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 Attribute groups are groups of attributes that are referenced by objects within
746 the IR. They are important for keeping ``.ll`` files readable, because a lot of
747 functions will use the same set of attributes. In the degenerative case of a
748 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
749 group will capture the important command line flags used to build that file.
751 An attribute group is a module-level object. To use an attribute group, an
752 object references the attribute group's ID (e.g. ``#37``). An object may refer
753 to more than one attribute group. In that situation, the attributes from the
754 different groups are merged.
756 Here is an example of attribute groups for a function that should always be
757 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
761 ; Target-independent attributes:
762 #0 = attributes { alwaysinline alignstack=4 }
764 ; Target-dependent attributes:
765 #1 = attributes { "no-sse" }
767 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
768 define void @f() #0 #1 { ... }
775 Function attributes are set to communicate additional information about
776 a function. Function attributes are considered to be part of the
777 function, not of the function type, so functions with different function
778 attributes can have the same function type.
780 Function attributes are simple keywords that follow the type specified.
781 If multiple attributes are needed, they are space separated. For
786 define void @f() noinline { ... }
787 define void @f() alwaysinline { ... }
788 define void @f() alwaysinline optsize { ... }
789 define void @f() optsize { ... }
792 This attribute indicates that the address safety analysis is enabled
795 This attribute indicates that, when emitting the prologue and
796 epilogue, the backend should forcibly align the stack pointer.
797 Specify the desired alignment, which must be a power of two, in
800 This attribute indicates that the inliner should attempt to inline
801 this function into callers whenever possible, ignoring any active
802 inlining size threshold for this caller.
804 This attribute suppresses lazy symbol binding for the function. This
805 may make calls to the function faster, at the cost of extra program
806 startup time if the function is not called during program startup.
808 This attribute indicates that the source code contained a hint that
809 inlining this function is desirable (such as the "inline" keyword in
810 C/C++). It is just a hint; it imposes no requirements on the
813 This attribute disables prologue / epilogue emission for the
814 function. This can have very system-specific consequences.
816 This attribute indicates that calls to the function cannot be
817 duplicated. A call to a ``noduplicate`` function may be moved
818 within its parent function, but may not be duplicated within
821 A function containing a ``noduplicate`` call may still
822 be an inlining candidate, provided that the call is not
823 duplicated by inlining. That implies that the function has
824 internal linkage and only has one call site, so the original
825 call is dead after inlining.
827 This attributes disables implicit floating point instructions.
829 This attribute indicates that the inliner should never inline this
830 function in any situation. This attribute may not be used together
831 with the ``alwaysinline`` attribute.
833 This attribute indicates that the code generator should not use a
834 red zone, even if the target-specific ABI normally permits it.
836 This function attribute indicates that the function never returns
837 normally. This produces undefined behavior at runtime if the
838 function ever does dynamically return.
840 This function attribute indicates that the function never returns
841 with an unwind or exceptional control flow. If the function does
842 unwind, its runtime behavior is undefined.
844 This attribute suggests that optimization passes and code generator
845 passes make choices that keep the code size of this function low,
846 and otherwise do optimizations specifically to reduce code size.
848 This attribute indicates that the function computes its result (or
849 decides to unwind an exception) based strictly on its arguments,
850 without dereferencing any pointer arguments or otherwise accessing
851 any mutable state (e.g. memory, control registers, etc) visible to
852 caller functions. It does not write through any pointer arguments
853 (including ``byval`` arguments) and never changes any state visible
854 to callers. This means that it cannot unwind exceptions by calling
855 the ``C++`` exception throwing methods.
857 This attribute indicates that the function does not write through
858 any pointer arguments (including ``byval`` arguments) or otherwise
859 modify any state (e.g. memory, control registers, etc) visible to
860 caller functions. It may dereference pointer arguments and read
861 state that may be set in the caller. A readonly function always
862 returns the same value (or unwinds an exception identically) when
863 called with the same set of arguments and global state. It cannot
864 unwind an exception by calling the ``C++`` exception throwing
867 This attribute indicates that this function can return twice. The C
868 ``setjmp`` is an example of such a function. The compiler disables
869 some optimizations (like tail calls) in the caller of these
872 This attribute indicates that the function should emit a stack
873 smashing protector. It is in the form of a "canary" --- a random value
874 placed on the stack before the local variables that's checked upon
875 return from the function to see if it has been overwritten. A
876 heuristic is used to determine if a function needs stack protectors
877 or not. The heuristic used will enable protectors for functions with:
879 - Character arrays larger than ``ssp-buffer-size`` (default 8).
880 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
881 - Calls to alloca() with variable sizes or constant sizes greater than
884 If a function that has an ``ssp`` attribute is inlined into a
885 function that doesn't have an ``ssp`` attribute, then the resulting
886 function will have an ``ssp`` attribute.
888 This attribute indicates that the function should *always* emit a
889 stack smashing protector. This overrides the ``ssp`` function
892 If a function that has an ``sspreq`` attribute is inlined into a
893 function that doesn't have an ``sspreq`` attribute or which has an
894 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
895 an ``sspreq`` attribute.
897 This attribute indicates that the function should emit a stack smashing
898 protector. This attribute causes a strong heuristic to be used when
899 determining if a function needs stack protectors. The strong heuristic
900 will enable protectors for functions with:
902 - Arrays of any size and type
903 - Aggregates containing an array of any size and type.
905 - Local variables that have had their address taken.
907 This overrides the ``ssp`` function attribute.
909 If a function that has an ``sspstrong`` attribute is inlined into a
910 function that doesn't have an ``sspstrong`` attribute, then the
911 resulting function will have an ``sspstrong`` attribute.
913 This attribute indicates that the thread safety analysis is enabled
915 ``uninitialized_checks``
916 This attribute indicates that the checks for uses of uninitialized
919 This attribute indicates that the ABI being targeted requires that
920 an unwind table entry be produce for this function even if we can
921 show that no exceptions passes by it. This is normally the case for
922 the ELF x86-64 abi, but it can be disabled for some compilation
927 Module-Level Inline Assembly
928 ----------------------------
930 Modules may contain "module-level inline asm" blocks, which corresponds
931 to the GCC "file scope inline asm" blocks. These blocks are internally
932 concatenated by LLVM and treated as a single unit, but may be separated
933 in the ``.ll`` file if desired. The syntax is very simple:
937 module asm "inline asm code goes here"
938 module asm "more can go here"
940 The strings can contain any character by escaping non-printable
941 characters. The escape sequence used is simply "\\xx" where "xx" is the
942 two digit hex code for the number.
944 The inline asm code is simply printed to the machine code .s file when
945 assembly code is generated.
950 A module may specify a target specific data layout string that specifies
951 how data is to be laid out in memory. The syntax for the data layout is
956 target datalayout = "layout specification"
958 The *layout specification* consists of a list of specifications
959 separated by the minus sign character ('-'). Each specification starts
960 with a letter and may include other information after the letter to
961 define some aspect of the data layout. The specifications accepted are
965 Specifies that the target lays out data in big-endian form. That is,
966 the bits with the most significance have the lowest address
969 Specifies that the target lays out data in little-endian form. That
970 is, the bits with the least significance have the lowest address
973 Specifies the natural alignment of the stack in bits. Alignment
974 promotion of stack variables is limited to the natural stack
975 alignment to avoid dynamic stack realignment. The stack alignment
976 must be a multiple of 8-bits. If omitted, the natural stack
977 alignment defaults to "unspecified", which does not prevent any
978 alignment promotions.
979 ``p[n]:<size>:<abi>:<pref>``
980 This specifies the *size* of a pointer and its ``<abi>`` and
981 ``<pref>``\erred alignments for address space ``n``. All sizes are in
982 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
983 preceding ``:`` should be omitted too. The address space, ``n`` is
984 optional, and if not specified, denotes the default address space 0.
985 The value of ``n`` must be in the range [1,2^23).
986 ``i<size>:<abi>:<pref>``
987 This specifies the alignment for an integer type of a given bit
988 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
989 ``v<size>:<abi>:<pref>``
990 This specifies the alignment for a vector type of a given bit
992 ``f<size>:<abi>:<pref>``
993 This specifies the alignment for a floating point type of a given bit
994 ``<size>``. Only values of ``<size>`` that are supported by the target
995 will work. 32 (float) and 64 (double) are supported on all targets; 80
996 or 128 (different flavors of long double) are also supported on some
998 ``a<size>:<abi>:<pref>``
999 This specifies the alignment for an aggregate type of a given bit
1001 ``s<size>:<abi>:<pref>``
1002 This specifies the alignment for a stack object of a given bit
1004 ``n<size1>:<size2>:<size3>...``
1005 This specifies a set of native integer widths for the target CPU in
1006 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1007 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1008 this set are considered to support most general arithmetic operations
1011 When constructing the data layout for a given target, LLVM starts with a
1012 default set of specifications which are then (possibly) overridden by
1013 the specifications in the ``datalayout`` keyword. The default
1014 specifications are given in this list:
1016 - ``E`` - big endian
1017 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment
1018 - ``S0`` - natural stack alignment is unspecified
1019 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1020 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1021 - ``i16:16:16`` - i16 is 16-bit aligned
1022 - ``i32:32:32`` - i32 is 32-bit aligned
1023 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1024 alignment of 64-bits
1025 - ``f16:16:16`` - half is 16-bit aligned
1026 - ``f32:32:32`` - float is 32-bit aligned
1027 - ``f64:64:64`` - double is 64-bit aligned
1028 - ``f128:128:128`` - quad is 128-bit aligned
1029 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1030 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1031 - ``a0:0:64`` - aggregates are 64-bit aligned
1033 When LLVM is determining the alignment for a given type, it uses the
1036 #. If the type sought is an exact match for one of the specifications,
1037 that specification is used.
1038 #. If no match is found, and the type sought is an integer type, then
1039 the smallest integer type that is larger than the bitwidth of the
1040 sought type is used. If none of the specifications are larger than
1041 the bitwidth then the largest integer type is used. For example,
1042 given the default specifications above, the i7 type will use the
1043 alignment of i8 (next largest) while both i65 and i256 will use the
1044 alignment of i64 (largest specified).
1045 #. If no match is found, and the type sought is a vector type, then the
1046 largest vector type that is smaller than the sought vector type will
1047 be used as a fall back. This happens because <128 x double> can be
1048 implemented in terms of 64 <2 x double>, for example.
1050 The function of the data layout string may not be what you expect.
1051 Notably, this is not a specification from the frontend of what alignment
1052 the code generator should use.
1054 Instead, if specified, the target data layout is required to match what
1055 the ultimate *code generator* expects. This string is used by the
1056 mid-level optimizers to improve code, and this only works if it matches
1057 what the ultimate code generator uses. If you would like to generate IR
1058 that does not embed this target-specific detail into the IR, then you
1059 don't have to specify the string. This will disable some optimizations
1060 that require precise layout information, but this also prevents those
1061 optimizations from introducing target specificity into the IR.
1063 .. _pointeraliasing:
1065 Pointer Aliasing Rules
1066 ----------------------
1068 Any memory access must be done through a pointer value associated with
1069 an address range of the memory access, otherwise the behavior is
1070 undefined. Pointer values are associated with address ranges according
1071 to the following rules:
1073 - A pointer value is associated with the addresses associated with any
1074 value it is *based* on.
1075 - An address of a global variable is associated with the address range
1076 of the variable's storage.
1077 - The result value of an allocation instruction is associated with the
1078 address range of the allocated storage.
1079 - A null pointer in the default address-space is associated with no
1081 - An integer constant other than zero or a pointer value returned from
1082 a function not defined within LLVM may be associated with address
1083 ranges allocated through mechanisms other than those provided by
1084 LLVM. Such ranges shall not overlap with any ranges of addresses
1085 allocated by mechanisms provided by LLVM.
1087 A pointer value is *based* on another pointer value according to the
1090 - A pointer value formed from a ``getelementptr`` operation is *based*
1091 on the first operand of the ``getelementptr``.
1092 - The result value of a ``bitcast`` is *based* on the operand of the
1094 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1095 values that contribute (directly or indirectly) to the computation of
1096 the pointer's value.
1097 - The "*based* on" relationship is transitive.
1099 Note that this definition of *"based"* is intentionally similar to the
1100 definition of *"based"* in C99, though it is slightly weaker.
1102 LLVM IR does not associate types with memory. The result type of a
1103 ``load`` merely indicates the size and alignment of the memory from
1104 which to load, as well as the interpretation of the value. The first
1105 operand type of a ``store`` similarly only indicates the size and
1106 alignment of the store.
1108 Consequently, type-based alias analysis, aka TBAA, aka
1109 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1110 :ref:`Metadata <metadata>` may be used to encode additional information
1111 which specialized optimization passes may use to implement type-based
1116 Volatile Memory Accesses
1117 ------------------------
1119 Certain memory accesses, such as :ref:`load <i_load>`'s,
1120 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1121 marked ``volatile``. The optimizers must not change the number of
1122 volatile operations or change their order of execution relative to other
1123 volatile operations. The optimizers *may* change the order of volatile
1124 operations relative to non-volatile operations. This is not Java's
1125 "volatile" and has no cross-thread synchronization behavior.
1127 IR-level volatile loads and stores cannot safely be optimized into
1128 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1129 flagged volatile. Likewise, the backend should never split or merge
1130 target-legal volatile load/store instructions.
1132 .. admonition:: Rationale
1134 Platforms may rely on volatile loads and stores of natively supported
1135 data width to be executed as single instruction. For example, in C
1136 this holds for an l-value of volatile primitive type with native
1137 hardware support, but not necessarily for aggregate types. The
1138 frontend upholds these expectations, which are intentionally
1139 unspecified in the IR. The rules above ensure that IR transformation
1140 do not violate the frontend's contract with the language.
1144 Memory Model for Concurrent Operations
1145 --------------------------------------
1147 The LLVM IR does not define any way to start parallel threads of
1148 execution or to register signal handlers. Nonetheless, there are
1149 platform-specific ways to create them, and we define LLVM IR's behavior
1150 in their presence. This model is inspired by the C++0x memory model.
1152 For a more informal introduction to this model, see the :doc:`Atomics`.
1154 We define a *happens-before* partial order as the least partial order
1157 - Is a superset of single-thread program order, and
1158 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1159 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1160 techniques, like pthread locks, thread creation, thread joining,
1161 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1162 Constraints <ordering>`).
1164 Note that program order does not introduce *happens-before* edges
1165 between a thread and signals executing inside that thread.
1167 Every (defined) read operation (load instructions, memcpy, atomic
1168 loads/read-modify-writes, etc.) R reads a series of bytes written by
1169 (defined) write operations (store instructions, atomic
1170 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1171 section, initialized globals are considered to have a write of the
1172 initializer which is atomic and happens before any other read or write
1173 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1174 may see any write to the same byte, except:
1176 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1177 write\ :sub:`2` happens before R\ :sub:`byte`, then
1178 R\ :sub:`byte` does not see write\ :sub:`1`.
1179 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1180 R\ :sub:`byte` does not see write\ :sub:`3`.
1182 Given that definition, R\ :sub:`byte` is defined as follows:
1184 - If R is volatile, the result is target-dependent. (Volatile is
1185 supposed to give guarantees which can support ``sig_atomic_t`` in
1186 C/C++, and may be used for accesses to addresses which do not behave
1187 like normal memory. It does not generally provide cross-thread
1189 - Otherwise, if there is no write to the same byte that happens before
1190 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1191 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1192 R\ :sub:`byte` returns the value written by that write.
1193 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1194 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1195 Memory Ordering Constraints <ordering>` section for additional
1196 constraints on how the choice is made.
1197 - Otherwise R\ :sub:`byte` returns ``undef``.
1199 R returns the value composed of the series of bytes it read. This
1200 implies that some bytes within the value may be ``undef`` **without**
1201 the entire value being ``undef``. Note that this only defines the
1202 semantics of the operation; it doesn't mean that targets will emit more
1203 than one instruction to read the series of bytes.
1205 Note that in cases where none of the atomic intrinsics are used, this
1206 model places only one restriction on IR transformations on top of what
1207 is required for single-threaded execution: introducing a store to a byte
1208 which might not otherwise be stored is not allowed in general.
1209 (Specifically, in the case where another thread might write to and read
1210 from an address, introducing a store can change a load that may see
1211 exactly one write into a load that may see multiple writes.)
1215 Atomic Memory Ordering Constraints
1216 ----------------------------------
1218 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1219 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1220 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1221 an ordering parameter that determines which other atomic instructions on
1222 the same address they *synchronize with*. These semantics are borrowed
1223 from Java and C++0x, but are somewhat more colloquial. If these
1224 descriptions aren't precise enough, check those specs (see spec
1225 references in the :doc:`atomics guide <Atomics>`).
1226 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1227 differently since they don't take an address. See that instruction's
1228 documentation for details.
1230 For a simpler introduction to the ordering constraints, see the
1234 The set of values that can be read is governed by the happens-before
1235 partial order. A value cannot be read unless some operation wrote
1236 it. This is intended to provide a guarantee strong enough to model
1237 Java's non-volatile shared variables. This ordering cannot be
1238 specified for read-modify-write operations; it is not strong enough
1239 to make them atomic in any interesting way.
1241 In addition to the guarantees of ``unordered``, there is a single
1242 total order for modifications by ``monotonic`` operations on each
1243 address. All modification orders must be compatible with the
1244 happens-before order. There is no guarantee that the modification
1245 orders can be combined to a global total order for the whole program
1246 (and this often will not be possible). The read in an atomic
1247 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1248 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1249 order immediately before the value it writes. If one atomic read
1250 happens before another atomic read of the same address, the later
1251 read must see the same value or a later value in the address's
1252 modification order. This disallows reordering of ``monotonic`` (or
1253 stronger) operations on the same address. If an address is written
1254 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1255 read that address repeatedly, the other threads must eventually see
1256 the write. This corresponds to the C++0x/C1x
1257 ``memory_order_relaxed``.
1259 In addition to the guarantees of ``monotonic``, a
1260 *synchronizes-with* edge may be formed with a ``release`` operation.
1261 This is intended to model C++'s ``memory_order_acquire``.
1263 In addition to the guarantees of ``monotonic``, if this operation
1264 writes a value which is subsequently read by an ``acquire``
1265 operation, it *synchronizes-with* that operation. (This isn't a
1266 complete description; see the C++0x definition of a release
1267 sequence.) This corresponds to the C++0x/C1x
1268 ``memory_order_release``.
1269 ``acq_rel`` (acquire+release)
1270 Acts as both an ``acquire`` and ``release`` operation on its
1271 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1272 ``seq_cst`` (sequentially consistent)
1273 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1274 operation which only reads, ``release`` for an operation which only
1275 writes), there is a global total order on all
1276 sequentially-consistent operations on all addresses, which is
1277 consistent with the *happens-before* partial order and with the
1278 modification orders of all the affected addresses. Each
1279 sequentially-consistent read sees the last preceding write to the
1280 same address in this global order. This corresponds to the C++0x/C1x
1281 ``memory_order_seq_cst`` and Java volatile.
1285 If an atomic operation is marked ``singlethread``, it only *synchronizes
1286 with* or participates in modification and seq\_cst total orderings with
1287 other operations running in the same thread (for example, in signal
1295 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1296 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1297 :ref:`frem <i_frem>`) have the following flags that can set to enable
1298 otherwise unsafe floating point operations
1301 No NaNs - Allow optimizations to assume the arguments and result are not
1302 NaN. Such optimizations are required to retain defined behavior over
1303 NaNs, but the value of the result is undefined.
1306 No Infs - Allow optimizations to assume the arguments and result are not
1307 +/-Inf. Such optimizations are required to retain defined behavior over
1308 +/-Inf, but the value of the result is undefined.
1311 No Signed Zeros - Allow optimizations to treat the sign of a zero
1312 argument or result as insignificant.
1315 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1316 argument rather than perform division.
1319 Fast - Allow algebraically equivalent transformations that may
1320 dramatically change results in floating point (e.g. reassociate). This
1321 flag implies all the others.
1328 The LLVM type system is one of the most important features of the
1329 intermediate representation. Being typed enables a number of
1330 optimizations to be performed on the intermediate representation
1331 directly, without having to do extra analyses on the side before the
1332 transformation. A strong type system makes it easier to read the
1333 generated code and enables novel analyses and transformations that are
1334 not feasible to perform on normal three address code representations.
1336 Type Classifications
1337 --------------------
1339 The types fall into a few useful classifications:
1348 * - :ref:`integer <t_integer>`
1349 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1352 * - :ref:`floating point <t_floating>`
1353 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1361 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1362 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1363 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1364 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1366 * - :ref:`primitive <t_primitive>`
1367 - :ref:`label <t_label>`,
1368 :ref:`void <t_void>`,
1369 :ref:`integer <t_integer>`,
1370 :ref:`floating point <t_floating>`,
1371 :ref:`x86mmx <t_x86mmx>`,
1372 :ref:`metadata <t_metadata>`.
1374 * - :ref:`derived <t_derived>`
1375 - :ref:`array <t_array>`,
1376 :ref:`function <t_function>`,
1377 :ref:`pointer <t_pointer>`,
1378 :ref:`structure <t_struct>`,
1379 :ref:`vector <t_vector>`,
1380 :ref:`opaque <t_opaque>`.
