ``store`` with other :ref:`volatile
4468 operations <volatile>`.
4470 If the ``store`` is marked as ``atomic``, it takes an extra
4471 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4472 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4473 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4474 when they may see multiple atomic stores. The type of the pointee must
4475 be an integer type whose bit width is a power of two greater than or
4476 equal to eight and less than or equal to a target-specific size limit.
4477 ``align`` must be explicitly specified on atomic stores, and the store
4478 has undefined behavior if the alignment is not set to a value which is
4479 at least the size in bytes of the pointee. ``!nontemporal`` does not
4480 have any defined semantics for atomic stores.
4482 The optional constant "align" argument specifies the alignment of the
4483 operation (that is, the alignment of the memory address). A value of 0
4484 or an omitted "align" argument means that the operation has the abi
4485 alignment for the target. It is the responsibility of the code emitter
4486 to ensure that the alignment information is correct. Overestimating the
4487 alignment results in an undefined behavior. Underestimating the
4488 alignment may produce less efficient code. An alignment of 1 is always
4491 The optional !nontemporal metadata must reference a single metatadata
4492 name <index> corresponding to a metadata node with one i32 entry of
4493 value 1. The existence of the !nontemporal metatadata on the instruction
4494 tells the optimizer and code generator that this load is not expected to
4495 be reused in the cache. The code generator may select special
4496 instructions to save cache bandwidth, such as the MOVNT instruction on
4502 The contents of memory are updated to contain '``<value>``' at the
4503 location specified by the '``<pointer>``' operand. If '``<value>``' is
4504 of scalar type then the number of bytes written does not exceed the
4505 minimum number of bytes needed to hold all bits of the type. For
4506 example, storing an ``i24`` writes at most three bytes. When writing a
4507 value of a type like ``i20`` with a size that is not an integral number
4508 of bytes, it is unspecified what happens to the extra bits that do not
4509 belong to the type, but they will typically be overwritten.
4514 .. code-block:: llvm
4516 %ptr = alloca i32 ; yields {i32*}:ptr
4517 store i32 3, i32* %ptr ; yields {void}
4518 %val = load i32* %ptr ; yields {i32}:val = i32 3
4522 '``fence``' Instruction
4523 ^^^^^^^^^^^^^^^^^^^^^^^
4530 fence [singlethread] <ordering> ; yields {void}
4535 The '``fence``' instruction is used to introduce happens-before edges
4541 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4542 defines what *synchronizes-with* edges they add. They can only be given
4543 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4548 A fence A which has (at least) ``release`` ordering semantics
4549 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4550 semantics if and only if there exist atomic operations X and Y, both
4551 operating on some atomic object M, such that A is sequenced before X, X
4552 modifies M (either directly or through some side effect of a sequence
4553 headed by X), Y is sequenced before B, and Y observes M. This provides a
4554 *happens-before* dependency between A and B. Rather than an explicit
4555 ``fence``, one (but not both) of the atomic operations X or Y might
4556 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4557 still *synchronize-with* the explicit ``fence`` and establish the
4558 *happens-before* edge.
4560 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4561 ``acquire`` and ``release`` semantics specified above, participates in
4562 the global program order of other ``seq_cst`` operations and/or fences.
4564 The optional ":ref:`singlethread <singlethread>`" argument specifies
4565 that the fence only synchronizes with other fences in the same thread.
4566 (This is useful for interacting with signal handlers.)
4571 .. code-block:: llvm
4573 fence acquire ; yields {void}
4574 fence singlethread seq_cst ; yields {void}
4578 '``cmpxchg``' Instruction
4579 ^^^^^^^^^^^^^^^^^^^^^^^^^
4586 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4591 The '``cmpxchg``' instruction is used to atomically modify memory. It
4592 loads a value in memory and compares it to a given value. If they are
4593 equal, it stores a new value into the memory.
4598 There are three arguments to the '``cmpxchg``' instruction: an address
4599 to operate on, a value to compare to the value currently be at that
4600 address, and a new value to place at that address if the compared values
4601 are equal. The type of '<cmp>' must be an integer type whose bit width
4602 is a power of two greater than or equal to eight and less than or equal
4603 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4604 type, and the type of '<pointer>' must be a pointer to that type. If the
4605 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4606 to modify the number or order of execution of this ``cmpxchg`` with
4607 other :ref:`volatile operations <volatile>`.
4609 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4610 synchronizes with other atomic operations.
4612 The optional "``singlethread``" argument declares that the ``cmpxchg``
4613 is only atomic with respect to code (usually signal handlers) running in
4614 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4615 respect to all other code in the system.
4617 The pointer passed into cmpxchg must have alignment greater than or
4618 equal to the size in memory of the operand.
4623 The contents of memory at the location specified by the '``<pointer>``'
4624 operand is read and compared to '``<cmp>``'; if the read value is the
4625 equal, '``<new>``' is written. The original value at the location is
4628 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4629 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4630 atomic load with an ordering parameter determined by dropping any
4631 ``release`` part of the ``cmpxchg``'s ordering.
4636 .. code-block:: llvm
4639 %orig = atomic load i32* %ptr unordered ; yields {i32}
4643 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4644 %squared = mul i32 %cmp, %cmp
4645 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4646 %success = icmp eq i32 %cmp, %old
4647 br i1 %success, label %done, label %loop
4654 '``atomicrmw``' Instruction
4655 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4662 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4667 The '``atomicrmw``' instruction is used to atomically modify memory.
4672 There are three arguments to the '``atomicrmw``' instruction: an
4673 operation to apply, an address whose value to modify, an argument to the
4674 operation. The operation must be one of the following keywords:
4688 The type of '<value>' must be an integer type whose bit width is a power
4689 of two greater than or equal to eight and less than or equal to a
4690 target-specific size limit. The type of the '``<pointer>``' operand must
4691 be a pointer to that type. If the ``atomicrmw`` is marked as
4692 ``volatile``, then the optimizer is not allowed to modify the number or
4693 order of execution of this ``atomicrmw`` with other :ref:`volatile
4694 operations <volatile>`.
4699 The contents of memory at the location specified by the '``<pointer>``'
4700 operand are atomically read, modified, and written back. The original
4701 value at the location is returned. The modification is specified by the
4704 - xchg: ``*ptr = val``
4705 - add: ``*ptr = *ptr + val``
4706 - sub: ``*ptr = *ptr - val``
4707 - and: ``*ptr = *ptr & val``
4708 - nand: ``*ptr = ~(*ptr & val)``
4709 - or: ``*ptr = *ptr | val``
4710 - xor: ``*ptr = *ptr ^ val``
4711 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4712 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4713 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4715 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4721 .. code-block:: llvm
4723 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4725 .. _i_getelementptr:
4727 '``getelementptr``' Instruction
4728 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4735 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4736 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4737 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
4742 The '``getelementptr``' instruction is used to get the address of a
4743 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
4744 address calculation only and does not access memory.
4749 The first argument is always a pointer or a vector of pointers, and
4750 forms the basis of the calculation. The remaining arguments are indices
4751 that indicate which of the elements of the aggregate object are indexed.
4752 The interpretation of each index is dependent on the type being indexed
4753 into. The first index always indexes the pointer value given as the
4754 first argument, the second index indexes a value of the type pointed to
4755 (not necessarily the value directly pointed to, since the first index
4756 can be non-zero), etc. The first type indexed into must be a pointer
4757 value, subsequent types can be arrays, vectors, and structs. Note that
4758 subsequent types being indexed into can never be pointers, since that
4759 would require loading the pointer before continuing calculation.
4761 The type of each index argument depends on the type it is indexing into.
4762 When indexing into a (optionally packed) structure, only ``i32`` integer
4763 **constants** are allowed (when using a vector of indices they must all
4764 be the **same** ``i32`` integer constant). When indexing into an array,
4765 pointer or vector, integers of any width are allowed, and they are not
4766 required to be constant. These integers are treated as signed values
4769 For example, let's consider a C code fragment and how it gets compiled
4785 int *foo(struct ST *s) {
4786 return &s[1].Z.B[5][13];
4789 The LLVM code generated by Clang is:
4791 .. code-block:: llvm
4793 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
4794 %struct.ST = type { i32, double, %struct.RT }
4796 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
4798 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
4805 In the example above, the first index is indexing into the
4806 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
4807 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
4808 indexes into the third element of the structure, yielding a
4809 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
4810 structure. The third index indexes into the second element of the
4811 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
4812 dimensions of the array are subscripted into, yielding an '``i32``'
4813 type. The '``getelementptr``' instruction returns a pointer to this
4814 element, thus computing a value of '``i32*``' type.
4816 Note that it is perfectly legal to index partially through a structure,
4817 returning a pointer to an inner element. Because of this, the LLVM code
4818 for the given testcase is equivalent to:
4820 .. code-block:: llvm
4822 define i32* @foo(%struct.ST* %s) {
4823 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
4824 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
4825 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
4826 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
4827 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
4831 If the ``inbounds`` keyword is present, the result value of the
4832 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
4833 pointer is not an *in bounds* address of an allocated object, or if any
4834 of the addresses that would be formed by successive addition of the
4835 offsets implied by the indices to the base address with infinitely
4836 precise signed arithmetic are not an *in bounds* address of that
4837 allocated object. The *in bounds* addresses for an allocated object are
4838 all the addresses that point into the object, plus the address one byte
4839 past the end. In cases where the base is a vector of pointers the
4840 ``inbounds`` keyword applies to each of the computations element-wise.
