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.
724 This indicates that the callee function at a call site is not
725 recognized as a built-in function. LLVM will retain the original call
726 and not replace it with equivalent code based on the semantics of the
731 Garbage Collector Names
732 -----------------------
734 Each function may specify a garbage collector name, which is simply a
739 define void @f() gc "name" { ... }
741 The compiler declares the supported values of *name*. Specifying a
742 collector which will cause the compiler to alter its output in order to
743 support the named garbage collection algorithm.
750 Attribute groups are groups of attributes that are referenced by objects within
751 the IR. They are important for keeping ``.ll`` files readable, because a lot of
752 functions will use the same set of attributes. In the degenerative case of a
753 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
754 group will capture the important command line flags used to build that file.
756 An attribute group is a module-level object. To use an attribute group, an
757 object references the attribute group's ID (e.g. ``#37``). An object may refer
758 to more than one attribute group. In that situation, the attributes from the
759 different groups are merged.
761 Here is an example of attribute groups for a function that should always be
762 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
766 ; Target-independent attributes:
767 #0 = attributes { alwaysinline alignstack=4 }
769 ; Target-dependent attributes:
770 #1 = attributes { "no-sse" }
772 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
773 define void @f() #0 #1 { ... }
780 Function attributes are set to communicate additional information about
781 a function. Function attributes are considered to be part of the
782 function, not of the function type, so functions with different function
783 attributes can have the same function type.
785 Function attributes are simple keywords that follow the type specified.
786 If multiple attributes are needed, they are space separated. For
791 define void @f() noinline { ... }
792 define void @f() alwaysinline { ... }
793 define void @f() alwaysinline optsize { ... }
794 define void @f() optsize { ... }
797 This attribute indicates that, when emitting the prologue and
798 epilogue, the backend should forcibly align the stack pointer.
799 Specify the desired alignment, which must be a power of two, in
802 This attribute indicates that the inliner should attempt to inline
803 this function into callers whenever possible, ignoring any active
804 inlining size threshold for this caller.
806 This attribute suppresses lazy symbol binding for the function. This
807 may make calls to the function faster, at the cost of extra program
808 startup time if the function is not called during program startup.
810 This attribute indicates that the source code contained a hint that
811 inlining this function is desirable (such as the "inline" keyword in
812 C/C++). It is just a hint; it imposes no requirements on the
815 This attribute disables prologue / epilogue emission for the
816 function. This can have very system-specific consequences.
818 This attribute indicates that calls to the function cannot be
819 duplicated. A call to a ``noduplicate`` function may be moved
820 within its parent function, but may not be duplicated within
823 A function containing a ``noduplicate`` call may still
824 be an inlining candidate, provided that the call is not
825 duplicated by inlining. That implies that the function has
826 internal linkage and only has one call site, so the original
827 call is dead after inlining.
829 This attributes disables implicit floating point instructions.
831 This attribute indicates that the inliner should never inline this
832 function in any situation. This attribute may not be used together
833 with the ``alwaysinline`` attribute.
835 This attribute indicates that the code generator should not use a
836 red zone, even if the target-specific ABI normally permits it.
838 This function attribute indicates that the function never returns
839 normally. This produces undefined behavior at runtime if the
840 function ever does dynamically return.
842 This function attribute indicates that the function never returns
843 with an unwind or exceptional control flow. If the function does
844 unwind, its runtime behavior is undefined.
846 This attribute suggests that optimization passes and code generator
847 passes make choices that keep the code size of this function low,
848 and otherwise do optimizations specifically to reduce code size.
850 This attribute indicates that the function computes its result (or
851 decides to unwind an exception) based strictly on its arguments,
852 without dereferencing any pointer arguments or otherwise accessing
853 any mutable state (e.g. memory, control registers, etc) visible to
854 caller functions. It does not write through any pointer arguments
855 (including ``byval`` arguments) and never changes any state visible
856 to callers. This means that it cannot unwind exceptions by calling
857 the ``C++`` exception throwing methods.
859 This attribute indicates that the function does not write through
860 any pointer arguments (including ``byval`` arguments) or otherwise
861 modify any state (e.g. memory, control registers, etc) visible to
862 caller functions. It may dereference pointer arguments and read
863 state that may be set in the caller. A readonly function always
864 returns the same value (or unwinds an exception identically) when
865 called with the same set of arguments and global state. It cannot
866 unwind an exception by calling the ``C++`` exception throwing
869 This attribute indicates that this function can return twice. The C
870 ``setjmp`` is an example of such a function. The compiler disables
871 some optimizations (like tail calls) in the caller of these
874 This attribute indicates that AddressSanitizer checks
875 (dynamic address safety analysis) are enabled for this function.
877 This attribute indicates that MemorySanitizer checks (dynamic detection
878 of accesses to uninitialized memory) are enabled for this function.
880 This attribute indicates that ThreadSanitizer checks
881 (dynamic thread safety analysis) are enabled for this function.
883 This attribute indicates that the function should emit a stack
884 smashing protector. It is in the form of a "canary" --- a random value
885 placed on the stack before the local variables that's checked upon
886 return from the function to see if it has been overwritten. A
887 heuristic is used to determine if a function needs stack protectors
888 or not. The heuristic used will enable protectors for functions with:
890 - Character arrays larger than ``ssp-buffer-size`` (default 8).
891 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
892 - Calls to alloca() with variable sizes or constant sizes greater than
895 If a function that has an ``ssp`` attribute is inlined into a
896 function that doesn't have an ``ssp`` attribute, then the resulting
897 function will have an ``ssp`` attribute.
899 This attribute indicates that the function should *always* emit a
900 stack smashing protector. This overrides the ``ssp`` function
903 If a function that has an ``sspreq`` attribute is inlined into a
904 function that doesn't have an ``sspreq`` attribute or which has an
905 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
906 an ``sspreq`` attribute.
908 This attribute indicates that the function should emit a stack smashing
909 protector. This attribute causes a strong heuristic to be used when
910 determining if a function needs stack protectors. The strong heuristic
911 will enable protectors for functions with:
913 - Arrays of any size and type
914 - Aggregates containing an array of any size and type.
916 - Local variables that have had their address taken.
918 This overrides the ``ssp`` function attribute.
920 If a function that has an ``sspstrong`` attribute is inlined into a
921 function that doesn't have an ``sspstrong`` attribute, then the
922 resulting function will have an ``sspstrong`` attribute.
924 This attribute indicates that the ABI being targeted requires that
925 an unwind table entry be produce for this function even if we can
926 show that no exceptions passes by it. This is normally the case for
927 the ELF x86-64 abi, but it can be disabled for some compilation
932 Module-Level Inline Assembly
933 ----------------------------
935 Modules may contain "module-level inline asm" blocks, which corresponds
936 to the GCC "file scope inline asm" blocks. These blocks are internally
937 concatenated by LLVM and treated as a single unit, but may be separated
938 in the ``.ll`` file if desired. The syntax is very simple:
942 module asm "inline asm code goes here"
943 module asm "more can go here"
945 The strings can contain any character by escaping non-printable
946 characters. The escape sequence used is simply "\\xx" where "xx" is the
947 two digit hex code for the number.
949 The inline asm code is simply printed to the machine code .s file when
950 assembly code is generated.
955 A module may specify a target specific data layout string that specifies
956 how data is to be laid out in memory. The syntax for the data layout is
961 target datalayout = "layout specification"
963 The *layout specification* consists of a list of specifications
964 separated by the minus sign character ('-'). Each specification starts
965 with a letter and may include other information after the letter to
966 define some aspect of the data layout. The specifications accepted are
970 Specifies that the target lays out data in big-endian form. That is,
971 the bits with the most significance have the lowest address
974 Specifies that the target lays out data in little-endian form. That
975 is, the bits with the least significance have the lowest address
978 Specifies the natural alignment of the stack in bits. Alignment
979 promotion of stack variables is limited to the natural stack
980 alignment to avoid dynamic stack realignment. The stack alignment
981 must be a multiple of 8-bits. If omitted, the natural stack
982 alignment defaults to "unspecified", which does not prevent any
983 alignment promotions.
984 ``p[n]:<size>:<abi>:<pref>``
985 This specifies the *size* of a pointer and its ``<abi>`` and
986 ``<pref>``\erred alignments for address space ``n``. All sizes are in
987 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
988 preceding ``:`` should be omitted too. The address space, ``n`` is
989 optional, and if not specified, denotes the default address space 0.
990 The value of ``n`` must be in the range [1,2^23).
991 ``i<size>:<abi>:<pref>``
992 This specifies the alignment for an integer type of a given bit
993 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
994 ``v<size>:<abi>:<pref>``
995 This specifies the alignment for a vector type of a given bit
997 ``f<size>:<abi>:<pref>``
998 This specifies the alignment for a floating point type of a given bit
999 ``<size>``. Only values of ``<size>`` that are supported by the target
1000 will work. 32 (float) and 64 (double) are supported on all targets; 80
1001 or 128 (different flavors of long double) are also supported on some
1003 ``a<size>:<abi>:<pref>``
1004 This specifies the alignment for an aggregate type of a given bit
1006 ``s<size>:<abi>:<pref>``
1007 This specifies the alignment for a stack object of a given bit
1009 ``n<size1>:<size2>:<size3>...``
1010 This specifies a set of native integer widths for the target CPU in
1011 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1012 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1013 this set are considered to support most general arithmetic operations
1016 When constructing the data layout for a given target, LLVM starts with a
1017 default set of specifications which are then (possibly) overridden by
1018 the specifications in the ``datalayout`` keyword. The default
1019 specifications are given in this list:
1021 - ``E`` - big endian
1022 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment
1023 - ``S0`` - natural stack alignment is unspecified
1024 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1025 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1026 - ``i16:16:16`` - i16 is 16-bit aligned
1027 - ``i32:32:32`` - i32 is 32-bit aligned
1028 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1029 alignment of 64-bits
1030 - ``f16:16:16`` - half is 16-bit aligned
1031 - ``f32:32:32`` - float is 32-bit aligned
1032 - ``f64:64:64`` - double is 64-bit aligned
1033 - ``f128:128:128`` - quad is 128-bit aligned
1034 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1035 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1036 - ``a0:0:64`` - aggregates are 64-bit aligned
1038 When LLVM is determining the alignment for a given type, it uses the
1041 #. If the type sought is an exact match for one of the specifications,
1042 that specification is used.
1043 #. If no match is found, and the type sought is an integer type, then
1044 the smallest integer type that is larger than the bitwidth of the
1045 sought type is used. If none of the specifications are larger than
1046 the bitwidth then the largest integer type is used. For example,
1047 given the default specifications above, the i7 type will use the
1048 alignment of i8 (next largest) while both i65 and i256 will use the
1049 alignment of i64 (largest specified).
1050 #. If no match is found, and the type sought is a vector type, then the
1051 largest vector type that is smaller than the sought vector type will
1052 be used as a fall back. This happens because <128 x double> can be
1053 implemented in terms of 64 <2 x double>, for example.
1055 The function of the data layout string may not be what you expect.
1056 Notably, this is not a specification from the frontend of what alignment
1057 the code generator should use.
1059 Instead, if specified, the target data layout is required to match what
1060 the ultimate *code generator* expects. This string is used by the
1061 mid-level optimizers to improve code, and this only works if it matches
1062 what the ultimate code generator uses. If you would like to generate IR
1063 that does not embed this target-specific detail into the IR, then you
1064 don't have to specify the string. This will disable some optimizations
1065 that require precise layout information, but this also prevents those
1066 optimizations from introducing target specificity into the IR.
1068 .. _pointeraliasing:
1070 Pointer Aliasing Rules
1071 ----------------------
1073 Any memory access must be done through a pointer value associated with
1074 an address range of the memory access, otherwise the behavior is
1075 undefined. Pointer values are associated with address ranges according
1076 to the following rules:
1078 - A pointer value is associated with the addresses associated with any
1079 value it is *based* on.
1080 - An address of a global variable is associated with the address range
1081 of the variable's storage.
1082 - The result value of an allocation instruction is associated with the
1083 address range of the allocated storage.
1084 - A null pointer in the default address-space is associated with no
1086 - An integer constant other than zero or a pointer value returned from
1087 a function not defined within LLVM may be associated with address
1088 ranges allocated through mechanisms other than those provided by
1089 LLVM. Such ranges shall not overlap with any ranges of addresses
1090 allocated by mechanisms provided by LLVM.
1092 A pointer value is *based* on another pointer value according to the
1095 - A pointer value formed from a ``getelementptr`` operation is *based*
1096 on the first operand of the ``getelementptr``.
1097 - The result value of a ``bitcast`` is *based* on the operand of the
1099 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1100 values that contribute (directly or indirectly) to the computation of
1101 the pointer's value.
1102 - The "*based* on" relationship is transitive.
1104 Note that this definition of *"based"* is intentionally similar to the
1105 definition of *"based"* in C99, though it is slightly weaker.
1107 LLVM IR does not associate types with memory. The result type of a
1108 ``load`` merely indicates the size and alignment of the memory from
1109 which to load, as well as the interpretation of the value. The first
1110 operand type of a ``store`` similarly only indicates the size and
1111 alignment of the store.
1113 Consequently, type-based alias analysis, aka TBAA, aka
1114 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1115 :ref:`Metadata <metadata>` may be used to encode additional information
1116 which specialized optimization passes may use to implement type-based
1121 Volatile Memory Accesses
1122 ------------------------
1124 Certain memory accesses, such as :ref:`load <i_load>`'s,
1125 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1126 marked ``volatile``. The optimizers must not change the number of
1127 volatile operations or change their order of execution relative to other
1128 volatile operations. The optimizers *may* change the order of volatile
1129 operations relative to non-volatile operations. This is not Java's
1130 "volatile" and has no cross-thread synchronization behavior.
1132 IR-level volatile loads and stores cannot safely be optimized into
1133 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1134 flagged volatile. Likewise, the backend should never split or merge
1135 target-legal volatile load/store instructions.
1137 .. admonition:: Rationale
1139 Platforms may rely on volatile loads and stores of natively supported
1140 data width to be executed as single instruction. For example, in C
1141 this holds for an l-value of volatile primitive type with native
1142 hardware support, but not necessarily for aggregate types. The
1143 frontend upholds these expectations, which are intentionally
1144 unspecified in the IR. The rules above ensure that IR transformation
1145 do not violate the frontend's contract with the language.
1149 Memory Model for Concurrent Operations
1150 --------------------------------------
1152 The LLVM IR does not define any way to start parallel threads of
1153 execution or to register signal handlers. Nonetheless, there are
1154 platform-specific ways to create them, and we define LLVM IR's behavior
1155 in their presence. This model is inspired by the C++0x memory model.
1157 For a more informal introduction to this model, see the :doc:`Atomics`.
1159 We define a *happens-before* partial order as the least partial order
1162 - Is a superset of single-thread program order, and
1163 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1164 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1165 techniques, like pthread locks, thread creation, thread joining,
1166 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1167 Constraints <ordering>`).
1169 Note that program order does not introduce *happens-before* edges
1170 between a thread and signals executing inside that thread.
1172 Every (defined) read operation (load instructions, memcpy, atomic
1173 loads/read-modify-writes, etc.) R reads a series of bytes written by
1174 (defined) write operations (store instructions, atomic
1175 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1176 section, initialized globals are considered to have a write of the
1177 initializer which is atomic and happens before any other read or write
1178 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1179 may see any write to the same byte, except:
1181 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1182 write\ :sub:`2` happens before R\ :sub:`byte`, then
1183 R\ :sub:`byte` does not see write\ :sub:`1`.
1184 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1185 R\ :sub:`byte` does not see write\ :sub:`3`.
1187 Given that definition, R\ :sub:`byte` is defined as follows:
1189 - If R is volatile, the result is target-dependent. (Volatile is
1190 supposed to give guarantees which can support ``sig_atomic_t`` in
1191 C/C++, and may be used for accesses to addresses which do not behave
1192 like normal memory. It does not generally provide cross-thread
1194 - Otherwise, if there is no write to the same byte that happens before
1195 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1196 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1197 R\ :sub:`byte` returns the value written by that write.
1198 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1199 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1200 Memory Ordering Constraints <ordering>` section for additional
1201 constraints on how the choice is made.
1202 - Otherwise R\ :sub:`byte` returns ``undef``.
1204 R returns the value composed of the series of bytes it read. This
1205 implies that some bytes within the value may be ``undef`` **without**
1206 the entire value being ``undef``. Note that this only defines the
1207 semantics of the operation; it doesn't mean that targets will emit more
1208 than one instruction to read the series of bytes.
1210 Note that in cases where none of the atomic intrinsics are used, this
1211 model places only one restriction on IR transformations on top of what
1212 is required for single-threaded execution: introducing a store to a byte
1213 which might not otherwise be stored is not allowed in general.
1214 (Specifically, in the case where another thread might write to and read
1215 from an address, introducing a store can change a load that may see
1216 exactly one write into a load that may see multiple writes.)
1220 Atomic Memory Ordering Constraints
1221 ----------------------------------
1223 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1224 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1225 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1226 an ordering parameter that determines which other atomic instructions on
1227 the same address they *synchronize with*. These semantics are borrowed
1228 from Java and C++0x, but are somewhat more colloquial. If these
1229 descriptions aren't precise enough, check those specs (see spec
1230 references in the :doc:`atomics guide <Atomics>`).
1231 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1232 differently since they don't take an address. See that instruction's
1233 documentation for details.
1235 For a simpler introduction to the ordering constraints, see the
1239 The set of values that can be read is governed by the happens-before
1240 partial order. A value cannot be read unless some operation wrote
1241 it. This is intended to provide a guarantee strong enough to model
1242 Java's non-volatile shared variables. This ordering cannot be
1243 specified for read-modify-write operations; it is not strong enough
1244 to make them atomic in any interesting way.
1246 In addition to the guarantees of ``unordered``, there is a single
1247 total order for modifications by ``monotonic`` operations on each
1248 address. All modification orders must be compatible with the
1249 happens-before order. There is no guarantee that the modification
1250 orders can be combined to a global total order for the whole program
1251 (and this often will not be possible). The read in an atomic
1252 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1253 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1254 order immediately before the value it writes. If one atomic read
1255 happens before another atomic read of the same address, the later
1256 read must see the same value or a later value in the address's
1257 modification order. This disallows reordering of ``monotonic`` (or
1258 stronger) operations on the same address. If an address is written
1259 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1260 read that address repeatedly, the other threads must eventually see
1261 the write. This corresponds to the C++0x/C1x
1262 ``memory_order_relaxed``.
1264 In addition to the guarantees of ``monotonic``, a
1265 *synchronizes-with* edge may be formed with a ``release`` operation.
1266 This is intended to model C++'s ``memory_order_acquire``.
1268 In addition to the guarantees of ``monotonic``, if this operation
1269 writes a value which is subsequently read by an ``acquire``
1270 operation, it *synchronizes-with* that operation. (This isn't a
1271 complete description; see the C++0x definition of a release
1272 sequence.) This corresponds to the C++0x/C1x
1273 ``memory_order_release``.
1274 ``acq_rel`` (acquire+release)
1275 Acts as both an ``acquire`` and ``release`` operation on its
1276 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1277 ``seq_cst`` (sequentially consistent)
1278 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1279 operation which only reads, ``release`` for an operation which only
1280 writes), there is a global total order on all
1281 sequentially-consistent operations on all addresses, which is
1282 consistent with the *happens-before* partial order and with the
1283 modification orders of all the affected addresses. Each
1284 sequentially-consistent read sees the last preceding write to the
1285 same address in this global order. This corresponds to the C++0x/C1x
1286 ``memory_order_seq_cst`` and Java volatile.
1290 If an atomic operation is marked ``singlethread``, it only *synchronizes
1291 with* or participates in modification and seq\_cst total orderings with
1292 other operations running in the same thread (for example, in signal
1300 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1301 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1302 :ref:`frem <i_frem>`) have the following flags that can set to enable
1303 otherwise unsafe floating point operations
1306 No NaNs - Allow optimizations to assume the arguments and result are not
1307 NaN. Such optimizations are required to retain defined behavior over
1308 NaNs, but the value of the result is undefined.
1311 No Infs - Allow optimizations to assume the arguments and result are not
1312 +/-Inf. Such optimizations are required to retain defined behavior over
1313 +/-Inf, but the value of the result is undefined.
1316 No Signed Zeros - Allow optimizations to treat the sign of a zero
1317 argument or result as insignificant.
1320 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1321 argument rather than perform division.
1324 Fast - Allow algebraically equivalent transformations that may
1325 dramatically change results in floating point (e.g. reassociate). This
1326 flag implies all the others.
1333 The LLVM type system is one of the most important features of the
1334 intermediate representation. Being typed enables a number of
1335 optimizations to be performed on the intermediate representation
1336 directly, without having to do extra analyses on the side before the
1337 transformation. A strong type system makes it easier to read the
1338 generated code and enables novel analyses and transformations that are
1339 not feasible to perform on normal three address code representations.
1341 Type Classifications
1342 --------------------
1344 The types fall into a few useful classifications:
1353 * - :ref:`integer <t_integer>`
1354 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1357 * - :ref:`floating point <t_floating>`
1358 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1366 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1367 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1368 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1369 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1371 * - :ref:`primitive <t_primitive>`
1372 - :ref:`label <t_label>`,
1373 :ref:`void <t_void>`,
1374 :ref:`integer <t_integer>`,
1375 :ref:`floating point <t_floating>`,
1376 :ref:`x86mmx <t_x86mmx>`,
1377 :ref:`metadata <t_metadata>`.