1382 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1383 Values of these types are the only ones which can be produced by
1391 The primitive types are the fundamental building blocks of the LLVM
1402 The integer type is a very simple type that simply specifies an
1403 arbitrary bit width for the integer type desired. Any bit width from 1
1404 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1413 The number of bits the integer will occupy is specified by the ``N``
1419 +----------------+------------------------------------------------+
1420 | ``i1`` | a single-bit integer. |
1421 +----------------+------------------------------------------------+
1422 | ``i32`` | a 32-bit integer. |
1423 +----------------+------------------------------------------------+
1424 | ``i1942652`` | a really big integer of over 1 million bits. |
1425 +----------------+------------------------------------------------+
1429 Floating Point Types
1430 ^^^^^^^^^^^^^^^^^^^^
1439 - 16-bit floating point value
1442 - 32-bit floating point value
1445 - 64-bit floating point value
1448 - 128-bit floating point value (112-bit mantissa)
1451 - 80-bit floating point value (X87)
1454 - 128-bit floating point value (two 64-bits)
1464 The x86mmx type represents a value held in an MMX register on an x86
1465 machine. The operations allowed on it are quite limited: parameters and
1466 return values, load and store, and bitcast. User-specified MMX
1467 instructions are represented as intrinsic or asm calls with arguments
1468 and/or results of this type. There are no arrays, vectors or constants
1486 The void type does not represent any value and has no size.
1503 The label type represents code labels.
1520 The metadata type represents embedded metadata. No derived types may be
1521 created from metadata except for :ref:`function <t_function>` arguments.
1535 The real power in LLVM comes from the derived types in the system. This
1536 is what allows a programmer to represent arrays, functions, pointers,
1537 and other useful types. Each of these types contain one or more element
1538 types which may be a primitive type, or another derived type. For
1539 example, it is possible to have a two dimensional array, using an array
1540 as the element type of another array.
1547 Aggregate Types are a subset of derived types that can contain multiple
1548 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1549 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1560 The array type is a very simple derived type that arranges elements
1561 sequentially in memory. The array type requires a size (number of
1562 elements) and an underlying data type.
1569 [<# elements> x <elementtype>]
1571 The number of elements is a constant integer value; ``elementtype`` may
1572 be any type with a size.
1577 +------------------+--------------------------------------+
1578 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1579 +------------------+--------------------------------------+
1580 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1581 +------------------+--------------------------------------+
1582 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1583 +------------------+--------------------------------------+
1585 Here are some examples of multidimensional arrays:
1587 +-----------------------------+----------------------------------------------------------+
1588 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1589 +-----------------------------+----------------------------------------------------------+
1590 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1591 +-----------------------------+----------------------------------------------------------+
1592 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1593 +-----------------------------+----------------------------------------------------------+
1595 There is no restriction on indexing beyond the end of the array implied
1596 by a static type (though there are restrictions on indexing beyond the
1597 bounds of an allocated object in some cases). This means that
1598 single-dimension 'variable sized array' addressing can be implemented in
1599 LLVM with a zero length array type. An implementation of 'pascal style
1600 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1611 The function type can be thought of as a function signature. It consists
1612 of a return type and a list of formal parameter types. The return type
1613 of a function type is a first class type or a void type.
1620 <returntype> (<parameter list>)
1622 ...where '``<parameter list>``' is a comma-separated list of type
1623 specifiers. Optionally, the parameter list may include a type ``...``,
1624 which indicates that the function takes a variable number of arguments.
1625 Variable argument functions can access their arguments with the
1626 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1627 '``<returntype>``' is any type except :ref:`label <t_label>`.
1632 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1633 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1634 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1635 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1636 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1637 | ``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. |
1638 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1639 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1640 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1650 The structure type is used to represent a collection of data members
1651 together in memory. The elements of a structure may be any type that has
1654 Structures in memory are accessed using '``load``' and '``store``' by
1655 getting a pointer to a field with the '``getelementptr``' instruction.
1656 Structures in registers are accessed using the '``extractvalue``' and
1657 '``insertvalue``' instructions.
1659 Structures may optionally be "packed" structures, which indicate that
1660 the alignment of the struct is one byte, and that there is no padding
1661 between the elements. In non-packed structs, padding between field types
1662 is inserted as defined by the DataLayout string in the module, which is
1663 required to match what the underlying code generator expects.
1665 Structures can either be "literal" or "identified". A literal structure
1666 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1667 identified types are always defined at the top level with a name.
1668 Literal types are uniqued by their contents and can never be recursive
1669 or opaque since there is no way to write one. Identified types can be
1670 recursive, can be opaqued, and are never uniqued.
1677 %T1 = type { <type list> } ; Identified normal struct type
1678 %T2 = type <{ <type list> }> ; Identified packed struct type
1683 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1684 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1685 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1686 | ``{ 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``. |
1687 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1688 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1689 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1693 Opaque Structure Types
1694 ^^^^^^^^^^^^^^^^^^^^^^
1699 Opaque structure types are used to represent named structure types that
1700 do not have a body specified. This corresponds (for example) to the C
1701 notion of a forward declared structure.
1714 +--------------+-------------------+
1715 | ``opaque`` | An opaque type. |
1716 +--------------+-------------------+
1726 The pointer type is used to specify memory locations. Pointers are
1727 commonly used to reference objects in memory.
1729 Pointer types may have an optional address space attribute defining the
1730 numbered address space where the pointed-to object resides. The default
1731 address space is number zero. The semantics of non-zero address spaces
1732 are target-specific.
1734 Note that LLVM does not permit pointers to void (``void*``) nor does it
1735 permit pointers to labels (``label*``). Use ``i8*`` instead.
1747 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1748 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1749 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1750 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1751 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1752 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1753 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1763 A vector type is a simple derived type that represents a vector of
1764 elements. Vector types are used when multiple primitive data are
1765 operated in parallel using a single instruction (SIMD). A vector type
1766 requires a size (number of elements) and an underlying primitive data
1767 type. Vector types are considered :ref:`first class <t_firstclass>`.
1774 < <# elements> x <elementtype> >
1776 The number of elements is a constant integer value larger than 0;
1777 elementtype may be any integer or floating point type, or a pointer to
1778 these types. Vectors of size zero are not allowed.
1783 +-------------------+--------------------------------------------------+
1784 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1785 +-------------------+--------------------------------------------------+
1786 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1787 +-------------------+--------------------------------------------------+
1788 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1789 +-------------------+--------------------------------------------------+
1790 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1791 +-------------------+--------------------------------------------------+
1796 LLVM has several different basic types of constants. This section
1797 describes them all and their syntax.
1802 **Boolean constants**
1803 The two strings '``true``' and '``false``' are both valid constants
1805 **Integer constants**
1806 Standard integers (such as '4') are constants of the
1807 :ref:`integer <t_integer>` type. Negative numbers may be used with
1809 **Floating point constants**
1810 Floating point constants use standard decimal notation (e.g.
1811 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1812 hexadecimal notation (see below). The assembler requires the exact
1813 decimal value of a floating-point constant. For example, the
1814 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1815 decimal in binary. Floating point constants must have a :ref:`floating
1816 point <t_floating>` type.
1817 **Null pointer constants**
1818 The identifier '``null``' is recognized as a null pointer constant
1819 and must be of :ref:`pointer type <t_pointer>`.
1821 The one non-intuitive notation for constants is the hexadecimal form of
1822 floating point constants. For example, the form
1823 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1824 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1825 constants are required (and the only time that they are generated by the
1826 disassembler) is when a floating point constant must be emitted but it
1827 cannot be represented as a decimal floating point number in a reasonable
1828 number of digits. For example, NaN's, infinities, and other special
1829 values are represented in their IEEE hexadecimal format so that assembly
1830 and disassembly do not cause any bits to change in the constants.
1832 When using the hexadecimal form, constants of types half, float, and
1833 double are represented using the 16-digit form shown above (which
1834 matches the IEEE754 representation for double); half and float values
1835 must, however, be exactly representable as IEEE 754 half and single
1836 precision, respectively. Hexadecimal format is always used for long
1837 double, and there are three forms of long double. The 80-bit format used
1838 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1839 128-bit format used by PowerPC (two adjacent doubles) is represented by
1840 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1841 represented by ``0xL`` followed by 32 hexadecimal digits; no currently
1842 supported target uses this format. Long doubles will only work if they
1843 match the long double format on your target. The IEEE 16-bit format
1844 (half precision) is represented by ``0xH`` followed by 4 hexadecimal
1845 digits. All hexadecimal formats are big-endian (sign bit at the left).
1847 There are no constants of type x86mmx.
1852 Complex constants are a (potentially recursive) combination of simple
1853 constants and smaller complex constants.
1855 **Structure constants**
1856 Structure constants are represented with notation similar to
1857 structure type definitions (a comma separated list of elements,
1858 surrounded by braces (``{}``)). For example:
1859 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1860 "``@G = external global i32``". Structure constants must have
1861 :ref:`structure type <t_struct>`, and the number and types of elements
1862 must match those specified by the type.
1864 Array constants are represented with notation similar to array type
1865 definitions (a comma separated list of elements, surrounded by
1866 square brackets (``[]``)). For example:
1867 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1868 :ref:`array type <t_array>`, and the number and types of elements must
1869 match those specified by the type.
1870 **Vector constants**
1871 Vector constants are represented with notation similar to vector
1872 type definitions (a comma separated list of elements, surrounded by
1873 less-than/greater-than's (``<>``)). For example:
1874 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1875 must have :ref:`vector type <t_vector>`, and the number and types of
1876 elements must match those specified by the type.
1877 **Zero initialization**
1878 The string '``zeroinitializer``' can be used to zero initialize a
1879 value to zero of *any* type, including scalar and
1880 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1881 having to print large zero initializers (e.g. for large arrays) and
1882 is always exactly equivalent to using explicit zero initializers.
1884 A metadata node is a structure-like constant with :ref:`metadata
1885 type <t_metadata>`. For example:
1886 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1887 constants that are meant to be interpreted as part of the
1888 instruction stream, metadata is a place to attach additional
1889 information such as debug info.
1891 Global Variable and Function Addresses
1892 --------------------------------------
1894 The addresses of :ref:`global variables <globalvars>` and
1895 :ref:`functions <functionstructure>` are always implicitly valid
1896 (link-time) constants. These constants are explicitly referenced when
1897 the :ref:`identifier for the global <identifiers>` is used and always have
1898 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1901 .. code-block:: llvm
1905 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1912 The string '``undef``' can be used anywhere a constant is expected, and
1913 indicates that the user of the value may receive an unspecified
1914 bit-pattern. Undefined values may be of any type (other than '``label``'
1915 or '``void``') and be used anywhere a constant is permitted.
1917 Undefined values are useful because they indicate to the compiler that
1918 the program is well defined no matter what value is used. This gives the
1919 compiler more freedom to optimize. Here are some examples of
1920 (potentially surprising) transformations that are valid (in pseudo IR):
1922 .. code-block:: llvm
1932 This is safe because all of the output bits are affected by the undef
1933 bits. Any output bit can have a zero or one depending on the input bits.
1935 .. code-block:: llvm
1946 These logical operations have bits that are not always affected by the
1947 input. For example, if ``%X`` has a zero bit, then the output of the
1948 '``and``' operation will always be a zero for that bit, no matter what
1949 the corresponding bit from the '``undef``' is. As such, it is unsafe to
1950 optimize or assume that the result of the '``and``' is '``undef``'.
1951 However, it is safe to assume that all bits of the '``undef``' could be
1952 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
1953 all the bits of the '``undef``' operand to the '``or``' could be set,
1954 allowing the '``or``' to be folded to -1.
1956 .. code-block:: llvm
1958 %A = select undef, %X, %Y
1959 %B = select undef, 42, %Y
1960 %C = select %X, %Y, undef
1970 This set of examples shows that undefined '``select``' (and conditional
1971 branch) conditions can go *either way*, but they have to come from one
1972 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
1973 both known to have a clear low bit, then ``%A`` would have to have a
1974 cleared low bit. However, in the ``%C`` example, the optimizer is
1975 allowed to assume that the '``undef``' operand could be the same as
1976 ``%Y``, allowing the whole '``select``' to be eliminated.
1978 .. code-block:: llvm
1980 %A = xor undef, undef
1997 This example points out that two '``undef``' operands are not
1998 necessarily the same. This can be surprising to people (and also matches
1999 C semantics) where they assume that "``X^X``" is always zero, even if
2000 ``X`` is undefined. This isn't true for a number of reasons, but the
2001 short answer is that an '``undef``' "variable" can arbitrarily change
2002 its value over its "live range". This is true because the variable
2003 doesn't actually *have a live range*. Instead, the value is logically
2004 read from arbitrary registers that happen to be around when needed, so
2005 the value is not necessarily consistent over time. In fact, ``%A`` and
2006 ``%C`` need to have the same semantics or the core LLVM "replace all
2007 uses with" concept would not hold.
2009 .. code-block:: llvm
2017 These examples show the crucial difference between an *undefined value*
2018 and *undefined behavior*. An undefined value (like '``undef``') is
2019 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2020 operation can be constant folded to '``undef``', because the '``undef``'
2021 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2022 However, in the second example, we can make a more aggressive
2023 assumption: because the ``undef`` is allowed to be an arbitrary value,
2024 we are allowed to assume that it could be zero. Since a divide by zero
2025 has *undefined behavior*, we are allowed to assume that the operation
2026 does not execute at all. This allows us to delete the divide and all
2027 code after it. Because the undefined operation "can't happen", the
2028 optimizer can assume that it occurs in dead code.
2030 .. code-block:: llvm
2032 a: store undef -> %X
2033 b: store %X -> undef
2038 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2039 value can be assumed to not have any effect; we can assume that the
2040 value is overwritten with bits that happen to match what was already
2041 there. However, a store *to* an undefined location could clobber
2042 arbitrary memory, therefore, it has undefined behavior.
2049 Poison values are similar to :ref:`undef values <undefvalues>`, however
2050 they also represent the fact that an instruction or constant expression
2051 which cannot evoke side effects has nevertheless detected a condition
2052 which results in undefined behavior.
2054 There is currently no way of representing a poison value in the IR; they
2055 only exist when produced by operations such as :ref:`add <i_add>` with
2058 Poison value behavior is defined in terms of value *dependence*:
2060 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2061 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2062 their dynamic predecessor basic block.
2063 - Function arguments depend on the corresponding actual argument values
2064 in the dynamic callers of their functions.
2065 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2066 instructions that dynamically transfer control back to them.
2067 - :ref:`Invoke <i_invoke>` instructions depend on the
2068 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2069 call instructions that dynamically transfer control back to them.
2070 - Non-volatile loads and stores depend on the most recent stores to all
2071 of the referenced memory addresses, following the order in the IR
2072 (including loads and stores implied by intrinsics such as
2073 :ref:`@llvm.memcpy <int_memcpy>`.)
2074 - An instruction with externally visible side effects depends on the
2075 most recent preceding instruction with externally visible side
2076 effects, following the order in the IR. (This includes :ref:`volatile
2077 operations <volatile>`.)
2078 - An instruction *control-depends* on a :ref:`terminator
2079 instruction <terminators>` if the terminator instruction has
2080 multiple successors and the instruction is always executed when
2081 control transfers to one of the successors, and may not be executed
2082 when control is transferred to another.
2083 - Additionally, an instruction also *control-depends* on a terminator
2084 instruction if the set of instructions it otherwise depends on would
2085 be different if the terminator had transferred control to a different
2087 - Dependence is transitive.
2089 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2090 with the additional affect that any instruction which has a *dependence*
2091 on a poison value has undefined behavior.
2093 Here are some examples:
2095 .. code-block:: llvm
2098 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2099 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2100 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2101 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2103 store i32 %poison, i32* @g ; Poison value stored to memory.
2104 %poison2 = load i32* @g ; Poison value loaded back from memory.
2106 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2108 %narrowaddr = bitcast i32* @g to i16*
2109 %wideaddr = bitcast i32* @g to i64*
2110 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2111 %poison4 = load i64* %wideaddr ; Returns a poison value.
2113 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2114 br i1 %cmp, label %true, label %end ; Branch to either destination.
2117 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2118 ; it has undefined behavior.
2122 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2123 ; Both edges into this PHI are
2124 ; control-dependent on %cmp, so this
2125 ; always results in a poison value.
2127 store volatile i32 0, i32* @g ; This would depend on the store in %true
2128 ; if %cmp is true, or the store in %entry
2129 ; otherwise, so this is undefined behavior.
2131 br i1 %cmp, label %second_true, label %second_end
2132 ; The same branch again, but this time the
2133 ; true block doesn't have side effects.
2140 store volatile i32 0, i32* @g ; This time, the instruction always depends
2141 ; on the store in %end. Also, it is
2142 ; control-equivalent to %end, so this is
2143 ; well-defined (ignoring earlier undefined
2144 ; behavior in this example).
2148 Addresses of Basic Blocks
2149 -------------------------
2151 ``blockaddress(@function, %block)``
2153 The '``blockaddress``' constant computes the address of the specified
2154 basic block in the specified function, and always has an ``i8*`` type.
2155 Taking the address of the entry block is illegal.
2157 This value only has defined behavior when used as an operand to the
2158 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2159 against null. Pointer equality tests between labels addresses results in
2160 undefined behavior --- though, again, comparison against null is ok, and
2161 no label is equal to the null pointer. This may be passed around as an
2162 opaque pointer sized value as long as the bits are not inspected. This
2163 allows ``ptrtoint`` and arithmetic to be performed on these values so
2164 long as the original value is reconstituted before the ``indirectbr``
2167 Finally, some targets may provide defined semantics when using the value
2168 as the operand to an inline assembly, but that is target specific.
2170 Constant Expressions
2171 --------------------
2173 Constant expressions are used to allow expressions involving other
2174 constants to be used as constants. Constant expressions may be of any
2175 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2176 that does not have side effects (e.g. load and call are not supported).
2177 The following is the syntax for constant expressions:
2179 ``trunc (CST to TYPE)``
2180 Truncate a constant to another type. The bit size of CST must be
2181 larger than the bit size of TYPE. Both types must be integers.
2182 ``zext (CST to TYPE)``
2183 Zero extend a constant to another type. The bit size of CST must be
2184 smaller than the bit size of TYPE. Both types must be integers.
2185 ``sext (CST to TYPE)``
2186 Sign extend a constant to another type. The bit size of CST must be
2187 smaller than the bit size of TYPE. Both types must be integers.
2188 ``fptrunc (CST to TYPE)``
2189 Truncate a floating point constant to another floating point type.
2190 The size of CST must be larger than the size of TYPE. Both types
2191 must be floating point.
2192 ``fpext (CST to TYPE)``
2193 Floating point extend a constant to another type. The size of CST
2194 must be smaller or equal to the size of TYPE. Both types must be
2196 ``fptoui (CST to TYPE)``
2197 Convert a floating point constant to the corresponding unsigned
2198 integer constant. TYPE must be a scalar or vector integer type. CST
2199 must be of scalar or vector floating point type. Both CST and TYPE
2200 must be scalars, or vectors of the same number of elements. If the
2201 value won't fit in the integer type, the results are undefined.
2202 ``fptosi (CST to TYPE)``
2203 Convert a floating point constant to the corresponding signed
2204 integer constant. TYPE must be a scalar or vector integer type. CST
2205 must be of scalar or vector floating point type. Both CST and TYPE
2206 must be scalars, or vectors of the same number of elements. If the
2207 value won't fit in the integer type, the results are undefined.
2208 ``uitofp (CST to TYPE)``
2209 Convert an unsigned integer constant to the corresponding floating
2210 point constant. TYPE must be a scalar or vector floating point type.
2211 CST must be of scalar or vector integer type. Both CST and TYPE must
2212 be scalars, or vectors of the same number of elements. If the value
2213 won't fit in the floating point type, the results are undefined.
2214 ``sitofp (CST to TYPE)``
2215 Convert a signed integer constant to the corresponding floating
2216 point constant. TYPE must be a scalar or vector floating point type.
2217 CST must be of scalar or vector integer type. Both CST and TYPE must
2218 be scalars, or vectors of the same number of elements. If the value
2219 won't fit in the floating point type, the results are undefined.
2220 ``ptrtoint (CST to TYPE)``
2221 Convert a pointer typed constant to the corresponding integer
2222 constant ``TYPE`` must be an integer type. ``CST`` must be of
2223 pointer type. The ``CST`` value is zero extended, truncated, or
2224 unchanged to make it fit in ``TYPE``.
2225 ``inttoptr (CST to TYPE)``
2226 Convert an integer constant to a pointer constant. TYPE must be a
2227 pointer type. CST must be of integer type. The CST value is zero
2228 extended, truncated, or unchanged to make it fit in a pointer size.
2229 This one is *really* dangerous!
2230 ``bitcast (CST to TYPE)``
2231 Convert a constant, CST, to another TYPE. The constraints of the
2232 operands are the same as those for the :ref:`bitcast
2233 instruction <i_bitcast>`.
2234 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2235 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2236 constants. As with the :ref:`getelementptr <i_getelementptr>`
2237 instruction, the index list may have zero or more indexes, which are
2238 required to make sense for the type of "CSTPTR".