4842 If the ``inbounds`` keyword is not present, the offsets are added to the
4843 base address with silently-wrapping two's complement arithmetic. If the
4844 offsets have a different width from the pointer, they are sign-extended
4845 or truncated to the width of the pointer. The result value of the
4846 ``getelementptr`` may be outside the object pointed to by the base
4847 pointer. The result value may not necessarily be used to access memory
4848 though, even if it happens to point into allocated storage. See the
4849 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
4852 The getelementptr instruction is often confusing. For some more insight
4853 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
4858 .. code-block:: llvm
4860 ; yields [12 x i8]*:aptr
4861 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4863 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4865 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4867 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4869 In cases where the pointer argument is a vector of pointers, each index
4870 must be a vector with the same number of elements. For example:
4872 .. code-block:: llvm
4874 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
4876 Conversion Operations
4877 ---------------------
4879 The instructions in this category are the conversion instructions
4880 (casting) which all take a single operand and a type. They perform
4881 various bit conversions on the operand.
4883 '``trunc .. to``' Instruction
4884 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4891 <result> = trunc <ty> <value> to <ty2> ; yields ty2
4896 The '``trunc``' instruction truncates its operand to the type ``ty2``.
4901 The '``trunc``' instruction takes a value to trunc, and a type to trunc
4902 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
4903 of the same number of integers. The bit size of the ``value`` must be
4904 larger than the bit size of the destination type, ``ty2``. Equal sized
4905 types are not allowed.
4910 The '``trunc``' instruction truncates the high order bits in ``value``
4911 and converts the remaining bits to ``ty2``. Since the source size must
4912 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
4913 It will always truncate bits.
4918 .. code-block:: llvm
4920 %X = trunc i32 257 to i8 ; yields i8:1
4921 %Y = trunc i32 123 to i1 ; yields i1:true
4922 %Z = trunc i32 122 to i1 ; yields i1:false
4923 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
4925 '``zext .. to``' Instruction
4926 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4933 <result> = zext <ty> <value> to <ty2> ; yields ty2
4938 The '``zext``' instruction zero extends its operand to type ``ty2``.
4943 The '``zext``' instruction takes a value to cast, and a type to cast it
4944 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
4945 the same number of integers. The bit size of the ``value`` must be
4946 smaller than the bit size of the destination type, ``ty2``.
4951 The ``zext`` fills the high order bits of the ``value`` with zero bits
4952 until it reaches the size of the destination type, ``ty2``.
4954 When zero extending from i1, the result will always be either 0 or 1.
4959 .. code-block:: llvm
4961 %X = zext i32 257 to i64 ; yields i64:257
4962 %Y = zext i1 true to i32 ; yields i32:1
4963 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
4965 '``sext .. to``' Instruction
4966 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4973 <result> = sext <ty> <value> to <ty2> ; yields ty2
4978 The '``sext``' sign extends ``value`` to the type ``ty2``.
4983 The '``sext``' instruction takes a value to cast, and a type to cast it
4984 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
4985 the same number of integers. The bit size of the ``value`` must be
4986 smaller than the bit size of the destination type, ``ty2``.
4991 The '``sext``' instruction performs a sign extension by copying the sign
4992 bit (highest order bit) of the ``value`` until it reaches the bit size
4993 of the type ``ty2``.
4995 When sign extending from i1, the extension always results in -1 or 0.
5000 .. code-block:: llvm
5002 %X = sext i8 -1 to i16 ; yields i16 :65535
5003 %Y = sext i1 true to i32 ; yields i32:-1
5004 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5006 '``fptrunc .. to``' Instruction
5007 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5014 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5019 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5024 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5025 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5026 The size of ``value`` must be larger than the size of ``ty2``. This
5027 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5032 The '``fptrunc``' instruction truncates a ``value`` from a larger
5033 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5034 point <t_floating>` type. If the value cannot fit within the
5035 destination type, ``ty2``, then the results are undefined.
5040 .. code-block:: llvm
5042 %X = fptrunc double 123.0 to float ; yields float:123.0
5043 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5045 '``fpext .. to``' Instruction
5046 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5053 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5058 The '``fpext``' extends a floating point ``value`` to a larger floating
5064 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5065 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5066 to. The source type must be smaller than the destination type.
5071 The '``fpext``' instruction extends the ``value`` from a smaller
5072 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5073 point <t_floating>` type. The ``fpext`` cannot be used to make a
5074 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5075 *no-op cast* for a floating point cast.
5080 .. code-block:: llvm
5082 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5083 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5085 '``fptoui .. to``' Instruction
5086 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5093 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5098 The '``fptoui``' converts a floating point ``value`` to its unsigned
5099 integer equivalent of type ``ty2``.
5104 The '``fptoui``' instruction takes a value to cast, which must be a
5105 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5106 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5107 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5108 type with the same number of elements as ``ty``
5113 The '``fptoui``' instruction converts its :ref:`floating
5114 point <t_floating>` operand into the nearest (rounding towards zero)
5115 unsigned integer value. If the value cannot fit in ``ty2``, the results
5121 .. code-block:: llvm
5123 %X = fptoui double 123.0 to i32 ; yields i32:123
5124 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5125 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5127 '``fptosi .. to``' Instruction
5128 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5135 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5140 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5141 ``value`` to type ``ty2``.
5146 The '``fptosi``' instruction takes a value to cast, which must be a
5147 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5148 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5149 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5150 type with the same number of elements as ``ty``
5155 The '``fptosi``' instruction converts its :ref:`floating
5156 point <t_floating>` operand into the nearest (rounding towards zero)
5157 signed integer value. If the value cannot fit in ``ty2``, the results
5163 .. code-block:: llvm
5165 %X = fptosi double -123.0 to i32 ; yields i32:-123
5166 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5167 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5169 '``uitofp .. to``' Instruction
5170 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5177 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5182 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5183 and converts that value to the ``ty2`` type.
5188 The '``uitofp``' instruction takes a value to cast, which must be a
5189 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5190 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5191 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5192 type with the same number of elements as ``ty``
5197 The '``uitofp``' instruction interprets its operand as an unsigned
5198 integer quantity and converts it to the corresponding floating point
5199 value. If the value cannot fit in the floating point value, the results
5205 .. code-block:: llvm
5207 %X = uitofp i32 257 to float ; yields float:257.0
5208 %Y = uitofp i8 -1 to double ; yields double:255.0
5210 '``sitofp .. to``' Instruction
5211 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5218 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5223 The '``sitofp``' instruction regards ``value`` as a signed integer and
5224 converts that value to the ``ty2`` type.
5229 The '``sitofp``' instruction takes a value to cast, which must be a
5230 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5231 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5232 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5233 type with the same number of elements as ``ty``
5238 The '``sitofp``' instruction interprets its operand as a signed integer
5239 quantity and converts it to the corresponding floating point value. If
5240 the value cannot fit in the floating point value, the results are
5246 .. code-block:: llvm
5248 %X = sitofp i32 257 to float ; yields float:257.0
5249 %Y = sitofp i8 -1 to double ; yields double:-1.0
5253 '``ptrtoint .. to``' Instruction
5254 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5261 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5266 The '``ptrtoint``' instruction converts the pointer or a vector of
5267 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5272 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5273 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5274 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5275 a vector of integers type.
5280 The '``ptrtoint``' instruction converts ``value`` to integer type
5281 ``ty2`` by interpreting the pointer value as an integer and either
5282 truncating or zero extending that value to the size of the integer type.
5283 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5284 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5285 the same size, then nothing is done (*no-op cast*) other than a type
5291 .. code-block:: llvm
5293 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5294 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5295 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5299 '``inttoptr .. to``' Instruction
5300 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5307 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5312 The '``inttoptr``' instruction converts an integer ``value`` to a
5313 pointer type, ``ty2``.
5318 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5319 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5325 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5326 applying either a zero extension or a truncation depending on the size
5327 of the integer ``value``. If ``value`` is larger than the size of a
5328 pointer then a truncation is done. If ``value`` is smaller than the size
5329 of a pointer then a zero extension is done. If they are the same size,
5330 nothing is done (*no-op cast*).
5335 .. code-block:: llvm
5337 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5338 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5339 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5340 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5344 '``bitcast .. to``' Instruction
5345 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5352 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5357 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5363 The '``bitcast``' instruction takes a value to cast, which must be a
5364 non-aggregate first class value, and a type to cast it to, which must
5365 also be a non-aggregate :ref:`first class <t_firstclass>` type. The bit
5366 sizes of ``value`` and the destination type, ``ty2``, must be identical.
5367 If the source type is a pointer, the destination type must also be a
5368 pointer. This instruction supports bitwise conversion of vectors to
5369 integers and to vectors of other types (as long as they have the same
5375 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It is
5376 always a *no-op cast* because no bits change with this conversion. The
5377 conversion is done as if the ``value`` had been stored to memory and
5378 read back as type ``ty2``. Pointer (or vector of pointers) types may
5379 only be converted to other pointer (or vector of pointers) types with
5380 this instruction. To convert pointers to other types, use the
5381 :ref:`inttoptr <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions
5387 .. code-block:: llvm
5389 %X = bitcast i8 255 to i8 ; yields i8 :-1
5390 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5391 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5392 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5399 The instructions in this category are the "miscellaneous" instructions,
5400 which defy better classification.
5404 '``icmp``' Instruction
5405 ^^^^^^^^^^^^^^^^^^^^^^
5412 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5417 The '``icmp``' instruction returns a boolean value or a vector of
5418 boolean values based on comparison of its two integer, integer vector,
5419 pointer, or pointer vector operands.
5424 The '``icmp``' instruction takes three operands. The first operand is
5425 the condition code indicating the kind of comparison to perform. It is
5426 not a value, just a keyword. The possible condition code are:
5429 #. ``ne``: not equal
5430 #. ``ugt``: unsigned greater than
5431 #. ``uge``: unsigned greater or equal
5432 #. ``ult``: unsigned less than
5433 #. ``ule``: unsigned less or equal
5434 #. ``sgt``: signed greater than
5435 #. ``sge``: signed greater or equal
5436 #. ``slt``: signed less than
5437 #. ``sle``: signed less or equal
5439 The remaining two arguments must be :ref:`integer <t_integer>` or
5440 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5441 must also be identical types.