1379 * - :ref:`derived <t_derived>`
1380 - :ref:`array <t_array>`,
1381 :ref:`function <t_function>`,
1382 :ref:`pointer <t_pointer>`,
1383 :ref:`structure <t_struct>`,
1384 :ref:`vector <t_vector>`,
1385 :ref:`opaque <t_opaque>`.
1387 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1388 Values of these types are the only ones which can be produced by
1396 The primitive types are the fundamental building blocks of the LLVM
1407 The integer type is a very simple type that simply specifies an
1408 arbitrary bit width for the integer type desired. Any bit width from 1
1409 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1418 The number of bits the integer will occupy is specified by the ``N``
1424 +----------------+------------------------------------------------+
1425 | ``i1`` | a single-bit integer. |
1426 +----------------+------------------------------------------------+
1427 | ``i32`` | a 32-bit integer. |
1428 +----------------+------------------------------------------------+
1429 | ``i1942652`` | a really big integer of over 1 million bits. |
1430 +----------------+------------------------------------------------+
1434 Floating Point Types
1435 ^^^^^^^^^^^^^^^^^^^^
1444 - 16-bit floating point value
1447 - 32-bit floating point value
1450 - 64-bit floating point value
1453 - 128-bit floating point value (112-bit mantissa)
1456 - 80-bit floating point value (X87)
1459 - 128-bit floating point value (two 64-bits)
1469 The x86mmx type represents a value held in an MMX register on an x86
1470 machine. The operations allowed on it are quite limited: parameters and
1471 return values, load and store, and bitcast. User-specified MMX
1472 instructions are represented as intrinsic or asm calls with arguments
1473 and/or results of this type. There are no arrays, vectors or constants
1491 The void type does not represent any value and has no size.
1508 The label type represents code labels.
1525 The metadata type represents embedded metadata. No derived types may be
1526 created from metadata except for :ref:`function <t_function>` arguments.
1540 The real power in LLVM comes from the derived types in the system. This
1541 is what allows a programmer to represent arrays, functions, pointers,
1542 and other useful types. Each of these types contain one or more element
1543 types which may be a primitive type, or another derived type. For
1544 example, it is possible to have a two dimensional array, using an array
1545 as the element type of another array.
1552 Aggregate Types are a subset of derived types that can contain multiple
1553 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1554 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1565 The array type is a very simple derived type that arranges elements
1566 sequentially in memory. The array type requires a size (number of
1567 elements) and an underlying data type.
1574 [<# elements> x <elementtype>]
1576 The number of elements is a constant integer value; ``elementtype`` may
1577 be any type with a size.
1582 +------------------+--------------------------------------+
1583 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1584 +------------------+--------------------------------------+
1585 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1586 +------------------+--------------------------------------+
1587 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1588 +------------------+--------------------------------------+
1590 Here are some examples of multidimensional arrays:
1592 +-----------------------------+----------------------------------------------------------+
1593 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1594 +-----------------------------+----------------------------------------------------------+
1595 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1596 +-----------------------------+----------------------------------------------------------+
1597 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1598 +-----------------------------+----------------------------------------------------------+
1600 There is no restriction on indexing beyond the end of the array implied
1601 by a static type (though there are restrictions on indexing beyond the
1602 bounds of an allocated object in some cases). This means that
1603 single-dimension 'variable sized array' addressing can be implemented in
1604 LLVM with a zero length array type. An implementation of 'pascal style
1605 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1616 The function type can be thought of as a function signature. It consists
1617 of a return type and a list of formal parameter types. The return type
1618 of a function type is a first class type or a void type.
1625 <returntype> (<parameter list>)
1627 ...where '``<parameter list>``' is a comma-separated list of type
1628 specifiers. Optionally, the parameter list may include a type ``...``,
1629 which indicates that the function takes a variable number of arguments.
1630 Variable argument functions can access their arguments with the
1631 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1632 '``<returntype>``' is any type except :ref:`label <t_label>`.
1637 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1638 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1639 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1640 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1641 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1642 | ``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. |
1643 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1644 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1645 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1655 The structure type is used to represent a collection of data members
1656 together in memory. The elements of a structure may be any type that has
1659 Structures in memory are accessed using '``load``' and '``store``' by
1660 getting a pointer to a field with the '``getelementptr``' instruction.
1661 Structures in registers are accessed using the '``extractvalue``' and
1662 '``insertvalue``' instructions.
1664 Structures may optionally be "packed" structures, which indicate that
1665 the alignment of the struct is one byte, and that there is no padding
1666 between the elements. In non-packed structs, padding between field types
1667 is inserted as defined by the DataLayout string in the module, which is
1668 required to match what the underlying code generator expects.
1670 Structures can either be "literal" or "identified". A literal structure
1671 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1672 identified types are always defined at the top level with a name.
1673 Literal types are uniqued by their contents and can never be recursive
1674 or opaque since there is no way to write one. Identified types can be
1675 recursive, can be opaqued, and are never uniqued.
1682 %T1 = type { <type list> } ; Identified normal struct type
1683 %T2 = type <{ <type list> }> ; Identified packed struct type
1688 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1689 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1690 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1691 | ``{ 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``. |
1692 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1693 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1694 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1698 Opaque Structure Types
1699 ^^^^^^^^^^^^^^^^^^^^^^
1704 Opaque structure types are used to represent named structure types that
1705 do not have a body specified. This corresponds (for example) to the C
1706 notion of a forward declared structure.
1719 +--------------+-------------------+
1720 | ``opaque`` | An opaque type. |
1721 +--------------+-------------------+
1731 The pointer type is used to specify memory locations. Pointers are
1732 commonly used to reference objects in memory.
1734 Pointer types may have an optional address space attribute defining the
1735 numbered address space where the pointed-to object resides. The default
1736 address space is number zero. The semantics of non-zero address spaces
1737 are target-specific.
1739 Note that LLVM does not permit pointers to void (``void*``) nor does it
1740 permit pointers to labels (``label*``). Use ``i8*`` instead.
1752 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1753 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1754 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1755 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1756 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1757 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1758 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1768 A vector type is a simple derived type that represents a vector of
1769 elements. Vector types are used when multiple primitive data are
1770 operated in parallel using a single instruction (SIMD). A vector type
1771 requires a size (number of elements) and an underlying primitive data
1772 type. Vector types are considered :ref:`first class <t_firstclass>`.
1779 < <# elements> x <elementtype> >
1781 The number of elements is a constant integer value larger than 0;
1782 elementtype may be any integer or floating point type, or a pointer to
1783 these types. Vectors of size zero are not allowed.
1788 +-------------------+--------------------------------------------------+
1789 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1790 +-------------------+--------------------------------------------------+
1791 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1792 +-------------------+--------------------------------------------------+
1793 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1794 +-------------------+--------------------------------------------------+
1795 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1796 +-------------------+--------------------------------------------------+
1801 LLVM has several different basic types of constants. This section
1802 describes them all and their syntax.
1807 **Boolean constants**
1808 The two strings '``true``' and '``false``' are both valid constants
1810 **Integer constants**
1811 Standard integers (such as '4') are constants of the
1812 :ref:`integer <t_integer>` type. Negative numbers may be used with
1814 **Floating point constants**
1815 Floating point constants use standard decimal notation (e.g.
1816 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1817 hexadecimal notation (see below). The assembler requires the exact
1818 decimal value of a floating-point constant. For example, the
1819 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1820 decimal in binary. Floating point constants must have a :ref:`floating
1821 point <t_floating>` type.
1822 **Null pointer constants**
1823 The identifier '``null``' is recognized as a null pointer constant
1824 and must be of :ref:`pointer type <t_pointer>`.
1826 The one non-intuitive notation for constants is the hexadecimal form of
1827 floating point constants. For example, the form
1828 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1829 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1830 constants are required (and the only time that they are generated by the
1831 disassembler) is when a floating point constant must be emitted but it
1832 cannot be represented as a decimal floating point number in a reasonable
1833 number of digits. For example, NaN's, infinities, and other special
1834 values are represented in their IEEE hexadecimal format so that assembly
1835 and disassembly do not cause any bits to change in the constants.
1837 When using the hexadecimal form, constants of types half, float, and
1838 double are represented using the 16-digit form shown above (which
1839 matches the IEEE754 representation for double); half and float values
1840 must, however, be exactly representable as IEEE 754 half and single
1841 precision, respectively. Hexadecimal format is always used for long
1842 double, and there are three forms of long double. The 80-bit format used
1843 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1844 128-bit format used by PowerPC (two adjacent doubles) is represented by
1845 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1846 represented by ``0xL`` followed by 32 hexadecimal digits; no currently
1847 supported target uses this format. Long doubles will only work if they
1848 match the long double format on your target. The IEEE 16-bit format
1849 (half precision) is represented by ``0xH`` followed by 4 hexadecimal
1850 digits. All hexadecimal formats are big-endian (sign bit at the left).
1852 There are no constants of type x86mmx.
1857 Complex constants are a (potentially recursive) combination of simple
1858 constants and smaller complex constants.
1860 **Structure constants**
1861 Structure constants are represented with notation similar to
1862 structure type definitions (a comma separated list of elements,
1863 surrounded by braces (``{}``)). For example:
1864 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1865 "``@G = external global i32``". Structure constants must have
1866 :ref:`structure type <t_struct>`, and the number and types of elements
1867 must match those specified by the type.
1869 Array constants are represented with notation similar to array type
1870 definitions (a comma separated list of elements, surrounded by
1871 square brackets (``[]``)). For example:
1872 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1873 :ref:`array type <t_array>`, and the number and types of elements must
1874 match those specified by the type.
1875 **Vector constants**
1876 Vector constants are represented with notation similar to vector
1877 type definitions (a comma separated list of elements, surrounded by
1878 less-than/greater-than's (``<>``)). For example:
1879 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1880 must have :ref:`vector type <t_vector>`, and the number and types of
1881 elements must match those specified by the type.
1882 **Zero initialization**
1883 The string '``zeroinitializer``' can be used to zero initialize a
1884 value to zero of *any* type, including scalar and
1885 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1886 having to print large zero initializers (e.g. for large arrays) and
1887 is always exactly equivalent to using explicit zero initializers.
1889 A metadata node is a structure-like constant with :ref:`metadata
1890 type <t_metadata>`. For example:
1891 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1892 constants that are meant to be interpreted as part of the
1893 instruction stream, metadata is a place to attach additional
1894 information such as debug info.
1896 Global Variable and Function Addresses
1897 --------------------------------------
1899 The addresses of :ref:`global variables <globalvars>` and
1900 :ref:`functions <functionstructure>` are always implicitly valid
1901 (link-time) constants. These constants are explicitly referenced when
1902 the :ref:`identifier for the global <identifiers>` is used and always have
1903 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1906 .. code-block:: llvm
1910 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1917 The string '``undef``' can be used anywhere a constant is expected, and
1918 indicates that the user of the value may receive an unspecified
1919 bit-pattern. Undefined values may be of any type (other than '``label``'
1920 or '``void``') and be used anywhere a constant is permitted.
1922 Undefined values are useful because they indicate to the compiler that
1923 the program is well defined no matter what value is used. This gives the
1924 compiler more freedom to optimize. Here are some examples of
1925 (potentially surprising) transformations that are valid (in pseudo IR):
1927 .. code-block:: llvm
1937 This is safe because all of the output bits are affected by the undef
1938 bits. Any output bit can have a zero or one depending on the input bits.
1940 .. code-block:: llvm
1951 These logical operations have bits that are not always affected by the
1952 input. For example, if ``%X`` has a zero bit, then the output of the
1953 '``and``' operation will always be a zero for that bit, no matter what
1954 the corresponding bit from the '``undef``' is. As such, it is unsafe to
1955 optimize or assume that the result of the '``and``' is '``undef``'.
1956 However, it is safe to assume that all bits of the '``undef``' could be
1957 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
1958 all the bits of the '``undef``' operand to the '``or``' could be set,
1959 allowing the '``or``' to be folded to -1.
1961 .. code-block:: llvm
1963 %A = select undef, %X, %Y
1964 %B = select undef, 42, %Y
1965 %C = select %X, %Y, undef
1975 This set of examples shows that undefined '``select``' (and conditional
1976 branch) conditions can go *either way*, but they have to come from one
1977 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
1978 both known to have a clear low bit, then ``%A`` would have to have a
1979 cleared low bit. However, in the ``%C`` example, the optimizer is
1980 allowed to assume that the '``undef``' operand could be the same as
1981 ``%Y``, allowing the whole '``select``' to be eliminated.
1983 .. code-block:: llvm
1985 %A = xor undef, undef
2002 This example points out that two '``undef``' operands are not
2003 necessarily the same. This can be surprising to people (and also matches
2004 C semantics) where they assume that "``X^X``" is always zero, even if
2005 ``X`` is undefined. This isn't true for a number of reasons, but the
2006 short answer is that an '``undef``' "variable" can arbitrarily change
2007 its value over its "live range". This is true because the variable
2008 doesn't actually *have a live range*. Instead, the value is logically
2009 read from arbitrary registers that happen to be around when needed, so
2010 the value is not necessarily consistent over time. In fact, ``%A`` and
2011 ``%C`` need to have the same semantics or the core LLVM "replace all
2012 uses with" concept would not hold.
2014 .. code-block:: llvm
2022 These examples show the crucial difference between an *undefined value*
2023 and *undefined behavior*. An undefined value (like '``undef``') is
2024 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2025 operation can be constant folded to '``undef``', because the '``undef``'
2026 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2027 However, in the second example, we can make a more aggressive
2028 assumption: because the ``undef`` is allowed to be an arbitrary value,
2029 we are allowed to assume that it could be zero. Since a divide by zero
2030 has *undefined behavior*, we are allowed to assume that the operation
2031 does not execute at all. This allows us to delete the divide and all
2032 code after it. Because the undefined operation "can't happen", the
2033 optimizer can assume that it occurs in dead code.
2035 .. code-block:: llvm
2037 a: store undef -> %X
2038 b: store %X -> undef
2043 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2044 value can be assumed to not have any effect; we can assume that the
2045 value is overwritten with bits that happen to match what was already
2046 there. However, a store *to* an undefined location could clobber
2047 arbitrary memory, therefore, it has undefined behavior.
2054 Poison values are similar to :ref:`undef values <undefvalues>`, however
2055 they also represent the fact that an instruction or constant expression
2056 which cannot evoke side effects has nevertheless detected a condition
2057 which results in undefined behavior.
2059 There is currently no way of representing a poison value in the IR; they
2060 only exist when produced by operations such as :ref:`add <i_add>` with
2063 Poison value behavior is defined in terms of value *dependence*:
2065 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2066 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2067 their dynamic predecessor basic block.
2068 - Function arguments depend on the corresponding actual argument values
2069 in the dynamic callers of their functions.
2070 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2071 instructions that dynamically transfer control back to them.
2072 - :ref:`Invoke <i_invoke>` instructions depend on the
2073 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2074 call instructions that dynamically transfer control back to them.
2075 - Non-volatile loads and stores depend on the most recent stores to all
2076 of the referenced memory addresses, following the order in the IR
2077 (including loads and stores implied by intrinsics such as
2078 :ref:`@llvm.memcpy <int_memcpy>`.)
2079 - An instruction with externally visible side effects depends on the
2080 most recent preceding instruction with externally visible side
2081 effects, following the order in the IR. (This includes :ref:`volatile
2082 operations <volatile>`.)
2083 - An instruction *control-depends* on a :ref:`terminator
2084 instruction <terminators>` if the terminator instruction has
2085 multiple successors and the instruction is always executed when
2086 control transfers to one of the successors, and may not be executed
2087 when control is transferred to another.
2088 - Additionally, an instruction also *control-depends* on a terminator
2089 instruction if the set of instructions it otherwise depends on would
2090 be different if the terminator had transferred control to a different
2092 - Dependence is transitive.
2094 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2095 with the additional affect that any instruction which has a *dependence*
2096 on a poison value has undefined behavior.
2098 Here are some examples:
2100 .. code-block:: llvm
2103 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2104 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2105 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2106 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2108 store i32 %poison, i32* @g ; Poison value stored to memory.
2109 %poison2 = load i32* @g ; Poison value loaded back from memory.
2111 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2113 %narrowaddr = bitcast i32* @g to i16*
2114 %wideaddr = bitcast i32* @g to i64*
2115 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2116 %poison4 = load i64* %wideaddr ; Returns a poison value.
2118 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2119 br i1 %cmp, label %true, label %end ; Branch to either destination.
2122 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2123 ; it has undefined behavior.
2127 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2128 ; Both edges into this PHI are
2129 ; control-dependent on %cmp, so this
2130 ; always results in a poison value.
2132 store volatile i32 0, i32* @g ; This would depend on the store in %true
2133 ; if %cmp is true, or the store in %entry
2134 ; otherwise, so this is undefined behavior.
2136 br i1 %cmp, label %second_true, label %second_end
2137 ; The same branch again, but this time the
2138 ; true block doesn't have side effects.
2145 store volatile i32 0, i32* @g ; This time, the instruction always depends
2146 ; on the store in %end. Also, it is
2147 ; control-equivalent to %end, so this is
2148 ; well-defined (ignoring earlier undefined
2149 ; behavior in this example).
2153 Addresses of Basic Blocks
2154 -------------------------
2156 ``blockaddress(@function, %block)``
2158 The '``blockaddress``' constant computes the address of the specified
2159 basic block in the specified function, and always has an ``i8*`` type.
2160 Taking the address of the entry block is illegal.
2162 This value only has defined behavior when used as an operand to the
2163 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2164 against null. Pointer equality tests between labels addresses results in
2165 undefined behavior --- though, again, comparison against null is ok, and
2166 no label is equal to the null pointer. This may be passed around as an
2167 opaque pointer sized value as long as the bits are not inspected. This
2168 allows ``ptrtoint`` and arithmetic to be performed on these values so
2169 long as the original value is reconstituted before the ``indirectbr``
2172 Finally, some targets may provide defined semantics when using the value
2173 as the operand to an inline assembly, but that is target specific.
2175 Constant Expressions
2176 --------------------
2178 Constant expressions are used to allow expressions involving other
2179 constants to be used as constants. Constant expressions may be of any
2180 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2181 that does not have side effects (e.g. load and call are not supported).
2182 The following is the syntax for constant expressions:
2184 ``trunc (CST to TYPE)``
2185 Truncate a constant to another type. The bit size of CST must be
2186 larger than the bit size of TYPE. Both types must be integers.
2187 ``zext (CST to TYPE)``
2188 Zero extend a constant to another type. The bit size of CST must be
2189 smaller than the bit size of TYPE. Both types must be integers.
2190 ``sext (CST to TYPE)``
2191 Sign extend a constant to another type. The bit size of CST must be
2192 smaller than the bit size of TYPE. Both types must be integers.
2193 ``fptrunc (CST to TYPE)``
2194 Truncate a floating point constant to another floating point type.
2195 The size of CST must be larger than the size of TYPE. Both types
2196 must be floating point.
2197 ``fpext (CST to TYPE)``
2198 Floating point extend a constant to another type. The size of CST
2199 must be smaller or equal to the size of TYPE. Both types must be
2201 ``fptoui (CST to TYPE)``
2202 Convert a floating point constant to the corresponding unsigned
2203 integer constant. TYPE must be a scalar or vector integer type. CST
2204 must be of scalar or vector floating point type. Both CST and TYPE
2205 must be scalars, or vectors of the same number of elements. If the
2206 value won't fit in the integer type, the results are undefined.
2207 ``fptosi (CST to TYPE)``
2208 Convert a floating point constant to the corresponding signed
2209 integer constant. TYPE must be a scalar or vector integer type. CST
2210 must be of scalar or vector floating point type. Both CST and TYPE
2211 must be scalars, or vectors of the same number of elements. If the
2212 value won't fit in the integer type, the results are undefined.
2213 ``uitofp (CST to TYPE)``
2214 Convert an unsigned integer constant to the corresponding floating
2215 point constant. TYPE must be a scalar or vector floating point type.
2216 CST must be of scalar or vector integer type. Both CST and TYPE must
2217 be scalars, or vectors of the same number of elements. If the value
2218 won't fit in the floating point type, the results are undefined.
2219 ``sitofp (CST to TYPE)``
2220 Convert a signed integer constant to the corresponding floating
2221 point constant. TYPE must be a scalar or vector floating point type.
2222 CST must be of scalar or vector integer type. Both CST and TYPE must
2223 be scalars, or vectors of the same number of elements. If the value
2224 won't fit in the floating point type, the results are undefined.
2225 ``ptrtoint (CST to TYPE)``
2226 Convert a pointer typed constant to the corresponding integer
2227 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2228 pointer type. The ``CST`` value is zero extended, truncated, or
2229 unchanged to make it fit in ``TYPE``.
2230 ``inttoptr (CST to TYPE)``
2231 Convert an integer constant to a pointer constant. TYPE must be a
2232 pointer type. CST must be of integer type. The CST value is zero
2233 extended, truncated, or unchanged to make it fit in a pointer size.
2234 This one is *really* dangerous!
2235 ``bitcast (CST to TYPE)``
2236 Convert a constant, CST, to another TYPE. The constraints of the
2237 operands are the same as those for the :ref:`bitcast
2238 instruction <i_bitcast>`.
2239 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2240 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2241 constants. As with the :ref:`getelementptr <i_getelementptr>`
2242 instruction, the index list may have zero or more indexes, which are
2243 required to make sense for the type of "CSTPTR".