2239 ``select (COND, VAL1, VAL2)``
2240 Perform the :ref:`select operation <i_select>` on constants.
2241 ``icmp COND (VAL1, VAL2)``
2242 Performs the :ref:`icmp operation <i_icmp>` on constants.
2243 ``fcmp COND (VAL1, VAL2)``
2244 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2245 ``extractelement (VAL, IDX)``
2246 Perform the :ref:`extractelement operation <i_extractelement>` on
2248 ``insertelement (VAL, ELT, IDX)``
2249 Perform the :ref:`insertelement operation <i_insertelement>` on
2251 ``shufflevector (VEC1, VEC2, IDXMASK)``
2252 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2254 ``extractvalue (VAL, IDX0, IDX1, ...)``
2255 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2256 constants. The index list is interpreted in a similar manner as
2257 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2258 least one index value must be specified.
2259 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2260 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2261 The index list is interpreted in a similar manner as indices in a
2262 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2263 value must be specified.
2264 ``OPCODE (LHS, RHS)``
2265 Perform the specified operation of the LHS and RHS constants. OPCODE
2266 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2267 binary <bitwiseops>` operations. The constraints on operands are
2268 the same as those for the corresponding instruction (e.g. no bitwise
2269 operations on floating point values are allowed).
2274 Inline Assembler Expressions
2275 ----------------------------
2277 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2278 Inline Assembly <moduleasm>`) through the use of a special value. This
2279 value represents the inline assembler as a string (containing the
2280 instructions to emit), a list of operand constraints (stored as a
2281 string), a flag that indicates whether or not the inline asm expression
2282 has side effects, and a flag indicating whether the function containing
2283 the asm needs to align its stack conservatively. An example inline
2284 assembler expression is:
2286 .. code-block:: llvm
2288 i32 (i32) asm "bswap $0", "=r,r"
2290 Inline assembler expressions may **only** be used as the callee operand
2291 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2292 Thus, typically we have:
2294 .. code-block:: llvm
2296 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2298 Inline asms with side effects not visible in the constraint list must be
2299 marked as having side effects. This is done through the use of the
2300 '``sideeffect``' keyword, like so:
2302 .. code-block:: llvm
2304 call void asm sideeffect "eieio", ""()
2306 In some cases inline asms will contain code that will not work unless
2307 the stack is aligned in some way, such as calls or SSE instructions on
2308 x86, yet will not contain code that does that alignment within the asm.
2309 The compiler should make conservative assumptions about what the asm
2310 might contain and should generate its usual stack alignment code in the
2311 prologue if the '``alignstack``' keyword is present:
2313 .. code-block:: llvm
2315 call void asm alignstack "eieio", ""()
2317 Inline asms also support using non-standard assembly dialects. The
2318 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2319 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2320 the only supported dialects. An example is:
2322 .. code-block:: llvm
2324 call void asm inteldialect "eieio", ""()
2326 If multiple keywords appear the '``sideeffect``' keyword must come
2327 first, the '``alignstack``' keyword second and the '``inteldialect``'
2333 The call instructions that wrap inline asm nodes may have a
2334 "``!srcloc``" MDNode attached to it that contains a list of constant
2335 integers. If present, the code generator will use the integer as the
2336 location cookie value when report errors through the ``LLVMContext``
2337 error reporting mechanisms. This allows a front-end to correlate backend
2338 errors that occur with inline asm back to the source code that produced
2341 .. code-block:: llvm
2343 call void asm sideeffect "something bad", ""(), !srcloc !42
2345 !42 = !{ i32 1234567 }
2347 It is up to the front-end to make sense of the magic numbers it places
2348 in the IR. If the MDNode contains multiple constants, the code generator
2349 will use the one that corresponds to the line of the asm that the error
2354 Metadata Nodes and Metadata Strings
2355 -----------------------------------
2357 LLVM IR allows metadata to be attached to instructions in the program
2358 that can convey extra information about the code to the optimizers and
2359 code generator. One example application of metadata is source-level
2360 debug information. There are two metadata primitives: strings and nodes.
2361 All metadata has the ``metadata`` type and is identified in syntax by a
2362 preceding exclamation point ('``!``').
2364 A metadata string is a string surrounded by double quotes. It can
2365 contain any character by escaping non-printable characters with
2366 "``\xx``" where "``xx``" is the two digit hex code. For example:
2369 Metadata nodes are represented with notation similar to structure
2370 constants (a comma separated list of elements, surrounded by braces and
2371 preceded by an exclamation point). Metadata nodes can have any values as
2372 their operand. For example:
2374 .. code-block:: llvm
2376 !{ metadata !"test\00", i32 10}
2378 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2379 metadata nodes, which can be looked up in the module symbol table. For
2382 .. code-block:: llvm
2384 !foo = metadata !{!4, !3}
2386 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2387 function is using two metadata arguments:
2389 .. code-block:: llvm
2391 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2393 Metadata can be attached with an instruction. Here metadata ``!21`` is
2394 attached to the ``add`` instruction using the ``!dbg`` identifier:
2396 .. code-block:: llvm
2398 %indvar.next = add i64 %indvar, 1, !dbg !21
2400 More information about specific metadata nodes recognized by the
2401 optimizers and code generator is found below.
2406 In LLVM IR, memory does not have types, so LLVM's own type system is not
2407 suitable for doing TBAA. Instead, metadata is added to the IR to
2408 describe a type system of a higher level language. This can be used to
2409 implement typical C/C++ TBAA, but it can also be used to implement
2410 custom alias analysis behavior for other languages.
2412 The current metadata format is very simple. TBAA metadata nodes have up
2413 to three fields, e.g.:
2415 .. code-block:: llvm
2417 !0 = metadata !{ metadata !"an example type tree" }
2418 !1 = metadata !{ metadata !"int", metadata !0 }
2419 !2 = metadata !{ metadata !"float", metadata !0 }
2420 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2422 The first field is an identity field. It can be any value, usually a
2423 metadata string, which uniquely identifies the type. The most important
2424 name in the tree is the name of the root node. Two trees with different
2425 root node names are entirely disjoint, even if they have leaves with
2428 The second field identifies the type's parent node in the tree, or is
2429 null or omitted for a root node. A type is considered to alias all of
2430 its descendants and all of its ancestors in the tree. Also, a type is
2431 considered to alias all types in other trees, so that bitcode produced
2432 from multiple front-ends is handled conservatively.
2434 If the third field is present, it's an integer which if equal to 1
2435 indicates that the type is "constant" (meaning
2436 ``pointsToConstantMemory`` should return true; see `other useful
2437 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2439 '``tbaa.struct``' Metadata
2440 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2442 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2443 aggregate assignment operations in C and similar languages, however it
2444 is defined to copy a contiguous region of memory, which is more than
2445 strictly necessary for aggregate types which contain holes due to
2446 padding. Also, it doesn't contain any TBAA information about the fields
2449 ``!tbaa.struct`` metadata can describe which memory subregions in a
2450 memcpy are padding and what the TBAA tags of the struct are.
2452 The current metadata format is very simple. ``!tbaa.struct`` metadata
2453 nodes are a list of operands which are in conceptual groups of three.
2454 For each group of three, the first operand gives the byte offset of a
2455 field in bytes, the second gives its size in bytes, and the third gives
2458 .. code-block:: llvm
2460 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2462 This describes a struct with two fields. The first is at offset 0 bytes
2463 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2464 and has size 4 bytes and has tbaa tag !2.
2466 Note that the fields need not be contiguous. In this example, there is a
2467 4 byte gap between the two fields. This gap represents padding which
2468 does not carry useful data and need not be preserved.
2470 '``fpmath``' Metadata
2471 ^^^^^^^^^^^^^^^^^^^^^
2473 ``fpmath`` metadata may be attached to any instruction of floating point
2474 type. It can be used to express the maximum acceptable error in the
2475 result of that instruction, in ULPs, thus potentially allowing the
2476 compiler to use a more efficient but less accurate method of computing
2477 it. ULP is defined as follows:
2479 If ``x`` is a real number that lies between two finite consecutive
2480 floating-point numbers ``a`` and ``b``, without being equal to one
2481 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2482 distance between the two non-equal finite floating-point numbers
2483 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2485 The metadata node shall consist of a single positive floating point
2486 number representing the maximum relative error, for example:
2488 .. code-block:: llvm
2490 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2492 '``range``' Metadata
2493 ^^^^^^^^^^^^^^^^^^^^
2495 ``range`` metadata may be attached only to loads of integer types. It
2496 expresses the possible ranges the loaded value is in. The ranges are
2497 represented with a flattened list of integers. The loaded value is known
2498 to be in the union of the ranges defined by each consecutive pair. Each
2499 pair has the following properties:
2501 - The type must match the type loaded by the instruction.
2502 - The pair ``a,b`` represents the range ``[a,b)``.
2503 - Both ``a`` and ``b`` are constants.
2504 - The range is allowed to wrap.
2505 - The range should not represent the full or empty set. That is,
2508 In addition, the pairs must be in signed order of the lower bound and
2509 they must be non-contiguous.
2513 .. code-block:: llvm
2515 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2516 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2517 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2518 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2520 !0 = metadata !{ i8 0, i8 2 }
2521 !1 = metadata !{ i8 255, i8 2 }
2522 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2523 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2528 It is sometimes useful to attach information to loop constructs. Currently,
2529 loop metadata is implemented as metadata attached to the branch instruction
2530 in the loop latch block. This type of metadata refer to a metadata node that is
2531 guaranteed to be separate for each loop. The loop-level metadata is prefixed
2534 The loop identifier metadata is implemented using a metadata that refers to
2537 .. code-block:: llvm
2539 !0 = metadata !{ metadata !0 }
2541 '``llvm.loop.parallel``' Metadata
2542 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2544 This loop metadata can be used to communicate that a loop should be considered
2545 a parallel loop. The semantics of parallel loops in this case is the one
2546 with the strongest cross-iteration instruction ordering freedom: the
2547 iterations in the loop can be considered completely independent of each
2548 other (also known as embarrassingly parallel loops).
2550 This metadata can originate from a programming language with parallel loop
2551 constructs. In such a case it is completely the programmer's responsibility
2552 to ensure the instructions from the different iterations of the loop can be
2553 executed in an arbitrary order, in parallel, or intertwined. No loop-carried
2554 dependency checking at all must be expected from the compiler.
2556 In order to fulfill the LLVM requirement for metadata to be safely ignored,
2557 it is important to ensure that a parallel loop is converted to
2558 a sequential loop in case an optimization (agnostic of the parallel loop
2559 semantics) converts the loop back to such. This happens when new memory
2560 accesses that do not fulfill the requirement of free ordering across iterations
2561 are added to the loop. Therefore, this metadata is required, but not
2562 sufficient, to consider the loop at hand a parallel loop. For a loop
2563 to be parallel, all its memory accessing instructions need to be
2564 marked with the ``llvm.mem.parallel_loop_access`` metadata that refer
2565 to the same loop identifier metadata that identify the loop at hand.
2570 Metadata types used to annotate memory accesses with information helpful
2571 for optimizations are prefixed with ``llvm.mem``.
2573 '``llvm.mem.parallel_loop_access``' Metadata
2574 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2576 For a loop to be parallel, in addition to using
2577 the ``llvm.loop.parallel`` metadata to mark the loop latch branch instruction,
2578 also all of the memory accessing instructions in the loop body need to be
2579 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2580 is at least one memory accessing instruction not marked with the metadata,
2581 the loop, despite it possibly using the ``llvm.loop.parallel`` metadata,
2582 must be considered a sequential loop. This causes parallel loops to be
2583 converted to sequential loops due to optimization passes that are unaware of
2584 the parallel semantics and that insert new memory instructions to the loop
2587 Example of a loop that is considered parallel due to its correct use of
2588 both ``llvm.loop.parallel`` and ``llvm.mem.parallel_loop_access``
2589 metadata types that refer to the same loop identifier metadata.
2591 .. code-block:: llvm
2595 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2597 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2599 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop.parallel !0
2603 !0 = metadata !{ metadata !0 }
2605 It is also possible to have nested parallel loops. In that case the
2606 memory accesses refer to a list of loop identifier metadata nodes instead of
2607 the loop identifier metadata node directly:
2609 .. code-block:: llvm
2616 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2618 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2620 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop.parallel !1
2624 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2626 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2628 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop.parallel !2
2630 outer.for.end: ; preds = %for.body
2632 !0 = metadata !{ metadata !1, metadata !2 } ; a list of parallel loop identifiers
2633 !1 = metadata !{ metadata !1 } ; an identifier for the inner parallel loop
2634 !2 = metadata !{ metadata !2 } ; an identifier for the outer parallel loop
2637 Module Flags Metadata
2638 =====================
2640 Information about the module as a whole is difficult to convey to LLVM's
2641 subsystems. The LLVM IR isn't sufficient to transmit this information.
2642 The ``llvm.module.flags`` named metadata exists in order to facilitate
2643 this. These flags are in the form of key / value pairs --- much like a
2644 dictionary --- making it easy for any subsystem who cares about a flag to
2647 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2648 Each triplet has the following form:
2650 - The first element is a *behavior* flag, which specifies the behavior
2651 when two (or more) modules are merged together, and it encounters two
2652 (or more) metadata with the same ID. The supported behaviors are
2654 - The second element is a metadata string that is a unique ID for the
2655 metadata. Each module may only have one flag entry for each unique ID (not
2656 including entries with the **Require** behavior).
2657 - The third element is the value of the flag.
2659 When two (or more) modules are merged together, the resulting
2660 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2661 each unique metadata ID string, there will be exactly one entry in the merged
2662 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2663 be determined by the merge behavior flag, as described below. The only exception
2664 is that entries with the *Require* behavior are always preserved.
2666 The following behaviors are supported:
2677 Emits an error if two values disagree, otherwise the resulting value
2678 is that of the operands.
2682 Emits a warning if two values disagree. The result value will be the
2683 operand for the flag from the first module being linked.
2687 Adds a requirement that another module flag be present and have a
2688 specified value after linking is performed. The value must be a
2689 metadata pair, where the first element of the pair is the ID of the
2690 module flag to be restricted, and the second element of the pair is
2691 the value the module flag should be restricted to. This behavior can
2692 be used to restrict the allowable results (via triggering of an
2693 error) of linking IDs with the **Override** behavior.
2697 Uses the specified value, regardless of the behavior or value of the
2698 other module. If both modules specify **Override**, but the values
2699 differ, an error will be emitted.
2703 Appends the two values, which are required to be metadata nodes.
2707 Appends the two values, which are required to be metadata
2708 nodes. However, duplicate entries in the second list are dropped
2709 during the append operation.
2711 It is an error for a particular unique flag ID to have multiple behaviors,
2712 except in the case of **Require** (which adds restrictions on another metadata
2713 value) or **Override**.
2715 An example of module flags:
2717 .. code-block:: llvm
2719 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2720 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2721 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2722 !3 = metadata !{ i32 3, metadata !"qux",
2724 metadata !"foo", i32 1
2727 !llvm.module.flags = !{ !0, !1, !2, !3 }
2729 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2730 if two or more ``!"foo"`` flags are seen is to emit an error if their
2731 values are not equal.
2733 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2734 behavior if two or more ``!"bar"`` flags are seen is to use the value
2737 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2738 behavior if two or more ``!"qux"`` flags are seen is to emit a
2739 warning if their values are not equal.
2741 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2745 metadata !{ metadata !"foo", i32 1 }
2747 The behavior is to emit an error if the ``llvm.module.flags`` does not
2748 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2751 Objective-C Garbage Collection Module Flags Metadata
2752 ----------------------------------------------------
2754 On the Mach-O platform, Objective-C stores metadata about garbage
2755 collection in a special section called "image info". The metadata
2756 consists of a version number and a bitmask specifying what types of
2757 garbage collection are supported (if any) by the file. If two or more
2758 modules are linked together their garbage collection metadata needs to
2759 be merged rather than appended together.
2761 The Objective-C garbage collection module flags metadata consists of the
2762 following key-value pairs:
2771 * - ``Objective-C Version``
2772 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2774 * - ``Objective-C Image Info Version``
2775 - **[Required]** --- The version of the image info section. Currently
2778 * - ``Objective-C Image Info Section``
2779 - **[Required]** --- The section to place the metadata. Valid values are
2780 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2781 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2782 Objective-C ABI version 2.
2784 * - ``Objective-C Garbage Collection``
2785 - **[Required]** --- Specifies whether garbage collection is supported or
2786 not. Valid values are 0, for no garbage collection, and 2, for garbage
2787 collection supported.
2789 * - ``Objective-C GC Only``
2790 - **[Optional]** --- Specifies that only garbage collection is supported.
2791 If present, its value must be 6. This flag requires that the
2792 ``Objective-C Garbage Collection`` flag have the value 2.
2794 Some important flag interactions:
2796 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2797 merged with a module with ``Objective-C Garbage Collection`` set to
2798 2, then the resulting module has the
2799 ``Objective-C Garbage Collection`` flag set to 0.
2800 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2801 merged with a module with ``Objective-C GC Only`` set to 6.
2803 Automatic Linker Flags Module Flags Metadata
2804 --------------------------------------------
2806 Some targets support embedding flags to the linker inside individual object
2807 files. Typically this is used in conjunction with language extensions which
2808 allow source files to explicitly declare the libraries they depend on, and have
2809 these automatically be transmitted to the linker via object files.
2811 These flags are encoded in the IR using metadata in the module flags section,
2812 using the ``Linker Options`` key. The merge behavior for this flag is required
2813 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2814 node which should be a list of other metadata nodes, each of which should be a
2815 list of metadata strings defining linker options.
2817 For example, the following metadata section specifies two separate sets of
2818 linker options, presumably to link against ``libz`` and the ``Cocoa``
2821 !0 = metadata !{ i32 6, metadata !"Linker Options",
2823 metadata !{ metadata !"-lz" },
2824 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2825 !llvm.module.flags = !{ !0 }
2827 The metadata encoding as lists of lists of options, as opposed to a collapsed
2828 list of options, is chosen so that the IR encoding can use multiple option
2829 strings to specify e.g., a single library, while still having that specifier be
2830 preserved as an atomic element that can be recognized by a target specific
2831 assembly writer or object file emitter.
2833 Each individual option is required to be either a valid option for the target's
2834 linker, or an option that is reserved by the target specific assembly writer or
2835 object file emitter. No other aspect of these options is defined by the IR.
2837 Intrinsic Global Variables
2838 ==========================
2840 LLVM has a number of "magic" global variables that contain data that
2841 affect code generation or other IR semantics. These are documented here.
2842 All globals of this sort should have a section specified as
2843 "``llvm.metadata``". This section and all globals that start with
2844 "``llvm.``" are reserved for use by LLVM.
2846 The '``llvm.used``' Global Variable
2847 -----------------------------------
2849 The ``@llvm.used`` global is an array with i8\* element type which has
2850 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2851 pointers to global variables and functions which may optionally have a
2852 pointer cast formed of bitcast or getelementptr. For example, a legal
2855 .. code-block:: llvm
2860 @llvm.used = appending global [2 x i8*] [
2862 i8* bitcast (i32* @Y to i8*)
2863 ], section "llvm.metadata"
2865 If a global variable appears in the ``@llvm.used`` list, then the
2866 compiler, assembler, and linker are required to treat the symbol as if
2867 there is a reference to the global that it cannot see. For example, if a
2868 variable has internal linkage and no references other than that from the
2869 ``@llvm.used`` list, it cannot be deleted. This is commonly used to
2870 represent references from inline asms and other things the compiler
2871 cannot "see", and corresponds to "``attribute((used))``" in GNU C.
2873 On some targets, the code generator must emit a directive to the
2874 assembler or object file to prevent the assembler and linker from
2875 molesting the symbol.
2877 The '``llvm.compiler.used``' Global Variable
2878 --------------------------------------------
2880 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2881 directive, except that it only prevents the compiler from touching the
2882 symbol. On targets that support it, this allows an intelligent linker to
2883 optimize references to the symbol without being impeded as it would be
2886 This is a rare construct that should only be used in rare circumstances,
2887 and should not be exposed to source languages.
2889 The '``llvm.global_ctors``' Global Variable
2890 -------------------------------------------
2892 .. code-block:: llvm
2894 %0 = type { i32, void ()* }
2895 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2897 The ``@llvm.global_ctors`` array contains a list of constructor
2898 functions and associated priorities. The functions referenced by this
2899 array will be called in ascending order of priority (i.e. lowest first)
2900 when the module is loaded. The order of functions with the same priority
2903 The '``llvm.global_dtors``' Global Variable
2904 -------------------------------------------
2906 .. code-block:: llvm
2908 %0 = type { i32, void ()* }
2909 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2911 The ``@llvm.global_dtors`` array contains a list of destructor functions
2912 and associated priorities. The functions referenced by this array will
2913 be called in descending order of priority (i.e. highest first) when the
2914 module is loaded. The order of functions with the same priority is not
2917 Instruction Reference
2918 =====================
2920 The LLVM instruction set consists of several different classifications
2921 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
2922 instructions <binaryops>`, :ref:`bitwise binary
2923 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
2924 :ref:`other instructions <otherops>`.
2928 Terminator Instructions
2929 -----------------------
2931 As mentioned :ref:`previously <functionstructure>`, every basic block in a
2932 program ends with a "Terminator" instruction, which indicates which
2933 block should be executed after the current block is finished. These
2934 terminator instructions typically yield a '``void``' value: they produce
2935 control flow, not values (the one exception being the
2936 ':ref:`invoke <i_invoke>`' instruction).
2938 The terminator instructions are: ':ref:`ret <i_ret>`',
2939 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
2940 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
2941 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
2945 '``ret``' Instruction
2946 ^^^^^^^^^^^^^^^^^^^^^
2953 ret <type> <value> ; Return a value from a non-void function
2954 ret void ; Return from void function
2959 The '``ret``' instruction is used to return control flow (and optionally
2960 a value) from a function back to the caller.
