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 attributes #0 = { alwaysinline alignstack=4 }
764 ; Target-dependent attributes:
765 attributes #1 = { "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, when emitting the prologue and
793 epilogue, the backend should forcibly align the stack pointer.
794 Specify the desired alignment, which must be a power of two, in
797 This attribute indicates that the inliner should attempt to inline
798 this function into callers whenever possible, ignoring any active
799 inlining size threshold for this caller.
801 This attribute suppresses lazy symbol binding for the function. This
802 may make calls to the function faster, at the cost of extra program
803 startup time if the function is not called during program startup.
805 This attribute indicates that the source code contained a hint that
806 inlining this function is desirable (such as the "inline" keyword in
807 C/C++). It is just a hint; it imposes no requirements on the
810 This attribute disables prologue / epilogue emission for the
811 function. This can have very system-specific consequences.
813 This indicates that the callee function at a call site is not
814 recognized as a built-in function. LLVM will retain the original call
815 and not replace it with equivalent code based on the semantics of the
816 built-in function. This is only valid at call sites, not on function
817 declarations or definitions.
819 This attribute indicates that calls to the function cannot be
820 duplicated. A call to a ``noduplicate`` function may be moved
821 within its parent function, but may not be duplicated within
824 A function containing a ``noduplicate`` call may still
825 be an inlining candidate, provided that the call is not
826 duplicated by inlining. That implies that the function has
827 internal linkage and only has one call site, so the original
828 call is dead after inlining.
830 This attributes disables implicit floating point instructions.
832 This attribute indicates that the inliner should never inline this
833 function in any situation. This attribute may not be used together
834 with the ``alwaysinline`` attribute.
836 This attribute indicates that the code generator should not use a
837 red zone, even if the target-specific ABI normally permits it.
839 This function attribute indicates that the function never returns
840 normally. This produces undefined behavior at runtime if the
841 function ever does dynamically return.
843 This function attribute indicates that the function never returns
844 with an unwind or exceptional control flow. If the function does
845 unwind, its runtime behavior is undefined.
847 This attribute suggests that optimization passes and code generator
848 passes make choices that keep the code size of this function low,
849 and otherwise do optimizations specifically to reduce code size.
851 This attribute indicates that the function computes its result (or
852 decides to unwind an exception) based strictly on its arguments,
853 without dereferencing any pointer arguments or otherwise accessing
854 any mutable state (e.g. memory, control registers, etc) visible to
855 caller functions. It does not write through any pointer arguments
856 (including ``byval`` arguments) and never changes any state visible
857 to callers. This means that it cannot unwind exceptions by calling
858 the ``C++`` exception throwing methods.
860 This attribute indicates that the function does not write through
861 any pointer arguments (including ``byval`` arguments) or otherwise
862 modify any state (e.g. memory, control registers, etc) visible to
863 caller functions. It may dereference pointer arguments and read
864 state that may be set in the caller. A readonly function always
865 returns the same value (or unwinds an exception identically) when
866 called with the same set of arguments and global state. It cannot
867 unwind an exception by calling the ``C++`` exception throwing
870 This attribute indicates that this function can return twice. The C
871 ``setjmp`` is an example of such a function. The compiler disables
872 some optimizations (like tail calls) in the caller of these
875 This attribute indicates that AddressSanitizer checks
876 (dynamic address safety analysis) are enabled for this function.
878 This attribute indicates that MemorySanitizer checks (dynamic detection
879 of accesses to uninitialized memory) are enabled for this function.
881 This attribute indicates that ThreadSanitizer checks
882 (dynamic thread safety analysis) are enabled for this function.
884 This attribute indicates that the function should emit a stack
885 smashing protector. It is in the form of a "canary" --- a random value
886 placed on the stack before the local variables that's checked upon
887 return from the function to see if it has been overwritten. A
888 heuristic is used to determine if a function needs stack protectors
889 or not. The heuristic used will enable protectors for functions with:
891 - Character arrays larger than ``ssp-buffer-size`` (default 8).
892 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
893 - Calls to alloca() with variable sizes or constant sizes greater than
896 If a function that has an ``ssp`` attribute is inlined into a
897 function that doesn't have an ``ssp`` attribute, then the resulting
898 function will have an ``ssp`` attribute.
900 This attribute indicates that the function should *always* emit a
901 stack smashing protector. This overrides the ``ssp`` function
904 If a function that has an ``sspreq`` attribute is inlined into a
905 function that doesn't have an ``sspreq`` attribute or which has an
906 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
907 an ``sspreq`` attribute.
909 This attribute indicates that the function should emit a stack smashing
910 protector. This attribute causes a strong heuristic to be used when
911 determining if a function needs stack protectors. The strong heuristic
912 will enable protectors for functions with:
914 - Arrays of any size and type
915 - Aggregates containing an array of any size and type.
917 - Local variables that have had their address taken.
919 This overrides the ``ssp`` function attribute.
921 If a function that has an ``sspstrong`` attribute is inlined into a
922 function that doesn't have an ``sspstrong`` attribute, then the
923 resulting function will have an ``sspstrong`` attribute.
925 This attribute indicates that the ABI being targeted requires that
926 an unwind table entry be produce for this function even if we can
927 show that no exceptions passes by it. This is normally the case for
928 the ELF x86-64 abi, but it can be disabled for some compilation
933 Module-Level Inline Assembly
934 ----------------------------
936 Modules may contain "module-level inline asm" blocks, which corresponds
937 to the GCC "file scope inline asm" blocks. These blocks are internally
938 concatenated by LLVM and treated as a single unit, but may be separated
939 in the ``.ll`` file if desired. The syntax is very simple:
943 module asm "inline asm code goes here"
944 module asm "more can go here"
946 The strings can contain any character by escaping non-printable
947 characters. The escape sequence used is simply "\\xx" where "xx" is the
948 two digit hex code for the number.
950 The inline asm code is simply printed to the machine code .s file when
951 assembly code is generated.
956 A module may specify a target specific data layout string that specifies
957 how data is to be laid out in memory. The syntax for the data layout is
962 target datalayout = "layout specification"
964 The *layout specification* consists of a list of specifications
965 separated by the minus sign character ('-'). Each specification starts
966 with a letter and may include other information after the letter to
967 define some aspect of the data layout. The specifications accepted are
971 Specifies that the target lays out data in big-endian form. That is,
972 the bits with the most significance have the lowest address
975 Specifies that the target lays out data in little-endian form. That
976 is, the bits with the least significance have the lowest address
979 Specifies the natural alignment of the stack in bits. Alignment
980 promotion of stack variables is limited to the natural stack
981 alignment to avoid dynamic stack realignment. The stack alignment
982 must be a multiple of 8-bits. If omitted, the natural stack
983 alignment defaults to "unspecified", which does not prevent any
984 alignment promotions.
985 ``p[n]:<size>:<abi>:<pref>``
986 This specifies the *size* of a pointer and its ``<abi>`` and
987 ``<pref>``\erred alignments for address space ``n``. All sizes are in
988 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
989 preceding ``:`` should be omitted too. The address space, ``n`` is
990 optional, and if not specified, denotes the default address space 0.
991 The value of ``n`` must be in the range [1,2^23).
992 ``i<size>:<abi>:<pref>``
993 This specifies the alignment for an integer type of a given bit
994 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
995 ``v<size>:<abi>:<pref>``
996 This specifies the alignment for a vector type of a given bit
998 ``f<size>:<abi>:<pref>``
999 This specifies the alignment for a floating point type of a given bit
1000 ``<size>``. Only values of ``<size>`` that are supported by the target
1001 will work. 32 (float) and 64 (double) are supported on all targets; 80
1002 or 128 (different flavors of long double) are also supported on some
1004 ``a<size>:<abi>:<pref>``
1005 This specifies the alignment for an aggregate type of a given bit
1007 ``s<size>:<abi>:<pref>``
1008 This specifies the alignment for a stack object of a given bit
1010 ``n<size1>:<size2>:<size3>...``
1011 This specifies a set of native integer widths for the target CPU in
1012 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1013 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1014 this set are considered to support most general arithmetic operations
1017 When constructing the data layout for a given target, LLVM starts with a
1018 default set of specifications which are then (possibly) overridden by
1019 the specifications in the ``datalayout`` keyword. The default
1020 specifications are given in this list:
1022 - ``E`` - big endian
1023 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment
1024 - ``S0`` - natural stack alignment is unspecified
1025 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1026 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1027 - ``i16:16:16`` - i16 is 16-bit aligned
1028 - ``i32:32:32`` - i32 is 32-bit aligned
1029 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1030 alignment of 64-bits
1031 - ``f16:16:16`` - half is 16-bit aligned
1032 - ``f32:32:32`` - float is 32-bit aligned
1033 - ``f64:64:64`` - double is 64-bit aligned
1034 - ``f128:128:128`` - quad is 128-bit aligned
1035 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1036 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1037 - ``a0:0:64`` - aggregates are 64-bit aligned
1039 When LLVM is determining the alignment for a given type, it uses the
1042 #. If the type sought is an exact match for one of the specifications,
1043 that specification is used.
1044 #. If no match is found, and the type sought is an integer type, then
1045 the smallest integer type that is larger than the bitwidth of the
1046 sought type is used. If none of the specifications are larger than
1047 the bitwidth then the largest integer type is used. For example,
1048 given the default specifications above, the i7 type will use the
1049 alignment of i8 (next largest) while both i65 and i256 will use the
1050 alignment of i64 (largest specified).
1051 #. If no match is found, and the type sought is a vector type, then the
1052 largest vector type that is smaller than the sought vector type will
1053 be used as a fall back. This happens because <128 x double> can be
1054 implemented in terms of 64 <2 x double>, for example.
1056 The function of the data layout string may not be what you expect.
1057 Notably, this is not a specification from the frontend of what alignment
1058 the code generator should use.
1060 Instead, if specified, the target data layout is required to match what
1061 the ultimate *code generator* expects. This string is used by the
1062 mid-level optimizers to improve code, and this only works if it matches
1063 what the ultimate code generator uses. If you would like to generate IR
1064 that does not embed this target-specific detail into the IR, then you
1065 don't have to specify the string. This will disable some optimizations
1066 that require precise layout information, but this also prevents those
1067 optimizations from introducing target specificity into the IR.
1069 .. _pointeraliasing:
1071 Pointer Aliasing Rules
1072 ----------------------
1074 Any memory access must be done through a pointer value associated with
1075 an address range of the memory access, otherwise the behavior is
1076 undefined. Pointer values are associated with address ranges according
1077 to the following rules:
1079 - A pointer value is associated with the addresses associated with any
1080 value it is *based* on.
1081 - An address of a global variable is associated with the address range
1082 of the variable's storage.
1083 - The result value of an allocation instruction is associated with the
1084 address range of the allocated storage.
1085 - A null pointer in the default address-space is associated with no
1087 - An integer constant other than zero or a pointer value returned from
1088 a function not defined within LLVM may be associated with address
1089 ranges allocated through mechanisms other than those provided by
1090 LLVM. Such ranges shall not overlap with any ranges of addresses
1091 allocated by mechanisms provided by LLVM.
1093 A pointer value is *based* on another pointer value according to the
1096 - A pointer value formed from a ``getelementptr`` operation is *based*
1097 on the first operand of the ``getelementptr``.
1098 - The result value of a ``bitcast`` is *based* on the operand of the
1100 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1101 values that contribute (directly or indirectly) to the computation of
1102 the pointer's value.
1103 - The "*based* on" relationship is transitive.
1105 Note that this definition of *"based"* is intentionally similar to the
1106 definition of *"based"* in C99, though it is slightly weaker.
1108 LLVM IR does not associate types with memory. The result type of a
1109 ``load`` merely indicates the size and alignment of the memory from
1110 which to load, as well as the interpretation of the value. The first
1111 operand type of a ``store`` similarly only indicates the size and
1112 alignment of the store.
1114 Consequently, type-based alias analysis, aka TBAA, aka
1115 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1116 :ref:`Metadata <metadata>` may be used to encode additional information
1117 which specialized optimization passes may use to implement type-based
1122 Volatile Memory Accesses
1123 ------------------------
1125 Certain memory accesses, such as :ref:`load <i_load>`'s,
1126 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1127 marked ``volatile``. The optimizers must not change the number of
1128 volatile operations or change their order of execution relative to other
1129 volatile operations. The optimizers *may* change the order of volatile
1130 operations relative to non-volatile operations. This is not Java's
1131 "volatile" and has no cross-thread synchronization behavior.
1133 IR-level volatile loads and stores cannot safely be optimized into
1134 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1135 flagged volatile. Likewise, the backend should never split or merge
1136 target-legal volatile load/store instructions.
1138 .. admonition:: Rationale
1140 Platforms may rely on volatile loads and stores of natively supported
1141 data width to be executed as single instruction. For example, in C
1142 this holds for an l-value of volatile primitive type with native
1143 hardware support, but not necessarily for aggregate types. The
1144 frontend upholds these expectations, which are intentionally
1145 unspecified in the IR. The rules above ensure that IR transformation
1146 do not violate the frontend's contract with the language.
1150 Memory Model for Concurrent Operations
1151 --------------------------------------
1153 The LLVM IR does not define any way to start parallel threads of
1154 execution or to register signal handlers. Nonetheless, there are
1155 platform-specific ways to create them, and we define LLVM IR's behavior
1156 in their presence. This model is inspired by the C++0x memory model.
1158 For a more informal introduction to this model, see the :doc:`Atomics`.
1160 We define a *happens-before* partial order as the least partial order
1163 - Is a superset of single-thread program order, and
1164 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1165 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1166 techniques, like pthread locks, thread creation, thread joining,
1167 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1168 Constraints <ordering>`).
1170 Note that program order does not introduce *happens-before* edges
1171 between a thread and signals executing inside that thread.
1173 Every (defined) read operation (load instructions, memcpy, atomic
1174 loads/read-modify-writes, etc.) R reads a series of bytes written by
1175 (defined) write operations (store instructions, atomic
1176 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1177 section, initialized globals are considered to have a write of the
1178 initializer which is atomic and happens before any other read or write
1179 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1180 may see any write to the same byte, except:
1182 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1183 write\ :sub:`2` happens before R\ :sub:`byte`, then
1184 R\ :sub:`byte` does not see write\ :sub:`1`.
1185 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1186 R\ :sub:`byte` does not see write\ :sub:`3`.
1188 Given that definition, R\ :sub:`byte` is defined as follows:
1190 - If R is volatile, the result is target-dependent. (Volatile is
1191 supposed to give guarantees which can support ``sig_atomic_t`` in
1192 C/C++, and may be used for accesses to addresses which do not behave
1193 like normal memory. It does not generally provide cross-thread
1195 - Otherwise, if there is no write to the same byte that happens before
1196 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1197 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1198 R\ :sub:`byte` returns the value written by that write.
1199 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1200 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1201 Memory Ordering Constraints <ordering>` section for additional
1202 constraints on how the choice is made.
1203 - Otherwise R\ :sub:`byte` returns ``undef``.
1205 R returns the value composed of the series of bytes it read. This
1206 implies that some bytes within the value may be ``undef`` **without**
1207 the entire value being ``undef``. Note that this only defines the
1208 semantics of the operation; it doesn't mean that targets will emit more
1209 than one instruction to read the series of bytes.
1211 Note that in cases where none of the atomic intrinsics are used, this
1212 model places only one restriction on IR transformations on top of what
1213 is required for single-threaded execution: introducing a store to a byte
1214 which might not otherwise be stored is not allowed in general.
1215 (Specifically, in the case where another thread might write to and read
1216 from an address, introducing a store can change a load that may see
1217 exactly one write into a load that may see multiple writes.)
1221 Atomic Memory Ordering Constraints
1222 ----------------------------------
1224 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1225 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1226 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1227 an ordering parameter that determines which other atomic instructions on
1228 the same address they *synchronize with*. These semantics are borrowed
1229 from Java and C++0x, but are somewhat more colloquial. If these
1230 descriptions aren't precise enough, check those specs (see spec
1231 references in the :doc:`atomics guide <Atomics>`).
1232 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1233 differently since they don't take an address. See that instruction's
1234 documentation for details.
1236 For a simpler introduction to the ordering constraints, see the
1240 The set of values that can be read is governed by the happens-before
1241 partial order. A value cannot be read unless some operation wrote
1242 it. This is intended to provide a guarantee strong enough to model
1243 Java's non-volatile shared variables. This ordering cannot be
1244 specified for read-modify-write operations; it is not strong enough
1245 to make them atomic in any interesting way.
1247 In addition to the guarantees of ``unordered``, there is a single
1248 total order for modifications by ``monotonic`` operations on each
1249 address. All modification orders must be compatible with the
1250 happens-before order. There is no guarantee that the modification
1251 orders can be combined to a global total order for the whole program
1252 (and this often will not be possible). The read in an atomic
1253 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1254 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1255 order immediately before the value it writes. If one atomic read
1256 happens before another atomic read of the same address, the later
1257 read must see the same value or a later value in the address's
1258 modification order. This disallows reordering of ``monotonic`` (or
1259 stronger) operations on the same address. If an address is written
1260 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1261 read that address repeatedly, the other threads must eventually see
1262 the write. This corresponds to the C++0x/C1x
1263 ``memory_order_relaxed``.
1265 In addition to the guarantees of ``monotonic``, a
1266 *synchronizes-with* edge may be formed with a ``release`` operation.
1267 This is intended to model C++'s ``memory_order_acquire``.
1269 In addition to the guarantees of ``monotonic``, if this operation
1270 writes a value which is subsequently read by an ``acquire``
1271 operation, it *synchronizes-with* that operation. (This isn't a
1272 complete description; see the C++0x definition of a release
1273 sequence.) This corresponds to the C++0x/C1x
1274 ``memory_order_release``.
1275 ``acq_rel`` (acquire+release)
1276 Acts as both an ``acquire`` and ``release`` operation on its
1277 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1278 ``seq_cst`` (sequentially consistent)
1279 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1280 operation which only reads, ``release`` for an operation which only
1281 writes), there is a global total order on all
1282 sequentially-consistent operations on all addresses, which is
1283 consistent with the *happens-before* partial order and with the
1284 modification orders of all the affected addresses. Each
1285 sequentially-consistent read sees the last preceding write to the
1286 same address in this global order. This corresponds to the C++0x/C1x
1287 ``memory_order_seq_cst`` and Java volatile.
1291 If an atomic operation is marked ``singlethread``, it only *synchronizes
1292 with* or participates in modification and seq\_cst total orderings with
1293 other operations running in the same thread (for example, in signal
1301 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1302 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1303 :ref:`frem <i_frem>`) have the following flags that can set to enable
1304 otherwise unsafe floating point operations
1307 No NaNs - Allow optimizations to assume the arguments and result are not
1308 NaN. Such optimizations are required to retain defined behavior over
1309 NaNs, but the value of the result is undefined.
1312 No Infs - Allow optimizations to assume the arguments and result are not
1313 +/-Inf. Such optimizations are required to retain defined behavior over
1314 +/-Inf, but the value of the result is undefined.
1317 No Signed Zeros - Allow optimizations to treat the sign of a zero
1318 argument or result as insignificant.
1321 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1322 argument rather than perform division.
1325 Fast - Allow algebraically equivalent transformations that may
1326 dramatically change results in floating point (e.g. reassociate). This
1327 flag implies all the others.
1334 The LLVM type system is one of the most important features of the
1335 intermediate representation. Being typed enables a number of
1336 optimizations to be performed on the intermediate representation
1337 directly, without having to do extra analyses on the side before the
1338 transformation. A strong type system makes it easier to read the
1339 generated code and enables novel analyses and transformations that are
1340 not feasible to perform on normal three address code representations.
1342 Type Classifications
1343 --------------------
1345 The types fall into a few useful classifications:
1354 * - :ref:`integer <t_integer>`
1355 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1358 * - :ref:`floating point <t_floating>`
1359 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1367 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1368 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1369 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1370 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1372 * - :ref:`primitive <t_primitive>`
1373 - :ref:`label <t_label>`,
1374 :ref:`void <t_void>`,
1375 :ref:`integer <t_integer>`,
1376 :ref:`floating point <t_floating>`,
1377 :ref:`x86mmx <t_x86mmx>`,
1378 :ref:`metadata <t_metadata>`.
1380 * - :ref:`derived <t_derived>`
1381 - :ref:`array <t_array>`,
1382 :ref:`function <t_function>`,
1383 :ref:`pointer <t_pointer>`,
1384 :ref:`structure <t_struct>`,
1385 :ref:`vector <t_vector>`,
1386 :ref:`opaque <t_opaque>`.
1388 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1389 Values of these types are the only ones which can be produced by
1397 The primitive types are the fundamental building blocks of the LLVM
1408 The integer type is a very simple type that simply specifies an
1409 arbitrary bit width for the integer type desired. Any bit width from 1
1410 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1419 The number of bits the integer will occupy is specified by the ``N``
1425 +----------------+------------------------------------------------+
1426 | ``i1`` | a single-bit integer. |
1427 +----------------+------------------------------------------------+
1428 | ``i32`` | a 32-bit integer. |
1429 +----------------+------------------------------------------------+
1430 | ``i1942652`` | a really big integer of over 1 million bits. |
1431 +----------------+------------------------------------------------+
1435 Floating Point Types
1436 ^^^^^^^^^^^^^^^^^^^^
1445 - 16-bit floating point value
1448 - 32-bit floating point value
1451 - 64-bit floating point value
1454 - 128-bit floating point value (112-bit mantissa)
1457 - 80-bit floating point value (X87)
1460 - 128-bit floating point value (two 64-bits)
1470 The x86mmx type represents a value held in an MMX register on an x86
1471 machine. The operations allowed on it are quite limited: parameters and
1472 return values, load and store, and bitcast. User-specified MMX
1473 instructions are represented as intrinsic or asm calls with arguments
1474 and/or results of this type. There are no arrays, vectors or constants
1492 The void type does not represent any value and has no size.
1509 The label type represents code labels.
1526 The metadata type represents embedded metadata. No derived types may be
1527 created from metadata except for :ref:`function <t_function>` arguments.