2244 ``select (COND, VAL1, VAL2)``
2245 Perform the :ref:`select operation <i_select>` on constants.
2246 ``icmp COND (VAL1, VAL2)``
2247 Performs the :ref:`icmp operation <i_icmp>` on constants.
2248 ``fcmp COND (VAL1, VAL2)``
2249 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2250 ``extractelement (VAL, IDX)``
2251 Perform the :ref:`extractelement operation <i_extractelement>` on
2253 ``insertelement (VAL, ELT, IDX)``
2254 Perform the :ref:`insertelement operation <i_insertelement>` on
2256 ``shufflevector (VEC1, VEC2, IDXMASK)``
2257 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2259 ``extractvalue (VAL, IDX0, IDX1, ...)``
2260 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2261 constants. The index list is interpreted in a similar manner as
2262 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2263 least one index value must be specified.
2264 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2265 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2266 The index list is interpreted in a similar manner as indices in a
2267 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2268 value must be specified.
2269 ``OPCODE (LHS, RHS)``
2270 Perform the specified operation of the LHS and RHS constants. OPCODE
2271 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2272 binary <bitwiseops>` operations. The constraints on operands are
2273 the same as those for the corresponding instruction (e.g. no bitwise
2274 operations on floating point values are allowed).
2279 Inline Assembler Expressions
2280 ----------------------------
2282 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2283 Inline Assembly <moduleasm>`) through the use of a special value. This
2284 value represents the inline assembler as a string (containing the
2285 instructions to emit), a list of operand constraints (stored as a
2286 string), a flag that indicates whether or not the inline asm expression
2287 has side effects, and a flag indicating whether the function containing
2288 the asm needs to align its stack conservatively. An example inline
2289 assembler expression is:
2291 .. code-block:: llvm
2293 i32 (i32) asm "bswap $0", "=r,r"
2295 Inline assembler expressions may **only** be used as the callee operand
2296 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2297 Thus, typically we have:
2299 .. code-block:: llvm
2301 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2303 Inline asms with side effects not visible in the constraint list must be
2304 marked as having side effects. This is done through the use of the
2305 '``sideeffect``' keyword, like so:
2307 .. code-block:: llvm
2309 call void asm sideeffect "eieio", ""()
2311 In some cases inline asms will contain code that will not work unless
2312 the stack is aligned in some way, such as calls or SSE instructions on
2313 x86, yet will not contain code that does that alignment within the asm.
2314 The compiler should make conservative assumptions about what the asm
2315 might contain and should generate its usual stack alignment code in the
2316 prologue if the '``alignstack``' keyword is present:
2318 .. code-block:: llvm
2320 call void asm alignstack "eieio", ""()
2322 Inline asms also support using non-standard assembly dialects. The
2323 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2324 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2325 the only supported dialects. An example is:
2327 .. code-block:: llvm
2329 call void asm inteldialect "eieio", ""()
2331 If multiple keywords appear the '``sideeffect``' keyword must come
2332 first, the '``alignstack``' keyword second and the '``inteldialect``'
2338 The call instructions that wrap inline asm nodes may have a
2339 "``!srcloc``" MDNode attached to it that contains a list of constant
2340 integers. If present, the code generator will use the integer as the
2341 location cookie value when report errors through the ``LLVMContext``
2342 error reporting mechanisms. This allows a front-end to correlate backend
2343 errors that occur with inline asm back to the source code that produced
2346 .. code-block:: llvm
2348 call void asm sideeffect "something bad", ""(), !srcloc !42
2350 !42 = !{ i32 1234567 }
2352 It is up to the front-end to make sense of the magic numbers it places
2353 in the IR. If the MDNode contains multiple constants, the code generator
2354 will use the one that corresponds to the line of the asm that the error
2359 Metadata Nodes and Metadata Strings
2360 -----------------------------------
2362 LLVM IR allows metadata to be attached to instructions in the program
2363 that can convey extra information about the code to the optimizers and
2364 code generator. One example application of metadata is source-level
2365 debug information. There are two metadata primitives: strings and nodes.
2366 All metadata has the ``metadata`` type and is identified in syntax by a
2367 preceding exclamation point ('``!``').
2369 A metadata string is a string surrounded by double quotes. It can
2370 contain any character by escaping non-printable characters with
2371 "``\xx``" where "``xx``" is the two digit hex code. For example:
2374 Metadata nodes are represented with notation similar to structure
2375 constants (a comma separated list of elements, surrounded by braces and
2376 preceded by an exclamation point). Metadata nodes can have any values as
2377 their operand. For example:
2379 .. code-block:: llvm
2381 !{ metadata !"test\00", i32 10}
2383 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2384 metadata nodes, which can be looked up in the module symbol table. For
2387 .. code-block:: llvm
2389 !foo = metadata !{!4, !3}
2391 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2392 function is using two metadata arguments:
2394 .. code-block:: llvm
2396 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2398 Metadata can be attached with an instruction. Here metadata ``!21`` is
2399 attached to the ``add`` instruction using the ``!dbg`` identifier:
2401 .. code-block:: llvm
2403 %indvar.next = add i64 %indvar, 1, !dbg !21
2405 More information about specific metadata nodes recognized by the
2406 optimizers and code generator is found below.
2411 In LLVM IR, memory does not have types, so LLVM's own type system is not
2412 suitable for doing TBAA. Instead, metadata is added to the IR to
2413 describe a type system of a higher level language. This can be used to
2414 implement typical C/C++ TBAA, but it can also be used to implement
2415 custom alias analysis behavior for other languages.
2417 The current metadata format is very simple. TBAA metadata nodes have up
2418 to three fields, e.g.:
2420 .. code-block:: llvm
2422 !0 = metadata !{ metadata !"an example type tree" }
2423 !1 = metadata !{ metadata !"int", metadata !0 }
2424 !2 = metadata !{ metadata !"float", metadata !0 }
2425 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2427 The first field is an identity field. It can be any value, usually a
2428 metadata string, which uniquely identifies the type. The most important
2429 name in the tree is the name of the root node. Two trees with different
2430 root node names are entirely disjoint, even if they have leaves with
2433 The second field identifies the type's parent node in the tree, or is
2434 null or omitted for a root node. A type is considered to alias all of
2435 its descendants and all of its ancestors in the tree. Also, a type is
2436 considered to alias all types in other trees, so that bitcode produced
2437 from multiple front-ends is handled conservatively.
2439 If the third field is present, it's an integer which if equal to 1
2440 indicates that the type is "constant" (meaning
2441 ``pointsToConstantMemory`` should return true; see `other useful
2442 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2444 '``tbaa.struct``' Metadata
2445 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2447 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2448 aggregate assignment operations in C and similar languages, however it
2449 is defined to copy a contiguous region of memory, which is more than
2450 strictly necessary for aggregate types which contain holes due to
2451 padding. Also, it doesn't contain any TBAA information about the fields
2454 ``!tbaa.struct`` metadata can describe which memory subregions in a
2455 memcpy are padding and what the TBAA tags of the struct are.
2457 The current metadata format is very simple. ``!tbaa.struct`` metadata
2458 nodes are a list of operands which are in conceptual groups of three.
2459 For each group of three, the first operand gives the byte offset of a
2460 field in bytes, the second gives its size in bytes, and the third gives
2463 .. code-block:: llvm
2465 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2467 This describes a struct with two fields. The first is at offset 0 bytes
2468 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2469 and has size 4 bytes and has tbaa tag !2.
2471 Note that the fields need not be contiguous. In this example, there is a
2472 4 byte gap between the two fields. This gap represents padding which
2473 does not carry useful data and need not be preserved.
2475 '``fpmath``' Metadata
2476 ^^^^^^^^^^^^^^^^^^^^^
2478 ``fpmath`` metadata may be attached to any instruction of floating point
2479 type. It can be used to express the maximum acceptable error in the
2480 result of that instruction, in ULPs, thus potentially allowing the
2481 compiler to use a more efficient but less accurate method of computing
2482 it. ULP is defined as follows:
2484 If ``x`` is a real number that lies between two finite consecutive
2485 floating-point numbers ``a`` and ``b``, without being equal to one
2486 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2487 distance between the two non-equal finite floating-point numbers
2488 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2490 The metadata node shall consist of a single positive floating point
2491 number representing the maximum relative error, for example:
2493 .. code-block:: llvm
2495 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2497 '``range``' Metadata
2498 ^^^^^^^^^^^^^^^^^^^^
2500 ``range`` metadata may be attached only to loads of integer types. It
2501 expresses the possible ranges the loaded value is in. The ranges are
2502 represented with a flattened list of integers. The loaded value is known
2503 to be in the union of the ranges defined by each consecutive pair. Each
2504 pair has the following properties:
2506 - The type must match the type loaded by the instruction.
2507 - The pair ``a,b`` represents the range ``[a,b)``.
2508 - Both ``a`` and ``b`` are constants.
2509 - The range is allowed to wrap.
2510 - The range should not represent the full or empty set. That is,
2513 In addition, the pairs must be in signed order of the lower bound and
2514 they must be non-contiguous.
2518 .. code-block:: llvm
2520 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2521 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2522 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2523 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2525 !0 = metadata !{ i8 0, i8 2 }
2526 !1 = metadata !{ i8 255, i8 2 }
2527 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2528 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2533 It is sometimes useful to attach information to loop constructs. Currently,
2534 loop metadata is implemented as metadata attached to the branch instruction
2535 in the loop latch block. This type of metadata refer to a metadata node that is
2536 guaranteed to be separate for each loop. The loop-level metadata is prefixed
2539 The loop identifier metadata is implemented using a metadata that refers to
2540 itself to avoid merging it with any other identifier metadata, e.g.,
2541 during module linkage or function inlining. That is, each loop should refer
2542 to their own identification metadata even if they reside in separate functions.
2543 The following example contains loop identifier metadata for two separate loop
2546 .. code-block:: llvm
2548 !0 = metadata !{ metadata !0 }
2549 !1 = metadata !{ metadata !1 }
2552 '``llvm.loop.parallel``' Metadata
2553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2555 This loop metadata can be used to communicate that a loop should be considered
2556 a parallel loop. The semantics of parallel loops in this case is the one
2557 with the strongest cross-iteration instruction ordering freedom: the
2558 iterations in the loop can be considered completely independent of each
2559 other (also known as embarrassingly parallel loops).
2561 This metadata can originate from a programming language with parallel loop
2562 constructs. In such a case it is completely the programmer's responsibility
2563 to ensure the instructions from the different iterations of the loop can be
2564 executed in an arbitrary order, in parallel, or intertwined. No loop-carried
2565 dependency checking at all must be expected from the compiler.
2567 In order to fulfill the LLVM requirement for metadata to be safely ignored,
2568 it is important to ensure that a parallel loop is converted to
2569 a sequential loop in case an optimization (agnostic of the parallel loop
2570 semantics) converts the loop back to such. This happens when new memory
2571 accesses that do not fulfill the requirement of free ordering across iterations
2572 are added to the loop. Therefore, this metadata is required, but not
2573 sufficient, to consider the loop at hand a parallel loop. For a loop
2574 to be parallel, all its memory accessing instructions need to be
2575 marked with the ``llvm.mem.parallel_loop_access`` metadata that refer
2576 to the same loop identifier metadata that identify the loop at hand.
2581 Metadata types used to annotate memory accesses with information helpful
2582 for optimizations are prefixed with ``llvm.mem``.
2584 '``llvm.mem.parallel_loop_access``' Metadata
2585 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2587 For a loop to be parallel, in addition to using
2588 the ``llvm.loop.parallel`` metadata to mark the loop latch branch instruction,
2589 also all of the memory accessing instructions in the loop body need to be
2590 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2591 is at least one memory accessing instruction not marked with the metadata,
2592 the loop, despite it possibly using the ``llvm.loop.parallel`` metadata,
2593 must be considered a sequential loop. This causes parallel loops to be
2594 converted to sequential loops due to optimization passes that are unaware of
2595 the parallel semantics and that insert new memory instructions to the loop
2598 Example of a loop that is considered parallel due to its correct use of
2599 both ``llvm.loop.parallel`` and ``llvm.mem.parallel_loop_access``
2600 metadata types that refer to the same loop identifier metadata.
2602 .. code-block:: llvm
2606 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2608 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2610 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop.parallel !0
2614 !0 = metadata !{ metadata !0 }
2616 It is also possible to have nested parallel loops. In that case the
2617 memory accesses refer to a list of loop identifier metadata nodes instead of
2618 the loop identifier metadata node directly:
2620 .. code-block:: llvm
2627 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2629 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2631 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop.parallel !1
2635 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2637 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2639 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop.parallel !2
2641 outer.for.end: ; preds = %for.body
2643 !0 = metadata !{ metadata !1, metadata !2 } ; a list of parallel loop identifiers
2644 !1 = metadata !{ metadata !1 } ; an identifier for the inner parallel loop
2645 !2 = metadata !{ metadata !2 } ; an identifier for the outer parallel loop
2648 Module Flags Metadata
2649 =====================
2651 Information about the module as a whole is difficult to convey to LLVM's
2652 subsystems. The LLVM IR isn't sufficient to transmit this information.
2653 The ``llvm.module.flags`` named metadata exists in order to facilitate
2654 this. These flags are in the form of key / value pairs --- much like a
2655 dictionary --- making it easy for any subsystem who cares about a flag to
2658 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2659 Each triplet has the following form:
2661 - The first element is a *behavior* flag, which specifies the behavior
2662 when two (or more) modules are merged together, and it encounters two
2663 (or more) metadata with the same ID. The supported behaviors are
2665 - The second element is a metadata string that is a unique ID for the
2666 metadata. Each module may only have one flag entry for each unique ID (not
2667 including entries with the **Require** behavior).
2668 - The third element is the value of the flag.
2670 When two (or more) modules are merged together, the resulting
2671 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2672 each unique metadata ID string, there will be exactly one entry in the merged
2673 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2674 be determined by the merge behavior flag, as described below. The only exception
2675 is that entries with the *Require* behavior are always preserved.
2677 The following behaviors are supported:
2688 Emits an error if two values disagree, otherwise the resulting value
2689 is that of the operands.
2693 Emits a warning if two values disagree. The result value will be the
2694 operand for the flag from the first module being linked.
2698 Adds a requirement that another module flag be present and have a
2699 specified value after linking is performed. The value must be a
2700 metadata pair, where the first element of the pair is the ID of the
2701 module flag to be restricted, and the second element of the pair is
2702 the value the module flag should be restricted to. This behavior can
2703 be used to restrict the allowable results (via triggering of an
2704 error) of linking IDs with the **Override** behavior.
2708 Uses the specified value, regardless of the behavior or value of the
2709 other module. If both modules specify **Override**, but the values
2710 differ, an error will be emitted.
2714 Appends the two values, which are required to be metadata nodes.
2718 Appends the two values, which are required to be metadata
2719 nodes. However, duplicate entries in the second list are dropped
2720 during the append operation.
2722 It is an error for a particular unique flag ID to have multiple behaviors,
2723 except in the case of **Require** (which adds restrictions on another metadata
2724 value) or **Override**.
2726 An example of module flags:
2728 .. code-block:: llvm
2730 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2731 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2732 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2733 !3 = metadata !{ i32 3, metadata !"qux",
2735 metadata !"foo", i32 1
2738 !llvm.module.flags = !{ !0, !1, !2, !3 }
2740 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2741 if two or more ``!"foo"`` flags are seen is to emit an error if their
2742 values are not equal.
2744 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2745 behavior if two or more ``!"bar"`` flags are seen is to use the value
2748 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2749 behavior if two or more ``!"qux"`` flags are seen is to emit a
2750 warning if their values are not equal.
2752 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2756 metadata !{ metadata !"foo", i32 1 }
2758 The behavior is to emit an error if the ``llvm.module.flags`` does not
2759 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2762 Objective-C Garbage Collection Module Flags Metadata
2763 ----------------------------------------------------
2765 On the Mach-O platform, Objective-C stores metadata about garbage
2766 collection in a special section called "image info". The metadata
2767 consists of a version number and a bitmask specifying what types of
2768 garbage collection are supported (if any) by the file. If two or more
2769 modules are linked together their garbage collection metadata needs to
2770 be merged rather than appended together.
2772 The Objective-C garbage collection module flags metadata consists of the
2773 following key-value pairs:
2782 * - ``Objective-C Version``
2783 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2785 * - ``Objective-C Image Info Version``
2786 - **[Required]** --- The version of the image info section. Currently
2789 * - ``Objective-C Image Info Section``
2790 - **[Required]** --- The section to place the metadata. Valid values are
2791 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2792 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2793 Objective-C ABI version 2.
2795 * - ``Objective-C Garbage Collection``
2796 - **[Required]** --- Specifies whether garbage collection is supported or
2797 not. Valid values are 0, for no garbage collection, and 2, for garbage
2798 collection supported.
2800 * - ``Objective-C GC Only``
2801 - **[Optional]** --- Specifies that only garbage collection is supported.
2802 If present, its value must be 6. This flag requires that the
2803 ``Objective-C Garbage Collection`` flag have the value 2.
2805 Some important flag interactions:
2807 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2808 merged with a module with ``Objective-C Garbage Collection`` set to
2809 2, then the resulting module has the
2810 ``Objective-C Garbage Collection`` flag set to 0.
2811 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2812 merged with a module with ``Objective-C GC Only`` set to 6.
2814 Automatic Linker Flags Module Flags Metadata
2815 --------------------------------------------
2817 Some targets support embedding flags to the linker inside individual object
2818 files. Typically this is used in conjunction with language extensions which
2819 allow source files to explicitly declare the libraries they depend on, and have
2820 these automatically be transmitted to the linker via object files.
2822 These flags are encoded in the IR using metadata in the module flags section,
2823 using the ``Linker Options`` key. The merge behavior for this flag is required
2824 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2825 node which should be a list of other metadata nodes, each of which should be a
2826 list of metadata strings defining linker options.
2828 For example, the following metadata section specifies two separate sets of
2829 linker options, presumably to link against ``libz`` and the ``Cocoa``
2832 !0 = metadata !{ i32 6, metadata !"Linker Options",
2834 metadata !{ metadata !"-lz" },
2835 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2836 !llvm.module.flags = !{ !0 }
2838 The metadata encoding as lists of lists of options, as opposed to a collapsed
2839 list of options, is chosen so that the IR encoding can use multiple option
2840 strings to specify e.g., a single library, while still having that specifier be
2841 preserved as an atomic element that can be recognized by a target specific
2842 assembly writer or object file emitter.
2844 Each individual option is required to be either a valid option for the target's
2845 linker, or an option that is reserved by the target specific assembly writer or
2846 object file emitter. No other aspect of these options is defined by the IR.
2848 Intrinsic Global Variables
2849 ==========================
2851 LLVM has a number of "magic" global variables that contain data that
2852 affect code generation or other IR semantics. These are documented here.
2853 All globals of this sort should have a section specified as
2854 "``llvm.metadata``". This section and all globals that start with
2855 "``llvm.``" are reserved for use by LLVM.
2857 The '``llvm.used``' Global Variable
2858 -----------------------------------
2860 The ``@llvm.used`` global is an array with i8\* element type which has
2861 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2862 pointers to global variables and functions which may optionally have a
2863 pointer cast formed of bitcast or getelementptr. For example, a legal
2866 .. code-block:: llvm
2871 @llvm.used = appending global [2 x i8*] [
2873 i8* bitcast (i32* @Y to i8*)
2874 ], section "llvm.metadata"
2876 If a global variable appears in the ``@llvm.used`` list, then the
2877 compiler, assembler, and linker are required to treat the symbol as if
2878 there is a reference to the global that it cannot see. For example, if a
2879 variable has internal linkage and no references other than that from the
2880 ``@llvm.used`` list, it cannot be deleted. This is commonly used to
2881 represent references from inline asms and other things the compiler
2882 cannot "see", and corresponds to "``attribute((used))``" in GNU C.
2884 On some targets, the code generator must emit a directive to the
2885 assembler or object file to prevent the assembler and linker from
2886 molesting the symbol.
2888 The '``llvm.compiler.used``' Global Variable
2889 --------------------------------------------
2891 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2892 directive, except that it only prevents the compiler from touching the
2893 symbol. On targets that support it, this allows an intelligent linker to
2894 optimize references to the symbol without being impeded as it would be
2897 This is a rare construct that should only be used in rare circumstances,
2898 and should not be exposed to source languages.
2900 The '``llvm.global_ctors``' Global Variable
2901 -------------------------------------------
2903 .. code-block:: llvm
2905 %0 = type { i32, void ()* }
2906 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2908 The ``@llvm.global_ctors`` array contains a list of constructor
2909 functions and associated priorities. The functions referenced by this
2910 array will be called in ascending order of priority (i.e. lowest first)
2911 when the module is loaded. The order of functions with the same priority
2914 The '``llvm.global_dtors``' Global Variable
2915 -------------------------------------------
2917 .. code-block:: llvm
2919 %0 = type { i32, void ()* }
2920 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2922 The ``@llvm.global_dtors`` array contains a list of destructor functions
2923 and associated priorities. The functions referenced by this array will
2924 be called in descending order of priority (i.e. highest first) when the
2925 module is loaded. The order of functions with the same priority is not
2928 Instruction Reference
2929 =====================
2931 The LLVM instruction set consists of several different classifications
2932 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
2933 instructions <binaryops>`, :ref:`bitwise binary
2934 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
2935 :ref:`other instructions <otherops>`.
2939 Terminator Instructions
2940 -----------------------
2942 As mentioned :ref:`previously <functionstructure>`, every basic block in a
2943 program ends with a "Terminator" instruction, which indicates which
2944 block should be executed after the current block is finished. These
2945 terminator instructions typically yield a '``void``' value: they produce
2946 control flow, not values (the one exception being the
2947 ':ref:`invoke <i_invoke>`' instruction).