5446 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5447 code given as ``cond``. The comparison performed always yields either an
5448 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5450 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5451 otherwise. No sign interpretation is necessary or performed.
5452 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5453 otherwise. No sign interpretation is necessary or performed.
5454 #. ``ugt``: interprets the operands as unsigned values and yields
5455 ``true`` if ``op1`` is greater than ``op2``.
5456 #. ``uge``: interprets the operands as unsigned values and yields
5457 ``true`` if ``op1`` is greater than or equal to ``op2``.
5458 #. ``ult``: interprets the operands as unsigned values and yields
5459 ``true`` if ``op1`` is less than ``op2``.
5460 #. ``ule``: interprets the operands as unsigned values and yields
5461 ``true`` if ``op1`` is less than or equal to ``op2``.
5462 #. ``sgt``: interprets the operands as signed values and yields ``true``
5463 if ``op1`` is greater than ``op2``.
5464 #. ``sge``: interprets the operands as signed values and yields ``true``
5465 if ``op1`` is greater than or equal to ``op2``.
5466 #. ``slt``: interprets the operands as signed values and yields ``true``
5467 if ``op1`` is less than ``op2``.
5468 #. ``sle``: interprets the operands as signed values and yields ``true``
5469 if ``op1`` is less than or equal to ``op2``.
5471 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5472 are compared as if they were integers.
5474 If the operands are integer vectors, then they are compared element by
5475 element. The result is an ``i1`` vector with the same number of elements
5476 as the values being compared. Otherwise, the result is an ``i1``.
5481 .. code-block:: llvm
5483 <result> = icmp eq i32 4, 5 ; yields: result=false
5484 <result> = icmp ne float* %X, %X ; yields: result=false
5485 <result> = icmp ult i16 4, 5 ; yields: result=true
5486 <result> = icmp sgt i16 4, 5 ; yields: result=false
5487 <result> = icmp ule i16 -4, 5 ; yields: result=false
5488 <result> = icmp sge i16 4, 5 ; yields: result=false
5490 Note that the code generator does not yet support vector types with the
5491 ``icmp`` instruction.
5495 '``fcmp``' Instruction
5496 ^^^^^^^^^^^^^^^^^^^^^^
5503 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5508 The '``fcmp``' instruction returns a boolean value or vector of boolean
5509 values based on comparison of its operands.
5511 If the operands are floating point scalars, then the result type is a
5512 boolean (:ref:`i1 <t_integer>`).
5514 If the operands are floating point vectors, then the result type is a
5515 vector of boolean with the same number of elements as the operands being
5521 The '``fcmp``' instruction takes three operands. The first operand is
5522 the condition code indicating the kind of comparison to perform. It is
5523 not a value, just a keyword. The possible condition code are:
5525 #. ``false``: no comparison, always returns false
5526 #. ``oeq``: ordered and equal
5527 #. ``ogt``: ordered and greater than
5528 #. ``oge``: ordered and greater than or equal
5529 #. ``olt``: ordered and less than
5530 #. ``ole``: ordered and less than or equal
5531 #. ``one``: ordered and not equal
5532 #. ``ord``: ordered (no nans)
5533 #. ``ueq``: unordered or equal
5534 #. ``ugt``: unordered or greater than
5535 #. ``uge``: unordered or greater than or equal
5536 #. ``ult``: unordered or less than
5537 #. ``ule``: unordered or less than or equal
5538 #. ``une``: unordered or not equal
5539 #. ``uno``: unordered (either nans)
5540 #. ``true``: no comparison, always returns true
5542 *Ordered* means that neither operand is a QNAN while *unordered* means
5543 that either operand may be a QNAN.
5545 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5546 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5547 type. They must have identical types.
5552 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5553 condition code given as ``cond``. If the operands are vectors, then the
5554 vectors are compared element by element. Each comparison performed
5555 always yields an :ref:`i1 <t_integer>` result, as follows:
5557 #. ``false``: always yields ``false``, regardless of operands.
5558 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5559 is equal to ``op2``.
5560 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5561 is greater than ``op2``.
5562 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5563 is greater than or equal to ``op2``.
5564 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5565 is less than ``op2``.
5566 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5567 is less than or equal to ``op2``.
5568 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5569 is not equal to ``op2``.
5570 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5571 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5573 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5574 greater than ``op2``.
5575 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5576 greater than or equal to ``op2``.
5577 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5579 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5580 less than or equal to ``op2``.
5581 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5582 not equal to ``op2``.
5583 #. ``uno``: yields ``true`` if either operand is a QNAN.
5584 #. ``true``: always yields ``true``, regardless of operands.
5589 .. code-block:: llvm
5591 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5592 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5593 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5594 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5596 Note that the code generator does not yet support vector types with the
5597 ``fcmp`` instruction.
5601 '``phi``' Instruction
5602 ^^^^^^^^^^^^^^^^^^^^^
5609 <result> = phi <ty> [ <val0>, <label0>], ...
5614 The '``phi``' instruction is used to implement the φ node in the SSA
5615 graph representing the function.
5620 The type of the incoming values is specified with the first type field.
5621 After this, the '``phi``' instruction takes a list of pairs as
5622 arguments, with one pair for each predecessor basic block of the current
5623 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5624 the value arguments to the PHI node. Only labels may be used as the
5627 There must be no non-phi instructions between the start of a basic block
5628 and the PHI instructions: i.e. PHI instructions must be first in a basic
5631 For the purposes of the SSA form, the use of each incoming value is
5632 deemed to occur on the edge from the corresponding predecessor block to
5633 the current block (but after any definition of an '``invoke``'
5634 instruction's return value on the same edge).
5639 At runtime, the '``phi``' instruction logically takes on the value
5640 specified by the pair corresponding to the predecessor basic block that
5641 executed just prior to the current block.
5646 .. code-block:: llvm
5648 Loop: ; Infinite loop that counts from 0 on up...
5649 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5650 %nextindvar = add i32 %indvar, 1
5655 '``select``' Instruction
5656 ^^^^^^^^^^^^^^^^^^^^^^^^
5663 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5665 selty is either i1 or {<N x i1>}
5670 The '``select``' instruction is used to choose one value based on a
5671 condition, without branching.
5676 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5677 values indicating the condition, and two values of the same :ref:`first
5678 class <t_firstclass>` type. If the val1/val2 are vectors and the
5679 condition is a scalar, then entire vectors are selected, not individual
5685 If the condition is an i1 and it evaluates to 1, the instruction returns
5686 the first value argument; otherwise, it returns the second value
5689 If the condition is a vector of i1, then the value arguments must be
5690 vectors of the same size, and the selection is done element by element.
5695 .. code-block:: llvm
5697 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5701 '``call``' Instruction
5702 ^^^^^^^^^^^^^^^^^^^^^^
5709 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5714 The '``call``' instruction represents a simple function call.
5719 This instruction requires several arguments:
5721 #. The optional "tail" marker indicates that the callee function does
5722 not access any allocas or varargs in the caller. Note that calls may
5723 be marked "tail" even if they do not occur before a
5724 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5725 function call is eligible for tail call optimization, but `might not
5726 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5727 The code generator may optimize calls marked "tail" with either 1)
5728 automatic `sibling call
5729 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5730 callee have matching signatures, or 2) forced tail call optimization
5731 when the following extra requirements are met:
5733 - Caller and callee both have the calling convention ``fastcc``.
5734 - The call is in tail position (ret immediately follows call and ret
5735 uses value of call or is void).
5736 - Option ``-tailcallopt`` is enabled, or
5737 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5738 - `Platform specific constraints are
5739 met. <CodeGenerator.html#tailcallopt>`_
5741 #. The optional "cconv" marker indicates which :ref:`calling
5742 convention <callingconv>` the call should use. If none is
5743 specified, the call defaults to using C calling conventions. The
5744 calling convention of the call must match the calling convention of
5745 the target function, or else the behavior is undefined.
5746 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5747 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5749 #. '``ty``': the type of the call instruction itself which is also the
5750 type of the return value. Functions that return no value are marked
5752 #. '``fnty``': shall be the signature of the pointer to function value
5753 being invoked. The argument types must match the types implied by
5754 this signature. This type can be omitted if the function is not
5755 varargs and if the function type does not return a pointer to a
5757 #. '``fnptrval``': An LLVM value containing a pointer to a function to
5758 be invoked. In most cases, this is a direct function invocation, but
5759 indirect ``call``'s are just as possible, calling an arbitrary pointer
5761 #. '``function args``': argument list whose types match the function
5762 signature argument types and parameter attributes. All arguments must
5763 be of :ref:`first class <t_firstclass>` type. If the function signature
5764 indicates the function accepts a variable number of arguments, the
5765 extra arguments can be specified.
5766 #. The optional :ref:`function attributes <fnattrs>` list. Only
5767 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5768 attributes are valid here.
5773 The '``call``' instruction is used to cause control flow to transfer to
5774 a specified function, with its incoming arguments bound to the specified
5775 values. Upon a '``ret``' instruction in the called function, control
5776 flow continues with the instruction after the function call, and the
5777 return value of the function is bound to the result argument.
5782 .. code-block:: llvm
5784 %retval = call i32 @test(i32 %argc)
5785 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
5786 %X = tail call i32 @foo() ; yields i32
5787 %Y = tail call fastcc i32 @foo() ; yields i32
5788 call void %foo(i8 97 signext)
5790 %struct.A = type { i32, i8 }
5791 %r = call %struct.A @foo() ; yields { 32, i8 }
5792 %gr = extractvalue %struct.A %r, 0 ; yields i32
5793 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
5794 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
5795 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
5797 llvm treats calls to some functions with names and arguments that match
5798 the standard C99 library as being the C99 library functions, and may
5799 perform optimizations or generate code for them under that assumption.