2962 There are two forms of the '``ret``' instruction: one that returns a
2963 value and then causes control flow, and one that just causes control
2969 The '``ret``' instruction optionally accepts a single argument, the
2970 return value. The type of the return value must be a ':ref:`first
2971 class <t_firstclass>`' type.
2973 A function is not :ref:`well formed <wellformed>` if it it has a non-void
2974 return type and contains a '``ret``' instruction with no return value or
2975 a return value with a type that does not match its type, or if it has a
2976 void return type and contains a '``ret``' instruction with a return
2982 When the '``ret``' instruction is executed, control flow returns back to
2983 the calling function's context. If the caller is a
2984 ":ref:`call <i_call>`" instruction, execution continues at the
2985 instruction after the call. If the caller was an
2986 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
2987 beginning of the "normal" destination block. If the instruction returns
2988 a value, that value shall set the call or invoke instruction's return
2994 .. code-block:: llvm
2996 ret i32 5 ; Return an integer value of 5
2997 ret void ; Return from a void function
2998 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3002 '``br``' Instruction
3003 ^^^^^^^^^^^^^^^^^^^^
3010 br i1 <cond>, label <iftrue>, label <iffalse>
3011 br label <dest> ; Unconditional branch
3016 The '``br``' instruction is used to cause control flow to transfer to a
3017 different basic block in the current function. There are two forms of
3018 this instruction, corresponding to a conditional branch and an
3019 unconditional branch.
3024 The conditional branch form of the '``br``' instruction takes a single
3025 '``i1``' value and two '``label``' values. The unconditional form of the
3026 '``br``' instruction takes a single '``label``' value as a target.
3031 Upon execution of a conditional '``br``' instruction, the '``i1``'
3032 argument is evaluated. If the value is ``true``, control flows to the
3033 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3034 to the '``iffalse``' ``label`` argument.
3039 .. code-block:: llvm
3042 %cond = icmp eq i32 %a, %b
3043 br i1 %cond, label %IfEqual, label %IfUnequal
3051 '``switch``' Instruction
3052 ^^^^^^^^^^^^^^^^^^^^^^^^
3059 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3064 The '``switch``' instruction is used to transfer control flow to one of
3065 several different places. It is a generalization of the '``br``'
3066 instruction, allowing a branch to occur to one of many possible
3072 The '``switch``' instruction uses three parameters: an integer
3073 comparison value '``value``', a default '``label``' destination, and an
3074 array of pairs of comparison value constants and '``label``'s. The table
3075 is not allowed to contain duplicate constant entries.
3080 The ``switch`` instruction specifies a table of values and destinations.
3081 When the '``switch``' instruction is executed, this table is searched
3082 for the given value. If the value is found, control flow is transferred
3083 to the corresponding destination; otherwise, control flow is transferred
3084 to the default destination.
3089 Depending on properties of the target machine and the particular
3090 ``switch`` instruction, this instruction may be code generated in
3091 different ways. For example, it could be generated as a series of
3092 chained conditional branches or with a lookup table.
3097 .. code-block:: llvm
3099 ; Emulate a conditional br instruction
3100 %Val = zext i1 %value to i32
3101 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3103 ; Emulate an unconditional br instruction
3104 switch i32 0, label %dest [ ]
3106 ; Implement a jump table:
3107 switch i32 %val, label %otherwise [ i32 0, label %onzero
3109 i32 2, label %ontwo ]
3113 '``indirectbr``' Instruction
3114 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3121 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3126 The '``indirectbr``' instruction implements an indirect branch to a
3127 label within the current function, whose address is specified by
3128 "``address``". Address must be derived from a
3129 :ref:`blockaddress <blockaddress>` constant.
3134 The '``address``' argument is the address of the label to jump to. The
3135 rest of the arguments indicate the full set of possible destinations
3136 that the address may point to. Blocks are allowed to occur multiple
3137 times in the destination list, though this isn't particularly useful.
3139 This destination list is required so that dataflow analysis has an
3140 accurate understanding of the CFG.
3145 Control transfers to the block specified in the address argument. All
3146 possible destination blocks must be listed in the label list, otherwise
3147 this instruction has undefined behavior. This implies that jumps to
3148 labels defined in other functions have undefined behavior as well.
3153 This is typically implemented with a jump through a register.
3158 .. code-block:: llvm
3160 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3164 '``invoke``' Instruction
3165 ^^^^^^^^^^^^^^^^^^^^^^^^
3172 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3173 to label <normal label> unwind label <exception label>
3178 The '``invoke``' instruction causes control to transfer to a specified
3179 function, with the possibility of control flow transfer to either the
3180 '``normal``' label or the '``exception``' label. If the callee function
3181 returns with the "``ret``" instruction, control flow will return to the
3182 "normal" label. If the callee (or any indirect callees) returns via the
3183 ":ref:`resume <i_resume>`" instruction or other exception handling
3184 mechanism, control is interrupted and continued at the dynamically
3185 nearest "exception" label.
3187 The '``exception``' label is a `landing
3188 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3189 '``exception``' label is required to have the
3190 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3191 information about the behavior of the program after unwinding happens,
3192 as its first non-PHI instruction. The restrictions on the
3193 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3194 instruction, so that the important information contained within the
3195 "``landingpad``" instruction can't be lost through normal code motion.
3200 This instruction requires several arguments:
3202 #. The optional "cconv" marker indicates which :ref:`calling
3203 convention <callingconv>` the call should use. If none is
3204 specified, the call defaults to using C calling conventions.
3205 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3206 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3208 #. '``ptr to function ty``': shall be the signature of the pointer to
3209 function value being invoked. In most cases, this is a direct
3210 function invocation, but indirect ``invoke``'s are just as possible,
3211 branching off an arbitrary pointer to function value.
3212 #. '``function ptr val``': An LLVM value containing a pointer to a
3213 function to be invoked.
3214 #. '``function args``': argument list whose types match the function
3215 signature argument types and parameter attributes. All arguments must
3216 be of :ref:`first class <t_firstclass>` type. If the function signature
3217 indicates the function accepts a variable number of arguments, the
3218 extra arguments can be specified.
3219 #. '``normal label``': the label reached when the called function
3220 executes a '``ret``' instruction.
3221 #. '``exception label``': the label reached when a callee returns via
3222 the :ref:`resume <i_resume>` instruction or other exception handling
3224 #. The optional :ref:`function attributes <fnattrs>` list. Only
3225 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3226 attributes are valid here.
3231 This instruction is designed to operate as a standard '``call``'
3232 instruction in most regards. The primary difference is that it
3233 establishes an association with a label, which is used by the runtime
3234 library to unwind the stack.
3236 This instruction is used in languages with destructors to ensure that
3237 proper cleanup is performed in the case of either a ``longjmp`` or a
3238 thrown exception. Additionally, this is important for implementation of
3239 '``catch``' clauses in high-level languages that support them.
3241 For the purposes of the SSA form, the definition of the value returned
3242 by the '``invoke``' instruction is deemed to occur on the edge from the
3243 current block to the "normal" label. If the callee unwinds then no
3244 return value is available.
3249 .. code-block:: llvm
3251 %retval = invoke i32 @Test(i32 15) to label %Continue
3252 unwind label %TestCleanup ; {i32}:retval set
3253 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3254 unwind label %TestCleanup ; {i32}:retval set
3258 '``resume``' Instruction
3259 ^^^^^^^^^^^^^^^^^^^^^^^^
3266 resume <type> <value>
3271 The '``resume``' instruction is a terminator instruction that has no
3277 The '``resume``' instruction requires one argument, which must have the
3278 same type as the result of any '``landingpad``' instruction in the same
3284 The '``resume``' instruction resumes propagation of an existing
3285 (in-flight) exception whose unwinding was interrupted with a
3286 :ref:`landingpad <i_landingpad>` instruction.
3291 .. code-block:: llvm
3293 resume { i8*, i32 } %exn
3297 '``unreachable``' Instruction
3298 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3310 The '``unreachable``' instruction has no defined semantics. This
3311 instruction is used to inform the optimizer that a particular portion of
3312 the code is not reachable. This can be used to indicate that the code
3313 after a no-return function cannot be reached, and other facts.
3318 The '``unreachable``' instruction has no defined semantics.
3325 Binary operators are used to do most of the computation in a program.
3326 They require two operands of the same type, execute an operation on
3327 them, and produce a single value. The operands might represent multiple
3328 data, as is the case with the :ref:`vector <t_vector>` data type. The
3329 result value has the same type as its operands.
3331 There are several different binary operators:
3335 '``add``' Instruction
3336 ^^^^^^^^^^^^^^^^^^^^^
3343 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3344 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3345 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3346 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3351 The '``add``' instruction returns the sum of its two operands.
3356 The two arguments to the '``add``' instruction must be
3357 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3358 arguments must have identical types.
3363 The value produced is the integer sum of the two operands.
3365 If the sum has unsigned overflow, the result returned is the
3366 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3369 Because LLVM integers use a two's complement representation, this
3370 instruction is appropriate for both signed and unsigned integers.
3372 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3373 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3374 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3375 unsigned and/or signed overflow, respectively, occurs.
3380 .. code-block:: llvm
3382 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3386 '``fadd``' Instruction
3387 ^^^^^^^^^^^^^^^^^^^^^^
3394 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3399 The '``fadd``' instruction returns the sum of its two operands.
3404 The two arguments to the '``fadd``' instruction must be :ref:`floating
3405 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3406 Both arguments must have identical types.
3411 The value produced is the floating point sum of the two operands. This
3412 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3413 which are optimization hints to enable otherwise unsafe floating point
3419 .. code-block:: llvm
3421 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3423 '``sub``' Instruction
3424 ^^^^^^^^^^^^^^^^^^^^^
3431 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3432 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3433 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3434 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3439 The '``sub``' instruction returns the difference of its two operands.
3441 Note that the '``sub``' instruction is used to represent the '``neg``'
3442 instruction present in most other intermediate representations.
3447 The two arguments to the '``sub``' instruction must be
3448 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3449 arguments must have identical types.
3454 The value produced is the integer difference of the two operands.
3456 If the difference has unsigned overflow, the result returned is the
3457 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3460 Because LLVM integers use a two's complement representation, this
3461 instruction is appropriate for both signed and unsigned integers.
3463 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3464 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3465 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3466 unsigned and/or signed overflow, respectively, occurs.
3471 .. code-block:: llvm
3473 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3474 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3478 '``fsub``' Instruction
3479 ^^^^^^^^^^^^^^^^^^^^^^
3486 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3491 The '``fsub``' instruction returns the difference of its two operands.
3493 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3494 instruction present in most other intermediate representations.
3499 The two arguments to the '``fsub``' instruction must be :ref:`floating
3500 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3501 Both arguments must have identical types.
3506 The value produced is the floating point difference of the two operands.
3507 This instruction can also take any number of :ref:`fast-math
3508 flags <fastmath>`, which are optimization hints to enable otherwise
3509 unsafe floating point optimizations:
3514 .. code-block:: llvm
3516 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3517 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3519 '``mul``' Instruction
3520 ^^^^^^^^^^^^^^^^^^^^^
3527 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3528 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3529 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3530 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3535 The '``mul``' instruction returns the product of its two operands.
3540 The two arguments to the '``mul``' instruction must be
3541 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3542 arguments must have identical types.
3547 The value produced is the integer product of the two operands.
3549 If the result of the multiplication has unsigned overflow, the result
3550 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3551 bit width of the result.
3553 Because LLVM integers use a two's complement representation, and the
3554 result is the same width as the operands, this instruction returns the
3555 correct result for both signed and unsigned integers. If a full product
3556 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3557 sign-extended or zero-extended as appropriate to the width of the full
3560 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3561 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3562 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3563 unsigned and/or signed overflow, respectively, occurs.
3568 .. code-block:: llvm
3570 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3574 '``fmul``' Instruction
3575 ^^^^^^^^^^^^^^^^^^^^^^
3582 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3587 The '``fmul``' instruction returns the product of its two operands.
3592 The two arguments to the '``fmul``' instruction must be :ref:`floating
3593 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3594 Both arguments must have identical types.
3599 The value produced is the floating point product of the two operands.
3600 This instruction can also take any number of :ref:`fast-math
3601 flags <fastmath>`, which are optimization hints to enable otherwise
3602 unsafe floating point optimizations:
3607 .. code-block:: llvm
3609 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3611 '``udiv``' Instruction
3612 ^^^^^^^^^^^^^^^^^^^^^^
3619 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3620 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3625 The '``udiv``' instruction returns the quotient of its two operands.
3630 The two arguments to the '``udiv``' instruction must be
3631 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3632 arguments must have identical types.
3637 The value produced is the unsigned integer quotient of the two operands.
3639 Note that unsigned integer division and signed integer division are
3640 distinct operations; for signed integer division, use '``sdiv``'.
3642 Division by zero leads to undefined behavior.
3644 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3645 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3646 such, "((a udiv exact b) mul b) == a").
3651 .. code-block:: llvm
3653 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3655 '``sdiv``' Instruction
3656 ^^^^^^^^^^^^^^^^^^^^^^
3663 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3664 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3669 The '``sdiv``' instruction returns the quotient of its two operands.
3674 The two arguments to the '``sdiv``' instruction must be
3675 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3676 arguments must have identical types.
3681 The value produced is the signed integer quotient of the two operands
3682 rounded towards zero.
3684 Note that signed integer division and unsigned integer division are
3685 distinct operations; for unsigned integer division, use '``udiv``'.
3687 Division by zero leads to undefined behavior. Overflow also leads to
3688 undefined behavior; this is a rare case, but can occur, for example, by
3689 doing a 32-bit division of -2147483648 by -1.
3691 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3692 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3697 .. code-block:: llvm
3699 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3703 '``fdiv``' Instruction
3704 ^^^^^^^^^^^^^^^^^^^^^^
3711 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3716 The '``fdiv``' instruction returns the quotient of its two operands.
3721 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3722 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3723 Both arguments must have identical types.
3728 The value produced is the floating point quotient of the two operands.
3729 This instruction can also take any number of :ref:`fast-math
3730 flags <fastmath>`, which are optimization hints to enable otherwise
3731 unsafe floating point optimizations:
3736 .. code-block:: llvm
3738 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3740 '``urem``' Instruction
3741 ^^^^^^^^^^^^^^^^^^^^^^
3748 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3753 The '``urem``' instruction returns the remainder from the unsigned
3754 division of its two arguments.
3759 The two arguments to the '``urem``' instruction must be
3760 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3761 arguments must have identical types.
3766 This instruction returns the unsigned integer *remainder* of a division.
3767 This instruction always performs an unsigned division to get the
3770 Note that unsigned integer remainder and signed integer remainder are
3771 distinct operations; for signed integer remainder, use '``srem``'.
3773 Taking the remainder of a division by zero leads to undefined behavior.
3778 .. code-block:: llvm
3780 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3782 '``srem``' Instruction
3783 ^^^^^^^^^^^^^^^^^^^^^^
3790 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3795 The '``srem``' instruction returns the remainder from the signed
3796 division of its two operands. This instruction can also take
3797 :ref:`vector <t_vector>` versions of the values in which case the elements
3803 The two arguments to the '``srem``' instruction must be
3804 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3805 arguments must have identical types.
3810 This instruction returns the *remainder* of a division (where the result
3811 is either zero or has the same sign as the dividend, ``op1``), not the
3812 *modulo* operator (where the result is either zero or has the same sign
3813 as the divisor, ``op2``) of a value. For more information about the
3814 difference, see `The Math
3815 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3816 table of how this is implemented in various languages, please see
3818 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3820 Note that signed integer remainder and unsigned integer remainder are
3821 distinct operations; for unsigned integer remainder, use '``urem``'.
3823 Taking the remainder of a division by zero leads to undefined behavior.
3824 Overflow also leads to undefined behavior; this is a rare case, but can
3825 occur, for example, by taking the remainder of a 32-bit division of
3826 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3827 rule lets srem be implemented using instructions that return both the
3828 result of the division and the remainder.)
3833 .. code-block:: llvm
3835 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3839 '``frem``' Instruction
3840 ^^^^^^^^^^^^^^^^^^^^^^
3847 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3852 The '``frem``' instruction returns the remainder from the division of
3858 The two arguments to the '``frem``' instruction must be :ref:`floating
3859 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3860 Both arguments must have identical types.
3865 This instruction returns the *remainder* of a division. The remainder
3866 has the same sign as the dividend. This instruction can also take any
3867 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3868 to enable otherwise unsafe floating point optimizations:
3873 .. code-block:: llvm
3875 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3879 Bitwise Binary Operations
3880 -------------------------
3882 Bitwise binary operators are used to do various forms of bit-twiddling
3883 in a program. They are generally very efficient instructions and can
3884 commonly be strength reduced from other instructions. They require two
3885 operands of the same type, execute an operation on them, and produce a
3886 single value. The resulting value is the same type as its operands.
3888 '``shl``' Instruction
3889 ^^^^^^^^^^^^^^^^^^^^^
3896 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
3897 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
3898 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
3899 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3904 The '``shl``' instruction returns the first operand shifted to the left
3905 a specified number of bits.
3910 Both arguments to the '``shl``' instruction must be the same
3911 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3912 '``op2``' is treated as an unsigned value.
3917 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
3918 where ``n`` is the width of the result. If ``op2`` is (statically or
3919 dynamically) negative or equal to or larger than the number of bits in
3920 ``op1``, the result is undefined. If the arguments are vectors, each
3921 vector element of ``op1`` is shifted by the corresponding shift amount
3924 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
3925 value <poisonvalues>` if it shifts out any non-zero bits. If the
3926 ``nsw`` keyword is present, then the shift produces a :ref:`poison
3927 value <poisonvalues>` if it shifts out any bits that disagree with the
3928 resultant sign bit. As such, NUW/NSW have the same semantics as they
3929 would if the shift were expressed as a mul instruction with the same
3930 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
3935 .. code-block:: llvm
3937 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
3938 <result> = shl i32 4, 2 ; yields {i32}: 16
3939 <result> = shl i32 1, 10 ; yields {i32}: 1024
3940 <result> = shl i32 1, 32 ; undefined
3941 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
3943 '``lshr``' Instruction
3944 ^^^^^^^^^^^^^^^^^^^^^^
3951 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
3952 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
3957 The '``lshr``' instruction (logical shift right) returns the first
3958 operand shifted to the right a specified number of bits with zero fill.
3963 Both arguments to the '``lshr``' instruction must be the same
3964 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3965 '``op2``' is treated as an unsigned value.
3970 This instruction always performs a logical shift right operation. The
3971 most significant bits of the result will be filled with zero bits after
3972 the shift. If ``op2`` is (statically or dynamically) equal to or larger
3973 than the number of bits in ``op1``, the result is undefined. If the
3974 arguments are vectors, each vector element of ``op1`` is shifted by the
3975 corresponding shift amount in ``op2``.
3977 If the ``exact`` keyword is present, the result value of the ``lshr`` is
3978 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3984 .. code-block:: llvm
3986 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
3987 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
3988 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
3989 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7FFFFFFF
3990 <result> = lshr i32 1, 32 ; undefined
3991 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
3993 '``ashr``' Instruction
3994 ^^^^^^^^^^^^^^^^^^^^^^
4001 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4002 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4007 The '``ashr``' instruction (arithmetic shift right) returns the first
4008 operand shifted to the right a specified number of bits with sign
4014 Both arguments to the '``ashr``' instruction must be the same
4015 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4016 '``op2``' is treated as an unsigned value.
4021 This instruction always performs an arithmetic shift right operation,
4022 The most significant bits of the result will be filled with the sign bit
4023 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4024 than the number of bits in ``op1``, the result is undefined. If the
4025 arguments are vectors, each vector element of ``op1`` is shifted by the
4026 corresponding shift amount in ``op2``.
4028 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4029 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4035 .. code-block:: llvm
4037 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4038 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4039 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4040 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4041 <result> = ashr i32 1, 32 ; undefined
4042 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4044 '``and``' Instruction
4045 ^^^^^^^^^^^^^^^^^^^^^
4052 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4057 The '``and``' instruction returns the bitwise logical and of its two
4063 The two arguments to the '``and``' instruction must be
4064 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4065 arguments must have identical types.
4070 The truth table used for the '``and``' instruction is:
4087 .. code-block:: llvm
4089 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4090 <result> = and i32 15, 40 ; yields {i32}:result = 8
4091 <result> = and i32 4, 8 ; yields {i32}:result = 0
4093 '``or``' Instruction
4094 ^^^^^^^^^^^^^^^^^^^^
4101 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4106 The '``or``' instruction returns the bitwise logical inclusive or of its
4112 The two arguments to the '``or``' instruction must be
4113 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4114 arguments must have identical types.
4119 The truth table used for the '``or``' instruction is:
4138 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4139 <result> = or i32 15, 40 ; yields {i32}:result = 47
4140 <result> = or i32 4, 8 ; yields {i32}:result = 12
4142 '``xor``' Instruction
4143 ^^^^^^^^^^^^^^^^^^^^^
4150 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4155 The '``xor``' instruction returns the bitwise logical exclusive or of
4156 its two operands. The ``xor`` is used to implement the "one's
4157 complement" operation, which is the "~" operator in C.
4162 The two arguments to the '``xor``' instruction must be
4163 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4164 arguments must have identical types.