2949 The terminator instructions are: ':ref:`ret <i_ret>`',
2950 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
2951 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
2952 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
2956 '``ret``' Instruction
2957 ^^^^^^^^^^^^^^^^^^^^^
2964 ret <type> <value> ; Return a value from a non-void function
2965 ret void ; Return from void function
2970 The '``ret``' instruction is used to return control flow (and optionally
2971 a value) from a function back to the caller.
2973 There are two forms of the '``ret``' instruction: one that returns a
2974 value and then causes control flow, and one that just causes control
2980 The '``ret``' instruction optionally accepts a single argument, the
2981 return value. The type of the return value must be a ':ref:`first
2982 class <t_firstclass>`' type.
2984 A function is not :ref:`well formed <wellformed>` if it it has a non-void
2985 return type and contains a '``ret``' instruction with no return value or
2986 a return value with a type that does not match its type, or if it has a
2987 void return type and contains a '``ret``' instruction with a return
2993 When the '``ret``' instruction is executed, control flow returns back to
2994 the calling function's context. If the caller is a
2995 ":ref:`call <i_call>`" instruction, execution continues at the
2996 instruction after the call. If the caller was an
2997 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
2998 beginning of the "normal" destination block. If the instruction returns
2999 a value, that value shall set the call or invoke instruction's return
3005 .. code-block:: llvm
3007 ret i32 5 ; Return an integer value of 5
3008 ret void ; Return from a void function
3009 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3013 '``br``' Instruction
3014 ^^^^^^^^^^^^^^^^^^^^
3021 br i1 <cond>, label <iftrue>, label <iffalse>
3022 br label <dest> ; Unconditional branch
3027 The '``br``' instruction is used to cause control flow to transfer to a
3028 different basic block in the current function. There are two forms of
3029 this instruction, corresponding to a conditional branch and an
3030 unconditional branch.
3035 The conditional branch form of the '``br``' instruction takes a single
3036 '``i1``' value and two '``label``' values. The unconditional form of the
3037 '``br``' instruction takes a single '``label``' value as a target.
3042 Upon execution of a conditional '``br``' instruction, the '``i1``'
3043 argument is evaluated. If the value is ``true``, control flows to the
3044 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3045 to the '``iffalse``' ``label`` argument.
3050 .. code-block:: llvm
3053 %cond = icmp eq i32 %a, %b
3054 br i1 %cond, label %IfEqual, label %IfUnequal
3062 '``switch``' Instruction
3063 ^^^^^^^^^^^^^^^^^^^^^^^^
3070 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3075 The '``switch``' instruction is used to transfer control flow to one of
3076 several different places. It is a generalization of the '``br``'
3077 instruction, allowing a branch to occur to one of many possible
3083 The '``switch``' instruction uses three parameters: an integer
3084 comparison value '``value``', a default '``label``' destination, and an
3085 array of pairs of comparison value constants and '``label``'s. The table
3086 is not allowed to contain duplicate constant entries.
3091 The ``switch`` instruction specifies a table of values and destinations.
3092 When the '``switch``' instruction is executed, this table is searched
3093 for the given value. If the value is found, control flow is transferred
3094 to the corresponding destination; otherwise, control flow is transferred
3095 to the default destination.
3100 Depending on properties of the target machine and the particular
3101 ``switch`` instruction, this instruction may be code generated in
3102 different ways. For example, it could be generated as a series of
3103 chained conditional branches or with a lookup table.
3108 .. code-block:: llvm
3110 ; Emulate a conditional br instruction
3111 %Val = zext i1 %value to i32
3112 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3114 ; Emulate an unconditional br instruction
3115 switch i32 0, label %dest [ ]
3117 ; Implement a jump table:
3118 switch i32 %val, label %otherwise [ i32 0, label %onzero
3120 i32 2, label %ontwo ]
3124 '``indirectbr``' Instruction
3125 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3132 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3137 The '``indirectbr``' instruction implements an indirect branch to a
3138 label within the current function, whose address is specified by
3139 "``address``". Address must be derived from a
3140 :ref:`blockaddress <blockaddress>` constant.
3145 The '``address``' argument is the address of the label to jump to. The
3146 rest of the arguments indicate the full set of possible destinations
3147 that the address may point to. Blocks are allowed to occur multiple
3148 times in the destination list, though this isn't particularly useful.
3150 This destination list is required so that dataflow analysis has an
3151 accurate understanding of the CFG.
3156 Control transfers to the block specified in the address argument. All
3157 possible destination blocks must be listed in the label list, otherwise
3158 this instruction has undefined behavior. This implies that jumps to
3159 labels defined in other functions have undefined behavior as well.
3164 This is typically implemented with a jump through a register.
3169 .. code-block:: llvm
3171 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3175 '``invoke``' Instruction
3176 ^^^^^^^^^^^^^^^^^^^^^^^^
3183 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3184 to label <normal label> unwind label <exception label>
3189 The '``invoke``' instruction causes control to transfer to a specified
3190 function, with the possibility of control flow transfer to either the
3191 '``normal``' label or the '``exception``' label. If the callee function
3192 returns with the "``ret``" instruction, control flow will return to the
3193 "normal" label. If the callee (or any indirect callees) returns via the
3194 ":ref:`resume <i_resume>`" instruction or other exception handling
3195 mechanism, control is interrupted and continued at the dynamically
3196 nearest "exception" label.
3198 The '``exception``' label is a `landing
3199 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3200 '``exception``' label is required to have the
3201 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3202 information about the behavior of the program after unwinding happens,
3203 as its first non-PHI instruction. The restrictions on the
3204 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3205 instruction, so that the important information contained within the
3206 "``landingpad``" instruction can't be lost through normal code motion.
3211 This instruction requires several arguments:
3213 #. The optional "cconv" marker indicates which :ref:`calling
3214 convention <callingconv>` the call should use. If none is
3215 specified, the call defaults to using C calling conventions.
3216 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3217 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3219 #. '``ptr to function ty``': shall be the signature of the pointer to
3220 function value being invoked. In most cases, this is a direct
3221 function invocation, but indirect ``invoke``'s are just as possible,
3222 branching off an arbitrary pointer to function value.
3223 #. '``function ptr val``': An LLVM value containing a pointer to a
3224 function to be invoked.
3225 #. '``function args``': argument list whose types match the function
3226 signature argument types and parameter attributes. All arguments must
3227 be of :ref:`first class <t_firstclass>` type. If the function signature
3228 indicates the function accepts a variable number of arguments, the
3229 extra arguments can be specified.
3230 #. '``normal label``': the label reached when the called function
3231 executes a '``ret``' instruction.
3232 #. '``exception label``': the label reached when a callee returns via
3233 the :ref:`resume <i_resume>` instruction or other exception handling
3235 #. The optional :ref:`function attributes <fnattrs>` list. Only
3236 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3237 attributes are valid here.
3242 This instruction is designed to operate as a standard '``call``'
3243 instruction in most regards. The primary difference is that it
3244 establishes an association with a label, which is used by the runtime
3245 library to unwind the stack.
3247 This instruction is used in languages with destructors to ensure that
3248 proper cleanup is performed in the case of either a ``longjmp`` or a
3249 thrown exception. Additionally, this is important for implementation of
3250 '``catch``' clauses in high-level languages that support them.
3252 For the purposes of the SSA form, the definition of the value returned
3253 by the '``invoke``' instruction is deemed to occur on the edge from the
3254 current block to the "normal" label. If the callee unwinds then no
3255 return value is available.
3260 .. code-block:: llvm
3262 %retval = invoke i32 @Test(i32 15) to label %Continue
3263 unwind label %TestCleanup ; {i32}:retval set
3264 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3265 unwind label %TestCleanup ; {i32}:retval set
3269 '``resume``' Instruction
3270 ^^^^^^^^^^^^^^^^^^^^^^^^
3277 resume <type> <value>
3282 The '``resume``' instruction is a terminator instruction that has no
3288 The '``resume``' instruction requires one argument, which must have the
3289 same type as the result of any '``landingpad``' instruction in the same
3295 The '``resume``' instruction resumes propagation of an existing
3296 (in-flight) exception whose unwinding was interrupted with a
3297 :ref:`landingpad <i_landingpad>` instruction.
3302 .. code-block:: llvm
3304 resume { i8*, i32 } %exn
3308 '``unreachable``' Instruction
3309 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3321 The '``unreachable``' instruction has no defined semantics. This
3322 instruction is used to inform the optimizer that a particular portion of
3323 the code is not reachable. This can be used to indicate that the code
3324 after a no-return function cannot be reached, and other facts.
3329 The '``unreachable``' instruction has no defined semantics.
3336 Binary operators are used to do most of the computation in a program.
3337 They require two operands of the same type, execute an operation on
3338 them, and produce a single value. The operands might represent multiple
3339 data, as is the case with the :ref:`vector <t_vector>` data type. The
3340 result value has the same type as its operands.
3342 There are several different binary operators:
3346 '``add``' Instruction
3347 ^^^^^^^^^^^^^^^^^^^^^
3354 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3355 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3356 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3357 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3362 The '``add``' instruction returns the sum of its two operands.
3367 The two arguments to the '``add``' instruction must be
3368 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3369 arguments must have identical types.
3374 The value produced is the integer sum of the two operands.
3376 If the sum has unsigned overflow, the result returned is the
3377 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3380 Because LLVM integers use a two's complement representation, this
3381 instruction is appropriate for both signed and unsigned integers.
3383 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3384 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3385 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3386 unsigned and/or signed overflow, respectively, occurs.
3391 .. code-block:: llvm
3393 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3397 '``fadd``' Instruction
3398 ^^^^^^^^^^^^^^^^^^^^^^
3405 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3410 The '``fadd``' instruction returns the sum of its two operands.
3415 The two arguments to the '``fadd``' instruction must be :ref:`floating
3416 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3417 Both arguments must have identical types.
3422 The value produced is the floating point sum of the two operands. This
3423 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3424 which are optimization hints to enable otherwise unsafe floating point
3430 .. code-block:: llvm
3432 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3434 '``sub``' Instruction
3435 ^^^^^^^^^^^^^^^^^^^^^
3442 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3443 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3444 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3445 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3450 The '``sub``' instruction returns the difference of its two operands.
3452 Note that the '``sub``' instruction is used to represent the '``neg``'
3453 instruction present in most other intermediate representations.
3458 The two arguments to the '``sub``' instruction must be
3459 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3460 arguments must have identical types.
3465 The value produced is the integer difference of the two operands.
3467 If the difference has unsigned overflow, the result returned is the
3468 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3471 Because LLVM integers use a two's complement representation, this
3472 instruction is appropriate for both signed and unsigned integers.
3474 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3475 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3476 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3477 unsigned and/or signed overflow, respectively, occurs.
3482 .. code-block:: llvm
3484 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3485 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3489 '``fsub``' Instruction
3490 ^^^^^^^^^^^^^^^^^^^^^^
3497 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3502 The '``fsub``' instruction returns the difference of its two operands.
3504 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3505 instruction present in most other intermediate representations.
3510 The two arguments to the '``fsub``' instruction must be :ref:`floating
3511 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3512 Both arguments must have identical types.
3517 The value produced is the floating point difference of the two operands.
3518 This instruction can also take any number of :ref:`fast-math
3519 flags <fastmath>`, which are optimization hints to enable otherwise
3520 unsafe floating point optimizations:
3525 .. code-block:: llvm
3527 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3528 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3530 '``mul``' Instruction
3531 ^^^^^^^^^^^^^^^^^^^^^
3538 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3539 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3540 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3541 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3546 The '``mul``' instruction returns the product of its two operands.
3551 The two arguments to the '``mul``' instruction must be
3552 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3553 arguments must have identical types.
3558 The value produced is the integer product of the two operands.
3560 If the result of the multiplication has unsigned overflow, the result
3561 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3562 bit width of the result.
3564 Because LLVM integers use a two's complement representation, and the
3565 result is the same width as the operands, this instruction returns the
3566 correct result for both signed and unsigned integers. If a full product
3567 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3568 sign-extended or zero-extended as appropriate to the width of the full
3571 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3572 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3573 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3574 unsigned and/or signed overflow, respectively, occurs.
3579 .. code-block:: llvm
3581 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3585 '``fmul``' Instruction
3586 ^^^^^^^^^^^^^^^^^^^^^^
3593 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3598 The '``fmul``' instruction returns the product of its two operands.
3603 The two arguments to the '``fmul``' instruction must be :ref:`floating
3604 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3605 Both arguments must have identical types.
3610 The value produced is the floating point product of the two operands.
3611 This instruction can also take any number of :ref:`fast-math
3612 flags <fastmath>`, which are optimization hints to enable otherwise
3613 unsafe floating point optimizations:
3618 .. code-block:: llvm
3620 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3622 '``udiv``' Instruction
3623 ^^^^^^^^^^^^^^^^^^^^^^
3630 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3631 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3636 The '``udiv``' instruction returns the quotient of its two operands.
3641 The two arguments to the '``udiv``' instruction must be
3642 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3643 arguments must have identical types.
3648 The value produced is the unsigned integer quotient of the two operands.
3650 Note that unsigned integer division and signed integer division are
3651 distinct operations; for signed integer division, use '``sdiv``'.
3653 Division by zero leads to undefined behavior.
3655 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3656 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3657 such, "((a udiv exact b) mul b) == a").
3662 .. code-block:: llvm
3664 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3666 '``sdiv``' Instruction
3667 ^^^^^^^^^^^^^^^^^^^^^^
3674 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3675 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3680 The '``sdiv``' instruction returns the quotient of its two operands.
3685 The two arguments to the '``sdiv``' instruction must be
3686 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3687 arguments must have identical types.
3692 The value produced is the signed integer quotient of the two operands
3693 rounded towards zero.
3695 Note that signed integer division and unsigned integer division are
3696 distinct operations; for unsigned integer division, use '``udiv``'.
3698 Division by zero leads to undefined behavior. Overflow also leads to
3699 undefined behavior; this is a rare case, but can occur, for example, by
3700 doing a 32-bit division of -2147483648 by -1.
3702 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3703 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3708 .. code-block:: llvm
3710 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3714 '``fdiv``' Instruction
3715 ^^^^^^^^^^^^^^^^^^^^^^
3722 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3727 The '``fdiv``' instruction returns the quotient of its two operands.
3732 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3733 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3734 Both arguments must have identical types.
3739 The value produced is the floating point quotient of the two operands.
3740 This instruction can also take any number of :ref:`fast-math
3741 flags <fastmath>`, which are optimization hints to enable otherwise
3742 unsafe floating point optimizations:
3747 .. code-block:: llvm
3749 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3751 '``urem``' Instruction
3752 ^^^^^^^^^^^^^^^^^^^^^^
3759 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3764 The '``urem``' instruction returns the remainder from the unsigned
3765 division of its two arguments.
3770 The two arguments to the '``urem``' instruction must be
3771 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3772 arguments must have identical types.
3777 This instruction returns the unsigned integer *remainder* of a division.
3778 This instruction always performs an unsigned division to get the
3781 Note that unsigned integer remainder and signed integer remainder are
3782 distinct operations; for signed integer remainder, use '``srem``'.
3784 Taking the remainder of a division by zero leads to undefined behavior.
3789 .. code-block:: llvm
3791 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3793 '``srem``' Instruction
3794 ^^^^^^^^^^^^^^^^^^^^^^
3801 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3806 The '``srem``' instruction returns the remainder from the signed
3807 division of its two operands. This instruction can also take
3808 :ref:`vector <t_vector>` versions of the values in which case the elements
3814 The two arguments to the '``srem``' instruction must be
3815 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3816 arguments must have identical types.
3821 This instruction returns the *remainder* of a division (where the result
3822 is either zero or has the same sign as the dividend, ``op1``), not the
3823 *modulo* operator (where the result is either zero or has the same sign
3824 as the divisor, ``op2``) of a value. For more information about the
3825 difference, see `The Math
3826 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3827 table of how this is implemented in various languages, please see
3829 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3831 Note that signed integer remainder and unsigned integer remainder are
3832 distinct operations; for unsigned integer remainder, use '``urem``'.
3834 Taking the remainder of a division by zero leads to undefined behavior.
3835 Overflow also leads to undefined behavior; this is a rare case, but can
3836 occur, for example, by taking the remainder of a 32-bit division of
3837 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3838 rule lets srem be implemented using instructions that return both the
3839 result of the division and the remainder.)
3844 .. code-block:: llvm
3846 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3850 '``frem``' Instruction
3851 ^^^^^^^^^^^^^^^^^^^^^^
3858 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3863 The '``frem``' instruction returns the remainder from the division of
3869 The two arguments to the '``frem``' instruction must be :ref:`floating
3870 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3871 Both arguments must have identical types.
3876 This instruction returns the *remainder* of a division. The remainder
3877 has the same sign as the dividend. This instruction can also take any
3878 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3879 to enable otherwise unsafe floating point optimizations:
3884 .. code-block:: llvm
3886 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3890 Bitwise Binary Operations
3891 -------------------------
3893 Bitwise binary operators are used to do various forms of bit-twiddling
3894 in a program. They are generally very efficient instructions and can
3895 commonly be strength reduced from other instructions. They require two
3896 operands of the same type, execute an operation on them, and produce a
3897 single value. The resulting value is the same type as its operands.
3899 '``shl``' Instruction
3900 ^^^^^^^^^^^^^^^^^^^^^
3907 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
3908 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
3909 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
3910 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3915 The '``shl``' instruction returns the first operand shifted to the left
3916 a specified number of bits.
3921 Both arguments to the '``shl``' instruction must be the same
3922 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3923 '``op2``' is treated as an unsigned value.
3928 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
3929 where ``n`` is the width of the result. If ``op2`` is (statically or
3930 dynamically) negative or equal to or larger than the number of bits in
3931 ``op1``, the result is undefined. If the arguments are vectors, each
3932 vector element of ``op1`` is shifted by the corresponding shift amount
3935 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
3936 value <poisonvalues>` if it shifts out any non-zero bits. If the
3937 ``nsw`` keyword is present, then the shift produces a :ref:`poison
3938 value <poisonvalues>` if it shifts out any bits that disagree with the
3939 resultant sign bit. As such, NUW/NSW have the same semantics as they
3940 would if the shift were expressed as a mul instruction with the same
3941 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
3946 .. code-block:: llvm
3948 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
3949 <result> = shl i32 4, 2 ; yields {i32}: 16
3950 <result> = shl i32 1, 10 ; yields {i32}: 1024
3951 <result> = shl i32 1, 32 ; undefined
3952 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
3954 '``lshr``' Instruction
3955 ^^^^^^^^^^^^^^^^^^^^^^
3962 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
3963 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
3968 The '``lshr``' instruction (logical shift right) returns the first
3969 operand shifted to the right a specified number of bits with zero fill.
3974 Both arguments to the '``lshr``' instruction must be the same
3975 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3976 '``op2``' is treated as an unsigned value.
3981 This instruction always performs a logical shift right operation. The
3982 most significant bits of the result will be filled with zero bits after
3983 the shift. If ``op2`` is (statically or dynamically) equal to or larger
3984 than the number of bits in ``op1``, the result is undefined. If the
3985 arguments are vectors, each vector element of ``op1`` is shifted by the
3986 corresponding shift amount in ``op2``.
3988 If the ``exact`` keyword is present, the result value of the ``lshr`` is
3989 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3995 .. code-block:: llvm
3997 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
3998 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
3999 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4000 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7FFFFFFF
4001 <result> = lshr i32 1, 32 ; undefined
4002 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4004 '``ashr``' Instruction
4005 ^^^^^^^^^^^^^^^^^^^^^^
4012 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4013 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4018 The '``ashr``' instruction (arithmetic shift right) returns the first
4019 operand shifted to the right a specified number of bits with sign
4025 Both arguments to the '``ashr``' instruction must be the same
4026 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4027 '``op2``' is treated as an unsigned value.
4032 This instruction always performs an arithmetic shift right operation,
4033 The most significant bits of the result will be filled with the sign bit
4034 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4035 than the number of bits in ``op1``, the result is undefined. If the
4036 arguments are vectors, each vector element of ``op1`` is shifted by the
4037 corresponding shift amount in ``op2``.
4039 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4040 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4046 .. code-block:: llvm
4048 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4049 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4050 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4051 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4052 <result> = ashr i32 1, 32 ; undefined
4053 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4055 '``and``' Instruction
4056 ^^^^^^^^^^^^^^^^^^^^^
4063 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4068 The '``and``' instruction returns the bitwise logical and of its two
4074 The two arguments to the '``and``' instruction must be
4075 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4076 arguments must have identical types.
4081 The truth table used for the '``and``' instruction is:
4098 .. code-block:: llvm
4100 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4101 <result> = and i32 15, 40 ; yields {i32}:result = 8
4102 <result> = and i32 4, 8 ; yields {i32}:result = 0
4104 '``or``' Instruction
4105 ^^^^^^^^^^^^^^^^^^^^
4112 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4117 The '``or``' instruction returns the bitwise logical inclusive or of its
4123 The two arguments to the '``or``' instruction must be
4124 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4125 arguments must have identical types.
4130 The truth table used for the '``or``' instruction is:
4149 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4150 <result> = or i32 15, 40 ; yields {i32}:result = 47
4151 <result> = or i32 4, 8 ; yields {i32}:result = 12
4153 '``xor``' Instruction
4154 ^^^^^^^^^^^^^^^^^^^^^
4161 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4166 The '``xor``' instruction returns the bitwise logical exclusive or of
4167 its two operands. The ``xor`` is used to implement the "one's
4168 complement" operation, which is the "~" operator in C.
4173 The two arguments to the '``xor``' instruction must be
4174 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4175 arguments must have identical types.