5800 This is something we'd like to change in the future to provide better
5801 support for freestanding environments and non-C-based languages.
5805 '``va_arg``' Instruction
5806 ^^^^^^^^^^^^^^^^^^^^^^^^
5813 <resultval> = va_arg <va_list*> <arglist>, <argty>
5818 The '``va_arg``' instruction is used to access arguments passed through
5819 the "variable argument" area of a function call. It is used to implement
5820 the ``va_arg`` macro in C.
5825 This instruction takes a ``va_list*`` value and the type of the
5826 argument. It returns a value of the specified argument type and
5827 increments the ``va_list`` to point to the next argument. The actual
5828 type of ``va_list`` is target specific.
5833 The '``va_arg``' instruction loads an argument of the specified type
5834 from the specified ``va_list`` and causes the ``va_list`` to point to
5835 the next argument. For more information, see the variable argument
5836 handling :ref:`Intrinsic Functions <int_varargs>`.
5838 It is legal for this instruction to be called in a function which does
5839 not take a variable number of arguments, for example, the ``vfprintf``
5842 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
5843 function <intrinsics>` because it takes a type as an argument.
5848 See the :ref:`variable argument processing <int_varargs>` section.
5850 Note that the code generator does not yet fully support va\_arg on many
5851 targets. Also, it does not currently support va\_arg with aggregate
5852 types on any target.
5856 '``landingpad``' Instruction
5857 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5864 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
5865 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
5867 <clause> := catch <type> <value>
5868 <clause> := filter <array constant type> <array constant>
5873 The '``landingpad``' instruction is used by `LLVM's exception handling
5874 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5875 is a landing pad --- one where the exception lands, and corresponds to the
5876 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
5877 defines values supplied by the personality function (``pers_fn``) upon
5878 re-entry to the function. The ``resultval`` has the type ``resultty``.
5883 This instruction takes a ``pers_fn`` value. This is the personality
5884 function associated with the unwinding mechanism. The optional
5885 ``cleanup`` flag indicates that the landing pad block is a cleanup.
5887 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
5888 contains the global variable representing the "type" that may be caught
5889 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
5890 clause takes an array constant as its argument. Use
5891 "``[0 x i8**] undef``" for a filter which cannot throw. The
5892 '``landingpad``' instruction must contain *at least* one ``clause`` or
5893 the ``cleanup`` flag.
5898 The '``landingpad``' instruction defines the values which are set by the
5899 personality function (``pers_fn``) upon re-entry to the function, and
5900 therefore the "result type" of the ``landingpad`` instruction. As with
5901 calling conventions, how the personality function results are
5902 represented in LLVM IR is target specific.
5904 The clauses are applied in order from top to bottom. If two
5905 ``landingpad`` instructions are merged together through inlining, the
5906 clauses from the calling function are appended to the list of clauses.
5907 When the call stack is being unwound due to an exception being thrown,
5908 the exception is compared against each ``clause`` in turn. If it doesn't
5909 match any of the clauses, and the ``cleanup`` flag is not set, then
5910 unwinding continues further up the call stack.
5912 The ``landingpad`` instruction has several restrictions:
5914 - A landing pad block is a basic block which is the unwind destination
5915 of an '``invoke``' instruction.
5916 - A landing pad block must have a '``landingpad``' instruction as its
5917 first non-PHI instruction.
5918 - There can be only one '``landingpad``' instruction within the landing
5920 - A basic block that is not a landing pad block may not include a
5921 '``landingpad``' instruction.
5922 - All '``landingpad``' instructions in a function must have the same
5923 personality function.
5928 .. code-block:: llvm
5930 ;; A landing pad which can catch an integer.
5931 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5933 ;; A landing pad that is a cleanup.
5934 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5936 ;; A landing pad which can catch an integer and can only throw a double.
5937 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5939 filter [1 x i8**] [@_ZTId]
5946 LLVM supports the notion of an "intrinsic function". These functions
5947 have well known names and semantics and are required to follow certain
5948 restrictions. Overall, these intrinsics represent an extension mechanism
5949 for the LLVM language that does not require changing all of the
5950 transformations in LLVM when adding to the language (or the bitcode
5951 reader/writer, the parser, etc...).
5953 Intrinsic function names must all start with an "``llvm.``" prefix. This
5954 prefix is reserved in LLVM for intrinsic names; thus, function names may
5955 not begin with this prefix. Intrinsic functions must always be external
5956 functions: you cannot define the body of intrinsic functions. Intrinsic
5957 functions may only be used in call or invoke instructions: it is illegal
5958 to take the address of an intrinsic function. Additionally, because
5959 intrinsic functions are part of the LLVM language, it is required if any
5960 are added that they be documented here.
5962 Some intrinsic functions can be overloaded, i.e., the intrinsic
5963 represents a family of functions that perform the same operation but on
5964 different data types. Because LLVM can represent over 8 million
5965 different integer types, overloading is used commonly to allow an
5966 intrinsic function to operate on any integer type. One or more of the
5967 argument types or the result type can be overloaded to accept any
5968 integer type. Argument types may also be defined as exactly matching a
5969 previous argument's type or the result type. This allows an intrinsic
5970 function which accepts multiple arguments, but needs all of them to be
5971 of the same type, to only be overloaded with respect to a single
5972 argument or the result.
5974 Overloaded intrinsics will have the names of its overloaded argument
5975 types encoded into its function name, each preceded by a period. Only
5976 those types which are overloaded result in a name suffix. Arguments
5977 whose type is matched against another type do not. For example, the
5978 ``llvm.ctpop`` function can take an integer of any width and returns an
5979 integer of exactly the same integer width. This leads to a family of
5980 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
5981 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
5982 overloaded, and only one type suffix is required. Because the argument's
5983 type is matched against the return type, it does not require its own
5986 To learn how to add an intrinsic function, please see the `Extending
5987 LLVM Guide <ExtendingLLVM.html>`_.
5991 Variable Argument Handling Intrinsics
5992 -------------------------------------
5994 Variable argument support is defined in LLVM with the
5995 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
5996 functions. These functions are related to the similarly named macros
5997 defined in the ``<stdarg.h>`` header file.
5999 All of these functions operate on arguments that use a target-specific
6000 value type "``va_list``". The LLVM assembly language reference manual
6001 does not define what this type is, so all transformations should be
6002 prepared to handle these functions regardless of the type used.
6004 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6005 variable argument handling intrinsic functions are used.
6007 .. code-block:: llvm
6009 define i32 @test(i32 %X, ...) {
6010 ; Initialize variable argument processing
6012 %ap2 = bitcast i8** %ap to i8*
6013 call void @llvm.va_start(i8* %ap2)
6015 ; Read a single integer argument
6016 %tmp = va_arg i8** %ap, i32
6018 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6020 %aq2 = bitcast i8** %aq to i8*
6021 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6022 call void @llvm.va_end(i8* %aq2)
6024 ; Stop processing of arguments.
6025 call void @llvm.va_end(i8* %ap2)
6029 declare void @llvm.va_start(i8*)
6030 declare void @llvm.va_copy(i8*, i8*)
6031 declare void @llvm.va_end(i8*)
6035 '``llvm.va_start``' Intrinsic
6036 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6043 declare void %llvm.va_start(i8* <arglist>)
6048 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6049 subsequent use by ``va_arg``.
6054 The argument is a pointer to a ``va_list`` element to initialize.
6059 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6060 available in C. In a target-dependent way, it initializes the
6061 ``va_list`` element to which the argument points, so that the next call
6062 to ``va_arg`` will produce the first variable argument passed to the
6063 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6064 to know the last argument of the function as the compiler can figure
6067 '``llvm.va_end``' Intrinsic
6068 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6075 declare void @llvm.va_end(i8* <arglist>)
6080 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6081 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6086 The argument is a pointer to a ``va_list`` to destroy.
6091 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6092 available in C. In a target-dependent way, it destroys the ``va_list``
6093 element to which the argument points. Calls to
6094 :ref:`llvm.va_start <int_va_start>` and
6095 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6100 '``llvm.va_copy``' Intrinsic
6101 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6108 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6113 The '``llvm.va_copy``' intrinsic copies the current argument position
6114 from the source argument list to the destination argument list.
6119 The first argument is a pointer to a ``va_list`` element to initialize.
6120 The second argument is a pointer to a ``va_list`` element to copy from.
6125 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6126 available in C. In a target-dependent way, it copies the source
6127 ``va_list`` element into the destination ``va_list`` element. This
6128 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6129 arbitrarily complex and require, for example, memory allocation.
6131 Accurate Garbage Collection Intrinsics
6132 --------------------------------------
6134 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6135 (GC) requires the implementation and generation of these intrinsics.
6136 These intrinsics allow identification of :ref:`GC roots on the
6137 stack <int_gcroot>`, as well as garbage collector implementations that
6138 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6139 Front-ends for type-safe garbage collected languages should generate
6140 these intrinsics to make use of the LLVM garbage collectors. For more
6141 details, see `Accurate Garbage Collection with
6142 LLVM <GarbageCollection.html>`_.
6144 The garbage collection intrinsics only operate on objects in the generic
6145 address space (address space zero).
6149 '``llvm.gcroot``' Intrinsic
6150 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6157 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6162 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6163 the code generator, and allows some metadata to be associated with it.
6168 The first argument specifies the address of a stack object that contains
6169 the root pointer. The second pointer (which must be either a constant or
6170 a global value address) contains the meta-data to be associated with the
6176 At runtime, a call to this intrinsic stores a null pointer into the
6177 "ptrloc" location. At compile-time, the code generator generates
6178 information to allow the runtime to find the pointer at GC safe points.