4169 The truth table used for the '``xor``' instruction is:
4186 .. code-block:: llvm
4188 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4189 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4190 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4191 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4196 LLVM supports several instructions to represent vector operations in a
4197 target-independent manner. These instructions cover the element-access
4198 and vector-specific operations needed to process vectors effectively.
4199 While LLVM does directly support these vector operations, many
4200 sophisticated algorithms will want to use target-specific intrinsics to
4201 take full advantage of a specific target.
4203 .. _i_extractelement:
4205 '``extractelement``' Instruction
4206 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4213 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4218 The '``extractelement``' instruction extracts a single scalar element
4219 from a vector at a specified index.
4224 The first operand of an '``extractelement``' instruction is a value of
4225 :ref:`vector <t_vector>` type. The second operand is an index indicating
4226 the position from which to extract the element. The index may be a
4232 The result is a scalar of the same type as the element type of ``val``.
4233 Its value is the value at position ``idx`` of ``val``. If ``idx``
4234 exceeds the length of ``val``, the results are undefined.
4239 .. code-block:: llvm
4241 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4243 .. _i_insertelement:
4245 '``insertelement``' Instruction
4246 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4253 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4258 The '``insertelement``' instruction inserts a scalar element into a
4259 vector at a specified index.
4264 The first operand of an '``insertelement``' instruction is a value of
4265 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4266 type must equal the element type of the first operand. The third operand
4267 is an index indicating the position at which to insert the value. The
4268 index may be a variable.
4273 The result is a vector of the same type as ``val``. Its element values
4274 are those of ``val`` except at position ``idx``, where it gets the value
4275 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4281 .. code-block:: llvm
4283 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4285 .. _i_shufflevector:
4287 '``shufflevector``' Instruction
4288 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4295 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4300 The '``shufflevector``' instruction constructs a permutation of elements
4301 from two input vectors, returning a vector with the same element type as
4302 the input and length that is the same as the shuffle mask.
4307 The first two operands of a '``shufflevector``' instruction are vectors
4308 with the same type. The third argument is a shuffle mask whose element
4309 type is always 'i32'. The result of the instruction is a vector whose
4310 length is the same as the shuffle mask and whose element type is the
4311 same as the element type of the first two operands.
4313 The shuffle mask operand is required to be a constant vector with either
4314 constant integer or undef values.
4319 The elements of the two input vectors are numbered from left to right
4320 across both of the vectors. The shuffle mask operand specifies, for each
4321 element of the result vector, which element of the two input vectors the
4322 result element gets. The element selector may be undef (meaning "don't
4323 care") and the second operand may be undef if performing a shuffle from
4329 .. code-block:: llvm
4331 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4332 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4333 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4334 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4335 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4336 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4337 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4338 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4340 Aggregate Operations
4341 --------------------
4343 LLVM supports several instructions for working with
4344 :ref:`aggregate <t_aggregate>` values.
4348 '``extractvalue``' Instruction
4349 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4356 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4361 The '``extractvalue``' instruction extracts the value of a member field
4362 from an :ref:`aggregate <t_aggregate>` value.
4367 The first operand of an '``extractvalue``' instruction is a value of
4368 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4369 constant indices to specify which value to extract in a similar manner
4370 as indices in a '``getelementptr``' instruction.
4372 The major differences to ``getelementptr`` indexing are:
4374 - Since the value being indexed is not a pointer, the first index is
4375 omitted and assumed to be zero.
4376 - At least one index must be specified.
4377 - Not only struct indices but also array indices must be in bounds.
4382 The result is the value at the position in the aggregate specified by
4388 .. code-block:: llvm
4390 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4394 '``insertvalue``' Instruction
4395 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4402 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4407 The '``insertvalue``' instruction inserts a value into a member field in
4408 an :ref:`aggregate <t_aggregate>` value.
4413 The first operand of an '``insertvalue``' instruction is a value of
4414 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4415 a first-class value to insert. The following operands are constant
4416 indices indicating the position at which to insert the value in a
4417 similar manner as indices in a '``extractvalue``' instruction. The value
4418 to insert must have the same type as the value identified by the
4424 The result is an aggregate of the same type as ``val``. Its value is
4425 that of ``val`` except that the value at the position specified by the
4426 indices is that of ``elt``.
4431 .. code-block:: llvm
4433 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4434 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4435 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4439 Memory Access and Addressing Operations
4440 ---------------------------------------
4442 A key design point of an SSA-based representation is how it represents
4443 memory. In LLVM, no memory locations are in SSA form, which makes things
4444 very simple. This section describes how to read, write, and allocate
4449 '``alloca``' Instruction
4450 ^^^^^^^^^^^^^^^^^^^^^^^^
4457 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4462 The '``alloca``' instruction allocates memory on the stack frame of the
4463 currently executing function, to be automatically released when this
4464 function returns to its caller. The object is always allocated in the
4465 generic address space (address space zero).
4470 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4471 bytes of memory on the runtime stack, returning a pointer of the
4472 appropriate type to the program. If "NumElements" is specified, it is
4473 the number of elements allocated, otherwise "NumElements" is defaulted
4474 to be one. If a constant alignment is specified, the value result of the
4475 allocation is guaranteed to be aligned to at least that boundary. If not
4476 specified, or if zero, the target can choose to align the allocation on
4477 any convenient boundary compatible with the type.
4479 '``type``' may be any sized type.
4484 Memory is allocated; a pointer is returned. The operation is undefined
4485 if there is insufficient stack space for the allocation. '``alloca``'d
4486 memory is automatically released when the function returns. The
4487 '``alloca``' instruction is commonly used to represent automatic
4488 variables that must have an address available. When the function returns
4489 (either with the ``ret`` or ``resume`` instructions), the memory is
4490 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4491 The order in which memory is allocated (ie., which way the stack grows)
4497 .. code-block:: llvm
4499 %ptr = alloca i32 ; yields {i32*}:ptr
4500 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4501 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4502 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4506 '``load``' Instruction
4507 ^^^^^^^^^^^^^^^^^^^^^^
4514 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4515 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4516 !<index> = !{ i32 1 }
4521 The '``load``' instruction is used to read from memory.
4526 The argument to the '``load``' instruction specifies the memory address
4527 from which to load. The pointer must point to a :ref:`first
4528 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4529 then the optimizer is not allowed to modify the number or order of
4530 execution of this ``load`` with other :ref:`volatile
4531 operations <volatile>`.
4533 If the ``load`` is marked as ``atomic``, it takes an extra
4534 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4535 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4536 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4537 when they may see multiple atomic stores. The type of the pointee must
4538 be an integer type whose bit width is a power of two greater than or
4539 equal to eight and less than or equal to a target-specific size limit.
4540 ``align`` must be explicitly specified on atomic loads, and the load has
4541 undefined behavior if the alignment is not set to a value which is at
4542 least the size in bytes of the pointee. ``!nontemporal`` does not have
4543 any defined semantics for atomic loads.
4545 The optional constant ``align`` argument specifies the alignment of the
4546 operation (that is, the alignment of the memory address). A value of 0
4547 or an omitted ``align`` argument means that the operation has the abi
4548 alignment for the target. It is the responsibility of the code emitter
4549 to ensure that the alignment information is correct. Overestimating the
4550 alignment results in undefined behavior. Underestimating the alignment
4551 may produce less efficient code. An alignment of 1 is always safe.
4553 The optional ``!nontemporal`` metadata must reference a single
4554 metatadata name <index> corresponding to a metadata node with one
4555 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4556 metatadata on the instruction tells the optimizer and code generator
4557 that this load is not expected to be reused in the cache. The code
4558 generator may select special instructions to save cache bandwidth, such
4559 as the ``MOVNT`` instruction on x86.
4561 The optional ``!invariant.load`` metadata must reference a single
4562 metatadata name <index> corresponding to a metadata node with no
4563 entries. The existence of the ``!invariant.load`` metatadata on the
4564 instruction tells the optimizer and code generator that this load
4565 address points to memory which does not change value during program
4566 execution. The optimizer may then move this load around, for example, by
4567 hoisting it out of loops using loop invariant code motion.
4572 The location of memory pointed to is loaded. If the value being loaded
4573 is of scalar type then the number of bytes read does not exceed the
4574 minimum number of bytes needed to hold all bits of the type. For
4575 example, loading an ``i24`` reads at most three bytes. When loading a
4576 value of a type like ``i20`` with a size that is not an integral number
4577 of bytes, the result is undefined if the value was not originally
4578 written using a store of the same type.
4583 .. code-block:: llvm
4585 %ptr = alloca i32 ; yields {i32*}:ptr
4586 store i32 3, i32* %ptr ; yields {void}
4587 %val = load i32* %ptr ; yields {i32}:val = i32 3
4591 '``store``' Instruction
4592 ^^^^^^^^^^^^^^^^^^^^^^^
4599 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4600 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4605 The '``store``' instruction is used to write to memory.
4610 There are two arguments to the '``store``' instruction: a value to store
4611 and an address at which to store it. The type of the '``<pointer>``'
4612 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4613 the '``<value>``' operand. If the ``store`` is marked as ``volatile``,
4614 then the optimizer is not allowed to modify the number or order of
4615 execution of this ``store`` with other :ref:`volatile
4616 operations <volatile>`.
4618 If the ``store`` is marked as ``atomic``, it takes an extra
4619 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4620 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4621 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4622 when they may see multiple atomic stores. The type of the pointee must
4623 be an integer type whose bit width is a power of two greater than or
4624 equal to eight and less than or equal to a target-specific size limit.
4625 ``align`` must be explicitly specified on atomic stores, and the store
4626 has undefined behavior if the alignment is not set to a value which is
4627 at least the size in bytes of the pointee. ``!nontemporal`` does not
4628 have any defined semantics for atomic stores.
4630 The optional constant "align" argument specifies the alignment of the
4631 operation (that is, the alignment of the memory address). A value of 0
4632 or an omitted "align" argument means that the operation has the abi
4633 alignment for the target. It is the responsibility of the code emitter
4634 to ensure that the alignment information is correct. Overestimating the
4635 alignment results in an undefined behavior. Underestimating the
4636 alignment may produce less efficient code. An alignment of 1 is always
4639 The optional !nontemporal metadata must reference a single metatadata
4640 name <index> corresponding to a metadata node with one i32 entry of
4641 value 1. The existence of the !nontemporal metatadata on the instruction
4642 tells the optimizer and code generator that this load is not expected to
4643 be reused in the cache. The code generator may select special
4644 instructions to save cache bandwidth, such as the MOVNT instruction on
4650 The contents of memory are updated to contain '``<value>``' at the
4651 location specified by the '``<pointer>``' operand. If '``<value>``' is
4652 of scalar type then the number of bytes written does not exceed the
4653 minimum number of bytes needed to hold all bits of the type. For
4654 example, storing an ``i24`` writes at most three bytes. When writing a
4655 value of a type like ``i20`` with a size that is not an integral number
4656 of bytes, it is unspecified what happens to the extra bits that do not
4657 belong to the type, but they will typically be overwritten.
4662 .. code-block:: llvm
4664 %ptr = alloca i32 ; yields {i32*}:ptr
4665 store i32 3, i32* %ptr ; yields {void}
4666 %val = load i32* %ptr ; yields {i32}:val = i32 3
4670 '``fence``' Instruction
4671 ^^^^^^^^^^^^^^^^^^^^^^^
4678 fence [singlethread] <ordering> ; yields {void}
4683 The '``fence``' instruction is used to introduce happens-before edges
4689 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4690 defines what *synchronizes-with* edges they add. They can only be given
4691 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4696 A fence A which has (at least) ``release`` ordering semantics
4697 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4698 semantics if and only if there exist atomic operations X and Y, both
4699 operating on some atomic object M, such that A is sequenced before X, X
4700 modifies M (either directly or through some side effect of a sequence
4701 headed by X), Y is sequenced before B, and Y observes M. This provides a
4702 *happens-before* dependency between A and B. Rather than an explicit
4703 ``fence``, one (but not both) of the atomic operations X or Y might
4704 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4705 still *synchronize-with* the explicit ``fence`` and establish the
4706 *happens-before* edge.
4708 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4709 ``acquire`` and ``release`` semantics specified above, participates in
4710 the global program order of other ``seq_cst`` operations and/or fences.
4712 The optional ":ref:`singlethread <singlethread>`" argument specifies
4713 that the fence only synchronizes with other fences in the same thread.
4714 (This is useful for interacting with signal handlers.)
4719 .. code-block:: llvm
4721 fence acquire ; yields {void}
4722 fence singlethread seq_cst ; yields {void}
4726 '``cmpxchg``' Instruction
4727 ^^^^^^^^^^^^^^^^^^^^^^^^^
4734 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4739 The '``cmpxchg``' instruction is used to atomically modify memory. It
4740 loads a value in memory and compares it to a given value. If they are
4741 equal, it stores a new value into the memory.
4746 There are three arguments to the '``cmpxchg``' instruction: an address
4747 to operate on, a value to compare to the value currently be at that
4748 address, and a new value to place at that address if the compared values
4749 are equal. The type of '<cmp>' must be an integer type whose bit width
4750 is a power of two greater than or equal to eight and less than or equal
4751 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4752 type, and the type of '<pointer>' must be a pointer to that type. If the
4753 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4754 to modify the number or order of execution of this ``cmpxchg`` with
4755 other :ref:`volatile operations <volatile>`.
4757 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4758 synchronizes with other atomic operations.
4760 The optional "``singlethread``" argument declares that the ``cmpxchg``
4761 is only atomic with respect to code (usually signal handlers) running in
4762 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4763 respect to all other code in the system.
4765 The pointer passed into cmpxchg must have alignment greater than or
4766 equal to the size in memory of the operand.
4771 The contents of memory at the location specified by the '``<pointer>``'
4772 operand is read and compared to '``<cmp>``'; if the read value is the
4773 equal, '``<new>``' is written. The original value at the location is
4776 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4777 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4778 atomic load with an ordering parameter determined by dropping any
4779 ``release`` part of the ``cmpxchg``'s ordering.
4784 .. code-block:: llvm
4787 %orig = atomic load i32* %ptr unordered ; yields {i32}
4791 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4792 %squared = mul i32 %cmp, %cmp
4793 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4794 %success = icmp eq i32 %cmp, %old
4795 br i1 %success, label %done, label %loop
4802 '``atomicrmw``' Instruction
4803 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4810 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4815 The '``atomicrmw``' instruction is used to atomically modify memory.
4820 There are three arguments to the '``atomicrmw``' instruction: an
4821 operation to apply, an address whose value to modify, an argument to the
4822 operation. The operation must be one of the following keywords:
4836 The type of '<value>' must be an integer type whose bit width is a power
4837 of two greater than or equal to eight and less than or equal to a
4838 target-specific size limit. The type of the '``<pointer>``' operand must
4839 be a pointer to that type. If the ``atomicrmw`` is marked as
4840 ``volatile``, then the optimizer is not allowed to modify the number or
4841 order of execution of this ``atomicrmw`` with other :ref:`volatile
4842 operations <volatile>`.
4847 The contents of memory at the location specified by the '``<pointer>``'
4848 operand are atomically read, modified, and written back. The original
4849 value at the location is returned. The modification is specified by the
4852 - xchg: ``*ptr = val``
4853 - add: ``*ptr = *ptr + val``
4854 - sub: ``*ptr = *ptr - val``
4855 - and: ``*ptr = *ptr & val``
4856 - nand: ``*ptr = ~(*ptr & val)``
4857 - or: ``*ptr = *ptr | val``
4858 - xor: ``*ptr = *ptr ^ val``
4859 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4860 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4861 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4863 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4869 .. code-block:: llvm
4871 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4873 .. _i_getelementptr:
4875 '``getelementptr``' Instruction
4876 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4883 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4884 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4885 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
4890 The '``getelementptr``' instruction is used to get the address of a
4891 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
4892 address calculation only and does not access memory.
4897 The first argument is always a pointer or a vector of pointers, and
4898 forms the basis of the calculation. The remaining arguments are indices
4899 that indicate which of the elements of the aggregate object are indexed.
4900 The interpretation of each index is dependent on the type being indexed
4901 into. The first index always indexes the pointer value given as the
4902 first argument, the second index indexes a value of the type pointed to
4903 (not necessarily the value directly pointed to, since the first index
4904 can be non-zero), etc. The first type indexed into must be a pointer
4905 value, subsequent types can be arrays, vectors, and structs. Note that
4906 subsequent types being indexed into can never be pointers, since that
4907 would require loading the pointer before continuing calculation.
4909 The type of each index argument depends on the type it is indexing into.
4910 When indexing into a (optionally packed) structure, only ``i32`` integer
4911 **constants** are allowed (when using a vector of indices they must all
4912 be the **same** ``i32`` integer constant). When indexing into an array,
4913 pointer or vector, integers of any width are allowed, and they are not
4914 required to be constant. These integers are treated as signed values
4917 For example, let's consider a C code fragment and how it gets compiled
4933 int *foo(struct ST *s) {
4934 return &s[1].Z.B[5][13];
4937 The LLVM code generated by Clang is:
4939 .. code-block:: llvm
4941 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
4942 %struct.ST = type { i32, double, %struct.RT }
4944 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
4946 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
4953 In the example above, the first index is indexing into the
4954 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
4955 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
4956 indexes into the third element of the structure, yielding a
4957 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
4958 structure. The third index indexes into the second element of the
4959 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
4960 dimensions of the array are subscripted into, yielding an '``i32``'
4961 type. The '``getelementptr``' instruction returns a pointer to this
4962 element, thus computing a value of '``i32*``' type.
4964 Note that it is perfectly legal to index partially through a structure,
4965 returning a pointer to an inner element. Because of this, the LLVM code
4966 for the given testcase is equivalent to:
4968 .. code-block:: llvm
4970 define i32* @foo(%struct.ST* %s) {
4971 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
4972 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
4973 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
4974 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
4975 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
4979 If the ``inbounds`` keyword is present, the result value of the
4980 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
4981 pointer is not an *in bounds* address of an allocated object, or if any
4982 of the addresses that would be formed by successive addition of the
4983 offsets implied by the indices to the base address with infinitely
4984 precise signed arithmetic are not an *in bounds* address of that
4985 allocated object. The *in bounds* addresses for an allocated object are
4986 all the addresses that point into the object, plus the address one byte
4987 past the end. In cases where the base is a vector of pointers the
4988 ``inbounds`` keyword applies to each of the computations element-wise.
4990 If the ``inbounds`` keyword is not present, the offsets are added to the
4991 base address with silently-wrapping two's complement arithmetic. If the
4992 offsets have a different width from the pointer, they are sign-extended
4993 or truncated to the width of the pointer. The result value of the
4994 ``getelementptr`` may be outside the object pointed to by the base
4995 pointer. The result value may not necessarily be used to access memory
4996 though, even if it happens to point into allocated storage. See the
4997 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5000 The getelementptr instruction is often confusing. For some more insight
5001 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5006 .. code-block:: llvm
5008 ; yields [12 x i8]*:aptr
5009 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5011 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5013 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5015 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5017 In cases where the pointer argument is a vector of pointers, each index
5018 must be a vector with the same number of elements. For example:
5020 .. code-block:: llvm
5022 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5024 Conversion Operations
5025 ---------------------
5027 The instructions in this category are the conversion instructions
5028 (casting) which all take a single operand and a type. They perform
5029 various bit conversions on the operand.
5031 '``trunc .. to``' Instruction
5032 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5039 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5044 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5049 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5050 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5051 of the same number of integers. The bit size of the ``value`` must be
5052 larger than the bit size of the destination type, ``ty2``. Equal sized
5053 types are not allowed.
5058 The '``trunc``' instruction truncates the high order bits in ``value``
5059 and converts the remaining bits to ``ty2``. Since the source size must
5060 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5061 It will always truncate bits.
5066 .. code-block:: llvm
5068 %X = trunc i32 257 to i8 ; yields i8:1
5069 %Y = trunc i32 123 to i1 ; yields i1:true
5070 %Z = trunc i32 122 to i1 ; yields i1:false
5071 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5073 '``zext .. to``' Instruction
5074 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5081 <result> = zext <ty> <value> to <ty2> ; yields ty2
5086 The '``zext``' instruction zero extends its operand to type ``ty2``.
5091 The '``zext``' instruction takes a value to cast, and a type to cast it
5092 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5093 the same number of integers. The bit size of the ``value`` must be
5094 smaller than the bit size of the destination type, ``ty2``.
5099 The ``zext`` fills the high order bits of the ``value`` with zero bits
5100 until it reaches the size of the destination type, ``ty2``.
5102 When zero extending from i1, the result will always be either 0 or 1.
5107 .. code-block:: llvm
5109 %X = zext i32 257 to i64 ; yields i64:257
5110 %Y = zext i1 true to i32 ; yields i32:1
5111 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5113 '``sext .. to``' Instruction
5114 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5121 <result> = sext <ty> <value> to <ty2> ; yields ty2
5126 The '``sext``' sign extends ``value`` to the type ``ty2``.
5131 The '``sext``' instruction takes a value to cast, and a type to cast it
5132 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5133 the same number of integers. The bit size of the ``value`` must be
5134 smaller than the bit size of the destination type, ``ty2``.
5139 The '``sext``' instruction performs a sign extension by copying the sign
5140 bit (highest order bit) of the ``value`` until it reaches the bit size
5141 of the type ``ty2``.