4180 The truth table used for the '``xor``' instruction is:
4197 .. code-block:: llvm
4199 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4200 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4201 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4202 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4207 LLVM supports several instructions to represent vector operations in a
4208 target-independent manner. These instructions cover the element-access
4209 and vector-specific operations needed to process vectors effectively.
4210 While LLVM does directly support these vector operations, many
4211 sophisticated algorithms will want to use target-specific intrinsics to
4212 take full advantage of a specific target.
4214 .. _i_extractelement:
4216 '``extractelement``' Instruction
4217 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4224 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4229 The '``extractelement``' instruction extracts a single scalar element
4230 from a vector at a specified index.
4235 The first operand of an '``extractelement``' instruction is a value of
4236 :ref:`vector <t_vector>` type. The second operand is an index indicating
4237 the position from which to extract the element. The index may be a
4243 The result is a scalar of the same type as the element type of ``val``.
4244 Its value is the value at position ``idx`` of ``val``. If ``idx``
4245 exceeds the length of ``val``, the results are undefined.
4250 .. code-block:: llvm
4252 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4254 .. _i_insertelement:
4256 '``insertelement``' Instruction
4257 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4264 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4269 The '``insertelement``' instruction inserts a scalar element into a
4270 vector at a specified index.
4275 The first operand of an '``insertelement``' instruction is a value of
4276 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4277 type must equal the element type of the first operand. The third operand
4278 is an index indicating the position at which to insert the value. The
4279 index may be a variable.
4284 The result is a vector of the same type as ``val``. Its element values
4285 are those of ``val`` except at position ``idx``, where it gets the value
4286 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4292 .. code-block:: llvm
4294 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4296 .. _i_shufflevector:
4298 '``shufflevector``' Instruction
4299 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4306 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4311 The '``shufflevector``' instruction constructs a permutation of elements
4312 from two input vectors, returning a vector with the same element type as
4313 the input and length that is the same as the shuffle mask.
4318 The first two operands of a '``shufflevector``' instruction are vectors
4319 with the same type. The third argument is a shuffle mask whose element
4320 type is always 'i32'. The result of the instruction is a vector whose
4321 length is the same as the shuffle mask and whose element type is the
4322 same as the element type of the first two operands.
4324 The shuffle mask operand is required to be a constant vector with either
4325 constant integer or undef values.
4330 The elements of the two input vectors are numbered from left to right
4331 across both of the vectors. The shuffle mask operand specifies, for each
4332 element of the result vector, which element of the two input vectors the
4333 result element gets. The element selector may be undef (meaning "don't
4334 care") and the second operand may be undef if performing a shuffle from
4340 .. code-block:: llvm
4342 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4343 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4344 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4345 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4346 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4347 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4348 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4349 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4351 Aggregate Operations
4352 --------------------
4354 LLVM supports several instructions for working with
4355 :ref:`aggregate <t_aggregate>` values.
4359 '``extractvalue``' Instruction
4360 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4367 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4372 The '``extractvalue``' instruction extracts the value of a member field
4373 from an :ref:`aggregate <t_aggregate>` value.
4378 The first operand of an '``extractvalue``' instruction is a value of
4379 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4380 constant indices to specify which value to extract in a similar manner
4381 as indices in a '``getelementptr``' instruction.
4383 The major differences to ``getelementptr`` indexing are:
4385 - Since the value being indexed is not a pointer, the first index is
4386 omitted and assumed to be zero.
4387 - At least one index must be specified.
4388 - Not only struct indices but also array indices must be in bounds.
4393 The result is the value at the position in the aggregate specified by
4399 .. code-block:: llvm
4401 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4405 '``insertvalue``' Instruction
4406 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4413 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4418 The '``insertvalue``' instruction inserts a value into a member field in
4419 an :ref:`aggregate <t_aggregate>` value.
4424 The first operand of an '``insertvalue``' instruction is a value of
4425 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4426 a first-class value to insert. The following operands are constant
4427 indices indicating the position at which to insert the value in a
4428 similar manner as indices in a '``extractvalue``' instruction. The value
4429 to insert must have the same type as the value identified by the
4435 The result is an aggregate of the same type as ``val``. Its value is
4436 that of ``val`` except that the value at the position specified by the
4437 indices is that of ``elt``.
4442 .. code-block:: llvm
4444 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4445 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4446 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4450 Memory Access and Addressing Operations
4451 ---------------------------------------
4453 A key design point of an SSA-based representation is how it represents
4454 memory. In LLVM, no memory locations are in SSA form, which makes things
4455 very simple. This section describes how to read, write, and allocate
4460 '``alloca``' Instruction
4461 ^^^^^^^^^^^^^^^^^^^^^^^^
4468 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4473 The '``alloca``' instruction allocates memory on the stack frame of the
4474 currently executing function, to be automatically released when this
4475 function returns to its caller. The object is always allocated in the
4476 generic address space (address space zero).
4481 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4482 bytes of memory on the runtime stack, returning a pointer of the
4483 appropriate type to the program. If "NumElements" is specified, it is
4484 the number of elements allocated, otherwise "NumElements" is defaulted
4485 to be one. If a constant alignment is specified, the value result of the
4486 allocation is guaranteed to be aligned to at least that boundary. If not
4487 specified, or if zero, the target can choose to align the allocation on
4488 any convenient boundary compatible with the type.
4490 '``type``' may be any sized type.
4495 Memory is allocated; a pointer is returned. The operation is undefined
4496 if there is insufficient stack space for the allocation. '``alloca``'d
4497 memory is automatically released when the function returns. The
4498 '``alloca``' instruction is commonly used to represent automatic
4499 variables that must have an address available. When the function returns
4500 (either with the ``ret`` or ``resume`` instructions), the memory is
4501 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4502 The order in which memory is allocated (ie., which way the stack grows)
4508 .. code-block:: llvm
4510 %ptr = alloca i32 ; yields {i32*}:ptr
4511 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4512 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4513 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4517 '``load``' Instruction
4518 ^^^^^^^^^^^^^^^^^^^^^^
4525 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4526 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4527 !<index> = !{ i32 1 }
4532 The '``load``' instruction is used to read from memory.
4537 The argument to the '``load``' instruction specifies the memory address
4538 from which to load. The pointer must point to a :ref:`first
4539 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4540 then the optimizer is not allowed to modify the number or order of
4541 execution of this ``load`` with other :ref:`volatile
4542 operations <volatile>`.
4544 If the ``load`` is marked as ``atomic``, it takes an extra
4545 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4546 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4547 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4548 when they may see multiple atomic stores. The type of the pointee must
4549 be an integer type whose bit width is a power of two greater than or
4550 equal to eight and less than or equal to a target-specific size limit.
4551 ``align`` must be explicitly specified on atomic loads, and the load has
4552 undefined behavior if the alignment is not set to a value which is at
4553 least the size in bytes of the pointee. ``!nontemporal`` does not have
4554 any defined semantics for atomic loads.
4556 The optional constant ``align`` argument specifies the alignment of the
4557 operation (that is, the alignment of the memory address). A value of 0
4558 or an omitted ``align`` argument means that the operation has the abi
4559 alignment for the target. It is the responsibility of the code emitter
4560 to ensure that the alignment information is correct. Overestimating the
4561 alignment results in undefined behavior. Underestimating the alignment
4562 may produce less efficient code. An alignment of 1 is always safe.
4564 The optional ``!nontemporal`` metadata must reference a single
4565 metatadata name <index> corresponding to a metadata node with one
4566 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4567 metatadata on the instruction tells the optimizer and code generator
4568 that this load is not expected to be reused in the cache. The code
4569 generator may select special instructions to save cache bandwidth, such
4570 as the ``MOVNT`` instruction on x86.
4572 The optional ``!invariant.load`` metadata must reference a single
4573 metatadata name <index> corresponding to a metadata node with no
4574 entries. The existence of the ``!invariant.load`` metatadata on the
4575 instruction tells the optimizer and code generator that this load
4576 address points to memory which does not change value during program
4577 execution. The optimizer may then move this load around, for example, by
4578 hoisting it out of loops using loop invariant code motion.
4583 The location of memory pointed to is loaded. If the value being loaded
4584 is of scalar type then the number of bytes read does not exceed the
4585 minimum number of bytes needed to hold all bits of the type. For
4586 example, loading an ``i24`` reads at most three bytes. When loading a
4587 value of a type like ``i20`` with a size that is not an integral number
4588 of bytes, the result is undefined if the value was not originally
4589 written using a store of the same type.
4594 .. code-block:: llvm
4596 %ptr = alloca i32 ; yields {i32*}:ptr
4597 store i32 3, i32* %ptr ; yields {void}
4598 %val = load i32* %ptr ; yields {i32}:val = i32 3
4602 '``store``' Instruction
4603 ^^^^^^^^^^^^^^^^^^^^^^^
4610 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4611 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4616 The '``store``' instruction is used to write to memory.
4621 There are two arguments to the '``store``' instruction: a value to store
4622 and an address at which to store it. The type of the '``<pointer>``'
4623 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4624 the '``<value>``' operand. If the ``store`` is marked as ``volatile``,
4625 then the optimizer is not allowed to modify the number or order of
4626 execution of this ``store`` with other :ref:`volatile
4627 operations <volatile>`.
4629 If the ``store`` is marked as ``atomic``, it takes an extra
4630 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4631 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4632 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4633 when they may see multiple atomic stores. The type of the pointee must
4634 be an integer type whose bit width is a power of two greater than or
4635 equal to eight and less than or equal to a target-specific size limit.
4636 ``align`` must be explicitly specified on atomic stores, and the store
4637 has undefined behavior if the alignment is not set to a value which is
4638 at least the size in bytes of the pointee. ``!nontemporal`` does not
4639 have any defined semantics for atomic stores.
4641 The optional constant "align" argument specifies the alignment of the
4642 operation (that is, the alignment of the memory address). A value of 0
4643 or an omitted "align" argument means that the operation has the abi
4644 alignment for the target. It is the responsibility of the code emitter
4645 to ensure that the alignment information is correct. Overestimating the
4646 alignment results in an undefined behavior. Underestimating the
4647 alignment may produce less efficient code. An alignment of 1 is always
4650 The optional !nontemporal metadata must reference a single metatadata
4651 name <index> corresponding to a metadata node with one i32 entry of
4652 value 1. The existence of the !nontemporal metatadata on the instruction
4653 tells the optimizer and code generator that this load is not expected to
4654 be reused in the cache. The code generator may select special
4655 instructions to save cache bandwidth, such as the MOVNT instruction on
4661 The contents of memory are updated to contain '``<value>``' at the
4662 location specified by the '``<pointer>``' operand. If '``<value>``' is
4663 of scalar type then the number of bytes written does not exceed the
4664 minimum number of bytes needed to hold all bits of the type. For
4665 example, storing an ``i24`` writes at most three bytes. When writing a
4666 value of a type like ``i20`` with a size that is not an integral number
4667 of bytes, it is unspecified what happens to the extra bits that do not
4668 belong to the type, but they will typically be overwritten.
4673 .. code-block:: llvm
4675 %ptr = alloca i32 ; yields {i32*}:ptr
4676 store i32 3, i32* %ptr ; yields {void}
4677 %val = load i32* %ptr ; yields {i32}:val = i32 3
4681 '``fence``' Instruction
4682 ^^^^^^^^^^^^^^^^^^^^^^^
4689 fence [singlethread] <ordering> ; yields {void}
4694 The '``fence``' instruction is used to introduce happens-before edges
4700 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4701 defines what *synchronizes-with* edges they add. They can only be given
4702 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4707 A fence A which has (at least) ``release`` ordering semantics
4708 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4709 semantics if and only if there exist atomic operations X and Y, both
4710 operating on some atomic object M, such that A is sequenced before X, X
4711 modifies M (either directly or through some side effect of a sequence
4712 headed by X), Y is sequenced before B, and Y observes M. This provides a
4713 *happens-before* dependency between A and B. Rather than an explicit
4714 ``fence``, one (but not both) of the atomic operations X or Y might
4715 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4716 still *synchronize-with* the explicit ``fence`` and establish the
4717 *happens-before* edge.
4719 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4720 ``acquire`` and ``release`` semantics specified above, participates in
4721 the global program order of other ``seq_cst`` operations and/or fences.
4723 The optional ":ref:`singlethread <singlethread>`" argument specifies
4724 that the fence only synchronizes with other fences in the same thread.
4725 (This is useful for interacting with signal handlers.)
4730 .. code-block:: llvm
4732 fence acquire ; yields {void}
4733 fence singlethread seq_cst ; yields {void}
4737 '``cmpxchg``' Instruction
4738 ^^^^^^^^^^^^^^^^^^^^^^^^^
4745 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4750 The '``cmpxchg``' instruction is used to atomically modify memory. It
4751 loads a value in memory and compares it to a given value. If they are
4752 equal, it stores a new value into the memory.
4757 There are three arguments to the '``cmpxchg``' instruction: an address
4758 to operate on, a value to compare to the value currently be at that
4759 address, and a new value to place at that address if the compared values
4760 are equal. The type of '<cmp>' must be an integer type whose bit width
4761 is a power of two greater than or equal to eight and less than or equal
4762 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4763 type, and the type of '<pointer>' must be a pointer to that type. If the
4764 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4765 to modify the number or order of execution of this ``cmpxchg`` with
4766 other :ref:`volatile operations <volatile>`.
4768 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4769 synchronizes with other atomic operations.
4771 The optional "``singlethread``" argument declares that the ``cmpxchg``
4772 is only atomic with respect to code (usually signal handlers) running in
4773 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4774 respect to all other code in the system.
4776 The pointer passed into cmpxchg must have alignment greater than or
4777 equal to the size in memory of the operand.
4782 The contents of memory at the location specified by the '``<pointer>``'
4783 operand is read and compared to '``<cmp>``'; if the read value is the
4784 equal, '``<new>``' is written. The original value at the location is
4787 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4788 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4789 atomic load with an ordering parameter determined by dropping any
4790 ``release`` part of the ``cmpxchg``'s ordering.
4795 .. code-block:: llvm
4798 %orig = atomic load i32* %ptr unordered ; yields {i32}
4802 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4803 %squared = mul i32 %cmp, %cmp
4804 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4805 %success = icmp eq i32 %cmp, %old
4806 br i1 %success, label %done, label %loop
4813 '``atomicrmw``' Instruction
4814 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4821 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4826 The '``atomicrmw``' instruction is used to atomically modify memory.
4831 There are three arguments to the '``atomicrmw``' instruction: an
4832 operation to apply, an address whose value to modify, an argument to the
4833 operation. The operation must be one of the following keywords:
4847 The type of '<value>' must be an integer type whose bit width is a power
4848 of two greater than or equal to eight and less than or equal to a
4849 target-specific size limit. The type of the '``<pointer>``' operand must
4850 be a pointer to that type. If the ``atomicrmw`` is marked as
4851 ``volatile``, then the optimizer is not allowed to modify the number or
4852 order of execution of this ``atomicrmw`` with other :ref:`volatile
4853 operations <volatile>`.
4858 The contents of memory at the location specified by the '``<pointer>``'
4859 operand are atomically read, modified, and written back. The original
4860 value at the location is returned. The modification is specified by the
4863 - xchg: ``*ptr = val``
4864 - add: ``*ptr = *ptr + val``
4865 - sub: ``*ptr = *ptr - val``
4866 - and: ``*ptr = *ptr & val``
4867 - nand: ``*ptr = ~(*ptr & val)``
4868 - or: ``*ptr = *ptr | val``
4869 - xor: ``*ptr = *ptr ^ val``
4870 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4871 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4872 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4874 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4880 .. code-block:: llvm
4882 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4884 .. _i_getelementptr:
4886 '``getelementptr``' Instruction
4887 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4894 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4895 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4896 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
4901 The '``getelementptr``' instruction is used to get the address of a
4902 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
4903 address calculation only and does not access memory.
4908 The first argument is always a pointer or a vector of pointers, and
4909 forms the basis of the calculation. The remaining arguments are indices
4910 that indicate which of the elements of the aggregate object are indexed.
4911 The interpretation of each index is dependent on the type being indexed
4912 into. The first index always indexes the pointer value given as the
4913 first argument, the second index indexes a value of the type pointed to
4914 (not necessarily the value directly pointed to, since the first index
4915 can be non-zero), etc. The first type indexed into must be a pointer
4916 value, subsequent types can be arrays, vectors, and structs. Note that
4917 subsequent types being indexed into can never be pointers, since that
4918 would require loading the pointer before continuing calculation.
4920 The type of each index argument depends on the type it is indexing into.
4921 When indexing into a (optionally packed) structure, only ``i32`` integer
4922 **constants** are allowed (when using a vector of indices they must all
4923 be the **same** ``i32`` integer constant). When indexing into an array,
4924 pointer or vector, integers of any width are allowed, and they are not
4925 required to be constant. These integers are treated as signed values
4928 For example, let's consider a C code fragment and how it gets compiled
4944 int *foo(struct ST *s) {
4945 return &s[1].Z.B[5][13];
4948 The LLVM code generated by Clang is:
4950 .. code-block:: llvm
4952 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
4953 %struct.ST = type { i32, double, %struct.RT }
4955 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
4957 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
4964 In the example above, the first index is indexing into the
4965 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
4966 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
4967 indexes into the third element of the structure, yielding a
4968 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
4969 structure. The third index indexes into the second element of the
4970 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
4971 dimensions of the array are subscripted into, yielding an '``i32``'
4972 type. The '``getelementptr``' instruction returns a pointer to this
4973 element, thus computing a value of '``i32*``' type.
4975 Note that it is perfectly legal to index partially through a structure,
4976 returning a pointer to an inner element. Because of this, the LLVM code
4977 for the given testcase is equivalent to:
4979 .. code-block:: llvm
4981 define i32* @foo(%struct.ST* %s) {
4982 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
4983 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
4984 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
4985 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
4986 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
4990 If the ``inbounds`` keyword is present, the result value of the
4991 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
4992 pointer is not an *in bounds* address of an allocated object, or if any
4993 of the addresses that would be formed by successive addition of the
4994 offsets implied by the indices to the base address with infinitely
4995 precise signed arithmetic are not an *in bounds* address of that
4996 allocated object. The *in bounds* addresses for an allocated object are
4997 all the addresses that point into the object, plus the address one byte
4998 past the end. In cases where the base is a vector of pointers the
4999 ``inbounds`` keyword applies to each of the computations element-wise.
5001 If the ``inbounds`` keyword is not present, the offsets are added to the
5002 base address with silently-wrapping two's complement arithmetic. If the
5003 offsets have a different width from the pointer, they are sign-extended
5004 or truncated to the width of the pointer. The result value of the
5005 ``getelementptr`` may be outside the object pointed to by the base
5006 pointer. The result value may not necessarily be used to access memory
5007 though, even if it happens to point into allocated storage. See the
5008 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5011 The getelementptr instruction is often confusing. For some more insight
5012 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5017 .. code-block:: llvm
5019 ; yields [12 x i8]*:aptr
5020 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5022 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5024 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5026 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5028 In cases where the pointer argument is a vector of pointers, each index
5029 must be a vector with the same number of elements. For example:
5031 .. code-block:: llvm
5033 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5035 Conversion Operations
5036 ---------------------
5038 The instructions in this category are the conversion instructions
5039 (casting) which all take a single operand and a type. They perform
5040 various bit conversions on the operand.
5042 '``trunc .. to``' Instruction
5043 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5050 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5055 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5060 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5061 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5062 of the same number of integers. The bit size of the ``value`` must be
5063 larger than the bit size of the destination type, ``ty2``. Equal sized
5064 types are not allowed.
5069 The '``trunc``' instruction truncates the high order bits in ``value``
5070 and converts the remaining bits to ``ty2``. Since the source size must
5071 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5072 It will always truncate bits.
5077 .. code-block:: llvm
5079 %X = trunc i32 257 to i8 ; yields i8:1
5080 %Y = trunc i32 123 to i1 ; yields i1:true
5081 %Z = trunc i32 122 to i1 ; yields i1:false
5082 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5084 '``zext .. to``' Instruction
5085 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5092 <result> = zext <ty> <value> to <ty2> ; yields ty2
5097 The '``zext``' instruction zero extends its operand to type ``ty2``.
5102 The '``zext``' instruction takes a value to cast, and a type to cast it
5103 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5104 the same number of integers. The bit size of the ``value`` must be
5105 smaller than the bit size of the destination type, ``ty2``.
5110 The ``zext`` fills the high order bits of the ``value`` with zero bits
5111 until it reaches the size of the destination type, ``ty2``.
5113 When zero extending from i1, the result will always be either 0 or 1.
5118 .. code-block:: llvm
5120 %X = zext i32 257 to i64 ; yields i64:257
5121 %Y = zext i1 true to i32 ; yields i32:1
5122 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5124 '``sext .. to``' Instruction
5125 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5132 <result> = sext <ty> <value> to <ty2> ; yields ty2
5137 The '``sext``' sign extends ``value`` to the type ``ty2``.
5142 The '``sext``' instruction takes a value to cast, and a type to cast it
5143 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5144 the same number of integers. The bit size of the ``value`` must be
5145 smaller than the bit size of the destination type, ``ty2``.
5150 The '``sext``' instruction performs a sign extension by copying the sign
5151 bit (highest order bit) of the ``value`` until it reaches the bit size
5152 of the type ``ty2``.