6179 The '``llvm.gcroot``' intrinsic may only be used in a function which
6180 :ref:`specifies a GC algorithm <gc>`.
6184 '``llvm.gcread``' Intrinsic
6185 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6192 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6197 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6198 locations, allowing garbage collector implementations that require read
6204 The second argument is the address to read from, which should be an
6205 address allocated from the garbage collector. The first object is a
6206 pointer to the start of the referenced object, if needed by the language
6207 runtime (otherwise null).
6212 The '``llvm.gcread``' intrinsic has the same semantics as a load
6213 instruction, but may be replaced with substantially more complex code by
6214 the garbage collector runtime, as needed. The '``llvm.gcread``'
6215 intrinsic may only be used in a function which :ref:`specifies a GC
6220 '``llvm.gcwrite``' Intrinsic
6221 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6228 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6233 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6234 locations, allowing garbage collector implementations that require write
6235 barriers (such as generational or reference counting collectors).
6240 The first argument is the reference to store, the second is the start of
6241 the object to store it to, and the third is the address of the field of
6242 Obj to store to. If the runtime does not require a pointer to the
6243 object, Obj may be null.
6248 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6249 instruction, but may be replaced with substantially more complex code by
6250 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6251 intrinsic may only be used in a function which :ref:`specifies a GC
6254 Code Generator Intrinsics
6255 -------------------------
6257 These intrinsics are provided by LLVM to expose special features that
6258 may only be implemented with code generator support.
6260 '``llvm.returnaddress``' Intrinsic
6261 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6268 declare i8 *@llvm.returnaddress(i32 <level>)
6273 The '``llvm.returnaddress``' intrinsic attempts to compute a
6274 target-specific value indicating the return address of the current
6275 function or one of its callers.
6280 The argument to this intrinsic indicates which function to return the
6281 address for. Zero indicates the calling function, one indicates its
6282 caller, etc. The argument is **required** to be a constant integer
6288 The '``llvm.returnaddress``' intrinsic either returns a pointer
6289 indicating the return address of the specified call frame, or zero if it
6290 cannot be identified. The value returned by this intrinsic is likely to
6291 be incorrect or 0 for arguments other than zero, so it should only be
6292 used for debugging purposes.
6294 Note that calling this intrinsic does not prevent function inlining or
6295 other aggressive transformations, so the value returned may not be that
6296 of the obvious source-language caller.
6298 '``llvm.frameaddress``' Intrinsic
6299 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6306 declare i8* @llvm.frameaddress(i32 <level>)
6311 The '``llvm.frameaddress``' intrinsic attempts to return the
6312 target-specific frame pointer value for the specified stack frame.
6317 The argument to this intrinsic indicates which function to return the
6318 frame pointer for. Zero indicates the calling function, one indicates
6319 its caller, etc. The argument is **required** to be a constant integer
6325 The '``llvm.frameaddress``' intrinsic either returns a pointer
6326 indicating the frame address of the specified call frame, or zero if it
6327 cannot be identified. The value returned by this intrinsic is likely to
6328 be incorrect or 0 for arguments other than zero, so it should only be
6329 used for debugging purposes.
6331 Note that calling this intrinsic does not prevent function inlining or
6332 other aggressive transformations, so the value returned may not be that
6333 of the obvious source-language caller.
6337 '``llvm.stacksave``' Intrinsic
6338 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6345 declare i8* @llvm.stacksave()
6350 The '``llvm.stacksave``' intrinsic is used to remember the current state
6351 of the function stack, for use with
6352 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6353 implementing language features like scoped automatic variable sized
6359 This intrinsic returns a opaque pointer value that can be passed to
6360 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6361 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6362 ``llvm.stacksave``, it effectively restores the state of the stack to
6363 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6364 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6365 were allocated after the ``llvm.stacksave`` was executed.
6367 .. _int_stackrestore:
6369 '``llvm.stackrestore``' Intrinsic
6370 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6377 declare void @llvm.stackrestore(i8* %ptr)
6382 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6383 the function stack to the state it was in when the corresponding
6384 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6385 useful for implementing language features like scoped automatic variable
6386 sized arrays in C99.
6391 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6393 '``llvm.prefetch``' Intrinsic
6394 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6401 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6406 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6407 insert a prefetch instruction if supported; otherwise, it is a noop.
6408 Prefetches have no effect on the behavior of the program but can change
6409 its performance characteristics.
6414 ``address`` is the address to be prefetched, ``rw`` is the specifier
6415 determining if the fetch should be for a read (0) or write (1), and
6416 ``locality`` is a temporal locality specifier ranging from (0) - no
6417 locality, to (3) - extremely local keep in cache. The ``cache type``
6418 specifies whether the prefetch is performed on the data (1) or
6419 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6420 arguments must be constant integers.
6425 This intrinsic does not modify the behavior of the program. In
6426 particular, prefetches cannot trap and do not produce a value. On
6427 targets that support this intrinsic, the prefetch can provide hints to
6428 the processor cache for better performance.
6430 '``llvm.pcmarker``' Intrinsic
6431 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6438 declare void @llvm.pcmarker(i32 <id>)
6443 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6444 Counter (PC) in a region of code to simulators and other tools. The
6445 method is target specific, but it is expected that the marker will use
6446 exported symbols to transmit the PC of the marker. The marker makes no
6447 guarantees that it will remain with any specific instruction after
6448 optimizations. It is possible that the presence of a marker will inhibit
6449 optimizations. The intended use is to be inserted after optimizations to
6450 allow correlations of simulation runs.
6455 ``id`` is a numerical id identifying the marker.
6460 This intrinsic does not modify the behavior of the program. Backends
6461 that do not support this intrinsic may ignore it.
6463 '``llvm.readcyclecounter``' Intrinsic
6464 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6471 declare i64 @llvm.readcyclecounter()
6476 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6477 counter register (or similar low latency, high accuracy clocks) on those
6478 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6479 should map to RPCC. As the backing counters overflow quickly (on the
6480 order of 9 seconds on alpha), this should only be used for small
6486 When directly supported, reading the cycle counter should not modify any
6487 memory. Implementations are allowed to either return a application
6488 specific value or a system wide value. On backends without support, this
6489 is lowered to a constant 0.
6491 Standard C Library Intrinsics
6492 -----------------------------
6494 LLVM provides intrinsics for a few important standard C library
6495 functions. These intrinsics allow source-language front-ends to pass
6496 information about the alignment of the pointer arguments to the code
6497 generator, providing opportunity for more efficient code generation.
6501 '``llvm.memcpy``' Intrinsic
6502 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6507 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6508 integer bit width and for different address spaces. Not all targets
6509 support all bit widths however.
6513 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6514 i32 <len>, i32 <align>, i1 <isvolatile>)
6515 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6516 i64 <len>, i32 <align>, i1 <isvolatile>)
6521 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6522 source location to the destination location.
6524 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6525 intrinsics do not return a value, takes extra alignment/isvolatile
6526 arguments and the pointers can be in specified address spaces.
6531 The first argument is a pointer to the destination, the second is a
6532 pointer to the source. The third argument is an integer argument
6533 specifying the number of bytes to copy, the fourth argument is the
6534 alignment of the source and destination locations, and the fifth is a
6535 boolean indicating a volatile access.
6537 If the call to this intrinsic has an alignment value that is not 0 or 1,
6538 then the caller guarantees that both the source and destination pointers
6539 are aligned to that boundary.
6541 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6542 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6543 very cleanly specified and it is unwise to depend on it.
6548 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6549 source location to the destination location, which are not allowed to
6550 overlap. It copies "len" bytes of memory over. If the argument is known
6551 to be aligned to some boundary, this can be specified as the fourth
6552 argument, otherwise it should be set to 0 or 1.
6554 '``llvm.memmove``' Intrinsic
6555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6560 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6561 bit width and for different address space. Not all targets support all
6566 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6567 i32 <len>, i32 <align>, i1 <isvolatile>)
6568 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6569 i64 <len>, i32 <align>, i1 <isvolatile>)
6574 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6575 source location to the destination location. It is similar to the
6576 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6579 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6580 intrinsics do not return a value, takes extra alignment/isvolatile
6581 arguments and the pointers can be in specified address spaces.
6586 The first argument is a pointer to the destination, the second is a
6587 pointer to the source. The third argument is an integer argument
6588 specifying the number of bytes to copy, the fourth argument is the
6589 alignment of the source and destination locations, and the fifth is a
6590 boolean indicating a volatile access.
6592 If the call to this intrinsic has an alignment value that is not 0 or 1,
6593 then the caller guarantees that the source and destination pointers are
6594 aligned to that boundary.
6596 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6597 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6598 not very cleanly specified and it is unwise to depend on it.
6603 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6604 source location to the destination location, which may overlap. It
6605 copies "len" bytes of memory over. If the argument is known to be
6606 aligned to some boundary, this can be specified as the fourth argument,
6607 otherwise it should be set to 0 or 1.
6609 '``llvm.memset.*``' Intrinsics
6610 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6615 This is an overloaded intrinsic. You can use llvm.memset on any integer
6616 bit width and for different address spaces. However, not all targets
6617 support all bit widths.
6621 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6622 i32 <len>, i32 <align>, i1 <isvolatile>)
6623 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6624 i64 <len>, i32 <align>, i1 <isvolatile>)
6629 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6630 particular byte value.
6632 Note that, unlike the standard libc function, the ``llvm.memset``
6633 intrinsic does not return a value and takes extra alignment/volatile
6634 arguments. Also, the destination can be in an arbitrary address space.
6639 The first argument is a pointer to the destination to fill, the second
6640 is the byte value with which to fill it, the third argument is an
6641 integer argument specifying the number of bytes to fill, and the fourth
6642 argument is the known alignment of the destination location.