1541 The real power in LLVM comes from the derived types in the system. This
1542 is what allows a programmer to represent arrays, functions, pointers,
1543 and other useful types. Each of these types contain one or more element
1544 types which may be a primitive type, or another derived type. For
1545 example, it is possible to have a two dimensional array, using an array
1546 as the element type of another array.
1553 Aggregate Types are a subset of derived types that can contain multiple
1554 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1555 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1566 The array type is a very simple derived type that arranges elements
1567 sequentially in memory. The array type requires a size (number of
1568 elements) and an underlying data type.
1575 [<# elements> x <elementtype>]
1577 The number of elements is a constant integer value; ``elementtype`` may
1578 be any type with a size.
1583 +------------------+--------------------------------------+
1584 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1585 +------------------+--------------------------------------+
1586 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1587 +------------------+--------------------------------------+
1588 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1589 +------------------+--------------------------------------+
1591 Here are some examples of multidimensional arrays:
1593 +-----------------------------+----------------------------------------------------------+
1594 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1595 +-----------------------------+----------------------------------------------------------+
1596 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1597 +-----------------------------+----------------------------------------------------------+
1598 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1599 +-----------------------------+----------------------------------------------------------+
1601 There is no restriction on indexing beyond the end of the array implied
1602 by a static type (though there are restrictions on indexing beyond the
1603 bounds of an allocated object in some cases). This means that
1604 single-dimension 'variable sized array' addressing can be implemented in
1605 LLVM with a zero length array type. An implementation of 'pascal style
1606 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1617 The function type can be thought of as a function signature. It consists
1618 of a return type and a list of formal parameter types. The return type
1619 of a function type is a first class type or a void type.
1626 <returntype> (<parameter list>)
1628 ...where '``<parameter list>``' is a comma-separated list of type
1629 specifiers. Optionally, the parameter list may include a type ``...``,
1630 which indicates that the function takes a variable number of arguments.
1631 Variable argument functions can access their arguments with the
1632 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1633 '``<returntype>``' is any type except :ref:`label <t_label>`.
1638 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1639 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1640 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1641 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1642 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1643 | ``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. |
1644 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1645 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1646 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1656 The structure type is used to represent a collection of data members
1657 together in memory. The elements of a structure may be any type that has
1660 Structures in memory are accessed using '``load``' and '``store``' by
1661 getting a pointer to a field with the '``getelementptr``' instruction.
1662 Structures in registers are accessed using the '``extractvalue``' and
1663 '``insertvalue``' instructions.
1665 Structures may optionally be "packed" structures, which indicate that
1666 the alignment of the struct is one byte, and that there is no padding
1667 between the elements. In non-packed structs, padding between field types
1668 is inserted as defined by the DataLayout string in the module, which is
1669 required to match what the underlying code generator expects.
1671 Structures can either be "literal" or "identified". A literal structure
1672 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1673 identified types are always defined at the top level with a name.
1674 Literal types are uniqued by their contents and can never be recursive
1675 or opaque since there is no way to write one. Identified types can be
1676 recursive, can be opaqued, and are never uniqued.
1683 %T1 = type { <type list> } ; Identified normal struct type
1684 %T2 = type <{ <type list> }> ; Identified packed struct type
1689 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1690 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1691 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1692 | ``{ 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``. |
1693 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1694 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1695 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1699 Opaque Structure Types
1700 ^^^^^^^^^^^^^^^^^^^^^^
1705 Opaque structure types are used to represent named structure types that
1706 do not have a body specified. This corresponds (for example) to the C
1707 notion of a forward declared structure.
1720 +--------------+-------------------+
1721 | ``opaque`` | An opaque type. |
1722 +--------------+-------------------+
1732 The pointer type is used to specify memory locations. Pointers are
1733 commonly used to reference objects in memory.
1735 Pointer types may have an optional address space attribute defining the
1736 numbered address space where the pointed-to object resides. The default
1737 address space is number zero. The semantics of non-zero address spaces
1738 are target-specific.
1740 Note that LLVM does not permit pointers to void (``void*``) nor does it
1741 permit pointers to labels (``label*``). Use ``i8*`` instead.
1753 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1754 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1755 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1756 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1757 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1758 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1759 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1769 A vector type is a simple derived type that represents a vector of
1770 elements. Vector types are used when multiple primitive data are
1771 operated in parallel using a single instruction (SIMD). A vector type
1772 requires a size (number of elements) and an underlying primitive data
1773 type. Vector types are considered :ref:`first class <t_firstclass>`.
1780 < <# elements> x <elementtype> >
1782 The number of elements is a constant integer value larger than 0;
1783 elementtype may be any integer or floating point type, or a pointer to
1784 these types. Vectors of size zero are not allowed.
1789 +-------------------+--------------------------------------------------+
1790 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1791 +-------------------+--------------------------------------------------+
1792 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1793 +-------------------+--------------------------------------------------+
1794 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1795 +-------------------+--------------------------------------------------+
1796 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1797 +-------------------+--------------------------------------------------+
1802 LLVM has several different basic types of constants. This section
1803 describes them all and their syntax.
1808 **Boolean constants**
1809 The two strings '``true``' and '``false``' are both valid constants
1811 **Integer constants**
1812 Standard integers (such as '4') are constants of the
1813 :ref:`integer <t_integer>` type. Negative numbers may be used with
1815 **Floating point constants**
1816 Floating point constants use standard decimal notation (e.g.
1817 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1818 hexadecimal notation (see below). The assembler requires the exact
1819 decimal value of a floating-point constant. For example, the
1820 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1821 decimal in binary. Floating point constants must have a :ref:`floating
1822 point <t_floating>` type.
1823 **Null pointer constants**
1824 The identifier '``null``' is recognized as a null pointer constant
1825 and must be of :ref:`pointer type <t_pointer>`.
1827 The one non-intuitive notation for constants is the hexadecimal form of
1828 floating point constants. For example, the form
1829 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1830 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1831 constants are required (and the only time that they are generated by the
1832 disassembler) is when a floating point constant must be emitted but it
1833 cannot be represented as a decimal floating point number in a reasonable
1834 number of digits. For example, NaN's, infinities, and other special
1835 values are represented in their IEEE hexadecimal format so that assembly
1836 and disassembly do not cause any bits to change in the constants.
1838 When using the hexadecimal form, constants of types half, float, and
1839 double are represented using the 16-digit form shown above (which
1840 matches the IEEE754 representation for double); half and float values
1841 must, however, be exactly representable as IEEE 754 half and single
1842 precision, respectively. Hexadecimal format is always used for long
1843 double, and there are three forms of long double. The 80-bit format used
1844 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1845 128-bit format used by PowerPC (two adjacent doubles) is represented by
1846 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1847 represented by ``0xL`` followed by 32 hexadecimal digits; no currently
1848 supported target uses this format. Long doubles will only work if they
1849 match the long double format on your target. The IEEE 16-bit format
1850 (half precision) is represented by ``0xH`` followed by 4 hexadecimal
1851 digits. All hexadecimal formats are big-endian (sign bit at the left).
1853 There are no constants of type x86mmx.
1858 Complex constants are a (potentially recursive) combination of simple
1859 constants and smaller complex constants.
1861 **Structure constants**
1862 Structure constants are represented with notation similar to
1863 structure type definitions (a comma separated list of elements,
1864 surrounded by braces (``{}``)). For example:
1865 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1866 "``@G = external global i32``". Structure constants must have
1867 :ref:`structure type <t_struct>`, and the number and types of elements
1868 must match those specified by the type.
1870 Array constants are represented with notation similar to array type
1871 definitions (a comma separated list of elements, surrounded by
1872 square brackets (``[]``)). For example:
1873 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1874 :ref:`array type <t_array>`, and the number and types of elements must
1875 match those specified by the type.
1876 **Vector constants**
1877 Vector constants are represented with notation similar to vector
1878 type definitions (a comma separated list of elements, surrounded by
1879 less-than/greater-than's (``<>``)). For example:
1880 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1881 must have :ref:`vector type <t_vector>`, and the number and types of
1882 elements must match those specified by the type.
1883 **Zero initialization**
1884 The string '``zeroinitializer``' can be used to zero initialize a
1885 value to zero of *any* type, including scalar and
1886 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1887 having to print large zero initializers (e.g. for large arrays) and
1888 is always exactly equivalent to using explicit zero initializers.
1890 A metadata node is a structure-like constant with :ref:`metadata
1891 type <t_metadata>`. For example:
1892 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1893 constants that are meant to be interpreted as part of the
1894 instruction stream, metadata is a place to attach additional
1895 information such as debug info.
1897 Global Variable and Function Addresses
1898 --------------------------------------
1900 The addresses of :ref:`global variables <globalvars>` and
1901 :ref:`functions <functionstructure>` are always implicitly valid
1902 (link-time) constants. These constants are explicitly referenced when
1903 the :ref:`identifier for the global <identifiers>` is used and always have
1904 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1907 .. code-block:: llvm
1911 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1918 The string '``undef``' can be used anywhere a constant is expected, and
1919 indicates that the user of the value may receive an unspecified
1920 bit-pattern. Undefined values may be of any type (other than '``label``'
1921 or '``void``') and be used anywhere a constant is permitted.
1923 Undefined values are useful because they indicate to the compiler that
1924 the program is well defined no matter what value is used. This gives the
1925 compiler more freedom to optimize. Here are some examples of
1926 (potentially surprising) transformations that are valid (in pseudo IR):
1928 .. code-block:: llvm
1938 This is safe because all of the output bits are affected by the undef
1939 bits. Any output bit can have a zero or one depending on the input bits.
1941 .. code-block:: llvm
1952 These logical operations have bits that are not always affected by the
1953 input. For example, if ``%X`` has a zero bit, then the output of the
1954 '``and``' operation will always be a zero for that bit, no matter what
1955 the corresponding bit from the '``undef``' is. As such, it is unsafe to
1956 optimize or assume that the result of the '``and``' is '``undef``'.
1957 However, it is safe to assume that all bits of the '``undef``' could be
1958 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
1959 all the bits of the '``undef``' operand to the '``or``' could be set,
1960 allowing the '``or``' to be folded to -1.
1962 .. code-block:: llvm
1964 %A = select undef, %X, %Y
1965 %B = select undef, 42, %Y
1966 %C = select %X, %Y, undef
1976 This set of examples shows that undefined '``select``' (and conditional
1977 branch) conditions can go *either way*, but they have to come from one
1978 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
1979 both known to have a clear low bit, then ``%A`` would have to have a
1980 cleared low bit. However, in the ``%C`` example, the optimizer is
1981 allowed to assume that the '``undef``' operand could be the same as
1982 ``%Y``, allowing the whole '``select``' to be eliminated.
1984 .. code-block:: llvm
1986 %A = xor undef, undef
2003 This example points out that two '``undef``' operands are not
2004 necessarily the same. This can be surprising to people (and also matches
2005 C semantics) where they assume that "``X^X``" is always zero, even if
2006 ``X`` is undefined. This isn't true for a number of reasons, but the
2007 short answer is that an '``undef``' "variable" can arbitrarily change
2008 its value over its "live range". This is true because the variable
2009 doesn't actually *have a live range*. Instead, the value is logically
2010 read from arbitrary registers that happen to be around when needed, so
2011 the value is not necessarily consistent over time. In fact, ``%A`` and
2012 ``%C`` need to have the same semantics or the core LLVM "replace all
2013 uses with" concept would not hold.
2015 .. code-block:: llvm
2023 These examples show the crucial difference between an *undefined value*
2024 and *undefined behavior*. An undefined value (like '``undef``') is
2025 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2026 operation can be constant folded to '``undef``', because the '``undef``'
2027 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2028 However, in the second example, we can make a more aggressive
2029 assumption: because the ``undef`` is allowed to be an arbitrary value,
2030 we are allowed to assume that it could be zero. Since a divide by zero
2031 has *undefined behavior*, we are allowed to assume that the operation
2032 does not execute at all. This allows us to delete the divide and all
2033 code after it. Because the undefined operation "can't happen", the
2034 optimizer can assume that it occurs in dead code.
2036 .. code-block:: llvm
2038 a: store undef -> %X
2039 b: store %X -> undef
2044 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2045 value can be assumed to not have any effect; we can assume that the
2046 value is overwritten with bits that happen to match what was already
2047 there. However, a store *to* an undefined location could clobber
2048 arbitrary memory, therefore, it has undefined behavior.
2055 Poison values are similar to :ref:`undef values <undefvalues>`, however
2056 they also represent the fact that an instruction or constant expression
2057 which cannot evoke side effects has nevertheless detected a condition
2058 which results in undefined behavior.
2060 There is currently no way of representing a poison value in the IR; they
2061 only exist when produced by operations such as :ref:`add <i_add>` with
2064 Poison value behavior is defined in terms of value *dependence*:
2066 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2067 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2068 their dynamic predecessor basic block.
2069 - Function arguments depend on the corresponding actual argument values
2070 in the dynamic callers of their functions.
2071 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2072 instructions that dynamically transfer control back to them.
2073 - :ref:`Invoke <i_invoke>` instructions depend on the
2074 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2075 call instructions that dynamically transfer control back to them.
2076 - Non-volatile loads and stores depend on the most recent stores to all
2077 of the referenced memory addresses, following the order in the IR
2078 (including loads and stores implied by intrinsics such as
2079 :ref:`@llvm.memcpy <int_memcpy>`.)
2080 - An instruction with externally visible side effects depends on the
2081 most recent preceding instruction with externally visible side
2082 effects, following the order in the IR. (This includes :ref:`volatile
2083 operations <volatile>`.)
2084 - An instruction *control-depends* on a :ref:`terminator
2085 instruction <terminators>` if the terminator instruction has
2086 multiple successors and the instruction is always executed when
2087 control transfers to one of the successors, and may not be executed
2088 when control is transferred to another.
2089 - Additionally, an instruction also *control-depends* on a terminator
2090 instruction if the set of instructions it otherwise depends on would
2091 be different if the terminator had transferred control to a different
2093 - Dependence is transitive.
2095 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2096 with the additional affect that any instruction which has a *dependence*
2097 on a poison value has undefined behavior.
2099 Here are some examples:
2101 .. code-block:: llvm
2104 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2105 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2106 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2107 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2109 store i32 %poison, i32* @g ; Poison value stored to memory.
2110 %poison2 = load i32* @g ; Poison value loaded back from memory.
2112 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2114 %narrowaddr = bitcast i32* @g to i16*
2115 %wideaddr = bitcast i32* @g to i64*
2116 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2117 %poison4 = load i64* %wideaddr ; Returns a poison value.
2119 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2120 br i1 %cmp, label %true, label %end ; Branch to either destination.
2123 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2124 ; it has undefined behavior.
2128 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2129 ; Both edges into this PHI are
2130 ; control-dependent on %cmp, so this
2131 ; always results in a poison value.
2133 store volatile i32 0, i32* @g ; This would depend on the store in %true
2134 ; if %cmp is true, or the store in %entry
2135 ; otherwise, so this is undefined behavior.
2137 br i1 %cmp, label %second_true, label %second_end
2138 ; The same branch again, but this time the
2139 ; true block doesn't have side effects.
2146 store volatile i32 0, i32* @g ; This time, the instruction always depends
2147 ; on the store in %end. Also, it is
2148 ; control-equivalent to %end, so this is
2149 ; well-defined (ignoring earlier undefined
2150 ; behavior in this example).
2154 Addresses of Basic Blocks
2155 -------------------------
2157 ``blockaddress(@function, %block)``
2159 The '``blockaddress``' constant computes the address of the specified
2160 basic block in the specified function, and always has an ``i8*`` type.
2161 Taking the address of the entry block is illegal.
2163 This value only has defined behavior when used as an operand to the
2164 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2165 against null. Pointer equality tests between labels addresses results in
2166 undefined behavior --- though, again, comparison against null is ok, and
2167 no label is equal to the null pointer. This may be passed around as an
2168 opaque pointer sized value as long as the bits are not inspected. This
2169 allows ``ptrtoint`` and arithmetic to be performed on these values so
2170 long as the original value is reconstituted before the ``indirectbr``
2173 Finally, some targets may provide defined semantics when using the value
2174 as the operand to an inline assembly, but that is target specific.
2176 Constant Expressions
2177 --------------------
2179 Constant expressions are used to allow expressions involving other
2180 constants to be used as constants. Constant expressions may be of any
2181 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2182 that does not have side effects (e.g. load and call are not supported).
2183 The following is the syntax for constant expressions:
2185 ``trunc (CST to TYPE)``
2186 Truncate a constant to another type. The bit size of CST must be
2187 larger than the bit size of TYPE. Both types must be integers.
2188 ``zext (CST to TYPE)``
2189 Zero extend a constant to another type. The bit size of CST must be
2190 smaller than the bit size of TYPE. Both types must be integers.
2191 ``sext (CST to TYPE)``
2192 Sign extend a constant to another type. The bit size of CST must be
2193 smaller than the bit size of TYPE. Both types must be integers.
2194 ``fptrunc (CST to TYPE)``
2195 Truncate a floating point constant to another floating point type.
2196 The size of CST must be larger than the size of TYPE. Both types
2197 must be floating point.
2198 ``fpext (CST to TYPE)``
2199 Floating point extend a constant to another type. The size of CST
2200 must be smaller or equal to the size of TYPE. Both types must be
2202 ``fptoui (CST to TYPE)``
2203 Convert a floating point constant to the corresponding unsigned
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 ``fptosi (CST to TYPE)``
2209 Convert a floating point constant to the corresponding signed
2210 integer constant. TYPE must be a scalar or vector integer type. CST
2211 must be of scalar or vector floating point type. Both CST and TYPE
2212 must be scalars, or vectors of the same number of elements. If the
2213 value won't fit in the integer type, the results are undefined.
2214 ``uitofp (CST to TYPE)``
2215 Convert an unsigned 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 ``sitofp (CST to TYPE)``
2221 Convert a signed integer constant to the corresponding floating
2222 point constant. TYPE must be a scalar or vector floating point type.
2223 CST must be of scalar or vector integer type. Both CST and TYPE must
2224 be scalars, or vectors of the same number of elements. If the value
2225 won't fit in the floating point type, the results are undefined.
2226 ``ptrtoint (CST to TYPE)``
2227 Convert a pointer typed constant to the corresponding integer
2228 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2229 pointer type. The ``CST`` value is zero extended, truncated, or
2230 unchanged to make it fit in ``TYPE``.
2231 ``inttoptr (CST to TYPE)``
2232 Convert an integer constant to a pointer constant. TYPE must be a
2233 pointer type. CST must be of integer type. The CST value is zero
2234 extended, truncated, or unchanged to make it fit in a pointer size.
2235 This one is *really* dangerous!
2236 ``bitcast (CST to TYPE)``
2237 Convert a constant, CST, to another TYPE. The constraints of the
2238 operands are the same as those for the :ref:`bitcast
2239 instruction <i_bitcast>`.
2240 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2241 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2242 constants. As with the :ref:`getelementptr <i_getelementptr>`
2243 instruction, the index list may have zero or more indexes, which are
2244 required to make sense for the type of "CSTPTR".
2245 ``select (COND, VAL1, VAL2)``
2246 Perform the :ref:`select operation <i_select>` on constants.
2247 ``icmp COND (VAL1, VAL2)``
2248 Performs the :ref:`icmp operation <i_icmp>` on constants.
2249 ``fcmp COND (VAL1, VAL2)``
2250 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2251 ``extractelement (VAL, IDX)``
2252 Perform the :ref:`extractelement operation <i_extractelement>` on
2254 ``insertelement (VAL, ELT, IDX)``
2255 Perform the :ref:`insertelement operation <i_insertelement>` on
2257 ``shufflevector (VEC1, VEC2, IDXMASK)``
2258 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2260 ``extractvalue (VAL, IDX0, IDX1, ...)``
2261 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2262 constants. The index list is interpreted in a similar manner as
2263 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2264 least one index value must be specified.
2265 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2266 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2267 The index list is interpreted in a similar manner as indices in a
2268 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2269 value must be specified.
2270 ``OPCODE (LHS, RHS)``
2271 Perform the specified operation of the LHS and RHS constants. OPCODE
2272 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2273 binary <bitwiseops>` operations. The constraints on operands are
2274 the same as those for the corresponding instruction (e.g. no bitwise
2275 operations on floating point values are allowed).