5154 When sign extending from i1, the extension always results in -1 or 0.
5159 .. code-block:: llvm
5161 %X = sext i8 -1 to i16 ; yields i16 :65535
5162 %Y = sext i1 true to i32 ; yields i32:-1
5163 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5165 '``fptrunc .. to``' Instruction
5166 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5173 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5178 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5183 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5184 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5185 The size of ``value`` must be larger than the size of ``ty2``. This
5186 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5191 The '``fptrunc``' instruction truncates a ``value`` from a larger
5192 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5193 point <t_floating>` type. If the value cannot fit within the
5194 destination type, ``ty2``, then the results are undefined.
5199 .. code-block:: llvm
5201 %X = fptrunc double 123.0 to float ; yields float:123.0
5202 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5204 '``fpext .. to``' Instruction
5205 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5212 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5217 The '``fpext``' extends a floating point ``value`` to a larger floating
5223 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5224 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5225 to. The source type must be smaller than the destination type.
5230 The '``fpext``' instruction extends the ``value`` from a smaller
5231 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5232 point <t_floating>` type. The ``fpext`` cannot be used to make a
5233 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5234 *no-op cast* for a floating point cast.
5239 .. code-block:: llvm
5241 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5242 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5244 '``fptoui .. to``' Instruction
5245 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5252 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5257 The '``fptoui``' converts a floating point ``value`` to its unsigned
5258 integer equivalent of type ``ty2``.
5263 The '``fptoui``' instruction takes a value to cast, which must be a
5264 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5265 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5266 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5267 type with the same number of elements as ``ty``
5272 The '``fptoui``' instruction converts its :ref:`floating
5273 point <t_floating>` operand into the nearest (rounding towards zero)
5274 unsigned integer value. If the value cannot fit in ``ty2``, the results
5280 .. code-block:: llvm
5282 %X = fptoui double 123.0 to i32 ; yields i32:123
5283 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5284 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5286 '``fptosi .. to``' Instruction
5287 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5294 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5299 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5300 ``value`` to type ``ty2``.
5305 The '``fptosi``' instruction takes a value to cast, which must be a
5306 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5307 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5308 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5309 type with the same number of elements as ``ty``
5314 The '``fptosi``' instruction converts its :ref:`floating
5315 point <t_floating>` operand into the nearest (rounding towards zero)
5316 signed integer value. If the value cannot fit in ``ty2``, the results
5322 .. code-block:: llvm
5324 %X = fptosi double -123.0 to i32 ; yields i32:-123
5325 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5326 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5328 '``uitofp .. to``' Instruction
5329 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5336 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5341 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5342 and converts that value to the ``ty2`` type.
5347 The '``uitofp``' instruction takes a value to cast, which must be a
5348 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5349 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5350 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5351 type with the same number of elements as ``ty``
5356 The '``uitofp``' instruction interprets its operand as an unsigned
5357 integer quantity and converts it to the corresponding floating point
5358 value. If the value cannot fit in the floating point value, the results
5364 .. code-block:: llvm
5366 %X = uitofp i32 257 to float ; yields float:257.0
5367 %Y = uitofp i8 -1 to double ; yields double:255.0
5369 '``sitofp .. to``' Instruction
5370 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5377 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5382 The '``sitofp``' instruction regards ``value`` as a signed integer and
5383 converts that value to the ``ty2`` type.
5388 The '``sitofp``' instruction takes a value to cast, which must be a
5389 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5390 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5391 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5392 type with the same number of elements as ``ty``
5397 The '``sitofp``' instruction interprets its operand as a signed integer
5398 quantity and converts it to the corresponding floating point value. If
5399 the value cannot fit in the floating point value, the results are
5405 .. code-block:: llvm
5407 %X = sitofp i32 257 to float ; yields float:257.0
5408 %Y = sitofp i8 -1 to double ; yields double:-1.0
5412 '``ptrtoint .. to``' Instruction
5413 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5420 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5425 The '``ptrtoint``' instruction converts the pointer or a vector of
5426 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5431 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5432 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5433 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5434 a vector of integers type.
5439 The '``ptrtoint``' instruction converts ``value`` to integer type
5440 ``ty2`` by interpreting the pointer value as an integer and either
5441 truncating or zero extending that value to the size of the integer type.
5442 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5443 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5444 the same size, then nothing is done (*no-op cast*) other than a type
5450 .. code-block:: llvm
5452 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5453 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5454 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5458 '``inttoptr .. to``' Instruction
5459 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5466 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5471 The '``inttoptr``' instruction converts an integer ``value`` to a
5472 pointer type, ``ty2``.
5477 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5478 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5484 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5485 applying either a zero extension or a truncation depending on the size
5486 of the integer ``value``. If ``value`` is larger than the size of a
5487 pointer then a truncation is done. If ``value`` is smaller than the size
5488 of a pointer then a zero extension is done. If they are the same size,
5489 nothing is done (*no-op cast*).
5494 .. code-block:: llvm
5496 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5497 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5498 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5499 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5503 '``bitcast .. to``' Instruction
5504 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5511 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5516 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5522 The '``bitcast``' instruction takes a value to cast, which must be a
5523 non-aggregate first class value, and a type to cast it to, which must
5524 also be a non-aggregate :ref:`first class <t_firstclass>` type. The bit
5525 sizes of ``value`` and the destination type, ``ty2``, must be identical.
5526 If the source type is a pointer, the destination type must also be a
5527 pointer. This instruction supports bitwise conversion of vectors to
5528 integers and to vectors of other types (as long as they have the same
5534 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It is
5535 always a *no-op cast* because no bits change with this conversion. The
5536 conversion is done as if the ``value`` had been stored to memory and
5537 read back as type ``ty2``. Pointer (or vector of pointers) types may
5538 only be converted to other pointer (or vector of pointers) types with
5539 this instruction. To convert pointers to other types, use the
5540 :ref:`inttoptr <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions
5546 .. code-block:: llvm
5548 %X = bitcast i8 255 to i8 ; yields i8 :-1
5549 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5550 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5551 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5558 The instructions in this category are the "miscellaneous" instructions,
5559 which defy better classification.
5563 '``icmp``' Instruction
5564 ^^^^^^^^^^^^^^^^^^^^^^
5571 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5576 The '``icmp``' instruction returns a boolean value or a vector of
5577 boolean values based on comparison of its two integer, integer vector,
5578 pointer, or pointer vector operands.
5583 The '``icmp``' instruction takes three operands. The first operand is
5584 the condition code indicating the kind of comparison to perform. It is
5585 not a value, just a keyword. The possible condition code are:
5588 #. ``ne``: not equal
5589 #. ``ugt``: unsigned greater than
5590 #. ``uge``: unsigned greater or equal
5591 #. ``ult``: unsigned less than
5592 #. ``ule``: unsigned less or equal
5593 #. ``sgt``: signed greater than
5594 #. ``sge``: signed greater or equal
5595 #. ``slt``: signed less than
5596 #. ``sle``: signed less or equal
5598 The remaining two arguments must be :ref:`integer <t_integer>` or
5599 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5600 must also be identical types.
5605 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5606 code given as ``cond``. The comparison performed always yields either an
5607 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5609 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5610 otherwise. No sign interpretation is necessary or performed.
5611 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5612 otherwise. No sign interpretation is necessary or performed.
5613 #. ``ugt``: interprets the operands as unsigned values and yields
5614 ``true`` if ``op1`` is greater than ``op2``.
5615 #. ``uge``: interprets the operands as unsigned values and yields
5616 ``true`` if ``op1`` is greater than or equal to ``op2``.
5617 #. ``ult``: interprets the operands as unsigned values and yields
5618 ``true`` if ``op1`` is less than ``op2``.
5619 #. ``ule``: interprets the operands as unsigned values and yields
5620 ``true`` if ``op1`` is less than or equal to ``op2``.
5621 #. ``sgt``: interprets the operands as signed values and yields ``true``
5622 if ``op1`` is greater than ``op2``.
5623 #. ``sge``: interprets the operands as signed values and yields ``true``
5624 if ``op1`` is greater than or equal to ``op2``.
5625 #. ``slt``: interprets the operands as signed values and yields ``true``
5626 if ``op1`` is less than ``op2``.
5627 #. ``sle``: interprets the operands as signed values and yields ``true``
5628 if ``op1`` is less than or equal to ``op2``.
5630 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5631 are compared as if they were integers.
5633 If the operands are integer vectors, then they are compared element by
5634 element. The result is an ``i1`` vector with the same number of elements
5635 as the values being compared. Otherwise, the result is an ``i1``.
5640 .. code-block:: llvm
5642 <result> = icmp eq i32 4, 5 ; yields: result=false
5643 <result> = icmp ne float* %X, %X ; yields: result=false
5644 <result> = icmp ult i16 4, 5 ; yields: result=true
5645 <result> = icmp sgt i16 4, 5 ; yields: result=false
5646 <result> = icmp ule i16 -4, 5 ; yields: result=false
5647 <result> = icmp sge i16 4, 5 ; yields: result=false
5649 Note that the code generator does not yet support vector types with the
5650 ``icmp`` instruction.
5654 '``fcmp``' Instruction
5655 ^^^^^^^^^^^^^^^^^^^^^^
5662 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5667 The '``fcmp``' instruction returns a boolean value or vector of boolean
5668 values based on comparison of its operands.
5670 If the operands are floating point scalars, then the result type is a
5671 boolean (:ref:`i1 <t_integer>`).
5673 If the operands are floating point vectors, then the result type is a
5674 vector of boolean with the same number of elements as the operands being
5680 The '``fcmp``' instruction takes three operands. The first operand is
5681 the condition code indicating the kind of comparison to perform. It is
5682 not a value, just a keyword. The possible condition code are:
5684 #. ``false``: no comparison, always returns false
5685 #. ``oeq``: ordered and equal
5686 #. ``ogt``: ordered and greater than
5687 #. ``oge``: ordered and greater than or equal
5688 #. ``olt``: ordered and less than
5689 #. ``ole``: ordered and less than or equal
5690 #. ``one``: ordered and not equal
5691 #. ``ord``: ordered (no nans)
5692 #. ``ueq``: unordered or equal
5693 #. ``ugt``: unordered or greater than
5694 #. ``uge``: unordered or greater than or equal
5695 #. ``ult``: unordered or less than
5696 #. ``ule``: unordered or less than or equal
5697 #. ``une``: unordered or not equal
5698 #. ``uno``: unordered (either nans)
5699 #. ``true``: no comparison, always returns true
5701 *Ordered* means that neither operand is a QNAN while *unordered* means
5702 that either operand may be a QNAN.
5704 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5705 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5706 type. They must have identical types.
5711 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5712 condition code given as ``cond``. If the operands are vectors, then the
5713 vectors are compared element by element. Each comparison performed
5714 always yields an :ref:`i1 <t_integer>` result, as follows:
5716 #. ``false``: always yields ``false``, regardless of operands.
5717 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5718 is equal to ``op2``.
5719 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5720 is greater than ``op2``.
5721 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5722 is greater than or equal to ``op2``.
5723 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5724 is less than ``op2``.
5725 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5726 is less than or equal to ``op2``.
5727 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5728 is not equal to ``op2``.
5729 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5730 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5732 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5733 greater than ``op2``.
5734 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5735 greater than or equal to ``op2``.
5736 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5738 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5739 less than or equal to ``op2``.
5740 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5741 not equal to ``op2``.
5742 #. ``uno``: yields ``true`` if either operand is a QNAN.
5743 #. ``true``: always yields ``true``, regardless of operands.
5748 .. code-block:: llvm
5750 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5751 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5752 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5753 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5755 Note that the code generator does not yet support vector types with the
5756 ``fcmp`` instruction.
5760 '``phi``' Instruction
5761 ^^^^^^^^^^^^^^^^^^^^^
5768 <result> = phi <ty> [ <val0>, <label0>], ...
5773 The '``phi``' instruction is used to implement the φ node in the SSA
5774 graph representing the function.
5779 The type of the incoming values is specified with the first type field.
5780 After this, the '``phi``' instruction takes a list of pairs as
5781 arguments, with one pair for each predecessor basic block of the current
5782 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5783 the value arguments to the PHI node. Only labels may be used as the
5786 There must be no non-phi instructions between the start of a basic block
5787 and the PHI instructions: i.e. PHI instructions must be first in a basic
5790 For the purposes of the SSA form, the use of each incoming value is
5791 deemed to occur on the edge from the corresponding predecessor block to
5792 the current block (but after any definition of an '``invoke``'
5793 instruction's return value on the same edge).
5798 At runtime, the '``phi``' instruction logically takes on the value
5799 specified by the pair corresponding to the predecessor basic block that
5800 executed just prior to the current block.
5805 .. code-block:: llvm
5807 Loop: ; Infinite loop that counts from 0 on up...
5808 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5809 %nextindvar = add i32 %indvar, 1
5814 '``select``' Instruction
5815 ^^^^^^^^^^^^^^^^^^^^^^^^
5822 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5824 selty is either i1 or {<N x i1>}
5829 The '``select``' instruction is used to choose one value based on a
5830 condition, without branching.
5835 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5836 values indicating the condition, and two values of the same :ref:`first
5837 class <t_firstclass>` type. If the val1/val2 are vectors and the
5838 condition is a scalar, then entire vectors are selected, not individual
5844 If the condition is an i1 and it evaluates to 1, the instruction returns
5845 the first value argument; otherwise, it returns the second value
5848 If the condition is a vector of i1, then the value arguments must be
5849 vectors of the same size, and the selection is done element by element.
5854 .. code-block:: llvm
5856 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5860 '``call``' Instruction
5861 ^^^^^^^^^^^^^^^^^^^^^^
5868 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5873 The '``call``' instruction represents a simple function call.
5878 This instruction requires several arguments:
5880 #. The optional "tail" marker indicates that the callee function does
5881 not access any allocas or varargs in the caller. Note that calls may
5882 be marked "tail" even if they do not occur before a
5883 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5884 function call is eligible for tail call optimization, but `might not
5885 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5886 The code generator may optimize calls marked "tail" with either 1)
5887 automatic `sibling call
5888 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5889 callee have matching signatures, or 2) forced tail call optimization
5890 when the following extra requirements are met:
5892 - Caller and callee both have the calling convention ``fastcc``.
5893 - The call is in tail position (ret immediately follows call and ret
5894 uses value of call or is void).
5895 - Option ``-tailcallopt`` is enabled, or
5896 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5897 - `Platform specific constraints are
5898 met. <CodeGenerator.html#tailcallopt>`_
5900 #. The optional "cconv" marker indicates which :ref:`calling
5901 convention <callingconv>` the call should use. If none is
5902 specified, the call defaults to using C calling conventions. The
5903 calling convention of the call must match the calling convention of
5904 the target function, or else the behavior is undefined.
5905 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5906 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5908 #. '``ty``': the type of the call instruction itself which is also the
5909 type of the return value. Functions that return no value are marked
5911 #. '``fnty``': shall be the signature of the pointer to function value
5912 being invoked. The argument types must match the types implied by
5913 this signature. This type can be omitted if the function is not
5914 varargs and if the function type does not return a pointer to a
5916 #. '``fnptrval``': An LLVM value containing a pointer to a function to
5917 be invoked. In most cases, this is a direct function invocation, but
5918 indirect ``call``'s are just as possible, calling an arbitrary pointer
5920 #. '``function args``': argument list whose types match the function
5921 signature argument types and parameter attributes. All arguments must
5922 be of :ref:`first class <t_firstclass>` type. If the function signature
5923 indicates the function accepts a variable number of arguments, the
5924 extra arguments can be specified.
5925 #. The optional :ref:`function attributes <fnattrs>` list. Only
5926 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5927 attributes are valid here.
5932 The '``call``' instruction is used to cause control flow to transfer to
5933 a specified function, with its incoming arguments bound to the specified
5934 values. Upon a '``ret``' instruction in the called function, control
5935 flow continues with the instruction after the function call, and the
5936 return value of the function is bound to the result argument.
5941 .. code-block:: llvm
5943 %retval = call i32 @test(i32 %argc)
5944 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
5945 %X = tail call i32 @foo() ; yields i32
5946 %Y = tail call fastcc i32 @foo() ; yields i32
5947 call void %foo(i8 97 signext)
5949 %struct.A = type { i32, i8 }
5950 %r = call %struct.A @foo() ; yields { 32, i8 }
5951 %gr = extractvalue %struct.A %r, 0 ; yields i32
5952 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
5953 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
5954 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
5956 llvm treats calls to some functions with names and arguments that match
5957 the standard C99 library as being the C99 library functions, and may
5958 perform optimizations or generate code for them under that assumption.
5959 This is something we'd like to change in the future to provide better
5960 support for freestanding environments and non-C-based languages.
5964 '``va_arg``' Instruction
5965 ^^^^^^^^^^^^^^^^^^^^^^^^
5972 <resultval> = va_arg <va_list*> <arglist>, <argty>
5977 The '``va_arg``' instruction is used to access arguments passed through
5978 the "variable argument" area of a function call. It is used to implement
5979 the ``va_arg`` macro in C.
5984 This instruction takes a ``va_list*`` value and the type of the
5985 argument. It returns a value of the specified argument type and
5986 increments the ``va_list`` to point to the next argument. The actual
5987 type of ``va_list`` is target specific.
5992 The '``va_arg``' instruction loads an argument of the specified type
5993 from the specified ``va_list`` and causes the ``va_list`` to point to
5994 the next argument. For more information, see the variable argument
5995 handling :ref:`Intrinsic Functions <int_varargs>`.
5997 It is legal for this instruction to be called in a function which does
5998 not take a variable number of arguments, for example, the ``vfprintf``
6001 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6002 function <intrinsics>` because it takes a type as an argument.
6007 See the :ref:`variable argument processing <int_varargs>` section.
6009 Note that the code generator does not yet fully support va\_arg on many
6010 targets. Also, it does not currently support va\_arg with aggregate
6011 types on any target.
6015 '``landingpad``' Instruction
6016 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6023 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6024 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6026 <clause> := catch <type> <value>
6027 <clause> := filter <array constant type> <array constant>
6032 The '``landingpad``' instruction is used by `LLVM's exception handling
6033 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6034 is a landing pad --- one where the exception lands, and corresponds to the
6035 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6036 defines values supplied by the personality function (``pers_fn``) upon
6037 re-entry to the function. The ``resultval`` has the type ``resultty``.
6042 This instruction takes a ``pers_fn`` value. This is the personality
6043 function associated with the unwinding mechanism. The optional
6044 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6046 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6047 contains the global variable representing the "type" that may be caught
6048 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6049 clause takes an array constant as its argument. Use
6050 "``[0 x i8**] undef``" for a filter which cannot throw. The
6051 '``landingpad``' instruction must contain *at least* one ``clause`` or
6052 the ``cleanup`` flag.
6057 The '``landingpad``' instruction defines the values which are set by the
6058 personality function (``pers_fn``) upon re-entry to the function, and
6059 therefore the "result type" of the ``landingpad`` instruction. As with
6060 calling conventions, how the personality function results are
6061 represented in LLVM IR is target specific.
6063 The clauses are applied in order from top to bottom. If two
6064 ``landingpad`` instructions are merged together through inlining, the
6065 clauses from the calling function are appended to the list of clauses.
6066 When the call stack is being unwound due to an exception being thrown,
6067 the exception is compared against each ``clause`` in turn. If it doesn't
6068 match any of the clauses, and the ``cleanup`` flag is not set, then
6069 unwinding continues further up the call stack.
6071 The ``landingpad`` instruction has several restrictions:
6073 - A landing pad block is a basic block which is the unwind destination
6074 of an '``invoke``' instruction.
6075 - A landing pad block must have a '``landingpad``' instruction as its
6076 first non-PHI instruction.
6077 - There can be only one '``landingpad``' instruction within the landing
6079 - A basic block that is not a landing pad block may not include a
6080 '``landingpad``' instruction.
6081 - All '``landingpad``' instructions in a function must have the same
6082 personality function.
6087 .. code-block:: llvm
6089 ;; A landing pad which can catch an integer.
6090 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6092 ;; A landing pad that is a cleanup.
6093 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6095 ;; A landing pad which can catch an integer and can only throw a double.
6096 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6098 filter [1 x i8**] [@_ZTId]
6105 LLVM supports the notion of an "intrinsic function". These functions
6106 have well known names and semantics and are required to follow certain
6107 restrictions. Overall, these intrinsics represent an extension mechanism
6108 for the LLVM language that does not require changing all of the
6109 transformations in LLVM when adding to the language (or the bitcode
6110 reader/writer, the parser, etc...).
6112 Intrinsic function names must all start with an "``llvm.``" prefix. This
6113 prefix is reserved in LLVM for intrinsic names; thus, function names may
6114 not begin with this prefix. Intrinsic functions must always be external
6115 functions: you cannot define the body of intrinsic functions. Intrinsic
6116 functions may only be used in call or invoke instructions: it is illegal
6117 to take the address of an intrinsic function. Additionally, because
6118 intrinsic functions are part of the LLVM language, it is required if any
6119 are added that they be documented here.
6121 Some intrinsic functions can be overloaded, i.e., the intrinsic
6122 represents a family of functions that perform the same operation but on
6123 different data types. Because LLVM can represent over 8 million
6124 different integer types, overloading is used commonly to allow an
6125 intrinsic function to operate on any integer type. One or more of the
6126 argument types or the result type can be overloaded to accept any
6127 integer type. Argument types may also be defined as exactly matching a
6128 previous argument's type or the result type. This allows an intrinsic
6129 function which accepts multiple arguments, but needs all of them to be
6130 of the same type, to only be overloaded with respect to a single
6131 argument or the result.