6644 If the call to this intrinsic has an alignment value that is not 0 or 1,
6645 then the caller guarantees that the destination pointer is aligned to
6648 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6649 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6650 very cleanly specified and it is unwise to depend on it.
6655 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6656 at the destination location. If the argument is known to be aligned to
6657 some boundary, this can be specified as the fourth argument, otherwise
6658 it should be set to 0 or 1.
6660 '``llvm.sqrt.*``' Intrinsic
6661 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6666 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6667 floating point or vector of floating point type. Not all targets support
6672 declare float @llvm.sqrt.f32(float %Val)
6673 declare double @llvm.sqrt.f64(double %Val)
6674 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6675 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6676 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6681 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6682 returning the same value as the libm '``sqrt``' functions would. Unlike
6683 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6684 negative numbers other than -0.0 (which allows for better optimization,
6685 because there is no need to worry about errno being set).
6686 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6691 The argument and return value are floating point numbers of the same
6697 This function returns the sqrt of the specified operand if it is a
6698 nonnegative floating point number.
6700 '``llvm.powi.*``' Intrinsic
6701 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6706 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6707 floating point or vector of floating point type. Not all targets support
6712 declare float @llvm.powi.f32(float %Val, i32 %power)
6713 declare double @llvm.powi.f64(double %Val, i32 %power)
6714 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6715 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6716 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6721 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6722 specified (positive or negative) power. The order of evaluation of
6723 multiplications is not defined. When a vector of floating point type is
6724 used, the second argument remains a scalar integer value.
6729 The second argument is an integer power, and the first is a value to
6730 raise to that power.
6735 This function returns the first value raised to the second power with an
6736 unspecified sequence of rounding operations.
6738 '``llvm.sin.*``' Intrinsic
6739 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6744 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
6745 floating point or vector of floating point type. Not all targets support
6750 declare float @llvm.sin.f32(float %Val)
6751 declare double @llvm.sin.f64(double %Val)
6752 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6753 declare fp128 @llvm.sin.f128(fp128 %Val)
6754 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6759 The '``llvm.sin.*``' intrinsics return the sine of the operand.
6764 The argument and return value are floating point numbers of the same
6770 This function returns the sine of the specified operand, returning the
6771 same values as the libm ``sin`` functions would, and handles error
6772 conditions in the same way.
6774 '``llvm.cos.*``' Intrinsic
6775 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6780 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
6781 floating point or vector of floating point type. Not all targets support
6786 declare float @llvm.cos.f32(float %Val)
6787 declare double @llvm.cos.f64(double %Val)
6788 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6789 declare fp128 @llvm.cos.f128(fp128 %Val)
6790 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6795 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
6800 The argument and return value are floating point numbers of the same
6806 This function returns the cosine of the specified operand, returning the
6807 same values as the libm ``cos`` functions would, and handles error
6808 conditions in the same way.
6810 '``llvm.pow.*``' Intrinsic
6811 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6816 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
6817 floating point or vector of floating point type. Not all targets support
6822 declare float @llvm.pow.f32(float %Val, float %Power)
6823 declare double @llvm.pow.f64(double %Val, double %Power)
6824 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6825 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6826 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6831 The '``llvm.pow.*``' intrinsics return the first operand raised to the
6832 specified (positive or negative) power.
6837 The second argument is a floating point power, and the first is a value
6838 to raise to that power.
6843 This function returns the first value raised to the second power,
6844 returning the same values as the libm ``pow`` functions would, and
6845 handles error conditions in the same way.
6847 '``llvm.exp.*``' Intrinsic
6848 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6853 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
6854 floating point or vector of floating point type. Not all targets support
6859 declare float @llvm.exp.f32(float %Val)
6860 declare double @llvm.exp.f64(double %Val)
6861 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
6862 declare fp128 @llvm.exp.f128(fp128 %Val)
6863 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
6868 The '``llvm.exp.*``' intrinsics perform the exp function.
6873 The argument and return value are floating point numbers of the same
6879 This function returns the same values as the libm ``exp`` functions
6880 would, and handles error conditions in the same way.
6882 '``llvm.exp2.*``' Intrinsic
6883 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6888 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
6889 floating point or vector of floating point type. Not all targets support
6894 declare float @llvm.exp2.f32(float %Val)
6895 declare double @llvm.exp2.f64(double %Val)
6896 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
6897 declare fp128 @llvm.exp2.f128(fp128 %Val)
6898 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
6903 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
6908 The argument and return value are floating point numbers of the same
6914 This function returns the same values as the libm ``exp2`` functions
6915 would, and handles error conditions in the same way.
6917 '``llvm.log.*``' Intrinsic
6918 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6923 This is an overloaded intrinsic. You can use ``llvm.log`` on any
6924 floating point or vector of floating point type. Not all targets support
6929 declare float @llvm.log.f32(float %Val)
6930 declare double @llvm.log.f64(double %Val)
6931 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
6932 declare fp128 @llvm.log.f128(fp128 %Val)
6933 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
6938 The '``llvm.log.*``' intrinsics perform the log function.
6943 The argument and return value are floating point numbers of the same
6949 This function returns the same values as the libm ``log`` functions
6950 would, and handles error conditions in the same way.
6952 '``llvm.log10.*``' Intrinsic
6953 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6958 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
6959 floating point or vector of floating point type. Not all targets support
6964 declare float @llvm.log10.f32(float %Val)
6965 declare double @llvm.log10.f64(double %Val)
6966 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
6967 declare fp128 @llvm.log10.f128(fp128 %Val)
6968 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
6973 The '``llvm.log10.*``' intrinsics perform the log10 function.
6978 The argument and return value are floating point numbers of the same
6984 This function returns the same values as the libm ``log10`` functions
6985 would, and handles error conditions in the same way.
6987 '``llvm.log2.*``' Intrinsic
6988 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6993 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
6994 floating point or vector of floating point type. Not all targets support
6999 declare float @llvm.log2.f32(float %Val)
7000 declare double @llvm.log2.f64(double %Val)
7001 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7002 declare fp128 @llvm.log2.f128(fp128 %Val)
7003 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7008 The '``llvm.log2.*``' intrinsics perform the log2 function.
7013 The argument and return value are floating point numbers of the same
7019 This function returns the same values as the libm ``log2`` functions
7020 would, and handles error conditions in the same way.
7022 '``llvm.fma.*``' Intrinsic
7023 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7028 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7029 floating point or vector of floating point type. Not all targets support
7034 declare float @llvm.fma.f32(float %a, float %b, float %c)
7035 declare double @llvm.fma.f64(double %a, double %b, double %c)
7036 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7037 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7038 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7043 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7049 The argument and return value are floating point numbers of the same
7055 This function returns the same values as the libm ``fma`` functions
7058 '``llvm.fabs.*``' Intrinsic
7059 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7064 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7065 floating point or vector of floating point type. Not all targets support
7070 declare float @llvm.fabs.f32(float %Val)
7071 declare double @llvm.fabs.f64(double %Val)
7072 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7073 declare fp128 @llvm.fabs.f128(fp128 %Val)
7074 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7079 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7085 The argument and return value are floating point numbers of the same
7091 This function returns the same values as the libm ``fabs`` functions
7092 would, and handles error conditions in the same way.
7094 '``llvm.floor.*``' Intrinsic
7095 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7100 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7101 floating point or vector of floating point type. Not all targets support
7106 declare float @llvm.floor.f32(float %Val)
7107 declare double @llvm.floor.f64(double %Val)
7108 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7109 declare fp128 @llvm.floor.f128(fp128 %Val)
7110 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7115 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7120 The argument and return value are floating point numbers of the same
7126 This function returns the same values as the libm ``floor`` functions
7127 would, and handles error conditions in the same way.
7129 '``llvm.ceil.*``' Intrinsic
7130 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7135 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7136 floating point or vector of floating point type. Not all targets support
7141 declare float @llvm.ceil.f32(float %Val)
7142 declare double @llvm.ceil.f64(double %Val)
7143 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7144 declare fp128 @llvm.ceil.f128(fp128 %Val)
7145 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7150 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7155 The argument and return value are floating point numbers of the same
7161 This function returns the same values as the libm ``ceil`` functions
7162 would, and handles error conditions in the same way.
7164 '``llvm.trunc.*``' Intrinsic
7165 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7170 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7171 floating point or vector of floating point type. Not all targets support
7176 declare float @llvm.trunc.f32(float %Val)
7177 declare double @llvm.trunc.f64(double %Val)
7178 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7179 declare fp128 @llvm.trunc.f128(fp128 %Val)
7180 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7185 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7186 nearest integer not larger in magnitude than the operand.
7191 The argument and return value are floating point numbers of the same
7197 This function returns the same values as the libm ``trunc`` functions
7198 would, and handles error conditions in the same way.
7200 '``llvm.rint.*``' Intrinsic
7201 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7206 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7207 floating point or vector of floating point type. Not all targets support
7212 declare float @llvm.rint.f32(float %Val)
7213 declare double @llvm.rint.f64(double %Val)
7214 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7215 declare fp128 @llvm.rint.f128(fp128 %Val)
7216 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7221 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7222 nearest integer. It may raise an inexact floating-point exception if the
7223 operand isn't an integer.
7228 The argument and return value are floating point numbers of the same
7234 This function returns the same values as the libm ``rint`` functions
7235 would, and handles error conditions in the same way.
7237 '``llvm.nearbyint.*``' Intrinsic
7238 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7243 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7244 floating point or vector of floating point type. Not all targets support
7249 declare float @llvm.nearbyint.f32(float %Val)
7250 declare double @llvm.nearbyint.f64(double %Val)
7251 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7252 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7253 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7258 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7264 The argument and return value are floating point numbers of the same
7270 This function returns the same values as the libm ``nearbyint``
7271 functions would, and handles error conditions in the same way.