2280 Inline Assembler Expressions
2281 ----------------------------
2283 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2284 Inline Assembly <moduleasm>`) through the use of a special value. This
2285 value represents the inline assembler as a string (containing the
2286 instructions to emit), a list of operand constraints (stored as a
2287 string), a flag that indicates whether or not the inline asm expression
2288 has side effects, and a flag indicating whether the function containing
2289 the asm needs to align its stack conservatively. An example inline
2290 assembler expression is:
2292 .. code-block:: llvm
2294 i32 (i32) asm "bswap $0", "=r,r"
2296 Inline assembler expressions may **only** be used as the callee operand
2297 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2298 Thus, typically we have:
2300 .. code-block:: llvm
2302 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2304 Inline asms with side effects not visible in the constraint list must be
2305 marked as having side effects. This is done through the use of the
2306 '``sideeffect``' keyword, like so:
2308 .. code-block:: llvm
2310 call void asm sideeffect "eieio", ""()
2312 In some cases inline asms will contain code that will not work unless
2313 the stack is aligned in some way, such as calls or SSE instructions on
2314 x86, yet will not contain code that does that alignment within the asm.
2315 The compiler should make conservative assumptions about what the asm
2316 might contain and should generate its usual stack alignment code in the
2317 prologue if the '``alignstack``' keyword is present:
2319 .. code-block:: llvm
2321 call void asm alignstack "eieio", ""()
2323 Inline asms also support using non-standard assembly dialects. The
2324 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2325 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2326 the only supported dialects. An example is:
2328 .. code-block:: llvm
2330 call void asm inteldialect "eieio", ""()
2332 If multiple keywords appear the '``sideeffect``' keyword must come
2333 first, the '``alignstack``' keyword second and the '``inteldialect``'
2339 The call instructions that wrap inline asm nodes may have a
2340 "``!srcloc``" MDNode attached to it that contains a list of constant
2341 integers. If present, the code generator will use the integer as the
2342 location cookie value when report errors through the ``LLVMContext``
2343 error reporting mechanisms. This allows a front-end to correlate backend
2344 errors that occur with inline asm back to the source code that produced
2347 .. code-block:: llvm
2349 call void asm sideeffect "something bad", ""(), !srcloc !42
2351 !42 = !{ i32 1234567 }
2353 It is up to the front-end to make sense of the magic numbers it places
2354 in the IR. If the MDNode contains multiple constants, the code generator
2355 will use the one that corresponds to the line of the asm that the error
2360 Metadata Nodes and Metadata Strings
2361 -----------------------------------
2363 LLVM IR allows metadata to be attached to instructions in the program
2364 that can convey extra information about the code to the optimizers and
2365 code generator. One example application of metadata is source-level
2366 debug information. There are two metadata primitives: strings and nodes.
2367 All metadata has the ``metadata`` type and is identified in syntax by a
2368 preceding exclamation point ('``!``').
2370 A metadata string is a string surrounded by double quotes. It can
2371 contain any character by escaping non-printable characters with
2372 "``\xx``" where "``xx``" is the two digit hex code. For example:
2375 Metadata nodes are represented with notation similar to structure
2376 constants (a comma separated list of elements, surrounded by braces and
2377 preceded by an exclamation point). Metadata nodes can have any values as
2378 their operand. For example:
2380 .. code-block:: llvm
2382 !{ metadata !"test\00", i32 10}
2384 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2385 metadata nodes, which can be looked up in the module symbol table. For
2388 .. code-block:: llvm
2390 !foo = metadata !{!4, !3}
2392 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2393 function is using two metadata arguments:
2395 .. code-block:: llvm
2397 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2399 Metadata can be attached with an instruction. Here metadata ``!21`` is
2400 attached to the ``add`` instruction using the ``!dbg`` identifier:
2402 .. code-block:: llvm
2404 %indvar.next = add i64 %indvar, 1, !dbg !21
2406 More information about specific metadata nodes recognized by the
2407 optimizers and code generator is found below.
2412 In LLVM IR, memory does not have types, so LLVM's own type system is not
2413 suitable for doing TBAA. Instead, metadata is added to the IR to
2414 describe a type system of a higher level language. This can be used to
2415 implement typical C/C++ TBAA, but it can also be used to implement
2416 custom alias analysis behavior for other languages.
2418 The current metadata format is very simple. TBAA metadata nodes have up
2419 to three fields, e.g.:
2421 .. code-block:: llvm
2423 !0 = metadata !{ metadata !"an example type tree" }
2424 !1 = metadata !{ metadata !"int", metadata !0 }
2425 !2 = metadata !{ metadata !"float", metadata !0 }
2426 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2428 The first field is an identity field. It can be any value, usually a
2429 metadata string, which uniquely identifies the type. The most important
2430 name in the tree is the name of the root node. Two trees with different
2431 root node names are entirely disjoint, even if they have leaves with
2434 The second field identifies the type's parent node in the tree, or is
2435 null or omitted for a root node. A type is considered to alias all of
2436 its descendants and all of its ancestors in the tree. Also, a type is
2437 considered to alias all types in other trees, so that bitcode produced
2438 from multiple front-ends is handled conservatively.
2440 If the third field is present, it's an integer which if equal to 1
2441 indicates that the type is "constant" (meaning
2442 ``pointsToConstantMemory`` should return true; see `other useful
2443 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2445 '``tbaa.struct``' Metadata
2446 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2448 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2449 aggregate assignment operations in C and similar languages, however it
2450 is defined to copy a contiguous region of memory, which is more than
2451 strictly necessary for aggregate types which contain holes due to
2452 padding. Also, it doesn't contain any TBAA information about the fields
2455 ``!tbaa.struct`` metadata can describe which memory subregions in a
2456 memcpy are padding and what the TBAA tags of the struct are.
2458 The current metadata format is very simple. ``!tbaa.struct`` metadata
2459 nodes are a list of operands which are in conceptual groups of three.
2460 For each group of three, the first operand gives the byte offset of a
2461 field in bytes, the second gives its size in bytes, and the third gives
2464 .. code-block:: llvm
2466 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2468 This describes a struct with two fields. The first is at offset 0 bytes
2469 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2470 and has size 4 bytes and has tbaa tag !2.
2472 Note that the fields need not be contiguous. In this example, there is a
2473 4 byte gap between the two fields. This gap represents padding which
2474 does not carry useful data and need not be preserved.
2476 '``fpmath``' Metadata
2477 ^^^^^^^^^^^^^^^^^^^^^
2479 ``fpmath`` metadata may be attached to any instruction of floating point
2480 type. It can be used to express the maximum acceptable error in the
2481 result of that instruction, in ULPs, thus potentially allowing the
2482 compiler to use a more efficient but less accurate method of computing
2483 it. ULP is defined as follows:
2485 If ``x`` is a real number that lies between two finite consecutive
2486 floating-point numbers ``a`` and ``b``, without being equal to one
2487 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2488 distance between the two non-equal finite floating-point numbers
2489 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2491 The metadata node shall consist of a single positive floating point
2492 number representing the maximum relative error, for example:
2494 .. code-block:: llvm
2496 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2498 '``range``' Metadata
2499 ^^^^^^^^^^^^^^^^^^^^
2501 ``range`` metadata may be attached only to loads of integer types. It
2502 expresses the possible ranges the loaded value is in. The ranges are
2503 represented with a flattened list of integers. The loaded value is known
2504 to be in the union of the ranges defined by each consecutive pair. Each
2505 pair has the following properties:
2507 - The type must match the type loaded by the instruction.
2508 - The pair ``a,b`` represents the range ``[a,b)``.
2509 - Both ``a`` and ``b`` are constants.
2510 - The range is allowed to wrap.
2511 - The range should not represent the full or empty set. That is,
2514 In addition, the pairs must be in signed order of the lower bound and
2515 they must be non-contiguous.
2519 .. code-block:: llvm
2521 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2522 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2523 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2524 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2526 !0 = metadata !{ i8 0, i8 2 }
2527 !1 = metadata !{ i8 255, i8 2 }
2528 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2529 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2534 It is sometimes useful to attach information to loop constructs. Currently,
2535 loop metadata is implemented as metadata attached to the branch instruction
2536 in the loop latch block. This type of metadata refer to a metadata node that is
2537 guaranteed to be separate for each loop. The loop-level metadata is prefixed
2540 The loop identifier metadata is implemented using a metadata that refers to
2541 itself to avoid merging it with any other identifier metadata, e.g.,
2542 during module linkage or function inlining. That is, each loop should refer
2543 to their own identification metadata even if they reside in separate functions.
2544 The following example contains loop identifier metadata for two separate loop
2547 .. code-block:: llvm
2549 !0 = metadata !{ metadata !0 }
2550 !1 = metadata !{ metadata !1 }
2553 '``llvm.loop.parallel``' Metadata
2554 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2556 This loop metadata can be used to communicate that a loop should be considered
2557 a parallel loop. The semantics of parallel loops in this case is the one
2558 with the strongest cross-iteration instruction ordering freedom: the
2559 iterations in the loop can be considered completely independent of each
2560 other (also known as embarrassingly parallel loops).
2562 This metadata can originate from a programming language with parallel loop
2563 constructs. In such a case it is completely the programmer's responsibility
2564 to ensure the instructions from the different iterations of the loop can be
2565 executed in an arbitrary order, in parallel, or intertwined. No loop-carried
2566 dependency checking at all must be expected from the compiler.
2568 In order to fulfill the LLVM requirement for metadata to be safely ignored,
2569 it is important to ensure that a parallel loop is converted to
2570 a sequential loop in case an optimization (agnostic of the parallel loop
2571 semantics) converts the loop back to such. This happens when new memory
2572 accesses that do not fulfill the requirement of free ordering across iterations
2573 are added to the loop. Therefore, this metadata is required, but not
2574 sufficient, to consider the loop at hand a parallel loop. For a loop
2575 to be parallel, all its memory accessing instructions need to be
2576 marked with the ``llvm.mem.parallel_loop_access`` metadata that refer
2577 to the same loop identifier metadata that identify the loop at hand.
2582 Metadata types used to annotate memory accesses with information helpful
2583 for optimizations are prefixed with ``llvm.mem``.
2585 '``llvm.mem.parallel_loop_access``' Metadata
2586 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2588 For a loop to be parallel, in addition to using
2589 the ``llvm.loop.parallel`` metadata to mark the loop latch branch instruction,
2590 also all of the memory accessing instructions in the loop body need to be
2591 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2592 is at least one memory accessing instruction not marked with the metadata,
2593 the loop, despite it possibly using the ``llvm.loop.parallel`` metadata,
2594 must be considered a sequential loop. This causes parallel loops to be
2595 converted to sequential loops due to optimization passes that are unaware of
2596 the parallel semantics and that insert new memory instructions to the loop
2599 Example of a loop that is considered parallel due to its correct use of
2600 both ``llvm.loop.parallel`` and ``llvm.mem.parallel_loop_access``
2601 metadata types that refer to the same loop identifier metadata.
2603 .. code-block:: llvm
2607 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2609 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2611 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop.parallel !0
2615 !0 = metadata !{ metadata !0 }
2617 It is also possible to have nested parallel loops. In that case the
2618 memory accesses refer to a list of loop identifier metadata nodes instead of
2619 the loop identifier metadata node directly:
2621 .. code-block:: llvm
2628 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2630 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2632 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop.parallel !1
2636 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2638 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2640 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop.parallel !2
2642 outer.for.end: ; preds = %for.body
2644 !0 = metadata !{ metadata !1, metadata !2 } ; a list of parallel loop identifiers
2645 !1 = metadata !{ metadata !1 } ; an identifier for the inner parallel loop
2646 !2 = metadata !{ metadata !2 } ; an identifier for the outer parallel loop
2649 Module Flags Metadata
2650 =====================
2652 Information about the module as a whole is difficult to convey to LLVM's
2653 subsystems. The LLVM IR isn't sufficient to transmit this information.
2654 The ``llvm.module.flags`` named metadata exists in order to facilitate
2655 this. These flags are in the form of key / value pairs --- much like a
2656 dictionary --- making it easy for any subsystem who cares about a flag to
2659 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2660 Each triplet has the following form:
2662 - The first element is a *behavior* flag, which specifies the behavior
2663 when two (or more) modules are merged together, and it encounters two
2664 (or more) metadata with the same ID. The supported behaviors are
2666 - The second element is a metadata string that is a unique ID for the
2667 metadata. Each module may only have one flag entry for each unique ID (not
2668 including entries with the **Require** behavior).
2669 - The third element is the value of the flag.
2671 When two (or more) modules are merged together, the resulting
2672 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2673 each unique metadata ID string, there will be exactly one entry in the merged
2674 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2675 be determined by the merge behavior flag, as described below. The only exception
2676 is that entries with the *Require* behavior are always preserved.
2678 The following behaviors are supported:
2689 Emits an error if two values disagree, otherwise the resulting value
2690 is that of the operands.
2694 Emits a warning if two values disagree. The result value will be the
2695 operand for the flag from the first module being linked.
2699 Adds a requirement that another module flag be present and have a
2700 specified value after linking is performed. The value must be a
2701 metadata pair, where the first element of the pair is the ID of the
2702 module flag to be restricted, and the second element of the pair is
2703 the value the module flag should be restricted to. This behavior can
2704 be used to restrict the allowable results (via triggering of an
2705 error) of linking IDs with the **Override** behavior.
2709 Uses the specified value, regardless of the behavior or value of the
2710 other module. If both modules specify **Override**, but the values
2711 differ, an error will be emitted.
2715 Appends the two values, which are required to be metadata nodes.
2719 Appends the two values, which are required to be metadata
2720 nodes. However, duplicate entries in the second list are dropped
2721 during the append operation.
2723 It is an error for a particular unique flag ID to have multiple behaviors,
2724 except in the case of **Require** (which adds restrictions on another metadata
2725 value) or **Override**.
2727 An example of module flags:
2729 .. code-block:: llvm
2731 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2732 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2733 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2734 !3 = metadata !{ i32 3, metadata !"qux",
2736 metadata !"foo", i32 1
2739 !llvm.module.flags = !{ !0, !1, !2, !3 }
2741 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2742 if two or more ``!"foo"`` flags are seen is to emit an error if their
2743 values are not equal.
2745 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2746 behavior if two or more ``!"bar"`` flags are seen is to use the value
2749 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2750 behavior if two or more ``!"qux"`` flags are seen is to emit a
2751 warning if their values are not equal.
2753 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2757 metadata !{ metadata !"foo", i32 1 }
2759 The behavior is to emit an error if the ``llvm.module.flags`` does not
2760 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2763 Objective-C Garbage Collection Module Flags Metadata
2764 ----------------------------------------------------
2766 On the Mach-O platform, Objective-C stores metadata about garbage
2767 collection in a special section called "image info". The metadata
2768 consists of a version number and a bitmask specifying what types of
2769 garbage collection are supported (if any) by the file. If two or more
2770 modules are linked together their garbage collection metadata needs to
2771 be merged rather than appended together.
2773 The Objective-C garbage collection module flags metadata consists of the
2774 following key-value pairs:
2783 * - ``Objective-C Version``
2784 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2786 * - ``Objective-C Image Info Version``
2787 - **[Required]** --- The version of the image info section. Currently
2790 * - ``Objective-C Image Info Section``
2791 - **[Required]** --- The section to place the metadata. Valid values are
2792 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2793 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2794 Objective-C ABI version 2.
2796 * - ``Objective-C Garbage Collection``
2797 - **[Required]** --- Specifies whether garbage collection is supported or
2798 not. Valid values are 0, for no garbage collection, and 2, for garbage
2799 collection supported.
2801 * - ``Objective-C GC Only``
2802 - **[Optional]** --- Specifies that only garbage collection is supported.
2803 If present, its value must be 6. This flag requires that the
2804 ``Objective-C Garbage Collection`` flag have the value 2.
2806 Some important flag interactions:
2808 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2809 merged with a module with ``Objective-C Garbage Collection`` set to
2810 2, then the resulting module has the
2811 ``Objective-C Garbage Collection`` flag set to 0.
2812 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2813 merged with a module with ``Objective-C GC Only`` set to 6.
2815 Automatic Linker Flags Module Flags Metadata
2816 --------------------------------------------
2818 Some targets support embedding flags to the linker inside individual object
2819 files. Typically this is used in conjunction with language extensions which
2820 allow source files to explicitly declare the libraries they depend on, and have
2821 these automatically be transmitted to the linker via object files.
2823 These flags are encoded in the IR using metadata in the module flags section,
2824 using the ``Linker Options`` key. The merge behavior for this flag is required
2825 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2826 node which should be a list of other metadata nodes, each of which should be a
2827 list of metadata strings defining linker options.
2829 For example, the following metadata section specifies two separate sets of
2830 linker options, presumably to link against ``libz`` and the ``Cocoa``
2833 !0 = metadata !{ i32 6, metadata !"Linker Options",
2835 metadata !{ metadata !"-lz" },
2836 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2837 !llvm.module.flags = !{ !0 }
2839 The metadata encoding as lists of lists of options, as opposed to a collapsed
2840 list of options, is chosen so that the IR encoding can use multiple option
2841 strings to specify e.g., a single library, while still having that specifier be
2842 preserved as an atomic element that can be recognized by a target specific
2843 assembly writer or object file emitter.
2845 Each individual option is required to be either a valid option for the target's
2846 linker, or an option that is reserved by the target specific assembly writer or
2847 object file emitter. No other aspect of these options is defined by the IR.
2849 Intrinsic Global Variables
2850 ==========================
2852 LLVM has a number of "magic" global variables that contain data that
2853 affect code generation or other IR semantics. These are documented here.
2854 All globals of this sort should have a section specified as
2855 "``llvm.metadata``". This section and all globals that start with
2856 "``llvm.``" are reserved for use by LLVM.
2858 The '``llvm.used``' Global Variable
2859 -----------------------------------
2861 The ``@llvm.used`` global is an array with i8\* element type which has
2862 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2863 pointers to global variables and functions which may optionally have a
2864 pointer cast formed of bitcast or getelementptr. For example, a legal
2867 .. code-block:: llvm
2872 @llvm.used = appending global [2 x i8*] [
2874 i8* bitcast (i32* @Y to i8*)
2875 ], section "llvm.metadata"
2877 If a global variable appears in the ``@llvm.used`` list, then the
2878 compiler, assembler, and linker are required to treat the symbol as if
2879 there is a reference to the global that it cannot see. For example, if a
2880 variable has internal linkage and no references other than that from the
2881 ``@llvm.used`` list, it cannot be deleted. This is commonly used to
2882 represent references from inline asms and other things the compiler
2883 cannot "see", and corresponds to "``attribute((used))``" in GNU C.
2885 On some targets, the code generator must emit a directive to the
2886 assembler or object file to prevent the assembler and linker from
2887 molesting the symbol.
2889 The '``llvm.compiler.used``' Global Variable
2890 --------------------------------------------
2892 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2893 directive, except that it only prevents the compiler from touching the
2894 symbol. On targets that support it, this allows an intelligent linker to
2895 optimize references to the symbol without being impeded as it would be
2898 This is a rare construct that should only be used in rare circumstances,
2899 and should not be exposed to source languages.
2901 The '``llvm.global_ctors``' Global Variable
2902 -------------------------------------------
2904 .. code-block:: llvm
2906 %0 = type { i32, void ()* }
2907 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2909 The ``@llvm.global_ctors`` array contains a list of constructor
2910 functions and associated priorities. The functions referenced by this
2911 array will be called in ascending order of priority (i.e. lowest first)
2912 when the module is loaded. The order of functions with the same priority
2915 The '``llvm.global_dtors``' Global Variable
2916 -------------------------------------------
2918 .. code-block:: llvm
2920 %0 = type { i32, void ()* }
2921 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2923 The ``@llvm.global_dtors`` array contains a list of destructor functions
2924 and associated priorities. The functions referenced by this array will
2925 be called in descending order of priority (i.e. highest first) when the
2926 module is loaded. The order of functions with the same priority is not
2929 Instruction Reference
2930 =====================
2932 The LLVM instruction set consists of several different classifications
2933 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
2934 instructions <binaryops>`, :ref:`bitwise binary
2935 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
2936 :ref:`other instructions <otherops>`.
2940 Terminator Instructions
2941 -----------------------
2943 As mentioned :ref:`previously <functionstructure>`, every basic block in a
2944 program ends with a "Terminator" instruction, which indicates which
2945 block should be executed after the current block is finished. These
2946 terminator instructions typically yield a '``void``' value: they produce
2947 control flow, not values (the one exception being the
2948 ':ref:`invoke <i_invoke>`' instruction).
2950 The terminator instructions are: ':ref:`ret <i_ret>`',
2951 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
2952 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
2953 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
2957 '``ret``' Instruction
2958 ^^^^^^^^^^^^^^^^^^^^^
2965 ret <type> <value> ; Return a value from a non-void function
2966 ret void ; Return from void function
2971 The '``ret``' instruction is used to return control flow (and optionally
2972 a value) from a function back to the caller.
2974 There are two forms of the '``ret``' instruction: one that returns a
2975 value and then causes control flow, and one that just causes control
2981 The '``ret``' instruction optionally accepts a single argument, the
2982 return value. The type of the return value must be a ':ref:`first
2983 class <t_firstclass>`' type.