6133 Overloaded intrinsics will have the names of its overloaded argument
6134 types encoded into its function name, each preceded by a period. Only
6135 those types which are overloaded result in a name suffix. Arguments
6136 whose type is matched against another type do not. For example, the
6137 ``llvm.ctpop`` function can take an integer of any width and returns an
6138 integer of exactly the same integer width. This leads to a family of
6139 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6140 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6141 overloaded, and only one type suffix is required. Because the argument's
6142 type is matched against the return type, it does not require its own
6145 To learn how to add an intrinsic function, please see the `Extending
6146 LLVM Guide <ExtendingLLVM.html>`_.
6150 Variable Argument Handling Intrinsics
6151 -------------------------------------
6153 Variable argument support is defined in LLVM with the
6154 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6155 functions. These functions are related to the similarly named macros
6156 defined in the ``<stdarg.h>`` header file.
6158 All of these functions operate on arguments that use a target-specific
6159 value type "``va_list``". The LLVM assembly language reference manual
6160 does not define what this type is, so all transformations should be
6161 prepared to handle these functions regardless of the type used.
6163 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6164 variable argument handling intrinsic functions are used.
6166 .. code-block:: llvm
6168 define i32 @test(i32 %X, ...) {
6169 ; Initialize variable argument processing
6171 %ap2 = bitcast i8** %ap to i8*
6172 call void @llvm.va_start(i8* %ap2)
6174 ; Read a single integer argument
6175 %tmp = va_arg i8** %ap, i32
6177 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6179 %aq2 = bitcast i8** %aq to i8*
6180 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6181 call void @llvm.va_end(i8* %aq2)
6183 ; Stop processing of arguments.
6184 call void @llvm.va_end(i8* %ap2)
6188 declare void @llvm.va_start(i8*)
6189 declare void @llvm.va_copy(i8*, i8*)
6190 declare void @llvm.va_end(i8*)
6194 '``llvm.va_start``' Intrinsic
6195 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6202 declare void %llvm.va_start(i8* <arglist>)
6207 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6208 subsequent use by ``va_arg``.
6213 The argument is a pointer to a ``va_list`` element to initialize.
6218 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6219 available in C. In a target-dependent way, it initializes the
6220 ``va_list`` element to which the argument points, so that the next call
6221 to ``va_arg`` will produce the first variable argument passed to the
6222 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6223 to know the last argument of the function as the compiler can figure
6226 '``llvm.va_end``' Intrinsic
6227 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6234 declare void @llvm.va_end(i8* <arglist>)
6239 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6240 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6245 The argument is a pointer to a ``va_list`` to destroy.
6250 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6251 available in C. In a target-dependent way, it destroys the ``va_list``
6252 element to which the argument points. Calls to
6253 :ref:`llvm.va_start <int_va_start>` and
6254 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6259 '``llvm.va_copy``' Intrinsic
6260 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6267 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6272 The '``llvm.va_copy``' intrinsic copies the current argument position
6273 from the source argument list to the destination argument list.
6278 The first argument is a pointer to a ``va_list`` element to initialize.
6279 The second argument is a pointer to a ``va_list`` element to copy from.
6284 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6285 available in C. In a target-dependent way, it copies the source
6286 ``va_list`` element into the destination ``va_list`` element. This
6287 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6288 arbitrarily complex and require, for example, memory allocation.
6290 Accurate Garbage Collection Intrinsics
6291 --------------------------------------
6293 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6294 (GC) requires the implementation and generation of these intrinsics.
6295 These intrinsics allow identification of :ref:`GC roots on the
6296 stack <int_gcroot>`, as well as garbage collector implementations that
6297 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6298 Front-ends for type-safe garbage collected languages should generate
6299 these intrinsics to make use of the LLVM garbage collectors. For more
6300 details, see `Accurate Garbage Collection with
6301 LLVM <GarbageCollection.html>`_.
6303 The garbage collection intrinsics only operate on objects in the generic
6304 address space (address space zero).
6308 '``llvm.gcroot``' Intrinsic
6309 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6316 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6321 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6322 the code generator, and allows some metadata to be associated with it.
6327 The first argument specifies the address of a stack object that contains
6328 the root pointer. The second pointer (which must be either a constant or
6329 a global value address) contains the meta-data to be associated with the
6335 At runtime, a call to this intrinsic stores a null pointer into the
6336 "ptrloc" location. At compile-time, the code generator generates
6337 information to allow the runtime to find the pointer at GC safe points.
6338 The '``llvm.gcroot``' intrinsic may only be used in a function which
6339 :ref:`specifies a GC algorithm <gc>`.
6343 '``llvm.gcread``' Intrinsic
6344 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6351 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6356 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6357 locations, allowing garbage collector implementations that require read
6363 The second argument is the address to read from, which should be an
6364 address allocated from the garbage collector. The first object is a
6365 pointer to the start of the referenced object, if needed by the language
6366 runtime (otherwise null).
6371 The '``llvm.gcread``' intrinsic has the same semantics as a load
6372 instruction, but may be replaced with substantially more complex code by
6373 the garbage collector runtime, as needed. The '``llvm.gcread``'
6374 intrinsic may only be used in a function which :ref:`specifies a GC
6379 '``llvm.gcwrite``' Intrinsic
6380 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6387 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6392 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6393 locations, allowing garbage collector implementations that require write
6394 barriers (such as generational or reference counting collectors).
6399 The first argument is the reference to store, the second is the start of
6400 the object to store it to, and the third is the address of the field of
6401 Obj to store to. If the runtime does not require a pointer to the
6402 object, Obj may be null.
6407 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6408 instruction, but may be replaced with substantially more complex code by
6409 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6410 intrinsic may only be used in a function which :ref:`specifies a GC
6413 Code Generator Intrinsics
6414 -------------------------
6416 These intrinsics are provided by LLVM to expose special features that
6417 may only be implemented with code generator support.
6419 '``llvm.returnaddress``' Intrinsic
6420 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6427 declare i8 *@llvm.returnaddress(i32 <level>)
6432 The '``llvm.returnaddress``' intrinsic attempts to compute a
6433 target-specific value indicating the return address of the current
6434 function or one of its callers.
6439 The argument to this intrinsic indicates which function to return the
6440 address for. Zero indicates the calling function, one indicates its
6441 caller, etc. The argument is **required** to be a constant integer
6447 The '``llvm.returnaddress``' intrinsic either returns a pointer
6448 indicating the return address of the specified call frame, or zero if it
6449 cannot be identified. The value returned by this intrinsic is likely to
6450 be incorrect or 0 for arguments other than zero, so it should only be
6451 used for debugging purposes.
6453 Note that calling this intrinsic does not prevent function inlining or
6454 other aggressive transformations, so the value returned may not be that
6455 of the obvious source-language caller.
6457 '``llvm.frameaddress``' Intrinsic
6458 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6465 declare i8* @llvm.frameaddress(i32 <level>)
6470 The '``llvm.frameaddress``' intrinsic attempts to return the
6471 target-specific frame pointer value for the specified stack frame.
6476 The argument to this intrinsic indicates which function to return the
6477 frame pointer for. Zero indicates the calling function, one indicates
6478 its caller, etc. The argument is **required** to be a constant integer
6484 The '``llvm.frameaddress``' intrinsic either returns a pointer
6485 indicating the frame address of the specified call frame, or zero if it
6486 cannot be identified. The value returned by this intrinsic is likely to
6487 be incorrect or 0 for arguments other than zero, so it should only be
6488 used for debugging purposes.
6490 Note that calling this intrinsic does not prevent function inlining or
6491 other aggressive transformations, so the value returned may not be that
6492 of the obvious source-language caller.
6496 '``llvm.stacksave``' Intrinsic
6497 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6504 declare i8* @llvm.stacksave()
6509 The '``llvm.stacksave``' intrinsic is used to remember the current state
6510 of the function stack, for use with
6511 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6512 implementing language features like scoped automatic variable sized
6518 This intrinsic returns a opaque pointer value that can be passed to
6519 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6520 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6521 ``llvm.stacksave``, it effectively restores the state of the stack to
6522 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6523 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6524 were allocated after the ``llvm.stacksave`` was executed.
6526 .. _int_stackrestore:
6528 '``llvm.stackrestore``' Intrinsic
6529 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6536 declare void @llvm.stackrestore(i8* %ptr)
6541 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6542 the function stack to the state it was in when the corresponding
6543 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6544 useful for implementing language features like scoped automatic variable
6545 sized arrays in C99.
6550 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6552 '``llvm.prefetch``' Intrinsic
6553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6560 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6565 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6566 insert a prefetch instruction if supported; otherwise, it is a noop.
6567 Prefetches have no effect on the behavior of the program but can change
6568 its performance characteristics.
6573 ``address`` is the address to be prefetched, ``rw`` is the specifier
6574 determining if the fetch should be for a read (0) or write (1), and
6575 ``locality`` is a temporal locality specifier ranging from (0) - no
6576 locality, to (3) - extremely local keep in cache. The ``cache type``
6577 specifies whether the prefetch is performed on the data (1) or
6578 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6579 arguments must be constant integers.
6584 This intrinsic does not modify the behavior of the program. In
6585 particular, prefetches cannot trap and do not produce a value. On
6586 targets that support this intrinsic, the prefetch can provide hints to
6587 the processor cache for better performance.
6589 '``llvm.pcmarker``' Intrinsic
6590 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6597 declare void @llvm.pcmarker(i32 <id>)
6602 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6603 Counter (PC) in a region of code to simulators and other tools. The
6604 method is target specific, but it is expected that the marker will use
6605 exported symbols to transmit the PC of the marker. The marker makes no
6606 guarantees that it will remain with any specific instruction after
6607 optimizations. It is possible that the presence of a marker will inhibit
6608 optimizations. The intended use is to be inserted after optimizations to
6609 allow correlations of simulation runs.
6614 ``id`` is a numerical id identifying the marker.
6619 This intrinsic does not modify the behavior of the program. Backends
6620 that do not support this intrinsic may ignore it.
6622 '``llvm.readcyclecounter``' Intrinsic
6623 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6630 declare i64 @llvm.readcyclecounter()
6635 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6636 counter register (or similar low latency, high accuracy clocks) on those
6637 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6638 should map to RPCC. As the backing counters overflow quickly (on the
6639 order of 9 seconds on alpha), this should only be used for small
6645 When directly supported, reading the cycle counter should not modify any
6646 memory. Implementations are allowed to either return a application
6647 specific value or a system wide value. On backends without support, this
6648 is lowered to a constant 0.
6650 Standard C Library Intrinsics
6651 -----------------------------
6653 LLVM provides intrinsics for a few important standard C library
6654 functions. These intrinsics allow source-language front-ends to pass
6655 information about the alignment of the pointer arguments to the code
6656 generator, providing opportunity for more efficient code generation.
6660 '``llvm.memcpy``' Intrinsic
6661 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6666 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6667 integer bit width and for different address spaces. Not all targets
6668 support all bit widths however.
6672 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6673 i32 <len>, i32 <align>, i1 <isvolatile>)
6674 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6675 i64 <len>, i32 <align>, i1 <isvolatile>)
6680 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6681 source location to the destination location.
6683 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6684 intrinsics do not return a value, takes extra alignment/isvolatile
6685 arguments and the pointers can be in specified address spaces.
6690 The first argument is a pointer to the destination, the second is a
6691 pointer to the source. The third argument is an integer argument
6692 specifying the number of bytes to copy, the fourth argument is the
6693 alignment of the source and destination locations, and the fifth is a
6694 boolean indicating a volatile access.
6696 If the call to this intrinsic has an alignment value that is not 0 or 1,
6697 then the caller guarantees that both the source and destination pointers
6698 are aligned to that boundary.
6700 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6701 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6702 very cleanly specified and it is unwise to depend on it.
6707 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6708 source location to the destination location, which are not allowed to
6709 overlap. It copies "len" bytes of memory over. If the argument is known
6710 to be aligned to some boundary, this can be specified as the fourth
6711 argument, otherwise it should be set to 0 or 1.
6713 '``llvm.memmove``' Intrinsic
6714 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6719 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6720 bit width and for different address space. Not all targets support all
6725 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6726 i32 <len>, i32 <align>, i1 <isvolatile>)
6727 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6728 i64 <len>, i32 <align>, i1 <isvolatile>)
6733 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6734 source location to the destination location. It is similar to the
6735 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6738 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6739 intrinsics do not return a value, takes extra alignment/isvolatile
6740 arguments and the pointers can be in specified address spaces.
6745 The first argument is a pointer to the destination, the second is a
6746 pointer to the source. The third argument is an integer argument
6747 specifying the number of bytes to copy, the fourth argument is the
6748 alignment of the source and destination locations, and the fifth is a
6749 boolean indicating a volatile access.
6751 If the call to this intrinsic has an alignment value that is not 0 or 1,
6752 then the caller guarantees that the source and destination pointers are
6753 aligned to that boundary.
6755 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6756 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6757 not very cleanly specified and it is unwise to depend on it.
6762 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6763 source location to the destination location, which may overlap. It
6764 copies "len" bytes of memory over. If the argument is known to be
6765 aligned to some boundary, this can be specified as the fourth argument,
6766 otherwise it should be set to 0 or 1.
6768 '``llvm.memset.*``' Intrinsics
6769 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6774 This is an overloaded intrinsic. You can use llvm.memset on any integer
6775 bit width and for different address spaces. However, not all targets
6776 support all bit widths.
6780 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6781 i32 <len>, i32 <align>, i1 <isvolatile>)
6782 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6783 i64 <len>, i32 <align>, i1 <isvolatile>)
6788 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6789 particular byte value.
6791 Note that, unlike the standard libc function, the ``llvm.memset``
6792 intrinsic does not return a value and takes extra alignment/volatile
6793 arguments. Also, the destination can be in an arbitrary address space.
6798 The first argument is a pointer to the destination to fill, the second
6799 is the byte value with which to fill it, the third argument is an
6800 integer argument specifying the number of bytes to fill, and the fourth
6801 argument is the known alignment of the destination location.
6803 If the call to this intrinsic has an alignment value that is not 0 or 1,
6804 then the caller guarantees that the destination pointer is aligned to
6807 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6808 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6809 very cleanly specified and it is unwise to depend on it.
6814 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6815 at the destination location. If the argument is known to be aligned to
6816 some boundary, this can be specified as the fourth argument, otherwise
6817 it should be set to 0 or 1.
6819 '``llvm.sqrt.*``' Intrinsic
6820 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6825 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6826 floating point or vector of floating point type. Not all targets support
6831 declare float @llvm.sqrt.f32(float %Val)
6832 declare double @llvm.sqrt.f64(double %Val)
6833 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6834 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6835 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6840 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6841 returning the same value as the libm '``sqrt``' functions would. Unlike
6842 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6843 negative numbers other than -0.0 (which allows for better optimization,
6844 because there is no need to worry about errno being set).
6845 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6850 The argument and return value are floating point numbers of the same
6856 This function returns the sqrt of the specified operand if it is a
6857 nonnegative floating point number.
6859 '``llvm.powi.*``' Intrinsic
6860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6865 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6866 floating point or vector of floating point type. Not all targets support
6871 declare float @llvm.powi.f32(float %Val, i32 %power)
6872 declare double @llvm.powi.f64(double %Val, i32 %power)
6873 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6874 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6875 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6880 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6881 specified (positive or negative) power. The order of evaluation of
6882 multiplications is not defined. When a vector of floating point type is
6883 used, the second argument remains a scalar integer value.
6888 The second argument is an integer power, and the first is a value to
6889 raise to that power.
6894 This function returns the first value raised to the second power with an
6895 unspecified sequence of rounding operations.
6897 '``llvm.sin.*``' Intrinsic
6898 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6903 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
6904 floating point or vector of floating point type. Not all targets support
6909 declare float @llvm.sin.f32(float %Val)
6910 declare double @llvm.sin.f64(double %Val)
6911 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6912 declare fp128 @llvm.sin.f128(fp128 %Val)
6913 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6918 The '``llvm.sin.*``' intrinsics return the sine of the operand.
6923 The argument and return value are floating point numbers of the same
6929 This function returns the sine of the specified operand, returning the
6930 same values as the libm ``sin`` functions would, and handles error
6931 conditions in the same way.
6933 '``llvm.cos.*``' Intrinsic
6934 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6939 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
6940 floating point or vector of floating point type. Not all targets support
6945 declare float @llvm.cos.f32(float %Val)
6946 declare double @llvm.cos.f64(double %Val)
6947 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6948 declare fp128 @llvm.cos.f128(fp128 %Val)
6949 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6954 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
6959 The argument and return value are floating point numbers of the same
6965 This function returns the cosine of the specified operand, returning the
6966 same values as the libm ``cos`` functions would, and handles error
6967 conditions in the same way.
6969 '``llvm.pow.*``' Intrinsic
6970 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6975 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
6976 floating point or vector of floating point type. Not all targets support
6981 declare float @llvm.pow.f32(float %Val, float %Power)
6982 declare double @llvm.pow.f64(double %Val, double %Power)
6983 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6984 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6985 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6990 The '``llvm.pow.*``' intrinsics return the first operand raised to the
6991 specified (positive or negative) power.
6996 The second argument is a floating point power, and the first is a value
6997 to raise to that power.
7002 This function returns the first value raised to the second power,
7003 returning the same values as the libm ``pow`` functions would, and
7004 handles error conditions in the same way.
7006 '``llvm.exp.*``' Intrinsic
7007 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7012 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7013 floating point or vector of floating point type. Not all targets support
7018 declare float @llvm.exp.f32(float %Val)
7019 declare double @llvm.exp.f64(double %Val)
7020 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7021 declare fp128 @llvm.exp.f128(fp128 %Val)
7022 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7027 The '``llvm.exp.*``' intrinsics perform the exp function.
7032 The argument and return value are floating point numbers of the same
7038 This function returns the same values as the libm ``exp`` functions
7039 would, and handles error conditions in the same way.
7041 '``llvm.exp2.*``' Intrinsic
7042 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7047 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7048 floating point or vector of floating point type. Not all targets support
7053 declare float @llvm.exp2.f32(float %Val)
7054 declare double @llvm.exp2.f64(double %Val)
7055 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7056 declare fp128 @llvm.exp2.f128(fp128 %Val)
7057 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7062 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7067 The argument and return value are floating point numbers of the same
7073 This function returns the same values as the libm ``exp2`` functions
7074 would, and handles error conditions in the same way.
7076 '``llvm.log.*``' Intrinsic
7077 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7082 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7083 floating point or vector of floating point type. Not all targets support
7088 declare float @llvm.log.f32(float %Val)
7089 declare double @llvm.log.f64(double %Val)
7090 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7091 declare fp128 @llvm.log.f128(fp128 %Val)
7092 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7097 The '``llvm.log.*``' intrinsics perform the log function.
7102 The argument and return value are floating point numbers of the same
7108 This function returns the same values as the libm ``log`` functions
7109 would, and handles error conditions in the same way.
7111 '``llvm.log10.*``' Intrinsic
7112 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7117 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7118 floating point or vector of floating point type. Not all targets support
7123 declare float @llvm.log10.f32(float %Val)
7124 declare double @llvm.log10.f64(double %Val)
7125 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7126 declare fp128 @llvm.log10.f128(fp128 %Val)
7127 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7132 The '``llvm.log10.*``' intrinsics perform the log10 function.
7137 The argument and return value are floating point numbers of the same
7143 This function returns the same values as the libm ``log10`` functions
7144 would, and handles error conditions in the same way.
7146 '``llvm.log2.*``' Intrinsic
7147 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7152 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7153 floating point or vector of floating point type. Not all targets support
7158 declare float @llvm.log2.f32(float %Val)
7159 declare double @llvm.log2.f64(double %Val)
7160 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7161 declare fp128 @llvm.log2.f128(fp128 %Val)
7162 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7167 The '``llvm.log2.*``' intrinsics perform the log2 function.
7172 The argument and return value are floating point numbers of the same
7178 This function returns the same values as the libm ``log2`` functions
7179 would, and handles error conditions in the same way.
7181 '``llvm.fma.*``' Intrinsic
7182 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7187 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7188 floating point or vector of floating point type. Not all targets support
7193 declare float @llvm.fma.f32(float %a, float %b, float %c)
7194 declare double @llvm.fma.f64(double %a, double %b, double %c)
7195 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7196 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7197 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7202 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7208 The argument and return value are floating point numbers of the same
7214 This function returns the same values as the libm ``fma`` functions
7217 '``llvm.fabs.*``' Intrinsic
7218 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7223 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7224 floating point or vector of floating point type. Not all targets support
7229 declare float @llvm.fabs.f32(float %Val)
7230 declare double @llvm.fabs.f64(double %Val)
7231 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7232 declare fp128 @llvm.fabs.f128(fp128 %Val)
7233 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7238 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7244 The argument and return value are floating point numbers of the same
7250 This function returns the same values as the libm ``fabs`` functions
7251 would, and handles error conditions in the same way.
7253 '``llvm.floor.*``' Intrinsic
7254 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7259 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7260 floating point or vector of floating point type. Not all targets support
7265 declare float @llvm.floor.f32(float %Val)
7266 declare double @llvm.floor.f64(double %Val)
7267 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7268 declare fp128 @llvm.floor.f128(fp128 %Val)
7269 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7274 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7279 The argument and return value are floating point numbers of the same
7285 This function returns the same values as the libm ``floor`` functions
7286 would, and handles error conditions in the same way.
7288 '``llvm.ceil.*``' Intrinsic
7289 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7294 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7295 floating point or vector of floating point type. Not all targets support
7300 declare float @llvm.ceil.f32(float %Val)
7301 declare double @llvm.ceil.f64(double %Val)
7302 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7303 declare fp128 @llvm.ceil.f128(fp128 %Val)
7304 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7309 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7314 The argument and return value are floating point numbers of the same
7320 This function returns the same values as the libm ``ceil`` functions
7321 would, and handles error conditions in the same way.