7273 Bit Manipulation Intrinsics
7274 ---------------------------
7276 LLVM provides intrinsics for a few important bit manipulation
7277 operations. These allow efficient code generation for some algorithms.
7279 '``llvm.bswap.*``' Intrinsics
7280 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7285 This is an overloaded intrinsic function. You can use bswap on any
7286 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7290 declare i16 @llvm.bswap.i16(i16 <id>)
7291 declare i32 @llvm.bswap.i32(i32 <id>)
7292 declare i64 @llvm.bswap.i64(i64 <id>)
7297 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7298 values with an even number of bytes (positive multiple of 16 bits).
7299 These are useful for performing operations on data that is not in the
7300 target's native byte order.
7305 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7306 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7307 intrinsic returns an i32 value that has the four bytes of the input i32
7308 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7309 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7310 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7311 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7314 '``llvm.ctpop.*``' Intrinsic
7315 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7320 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7321 bit width, or on any vector with integer elements. Not all targets
7322 support all bit widths or vector types, however.
7326 declare i8 @llvm.ctpop.i8(i8 <src>)
7327 declare i16 @llvm.ctpop.i16(i16 <src>)
7328 declare i32 @llvm.ctpop.i32(i32 <src>)
7329 declare i64 @llvm.ctpop.i64(i64 <src>)
7330 declare i256 @llvm.ctpop.i256(i256 <src>)
7331 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7336 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7342 The only argument is the value to be counted. The argument may be of any
7343 integer type, or a vector with integer elements. The return type must
7344 match the argument type.
7349 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7350 each element of a vector.
7352 '``llvm.ctlz.*``' Intrinsic
7353 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7358 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7359 integer bit width, or any vector whose elements are integers. Not all
7360 targets support all bit widths or vector types, however.
7364 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7365 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7366 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7367 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7368 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7369 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7374 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7375 leading zeros in a variable.
7380 The first argument is the value to be counted. This argument may be of
7381 any integer type, or a vectory with integer element type. The return
7382 type must match the first argument type.
7384 The second argument must be a constant and is a flag to indicate whether
7385 the intrinsic should ensure that a zero as the first argument produces a
7386 defined result. Historically some architectures did not provide a
7387 defined result for zero values as efficiently, and many algorithms are
7388 now predicated on avoiding zero-value inputs.
7393 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7394 zeros in a variable, or within each element of the vector. If
7395 ``src == 0`` then the result is the size in bits of the type of ``src``
7396 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7397 ``llvm.ctlz(i32 2) = 30``.
7399 '``llvm.cttz.*``' Intrinsic
7400 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7405 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7406 integer bit width, or any vector of integer elements. Not all targets
7407 support all bit widths or vector types, however.
7411 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7412 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7413 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7414 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7415 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7416 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7421 The '``llvm.cttz``' family of intrinsic functions counts the number of
7427 The first argument is the value to be counted. This argument may be of
7428 any integer type, or a vectory with integer element type. The return
7429 type must match the first argument type.
7431 The second argument must be a constant and is a flag to indicate whether
7432 the intrinsic should ensure that a zero as the first argument produces a
7433 defined result. Historically some architectures did not provide a
7434 defined result for zero values as efficiently, and many algorithms are
7435 now predicated on avoiding zero-value inputs.
7440 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7441 zeros in a variable, or within each element of a vector. If ``src == 0``
7442 then the result is the size in bits of the type of ``src`` if
7443 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7444 ``llvm.cttz(2) = 1``.
7446 Arithmetic with Overflow Intrinsics
7447 -----------------------------------
7449 LLVM provides intrinsics for some arithmetic with overflow operations.
7451 '``llvm.sadd.with.overflow.*``' Intrinsics
7452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7457 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7458 on any integer bit width.
7462 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7463 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7464 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7469 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7470 a signed addition of the two arguments, and indicate whether an overflow
7471 occurred during the signed summation.
7476 The arguments (%a and %b) and the first element of the result structure
7477 may be of integer types of any bit width, but they must have the same
7478 bit width. The second element of the result structure must be of type
7479 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7485 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7486 a signed addition of the two variables. They return a structure --- the
7487 first element of which is the signed summation, and the second element
7488 of which is a bit specifying if the signed summation resulted in an
7494 .. code-block:: llvm
7496 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7497 %sum = extractvalue {i32, i1} %res, 0
7498 %obit = extractvalue {i32, i1} %res, 1
7499 br i1 %obit, label %overflow, label %normal
7501 '``llvm.uadd.with.overflow.*``' Intrinsics
7502 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7507 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7508 on any integer bit width.
7512 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7513 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7514 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7519 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7520 an unsigned addition of the two arguments, and indicate whether a carry
7521 occurred during the unsigned summation.
7526 The arguments (%a and %b) and the first element of the result structure
7527 may be of integer types of any bit width, but they must have the same
7528 bit width. The second element of the result structure must be of type
7529 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7535 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7536 an unsigned addition of the two arguments. They return a structure --- the
7537 first element of which is the sum, and the second element of which is a
7538 bit specifying if the unsigned summation resulted in a carry.
7543 .. code-block:: llvm
7545 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7546 %sum = extractvalue {i32, i1} %res, 0
7547 %obit = extractvalue {i32, i1} %res, 1
7548 br i1 %obit, label %carry, label %normal
7550 '``llvm.ssub.with.overflow.*``' Intrinsics
7551 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7556 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7557 on any integer bit width.
7561 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7562 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7563 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7568 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7569 a signed subtraction of the two arguments, and indicate whether an
7570 overflow occurred during the signed subtraction.
7575 The arguments (%a and %b) and the first element of the result structure
7576 may be of integer types of any bit width, but they must have the same
7577 bit width. The second element of the result structure must be of type
7578 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7584 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7585 a signed subtraction of the two arguments. They return a structure --- the
7586 first element of which is the subtraction, and the second element of
7587 which is a bit specifying if the signed subtraction resulted in an
7593 .. code-block:: llvm
7595 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7596 %sum = extractvalue {i32, i1} %res, 0
7597 %obit = extractvalue {i32, i1} %res, 1
7598 br i1 %obit, label %overflow, label %normal
7600 '``llvm.usub.with.overflow.*``' Intrinsics
7601 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7606 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7607 on any integer bit width.
7611 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7612 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7613 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7618 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7619 an unsigned subtraction of the two arguments, and indicate whether an
7620 overflow occurred during the unsigned subtraction.
7625 The arguments (%a and %b) and the first element of the result structure
7626 may be of integer types of any bit width, but they must have the same
7627 bit width. The second element of the result structure must be of type
7628 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7634 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7635 an unsigned subtraction of the two arguments. They return a structure ---
7636 the first element of which is the subtraction, and the second element of
7637 which is a bit specifying if the unsigned subtraction resulted in an
7643 .. code-block:: llvm
7645 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7646 %sum = extractvalue {i32, i1} %res, 0
7647 %obit = extractvalue {i32, i1} %res, 1
7648 br i1 %obit, label %overflow, label %normal
7650 '``llvm.smul.with.overflow.*``' Intrinsics
7651 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7656 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7657 on any integer bit width.
7661 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7662 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7663 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7668 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7669 a signed multiplication of the two arguments, and indicate whether an
7670 overflow occurred during the signed multiplication.
7675 The arguments (%a and %b) and the first element of the result structure
7676 may be of integer types of any bit width, but they must have the same
7677 bit width. The second element of the result structure must be of type
7678 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7684 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7685 a signed multiplication of the two arguments. They return a structure ---
7686 the first element of which is the multiplication, and the second element
7687 of which is a bit specifying if the signed multiplication resulted in an
7693 .. code-block:: llvm
7695 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7696 %sum = extractvalue {i32, i1} %res, 0
7697 %obit = extractvalue {i32, i1} %res, 1
7698 br i1 %obit, label %overflow, label %normal
7700 '``llvm.umul.with.overflow.*``' Intrinsics
7701 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7706 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
7707 on any integer bit width.
7711 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7712 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7713 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7718 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7719 a unsigned multiplication of the two arguments, and indicate whether an
7720 overflow occurred during the unsigned multiplication.
7725 The arguments (%a and %b) and the first element of the result structure
7726 may be of integer types of any bit width, but they must have the same
7727 bit width. The second element of the result structure must be of type
7728 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7734 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7735 an unsigned multiplication of the two arguments. They return a structure ---
7736 the first element of which is the multiplication, and the second
7737 element of which is a bit specifying if the unsigned multiplication
7738 resulted in an overflow.
7743 .. code-block:: llvm
7745 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7746 %sum = extractvalue {i32, i1} %res, 0
7747 %obit = extractvalue {i32, i1} %res, 1
7748 br i1 %obit, label %overflow, label %normal
7750 Specialised Arithmetic Intrinsics
7751 ---------------------------------
7753 '``llvm.fmuladd.*``' Intrinsic
7754 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7761 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
7762 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
7767 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
7768 expressions that can be fused if the code generator determines that (a) the
7769 target instruction set has support for a fused operation, and (b) that the
7770 fused operation is more efficient than the equivalent, separate pair of mul
7771 and add instructions.
7776 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
7777 multiplicands, a and b, and an addend c.
7786 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
7788 is equivalent to the expression a \* b + c, except that rounding will
7789 not be performed between the multiplication and addition steps if the
7790 code generator fuses the operations. Fusion is not guaranteed, even if
7791 the target platform supports it. If a fused multiply-add is required the
7792 corresponding llvm.fma.\* intrinsic function should be used instead.
7797 .. code-block:: llvm
7799 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
7801 Half Precision Floating Point Intrinsics
7802 ----------------------------------------
7804 For most target platforms, half precision floating point is a
7805 storage-only format. This means that it is a dense encoding (in memory)
7806 but does not support computation in the format.