2985 A function is not :ref:`well formed <wellformed>` if it it has a non-void
2986 return type and contains a '``ret``' instruction with no return value or
2987 a return value with a type that does not match its type, or if it has a
2988 void return type and contains a '``ret``' instruction with a return
2994 When the '``ret``' instruction is executed, control flow returns back to
2995 the calling function's context. If the caller is a
2996 ":ref:`call <i_call>`" instruction, execution continues at the
2997 instruction after the call. If the caller was an
2998 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
2999 beginning of the "normal" destination block. If the instruction returns
3000 a value, that value shall set the call or invoke instruction's return
3006 .. code-block:: llvm
3008 ret i32 5 ; Return an integer value of 5
3009 ret void ; Return from a void function
3010 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3014 '``br``' Instruction
3015 ^^^^^^^^^^^^^^^^^^^^
3022 br i1 <cond>, label <iftrue>, label <iffalse>
3023 br label <dest> ; Unconditional branch
3028 The '``br``' instruction is used to cause control flow to transfer to a
3029 different basic block in the current function. There are two forms of
3030 this instruction, corresponding to a conditional branch and an
3031 unconditional branch.
3036 The conditional branch form of the '``br``' instruction takes a single
3037 '``i1``' value and two '``label``' values. The unconditional form of the
3038 '``br``' instruction takes a single '``label``' value as a target.
3043 Upon execution of a conditional '``br``' instruction, the '``i1``'
3044 argument is evaluated. If the value is ``true``, control flows to the
3045 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3046 to the '``iffalse``' ``label`` argument.
3051 .. code-block:: llvm
3054 %cond = icmp eq i32 %a, %b
3055 br i1 %cond, label %IfEqual, label %IfUnequal
3063 '``switch``' Instruction
3064 ^^^^^^^^^^^^^^^^^^^^^^^^
3071 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3076 The '``switch``' instruction is used to transfer control flow to one of
3077 several different places. It is a generalization of the '``br``'
3078 instruction, allowing a branch to occur to one of many possible
3084 The '``switch``' instruction uses three parameters: an integer
3085 comparison value '``value``', a default '``label``' destination, and an
3086 array of pairs of comparison value constants and '``label``'s. The table
3087 is not allowed to contain duplicate constant entries.
3092 The ``switch`` instruction specifies a table of values and destinations.
3093 When the '``switch``' instruction is executed, this table is searched
3094 for the given value. If the value is found, control flow is transferred
3095 to the corresponding destination; otherwise, control flow is transferred
3096 to the default destination.
3101 Depending on properties of the target machine and the particular
3102 ``switch`` instruction, this instruction may be code generated in
3103 different ways. For example, it could be generated as a series of
3104 chained conditional branches or with a lookup table.
3109 .. code-block:: llvm
3111 ; Emulate a conditional br instruction
3112 %Val = zext i1 %value to i32
3113 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3115 ; Emulate an unconditional br instruction
3116 switch i32 0, label %dest [ ]
3118 ; Implement a jump table:
3119 switch i32 %val, label %otherwise [ i32 0, label %onzero
3121 i32 2, label %ontwo ]
3125 '``indirectbr``' Instruction
3126 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3133 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3138 The '``indirectbr``' instruction implements an indirect branch to a
3139 label within the current function, whose address is specified by
3140 "``address``". Address must be derived from a
3141 :ref:`blockaddress <blockaddress>` constant.
3146 The '``address``' argument is the address of the label to jump to. The
3147 rest of the arguments indicate the full set of possible destinations
3148 that the address may point to. Blocks are allowed to occur multiple
3149 times in the destination list, though this isn't particularly useful.
3151 This destination list is required so that dataflow analysis has an
3152 accurate understanding of the CFG.
3157 Control transfers to the block specified in the address argument. All
3158 possible destination blocks must be listed in the label list, otherwise
3159 this instruction has undefined behavior. This implies that jumps to
3160 labels defined in other functions have undefined behavior as well.
3165 This is typically implemented with a jump through a register.
3170 .. code-block:: llvm
3172 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3176 '``invoke``' Instruction
3177 ^^^^^^^^^^^^^^^^^^^^^^^^
3184 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3185 to label <normal label> unwind label <exception label>
3190 The '``invoke``' instruction causes control to transfer to a specified
3191 function, with the possibility of control flow transfer to either the
3192 '``normal``' label or the '``exception``' label. If the callee function
3193 returns with the "``ret``" instruction, control flow will return to the
3194 "normal" label. If the callee (or any indirect callees) returns via the
3195 ":ref:`resume <i_resume>`" instruction or other exception handling
3196 mechanism, control is interrupted and continued at the dynamically
3197 nearest "exception" label.
3199 The '``exception``' label is a `landing
3200 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3201 '``exception``' label is required to have the
3202 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3203 information about the behavior of the program after unwinding happens,
3204 as its first non-PHI instruction. The restrictions on the
3205 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3206 instruction, so that the important information contained within the
3207 "``landingpad``" instruction can't be lost through normal code motion.
3212 This instruction requires several arguments:
3214 #. The optional "cconv" marker indicates which :ref:`calling
3215 convention <callingconv>` the call should use. If none is
3216 specified, the call defaults to using C calling conventions.
3217 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3218 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3220 #. '``ptr to function ty``': shall be the signature of the pointer to
3221 function value being invoked. In most cases, this is a direct
3222 function invocation, but indirect ``invoke``'s are just as possible,
3223 branching off an arbitrary pointer to function value.
3224 #. '``function ptr val``': An LLVM value containing a pointer to a
3225 function to be invoked.
3226 #. '``function args``': argument list whose types match the function
3227 signature argument types and parameter attributes. All arguments must
3228 be of :ref:`first class <t_firstclass>` type. If the function signature
3229 indicates the function accepts a variable number of arguments, the
3230 extra arguments can be specified.
3231 #. '``normal label``': the label reached when the called function
3232 executes a '``ret``' instruction.
3233 #. '``exception label``': the label reached when a callee returns via
3234 the :ref:`resume <i_resume>` instruction or other exception handling
3236 #. The optional :ref:`function attributes <fnattrs>` list. Only
3237 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3238 attributes are valid here.
3243 This instruction is designed to operate as a standard '``call``'
3244 instruction in most regards. The primary difference is that it
3245 establishes an association with a label, which is used by the runtime
3246 library to unwind the stack.
3248 This instruction is used in languages with destructors to ensure that
3249 proper cleanup is performed in the case of either a ``longjmp`` or a
3250 thrown exception. Additionally, this is important for implementation of
3251 '``catch``' clauses in high-level languages that support them.
3253 For the purposes of the SSA form, the definition of the value returned
3254 by the '``invoke``' instruction is deemed to occur on the edge from the
3255 current block to the "normal" label. If the callee unwinds then no
3256 return value is available.
3261 .. code-block:: llvm
3263 %retval = invoke i32 @Test(i32 15) to label %Continue
3264 unwind label %TestCleanup ; {i32}:retval set
3265 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3266 unwind label %TestCleanup ; {i32}:retval set
3270 '``resume``' Instruction
3271 ^^^^^^^^^^^^^^^^^^^^^^^^
3278 resume <type> <value>
3283 The '``resume``' instruction is a terminator instruction that has no
3289 The '``resume``' instruction requires one argument, which must have the
3290 same type as the result of any '``landingpad``' instruction in the same
3296 The '``resume``' instruction resumes propagation of an existing
3297 (in-flight) exception whose unwinding was interrupted with a
3298 :ref:`landingpad <i_landingpad>` instruction.
3303 .. code-block:: llvm
3305 resume { i8*, i32 } %exn
3309 '``unreachable``' Instruction
3310 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3322 The '``unreachable``' instruction has no defined semantics. This
3323 instruction is used to inform the optimizer that a particular portion of
3324 the code is not reachable. This can be used to indicate that the code
3325 after a no-return function cannot be reached, and other facts.
3330 The '``unreachable``' instruction has no defined semantics.
3337 Binary operators are used to do most of the computation in a program.
3338 They require two operands of the same type, execute an operation on
3339 them, and produce a single value. The operands might represent multiple
3340 data, as is the case with the :ref:`vector <t_vector>` data type. The
3341 result value has the same type as its operands.
3343 There are several different binary operators:
3347 '``add``' Instruction
3348 ^^^^^^^^^^^^^^^^^^^^^
3355 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3356 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3357 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3358 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3363 The '``add``' instruction returns the sum of its two operands.
3368 The two arguments to the '``add``' instruction must be
3369 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3370 arguments must have identical types.
3375 The value produced is the integer sum of the two operands.
3377 If the sum has unsigned overflow, the result returned is the
3378 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3381 Because LLVM integers use a two's complement representation, this
3382 instruction is appropriate for both signed and unsigned integers.
3384 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3385 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3386 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3387 unsigned and/or signed overflow, respectively, occurs.
3392 .. code-block:: llvm
3394 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3398 '``fadd``' Instruction
3399 ^^^^^^^^^^^^^^^^^^^^^^
3406 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3411 The '``fadd``' instruction returns the sum of its two operands.
3416 The two arguments to the '``fadd``' instruction must be :ref:`floating
3417 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3418 Both arguments must have identical types.
3423 The value produced is the floating point sum of the two operands. This
3424 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3425 which are optimization hints to enable otherwise unsafe floating point
3431 .. code-block:: llvm
3433 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3435 '``sub``' Instruction
3436 ^^^^^^^^^^^^^^^^^^^^^
3443 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3444 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3445 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3446 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3451 The '``sub``' instruction returns the difference of its two operands.
3453 Note that the '``sub``' instruction is used to represent the '``neg``'
3454 instruction present in most other intermediate representations.
3459 The two arguments to the '``sub``' instruction must be
3460 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3461 arguments must have identical types.
3466 The value produced is the integer difference of the two operands.
3468 If the difference has unsigned overflow, the result returned is the
3469 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3472 Because LLVM integers use a two's complement representation, this
3473 instruction is appropriate for both signed and unsigned integers.
3475 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3476 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3477 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3478 unsigned and/or signed overflow, respectively, occurs.
3483 .. code-block:: llvm
3485 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3486 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3490 '``fsub``' Instruction
3491 ^^^^^^^^^^^^^^^^^^^^^^
3498 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3503 The '``fsub``' instruction returns the difference of its two operands.
3505 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3506 instruction present in most other intermediate representations.
3511 The two arguments to the '``fsub``' instruction must be :ref:`floating
3512 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3513 Both arguments must have identical types.
3518 The value produced is the floating point difference of the two operands.
3519 This instruction can also take any number of :ref:`fast-math
3520 flags <fastmath>`, which are optimization hints to enable otherwise
3521 unsafe floating point optimizations:
3526 .. code-block:: llvm
3528 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3529 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3531 '``mul``' Instruction
3532 ^^^^^^^^^^^^^^^^^^^^^
3539 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3540 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3541 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3542 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3547 The '``mul``' instruction returns the product of its two operands.
3552 The two arguments to the '``mul``' instruction must be
3553 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3554 arguments must have identical types.
3559 The value produced is the integer product of the two operands.
3561 If the result of the multiplication has unsigned overflow, the result
3562 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3563 bit width of the result.
3565 Because LLVM integers use a two's complement representation, and the
3566 result is the same width as the operands, this instruction returns the
3567 correct result for both signed and unsigned integers. If a full product
3568 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3569 sign-extended or zero-extended as appropriate to the width of the full
3572 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3573 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3574 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3575 unsigned and/or signed overflow, respectively, occurs.
3580 .. code-block:: llvm
3582 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3586 '``fmul``' Instruction
3587 ^^^^^^^^^^^^^^^^^^^^^^
3594 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3599 The '``fmul``' instruction returns the product of its two operands.
3604 The two arguments to the '``fmul``' instruction must be :ref:`floating
3605 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3606 Both arguments must have identical types.
3611 The value produced is the floating point product of the two operands.
3612 This instruction can also take any number of :ref:`fast-math
3613 flags <fastmath>`, which are optimization hints to enable otherwise
3614 unsafe floating point optimizations:
3619 .. code-block:: llvm
3621 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3623 '``udiv``' Instruction
3624 ^^^^^^^^^^^^^^^^^^^^^^
3631 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3632 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3637 The '``udiv``' instruction returns the quotient of its two operands.
3642 The two arguments to the '``udiv``' instruction must be
3643 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3644 arguments must have identical types.
3649 The value produced is the unsigned integer quotient of the two operands.
3651 Note that unsigned integer division and signed integer division are
3652 distinct operations; for signed integer division, use '``sdiv``'.
3654 Division by zero leads to undefined behavior.
3656 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3657 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3658 such, "((a udiv exact b) mul b) == a").
3663 .. code-block:: llvm
3665 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3667 '``sdiv``' Instruction
3668 ^^^^^^^^^^^^^^^^^^^^^^
3675 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3676 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3681 The '``sdiv``' instruction returns the quotient of its two operands.
3686 The two arguments to the '``sdiv``' instruction must be
3687 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3688 arguments must have identical types.
3693 The value produced is the signed integer quotient of the two operands
3694 rounded towards zero.
3696 Note that signed integer division and unsigned integer division are
3697 distinct operations; for unsigned integer division, use '``udiv``'.
3699 Division by zero leads to undefined behavior. Overflow also leads to
3700 undefined behavior; this is a rare case, but can occur, for example, by
3701 doing a 32-bit division of -2147483648 by -1.
3703 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3704 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3709 .. code-block:: llvm
3711 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3715 '``fdiv``' Instruction
3716 ^^^^^^^^^^^^^^^^^^^^^^
3723 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3728 The '``fdiv``' instruction returns the quotient of its two operands.
3733 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3734 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3735 Both arguments must have identical types.
3740 The value produced is the floating point quotient of the two operands.
3741 This instruction can also take any number of :ref:`fast-math
3742 flags <fastmath>`, which are optimization hints to enable otherwise
3743 unsafe floating point optimizations:
3748 .. code-block:: llvm
3750 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3752 '``urem``' Instruction
3753 ^^^^^^^^^^^^^^^^^^^^^^
3760 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3765 The '``urem``' instruction returns the remainder from the unsigned
3766 division of its two arguments.
3771 The two arguments to the '``urem``' instruction must be
3772 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3773 arguments must have identical types.
3778 This instruction returns the unsigned integer *remainder* of a division.
3779 This instruction always performs an unsigned division to get the
3782 Note that unsigned integer remainder and signed integer remainder are
3783 distinct operations; for signed integer remainder, use '``srem``'.
3785 Taking the remainder of a division by zero leads to undefined behavior.
3790 .. code-block:: llvm
3792 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3794 '``srem``' Instruction
3795 ^^^^^^^^^^^^^^^^^^^^^^
3802 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3807 The '``srem``' instruction returns the remainder from the signed
3808 division of its two operands. This instruction can also take
3809 :ref:`vector <t_vector>` versions of the values in which case the elements
3815 The two arguments to the '``srem``' instruction must be
3816 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3817 arguments must have identical types.
3822 This instruction returns the *remainder* of a division (where the result
3823 is either zero or has the same sign as the dividend, ``op1``), not the
3824 *modulo* operator (where the result is either zero or has the same sign
3825 as the divisor, ``op2``) of a value. For more information about the
3826 difference, see `The Math
3827 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3828 table of how this is implemented in various languages, please see
3830 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3832 Note that signed integer remainder and unsigned integer remainder are
3833 distinct operations; for unsigned integer remainder, use '``urem``'.
3835 Taking the remainder of a division by zero leads to undefined behavior.
3836 Overflow also leads to undefined behavior; this is a rare case, but can
3837 occur, for example, by taking the remainder of a 32-bit division of
3838 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3839 rule lets srem be implemented using instructions that return both the
3840 result of the division and the remainder.)
3845 .. code-block:: llvm
3847 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3851 '``frem``' Instruction
3852 ^^^^^^^^^^^^^^^^^^^^^^
3859 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3864 The '``frem``' instruction returns the remainder from the division of
3870 The two arguments to the '``frem``' instruction must be :ref:`floating
3871 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3872 Both arguments must have identical types.
3877 This instruction returns the *remainder* of a division. The remainder
3878 has the same sign as the dividend. This instruction can also take any
3879 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3880 to enable otherwise unsafe floating point optimizations:
3885 .. code-block:: llvm
3887 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3891 Bitwise Binary Operations
3892 -------------------------
3894 Bitwise binary operators are used to do various forms of bit-twiddling
3895 in a program. They are generally very efficient instructions and can
3896 commonly be strength reduced from other instructions. They require two
3897 operands of the same type, execute an operation on them, and produce a
3898 single value. The resulting value is the same type as its operands.
3900 '``shl``' Instruction
3901 ^^^^^^^^^^^^^^^^^^^^^
3908 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
3909 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
3910 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
3911 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3916 The '``shl``' instruction returns the first operand shifted to the left
3917 a specified number of bits.
3922 Both arguments to the '``shl``' instruction must be the same
3923 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3924 '``op2``' is treated as an unsigned value.
3929 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
3930 where ``n`` is the width of the result. If ``op2`` is (statically or
3931 dynamically) negative or equal to or larger than the number of bits in
3932 ``op1``, the result is undefined. If the arguments are vectors, each
3933 vector element of ``op1`` is shifted by the corresponding shift amount
3936 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
3937 value <poisonvalues>` if it shifts out any non-zero bits. If the
3938 ``nsw`` keyword is present, then the shift produces a :ref:`poison
3939 value <poisonvalues>` if it shifts out any bits that disagree with the
3940 resultant sign bit. As such, NUW/NSW have the same semantics as they
3941 would if the shift were expressed as a mul instruction with the same
3942 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
3947 .. code-block:: llvm
3949 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
3950 <result> = shl i32 4, 2 ; yields {i32}: 16
3951 <result> = shl i32 1, 10 ; yields {i32}: 1024
3952 <result> = shl i32 1, 32 ; undefined
3953 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
3955 '``lshr``' Instruction
3956 ^^^^^^^^^^^^^^^^^^^^^^
3963 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
3964 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
3969 The '``lshr``' instruction (logical shift right) returns the first
3970 operand shifted to the right a specified number of bits with zero fill.
3975 Both arguments to the '``lshr``' instruction must be the same
3976 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3977 '``op2``' is treated as an unsigned value.
3982 This instruction always performs a logical shift right operation. The
3983 most significant bits of the result will be filled with zero bits after
3984 the shift. If ``op2`` is (statically or dynamically) equal to or larger
3985 than the number of bits in ``op1``, the result is undefined. If the
3986 arguments are vectors, each vector element of ``op1`` is shifted by the
3987 corresponding shift amount in ``op2``.
3989 If the ``exact`` keyword is present, the result value of the ``lshr`` is
3990 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3996 .. code-block:: llvm
3998 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
3999 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4000 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4001 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7FFFFFFF
4002 <result> = lshr i32 1, 32 ; undefined
4003 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4005 '``ashr``' Instruction
4006 ^^^^^^^^^^^^^^^^^^^^^^
4013 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4014 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4019 The '``ashr``' instruction (arithmetic shift right) returns the first
4020 operand shifted to the right a specified number of bits with sign
4026 Both arguments to the '``ashr``' instruction must be the same
4027 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4028 '``op2``' is treated as an unsigned value.
4033 This instruction always performs an arithmetic shift right operation,
4034 The most significant bits of the result will be filled with the sign bit
4035 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4036 than the number of bits in ``op1``, the result is undefined. If the
4037 arguments are vectors, each vector element of ``op1`` is shifted by the
4038 corresponding shift amount in ``op2``.
4040 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4041 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4047 .. code-block:: llvm
4049 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4050 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4051 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4052 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4053 <result> = ashr i32 1, 32 ; undefined
4054 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4056 '``and``' Instruction
4057 ^^^^^^^^^^^^^^^^^^^^^
4064 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4069 The '``and``' instruction returns the bitwise logical and of its two
4075 The two arguments to the '``and``' instruction must be
4076 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4077 arguments must have identical types.
4082 The truth table used for the '``and``' instruction is:
4099 .. code-block:: llvm
4101 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4102 <result> = and i32 15, 40 ; yields {i32}:result = 8
4103 <result> = and i32 4, 8 ; yields {i32}:result = 0
4105 '``or``' Instruction
4106 ^^^^^^^^^^^^^^^^^^^^
4113 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4118 The '``or``' instruction returns the bitwise logical inclusive or of its
4124 The two arguments to the '``or``' instruction must be
4125 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4126 arguments must have identical types.
4131 The truth table used for the '``or``' instruction is:
4150 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4151 <result> = or i32 15, 40 ; yields {i32}:result = 47
4152 <result> = or i32 4, 8 ; yields {i32}:result = 12
4154 '``xor``' Instruction
4155 ^^^^^^^^^^^^^^^^^^^^^
4162 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4167 The '``xor``' instruction returns the bitwise logical exclusive or of
4168 its two operands. The ``xor`` is used to implement the "one's
4169 complement" operation, which is the "~" operator in C.
4174 The two arguments to the '``xor``' instruction must be
4175 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4176 arguments must have identical types.
4181 The truth table used for the '``xor``' instruction is:
4198 .. code-block:: llvm
4200 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4201 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4202 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4203 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4208 LLVM supports several instructions to represent vector operations in a
4209 target-independent manner. These instructions cover the element-access
4210 and vector-specific operations needed to process vectors effectively.
4211 While LLVM does directly support these vector operations, many
4212 sophisticated algorithms will want to use target-specific intrinsics to
4213 take full advantage of a specific target.
4215 .. _i_extractelement:
4217 '``extractelement``' Instruction
4218 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4225 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4230 The '``extractelement``' instruction extracts a single scalar element
4231 from a vector at a specified index.
4236 The first operand of an '``extractelement``' instruction is a value of
4237 :ref:`vector <t_vector>` type. The second operand is an index indicating
4238 the position from which to extract the element. The index may be a
4244 The result is a scalar of the same type as the element type of ``val``.
4245 Its value is the value at position ``idx`` of ``val``. If ``idx``
4246 exceeds the length of ``val``, the results are undefined.