5143 When sign extending from i1, the extension always results in -1 or 0.
5148 .. code-block:: llvm
5150 %X = sext i8 -1 to i16 ; yields i16 :65535
5151 %Y = sext i1 true to i32 ; yields i32:-1
5152 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5154 '``fptrunc .. to``' Instruction
5155 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5162 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5167 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5172 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5173 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5174 The size of ``value`` must be larger than the size of ``ty2``. This
5175 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5180 The '``fptrunc``' instruction truncates a ``value`` from a larger
5181 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5182 point <t_floating>` type. If the value cannot fit within the
5183 destination type, ``ty2``, then the results are undefined.
5188 .. code-block:: llvm
5190 %X = fptrunc double 123.0 to float ; yields float:123.0
5191 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5193 '``fpext .. to``' Instruction
5194 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5201 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5206 The '``fpext``' extends a floating point ``value`` to a larger floating
5212 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5213 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5214 to. The source type must be smaller than the destination type.
5219 The '``fpext``' instruction extends the ``value`` from a smaller
5220 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5221 point <t_floating>` type. The ``fpext`` cannot be used to make a
5222 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5223 *no-op cast* for a floating point cast.
5228 .. code-block:: llvm
5230 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5231 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5233 '``fptoui .. to``' Instruction
5234 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5241 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5246 The '``fptoui``' converts a floating point ``value`` to its unsigned
5247 integer equivalent of type ``ty2``.
5252 The '``fptoui``' instruction takes a value to cast, which must be a
5253 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5254 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5255 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5256 type with the same number of elements as ``ty``
5261 The '``fptoui``' instruction converts its :ref:`floating
5262 point <t_floating>` operand into the nearest (rounding towards zero)
5263 unsigned integer value. If the value cannot fit in ``ty2``, the results
5269 .. code-block:: llvm
5271 %X = fptoui double 123.0 to i32 ; yields i32:123
5272 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5273 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5275 '``fptosi .. to``' Instruction
5276 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5283 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5288 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5289 ``value`` to type ``ty2``.
5294 The '``fptosi``' instruction takes a value to cast, which must be a
5295 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5296 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5297 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5298 type with the same number of elements as ``ty``
5303 The '``fptosi``' instruction converts its :ref:`floating
5304 point <t_floating>` operand into the nearest (rounding towards zero)
5305 signed integer value. If the value cannot fit in ``ty2``, the results
5311 .. code-block:: llvm
5313 %X = fptosi double -123.0 to i32 ; yields i32:-123
5314 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5315 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5317 '``uitofp .. to``' Instruction
5318 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5325 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5330 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5331 and converts that value to the ``ty2`` type.
5336 The '``uitofp``' instruction takes a value to cast, which must be a
5337 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5338 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5339 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5340 type with the same number of elements as ``ty``
5345 The '``uitofp``' instruction interprets its operand as an unsigned
5346 integer quantity and converts it to the corresponding floating point
5347 value. If the value cannot fit in the floating point value, the results
5353 .. code-block:: llvm
5355 %X = uitofp i32 257 to float ; yields float:257.0
5356 %Y = uitofp i8 -1 to double ; yields double:255.0
5358 '``sitofp .. to``' Instruction
5359 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5366 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5371 The '``sitofp``' instruction regards ``value`` as a signed integer and
5372 converts that value to the ``ty2`` type.
5377 The '``sitofp``' instruction takes a value to cast, which must be a
5378 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5379 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5380 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5381 type with the same number of elements as ``ty``
5386 The '``sitofp``' instruction interprets its operand as a signed integer
5387 quantity and converts it to the corresponding floating point value. If
5388 the value cannot fit in the floating point value, the results are
5394 .. code-block:: llvm
5396 %X = sitofp i32 257 to float ; yields float:257.0
5397 %Y = sitofp i8 -1 to double ; yields double:-1.0
5401 '``ptrtoint .. to``' Instruction
5402 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5409 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5414 The '``ptrtoint``' instruction converts the pointer or a vector of
5415 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5420 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5421 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5422 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5423 a vector of integers type.
5428 The '``ptrtoint``' instruction converts ``value`` to integer type
5429 ``ty2`` by interpreting the pointer value as an integer and either
5430 truncating or zero extending that value to the size of the integer type.
5431 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5432 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5433 the same size, then nothing is done (*no-op cast*) other than a type
5439 .. code-block:: llvm
5441 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5442 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5443 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5447 '``inttoptr .. to``' Instruction
5448 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5455 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5460 The '``inttoptr``' instruction converts an integer ``value`` to a
5461 pointer type, ``ty2``.
5466 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5467 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5473 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5474 applying either a zero extension or a truncation depending on the size
5475 of the integer ``value``. If ``value`` is larger than the size of a
5476 pointer then a truncation is done. If ``value`` is smaller than the size
5477 of a pointer then a zero extension is done. If they are the same size,
5478 nothing is done (*no-op cast*).
5483 .. code-block:: llvm
5485 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5486 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5487 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5488 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5492 '``bitcast .. to``' Instruction
5493 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5500 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5505 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5511 The '``bitcast``' instruction takes a value to cast, which must be a
5512 non-aggregate first class value, and a type to cast it to, which must
5513 also be a non-aggregate :ref:`first class <t_firstclass>` type. The bit
5514 sizes of ``value`` and the destination type, ``ty2``, must be identical.
5515 If the source type is a pointer, the destination type must also be a
5516 pointer. This instruction supports bitwise conversion of vectors to
5517 integers and to vectors of other types (as long as they have the same
5523 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It is
5524 always a *no-op cast* because no bits change with this conversion. The
5525 conversion is done as if the ``value`` had been stored to memory and
5526 read back as type ``ty2``. Pointer (or vector of pointers) types may
5527 only be converted to other pointer (or vector of pointers) types with
5528 this instruction. To convert pointers to other types, use the
5529 :ref:`inttoptr <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions
5535 .. code-block:: llvm
5537 %X = bitcast i8 255 to i8 ; yields i8 :-1
5538 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5539 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5540 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5547 The instructions in this category are the "miscellaneous" instructions,
5548 which defy better classification.
5552 '``icmp``' Instruction
5553 ^^^^^^^^^^^^^^^^^^^^^^
5560 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5565 The '``icmp``' instruction returns a boolean value or a vector of
5566 boolean values based on comparison of its two integer, integer vector,
5567 pointer, or pointer vector operands.
5572 The '``icmp``' instruction takes three operands. The first operand is
5573 the condition code indicating the kind of comparison to perform. It is
5574 not a value, just a keyword. The possible condition code are:
5577 #. ``ne``: not equal
5578 #. ``ugt``: unsigned greater than
5579 #. ``uge``: unsigned greater or equal
5580 #. ``ult``: unsigned less than
5581 #. ``ule``: unsigned less or equal
5582 #. ``sgt``: signed greater than
5583 #. ``sge``: signed greater or equal
5584 #. ``slt``: signed less than
5585 #. ``sle``: signed less or equal
5587 The remaining two arguments must be :ref:`integer <t_integer>` or
5588 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5589 must also be identical types.
5594 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5595 code given as ``cond``. The comparison performed always yields either an
5596 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5598 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5599 otherwise. No sign interpretation is necessary or performed.
5600 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5601 otherwise. No sign interpretation is necessary or performed.
5602 #. ``ugt``: interprets the operands as unsigned values and yields
5603 ``true`` if ``op1`` is greater than ``op2``.
5604 #. ``uge``: interprets the operands as unsigned values and yields
5605 ``true`` if ``op1`` is greater than or equal to ``op2``.
5606 #. ``ult``: interprets the operands as unsigned values and yields
5607 ``true`` if ``op1`` is less than ``op2``.
5608 #. ``ule``: interprets the operands as unsigned values and yields
5609 ``true`` if ``op1`` is less than or equal to ``op2``.
5610 #. ``sgt``: interprets the operands as signed values and yields ``true``
5611 if ``op1`` is greater than ``op2``.
5612 #. ``sge``: interprets the operands as signed values and yields ``true``
5613 if ``op1`` is greater than or equal to ``op2``.
5614 #. ``slt``: interprets the operands as signed values and yields ``true``
5615 if ``op1`` is less than ``op2``.
5616 #. ``sle``: interprets the operands as signed values and yields ``true``
5617 if ``op1`` is less than or equal to ``op2``.
5619 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5620 are compared as if they were integers.
5622 If the operands are integer vectors, then they are compared element by
5623 element. The result is an ``i1`` vector with the same number of elements
5624 as the values being compared. Otherwise, the result is an ``i1``.
5629 .. code-block:: llvm
5631 <result> = icmp eq i32 4, 5 ; yields: result=false
5632 <result> = icmp ne float* %X, %X ; yields: result=false
5633 <result> = icmp ult i16 4, 5 ; yields: result=true
5634 <result> = icmp sgt i16 4, 5 ; yields: result=false
5635 <result> = icmp ule i16 -4, 5 ; yields: result=false
5636 <result> = icmp sge i16 4, 5 ; yields: result=false
5638 Note that the code generator does not yet support vector types with the
5639 ``icmp`` instruction.
5643 '``fcmp``' Instruction
5644 ^^^^^^^^^^^^^^^^^^^^^^
5651 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5656 The '``fcmp``' instruction returns a boolean value or vector of boolean
5657 values based on comparison of its operands.
5659 If the operands are floating point scalars, then the result type is a
5660 boolean (:ref:`i1 <t_integer>`).
5662 If the operands are floating point vectors, then the result type is a
5663 vector of boolean with the same number of elements as the operands being
5669 The '``fcmp``' instruction takes three operands. The first operand is
5670 the condition code indicating the kind of comparison to perform. It is
5671 not a value, just a keyword. The possible condition code are:
5673 #. ``false``: no comparison, always returns false
5674 #. ``oeq``: ordered and equal
5675 #. ``ogt``: ordered and greater than
5676 #. ``oge``: ordered and greater than or equal
5677 #. ``olt``: ordered and less than
5678 #. ``ole``: ordered and less than or equal
5679 #. ``one``: ordered and not equal
5680 #. ``ord``: ordered (no nans)
5681 #. ``ueq``: unordered or equal
5682 #. ``ugt``: unordered or greater than
5683 #. ``uge``: unordered or greater than or equal
5684 #. ``ult``: unordered or less than
5685 #. ``ule``: unordered or less than or equal
5686 #. ``une``: unordered or not equal
5687 #. ``uno``: unordered (either nans)
5688 #. ``true``: no comparison, always returns true
5690 *Ordered* means that neither operand is a QNAN while *unordered* means
5691 that either operand may be a QNAN.
5693 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5694 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5695 type. They must have identical types.
5700 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5701 condition code given as ``cond``. If the operands are vectors, then the
5702 vectors are compared element by element. Each comparison performed
5703 always yields an :ref:`i1 <t_integer>` result, as follows:
5705 #. ``false``: always yields ``false``, regardless of operands.
5706 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5707 is equal to ``op2``.
5708 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5709 is greater than ``op2``.
5710 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5711 is greater than or equal to ``op2``.
5712 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5713 is less than ``op2``.
5714 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5715 is less than or equal to ``op2``.
5716 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5717 is not equal to ``op2``.
5718 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5719 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5721 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5722 greater than ``op2``.
5723 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5724 greater than or equal to ``op2``.
5725 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5727 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5728 less than or equal to ``op2``.
5729 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5730 not equal to ``op2``.
5731 #. ``uno``: yields ``true`` if either operand is a QNAN.
5732 #. ``true``: always yields ``true``, regardless of operands.
5737 .. code-block:: llvm
5739 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5740 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5741 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5742 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5744 Note that the code generator does not yet support vector types with the
5745 ``fcmp`` instruction.
5749 '``phi``' Instruction
5750 ^^^^^^^^^^^^^^^^^^^^^
5757 <result> = phi <ty> [ <val0>, <label0>], ...
5762 The '``phi``' instruction is used to implement the φ node in the SSA
5763 graph representing the function.
5768 The type of the incoming values is specified with the first type field.
5769 After this, the '``phi``' instruction takes a list of pairs as
5770 arguments, with one pair for each predecessor basic block of the current
5771 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5772 the value arguments to the PHI node. Only labels may be used as the
5775 There must be no non-phi instructions between the start of a basic block
5776 and the PHI instructions: i.e. PHI instructions must be first in a basic
5779 For the purposes of the SSA form, the use of each incoming value is
5780 deemed to occur on the edge from the corresponding predecessor block to
5781 the current block (but after any definition of an '``invoke``'
5782 instruction's return value on the same edge).
5787 At runtime, the '``phi``' instruction logically takes on the value
5788 specified by the pair corresponding to the predecessor basic block that
5789 executed just prior to the current block.
5794 .. code-block:: llvm
5796 Loop: ; Infinite loop that counts from 0 on up...
5797 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5798 %nextindvar = add i32 %indvar, 1
5803 '``select``' Instruction
5804 ^^^^^^^^^^^^^^^^^^^^^^^^
5811 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5813 selty is either i1 or {<N x i1>}
5818 The '``select``' instruction is used to choose one value based on a
5819 condition, without branching.
5824 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5825 values indicating the condition, and two values of the same :ref:`first
5826 class <t_firstclass>` type. If the val1/val2 are vectors and the
5827 condition is a scalar, then entire vectors are selected, not individual
5833 If the condition is an i1 and it evaluates to 1, the instruction returns
5834 the first value argument; otherwise, it returns the second value
5837 If the condition is a vector of i1, then the value arguments must be
5838 vectors of the same size, and the selection is done element by element.
5843 .. code-block:: llvm
5845 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5849 '``call``' Instruction
5850 ^^^^^^^^^^^^^^^^^^^^^^
5857 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5862 The '``call``' instruction represents a simple function call.
5867 This instruction requires several arguments:
5869 #. The optional "tail" marker indicates that the callee function does
5870 not access any allocas or varargs in the caller. Note that calls may
5871 be marked "tail" even if they do not occur before a
5872 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5873 function call is eligible for tail call optimization, but `might not
5874 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5875 The code generator may optimize calls marked "tail" with either 1)
5876 automatic `sibling call
5877 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5878 callee have matching signatures, or 2) forced tail call optimization
5879 when the following extra requirements are met:
5881 - Caller and callee both have the calling convention ``fastcc``.
5882 - The call is in tail position (ret immediately follows call and ret
5883 uses value of call or is void).
5884 - Option ``-tailcallopt`` is enabled, or
5885 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5886 - `Platform specific constraints are
5887 met. <CodeGenerator.html#tailcallopt>`_
5889 #. The optional "cconv" marker indicates which :ref:`calling
5890 convention <callingconv>` the call should use. If none is
5891 specified, the call defaults to using C calling conventions. The
5892 calling convention of the call must match the calling convention of
5893 the target function, or else the behavior is undefined.
5894 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5895 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5897 #. '``ty``': the type of the call instruction itself which is also the
5898 type of the return value. Functions that return no value are marked
5900 #. '``fnty``': shall be the signature of the pointer to function value
5901 being invoked. The argument types must match the types implied by
5902 this signature. This type can be omitted if the function is not
5903 varargs and if the function type does not return a pointer to a
5905 #. '``fnptrval``': An LLVM value containing a pointer to a function to
5906 be invoked. In most cases, this is a direct function invocation, but
5907 indirect ``call``'s are just as possible, calling an arbitrary pointer
5909 #. '``function args``': argument list whose types match the function
5910 signature argument types and parameter attributes. All arguments must
5911 be of :ref:`first class <t_firstclass>` type. If the function signature
5912 indicates the function accepts a variable number of arguments, the
5913 extra arguments can be specified.
5914 #. The optional :ref:`function attributes <fnattrs>` list. Only
5915 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5916 attributes are valid here.
5921 The '``call``' instruction is used to cause control flow to transfer to
5922 a specified function, with its incoming arguments bound to the specified
5923 values. Upon a '``ret``' instruction in the called function, control
5924 flow continues with the instruction after the function call, and the
5925 return value of the function is bound to the result argument.
5930 .. code-block:: llvm
5932 %retval = call i32 @test(i32 %argc)
5933 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
5934 %X = tail call i32 @foo() ; yields i32
5935 %Y = tail call fastcc i32 @foo() ; yields i32
5936 call void %foo(i8 97 signext)
5938 %struct.A = type { i32, i8 }
5939 %r = call %struct.A @foo() ; yields { 32, i8 }
5940 %gr = extractvalue %struct.A %r, 0 ; yields i32
5941 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
5942 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
5943 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
5945 llvm treats calls to some functions with names and arguments that match
5946 the standard C99 library as being the C99 library functions, and may
5947 perform optimizations or generate code for them under that assumption.
5948 This is something we'd like to change in the future to provide better
5949 support for freestanding environments and non-C-based languages.
5953 '``va_arg``' Instruction
5954 ^^^^^^^^^^^^^^^^^^^^^^^^
5961 <resultval> = va_arg <va_list*> <arglist>, <argty>
5966 The '``va_arg``' instruction is used to access arguments passed through
5967 the "variable argument" area of a function call. It is used to implement
5968 the ``va_arg`` macro in C.
5973 This instruction takes a ``va_list*`` value and the type of the
5974 argument. It returns a value of the specified argument type and
5975 increments the ``va_list`` to point to the next argument. The actual
5976 type of ``va_list`` is target specific.
5981 The '``va_arg``' instruction loads an argument of the specified type
5982 from the specified ``va_list`` and causes the ``va_list`` to point to
5983 the next argument. For more information, see the variable argument
5984 handling :ref:`Intrinsic Functions <int_varargs>`.
5986 It is legal for this instruction to be called in a function which does
5987 not take a variable number of arguments, for example, the ``vfprintf``
5990 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
5991 function <intrinsics>` because it takes a type as an argument.
5996 See the :ref:`variable argument processing <int_varargs>` section.
5998 Note that the code generator does not yet fully support va\_arg on many
5999 targets. Also, it does not currently support va\_arg with aggregate
6000 types on any target.
6004 '``landingpad``' Instruction
6005 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6012 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6013 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6015 <clause> := catch <type> <value>
6016 <clause> := filter <array constant type> <array constant>
6021 The '``landingpad``' instruction is used by `LLVM's exception handling
6022 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6023 is a landing pad --- one where the exception lands, and corresponds to the
6024 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6025 defines values supplied by the personality function (``pers_fn``) upon
6026 re-entry to the function. The ``resultval`` has the type ``resultty``.
6031 This instruction takes a ``pers_fn`` value. This is the personality
6032 function associated with the unwinding mechanism. The optional
6033 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6035 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6036 contains the global variable representing the "type" that may be caught
6037 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6038 clause takes an array constant as its argument. Use
6039 "``[0 x i8**] undef``" for a filter which cannot throw. The
6040 '``landingpad``' instruction must contain *at least* one ``clause`` or
6041 the ``cleanup`` flag.
6046 The '``landingpad``' instruction defines the values which are set by the
6047 personality function (``pers_fn``) upon re-entry to the function, and
6048 therefore the "result type" of the ``landingpad`` instruction. As with
6049 calling conventions, how the personality function results are
6050 represented in LLVM IR is target specific.
6052 The clauses are applied in order from top to bottom. If two
6053 ``landingpad`` instructions are merged together through inlining, the
6054 clauses from the calling function are appended to the list of clauses.
6055 When the call stack is being unwound due to an exception being thrown,
6056 the exception is compared against each ``clause`` in turn. If it doesn't
6057 match any of the clauses, and the ``cleanup`` flag is not set, then
6058 unwinding continues further up the call stack.
6060 The ``landingpad`` instruction has several restrictions:
6062 - A landing pad block is a basic block which is the unwind destination
6063 of an '``invoke``' instruction.
6064 - A landing pad block must have a '``landingpad``' instruction as its
6065 first non-PHI instruction.
6066 - There can be only one '``landingpad``' instruction within the landing
6068 - A basic block that is not a landing pad block may not include a
6069 '``landingpad``' instruction.
6070 - All '``landingpad``' instructions in a function must have the same
6071 personality function.
6076 .. code-block:: llvm
6078 ;; A landing pad which can catch an integer.
6079 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6081 ;; A landing pad that is a cleanup.
6082 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6084 ;; A landing pad which can catch an integer and can only throw a double.
6085 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6087 filter [1 x i8**] [@_ZTId]
6094 LLVM supports the notion of an "intrinsic function". These functions
6095 have well known names and semantics and are required to follow certain
6096 restrictions. Overall, these intrinsics represent an extension mechanism
6097 for the LLVM language that does not require changing all of the
6098 transformations in LLVM when adding to the language (or the bitcode
6099 reader/writer, the parser, etc...).
6101 Intrinsic function names must all start with an "``llvm.``" prefix. This
6102 prefix is reserved in LLVM for intrinsic names; thus, function names may
6103 not begin with this prefix. Intrinsic functions must always be external
6104 functions: you cannot define the body of intrinsic functions. Intrinsic
6105 functions may only be used in call or invoke instructions: it is illegal
6106 to take the address of an intrinsic function. Additionally, because
6107 intrinsic functions are part of the LLVM language, it is required if any
6108 are added that they be documented here.
6110 Some intrinsic functions can be overloaded, i.e., the intrinsic
6111 represents a family of functions that perform the same operation but on
6112 different data types. Because LLVM can represent over 8 million
6113 different integer types, overloading is used commonly to allow an
6114 intrinsic function to operate on any integer type. One or more of the
6115 argument types or the result type can be overloaded to accept any
6116 integer type. Argument types may also be defined as exactly matching a
6117 previous argument's type or the result type. This allows an intrinsic
6118 function which accepts multiple arguments, but needs all of them to be
6119 of the same type, to only be overloaded with respect to a single
6120 argument or the result.