7323 '``llvm.trunc.*``' Intrinsic
7324 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7329 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7330 floating point or vector of floating point type. Not all targets support
7335 declare float @llvm.trunc.f32(float %Val)
7336 declare double @llvm.trunc.f64(double %Val)
7337 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7338 declare fp128 @llvm.trunc.f128(fp128 %Val)
7339 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7344 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7345 nearest integer not larger in magnitude than the operand.
7350 The argument and return value are floating point numbers of the same
7356 This function returns the same values as the libm ``trunc`` functions
7357 would, and handles error conditions in the same way.
7359 '``llvm.rint.*``' Intrinsic
7360 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7365 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7366 floating point or vector of floating point type. Not all targets support
7371 declare float @llvm.rint.f32(float %Val)
7372 declare double @llvm.rint.f64(double %Val)
7373 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7374 declare fp128 @llvm.rint.f128(fp128 %Val)
7375 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7380 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7381 nearest integer. It may raise an inexact floating-point exception if the
7382 operand isn't an integer.
7387 The argument and return value are floating point numbers of the same
7393 This function returns the same values as the libm ``rint`` functions
7394 would, and handles error conditions in the same way.
7396 '``llvm.nearbyint.*``' Intrinsic
7397 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7402 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7403 floating point or vector of floating point type. Not all targets support
7408 declare float @llvm.nearbyint.f32(float %Val)
7409 declare double @llvm.nearbyint.f64(double %Val)
7410 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7411 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7412 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7417 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7423 The argument and return value are floating point numbers of the same
7429 This function returns the same values as the libm ``nearbyint``
7430 functions would, and handles error conditions in the same way.
7432 Bit Manipulation Intrinsics
7433 ---------------------------
7435 LLVM provides intrinsics for a few important bit manipulation
7436 operations. These allow efficient code generation for some algorithms.
7438 '``llvm.bswap.*``' Intrinsics
7439 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7444 This is an overloaded intrinsic function. You can use bswap on any
7445 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7449 declare i16 @llvm.bswap.i16(i16 <id>)
7450 declare i32 @llvm.bswap.i32(i32 <id>)
7451 declare i64 @llvm.bswap.i64(i64 <id>)
7456 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7457 values with an even number of bytes (positive multiple of 16 bits).
7458 These are useful for performing operations on data that is not in the
7459 target's native byte order.
7464 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7465 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7466 intrinsic returns an i32 value that has the four bytes of the input i32
7467 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7468 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7469 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7470 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7473 '``llvm.ctpop.*``' Intrinsic
7474 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7479 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7480 bit width, or on any vector with integer elements. Not all targets
7481 support all bit widths or vector types, however.
7485 declare i8 @llvm.ctpop.i8(i8 <src>)
7486 declare i16 @llvm.ctpop.i16(i16 <src>)
7487 declare i32 @llvm.ctpop.i32(i32 <src>)
7488 declare i64 @llvm.ctpop.i64(i64 <src>)
7489 declare i256 @llvm.ctpop.i256(i256 <src>)
7490 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7495 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7501 The only argument is the value to be counted. The argument may be of any
7502 integer type, or a vector with integer elements. The return type must
7503 match the argument type.
7508 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7509 each element of a vector.
7511 '``llvm.ctlz.*``' Intrinsic
7512 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7517 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7518 integer bit width, or any vector whose elements are integers. Not all
7519 targets support all bit widths or vector types, however.
7523 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7524 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7525 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7526 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7527 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7528 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7533 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7534 leading zeros in a variable.
7539 The first argument is the value to be counted. This argument may be of
7540 any integer type, or a vectory with integer element type. The return
7541 type must match the first argument type.
7543 The second argument must be a constant and is a flag to indicate whether
7544 the intrinsic should ensure that a zero as the first argument produces a
7545 defined result. Historically some architectures did not provide a
7546 defined result for zero values as efficiently, and many algorithms are
7547 now predicated on avoiding zero-value inputs.
7552 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7553 zeros in a variable, or within each element of the vector. If
7554 ``src == 0`` then the result is the size in bits of the type of ``src``
7555 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7556 ``llvm.ctlz(i32 2) = 30``.
7558 '``llvm.cttz.*``' Intrinsic
7559 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7564 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7565 integer bit width, or any vector of integer elements. Not all targets
7566 support all bit widths or vector types, however.
7570 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7571 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7572 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7573 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7574 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7575 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7580 The '``llvm.cttz``' family of intrinsic functions counts the number of
7586 The first argument is the value to be counted. This argument may be of
7587 any integer type, or a vectory with integer element type. The return
7588 type must match the first argument type.
7590 The second argument must be a constant and is a flag to indicate whether
7591 the intrinsic should ensure that a zero as the first argument produces a
7592 defined result. Historically some architectures did not provide a
7593 defined result for zero values as efficiently, and many algorithms are
7594 now predicated on avoiding zero-value inputs.
7599 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7600 zeros in a variable, or within each element of a vector. If ``src == 0``
7601 then the result is the size in bits of the type of ``src`` if
7602 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7603 ``llvm.cttz(2) = 1``.
7605 Arithmetic with Overflow Intrinsics
7606 -----------------------------------
7608 LLVM provides intrinsics for some arithmetic with overflow operations.
7610 '``llvm.sadd.with.overflow.*``' Intrinsics
7611 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7616 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7617 on any integer bit width.
7621 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7622 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7623 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7628 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7629 a signed addition of the two arguments, and indicate whether an overflow
7630 occurred during the signed summation.
7635 The arguments (%a and %b) and the first element of the result structure
7636 may be of integer types of any bit width, but they must have the same
7637 bit width. The second element of the result structure must be of type
7638 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7644 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7645 a signed addition of the two variables. They return a structure --- the
7646 first element of which is the signed summation, and the second element
7647 of which is a bit specifying if the signed summation resulted in an
7653 .. code-block:: llvm
7655 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7656 %sum = extractvalue {i32, i1} %res, 0
7657 %obit = extractvalue {i32, i1} %res, 1
7658 br i1 %obit, label %overflow, label %normal
7660 '``llvm.uadd.with.overflow.*``' Intrinsics
7661 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7666 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7667 on any integer bit width.
7671 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7672 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7673 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7678 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7679 an unsigned addition of the two arguments, and indicate whether a carry
7680 occurred during the unsigned summation.
7685 The arguments (%a and %b) and the first element of the result structure
7686 may be of integer types of any bit width, but they must have the same
7687 bit width. The second element of the result structure must be of type
7688 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7694 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7695 an unsigned addition of the two arguments. They return a structure --- the
7696 first element of which is the sum, and the second element of which is a
7697 bit specifying if the unsigned summation resulted in a carry.
7702 .. code-block:: llvm
7704 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7705 %sum = extractvalue {i32, i1} %res, 0
7706 %obit = extractvalue {i32, i1} %res, 1
7707 br i1 %obit, label %carry, label %normal
7709 '``llvm.ssub.with.overflow.*``' Intrinsics
7710 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7715 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7716 on any integer bit width.
7720 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7721 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7722 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7727 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7728 a signed subtraction of the two arguments, and indicate whether an
7729 overflow occurred during the signed subtraction.
7734 The arguments (%a and %b) and the first element of the result structure
7735 may be of integer types of any bit width, but they must have the same
7736 bit width. The second element of the result structure must be of type
7737 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7743 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7744 a signed subtraction of the two arguments. They return a structure --- the
7745 first element of which is the subtraction, and the second element of
7746 which is a bit specifying if the signed subtraction resulted in an
7752 .. code-block:: llvm
7754 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7755 %sum = extractvalue {i32, i1} %res, 0
7756 %obit = extractvalue {i32, i1} %res, 1
7757 br i1 %obit, label %overflow, label %normal
7759 '``llvm.usub.with.overflow.*``' Intrinsics
7760 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7765 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7766 on any integer bit width.
7770 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7771 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7772 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7777 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7778 an unsigned subtraction of the two arguments, and indicate whether an
7779 overflow occurred during the unsigned subtraction.
7784 The arguments (%a and %b) and the first element of the result structure
7785 may be of integer types of any bit width, but they must have the same
7786 bit width. The second element of the result structure must be of type
7787 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7793 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7794 an unsigned subtraction of the two arguments. They return a structure ---
7795 the first element of which is the subtraction, and the second element of
7796 which is a bit specifying if the unsigned subtraction resulted in an
7802 .. code-block:: llvm
7804 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7805 %sum = extractvalue {i32, i1} %res, 0
7806 %obit = extractvalue {i32, i1} %res, 1
7807 br i1 %obit, label %overflow, label %normal
7809 '``llvm.smul.with.overflow.*``' Intrinsics
7810 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7815 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7816 on any integer bit width.
7820 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7821 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7822 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7827 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7828 a signed multiplication of the two arguments, and indicate whether an
7829 overflow occurred during the signed multiplication.
7834 The arguments (%a and %b) and the first element of the result structure
7835 may be of integer types of any bit width, but they must have the same
7836 bit width. The second element of the result structure must be of type
7837 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7843 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7844 a signed multiplication of the two arguments. They return a structure ---
7845 the first element of which is the multiplication, and the second element
7846 of which is a bit specifying if the signed multiplication resulted in an
7852 .. code-block:: llvm
7854 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7855 %sum = extractvalue {i32, i1} %res, 0
7856 %obit = extractvalue {i32, i1} %res, 1
7857 br i1 %obit, label %overflow, label %normal
7859 '``llvm.umul.with.overflow.*``' Intrinsics
7860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7865 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
7866 on any integer bit width.
7870 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7871 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7872 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7877 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7878 a unsigned multiplication of the two arguments, and indicate whether an
7879 overflow occurred during the unsigned multiplication.
7884 The arguments (%a and %b) and the first element of the result structure
7885 may be of integer types of any bit width, but they must have the same
7886 bit width. The second element of the result structure must be of type
7887 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7893 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7894 an unsigned multiplication of the two arguments. They return a structure ---
7895 the first element of which is the multiplication, and the second
7896 element of which is a bit specifying if the unsigned multiplication
7897 resulted in an overflow.
7902 .. code-block:: llvm
7904 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7905 %sum = extractvalue {i32, i1} %res, 0
7906 %obit = extractvalue {i32, i1} %res, 1
7907 br i1 %obit, label %overflow, label %normal
7909 Specialised Arithmetic Intrinsics
7910 ---------------------------------
7912 '``llvm.fmuladd.*``' Intrinsic
7913 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7920 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
7921 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
7926 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
7927 expressions that can be fused if the code generator determines that (a) the
7928 target instruction set has support for a fused operation, and (b) that the
7929 fused operation is more efficient than the equivalent, separate pair of mul
7930 and add instructions.
7935 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
7936 multiplicands, a and b, and an addend c.
7945 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
7947 is equivalent to the expression a \* b + c, except that rounding will
7948 not be performed between the multiplication and addition steps if the
7949 code generator fuses the operations. Fusion is not guaranteed, even if
7950 the target platform supports it. If a fused multiply-add is required the
7951 corresponding llvm.fma.\* intrinsic function should be used instead.
7956 .. code-block:: llvm
7958 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
7960 Half Precision Floating Point Intrinsics
7961 ----------------------------------------
7963 For most target platforms, half precision floating point is a
7964 storage-only format. This means that it is a dense encoding (in memory)
7965 but does not support computation in the format.
7967 This means that code must first load the half-precision floating point
7968 value as an i16, then convert it to float with
7969 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
7970 then be performed on the float value (including extending to double
7971 etc). To store the value back to memory, it is first converted to float
7972 if needed, then converted to i16 with
7973 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
7976 .. _int_convert_to_fp16:
7978 '``llvm.convert.to.fp16``' Intrinsic
7979 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7986 declare i16 @llvm.convert.to.fp16(f32 %a)
7991 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7992 from single precision floating point format to half precision floating
7998 The intrinsic function contains single argument - the value to be
8004 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8005 from single precision floating point format to half precision floating
8006 point format. The return value is an ``i16`` which contains the
8012 .. code-block:: llvm
8014 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8015 store i16 %res, i16* @x, align 2
8017 .. _int_convert_from_fp16:
8019 '``llvm.convert.from.fp16``' Intrinsic
8020 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8027 declare f32 @llvm.convert.from.fp16(i16 %a)
8032 The '``llvm.convert.from.fp16``' intrinsic function performs a
8033 conversion from half precision floating point format to single precision
8034 floating point format.
8039 The intrinsic function contains single argument - the value to be
8045 The '``llvm.convert.from.fp16``' intrinsic function performs a
8046 conversion from half single precision floating point format to single
8047 precision floating point format. The input half-float value is
8048 represented by an ``i16`` value.
8053 .. code-block:: llvm
8055 %a = load i16* @x, align 2
8056 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8061 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8062 prefix), are described in the `LLVM Source Level
8063 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8066 Exception Handling Intrinsics
8067 -----------------------------
8069 The LLVM exception handling intrinsics (which all start with
8070 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8071 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8075 Trampoline Intrinsics
8076 ---------------------
8078 These intrinsics make it possible to excise one parameter, marked with
8079 the :ref:`nest <nest>` attribute, from a function. The result is a
8080 callable function pointer lacking the nest parameter - the caller does
8081 not need to provide a value for it. Instead, the value to use is stored
8082 in advance in a "trampoline", a block of memory usually allocated on the
8083 stack, which also contains code to splice the nest value into the
8084 argument list. This is used to implement the GCC nested function address
8087 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8088 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8089 It can be created as follows:
8091 .. code-block:: llvm
8093 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8094 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8095 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8096 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8097 %fp = bitcast i8* %p to i32 (i32, i32)*
8099 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8100 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8104 '``llvm.init.trampoline``' Intrinsic
8105 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8112 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8117 This fills the memory pointed to by ``tramp`` with executable code,
8118 turning it into a trampoline.
8123 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8124 pointers. The ``tramp`` argument must point to a sufficiently large and
8125 sufficiently aligned block of memory; this memory is written to by the
8126 intrinsic. Note that the size and the alignment are target-specific -
8127 LLVM currently provides no portable way of determining them, so a
8128 front-end that generates this intrinsic needs to have some
8129 target-specific knowledge. The ``func`` argument must hold a function
8130 bitcast to an ``i8*``.
8135 The block of memory pointed to by ``tramp`` is filled with target
8136 dependent code, turning it into a function. Then ``tramp`` needs to be
8137 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8138 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8139 function's signature is the same as that of ``func`` with any arguments
8140 marked with the ``nest`` attribute removed. At most one such ``nest``
8141 argument is allowed, and it must be of pointer type. Calling the new
8142 function is equivalent to calling ``func`` with the same argument list,
8143 but with ``nval`` used for the missing ``nest`` argument. If, after
8144 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8145 modified, then the effect of any later call to the returned function
8146 pointer is undefined.
8150 '``llvm.adjust.trampoline``' Intrinsic
8151 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8158 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8163 This performs any required machine-specific adjustment to the address of
8164 a trampoline (passed as ``tramp``).
8169 ``tramp`` must point to a block of memory which already has trampoline
8170 code filled in by a previous call to
8171 :ref:`llvm.init.trampoline <int_it>`.
8176 On some architectures the address of the code to be executed needs to be
8177 different to the address where the trampoline is actually stored. This
8178 intrinsic returns the executable address corresponding to ``tramp``
8179 after performing the required machine specific adjustments. The pointer
8180 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8185 This class of intrinsics exists to information about the lifetime of
8186 memory objects and ranges where variables are immutable.
8188 '``llvm.lifetime.start``' Intrinsic
8189 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8196 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8201 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8207 The first argument is a constant integer representing the size of the
8208 object, or -1 if it is variable sized. The second argument is a pointer
8214 This intrinsic indicates that before this point in the code, the value
8215 of the memory pointed to by ``ptr`` is dead. This means that it is known
8216 to never be used and has an undefined value. A load from the pointer
8217 that precedes this intrinsic can be replaced with ``'undef'``.
8219 '``llvm.lifetime.end``' Intrinsic
8220 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8227 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8232 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8238 The first argument is a constant integer representing the size of the
8239 object, or -1 if it is variable sized. The second argument is a pointer
8245 This intrinsic indicates that after this point in the code, the value of
8246 the memory pointed to by ``ptr`` is dead. This means that it is known to
8247 never be used and has an undefined value. Any stores into the memory
8248 object following this intrinsic may be removed as dead.
8250 '``llvm.invariant.start``' Intrinsic
8251 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8258 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8263 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8264 a memory object will not change.
8269 The first argument is a constant integer representing the size of the
8270 object, or -1 if it is variable sized. The second argument is a pointer
8276 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8277 the return value, the referenced memory location is constant and
8280 '``llvm.invariant.end``' Intrinsic
8281 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8288 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8293 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8294 memory object are mutable.
8299 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8300 The second argument is a constant integer representing the size of the
8301 object, or -1 if it is variable sized and the third argument is a
8302 pointer to the object.
8307 This intrinsic indicates that the memory is mutable again.
8312 This class of intrinsics is designed to be generic and has no specific
8315 '``llvm.var.annotation``' Intrinsic
8316 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8323 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8328 The '``llvm.var.annotation``' intrinsic.
8333 The first argument is a pointer to a value, the second is a pointer to a
8334 global string, the third is a pointer to a global string which is the
8335 source file name, and the last argument is the line number.
8340 This intrinsic allows annotation of local variables with arbitrary
8341 strings. This can be useful for special purpose optimizations that want
8342 to look for these annotations. These have no other defined use; they are
8343 ignored by code generation and optimization.
8345 '``llvm.ptr.annotation.*``' Intrinsic
8346 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8351 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8352 pointer to an integer of any width. *NOTE* you must specify an address space for
8353 the pointer. The identifier for the default address space is the integer
8358 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8359 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8360 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8361 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8362 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8367 The '``llvm.ptr.annotation``' intrinsic.
8372 The first argument is a pointer to an integer value of arbitrary bitwidth
8373 (result of some expression), the second is a pointer to a global string, the
8374 third is a pointer to a global string which is the source file name, and the
8375 last argument is the line number. It returns the value of the first argument.
8380 This intrinsic allows annotation of a pointer to an integer with arbitrary
8381 strings. This can be useful for special purpose optimizations that want to look
8382 for these annotations. These have no other defined use; they are ignored by code
8383 generation and optimization.
8385 '``llvm.annotation.*``' Intrinsic
8386 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8391 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8392 any integer bit width.
8396 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8397 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8398 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8399 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8400 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8405 The '``llvm.annotation``' intrinsic.
8410 The first argument is an integer value (result of some expression), the
8411 second is a pointer to a global string, the third is a pointer to a
8412 global string which is the source file name, and the last argument is
8413 the line number. It returns the value of the first argument.
8418 This intrinsic allows annotations to be put on arbitrary expressions
8419 with arbitrary strings. This can be useful for special purpose
8420 optimizations that want to look for these annotations. These have no
8421 other defined use; they are ignored by code generation and optimization.
8423 '``llvm.trap``' Intrinsic
8424 ^^^^^^^^^^^^^^^^^^^^^^^^^
8431 declare void @llvm.trap() noreturn nounwind
8436 The '``llvm.trap``' intrinsic.
8446 This intrinsic is lowered to the target dependent trap instruction. If
8447 the target does not have a trap instruction, this intrinsic will be
8448 lowered to a call of the ``abort()`` function.
8450 '``llvm.debugtrap``' Intrinsic
8451 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8458 declare void @llvm.debugtrap() nounwind
8463 The '``llvm.debugtrap``' intrinsic.
8473 This intrinsic is lowered to code which is intended to cause an
8474 execution trap with the intention of requesting the attention of a
8477 '``llvm.stackprotector``' Intrinsic
8478 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8485 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8490 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8491 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8492 is placed on the stack before local variables.
8497 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8498 The first argument is the value loaded from the stack guard
8499 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8500 enough space to hold the value of the guard.
8505 This intrinsic causes the prologue/epilogue inserter to force the
8506 position of the ``AllocaInst`` stack slot to be before local variables
8507 on the stack. This is to ensure that if a local variable on the stack is
8508 overwritten, it will destroy the value of the guard. When the function
8509 exits, the guard on the stack is checked against the original guard. If
8510 they are different, then the program aborts by calling the
8511 ``__stack_chk_fail()`` function.
8513 '``llvm.objectsize``' Intrinsic
8514 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8521 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8522 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8527 The ``llvm.objectsize`` intrinsic is designed to provide information to
8528 the optimizers to determine at compile time whether a) an operation
8529 (like memcpy) will overflow a buffer that corresponds to an object, or
8530 b) that a runtime check for overflow isn't necessary. An object in this
8531 context means an allocation of a specific class, structure, array, or
8537 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8538 argument is a pointer to or into the ``object``. The second argument is
8539 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8540 or -1 (if false) when the object size is unknown. The second argument
8541 only accepts constants.
8546 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8547 the size of the object concerned. If the size cannot be determined at
8548 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8549 on the ``min`` argument).
8551 '``llvm.expect``' Intrinsic
8552 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8559 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8560 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8565 The ``llvm.expect`` intrinsic provides information about expected (the
8566 most probable) value of ``val``, which can be used by optimizers.
8571 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8572 a value. The second argument is an expected value, this needs to be a
8573 constant value, variables are not allowed.
8578 This intrinsic is lowered to the ``val``.
8580 '``llvm.donothing``' Intrinsic
8581 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8588 declare void @llvm.donothing() nounwind readnone
8593 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8594 only intrinsic that can be called with an invoke instruction.
8604 This intrinsic does nothing, and it's removed by optimizers and ignored