4251 .. code-block:: llvm
4253 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4255 .. _i_insertelement:
4257 '``insertelement``' Instruction
4258 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4265 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4270 The '``insertelement``' instruction inserts a scalar element into a
4271 vector at a specified index.
4276 The first operand of an '``insertelement``' instruction is a value of
4277 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4278 type must equal the element type of the first operand. The third operand
4279 is an index indicating the position at which to insert the value. The
4280 index may be a variable.
4285 The result is a vector of the same type as ``val``. Its element values
4286 are those of ``val`` except at position ``idx``, where it gets the value
4287 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4293 .. code-block:: llvm
4295 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4297 .. _i_shufflevector:
4299 '``shufflevector``' Instruction
4300 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4307 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4312 The '``shufflevector``' instruction constructs a permutation of elements
4313 from two input vectors, returning a vector with the same element type as
4314 the input and length that is the same as the shuffle mask.
4319 The first two operands of a '``shufflevector``' instruction are vectors
4320 with the same type. The third argument is a shuffle mask whose element
4321 type is always 'i32'. The result of the instruction is a vector whose
4322 length is the same as the shuffle mask and whose element type is the
4323 same as the element type of the first two operands.
4325 The shuffle mask operand is required to be a constant vector with either
4326 constant integer or undef values.
4331 The elements of the two input vectors are numbered from left to right
4332 across both of the vectors. The shuffle mask operand specifies, for each
4333 element of the result vector, which element of the two input vectors the
4334 result element gets. The element selector may be undef (meaning "don't
4335 care") and the second operand may be undef if performing a shuffle from
4341 .. code-block:: llvm
4343 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4344 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4345 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4346 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4347 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4348 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4349 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4350 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4352 Aggregate Operations
4353 --------------------
4355 LLVM supports several instructions for working with
4356 :ref:`aggregate <t_aggregate>` values.
4360 '``extractvalue``' Instruction
4361 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4368 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4373 The '``extractvalue``' instruction extracts the value of a member field
4374 from an :ref:`aggregate <t_aggregate>` value.
4379 The first operand of an '``extractvalue``' instruction is a value of
4380 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4381 constant indices to specify which value to extract in a similar manner
4382 as indices in a '``getelementptr``' instruction.
4384 The major differences to ``getelementptr`` indexing are:
4386 - Since the value being indexed is not a pointer, the first index is
4387 omitted and assumed to be zero.
4388 - At least one index must be specified.
4389 - Not only struct indices but also array indices must be in bounds.
4394 The result is the value at the position in the aggregate specified by
4400 .. code-block:: llvm
4402 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4406 '``insertvalue``' Instruction
4407 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4414 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4419 The '``insertvalue``' instruction inserts a value into a member field in
4420 an :ref:`aggregate <t_aggregate>` value.
4425 The first operand of an '``insertvalue``' instruction is a value of
4426 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4427 a first-class value to insert. The following operands are constant
4428 indices indicating the position at which to insert the value in a
4429 similar manner as indices in a '``extractvalue``' instruction. The value
4430 to insert must have the same type as the value identified by the
4436 The result is an aggregate of the same type as ``val``. Its value is
4437 that of ``val`` except that the value at the position specified by the
4438 indices is that of ``elt``.
4443 .. code-block:: llvm
4445 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4446 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4447 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4451 Memory Access and Addressing Operations
4452 ---------------------------------------
4454 A key design point of an SSA-based representation is how it represents
4455 memory. In LLVM, no memory locations are in SSA form, which makes things
4456 very simple. This section describes how to read, write, and allocate
4461 '``alloca``' Instruction
4462 ^^^^^^^^^^^^^^^^^^^^^^^^
4469 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4474 The '``alloca``' instruction allocates memory on the stack frame of the
4475 currently executing function, to be automatically released when this
4476 function returns to its caller. The object is always allocated in the
4477 generic address space (address space zero).
4482 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4483 bytes of memory on the runtime stack, returning a pointer of the
4484 appropriate type to the program. If "NumElements" is specified, it is
4485 the number of elements allocated, otherwise "NumElements" is defaulted
4486 to be one. If a constant alignment is specified, the value result of the
4487 allocation is guaranteed to be aligned to at least that boundary. If not
4488 specified, or if zero, the target can choose to align the allocation on
4489 any convenient boundary compatible with the type.
4491 '``type``' may be any sized type.
4496 Memory is allocated; a pointer is returned. The operation is undefined
4497 if there is insufficient stack space for the allocation. '``alloca``'d
4498 memory is automatically released when the function returns. The
4499 '``alloca``' instruction is commonly used to represent automatic
4500 variables that must have an address available. When the function returns
4501 (either with the ``ret`` or ``resume`` instructions), the memory is
4502 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4503 The order in which memory is allocated (ie., which way the stack grows)
4509 .. code-block:: llvm
4511 %ptr = alloca i32 ; yields {i32*}:ptr
4512 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4513 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4514 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4518 '``load``' Instruction
4519 ^^^^^^^^^^^^^^^^^^^^^^
4526 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4527 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4528 !<index> = !{ i32 1 }
4533 The '``load``' instruction is used to read from memory.
4538 The argument to the ``load`` instruction specifies the memory address
4539 from which to load. The pointer must point to a :ref:`first
4540 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4541 then the optimizer is not allowed to modify the number or order of
4542 execution of this ``load`` with other :ref:`volatile
4543 operations <volatile>`.
4545 If the ``load`` is marked as ``atomic``, it takes an extra
4546 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4547 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4548 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4549 when they may see multiple atomic stores. The type of the pointee must
4550 be an integer type whose bit width is a power of two greater than or
4551 equal to eight and less than or equal to a target-specific size limit.
4552 ``align`` must be explicitly specified on atomic loads, and the load has
4553 undefined behavior if the alignment is not set to a value which is at
4554 least the size in bytes of the pointee. ``!nontemporal`` does not have
4555 any defined semantics for atomic loads.
4557 The optional constant ``align`` argument specifies the alignment of the
4558 operation (that is, the alignment of the memory address). A value of 0
4559 or an omitted ``align`` argument means that the operation has the ABI
4560 alignment for the target. It is the responsibility of the code emitter
4561 to ensure that the alignment information is correct. Overestimating the
4562 alignment results in undefined behavior. Underestimating the alignment
4563 may produce less efficient code. An alignment of 1 is always safe.
4565 The optional ``!nontemporal`` metadata must reference a single
4566 metatadata name ``<index>`` corresponding to a metadata node with one
4567 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4568 metatadata on the instruction tells the optimizer and code generator
4569 that this load is not expected to be reused in the cache. The code
4570 generator may select special instructions to save cache bandwidth, such
4571 as the ``MOVNT`` instruction on x86.
4573 The optional ``!invariant.load`` metadata must reference a single
4574 metatadata name ``<index>`` corresponding to a metadata node with no
4575 entries. The existence of the ``!invariant.load`` metatadata on the
4576 instruction tells the optimizer and code generator that this load
4577 address points to memory which does not change value during program
4578 execution. The optimizer may then move this load around, for example, by
4579 hoisting it out of loops using loop invariant code motion.
4584 The location of memory pointed to is loaded. If the value being loaded
4585 is of scalar type then the number of bytes read does not exceed the
4586 minimum number of bytes needed to hold all bits of the type. For
4587 example, loading an ``i24`` reads at most three bytes. When loading a
4588 value of a type like ``i20`` with a size that is not an integral number
4589 of bytes, the result is undefined if the value was not originally
4590 written using a store of the same type.
4595 .. code-block:: llvm
4597 %ptr = alloca i32 ; yields {i32*}:ptr
4598 store i32 3, i32* %ptr ; yields {void}
4599 %val = load i32* %ptr ; yields {i32}:val = i32 3
4603 '``store``' Instruction
4604 ^^^^^^^^^^^^^^^^^^^^^^^
4611 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4612 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4617 The '``store``' instruction is used to write to memory.
4622 There are two arguments to the ``store`` instruction: a value to store
4623 and an address at which to store it. The type of the ``<pointer>``
4624 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4625 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4626 then the optimizer is not allowed to modify the number or order of
4627 execution of this ``store`` with other :ref:`volatile
4628 operations <volatile>`.
4630 If the ``store`` is marked as ``atomic``, it takes an extra
4631 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4632 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4633 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4634 when they may see multiple atomic stores. The type of the pointee must
4635 be an integer type whose bit width is a power of two greater than or
4636 equal to eight and less than or equal to a target-specific size limit.
4637 ``align`` must be explicitly specified on atomic stores, and the store
4638 has undefined behavior if the alignment is not set to a value which is
4639 at least the size in bytes of the pointee. ``!nontemporal`` does not
4640 have any defined semantics for atomic stores.
4642 The optional constant ``align`` argument specifies the alignment of the
4643 operation (that is, the alignment of the memory address). A value of 0
4644 or an omitted ``align`` argument means that the operation has the ABI
4645 alignment for the target. It is the responsibility of the code emitter
4646 to ensure that the alignment information is correct. Overestimating the
4647 alignment results in undefined behavior. Underestimating the
4648 alignment may produce less efficient code. An alignment of 1 is always
4651 The optional ``!nontemporal`` metadata must reference a single metatadata
4652 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4653 value 1. The existence of the ``!nontemporal`` metatadata on the instruction
4654 tells the optimizer and code generator that this load is not expected to
4655 be reused in the cache. The code generator may select special
4656 instructions to save cache bandwidth, such as the MOVNT instruction on
4662 The contents of memory are updated to contain ``<value>`` at the
4663 location specified by the ``<pointer>`` operand. If ``<value>`` is
4664 of scalar type then the number of bytes written does not exceed the
4665 minimum number of bytes needed to hold all bits of the type. For
4666 example, storing an ``i24`` writes at most three bytes. When writing a
4667 value of a type like ``i20`` with a size that is not an integral number
4668 of bytes, it is unspecified what happens to the extra bits that do not
4669 belong to the type, but they will typically be overwritten.
4674 .. code-block:: llvm
4676 %ptr = alloca i32 ; yields {i32*}:ptr
4677 store i32 3, i32* %ptr ; yields {void}
4678 %val = load i32* %ptr ; yields {i32}:val = i32 3
4682 '``fence``' Instruction
4683 ^^^^^^^^^^^^^^^^^^^^^^^
4690 fence [singlethread] <ordering> ; yields {void}
4695 The '``fence``' instruction is used to introduce happens-before edges
4701 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4702 defines what *synchronizes-with* edges they add. They can only be given
4703 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4708 A fence A which has (at least) ``release`` ordering semantics
4709 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4710 semantics if and only if there exist atomic operations X and Y, both
4711 operating on some atomic object M, such that A is sequenced before X, X
4712 modifies M (either directly or through some side effect of a sequence
4713 headed by X), Y is sequenced before B, and Y observes M. This provides a
4714 *happens-before* dependency between A and B. Rather than an explicit
4715 ``fence``, one (but not both) of the atomic operations X or Y might
4716 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4717 still *synchronize-with* the explicit ``fence`` and establish the
4718 *happens-before* edge.
4720 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4721 ``acquire`` and ``release`` semantics specified above, participates in
4722 the global program order of other ``seq_cst`` operations and/or fences.
4724 The optional ":ref:`singlethread <singlethread>`" argument specifies
4725 that the fence only synchronizes with other fences in the same thread.
4726 (This is useful for interacting with signal handlers.)
4731 .. code-block:: llvm
4733 fence acquire ; yields {void}
4734 fence singlethread seq_cst ; yields {void}
4738 '``cmpxchg``' Instruction
4739 ^^^^^^^^^^^^^^^^^^^^^^^^^
4746 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4751 The '``cmpxchg``' instruction is used to atomically modify memory. It
4752 loads a value in memory and compares it to a given value. If they are
4753 equal, it stores a new value into the memory.
4758 There are three arguments to the '``cmpxchg``' instruction: an address
4759 to operate on, a value to compare to the value currently be at that
4760 address, and a new value to place at that address if the compared values
4761 are equal. The type of '<cmp>' must be an integer type whose bit width
4762 is a power of two greater than or equal to eight and less than or equal
4763 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4764 type, and the type of '<pointer>' must be a pointer to that type. If the
4765 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4766 to modify the number or order of execution of this ``cmpxchg`` with
4767 other :ref:`volatile operations <volatile>`.
4769 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4770 synchronizes with other atomic operations.
4772 The optional "``singlethread``" argument declares that the ``cmpxchg``
4773 is only atomic with respect to code (usually signal handlers) running in
4774 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4775 respect to all other code in the system.
4777 The pointer passed into cmpxchg must have alignment greater than or
4778 equal to the size in memory of the operand.
4783 The contents of memory at the location specified by the '``<pointer>``'
4784 operand is read and compared to '``<cmp>``'; if the read value is the
4785 equal, '``<new>``' is written. The original value at the location is
4788 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4789 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4790 atomic load with an ordering parameter determined by dropping any
4791 ``release`` part of the ``cmpxchg``'s ordering.
4796 .. code-block:: llvm
4799 %orig = atomic load i32* %ptr unordered ; yields {i32}
4803 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4804 %squared = mul i32 %cmp, %cmp
4805 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4806 %success = icmp eq i32 %cmp, %old
4807 br i1 %success, label %done, label %loop
4814 '``atomicrmw``' Instruction
4815 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4822 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4827 The '``atomicrmw``' instruction is used to atomically modify memory.
4832 There are three arguments to the '``atomicrmw``' instruction: an
4833 operation to apply, an address whose value to modify, an argument to the
4834 operation. The operation must be one of the following keywords:
4848 The type of '<value>' must be an integer type whose bit width is a power
4849 of two greater than or equal to eight and less than or equal to a
4850 target-specific size limit. The type of the '``<pointer>``' operand must
4851 be a pointer to that type. If the ``atomicrmw`` is marked as
4852 ``volatile``, then the optimizer is not allowed to modify the number or
4853 order of execution of this ``atomicrmw`` with other :ref:`volatile
4854 operations <volatile>`.
4859 The contents of memory at the location specified by the '``<pointer>``'
4860 operand are atomically read, modified, and written back. The original
4861 value at the location is returned. The modification is specified by the
4864 - xchg: ``*ptr = val``
4865 - add: ``*ptr = *ptr + val``
4866 - sub: ``*ptr = *ptr - val``
4867 - and: ``*ptr = *ptr & val``
4868 - nand: ``*ptr = ~(*ptr & val)``
4869 - or: ``*ptr = *ptr | val``
4870 - xor: ``*ptr = *ptr ^ val``
4871 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4872 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4873 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4875 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4881 .. code-block:: llvm
4883 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4885 .. _i_getelementptr:
4887 '``getelementptr``' Instruction
4888 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4895 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4896 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4897 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
4902 The '``getelementptr``' instruction is used to get the address of a
4903 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
4904 address calculation only and does not access memory.
4909 The first argument is always a pointer or a vector of pointers, and
4910 forms the basis of the calculation. The remaining arguments are indices
4911 that indicate which of the elements of the aggregate object are indexed.
4912 The interpretation of each index is dependent on the type being indexed
4913 into. The first index always indexes the pointer value given as the
4914 first argument, the second index indexes a value of the type pointed to
4915 (not necessarily the value directly pointed to, since the first index
4916 can be non-zero), etc. The first type indexed into must be a pointer
4917 value, subsequent types can be arrays, vectors, and structs. Note that
4918 subsequent types being indexed into can never be pointers, since that
4919 would require loading the pointer before continuing calculation.
4921 The type of each index argument depends on the type it is indexing into.
4922 When indexing into a (optionally packed) structure, only ``i32`` integer
4923 **constants** are allowed (when using a vector of indices they must all
4924 be the **same** ``i32`` integer constant). When indexing into an array,
4925 pointer or vector, integers of any width are allowed, and they are not
4926 required to be constant. These integers are treated as signed values
4929 For example, let's consider a C code fragment and how it gets compiled
4945 int *foo(struct ST *s) {
4946 return &s[1].Z.B[5][13];
4949 The LLVM code generated by Clang is:
4951 .. code-block:: llvm
4953 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
4954 %struct.ST = type { i32, double, %struct.RT }
4956 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
4958 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
4965 In the example above, the first index is indexing into the
4966 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
4967 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
4968 indexes into the third element of the structure, yielding a
4969 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
4970 structure. The third index indexes into the second element of the
4971 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
4972 dimensions of the array are subscripted into, yielding an '``i32``'
4973 type. The '``getelementptr``' instruction returns a pointer to this
4974 element, thus computing a value of '``i32*``' type.
4976 Note that it is perfectly legal to index partially through a structure,
4977 returning a pointer to an inner element. Because of this, the LLVM code
4978 for the given testcase is equivalent to:
4980 .. code-block:: llvm
4982 define i32* @foo(%struct.ST* %s) {
4983 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
4984 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
4985 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
4986 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
4987 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
4991 If the ``inbounds`` keyword is present, the result value of the
4992 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
4993 pointer is not an *in bounds* address of an allocated object, or if any
4994 of the addresses that would be formed by successive addition of the
4995 offsets implied by the indices to the base address with infinitely
4996 precise signed arithmetic are not an *in bounds* address of that
4997 allocated object. The *in bounds* addresses for an allocated object are
4998 all the addresses that point into the object, plus the address one byte
4999 past the end. In cases where the base is a vector of pointers the
5000 ``inbounds`` keyword applies to each of the computations element-wise.
5002 If the ``inbounds`` keyword is not present, the offsets are added to the
5003 base address with silently-wrapping two's complement arithmetic. If the
5004 offsets have a different width from the pointer, they are sign-extended
5005 or truncated to the width of the pointer. The result value of the
5006 ``getelementptr`` may be outside the object pointed to by the base
5007 pointer. The result value may not necessarily be used to access memory
5008 though, even if it happens to point into allocated storage. See the
5009 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5012 The getelementptr instruction is often confusing. For some more insight
5013 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5018 .. code-block:: llvm
5020 ; yields [12 x i8]*:aptr
5021 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5023 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5025 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5027 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5029 In cases where the pointer argument is a vector of pointers, each index
5030 must be a vector with the same number of elements. For example:
5032 .. code-block:: llvm
5034 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5036 Conversion Operations
5037 ---------------------
5039 The instructions in this category are the conversion instructions
5040 (casting) which all take a single operand and a type. They perform
5041 various bit conversions on the operand.
5043 '``trunc .. to``' Instruction
5044 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5051 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5056 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5061 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5062 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5063 of the same number of integers. The bit size of the ``value`` must be
5064 larger than the bit size of the destination type, ``ty2``. Equal sized
5065 types are not allowed.
5070 The '``trunc``' instruction truncates the high order bits in ``value``
5071 and converts the remaining bits to ``ty2``. Since the source size must
5072 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5073 It will always truncate bits.
5078 .. code-block:: llvm
5080 %X = trunc i32 257 to i8 ; yields i8:1
5081 %Y = trunc i32 123 to i1 ; yields i1:true
5082 %Z = trunc i32 122 to i1 ; yields i1:false
5083 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5085 '``zext .. to``' Instruction
5086 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5093 <result> = zext <ty> <value> to <ty2> ; yields ty2
5098 The '``zext``' instruction zero extends its operand to type ``ty2``.
5103 The '``zext``' instruction takes a value to cast, and a type to cast it
5104 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5105 the same number of integers. The bit size of the ``value`` must be
5106 smaller than the bit size of the destination type, ``ty2``.
5111 The ``zext`` fills the high order bits of the ``value`` with zero bits
5112 until it reaches the size of the destination type, ``ty2``.
5114 When zero extending from i1, the result will always be either 0 or 1.
5119 .. code-block:: llvm
5121 %X = zext i32 257 to i64 ; yields i64:257
5122 %Y = zext i1 true to i32 ; yields i32:1
5123 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5125 '``sext .. to``' Instruction
5126 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5133 <result> = sext <ty> <value> to <ty2> ; yields ty2
5138 The '``sext``' sign extends ``value`` to the type ``ty2``.
5143 The '``sext``' instruction takes a value to cast, and a type to cast it
5144 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5145 the same number of integers. The bit size of the ``value`` must be
5146 smaller than the bit size of the destination type, ``ty2``.
5151 The '``sext``' instruction performs a sign extension by copying the sign
5152 bit (highest order bit) of the ``value`` until it reaches the bit size
5153 of the type ``ty2``.
5155 When sign extending from i1, the extension always results in -1 or 0.
5160 .. code-block:: llvm
5162 %X = sext i8 -1 to i16 ; yields i16 :65535
5163 %Y = sext i1 true to i32 ; yields i32:-1
5164 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5166 '``fptrunc .. to``' Instruction
5167 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5174 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5179 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5184 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5185 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5186 The size of ``value`` must be larger than the size of ``ty2``. This
5187 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5192 The '``fptrunc``' instruction truncates a ``value`` from a larger
5193 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5194 point <t_floating>` type. If the value cannot fit within the
5195 destination type, ``ty2``, then the results are undefined.
5200 .. code-block:: llvm
5202 %X = fptrunc double 123.0 to float ; yields float:123.0
5203 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5205 '``fpext .. to``' Instruction
5206 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5213 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5218 The '``fpext``' extends a floating point ``value`` to a larger floating
5224 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5225 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5226 to. The source type must be smaller than the destination type.