6122 Overloaded intrinsics will have the names of its overloaded argument
6123 types encoded into its function name, each preceded by a period. Only
6124 those types which are overloaded result in a name suffix. Arguments
6125 whose type is matched against another type do not. For example, the
6126 ``llvm.ctpop`` function can take an integer of any width and returns an
6127 integer of exactly the same integer width. This leads to a family of
6128 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6129 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6130 overloaded, and only one type suffix is required. Because the argument's
6131 type is matched against the return type, it does not require its own
6134 To learn how to add an intrinsic function, please see the `Extending
6135 LLVM Guide <ExtendingLLVM.html>`_.
6139 Variable Argument Handling Intrinsics
6140 -------------------------------------
6142 Variable argument support is defined in LLVM with the
6143 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6144 functions. These functions are related to the similarly named macros
6145 defined in the ``<stdarg.h>`` header file.
6147 All of these functions operate on arguments that use a target-specific
6148 value type "``va_list``". The LLVM assembly language reference manual
6149 does not define what this type is, so all transformations should be
6150 prepared to handle these functions regardless of the type used.
6152 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6153 variable argument handling intrinsic functions are used.
6155 .. code-block:: llvm
6157 define i32 @test(i32 %X, ...) {
6158 ; Initialize variable argument processing
6160 %ap2 = bitcast i8** %ap to i8*
6161 call void @llvm.va_start(i8* %ap2)
6163 ; Read a single integer argument
6164 %tmp = va_arg i8** %ap, i32
6166 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6168 %aq2 = bitcast i8** %aq to i8*
6169 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6170 call void @llvm.va_end(i8* %aq2)
6172 ; Stop processing of arguments.
6173 call void @llvm.va_end(i8* %ap2)
6177 declare void @llvm.va_start(i8*)
6178 declare void @llvm.va_copy(i8*, i8*)
6179 declare void @llvm.va_end(i8*)
6183 '``llvm.va_start``' Intrinsic
6184 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6191 declare void %llvm.va_start(i8* <arglist>)
6196 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6197 subsequent use by ``va_arg``.
6202 The argument is a pointer to a ``va_list`` element to initialize.
6207 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6208 available in C. In a target-dependent way, it initializes the
6209 ``va_list`` element to which the argument points, so that the next call
6210 to ``va_arg`` will produce the first variable argument passed to the
6211 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6212 to know the last argument of the function as the compiler can figure
6215 '``llvm.va_end``' Intrinsic
6216 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6223 declare void @llvm.va_end(i8* <arglist>)
6228 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6229 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6234 The argument is a pointer to a ``va_list`` to destroy.
6239 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6240 available in C. In a target-dependent way, it destroys the ``va_list``
6241 element to which the argument points. Calls to
6242 :ref:`llvm.va_start <int_va_start>` and
6243 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6248 '``llvm.va_copy``' Intrinsic
6249 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6256 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6261 The '``llvm.va_copy``' intrinsic copies the current argument position
6262 from the source argument list to the destination argument list.
6267 The first argument is a pointer to a ``va_list`` element to initialize.
6268 The second argument is a pointer to a ``va_list`` element to copy from.
6273 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6274 available in C. In a target-dependent way, it copies the source
6275 ``va_list`` element into the destination ``va_list`` element. This
6276 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6277 arbitrarily complex and require, for example, memory allocation.
6279 Accurate Garbage Collection Intrinsics
6280 --------------------------------------
6282 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6283 (GC) requires the implementation and generation of these intrinsics.
6284 These intrinsics allow identification of :ref:`GC roots on the
6285 stack <int_gcroot>`, as well as garbage collector implementations that
6286 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6287 Front-ends for type-safe garbage collected languages should generate
6288 these intrinsics to make use of the LLVM garbage collectors. For more
6289 details, see `Accurate Garbage Collection with
6290 LLVM <GarbageCollection.html>`_.
6292 The garbage collection intrinsics only operate on objects in the generic
6293 address space (address space zero).
6297 '``llvm.gcroot``' Intrinsic
6298 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6305 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6310 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6311 the code generator, and allows some metadata to be associated with it.
6316 The first argument specifies the address of a stack object that contains
6317 the root pointer. The second pointer (which must be either a constant or
6318 a global value address) contains the meta-data to be associated with the
6324 At runtime, a call to this intrinsic stores a null pointer into the
6325 "ptrloc" location. At compile-time, the code generator generates
6326 information to allow the runtime to find the pointer at GC safe points.
6327 The '``llvm.gcroot``' intrinsic may only be used in a function which
6328 :ref:`specifies a GC algorithm <gc>`.
6332 '``llvm.gcread``' Intrinsic
6333 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6340 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6345 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6346 locations, allowing garbage collector implementations that require read
6352 The second argument is the address to read from, which should be an
6353 address allocated from the garbage collector. The first object is a
6354 pointer to the start of the referenced object, if needed by the language
6355 runtime (otherwise null).
6360 The '``llvm.gcread``' intrinsic has the same semantics as a load
6361 instruction, but may be replaced with substantially more complex code by
6362 the garbage collector runtime, as needed. The '``llvm.gcread``'
6363 intrinsic may only be used in a function which :ref:`specifies a GC
6368 '``llvm.gcwrite``' Intrinsic
6369 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6376 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6381 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6382 locations, allowing garbage collector implementations that require write
6383 barriers (such as generational or reference counting collectors).
6388 The first argument is the reference to store, the second is the start of
6389 the object to store it to, and the third is the address of the field of
6390 Obj to store to. If the runtime does not require a pointer to the
6391 object, Obj may be null.
6396 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6397 instruction, but may be replaced with substantially more complex code by
6398 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6399 intrinsic may only be used in a function which :ref:`specifies a GC
6402 Code Generator Intrinsics
6403 -------------------------
6405 These intrinsics are provided by LLVM to expose special features that
6406 may only be implemented with code generator support.
6408 '``llvm.returnaddress``' Intrinsic
6409 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6416 declare i8 *@llvm.returnaddress(i32 <level>)
6421 The '``llvm.returnaddress``' intrinsic attempts to compute a
6422 target-specific value indicating the return address of the current
6423 function or one of its callers.
6428 The argument to this intrinsic indicates which function to return the
6429 address for. Zero indicates the calling function, one indicates its
6430 caller, etc. The argument is **required** to be a constant integer
6436 The '``llvm.returnaddress``' intrinsic either returns a pointer
6437 indicating the return address of the specified call frame, or zero if it
6438 cannot be identified. The value returned by this intrinsic is likely to
6439 be incorrect or 0 for arguments other than zero, so it should only be
6440 used for debugging purposes.
6442 Note that calling this intrinsic does not prevent function inlining or
6443 other aggressive transformations, so the value returned may not be that
6444 of the obvious source-language caller.
6446 '``llvm.frameaddress``' Intrinsic
6447 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6454 declare i8* @llvm.frameaddress(i32 <level>)
6459 The '``llvm.frameaddress``' intrinsic attempts to return the
6460 target-specific frame pointer value for the specified stack frame.
6465 The argument to this intrinsic indicates which function to return the
6466 frame pointer for. Zero indicates the calling function, one indicates
6467 its caller, etc. The argument is **required** to be a constant integer
6473 The '``llvm.frameaddress``' intrinsic either returns a pointer
6474 indicating the frame address of the specified call frame, or zero if it
6475 cannot be identified. The value returned by this intrinsic is likely to
6476 be incorrect or 0 for arguments other than zero, so it should only be
6477 used for debugging purposes.
6479 Note that calling this intrinsic does not prevent function inlining or
6480 other aggressive transformations, so the value returned may not be that
6481 of the obvious source-language caller.
6485 '``llvm.stacksave``' Intrinsic
6486 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6493 declare i8* @llvm.stacksave()
6498 The '``llvm.stacksave``' intrinsic is used to remember the current state
6499 of the function stack, for use with
6500 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6501 implementing language features like scoped automatic variable sized
6507 This intrinsic returns a opaque pointer value that can be passed to
6508 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6509 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6510 ``llvm.stacksave``, it effectively restores the state of the stack to
6511 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6512 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6513 were allocated after the ``llvm.stacksave`` was executed.
6515 .. _int_stackrestore:
6517 '``llvm.stackrestore``' Intrinsic
6518 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6525 declare void @llvm.stackrestore(i8* %ptr)
6530 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6531 the function stack to the state it was in when the corresponding
6532 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6533 useful for implementing language features like scoped automatic variable
6534 sized arrays in C99.
6539 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6541 '``llvm.prefetch``' Intrinsic
6542 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6549 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6554 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6555 insert a prefetch instruction if supported; otherwise, it is a noop.
6556 Prefetches have no effect on the behavior of the program but can change
6557 its performance characteristics.
6562 ``address`` is the address to be prefetched, ``rw`` is the specifier
6563 determining if the fetch should be for a read (0) or write (1), and
6564 ``locality`` is a temporal locality specifier ranging from (0) - no
6565 locality, to (3) - extremely local keep in cache. The ``cache type``
6566 specifies whether the prefetch is performed on the data (1) or
6567 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6568 arguments must be constant integers.
6573 This intrinsic does not modify the behavior of the program. In
6574 particular, prefetches cannot trap and do not produce a value. On
6575 targets that support this intrinsic, the prefetch can provide hints to
6576 the processor cache for better performance.
6578 '``llvm.pcmarker``' Intrinsic
6579 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6586 declare void @llvm.pcmarker(i32 <id>)
6591 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6592 Counter (PC) in a region of code to simulators and other tools. The
6593 method is target specific, but it is expected that the marker will use
6594 exported symbols to transmit the PC of the marker. The marker makes no
6595 guarantees that it will remain with any specific instruction after
6596 optimizations. It is possible that the presence of a marker will inhibit
6597 optimizations. The intended use is to be inserted after optimizations to
6598 allow correlations of simulation runs.
6603 ``id`` is a numerical id identifying the marker.
6608 This intrinsic does not modify the behavior of the program. Backends
6609 that do not support this intrinsic may ignore it.
6611 '``llvm.readcyclecounter``' Intrinsic
6612 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6619 declare i64 @llvm.readcyclecounter()
6624 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6625 counter register (or similar low latency, high accuracy clocks) on those
6626 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6627 should map to RPCC. As the backing counters overflow quickly (on the
6628 order of 9 seconds on alpha), this should only be used for small
6634 When directly supported, reading the cycle counter should not modify any
6635 memory. Implementations are allowed to either return a application
6636 specific value or a system wide value. On backends without support, this
6637 is lowered to a constant 0.
6639 Standard C Library Intrinsics
6640 -----------------------------
6642 LLVM provides intrinsics for a few important standard C library
6643 functions. These intrinsics allow source-language front-ends to pass
6644 information about the alignment of the pointer arguments to the code
6645 generator, providing opportunity for more efficient code generation.
6649 '``llvm.memcpy``' Intrinsic
6650 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6655 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6656 integer bit width and for different address spaces. Not all targets
6657 support all bit widths however.
6661 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6662 i32 <len>, i32 <align>, i1 <isvolatile>)
6663 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6664 i64 <len>, i32 <align>, i1 <isvolatile>)
6669 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6670 source location to the destination location.
6672 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6673 intrinsics do not return a value, takes extra alignment/isvolatile
6674 arguments and the pointers can be in specified address spaces.
6679 The first argument is a pointer to the destination, the second is a
6680 pointer to the source. The third argument is an integer argument
6681 specifying the number of bytes to copy, the fourth argument is the
6682 alignment of the source and destination locations, and the fifth is a
6683 boolean indicating a volatile access.
6685 If the call to this intrinsic has an alignment value that is not 0 or 1,
6686 then the caller guarantees that both the source and destination pointers
6687 are aligned to that boundary.
6689 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6690 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6691 very cleanly specified and it is unwise to depend on it.
6696 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6697 source location to the destination location, which are not allowed to
6698 overlap. It copies "len" bytes of memory over. If the argument is known
6699 to be aligned to some boundary, this can be specified as the fourth
6700 argument, otherwise it should be set to 0 or 1.
6702 '``llvm.memmove``' Intrinsic
6703 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6708 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6709 bit width and for different address space. Not all targets support all
6714 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6715 i32 <len>, i32 <align>, i1 <isvolatile>)
6716 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6717 i64 <len>, i32 <align>, i1 <isvolatile>)
6722 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6723 source location to the destination location. It is similar to the
6724 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6727 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6728 intrinsics do not return a value, takes extra alignment/isvolatile
6729 arguments and the pointers can be in specified address spaces.
6734 The first argument is a pointer to the destination, the second is a
6735 pointer to the source. The third argument is an integer argument
6736 specifying the number of bytes to copy, the fourth argument is the
6737 alignment of the source and destination locations, and the fifth is a
6738 boolean indicating a volatile access.
6740 If the call to this intrinsic has an alignment value that is not 0 or 1,
6741 then the caller guarantees that the source and destination pointers are
6742 aligned to that boundary.
6744 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6745 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6746 not very cleanly specified and it is unwise to depend on it.
6751 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6752 source location to the destination location, which may overlap. It
6753 copies "len" bytes of memory over. If the argument is known to be
6754 aligned to some boundary, this can be specified as the fourth argument,
6755 otherwise it should be set to 0 or 1.
6757 '``llvm.memset.*``' Intrinsics
6758 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6763 This is an overloaded intrinsic. You can use llvm.memset on any integer
6764 bit width and for different address spaces. However, not all targets
6765 support all bit widths.
6769 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6770 i32 <len>, i32 <align>, i1 <isvolatile>)
6771 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6772 i64 <len>, i32 <align>, i1 <isvolatile>)
6777 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6778 particular byte value.
6780 Note that, unlike the standard libc function, the ``llvm.memset``
6781 intrinsic does not return a value and takes extra alignment/volatile
6782 arguments. Also, the destination can be in an arbitrary address space.
6787 The first argument is a pointer to the destination to fill, the second
6788 is the byte value with which to fill it, the third argument is an
6789 integer argument specifying the number of bytes to fill, and the fourth
6790 argument is the known alignment of the destination location.
6792 If the call to this intrinsic has an alignment value that is not 0 or 1,
6793 then the caller guarantees that the destination pointer is aligned to
6796 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6797 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6798 very cleanly specified and it is unwise to depend on it.
6803 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6804 at the destination location. If the argument is known to be aligned to
6805 some boundary, this can be specified as the fourth argument, otherwise
6806 it should be set to 0 or 1.
6808 '``llvm.sqrt.*``' Intrinsic
6809 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6814 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6815 floating point or vector of floating point type. Not all targets support
6820 declare float @llvm.sqrt.f32(float %Val)
6821 declare double @llvm.sqrt.f64(double %Val)
6822 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6823 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6824 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6829 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6830 returning the same value as the libm '``sqrt``' functions would. Unlike
6831 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6832 negative numbers other than -0.0 (which allows for better optimization,
6833 because there is no need to worry about errno being set).
6834 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6839 The argument and return value are floating point numbers of the same
6845 This function returns the sqrt of the specified operand if it is a
6846 nonnegative floating point number.
6848 '``llvm.powi.*``' Intrinsic
6849 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6854 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6855 floating point or vector of floating point type. Not all targets support
6860 declare float @llvm.powi.f32(float %Val, i32 %power)
6861 declare double @llvm.powi.f64(double %Val, i32 %power)
6862 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6863 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6864 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6869 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6870 specified (positive or negative) power. The order of evaluation of
6871 multiplications is not defined. When a vector of floating point type is
6872 used, the second argument remains a scalar integer value.
6877 The second argument is an integer power, and the first is a value to
6878 raise to that power.
6883 This function returns the first value raised to the second power with an
6884 unspecified sequence of rounding operations.
6886 '``llvm.sin.*``' Intrinsic
6887 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6892 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
6893 floating point or vector of floating point type. Not all targets support
6898 declare float @llvm.sin.f32(float %Val)
6899 declare double @llvm.sin.f64(double %Val)
6900 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6901 declare fp128 @llvm.sin.f128(fp128 %Val)
6902 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6907 The '``llvm.sin.*``' intrinsics return the sine of the operand.
6912 The argument and return value are floating point numbers of the same
6918 This function returns the sine of the specified operand, returning the
6919 same values as the libm ``sin`` functions would, and handles error
6920 conditions in the same way.
6922 '``llvm.cos.*``' Intrinsic
6923 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6928 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
6929 floating point or vector of floating point type. Not all targets support
6934 declare float @llvm.cos.f32(float %Val)
6935 declare double @llvm.cos.f64(double %Val)
6936 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6937 declare fp128 @llvm.cos.f128(fp128 %Val)
6938 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6943 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
6948 The argument and return value are floating point numbers of the same
6954 This function returns the cosine of the specified operand, returning the
6955 same values as the libm ``cos`` functions would, and handles error
6956 conditions in the same way.
6958 '``llvm.pow.*``' Intrinsic
6959 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6964 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
6965 floating point or vector of floating point type. Not all targets support
6970 declare float @llvm.pow.f32(float %Val, float %Power)
6971 declare double @llvm.pow.f64(double %Val, double %Power)
6972 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6973 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6974 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6979 The '``llvm.pow.*``' intrinsics return the first operand raised to the
6980 specified (positive or negative) power.
6985 The second argument is a floating point power, and the first is a value
6986 to raise to that power.
6991 This function returns the first value raised to the second power,
6992 returning the same values as the libm ``pow`` functions would, and
6993 handles error conditions in the same way.
6995 '``llvm.exp.*``' Intrinsic
6996 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7001 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7002 floating point or vector of floating point type. Not all targets support
7007 declare float @llvm.exp.f32(float %Val)
7008 declare double @llvm.exp.f64(double %Val)
7009 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7010 declare fp128 @llvm.exp.f128(fp128 %Val)
7011 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7016 The '``llvm.exp.*``' intrinsics perform the exp function.
7021 The argument and return value are floating point numbers of the same
7027 This function returns the same values as the libm ``exp`` functions
7028 would, and handles error conditions in the same way.
7030 '``llvm.exp2.*``' Intrinsic
7031 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7036 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7037 floating point or vector of floating point type. Not all targets support
7042 declare float @llvm.exp2.f32(float %Val)
7043 declare double @llvm.exp2.f64(double %Val)
7044 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7045 declare fp128 @llvm.exp2.f128(fp128 %Val)
7046 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7051 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7056 The argument and return value are floating point numbers of the same
7062 This function returns the same values as the libm ``exp2`` functions
7063 would, and handles error conditions in the same way.
7065 '``llvm.log.*``' Intrinsic
7066 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7071 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7072 floating point or vector of floating point type. Not all targets support
7077 declare float @llvm.log.f32(float %Val)
7078 declare double @llvm.log.f64(double %Val)
7079 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7080 declare fp128 @llvm.log.f128(fp128 %Val)
7081 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7086 The '``llvm.log.*``' intrinsics perform the log function.
7091 The argument and return value are floating point numbers of the same
7097 This function returns the same values as the libm ``log`` functions
7098 would, and handles error conditions in the same way.
7100 '``llvm.log10.*``' Intrinsic
7101 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7106 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7107 floating point or vector of floating point type. Not all targets support
7112 declare float @llvm.log10.f32(float %Val)
7113 declare double @llvm.log10.f64(double %Val)
7114 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7115 declare fp128 @llvm.log10.f128(fp128 %Val)
7116 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7121 The '``llvm.log10.*``' intrinsics perform the log10 function.
7126 The argument and return value are floating point numbers of the same
7132 This function returns the same values as the libm ``log10`` functions
7133 would, and handles error conditions in the same way.
7135 '``llvm.log2.*``' Intrinsic
7136 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7141 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7142 floating point or vector of floating point type. Not all targets support
7147 declare float @llvm.log2.f32(float %Val)
7148 declare double @llvm.log2.f64(double %Val)
7149 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7150 declare fp128 @llvm.log2.f128(fp128 %Val)
7151 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7156 The '``llvm.log2.*``' intrinsics perform the log2 function.
7161 The argument and return value are floating point numbers of the same
7167 This function returns the same values as the libm ``log2`` functions
7168 would, and handles error conditions in the same way.
7170 '``llvm.fma.*``' Intrinsic
7171 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7176 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7177 floating point or vector of floating point type. Not all targets support
7182 declare float @llvm.fma.f32(float %a, float %b, float %c)
7183 declare double @llvm.fma.f64(double %a, double %b, double %c)
7184 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7185 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7186 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7191 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7197 The argument and return value are floating point numbers of the same
7203 This function returns the same values as the libm ``fma`` functions
7206 '``llvm.fabs.*``' Intrinsic
7207 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7212 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7213 floating point or vector of floating point type. Not all targets support
7218 declare float @llvm.fabs.f32(float %Val)
7219 declare double @llvm.fabs.f64(double %Val)
7220 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7221 declare fp128 @llvm.fabs.f128(fp128 %Val)
7222 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7227 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7233 The argument and return value are floating point numbers of the same
7239 This function returns the same values as the libm ``fabs`` functions
7240 would, and handles error conditions in the same way.
7242 '``llvm.floor.*``' Intrinsic
7243 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7248 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7249 floating point or vector of floating point type. Not all targets support
7254 declare float @llvm.floor.f32(float %Val)
7255 declare double @llvm.floor.f64(double %Val)
7256 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7257 declare fp128 @llvm.floor.f128(fp128 %Val)
7258 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7263 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7268 The argument and return value are floating point numbers of the same
7274 This function returns the same values as the libm ``floor`` functions
7275 would, and handles error conditions in the same way.