7808 This means that code must first load the half-precision floating point
7809 value as an i16, then convert it to float with
7810 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
7811 then be performed on the float value (including extending to double
7812 etc). To store the value back to memory, it is first converted to float
7813 if needed, then converted to i16 with
7814 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
7817 .. _int_convert_to_fp16:
7819 '``llvm.convert.to.fp16``' Intrinsic
7820 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7827 declare i16 @llvm.convert.to.fp16(f32 %a)
7832 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7833 from single precision floating point format to half precision floating
7839 The intrinsic function contains single argument - the value to be
7845 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7846 from single precision floating point format to half precision floating
7847 point format. The return value is an ``i16`` which contains the
7853 .. code-block:: llvm
7855 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7856 store i16 %res, i16* @x, align 2
7858 .. _int_convert_from_fp16:
7860 '``llvm.convert.from.fp16``' Intrinsic
7861 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7868 declare f32 @llvm.convert.from.fp16(i16 %a)
7873 The '``llvm.convert.from.fp16``' intrinsic function performs a
7874 conversion from half precision floating point format to single precision
7875 floating point format.
7880 The intrinsic function contains single argument - the value to be
7886 The '``llvm.convert.from.fp16``' intrinsic function performs a
7887 conversion from half single precision floating point format to single
7888 precision floating point format. The input half-float value is
7889 represented by an ``i16`` value.
7894 .. code-block:: llvm
7896 %a = load i16* @x, align 2
7897 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7902 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
7903 prefix), are described in the `LLVM Source Level
7904 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
7907 Exception Handling Intrinsics
7908 -----------------------------
7910 The LLVM exception handling intrinsics (which all start with
7911 ``llvm.eh.`` prefix), are described in the `LLVM Exception
7912 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
7916 Trampoline Intrinsics
7917 ---------------------
7919 These intrinsics make it possible to excise one parameter, marked with
7920 the :ref:`nest <nest>` attribute, from a function. The result is a
7921 callable function pointer lacking the nest parameter - the caller does
7922 not need to provide a value for it. Instead, the value to use is stored
7923 in advance in a "trampoline", a block of memory usually allocated on the
7924 stack, which also contains code to splice the nest value into the
7925 argument list. This is used to implement the GCC nested function address
7928 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
7929 then the resulting function pointer has signature ``i32 (i32, i32)*``.
7930 It can be created as follows:
7932 .. code-block:: llvm
7934 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7935 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
7936 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
7937 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
7938 %fp = bitcast i8* %p to i32 (i32, i32)*
7940 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
7941 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
7945 '``llvm.init.trampoline``' Intrinsic
7946 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7953 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
7958 This fills the memory pointed to by ``tramp`` with executable code,
7959 turning it into a trampoline.
7964 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
7965 pointers. The ``tramp`` argument must point to a sufficiently large and
7966 sufficiently aligned block of memory; this memory is written to by the
7967 intrinsic. Note that the size and the alignment are target-specific -
7968 LLVM currently provides no portable way of determining them, so a
7969 front-end that generates this intrinsic needs to have some
7970 target-specific knowledge. The ``func`` argument must hold a function
7971 bitcast to an ``i8*``.
7976 The block of memory pointed to by ``tramp`` is filled with target
7977 dependent code, turning it into a function. Then ``tramp`` needs to be
7978 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
7979 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
7980 function's signature is the same as that of ``func`` with any arguments
7981 marked with the ``nest`` attribute removed. At most one such ``nest``
7982 argument is allowed, and it must be of pointer type. Calling the new
7983 function is equivalent to calling ``func`` with the same argument list,
7984 but with ``nval`` used for the missing ``nest`` argument. If, after
7985 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
7986 modified, then the effect of any later call to the returned function
7987 pointer is undefined.
7991 '``llvm.adjust.trampoline``' Intrinsic
7992 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7999 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8004 This performs any required machine-specific adjustment to the address of
8005 a trampoline (passed as ``tramp``).
8010 ``tramp`` must point to a block of memory which already has trampoline
8011 code filled in by a previous call to
8012 :ref:`llvm.init.trampoline <int_it>`.
8017 On some architectures the address of the code to be executed needs to be
8018 different to the address where the trampoline is actually stored. This
8019 intrinsic returns the executable address corresponding to ``tramp``
8020 after performing the required machine specific adjustments. The pointer
8021 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8026 This class of intrinsics exists to information about the lifetime of
8027 memory objects and ranges where variables are immutable.
8029 '``llvm.lifetime.start``' Intrinsic
8030 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8037 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8042 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8048 The first argument is a constant integer representing the size of the
8049 object, or -1 if it is variable sized. The second argument is a pointer
8055 This intrinsic indicates that before this point in the code, the value
8056 of the memory pointed to by ``ptr`` is dead. This means that it is known
8057 to never be used and has an undefined value. A load from the pointer
8058 that precedes this intrinsic can be replaced with ``'undef'``.
8060 '``llvm.lifetime.end``' Intrinsic
8061 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8068 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8073 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8079 The first argument is a constant integer representing the size of the
8080 object, or -1 if it is variable sized. The second argument is a pointer
8086 This intrinsic indicates that after this point in the code, the value of
8087 the memory pointed to by ``ptr`` is dead. This means that it is known to
8088 never be used and has an undefined value. Any stores into the memory
8089 object following this intrinsic may be removed as dead.
8091 '``llvm.invariant.start``' Intrinsic
8092 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8099 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8104 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8105 a memory object will not change.
8110 The first argument is a constant integer representing the size of the
8111 object, or -1 if it is variable sized. The second argument is a pointer
8117 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8118 the return value, the referenced memory location is constant and
8121 '``llvm.invariant.end``' Intrinsic
8122 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8129 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8134 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8135 memory object are mutable.
8140 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8141 The second argument is a constant integer representing the size of the
8142 object, or -1 if it is variable sized and the third argument is a
8143 pointer to the object.
8148 This intrinsic indicates that the memory is mutable again.
8153 This class of intrinsics is designed to be generic and has no specific
8156 '``llvm.var.annotation``' Intrinsic
8157 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8164 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8169 The '``llvm.var.annotation``' intrinsic.
8174 The first argument is a pointer to a value, the second is a pointer to a
8175 global string, the third is a pointer to a global string which is the
8176 source file name, and the last argument is the line number.
8181 This intrinsic allows annotation of local variables with arbitrary
8182 strings. This can be useful for special purpose optimizations that want
8183 to look for these annotations. These have no other defined use; they are
8184 ignored by code generation and optimization.
8186 '``llvm.annotation.*``' Intrinsic
8187 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8192 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8193 any integer bit width.
8197 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8198 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8199 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8200 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8201 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8206 The '``llvm.annotation``' intrinsic.
8211 The first argument is an integer value (result of some expression), the
8212 second is a pointer to a global string, the third is a pointer to a
8213 global string which is the source file name, and the last argument is
8214 the line number. It returns the value of the first argument.
8219 This intrinsic allows annotations to be put on arbitrary expressions
8220 with arbitrary strings. This can be useful for special purpose
8221 optimizations that want to look for these annotations. These have no
8222 other defined use; they are ignored by code generation and optimization.
8224 '``llvm.trap``' Intrinsic
8225 ^^^^^^^^^^^^^^^^^^^^^^^^^
8232 declare void @llvm.trap() noreturn nounwind
8237 The '``llvm.trap``' intrinsic.
8247 This intrinsic is lowered to the target dependent trap instruction. If
8248 the target does not have a trap instruction, this intrinsic will be
8249 lowered to a call of the ``abort()`` function.
8251 '``llvm.debugtrap``' Intrinsic
8252 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8259 declare void @llvm.debugtrap() nounwind
8264 The '``llvm.debugtrap``' intrinsic.
8274 This intrinsic is lowered to code which is intended to cause an
8275 execution trap with the intention of requesting the attention of a
8278 '``llvm.stackprotector``' Intrinsic
8279 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8286 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8291 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8292 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8293 is placed on the stack before local variables.
8298 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8299 The first argument is the value loaded from the stack guard
8300 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8301 enough space to hold the value of the guard.
8306 This intrinsic causes the prologue/epilogue inserter to force the
8307 position of the ``AllocaInst`` stack slot to be before local variables
8308 on the stack. This is to ensure that if a local variable on the stack is
8309 overwritten, it will destroy the value of the guard. When the function
8310 exits, the guard on the stack is checked against the original guard. If
8311 they are different, then the program aborts by calling the
8312 ``__stack_chk_fail()`` function.
8314 '``llvm.objectsize``' Intrinsic
8315 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8322 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8323 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8328 The ``llvm.objectsize`` intrinsic is designed to provide information to
8329 the optimizers to determine at compile time whether a) an operation
8330 (like memcpy) will overflow a buffer that corresponds to an object, or
8331 b) that a runtime check for overflow isn't necessary. An object in this
8332 context means an allocation of a specific class, structure, array, or
8338 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8339 argument is a pointer to or into the ``object``. The second argument is
8340 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8341 or -1 (if false) when the object size is unknown. The second argument
8342 only accepts constants.
8347 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8348 the size of the object concerned. If the size cannot be determined at
8349 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8350 on the ``min`` argument).
8352 '``llvm.expect``' Intrinsic
8353 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8360 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8361 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8366 The ``llvm.expect`` intrinsic provides information about expected (the
8367 most probable) value of ``val``, which can be used by optimizers.
8372 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8373 a value. The second argument is an expected value, this needs to be a
8374 constant value, variables are not allowed.
8379 This intrinsic is lowered to the ``val``.
8381 '``llvm.donothing``' Intrinsic
8382 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8389 declare void @llvm.donothing() nounwind readnone
8394 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8395 only intrinsic that can be called with an invoke instruction.
8405 This intrinsic does nothing, and it's removed by optimizers and ignored