5231 The '``fpext``' instruction extends the ``value`` from a smaller
5232 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5233 point <t_floating>` type. The ``fpext`` cannot be used to make a
5234 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5235 *no-op cast* for a floating point cast.
5240 .. code-block:: llvm
5242 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5243 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5245 '``fptoui .. to``' Instruction
5246 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5253 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5258 The '``fptoui``' converts a floating point ``value`` to its unsigned
5259 integer equivalent of type ``ty2``.
5264 The '``fptoui``' instruction takes a value to cast, which must be a
5265 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5266 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5267 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5268 type with the same number of elements as ``ty``
5273 The '``fptoui``' instruction converts its :ref:`floating
5274 point <t_floating>` operand into the nearest (rounding towards zero)
5275 unsigned integer value. If the value cannot fit in ``ty2``, the results
5281 .. code-block:: llvm
5283 %X = fptoui double 123.0 to i32 ; yields i32:123
5284 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5285 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5287 '``fptosi .. to``' Instruction
5288 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5295 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5300 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5301 ``value`` to type ``ty2``.
5306 The '``fptosi``' instruction takes a value to cast, which must be a
5307 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5308 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5309 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5310 type with the same number of elements as ``ty``
5315 The '``fptosi``' instruction converts its :ref:`floating
5316 point <t_floating>` operand into the nearest (rounding towards zero)
5317 signed integer value. If the value cannot fit in ``ty2``, the results
5323 .. code-block:: llvm
5325 %X = fptosi double -123.0 to i32 ; yields i32:-123
5326 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5327 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5329 '``uitofp .. to``' Instruction
5330 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5337 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5342 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5343 and converts that value to the ``ty2`` type.
5348 The '``uitofp``' instruction takes a value to cast, which must be a
5349 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5350 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5351 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5352 type with the same number of elements as ``ty``
5357 The '``uitofp``' instruction interprets its operand as an unsigned
5358 integer quantity and converts it to the corresponding floating point
5359 value. If the value cannot fit in the floating point value, the results
5365 .. code-block:: llvm
5367 %X = uitofp i32 257 to float ; yields float:257.0
5368 %Y = uitofp i8 -1 to double ; yields double:255.0
5370 '``sitofp .. to``' Instruction
5371 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5378 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5383 The '``sitofp``' instruction regards ``value`` as a signed integer and
5384 converts that value to the ``ty2`` type.
5389 The '``sitofp``' instruction takes a value to cast, which must be a
5390 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5391 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5392 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5393 type with the same number of elements as ``ty``
5398 The '``sitofp``' instruction interprets its operand as a signed integer
5399 quantity and converts it to the corresponding floating point value. If
5400 the value cannot fit in the floating point value, the results are
5406 .. code-block:: llvm
5408 %X = sitofp i32 257 to float ; yields float:257.0
5409 %Y = sitofp i8 -1 to double ; yields double:-1.0
5413 '``ptrtoint .. to``' Instruction
5414 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5421 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5426 The '``ptrtoint``' instruction converts the pointer or a vector of
5427 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5432 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5433 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5434 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5435 a vector of integers type.
5440 The '``ptrtoint``' instruction converts ``value`` to integer type
5441 ``ty2`` by interpreting the pointer value as an integer and either
5442 truncating or zero extending that value to the size of the integer type.
5443 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5444 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5445 the same size, then nothing is done (*no-op cast*) other than a type
5451 .. code-block:: llvm
5453 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5454 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5455 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5459 '``inttoptr .. to``' Instruction
5460 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5467 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5472 The '``inttoptr``' instruction converts an integer ``value`` to a
5473 pointer type, ``ty2``.
5478 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5479 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5485 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5486 applying either a zero extension or a truncation depending on the size
5487 of the integer ``value``. If ``value`` is larger than the size of a
5488 pointer then a truncation is done. If ``value`` is smaller than the size
5489 of a pointer then a zero extension is done. If they are the same size,
5490 nothing is done (*no-op cast*).
5495 .. code-block:: llvm
5497 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5498 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5499 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5500 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5504 '``bitcast .. to``' Instruction
5505 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5512 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5517 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5523 The '``bitcast``' instruction takes a value to cast, which must be a
5524 non-aggregate first class value, and a type to cast it to, which must
5525 also be a non-aggregate :ref:`first class <t_firstclass>` type. The bit
5526 sizes of ``value`` and the destination type, ``ty2``, must be identical.
5527 If the source type is a pointer, the destination type must also be a
5528 pointer. This instruction supports bitwise conversion of vectors to
5529 integers and to vectors of other types (as long as they have the same
5535 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It is
5536 always a *no-op cast* because no bits change with this conversion. The
5537 conversion is done as if the ``value`` had been stored to memory and
5538 read back as type ``ty2``. Pointer (or vector of pointers) types may
5539 only be converted to other pointer (or vector of pointers) types with
5540 this instruction. To convert pointers to other types, use the
5541 :ref:`inttoptr <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions
5547 .. code-block:: llvm
5549 %X = bitcast i8 255 to i8 ; yields i8 :-1
5550 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5551 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5552 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5559 The instructions in this category are the "miscellaneous" instructions,
5560 which defy better classification.
5564 '``icmp``' Instruction
5565 ^^^^^^^^^^^^^^^^^^^^^^
5572 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5577 The '``icmp``' instruction returns a boolean value or a vector of
5578 boolean values based on comparison of its two integer, integer vector,
5579 pointer, or pointer vector operands.
5584 The '``icmp``' instruction takes three operands. The first operand is
5585 the condition code indicating the kind of comparison to perform. It is
5586 not a value, just a keyword. The possible condition code are:
5589 #. ``ne``: not equal
5590 #. ``ugt``: unsigned greater than
5591 #. ``uge``: unsigned greater or equal
5592 #. ``ult``: unsigned less than
5593 #. ``ule``: unsigned less or equal
5594 #. ``sgt``: signed greater than
5595 #. ``sge``: signed greater or equal
5596 #. ``slt``: signed less than
5597 #. ``sle``: signed less or equal
5599 The remaining two arguments must be :ref:`integer <t_integer>` or
5600 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5601 must also be identical types.
5606 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5607 code given as ``cond``. The comparison performed always yields either an
5608 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5610 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5611 otherwise. No sign interpretation is necessary or performed.
5612 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5613 otherwise. No sign interpretation is necessary or performed.
5614 #. ``ugt``: interprets the operands as unsigned values and yields
5615 ``true`` if ``op1`` is greater than ``op2``.
5616 #. ``uge``: interprets the operands as unsigned values and yields
5617 ``true`` if ``op1`` is greater than or equal to ``op2``.
5618 #. ``ult``: interprets the operands as unsigned values and yields
5619 ``true`` if ``op1`` is less than ``op2``.
5620 #. ``ule``: interprets the operands as unsigned values and yields
5621 ``true`` if ``op1`` is less than or equal to ``op2``.
5622 #. ``sgt``: interprets the operands as signed values and yields ``true``
5623 if ``op1`` is greater than ``op2``.
5624 #. ``sge``: interprets the operands as signed values and yields ``true``
5625 if ``op1`` is greater than or equal to ``op2``.
5626 #. ``slt``: interprets the operands as signed values and yields ``true``
5627 if ``op1`` is less than ``op2``.
5628 #. ``sle``: interprets the operands as signed values and yields ``true``
5629 if ``op1`` is less than or equal to ``op2``.
5631 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5632 are compared as if they were integers.
5634 If the operands are integer vectors, then they are compared element by
5635 element. The result is an ``i1`` vector with the same number of elements
5636 as the values being compared. Otherwise, the result is an ``i1``.
5641 .. code-block:: llvm
5643 <result> = icmp eq i32 4, 5 ; yields: result=false
5644 <result> = icmp ne float* %X, %X ; yields: result=false
5645 <result> = icmp ult i16 4, 5 ; yields: result=true
5646 <result> = icmp sgt i16 4, 5 ; yields: result=false
5647 <result> = icmp ule i16 -4, 5 ; yields: result=false
5648 <result> = icmp sge i16 4, 5 ; yields: result=false
5650 Note that the code generator does not yet support vector types with the
5651 ``icmp`` instruction.
5655 '``fcmp``' Instruction
5656 ^^^^^^^^^^^^^^^^^^^^^^
5663 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5668 The '``fcmp``' instruction returns a boolean value or vector of boolean
5669 values based on comparison of its operands.
5671 If the operands are floating point scalars, then the result type is a
5672 boolean (:ref:`i1 <t_integer>`).
5674 If the operands are floating point vectors, then the result type is a
5675 vector of boolean with the same number of elements as the operands being
5681 The '``fcmp``' instruction takes three operands. The first operand is
5682 the condition code indicating the kind of comparison to perform. It is
5683 not a value, just a keyword. The possible condition code are:
5685 #. ``false``: no comparison, always returns false
5686 #. ``oeq``: ordered and equal
5687 #. ``ogt``: ordered and greater than
5688 #. ``oge``: ordered and greater than or equal
5689 #. ``olt``: ordered and less than
5690 #. ``ole``: ordered and less than or equal
5691 #. ``one``: ordered and not equal
5692 #. ``ord``: ordered (no nans)
5693 #. ``ueq``: unordered or equal
5694 #. ``ugt``: unordered or greater than
5695 #. ``uge``: unordered or greater than or equal
5696 #. ``ult``: unordered or less than
5697 #. ``ule``: unordered or less than or equal
5698 #. ``une``: unordered or not equal
5699 #. ``uno``: unordered (either nans)
5700 #. ``true``: no comparison, always returns true
5702 *Ordered* means that neither operand is a QNAN while *unordered* means
5703 that either operand may be a QNAN.
5705 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5706 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5707 type. They must have identical types.
5712 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5713 condition code given as ``cond``. If the operands are vectors, then the
5714 vectors are compared element by element. Each comparison performed
5715 always yields an :ref:`i1 <t_integer>` result, as follows:
5717 #. ``false``: always yields ``false``, regardless of operands.
5718 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5719 is equal to ``op2``.
5720 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5721 is greater than ``op2``.
5722 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5723 is greater than or equal to ``op2``.
5724 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5725 is less than ``op2``.
5726 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5727 is less than or equal to ``op2``.
5728 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5729 is not equal to ``op2``.
5730 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5731 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5733 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5734 greater than ``op2``.
5735 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5736 greater than or equal to ``op2``.
5737 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5739 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5740 less than or equal to ``op2``.
5741 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5742 not equal to ``op2``.
5743 #. ``uno``: yields ``true`` if either operand is a QNAN.
5744 #. ``true``: always yields ``true``, regardless of operands.
5749 .. code-block:: llvm
5751 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5752 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5753 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5754 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5756 Note that the code generator does not yet support vector types with the
5757 ``fcmp`` instruction.
5761 '``phi``' Instruction
5762 ^^^^^^^^^^^^^^^^^^^^^
5769 <result> = phi <ty> [ <val0>, <label0>], ...
5774 The '``phi``' instruction is used to implement the φ node in the SSA
5775 graph representing the function.
5780 The type of the incoming values is specified with the first type field.
5781 After this, the '``phi``' instruction takes a list of pairs as
5782 arguments, with one pair for each predecessor basic block of the current
5783 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5784 the value arguments to the PHI node. Only labels may be used as the
5787 There must be no non-phi instructions between the start of a basic block
5788 and the PHI instructions: i.e. PHI instructions must be first in a basic
5791 For the purposes of the SSA form, the use of each incoming value is
5792 deemed to occur on the edge from the corresponding predecessor block to
5793 the current block (but after any definition of an '``invoke``'
5794 instruction's return value on the same edge).
5799 At runtime, the '``phi``' instruction logically takes on the value
5800 specified by the pair corresponding to the predecessor basic block that
5801 executed just prior to the current block.
5806 .. code-block:: llvm
5808 Loop: ; Infinite loop that counts from 0 on up...
5809 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5810 %nextindvar = add i32 %indvar, 1
5815 '``select``' Instruction
5816 ^^^^^^^^^^^^^^^^^^^^^^^^
5823 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5825 selty is either i1 or {<N x i1>}
5830 The '``select``' instruction is used to choose one value based on a
5831 condition, without branching.
5836 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5837 values indicating the condition, and two values of the same :ref:`first
5838 class <t_firstclass>` type. If the val1/val2 are vectors and the
5839 condition is a scalar, then entire vectors are selected, not individual
5845 If the condition is an i1 and it evaluates to 1, the instruction returns
5846 the first value argument; otherwise, it returns the second value
5849 If the condition is a vector of i1, then the value arguments must be
5850 vectors of the same size, and the selection is done element by element.
5855 .. code-block:: llvm
5857 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5861 '``call``' Instruction
5862 ^^^^^^^^^^^^^^^^^^^^^^
5869 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5874 The '``call``' instruction represents a simple function call.
5879 This instruction requires several arguments:
5881 #. The optional "tail" marker indicates that the callee function does
5882 not access any allocas or varargs in the caller. Note that calls may
5883 be marked "tail" even if they do not occur before a
5884 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5885 function call is eligible for tail call optimization, but `might not
5886 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5887 The code generator may optimize calls marked "tail" with either 1)
5888 automatic `sibling call
5889 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5890 callee have matching signatures, or 2) forced tail call optimization
5891 when the following extra requirements are met:
5893 - Caller and callee both have the calling convention ``fastcc``.
5894 - The call is in tail position (ret immediately follows call and ret
5895 uses value of call or is void).
5896 - Option ``-tailcallopt`` is enabled, or
5897 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5898 - `Platform specific constraints are
5899 met. <CodeGenerator.html#tailcallopt>`_
5901 #. The optional "cconv" marker indicates which :ref:`calling
5902 convention <callingconv>` the call should use. If none is
5903 specified, the call defaults to using C calling conventions. The
5904 calling convention of the call must match the calling convention of
5905 the target function, or else the behavior is undefined.
5906 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5907 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5909 #. '``ty``': the type of the call instruction itself which is also the
5910 type of the return value. Functions that return no value are marked
5912 #. '``fnty``': shall be the signature of the pointer to function value
5913 being invoked. The argument types must match the types implied by
5914 this signature. This type can be omitted if the function is not
5915 varargs and if the function type does not return a pointer to a
5917 #. '``fnptrval``': An LLVM value containing a pointer to a function to
5918 be invoked. In most cases, this is a direct function invocation, but
5919 indirect ``call``'s are just as possible, calling an arbitrary pointer
5921 #. '``function args``': argument list whose types match the function
5922 signature argument types and parameter attributes. All arguments must
5923 be of :ref:`first class <t_firstclass>` type. If the function signature
5924 indicates the function accepts a variable number of arguments, the
5925 extra arguments can be specified.
5926 #. The optional :ref:`function attributes <fnattrs>` list. Only
5927 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5928 attributes are valid here.
5933 The '``call``' instruction is used to cause control flow to transfer to
5934 a specified function, with its incoming arguments bound to the specified
5935 values. Upon a '``ret``' instruction in the called function, control
5936 flow continues with the instruction after the function call, and the
5937 return value of the function is bound to the result argument.
5942 .. code-block:: llvm
5944 %retval = call i32 @test(i32 %argc)
5945 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
5946 %X = tail call i32 @foo() ; yields i32
5947 %Y = tail call fastcc i32 @foo() ; yields i32
5948 call void %foo(i8 97 signext)
5950 %struct.A = type { i32, i8 }
5951 %r = call %struct.A @foo() ; yields { 32, i8 }
5952 %gr = extractvalue %struct.A %r, 0 ; yields i32
5953 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
5954 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
5955 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
5957 llvm treats calls to some functions with names and arguments that match
5958 the standard C99 library as being the C99 library functions, and may
5959 perform optimizations or generate code for them under that assumption.
5960 This is something we'd like to change in the future to provide better
5961 support for freestanding environments and non-C-based languages.
5965 '``va_arg``' Instruction
5966 ^^^^^^^^^^^^^^^^^^^^^^^^
5973 <resultval> = va_arg <va_list*> <arglist>, <argty>
5978 The '``va_arg``' instruction is used to access arguments passed through
5979 the "variable argument" area of a function call. It is used to implement
5980 the ``va_arg`` macro in C.
5985 This instruction takes a ``va_list*`` value and the type of the
5986 argument. It returns a value of the specified argument type and
5987 increments the ``va_list`` to point to the next argument. The actual
5988 type of ``va_list`` is target specific.
5993 The '``va_arg``' instruction loads an argument of the specified type
5994 from the specified ``va_list`` and causes the ``va_list`` to point to
5995 the next argument. For more information, see the variable argument
5996 handling :ref:`Intrinsic Functions <int_varargs>`.
5998 It is legal for this instruction to be called in a function which does
5999 not take a variable number of arguments, for example, the ``vfprintf``
6002 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6003 function <intrinsics>` because it takes a type as an argument.
6008 See the :ref:`variable argument processing <int_varargs>` section.
6010 Note that the code generator does not yet fully support va\_arg on many
6011 targets. Also, it does not currently support va\_arg with aggregate
6012 types on any target.
6016 '``landingpad``' Instruction
6017 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6024 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6025 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6027 <clause> := catch <type> <value>
6028 <clause> := filter <array constant type> <array constant>
6033 The '``landingpad``' instruction is used by `LLVM's exception handling
6034 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6035 is a landing pad --- one where the exception lands, and corresponds to the
6036 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6037 defines values supplied by the personality function (``pers_fn``) upon
6038 re-entry to the function. The ``resultval`` has the type ``resultty``.
6043 This instruction takes a ``pers_fn`` value. This is the personality
6044 function associated with the unwinding mechanism. The optional
6045 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6047 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6048 contains the global variable representing the "type" that may be caught
6049 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6050 clause takes an array constant as its argument. Use
6051 "``[0 x i8**] undef``" for a filter which cannot throw. The
6052 '``landingpad``' instruction must contain *at least* one ``clause`` or
6053 the ``cleanup`` flag.
6058 The '``landingpad``' instruction defines the values which are set by the
6059 personality function (``pers_fn``) upon re-entry to the function, and
6060 therefore the "result type" of the ``landingpad`` instruction. As with
6061 calling conventions, how the personality function results are
6062 represented in LLVM IR is target specific.
6064 The clauses are applied in order from top to bottom. If two
6065 ``landingpad`` instructions are merged together through inlining, the
6066 clauses from the calling function are appended to the list of clauses.
6067 When the call stack is being unwound due to an exception being thrown,
6068 the exception is compared against each ``clause`` in turn. If it doesn't
6069 match any of the clauses, and the ``cleanup`` flag is not set, then
6070 unwinding continues further up the call stack.
6072 The ``landingpad`` instruction has several restrictions:
6074 - A landing pad block is a basic block which is the unwind destination
6075 of an '``invoke``' instruction.
6076 - A landing pad block must have a '``landingpad``' instruction as its
6077 first non-PHI instruction.
6078 - There can be only one '``landingpad``' instruction within the landing
6080 - A basic block that is not a landing pad block may not include a
6081 '``landingpad``' instruction.
6082 - All '``landingpad``' instructions in a function must have the same
6083 personality function.
6088 .. code-block:: llvm
6090 ;; A landing pad which can catch an integer.
6091 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6093 ;; A landing pad that is a cleanup.
6094 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6096 ;; A landing pad which can catch an integer and can only throw a double.
6097 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6099 filter [1 x i8**] [@_ZTId]
6106 LLVM supports the notion of an "intrinsic function". These functions
6107 have well known names and semantics and are required to follow certain
6108 restrictions. Overall, these intrinsics represent an extension mechanism
6109 for the LLVM language that does not require changing all of the
6110 transformations in LLVM when adding to the language (or the bitcode
6111 reader/writer, the parser, etc...).
6113 Intrinsic function names must all start with an "``llvm.``" prefix. This
6114 prefix is reserved in LLVM for intrinsic names; thus, function names may
6115 not begin with this prefix. Intrinsic functions must always be external
6116 functions: you cannot define the body of intrinsic functions. Intrinsic
6117 functions may only be used in call or invoke instructions: it is illegal
6118 to take the address of an intrinsic function. Additionally, because
6119 intrinsic functions are part of the LLVM language, it is required if any
6120 are added that they be documented here.
6122 Some intrinsic functions can be overloaded, i.e., the intrinsic
6123 represents a family of functions that perform the same operation but on
6124 different data types. Because LLVM can represent over 8 million
6125 different integer types, overloading is used commonly to allow an
6126 intrinsic function to operate on any integer type. One or more of the
6127 argument types or the result type can be overloaded to accept any
6128 integer type. Argument types may also be defined as exactly matching a
6129 previous argument's type or the result type. This allows an intrinsic
6130 function which accepts multiple arguments, but needs all of them to be
6131 of the same type, to only be overloaded with respect to a single
6132 argument or the result.
6134 Overloaded intrinsics will have the names of its overloaded argument
6135 types encoded into its function name, each preceded by a period. Only
6136 those types which are overloaded result in a name suffix. Arguments
6137 whose type is matched against another type do not. For example, the
6138 ``llvm.ctpop`` function can take an integer of any width and returns an
6139 integer of exactly the same integer width. This leads to a family of
6140 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6141 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6142 overloaded, and only one type suffix is required. Because the argument's
6143 type is matched against the return type, it does not require its own
6146 To learn how to add an intrinsic function, please see the `Extending
6147 LLVM Guide <ExtendingLLVM.html>`_.
6151 Variable Argument Handling Intrinsics
6152 -------------------------------------
6154 Variable argument support is defined in LLVM with the
6155 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6156 functions. These functions are related to the similarly named macros
6157 defined in the ``<stdarg.h>`` header file.