7277 '``llvm.ceil.*``' Intrinsic
7278 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7283 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7284 floating point or vector of floating point type. Not all targets support
7289 declare float @llvm.ceil.f32(float %Val)
7290 declare double @llvm.ceil.f64(double %Val)
7291 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7292 declare fp128 @llvm.ceil.f128(fp128 %Val)
7293 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7298 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7303 The argument and return value are floating point numbers of the same
7309 This function returns the same values as the libm ``ceil`` functions
7310 would, and handles error conditions in the same way.
7312 '``llvm.trunc.*``' Intrinsic
7313 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7318 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7319 floating point or vector of floating point type. Not all targets support
7324 declare float @llvm.trunc.f32(float %Val)
7325 declare double @llvm.trunc.f64(double %Val)
7326 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7327 declare fp128 @llvm.trunc.f128(fp128 %Val)
7328 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7333 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7334 nearest integer not larger in magnitude than the operand.
7339 The argument and return value are floating point numbers of the same
7345 This function returns the same values as the libm ``trunc`` functions
7346 would, and handles error conditions in the same way.
7348 '``llvm.rint.*``' Intrinsic
7349 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7354 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7355 floating point or vector of floating point type. Not all targets support
7360 declare float @llvm.rint.f32(float %Val)
7361 declare double @llvm.rint.f64(double %Val)
7362 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7363 declare fp128 @llvm.rint.f128(fp128 %Val)
7364 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7369 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7370 nearest integer. It may raise an inexact floating-point exception if the
7371 operand isn't an integer.
7376 The argument and return value are floating point numbers of the same
7382 This function returns the same values as the libm ``rint`` functions
7383 would, and handles error conditions in the same way.
7385 '``llvm.nearbyint.*``' Intrinsic
7386 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7391 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7392 floating point or vector of floating point type. Not all targets support
7397 declare float @llvm.nearbyint.f32(float %Val)
7398 declare double @llvm.nearbyint.f64(double %Val)
7399 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7400 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7401 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7406 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7412 The argument and return value are floating point numbers of the same
7418 This function returns the same values as the libm ``nearbyint``
7419 functions would, and handles error conditions in the same way.
7421 Bit Manipulation Intrinsics
7422 ---------------------------
7424 LLVM provides intrinsics for a few important bit manipulation
7425 operations. These allow efficient code generation for some algorithms.
7427 '``llvm.bswap.*``' Intrinsics
7428 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7433 This is an overloaded intrinsic function. You can use bswap on any
7434 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7438 declare i16 @llvm.bswap.i16(i16 <id>)
7439 declare i32 @llvm.bswap.i32(i32 <id>)
7440 declare i64 @llvm.bswap.i64(i64 <id>)
7445 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7446 values with an even number of bytes (positive multiple of 16 bits).
7447 These are useful for performing operations on data that is not in the
7448 target's native byte order.
7453 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7454 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7455 intrinsic returns an i32 value that has the four bytes of the input i32
7456 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7457 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7458 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7459 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7462 '``llvm.ctpop.*``' Intrinsic
7463 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7468 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7469 bit width, or on any vector with integer elements. Not all targets
7470 support all bit widths or vector types, however.
7474 declare i8 @llvm.ctpop.i8(i8 <src>)
7475 declare i16 @llvm.ctpop.i16(i16 <src>)
7476 declare i32 @llvm.ctpop.i32(i32 <src>)
7477 declare i64 @llvm.ctpop.i64(i64 <src>)
7478 declare i256 @llvm.ctpop.i256(i256 <src>)
7479 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7484 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7490 The only argument is the value to be counted. The argument may be of any
7491 integer type, or a vector with integer elements. The return type must
7492 match the argument type.
7497 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7498 each element of a vector.
7500 '``llvm.ctlz.*``' Intrinsic
7501 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7506 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7507 integer bit width, or any vector whose elements are integers. Not all
7508 targets support all bit widths or vector types, however.
7512 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7513 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7514 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7515 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7516 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7517 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7522 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7523 leading zeros in a variable.
7528 The first argument is the value to be counted. This argument may be of
7529 any integer type, or a vectory with integer element type. The return
7530 type must match the first argument type.
7532 The second argument must be a constant and is a flag to indicate whether
7533 the intrinsic should ensure that a zero as the first argument produces a
7534 defined result. Historically some architectures did not provide a
7535 defined result for zero values as efficiently, and many algorithms are
7536 now predicated on avoiding zero-value inputs.
7541 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7542 zeros in a variable, or within each element of the vector. If
7543 ``src == 0`` then the result is the size in bits of the type of ``src``
7544 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7545 ``llvm.ctlz(i32 2) = 30``.
7547 '``llvm.cttz.*``' Intrinsic
7548 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7553 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7554 integer bit width, or any vector of integer elements. Not all targets
7555 support all bit widths or vector types, however.
7559 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7560 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7561 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7562 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7563 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7564 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7569 The '``llvm.cttz``' family of intrinsic functions counts the number of
7575 The first argument is the value to be counted. This argument may be of
7576 any integer type, or a vectory with integer element type. The return
7577 type must match the first argument type.
7579 The second argument must be a constant and is a flag to indicate whether
7580 the intrinsic should ensure that a zero as the first argument produces a
7581 defined result. Historically some architectures did not provide a
7582 defined result for zero values as efficiently, and many algorithms are
7583 now predicated on avoiding zero-value inputs.
7588 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7589 zeros in a variable, or within each element of a vector. If ``src == 0``
7590 then the result is the size in bits of the type of ``src`` if
7591 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7592 ``llvm.cttz(2) = 1``.
7594 Arithmetic with Overflow Intrinsics
7595 -----------------------------------
7597 LLVM provides intrinsics for some arithmetic with overflow operations.
7599 '``llvm.sadd.with.overflow.*``' Intrinsics
7600 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7605 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7606 on any integer bit width.
7610 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7611 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7612 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7617 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7618 a signed addition of the two arguments, and indicate whether an overflow
7619 occurred during the signed summation.
7624 The arguments (%a and %b) and the first element of the result structure
7625 may be of integer types of any bit width, but they must have the same
7626 bit width. The second element of the result structure must be of type
7627 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7633 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7634 a signed addition of the two variables. They return a structure --- the
7635 first element of which is the signed summation, and the second element
7636 of which is a bit specifying if the signed summation resulted in an
7642 .. code-block:: llvm
7644 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7645 %sum = extractvalue {i32, i1} %res, 0
7646 %obit = extractvalue {i32, i1} %res, 1
7647 br i1 %obit, label %overflow, label %normal
7649 '``llvm.uadd.with.overflow.*``' Intrinsics
7650 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7655 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7656 on any integer bit width.
7660 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7661 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7662 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7667 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7668 an unsigned addition of the two arguments, and indicate whether a carry
7669 occurred during the unsigned summation.
7674 The arguments (%a and %b) and the first element of the result structure
7675 may be of integer types of any bit width, but they must have the same
7676 bit width. The second element of the result structure must be of type
7677 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7683 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7684 an unsigned addition of the two arguments. They return a structure --- the
7685 first element of which is the sum, and the second element of which is a
7686 bit specifying if the unsigned summation resulted in a carry.
7691 .. code-block:: llvm
7693 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7694 %sum = extractvalue {i32, i1} %res, 0
7695 %obit = extractvalue {i32, i1} %res, 1
7696 br i1 %obit, label %carry, label %normal
7698 '``llvm.ssub.with.overflow.*``' Intrinsics
7699 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7704 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7705 on any integer bit width.
7709 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7710 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7711 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7716 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7717 a signed subtraction of the two arguments, and indicate whether an
7718 overflow occurred during the signed subtraction.
7723 The arguments (%a and %b) and the first element of the result structure
7724 may be of integer types of any bit width, but they must have the same
7725 bit width. The second element of the result structure must be of type
7726 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7732 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7733 a signed subtraction of the two arguments. They return a structure --- the
7734 first element of which is the subtraction, and the second element of
7735 which is a bit specifying if the signed subtraction resulted in an
7741 .. code-block:: llvm
7743 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7744 %sum = extractvalue {i32, i1} %res, 0
7745 %obit = extractvalue {i32, i1} %res, 1
7746 br i1 %obit, label %overflow, label %normal
7748 '``llvm.usub.with.overflow.*``' Intrinsics
7749 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7754 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7755 on any integer bit width.
7759 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7760 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7761 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7766 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7767 an unsigned subtraction of the two arguments, and indicate whether an
7768 overflow occurred during the unsigned subtraction.
7773 The arguments (%a and %b) and the first element of the result structure
7774 may be of integer types of any bit width, but they must have the same
7775 bit width. The second element of the result structure must be of type
7776 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7782 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7783 an unsigned subtraction of the two arguments. They return a structure ---
7784 the first element of which is the subtraction, and the second element of
7785 which is a bit specifying if the unsigned subtraction resulted in an
7791 .. code-block:: llvm
7793 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7794 %sum = extractvalue {i32, i1} %res, 0
7795 %obit = extractvalue {i32, i1} %res, 1
7796 br i1 %obit, label %overflow, label %normal
7798 '``llvm.smul.with.overflow.*``' Intrinsics
7799 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7804 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7805 on any integer bit width.
7809 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7810 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7811 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7816 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7817 a signed multiplication of the two arguments, and indicate whether an
7818 overflow occurred during the signed multiplication.
7823 The arguments (%a and %b) and the first element of the result structure
7824 may be of integer types of any bit width, but they must have the same
7825 bit width. The second element of the result structure must be of type
7826 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7832 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7833 a signed multiplication of the two arguments. They return a structure ---
7834 the first element of which is the multiplication, and the second element
7835 of which is a bit specifying if the signed multiplication resulted in an
7841 .. code-block:: llvm
7843 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7844 %sum = extractvalue {i32, i1} %res, 0
7845 %obit = extractvalue {i32, i1} %res, 1
7846 br i1 %obit, label %overflow, label %normal
7848 '``llvm.umul.with.overflow.*``' Intrinsics
7849 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7854 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
7855 on any integer bit width.
7859 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7860 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7861 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7866 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7867 a unsigned multiplication of the two arguments, and indicate whether an
7868 overflow occurred during the unsigned multiplication.
7873 The arguments (%a and %b) and the first element of the result structure
7874 may be of integer types of any bit width, but they must have the same
7875 bit width. The second element of the result structure must be of type
7876 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7882 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7883 an unsigned multiplication of the two arguments. They return a structure ---
7884 the first element of which is the multiplication, and the second
7885 element of which is a bit specifying if the unsigned multiplication
7886 resulted in an overflow.
7891 .. code-block:: llvm
7893 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7894 %sum = extractvalue {i32, i1} %res, 0
7895 %obit = extractvalue {i32, i1} %res, 1
7896 br i1 %obit, label %overflow, label %normal
7898 Specialised Arithmetic Intrinsics
7899 ---------------------------------
7901 '``llvm.fmuladd.*``' Intrinsic
7902 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7909 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
7910 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
7915 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
7916 expressions that can be fused if the code generator determines that (a) the
7917 target instruction set has support for a fused operation, and (b) that the
7918 fused operation is more efficient than the equivalent, separate pair of mul
7919 and add instructions.
7924 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
7925 multiplicands, a and b, and an addend c.
7934 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
7936 is equivalent to the expression a \* b + c, except that rounding will
7937 not be performed between the multiplication and addition steps if the
7938 code generator fuses the operations. Fusion is not guaranteed, even if
7939 the target platform supports it. If a fused multiply-add is required the
7940 corresponding llvm.fma.\* intrinsic function should be used instead.
7945 .. code-block:: llvm
7947 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
7949 Half Precision Floating Point Intrinsics
7950 ----------------------------------------
7952 For most target platforms, half precision floating point is a
7953 storage-only format. This means that it is a dense encoding (in memory)
7954 but does not support computation in the format.
7956 This means that code must first load the half-precision floating point
7957 value as an i16, then convert it to float with
7958 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
7959 then be performed on the float value (including extending to double
7960 etc). To store the value back to memory, it is first converted to float
7961 if needed, then converted to i16 with
7962 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
7965 .. _int_convert_to_fp16:
7967 '``llvm.convert.to.fp16``' Intrinsic
7968 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7975 declare i16 @llvm.convert.to.fp16(f32 %a)
7980 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7981 from single precision floating point format to half precision floating
7987 The intrinsic function contains single argument - the value to be
7993 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7994 from single precision floating point format to half precision floating
7995 point format. The return value is an ``i16`` which contains the
8001 .. code-block:: llvm
8003 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8004 store i16 %res, i16* @x, align 2
8006 .. _int_convert_from_fp16:
8008 '``llvm.convert.from.fp16``' Intrinsic
8009 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8016 declare f32 @llvm.convert.from.fp16(i16 %a)
8021 The '``llvm.convert.from.fp16``' intrinsic function performs a
8022 conversion from half precision floating point format to single precision
8023 floating point format.
8028 The intrinsic function contains single argument - the value to be
8034 The '``llvm.convert.from.fp16``' intrinsic function performs a
8035 conversion from half single precision floating point format to single
8036 precision floating point format. The input half-float value is
8037 represented by an ``i16`` value.
8042 .. code-block:: llvm
8044 %a = load i16* @x, align 2
8045 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8050 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8051 prefix), are described in the `LLVM Source Level
8052 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8055 Exception Handling Intrinsics
8056 -----------------------------
8058 The LLVM exception handling intrinsics (which all start with
8059 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8060 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8064 Trampoline Intrinsics
8065 ---------------------
8067 These intrinsics make it possible to excise one parameter, marked with
8068 the :ref:`nest <nest>` attribute, from a function. The result is a
8069 callable function pointer lacking the nest parameter - the caller does
8070 not need to provide a value for it. Instead, the value to use is stored
8071 in advance in a "trampoline", a block of memory usually allocated on the
8072 stack, which also contains code to splice the nest value into the
8073 argument list. This is used to implement the GCC nested function address
8076 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8077 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8078 It can be created as follows:
8080 .. code-block:: llvm
8082 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8083 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8084 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8085 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8086 %fp = bitcast i8* %p to i32 (i32, i32)*
8088 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8089 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8093 '``llvm.init.trampoline``' Intrinsic
8094 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8101 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8106 This fills the memory pointed to by ``tramp`` with executable code,
8107 turning it into a trampoline.
8112 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8113 pointers. The ``tramp`` argument must point to a sufficiently large and
8114 sufficiently aligned block of memory; this memory is written to by the
8115 intrinsic. Note that the size and the alignment are target-specific -
8116 LLVM currently provides no portable way of determining them, so a
8117 front-end that generates this intrinsic needs to have some
8118 target-specific knowledge. The ``func`` argument must hold a function
8119 bitcast to an ``i8*``.
8124 The block of memory pointed to by ``tramp`` is filled with target
8125 dependent code, turning it into a function. Then ``tramp`` needs to be
8126 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8127 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8128 function's signature is the same as that of ``func`` with any arguments
8129 marked with the ``nest`` attribute removed. At most one such ``nest``
8130 argument is allowed, and it must be of pointer type. Calling the new
8131 function is equivalent to calling ``func`` with the same argument list,
8132 but with ``nval`` used for the missing ``nest`` argument. If, after
8133 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8134 modified, then the effect of any later call to the returned function
8135 pointer is undefined.
8139 '``llvm.adjust.trampoline``' Intrinsic
8140 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8147 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8152 This performs any required machine-specific adjustment to the address of
8153 a trampoline (passed as ``tramp``).
8158 ``tramp`` must point to a block of memory which already has trampoline
8159 code filled in by a previous call to
8160 :ref:`llvm.init.trampoline <int_it>`.
8165 On some architectures the address of the code to be executed needs to be
8166 different to the address where the trampoline is actually stored. This
8167 intrinsic returns the executable address corresponding to ``tramp``
8168 after performing the required machine specific adjustments. The pointer
8169 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8174 This class of intrinsics exists to information about the lifetime of
8175 memory objects and ranges where variables are immutable.
8177 '``llvm.lifetime.start``' Intrinsic
8178 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8185 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8190 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8196 The first argument is a constant integer representing the size of the
8197 object, or -1 if it is variable sized. The second argument is a pointer
8203 This intrinsic indicates that before this point in the code, the value
8204 of the memory pointed to by ``ptr`` is dead. This means that it is known
8205 to never be used and has an undefined value. A load from the pointer
8206 that precedes this intrinsic can be replaced with ``'undef'``.
8208 '``llvm.lifetime.end``' Intrinsic
8209 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8216 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8221 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8227 The first argument is a constant integer representing the size of the
8228 object, or -1 if it is variable sized. The second argument is a pointer
8234 This intrinsic indicates that after this point in the code, the value of
8235 the memory pointed to by ``ptr`` is dead. This means that it is known to
8236 never be used and has an undefined value. Any stores into the memory
8237 object following this intrinsic may be removed as dead.
8239 '``llvm.invariant.start``' Intrinsic
8240 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8247 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8252 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8253 a memory object will not change.
8258 The first argument is a constant integer representing the size of the
8259 object, or -1 if it is variable sized. The second argument is a pointer
8265 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8266 the return value, the referenced memory location is constant and
8269 '``llvm.invariant.end``' Intrinsic
8270 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8277 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8282 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8283 memory object are mutable.
8288 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8289 The second argument is a constant integer representing the size of the
8290 object, or -1 if it is variable sized and the third argument is a
8291 pointer to the object.
8296 This intrinsic indicates that the memory is mutable again.
8301 This class of intrinsics is designed to be generic and has no specific
8304 '``llvm.var.annotation``' Intrinsic
8305 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8312 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8317 The '``llvm.var.annotation``' intrinsic.
8322 The first argument is a pointer to a value, the second is a pointer to a
8323 global string, the third is a pointer to a global string which is the
8324 source file name, and the last argument is the line number.
8329 This intrinsic allows annotation of local variables with arbitrary
8330 strings. This can be useful for special purpose optimizations that want
8331 to look for these annotations. These have no other defined use; they are
8332 ignored by code generation and optimization.
8334 '``llvm.annotation.*``' Intrinsic
8335 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8340 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8341 any integer bit width.
8345 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8346 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8347 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8348 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8349 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8354 The '``llvm.annotation``' intrinsic.
8359 The first argument is an integer value (result of some expression), the
8360 second is a pointer to a global string, the third is a pointer to a
8361 global string which is the source file name, and the last argument is
8362 the line number. It returns the value of the first argument.
8367 This intrinsic allows annotations to be put on arbitrary expressions
8368 with arbitrary strings. This can be useful for special purpose
8369 optimizations that want to look for these annotations. These have no
8370 other defined use; they are ignored by code generation and optimization.
8372 '``llvm.trap``' Intrinsic
8373 ^^^^^^^^^^^^^^^^^^^^^^^^^
8380 declare void @llvm.trap() noreturn nounwind
8385 The '``llvm.trap``' intrinsic.
8395 This intrinsic is lowered to the target dependent trap instruction. If
8396 the target does not have a trap instruction, this intrinsic will be
8397 lowered to a call of the ``abort()`` function.
8399 '``llvm.debugtrap``' Intrinsic
8400 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8407 declare void @llvm.debugtrap() nounwind
8412 The '``llvm.debugtrap``' intrinsic.
8422 This intrinsic is lowered to code which is intended to cause an
8423 execution trap with the intention of requesting the attention of a
8426 '``llvm.stackprotector``' Intrinsic
8427 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8434 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8439 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8440 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8441 is placed on the stack before local variables.
8446 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8447 The first argument is the value loaded from the stack guard
8448 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8449 enough space to hold the value of the guard.
8454 This intrinsic causes the prologue/epilogue inserter to force the
8455 position of the ``AllocaInst`` stack slot to be before local variables
8456 on the stack. This is to ensure that if a local variable on the stack is
8457 overwritten, it will destroy the value of the guard. When the function
8458 exits, the guard on the stack is checked against the original guard. If
8459 they are different, then the program aborts by calling the
8460 ``__stack_chk_fail()`` function.
8462 '``llvm.objectsize``' Intrinsic
8463 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8470 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8471 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8476 The ``llvm.objectsize`` intrinsic is designed to provide information to
8477 the optimizers to determine at compile time whether a) an operation
8478 (like memcpy) will overflow a buffer that corresponds to an object, or
8479 b) that a runtime check for overflow isn't necessary. An object in this
8480 context means an allocation of a specific class, structure, array, or
8486 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8487 argument is a pointer to or into the ``object``. The second argument is
8488 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8489 or -1 (if false) when the object size is unknown. The second argument
8490 only accepts constants.
8495 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8496 the size of the object concerned. If the size cannot be determined at
8497 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8498 on the ``min`` argument).
8500 '``llvm.expect``' Intrinsic
8501 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8508 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8509 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8514 The ``llvm.expect`` intrinsic provides information about expected (the
8515 most probable) value of ``val``, which can be used by optimizers.
8520 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8521 a value. The second argument is an expected value, this needs to be a
8522 constant value, variables are not allowed.
8527 This intrinsic is lowered to the ``val``.
8529 '``llvm.donothing``' Intrinsic
8530 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8537 declare void @llvm.donothing() nounwind readnone
8542 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8543 only intrinsic that can be called with an invoke instruction.
8553 This intrinsic does nothing, and it's removed by optimizers and ignored