6159 All of these functions operate on arguments that use a target-specific
6160 value type "``va_list``". The LLVM assembly language reference manual
6161 does not define what this type is, so all transformations should be
6162 prepared to handle these functions regardless of the type used.
6164 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6165 variable argument handling intrinsic functions are used.
6167 .. code-block:: llvm
6169 define i32 @test(i32 %X, ...) {
6170 ; Initialize variable argument processing
6172 %ap2 = bitcast i8** %ap to i8*
6173 call void @llvm.va_start(i8* %ap2)
6175 ; Read a single integer argument
6176 %tmp = va_arg i8** %ap, i32
6178 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6180 %aq2 = bitcast i8** %aq to i8*
6181 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6182 call void @llvm.va_end(i8* %aq2)
6184 ; Stop processing of arguments.
6185 call void @llvm.va_end(i8* %ap2)
6189 declare void @llvm.va_start(i8*)
6190 declare void @llvm.va_copy(i8*, i8*)
6191 declare void @llvm.va_end(i8*)
6195 '``llvm.va_start``' Intrinsic
6196 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6203 declare void %llvm.va_start(i8* <arglist>)
6208 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6209 subsequent use by ``va_arg``.
6214 The argument is a pointer to a ``va_list`` element to initialize.
6219 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6220 available in C. In a target-dependent way, it initializes the
6221 ``va_list`` element to which the argument points, so that the next call
6222 to ``va_arg`` will produce the first variable argument passed to the
6223 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6224 to know the last argument of the function as the compiler can figure
6227 '``llvm.va_end``' Intrinsic
6228 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6235 declare void @llvm.va_end(i8* <arglist>)
6240 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6241 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6246 The argument is a pointer to a ``va_list`` to destroy.
6251 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6252 available in C. In a target-dependent way, it destroys the ``va_list``
6253 element to which the argument points. Calls to
6254 :ref:`llvm.va_start <int_va_start>` and
6255 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6260 '``llvm.va_copy``' Intrinsic
6261 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6268 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6273 The '``llvm.va_copy``' intrinsic copies the current argument position
6274 from the source argument list to the destination argument list.
6279 The first argument is a pointer to a ``va_list`` element to initialize.
6280 The second argument is a pointer to a ``va_list`` element to copy from.
6285 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6286 available in C. In a target-dependent way, it copies the source
6287 ``va_list`` element into the destination ``va_list`` element. This
6288 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6289 arbitrarily complex and require, for example, memory allocation.
6291 Accurate Garbage Collection Intrinsics
6292 --------------------------------------
6294 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6295 (GC) requires the implementation and generation of these intrinsics.
6296 These intrinsics allow identification of :ref:`GC roots on the
6297 stack <int_gcroot>`, as well as garbage collector implementations that
6298 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6299 Front-ends for type-safe garbage collected languages should generate
6300 these intrinsics to make use of the LLVM garbage collectors. For more
6301 details, see `Accurate Garbage Collection with
6302 LLVM <GarbageCollection.html>`_.
6304 The garbage collection intrinsics only operate on objects in the generic
6305 address space (address space zero).
6309 '``llvm.gcroot``' Intrinsic
6310 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6317 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6322 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6323 the code generator, and allows some metadata to be associated with it.
6328 The first argument specifies the address of a stack object that contains
6329 the root pointer. The second pointer (which must be either a constant or
6330 a global value address) contains the meta-data to be associated with the
6336 At runtime, a call to this intrinsic stores a null pointer into the
6337 "ptrloc" location. At compile-time, the code generator generates
6338 information to allow the runtime to find the pointer at GC safe points.
6339 The '``llvm.gcroot``' intrinsic may only be used in a function which
6340 :ref:`specifies a GC algorithm <gc>`.
6344 '``llvm.gcread``' Intrinsic
6345 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6352 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6357 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6358 locations, allowing garbage collector implementations that require read
6364 The second argument is the address to read from, which should be an
6365 address allocated from the garbage collector. The first object is a
6366 pointer to the start of the referenced object, if needed by the language
6367 runtime (otherwise null).
6372 The '``llvm.gcread``' intrinsic has the same semantics as a load
6373 instruction, but may be replaced with substantially more complex code by
6374 the garbage collector runtime, as needed. The '``llvm.gcread``'
6375 intrinsic may only be used in a function which :ref:`specifies a GC
6380 '``llvm.gcwrite``' Intrinsic
6381 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6388 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6393 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6394 locations, allowing garbage collector implementations that require write
6395 barriers (such as generational or reference counting collectors).
6400 The first argument is the reference to store, the second is the start of
6401 the object to store it to, and the third is the address of the field of
6402 Obj to store to. If the runtime does not require a pointer to the
6403 object, Obj may be null.
6408 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6409 instruction, but may be replaced with substantially more complex code by
6410 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6411 intrinsic may only be used in a function which :ref:`specifies a GC
6414 Code Generator Intrinsics
6415 -------------------------
6417 These intrinsics are provided by LLVM to expose special features that
6418 may only be implemented with code generator support.
6420 '``llvm.returnaddress``' Intrinsic
6421 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6428 declare i8 *@llvm.returnaddress(i32 <level>)
6433 The '``llvm.returnaddress``' intrinsic attempts to compute a
6434 target-specific value indicating the return address of the current
6435 function or one of its callers.
6440 The argument to this intrinsic indicates which function to return the
6441 address for. Zero indicates the calling function, one indicates its
6442 caller, etc. The argument is **required** to be a constant integer
6448 The '``llvm.returnaddress``' intrinsic either returns a pointer
6449 indicating the return address of the specified call frame, or zero if it
6450 cannot be identified. The value returned by this intrinsic is likely to
6451 be incorrect or 0 for arguments other than zero, so it should only be
6452 used for debugging purposes.
6454 Note that calling this intrinsic does not prevent function inlining or
6455 other aggressive transformations, so the value returned may not be that
6456 of the obvious source-language caller.
6458 '``llvm.frameaddress``' Intrinsic
6459 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6466 declare i8* @llvm.frameaddress(i32 <level>)
6471 The '``llvm.frameaddress``' intrinsic attempts to return the
6472 target-specific frame pointer value for the specified stack frame.
6477 The argument to this intrinsic indicates which function to return the
6478 frame pointer for. Zero indicates the calling function, one indicates
6479 its caller, etc. The argument is **required** to be a constant integer
6485 The '``llvm.frameaddress``' intrinsic either returns a pointer
6486 indicating the frame address of the specified call frame, or zero if it
6487 cannot be identified. The value returned by this intrinsic is likely to
6488 be incorrect or 0 for arguments other than zero, so it should only be
6489 used for debugging purposes.
6491 Note that calling this intrinsic does not prevent function inlining or
6492 other aggressive transformations, so the value returned may not be that
6493 of the obvious source-language caller.
6497 '``llvm.stacksave``' Intrinsic
6498 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6505 declare i8* @llvm.stacksave()
6510 The '``llvm.stacksave``' intrinsic is used to remember the current state
6511 of the function stack, for use with
6512 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6513 implementing language features like scoped automatic variable sized
6519 This intrinsic returns a opaque pointer value that can be passed to
6520 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6521 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6522 ``llvm.stacksave``, it effectively restores the state of the stack to
6523 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6524 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6525 were allocated after the ``llvm.stacksave`` was executed.
6527 .. _int_stackrestore:
6529 '``llvm.stackrestore``' Intrinsic
6530 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6537 declare void @llvm.stackrestore(i8* %ptr)
6542 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6543 the function stack to the state it was in when the corresponding
6544 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6545 useful for implementing language features like scoped automatic variable
6546 sized arrays in C99.
6551 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6553 '``llvm.prefetch``' Intrinsic
6554 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6561 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6566 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6567 insert a prefetch instruction if supported; otherwise, it is a noop.
6568 Prefetches have no effect on the behavior of the program but can change
6569 its performance characteristics.
6574 ``address`` is the address to be prefetched, ``rw`` is the specifier
6575 determining if the fetch should be for a read (0) or write (1), and
6576 ``locality`` is a temporal locality specifier ranging from (0) - no
6577 locality, to (3) - extremely local keep in cache. The ``cache type``
6578 specifies whether the prefetch is performed on the data (1) or
6579 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6580 arguments must be constant integers.
6585 This intrinsic does not modify the behavior of the program. In
6586 particular, prefetches cannot trap and do not produce a value. On
6587 targets that support this intrinsic, the prefetch can provide hints to
6588 the processor cache for better performance.
6590 '``llvm.pcmarker``' Intrinsic
6591 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6598 declare void @llvm.pcmarker(i32 <id>)
6603 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6604 Counter (PC) in a region of code to simulators and other tools. The
6605 method is target specific, but it is expected that the marker will use
6606 exported symbols to transmit the PC of the marker. The marker makes no
6607 guarantees that it will remain with any specific instruction after
6608 optimizations. It is possible that the presence of a marker will inhibit
6609 optimizations. The intended use is to be inserted after optimizations to
6610 allow correlations of simulation runs.
6615 ``id`` is a numerical id identifying the marker.
6620 This intrinsic does not modify the behavior of the program. Backends
6621 that do not support this intrinsic may ignore it.
6623 '``llvm.readcyclecounter``' Intrinsic
6624 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6631 declare i64 @llvm.readcyclecounter()
6636 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6637 counter register (or similar low latency, high accuracy clocks) on those
6638 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6639 should map to RPCC. As the backing counters overflow quickly (on the
6640 order of 9 seconds on alpha), this should only be used for small
6646 When directly supported, reading the cycle counter should not modify any
6647 memory. Implementations are allowed to either return a application
6648 specific value or a system wide value. On backends without support, this
6649 is lowered to a constant 0.
6651 Standard C Library Intrinsics
6652 -----------------------------
6654 LLVM provides intrinsics for a few important standard C library
6655 functions. These intrinsics allow source-language front-ends to pass
6656 information about the alignment of the pointer arguments to the code
6657 generator, providing opportunity for more efficient code generation.
6661 '``llvm.memcpy``' Intrinsic
6662 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6667 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6668 integer bit width and for different address spaces. Not all targets
6669 support all bit widths however.
6673 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6674 i32 <len>, i32 <align>, i1 <isvolatile>)
6675 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6676 i64 <len>, i32 <align>, i1 <isvolatile>)
6681 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6682 source location to the destination location.
6684 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6685 intrinsics do not return a value, takes extra alignment/isvolatile
6686 arguments and the pointers can be in specified address spaces.
6691 The first argument is a pointer to the destination, the second is a
6692 pointer to the source. The third argument is an integer argument
6693 specifying the number of bytes to copy, the fourth argument is the
6694 alignment of the source and destination locations, and the fifth is a
6695 boolean indicating a volatile access.
6697 If the call to this intrinsic has an alignment value that is not 0 or 1,
6698 then the caller guarantees that both the source and destination pointers
6699 are aligned to that boundary.
6701 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6702 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6703 very cleanly specified and it is unwise to depend on it.
6708 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6709 source location to the destination location, which are not allowed to
6710 overlap. It copies "len" bytes of memory over. If the argument is known
6711 to be aligned to some boundary, this can be specified as the fourth
6712 argument, otherwise it should be set to 0 or 1.
6714 '``llvm.memmove``' Intrinsic
6715 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6720 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6721 bit width and for different address space. Not all targets support all
6726 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6727 i32 <len>, i32 <align>, i1 <isvolatile>)
6728 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6729 i64 <len>, i32 <align>, i1 <isvolatile>)
6734 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6735 source location to the destination location. It is similar to the
6736 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6739 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6740 intrinsics do not return a value, takes extra alignment/isvolatile
6741 arguments and the pointers can be in specified address spaces.
6746 The first argument is a pointer to the destination, the second is a
6747 pointer to the source. The third argument is an integer argument
6748 specifying the number of bytes to copy, the fourth argument is the
6749 alignment of the source and destination locations, and the fifth is a
6750 boolean indicating a volatile access.
6752 If the call to this intrinsic has an alignment value that is not 0 or 1,
6753 then the caller guarantees that the source and destination pointers are
6754 aligned to that boundary.
6756 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6757 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6758 not very cleanly specified and it is unwise to depend on it.
6763 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6764 source location to the destination location, which may overlap. It
6765 copies "len" bytes of memory over. If the argument is known to be
6766 aligned to some boundary, this can be specified as the fourth argument,
6767 otherwise it should be set to 0 or 1.
6769 '``llvm.memset.*``' Intrinsics
6770 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6775 This is an overloaded intrinsic. You can use llvm.memset on any integer
6776 bit width and for different address spaces. However, not all targets
6777 support all bit widths.
6781 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6782 i32 <len>, i32 <align>, i1 <isvolatile>)
6783 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6784 i64 <len>, i32 <align>, i1 <isvolatile>)
6789 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6790 particular byte value.
6792 Note that, unlike the standard libc function, the ``llvm.memset``
6793 intrinsic does not return a value and takes extra alignment/volatile
6794 arguments. Also, the destination can be in an arbitrary address space.
6799 The first argument is a pointer to the destination to fill, the second
6800 is the byte value with which to fill it, the third argument is an
6801 integer argument specifying the number of bytes to fill, and the fourth
6802 argument is the known alignment of the destination location.
6804 If the call to this intrinsic has an alignment value that is not 0 or 1,
6805 then the caller guarantees that the destination pointer is aligned to
6808 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6809 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6810 very cleanly specified and it is unwise to depend on it.
6815 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6816 at the destination location. If the argument is known to be aligned to
6817 some boundary, this can be specified as the fourth argument, otherwise
6818 it should be set to 0 or 1.
6820 '``llvm.sqrt.*``' Intrinsic
6821 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6826 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6827 floating point or vector of floating point type. Not all targets support
6832 declare float @llvm.sqrt.f32(float %Val)
6833 declare double @llvm.sqrt.f64(double %Val)
6834 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6835 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6836 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6841 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6842 returning the same value as the libm '``sqrt``' functions would. Unlike
6843 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6844 negative numbers other than -0.0 (which allows for better optimization,
6845 because there is no need to worry about errno being set).
6846 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6851 The argument and return value are floating point numbers of the same
6857 This function returns the sqrt of the specified operand if it is a
6858 nonnegative floating point number.
6860 '``llvm.powi.*``' Intrinsic
6861 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6866 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6867 floating point or vector of floating point type. Not all targets support
6872 declare float @llvm.powi.f32(float %Val, i32 %power)
6873 declare double @llvm.powi.f64(double %Val, i32 %power)
6874 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6875 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6876 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6881 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6882 specified (positive or negative) power. The order of evaluation of
6883 multiplications is not defined. When a vector of floating point type is
6884 used, the second argument remains a scalar integer value.
6889 The second argument is an integer power, and the first is a value to
6890 raise to that power.
6895 This function returns the first value raised to the second power with an
6896 unspecified sequence of rounding operations.
6898 '``llvm.sin.*``' Intrinsic
6899 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6904 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
6905 floating point or vector of floating point type. Not all targets support
6910 declare float @llvm.sin.f32(float %Val)
6911 declare double @llvm.sin.f64(double %Val)
6912 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6913 declare fp128 @llvm.sin.f128(fp128 %Val)
6914 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6919 The '``llvm.sin.*``' intrinsics return the sine of the operand.
6924 The argument and return value are floating point numbers of the same
6930 This function returns the sine of the specified operand, returning the
6931 same values as the libm ``sin`` functions would, and handles error
6932 conditions in the same way.
6934 '``llvm.cos.*``' Intrinsic
6935 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6940 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
6941 floating point or vector of floating point type. Not all targets support
6946 declare float @llvm.cos.f32(float %Val)
6947 declare double @llvm.cos.f64(double %Val)
6948 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6949 declare fp128 @llvm.cos.f128(fp128 %Val)
6950 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6955 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
6960 The argument and return value are floating point numbers of the same
6966 This function returns the cosine of the specified operand, returning the
6967 same values as the libm ``cos`` functions would, and handles error
6968 conditions in the same way.
6970 '``llvm.pow.*``' Intrinsic
6971 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6976 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
6977 floating point or vector of floating point type. Not all targets support
6982 declare float @llvm.pow.f32(float %Val, float %Power)
6983 declare double @llvm.pow.f64(double %Val, double %Power)
6984 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6985 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6986 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6991 The '``llvm.pow.*``' intrinsics return the first operand raised to the
6992 specified (positive or negative) power.
6997 The second argument is a floating point power, and the first is a value
6998 to raise to that power.
7003 This function returns the first value raised to the second power,
7004 returning the same values as the libm ``pow`` functions would, and
7005 handles error conditions in the same way.
7007 '``llvm.exp.*``' Intrinsic
7008 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7013 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7014 floating point or vector of floating point type. Not all targets support
7019 declare float @llvm.exp.f32(float %Val)
7020 declare double @llvm.exp.f64(double %Val)
7021 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7022 declare fp128 @llvm.exp.f128(fp128 %Val)
7023 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7028 The '``llvm.exp.*``' intrinsics perform the exp function.
7033 The argument and return value are floating point numbers of the same
7039 This function returns the same values as the libm ``exp`` functions
7040 would, and handles error conditions in the same way.
7042 '``llvm.exp2.*``' Intrinsic
7043 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7048 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7049 floating point or vector of floating point type. Not all targets support
7054 declare float @llvm.exp2.f32(float %Val)
7055 declare double @llvm.exp2.f64(double %Val)
7056 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7057 declare fp128 @llvm.exp2.f128(fp128 %Val)
7058 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7063 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7068 The argument and return value are floating point numbers of the same
7074 This function returns the same values as the libm ``exp2`` functions
7075 would, and handles error conditions in the same way.
7077 '``llvm.log.*``' Intrinsic
7078 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7083 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7084 floating point or vector of floating point type. Not all targets support
7089 declare float @llvm.log.f32(float %Val)
7090 declare double @llvm.log.f64(double %Val)
7091 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7092 declare fp128 @llvm.log.f128(fp128 %Val)
7093 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7098 The '``llvm.log.*``' intrinsics perform the log function.
7103 The argument and return value are floating point numbers of the same
7109 This function returns the same values as the libm ``log`` functions
7110 would, and handles error conditions in the same way.
7112 '``llvm.log10.*``' Intrinsic
7113 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7118 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7119 floating point or vector of floating point type. Not all targets support
7124 declare float @llvm.log10.f32(float %Val)
7125 declare double @llvm.log10.f64(double %Val)
7126 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7127 declare fp128 @llvm.log10.f128(fp128 %Val)
7128 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7133 The '``llvm.log10.*``' intrinsics perform the log10 function.
7138 The argument and return value are floating point numbers of the same
7144 This function returns the same values as the libm ``log10`` functions
7145 would, and handles error conditions in the same way.
7147 '``llvm.log2.*``' Intrinsic
7148 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7153 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7154 floating point or vector of floating point type. Not all targets support
7159 declare float @llvm.log2.f32(float %Val)
7160 declare double @llvm.log2.f64(double %Val)
7161 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7162 declare fp128 @llvm.log2.f128(fp128 %Val)
7163 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7168 The '``llvm.log2.*``' intrinsics perform the log2 function.
7173 The argument and return value are floating point numbers of the same
7179 This function returns the same values as the libm ``log2`` functions
7180 would, and handles error conditions in the same way.
7182 '``llvm.fma.*``' Intrinsic
7183 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7188 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7189 floating point or vector of floating point type. Not all targets support
7194 declare float @llvm.fma.f32(float %a, float %b, float %c)
7195 declare double @llvm.fma.f64(double %a, double %b, double %c)
7196 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7197 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7198 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7203 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7209 The argument and return value are floating point numbers of the same
7215 This function returns the same values as the libm ``fma`` functions
7218 '``llvm.fabs.*``' Intrinsic
7219 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7224 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7225 floating point or vector of floating point type. Not all targets support
7230 declare float @llvm.fabs.f32(float %Val)
7231 declare double @llvm.fabs.f64(double %Val)
7232 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7233 declare fp128 @llvm.fabs.f128(fp128 %Val)
7234 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7239 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7245 The argument and return value are floating point numbers of the same
7251 This function returns the same values as the libm ``fabs`` functions
7252 would, and handles error conditions in the same way.
7254 '``llvm.floor.*``' Intrinsic
7255 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7260 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7261 floating point or vector of floating point type. Not all targets support
7266 declare float @llvm.floor.f32(float %Val)
7267 declare double @llvm.floor.f64(double %Val)
7268 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7269 declare fp128 @llvm.floor.f128(fp128 %Val)
7270 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7275 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7280 The argument and return value are floating point numbers of the same
7286 This function returns the same values as the libm ``floor`` functions
7287 would, and handles error conditions in the same way.
7289 '``llvm.ceil.*``' Intrinsic
7290 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7295 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7296 floating point or vector of floating point type. Not all targets support
7301 declare float @llvm.ceil.f32(float %Val)
7302 declare double @llvm.ceil.f64(double %Val)
7303 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7304 declare fp128 @llvm.ceil.f128(fp128 %Val)
7305 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7310 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7315 The argument and return value are floating point numbers of the same
7321 This function returns the same values as the libm ``ceil`` functions
7322 would, and handles error conditions in the same way.
7324 '``llvm.trunc.*``' Intrinsic
7325 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7330 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7331 floating point or vector of floating point type. Not all targets support
7336 declare float @llvm.trunc.f32(float %Val)
7337 declare double @llvm.trunc.f64(double %Val)
7338 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7339 declare fp128 @llvm.trunc.f128(fp128 %Val)
7340 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7345 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7346 nearest integer not larger in magnitude than the operand.
7351 The argument and return value are floating point numbers of the same
7357 This function returns the same values as the libm ``trunc`` functions
7358 would, and handles error conditions in the same way.
7360 '``llvm.rint.*``' Intrinsic
7361 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7366 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7367 floating point or vector of floating point type. Not all targets support
7372 declare float @llvm.rint.f32(float %Val)
7373 declare double @llvm.rint.f64(double %Val)
7374 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7375 declare fp128 @llvm.rint.f128(fp128 %Val)
7376 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7381 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7382 nearest integer. It may raise an inexact floating-point exception if the
7383 operand isn't an integer.
7388 The argument and return value are floating point numbers of the same
7394 This function returns the same values as the libm ``rint`` functions
7395 would, and handles error conditions in the same way.
7397 '``llvm.nearbyint.*``' Intrinsic
7398 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7403 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7404 floating point or vector of floating point type. Not all targets support
7409 declare float @llvm.nearbyint.f32(float %Val)
7410 declare double @llvm.nearbyint.f64(double %Val)
7411 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7412 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7413 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7418 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7424 The argument and return value are floating point numbers of the same
7430 This function returns the same values as the libm ``nearbyint``
7431 functions would, and handles error conditions in the same way.
7433 Bit Manipulation Intrinsics
7434 ---------------------------
7436 LLVM provides intrinsics for a few important bit manipulation
7437 operations. These allow efficient code generation for some algorithms.
7439 '``llvm.bswap.*``' Intrinsics
7440 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7445 This is an overloaded intrinsic function. You can use bswap on any
7446 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7450 declare i16 @llvm.bswap.i16(i16 <id>)
7451 declare i32 @llvm.bswap.i32(i32 <id>)
7452 declare i64 @llvm.bswap.i64(i64 <id>)
7457 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7458 values with an even number of bytes (positive multiple of 16 bits).
7459 These are useful for performing operations on data that is not in the
7460 target's native byte order.
7465 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7466 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7467 intrinsic returns an i32 value that has the four bytes of the input i32
7468 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7469 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7470 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7471 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7474 '``llvm.ctpop.*``' Intrinsic
7475 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7480 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7481 bit width, or on any vector with integer elements. Not all targets
7482 support all bit widths or vector types, however.
7486 declare i8 @llvm.ctpop.i8(i8 <src>)
7487 declare i16 @llvm.ctpop.i16(i16 <src>)
7488 declare i32 @llvm.ctpop.i32(i32 <src>)
7489 declare i64 @llvm.ctpop.i64(i64 <src>)
7490 declare i256 @llvm.ctpop.i256(i256 <src>)
7491 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7496 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7502 The only argument is the value to be counted. The argument may be of any
7503 integer type, or a vector with integer elements. The return type must
7504 match the argument type.
7509 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7510 each element of a vector.
7512 '``llvm.ctlz.*``' Intrinsic
7513 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7518 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7519 integer bit width, or any vector whose elements are integers. Not all
7520 targets support all bit widths or vector types, however.
7524 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7525 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7526 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7527 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7528 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7529 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7534 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7535 leading zeros in a variable.
7540 The first argument is the value to be counted. This argument may be of
7541 any integer type, or a vectory with integer element type. The return
7542 type must match the first argument type.
7544 The second argument must be a constant and is a flag to indicate whether
7545 the intrinsic should ensure that a zero as the first argument produces a
7546 defined result. Historically some architectures did not provide a
7547 defined result for zero values as efficiently, and many algorithms are
7548 now predicated on avoiding zero-value inputs.
7553 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7554 zeros in a variable, or within each element of the vector. If
7555 ``src == 0`` then the result is the size in bits of the type of ``src``
7556 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7557 ``llvm.ctlz(i32 2) = 30``.
7559 '``llvm.cttz.*``' Intrinsic
7560 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7565 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7566 integer bit width, or any vector of integer elements. Not all targets
7567 support all bit widths or vector types, however.
7571 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7572 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7573 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7574 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7575 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7576 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7581 The '``llvm.cttz``' family of intrinsic functions counts the number of
7587 The first argument is the value to be counted. This argument may be of
7588 any integer type, or a vectory with integer element type. The return
7589 type must match the first argument type.
7591 The second argument must be a constant and is a flag to indicate whether
7592 the intrinsic should ensure that a zero as the first argument produces a
7593 defined result. Historically some architectures did not provide a
7594 defined result for zero values as efficiently, and many algorithms are
7595 now predicated on avoiding zero-value inputs.
7600 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7601 zeros in a variable, or within each element of a vector. If ``src == 0``
7602 then the result is the size in bits of the type of ``src`` if
7603 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7604 ``llvm.cttz(2) = 1``.
7606 Arithmetic with Overflow Intrinsics
7607 -----------------------------------
7609 LLVM provides intrinsics for some arithmetic with overflow operations.
7611 '``llvm.sadd.with.overflow.*``' Intrinsics
7612 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7617 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7618 on any integer bit width.
7622 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7623 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7624 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7629 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7630 a signed addition of the two arguments, and indicate whether an overflow
7631 occurred during the signed summation.
7636 The arguments (%a and %b) and the first element of the result structure
7637 may be of integer types of any bit width, but they must have the same
7638 bit width. The second element of the result structure must be of type
7639 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7645 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7646 a signed addition of the two variables. They return a structure --- the
7647 first element of which is the signed summation, and the second element
7648 of which is a bit specifying if the signed summation resulted in an
7654 .. code-block:: llvm
7656 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7657 %sum = extractvalue {i32, i1} %res, 0
7658 %obit = extractvalue {i32, i1} %res, 1
7659 br i1 %obit, label %overflow, label %normal
7661 '``llvm.uadd.with.overflow.*``' Intrinsics
7662 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7667 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7668 on any integer bit width.
7672 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7673 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7674 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7679 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7680 an unsigned addition of the two arguments, and indicate whether a carry
7681 occurred during the unsigned summation.
7686 The arguments (%a and %b) and the first element of the result structure
7687 may be of integer types of any bit width, but they must have the same
7688 bit width. The second element of the result structure must be of type
7689 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7695 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7696 an unsigned addition of the two arguments. They return a structure --- the
7697 first element of which is the sum, and the second element of which is a
7698 bit specifying if the unsigned summation resulted in a carry.
7703 .. code-block:: llvm
7705 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7706 %sum = extractvalue {i32, i1} %res, 0
7707 %obit = extractvalue {i32, i1} %res, 1
7708 br i1 %obit, label %carry, label %normal
7710 '``llvm.ssub.with.overflow.*``' Intrinsics
7711 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7716 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7717 on any integer bit width.
7721 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7722 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7723 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7728 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7729 a signed subtraction of the two arguments, and indicate whether an
7730 overflow occurred during the signed subtraction.
7735 The arguments (%a and %b) and the first element of the result structure
7736 may be of integer types of any bit width, but they must have the same
7737 bit width. The second element of the result structure must be of type
7738 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7744 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7745 a signed subtraction of the two arguments. They return a structure --- the
7746 first element of which is the subtraction, and the second element of
7747 which is a bit specifying if the signed subtraction resulted in an
7753 .. code-block:: llvm
7755 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7756 %sum = extractvalue {i32, i1} %res, 0
7757 %obit = extractvalue {i32, i1} %res, 1
7758 br i1 %obit, label %overflow, label %normal
7760 '``llvm.usub.with.overflow.*``' Intrinsics
7761 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7766 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7767 on any integer bit width.
7771 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7772 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7773 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7778 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7779 an unsigned subtraction of the two arguments, and indicate whether an
7780 overflow occurred during the unsigned subtraction.
7785 The arguments (%a and %b) and the first element of the result structure
7786 may be of integer types of any bit width, but they must have the same
7787 bit width. The second element of the result structure must be of type
7788 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7794 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7795 an unsigned subtraction of the two arguments. They return a structure ---
7796 the first element of which is the subtraction, and the second element of
7797 which is a bit specifying if the unsigned subtraction resulted in an
7803 .. code-block:: llvm
7805 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7806 %sum = extractvalue {i32, i1} %res, 0
7807 %obit = extractvalue {i32, i1} %res, 1
7808 br i1 %obit, label %overflow, label %normal
7810 '``llvm.smul.with.overflow.*``' Intrinsics
7811 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7816 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7817 on any integer bit width.
7821 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7822 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7823 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7828 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7829 a signed multiplication of the two arguments, and indicate whether an
7830 overflow occurred during the signed multiplication.
7835 The arguments (%a and %b) and the first element of the result structure
7836 may be of integer types of any bit width, but they must have the same
7837 bit width. The second element of the result structure must be of type
7838 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7844 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7845 a signed multiplication of the two arguments. They return a structure ---
7846 the first element of which is the multiplication, and the second element
7847 of which is a bit specifying if the signed multiplication resulted in an
7853 .. code-block:: llvm
7855 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7856 %sum = extractvalue {i32, i1} %res, 0
7857 %obit = extractvalue {i32, i1} %res, 1
7858 br i1 %obit, label %overflow, label %normal
7860 '``llvm.umul.with.overflow.*``' Intrinsics
7861 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7866 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
7867 on any integer bit width.
7871 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7872 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7873 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7878 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7879 a unsigned multiplication of the two arguments, and indicate whether an
7880 overflow occurred during the unsigned multiplication.
7885 The arguments (%a and %b) and the first element of the result structure
7886 may be of integer types of any bit width, but they must have the same
7887 bit width. The second element of the result structure must be of type
7888 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7894 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7895 an unsigned multiplication of the two arguments. They return a structure ---
7896 the first element of which is the multiplication, and the second
7897 element of which is a bit specifying if the unsigned multiplication
7898 resulted in an overflow.
7903 .. code-block:: llvm
7905 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7906 %sum = extractvalue {i32, i1} %res, 0
7907 %obit = extractvalue {i32, i1} %res, 1
7908 br i1 %obit, label %overflow, label %normal
7910 Specialised Arithmetic Intrinsics
7911 ---------------------------------
7913 '``llvm.fmuladd.*``' Intrinsic
7914 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7921 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
7922 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
7927 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
7928 expressions that can be fused if the code generator determines that (a) the
7929 target instruction set has support for a fused operation, and (b) that the
7930 fused operation is more efficient than the equivalent, separate pair of mul
7931 and add instructions.
7936 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
7937 multiplicands, a and b, and an addend c.
7946 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
7948 is equivalent to the expression a \* b + c, except that rounding will
7949 not be performed between the multiplication and addition steps if the
7950 code generator fuses the operations. Fusion is not guaranteed, even if
7951 the target platform supports it. If a fused multiply-add is required the
7952 corresponding llvm.fma.\* intrinsic function should be used instead.
7957 .. code-block:: llvm
7959 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
7961 Half Precision Floating Point Intrinsics
7962 ----------------------------------------
7964 For most target platforms, half precision floating point is a
7965 storage-only format. This means that it is a dense encoding (in memory)
7966 but does not support computation in the format.
7968 This means that code must first load the half-precision floating point
7969 value as an i16, then convert it to float with
7970 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
7971 then be performed on the float value (including extending to double
7972 etc). To store the value back to memory, it is first converted to float
7973 if needed, then converted to i16 with
7974 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
7977 .. _int_convert_to_fp16:
7979 '``llvm.convert.to.fp16``' Intrinsic
7980 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7987 declare i16 @llvm.convert.to.fp16(f32 %a)
7992 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7993 from single precision floating point format to half precision floating
7999 The intrinsic function contains single argument - the value to be
8005 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8006 from single precision floating point format to half precision floating
8007 point format. The return value is an ``i16`` which contains the
8013 .. code-block:: llvm
8015 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8016 store i16 %res, i16* @x, align 2
8018 .. _int_convert_from_fp16:
8020 '``llvm.convert.from.fp16``' Intrinsic
8021 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8028 declare f32 @llvm.convert.from.fp16(i16 %a)
8033 The '``llvm.convert.from.fp16``' intrinsic function performs a
8034 conversion from half precision floating point format to single precision
8035 floating point format.
8040 The intrinsic function contains single argument - the value to be
8046 The '``llvm.convert.from.fp16``' intrinsic function performs a
8047 conversion from half single precision floating point format to single
8048 precision floating point format. The input half-float value is
8049 represented by an ``i16`` value.
8054 .. code-block:: llvm
8056 %a = load i16* @x, align 2
8057 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8062 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8063 prefix), are described in the `LLVM Source Level
8064 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8067 Exception Handling Intrinsics
8068 -----------------------------
8070 The LLVM exception handling intrinsics (which all start with
8071 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8072 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8076 Trampoline Intrinsics
8077 ---------------------
8079 These intrinsics make it possible to excise one parameter, marked with
8080 the :ref:`nest <nest>` attribute, from a function. The result is a
8081 callable function pointer lacking the nest parameter - the caller does
8082 not need to provide a value for it. Instead, the value to use is stored
8083 in advance in a "trampoline", a block of memory usually allocated on the
8084 stack, which also contains code to splice the nest value into the
8085 argument list. This is used to implement the GCC nested function address
8088 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8089 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8090 It can be created as follows:
8092 .. code-block:: llvm
8094 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8095 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8096 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8097 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8098 %fp = bitcast i8* %p to i32 (i32, i32)*
8100 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8101 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8105 '``llvm.init.trampoline``' Intrinsic
8106 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8113 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8118 This fills the memory pointed to by ``tramp`` with executable code,
8119 turning it into a trampoline.
8124 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8125 pointers. The ``tramp`` argument must point to a sufficiently large and
8126 sufficiently aligned block of memory; this memory is written to by the
8127 intrinsic. Note that the size and the alignment are target-specific -
8128 LLVM currently provides no portable way of determining them, so a
8129 front-end that generates this intrinsic needs to have some
8130 target-specific knowledge. The ``func`` argument must hold a function
8131 bitcast to an ``i8*``.
8136 The block of memory pointed to by ``tramp`` is filled with target
8137 dependent code, turning it into a function. Then ``tramp`` needs to be
8138 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8139 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8140 function's signature is the same as that of ``func`` with any arguments
8141 marked with the ``nest`` attribute removed. At most one such ``nest``
8142 argument is allowed, and it must be of pointer type. Calling the new
8143 function is equivalent to calling ``func`` with the same argument list,
8144 but with ``nval`` used for the missing ``nest`` argument. If, after
8145 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8146 modified, then the effect of any later call to the returned function
8147 pointer is undefined.
8151 '``llvm.adjust.trampoline``' Intrinsic
8152 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8159 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8164 This performs any required machine-specific adjustment to the address of
8165 a trampoline (passed as ``tramp``).
8170 ``tramp`` must point to a block of memory which already has trampoline
8171 code filled in by a previous call to
8172 :ref:`llvm.init.trampoline <int_it>`.
8177 On some architectures the address of the code to be executed needs to be
8178 different to the address where the trampoline is actually stored. This
8179 intrinsic returns the executable address corresponding to ``tramp``
8180 after performing the required machine specific adjustments. The pointer
8181 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8186 This class of intrinsics exists to information about the lifetime of
8187 memory objects and ranges where variables are immutable.
8189 '``llvm.lifetime.start``' Intrinsic
8190 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8197 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8202 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8208 The first argument is a constant integer representing the size of the
8209 object, or -1 if it is variable sized. The second argument is a pointer
8215 This intrinsic indicates that before this point in the code, the value
8216 of the memory pointed to by ``ptr`` is dead. This means that it is known
8217 to never be used and has an undefined value. A load from the pointer
8218 that precedes this intrinsic can be replaced with ``'undef'``.
8220 '``llvm.lifetime.end``' Intrinsic
8221 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8228 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8233 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8239 The first argument is a constant integer representing the size of the
8240 object, or -1 if it is variable sized. The second argument is a pointer
8246 This intrinsic indicates that after this point in the code, the value of
8247 the memory pointed to by ``ptr`` is dead. This means that it is known to
8248 never be used and has an undefined value. Any stores into the memory
8249 object following this intrinsic may be removed as dead.
8251 '``llvm.invariant.start``' Intrinsic
8252 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8259 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8264 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8265 a memory object will not change.
8270 The first argument is a constant integer representing the size of the
8271 object, or -1 if it is variable sized. The second argument is a pointer
8277 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8278 the return value, the referenced memory location is constant and
8281 '``llvm.invariant.end``' Intrinsic
8282 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8289 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8294 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8295 memory object are mutable.
8300 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8301 The second argument is a constant integer representing the size of the
8302 object, or -1 if it is variable sized and the third argument is a
8303 pointer to the object.
8308 This intrinsic indicates that the memory is mutable again.
8313 This class of intrinsics is designed to be generic and has no specific
8316 '``llvm.var.annotation``' Intrinsic
8317 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8324 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8329 The '``llvm.var.annotation``' intrinsic.
8334 The first argument is a pointer to a value, the second is a pointer to a
8335 global string, the third is a pointer to a global string which is the
8336 source file name, and the last argument is the line number.
8341 This intrinsic allows annotation of local variables with arbitrary
8342 strings. This can be useful for special purpose optimizations that want
8343 to look for these annotations. These have no other defined use; they are
8344 ignored by code generation and optimization.
8346 '``llvm.ptr.annotation.*``' Intrinsic
8347 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8352 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8353 pointer to an integer of any width. *NOTE* you must specify an address space for
8354 the pointer. The identifier for the default address space is the integer
8359 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8360 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8361 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8362 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8363 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8368 The '``llvm.ptr.annotation``' intrinsic.
8373 The first argument is a pointer to an integer value of arbitrary bitwidth
8374 (result of some expression), the second is a pointer to a global string, the
8375 third is a pointer to a global string which is the source file name, and the
8376 last argument is the line number. It returns the value of the first argument.
8381 This intrinsic allows annotation of a pointer to an integer with arbitrary
8382 strings. This can be useful for special purpose optimizations that want to look
8383 for these annotations. These have no other defined use; they are ignored by code
8384 generation and optimization.
8386 '``llvm.annotation.*``' Intrinsic
8387 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8392 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8393 any integer bit width.
8397 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8398 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8399 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8400 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8401 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8406 The '``llvm.annotation``' intrinsic.
8411 The first argument is an integer value (result of some expression), the
8412 second is a pointer to a global string, the third is a pointer to a
8413 global string which is the source file name, and the last argument is
8414 the line number. It returns the value of the first argument.
8419 This intrinsic allows annotations to be put on arbitrary expressions
8420 with arbitrary strings. This can be useful for special purpose
8421 optimizations that want to look for these annotations. These have no
8422 other defined use; they are ignored by code generation and optimization.
8424 '``llvm.trap``' Intrinsic
8425 ^^^^^^^^^^^^^^^^^^^^^^^^^
8432 declare void @llvm.trap() noreturn nounwind
8437 The '``llvm.trap``' intrinsic.
8447 This intrinsic is lowered to the target dependent trap instruction. If
8448 the target does not have a trap instruction, this intrinsic will be
8449 lowered to a call of the ``abort()`` function.
8451 '``llvm.debugtrap``' Intrinsic
8452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8459 declare void @llvm.debugtrap() nounwind
8464 The '``llvm.debugtrap``' intrinsic.
8474 This intrinsic is lowered to code which is intended to cause an
8475 execution trap with the intention of requesting the attention of a
8478 '``llvm.stackprotector``' Intrinsic
8479 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8486 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8491 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8492 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8493 is placed on the stack before local variables.
8498 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8499 The first argument is the value loaded from the stack guard
8500 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8501 enough space to hold the value of the guard.
8506 This intrinsic causes the prologue/epilogue inserter to force the
8507 position of the ``AllocaInst`` stack slot to be before local variables
8508 on the stack. This is to ensure that if a local variable on the stack is
8509 overwritten, it will destroy the value of the guard. When the function
8510 exits, the guard on the stack is checked against the original guard. If
8511 they are different, then the program aborts by calling the
8512 ``__stack_chk_fail()`` function.
8514 '``llvm.objectsize``' Intrinsic
8515 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8522 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8523 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8528 The ``llvm.objectsize`` intrinsic is designed to provide information to
8529 the optimizers to determine at compile time whether a) an operation
8530 (like memcpy) will overflow a buffer that corresponds to an object, or
8531 b) that a runtime check for overflow isn't necessary. An object in this
8532 context means an allocation of a specific class, structure, array, or
8538 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8539 argument is a pointer to or into the ``object``. The second argument is
8540 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8541 or -1 (if false) when the object size is unknown. The second argument
8542 only accepts constants.
8547 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8548 the size of the object concerned. If the size cannot be determined at
8549 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8550 on the ``min`` argument).
8552 '``llvm.expect``' Intrinsic
8553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8560 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8561 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8566 The ``llvm.expect`` intrinsic provides information about expected (the
8567 most probable) value of ``val``, which can be used by optimizers.
8572 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8573 a value. The second argument is an expected value, this needs to be a
8574 constant value, variables are not allowed.
8579 This intrinsic is lowered to the ``val``.
8581 '``llvm.donothing``' Intrinsic
8582 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8589 declare void @llvm.donothing() nounwind readnone
8594 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8595 only intrinsic that can be called with an invoke instruction.
8605 This intrinsic does nothing, and it's removed by optimizers and ignored